DECISION SUPPORT PROJECT BP-2 Spatial & Economic Analysis for Policy and Decision Support in Agriculture and Environment w-~~~-... ,... -,.,~,:-'~~-= - .· ~ .. '.- -· ~~ 11( ' - • f ' .. -- .._ ( • •. 1 ,_ ... - .... - i i ·• \ ,, ·: U.'\l i At) O~ :. ,¡O .W.ACIUN Y LLilUfva:N 1 ACION AnnuaJ Report 2006 r• csorull , Centro lnlemadonal de AQI1eutuo Tropical lnterna•a"lCII Cent• 1or Tropical AQIIcutue TABLE OF CONTENTS Decision Support Project Annual Report Page ~lJ~ ~~()lt1r ~1]~~1{ • .. •. •••••.•.••••.• •. •• ••• . ...•..•.••• . •••.••• ••. •.••.•. • .• •• •••. 5 TREME 1: UNDERSTANDING SPATIAL AND TEMPO~ V ARIABILITY OF PLANT -ENVIRONMENT AL INTERACTIONS WITHIN AN ACROSS LANDSCAPES Papers Phenotypic variation in important economic species of medicinal plants and its implications for product quality management in higb val u e cbains ... . .. .. . .. . . . . ... .. . . .. ... 21 Tbe effect of climate change on crop wild relatives..... .... .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 34 Anotber dimension to grazing systems: Soil carbon.. ... .... ... ..... .. .. ... .. ..... .. .. . . . . . . . . . . . . . . . . 49 THEME 2: ~ALIZING THE POTENTIAL OF SPATIAL AND TEMPO~ V ARIABILITY OF PLANT -ENVIRONMENT INTERACTIONS Papers Systematic agronomic farm management for ímproved coffee quality ... ... ......... .... .. ... .... 74 Geographical analyses to explore interactions between inberent coffee quality and production environment. .. .. . ... . .. . .. ... . . . ...... . ... .... ... .. . ... ... . . . ... .. . .. . . .. 94 TREME 3: IDENTIFYING THE CONDITIONS UNDER WlllCH ENVIRONMENTAL GOODS AND SERVICES CONTRIBUTE TO EQIDTABLE AND SUSTAINABLE DEVELOPMENT Papers High-value agricultural products: Can smallholder farmers also benefit? ..... .... ...... JOS Watershed management and poverty alleviation in the Colombian Andes ......... .. ..... ... ............. ... ........ .. ..... . ......... .. .... .. .... ...... ... ... ...... .... .. .... .. .... .. 116 Dynamics and def"mition of poverty in the Colombian Andes: Participative vs. objective approaches ... .... .. .... .. ... ...... .. .. ... .. .... .. .. ..... .... .. ........ ..... 136 Agricultural water productivity: Issues, concepts and approaches Basin Focal Project Working Paper no. 2a .................. ... ..... . ... .. . ... .. ............. ... ... . 158 Water productivity: Measuring and mapping in benchmark basins Basin Focal Project Working Paper no. 2 ...................... .. .. .. ...... .. ....... .... .. .. ......... 176 Analyzing water poverty: water, agriculture and poverty in basins Basin Focal Project Working Paper no. 3 .................. .. ....... .............. .... ............... 193 Environmental and socio-economic evaluation of prototype forest plantations in Cordoba department, Colombia ...... .... .................. .. ...................... 212 lmpacts and indicators of impact of fair trade, fair trade organic, specialty coffee ....................... .......... ......... .. .. ............. ..... .. ............ ... ..... ....... ... 223 THEME 4: ESTIMATING THE MAGNITUDE AND DISTRIBUTION OF SOCIO-ECONOMIC AND ENVIRONMENT AL IMP ACTS OF AGRICUL TURAL R&D ON POVERTY AND VULNERABILITY IN RURAL COMMUNITIES Papers Strategic approaches to targeting technology generation: Assessing the coinciden ce of poverty and drougbt-prone crop production ........................ .. ...... ... 237 How cost-effective is biofortification in combating micronutrient malnutrition? An ex-ante assessment ........ . ............ ........... .................................. 256 Spatial trade-off analyses for site-sensitive development interventions in upland systems of Southeast Asia .......... .. ..................... ....... . ... ...... ............ ......... 277 Assessing the potential impact of the consortium for improved agriculture-based liveliboods in Central Mrica (CIALCA): Spatial targeting of research activities........... ............. .. . . . . . .. . .. . . . . . . . . . . . . . .. . . . . . . .. . .. . . . . . .. .. ... 283 THEME 5: OBTAINING, DEVELOPING, AND MANAGING DATA AND INFORMATION Papers An evaluation of void-filling interpolation metbods for SRTM data ... .................. ... 294 Information management for product differentiation in supply cbains: The case of speciality coffee .................................... . ........... ...... ............ .... ... ... ... 319 Satellite imagery and information networks for monitoring climate and vegetation in Colombia ............................ ........ ........... ............................... 338 Assessment of high speed internet for remote sensing data acquisition and exchange in Colombia and Latin America .................................... 345 Method of processing MODIS images for Colombia ...... ...... ............... .. .......... ... ... 352 ldentifying Critical Issues to Promote Technical Change and Enhance the Efficiency and Competitiveness of the Beef Sector in Costa Rica... ... ... ... ............ .. .... ..... .. .................................. . ........... ........ . ...... 360 New systems of agricultural production and environmental services: an economic evaluation in the Altillanura of Colombia ...................................... 367 Communities and Watershed project ....................... ....... ..................... ............... 372 Protocol for the characterization of carbon and water in high mountain ecosystems ... .... ............... ... ... .. ....... .. ... ....... .... ... .... ................ ................ ........ .. 373 Youth Bolivia: alliance for water-science and the future CGIAR- CIDA Canada Linkage Fund ..... ................................... ............ .............................. .. .. 376 Role of Andean wetlands in water availability for downstream users, Barbas watershed Colombia ....... ........ ... .. .. ............................................ .............. ... ..................... 380 Youth, leadership and researcb: lmproving education for development ..............•.. 384 ANNUALREPORTS~ARY Project BP-2: DECISION SUPPORT l. Project Logframe Our goal is to improve the targeting of investments in agricultura! and NRM research and development through economic and geographic analysis. Our objective is to develop and provide analysis, information and tools to improve decisions about where, wben and how innovations can be implemented to enhance rural livelihoods in a sustainable and equitable manner. 5 Project Logframe (MTP 2006 - 2008) O utputs lntended User O utcome l mpact lmplications ofaltemative Scientists and research Decision-makers informed lmpacts to R&D R&D decisions analyzed managers; development regarding potential tradtoffs investments are more ... planners and rtsu/Jingfrom the a/location of efficient, equitable, !:; practitioners; research or development funds, sustainable ~ policymakers; donors either directly or indirectly via :;:¡ and others who make changts in policy o decision about how R&D resources are invested. Valuation ofproductivity Research managers in Depending on the results, R&D investrnents benefits and environmental ClA T; Policymakers and research priorities for R&D are generate beneficia( services generated by land planners in Colombia and confirmed or revised. impacts, minimizing use systerns in Colombia other countries with tradeoffs between similar ecosysterns environmental and productivity benefits. Assessrnent of early CIA T forages project and Researchers, research managers Benefits of improved :¡ adoption and impact of national partners; CIA T and extensionists better forages on farmer welfare ~ improved forages in Asia research management understand what did and didn ' t are larger and more ... work, and use the knowledge to widespread Cll ¡.... institutionalize results of current project and to improve design of future projects. Analysis of the potential CPWF and research An estimation ofthe likely impact R&D investments on water impact of water research partners including NGO's, on poverty and improvernents in related project better projects under Universities and water management as a resultan! of targeted. irnplementation by the CPWF environmental authorities. CPWF implementation in prioritized basins. M Frarneworks and tools for Researchers and analysts Researchers use their better R&D efforts more ¡.... evaluating and targeting in ClA T and partner conceptual and empirical effective, equitable and :;:¡ technology andlor organizations understanding ofhow impact sustainable. c.. management alternatives in occurs and is measured to design ¡.... ;::¡ agriculture and NRM R&D more impact-oriented projects. o User friendly empírica! too! Researchers and planners Projects and policies about Economic and for quantifying and valuing working on economics of payment for environmental environmental impacts of environmental services environmental services services schemes are better payment for developed designed and targeted environrnental services schemes are more effective, equitable, wide- spread and sustainable . ., ... Homologue concept Decision makers in Tools are usedfor identijication of More effeelive locating and "' to.o ... demonstrated, veri fied and producer associations, genetic resourus that are deployed targeting of gumplasm Cll ¡.... published. NGOs, and GOs. to supporl agricu/Jural leads to higher welfare and development. environmental benefits Concepts and principies and Policy makers and Identification of sites that can most Widespread implementation potential for site sensitive planners in agriculture and benefit from natural hazard of insurance for smallholder natural hazard insurance. fmance ministries, insurance. farmers. Enhanced incorne, Organizations that work Enhanced effectiveness of scheme equity and land for and with producers. implementation management. Spatial, economic and other Researchers interna! and Researchers and decision makers Better analysis and t') information and data externa! to ClA T, ha ve readily-accessible accurate decisions are made thereby ¡... ;::¡ developed, maintained and agricultural decision and appropriate information from enhancing impacts ¡: made available to interna! and makers. which to conduct analysis and base :;:¡ externa! users. actions. o Global derivates ofhigh- Researchers interna! and Accurate topographic informa/ion resolution digital elevation externa! to ClA T. National incorporated in analysts of - "' models for tropical areas. agricultura! and agrobiodiversity and in reuarch to.o environmental NGOs and on soi/ and water management ... Cll GOs. ¡.... 6 2. Output Targets OUTPUT 1: Implications of alterna ti ve R&D decisions analyzed Output t.argets 2006 and Achievement Evidence Valuation of productivity benefits Rubiano, J., Quintero, M., Estrada. R. & Moreno, A. (2006). and environmental services 100% Multiscale Analysis for Promoting lntegrated Watershed generated by land use systems in Management. Water lntemational Vol 31, No.3. Forthcoming Colombia lmpact pathways constructed for 6 out of9 CPWF basins http://impactpathways.pbwiki.com Analysis of the potential impact of Draft impact pathway narratives constructed for 3 basins and water research projects under 500/o posted on the web. implementation by the CPWF in Extrapolation domain analysis completed for 3 basins. prioritized basins. Impact pathways methodology developed and published (see lPRA 2006 Output) OUTPUT 2: Frameworks, tools for evaluating and targeting technology andlor management altematives U ser friendly empírica! too! for Quintero, M., Estrada, R.O. y García, J. 2006. A Manual for ECOSAUT: A Model for the Economic, Social and Environmental quantifying and valuing 100% Evaluation ofLand Use. CIAT-CIP-GTZ-CONDESAN-WFCP. environmental services developed Centro Internacional de la Papa. Lima, Perú. 86 p.(CD-ROM) Homologue concept demonstrated, Homologue™ Version Beta a.O. A computer system for identifying 95% similar environments throughout the tropical world. Jones, Diaz, verified and published Cock. CD with software and users manual. (CD-ROM) Final Report: A system of drought insurance for poverty alleviat.ion Concepts and principies and potential in rural areas. A feasibiJity study of a practica! method of drought for site sensitive natural hazard 100% insurance that is self-sustaining and ready for use by poor farmers, insurance NGOs or other development organizations. Diaz Nieto et al. 95 pages. OUTPUT 3: Spatial, economic and other information and data Oownload webpage @ htto://srtm.csi.cgiar.org/; Reuter H.I, A. Nelson, A.Jarvis, accepted, An evaluation ofvoid Global derivates of high-resolution f:illing interpolation methods ods for SRTM data, lnternational digital elevation models for tropical 100% Journal of Geographic lnformation Science. areas Jarvis, A., Rubiano, J., Nelson, A., Farrow, A., & Mulligan, M. Practica! use of SRBM data in the tropics- Comparisons with digital elevation models generated from cartographics data. Working Document no. 198, 32 pp. CIAT, Cali, Colombia. 3. Research Highlights • Increased User Utility for the Topographic Data Base The Digital Elevation Model that has been derived from the February 2000 Shuttle Radar Topography Mission (SRTM) has been one of the most important publicly available new spatiaJ datasets in recent years. However, the ' finished' grade version of the data stiU contains data voids (sorne 836,000 km2) - and other anomalies - that prevent immediate use for a wide range of applications. These voids can be filled using a range of interpolation aJgorithms in conjunction with other sources of elevation data, but there is little guidance on the most appropriate void filling 7 method. Project scientists of BP2 and their partners developed (i) a method to fill voids using a variety of interpolators, (ii) a method to determine the most appropriate void filling algorithms using a classification of the voids based on their size and a typology of their surrounding terrain; and (iii) the classification of the most appropriate algorithm for each of the 3,339,913 voids in the SRTM data. Based on a sample of 1,304 artificial but realistic voids across six terrain types and eigbt void size classes, it was found that the choice of void filling algorithm is dependent on both the size and terrain type of the void. The best methods can be generalised as: Kriging or Inverse Distance Weighting interpolation for small and medium size voids in relatively flat Jow-lying areas; Spline interpolation for small and medium sized voids in higb altitude and dissected terrain; Triangular Irregular Network or lnverse Distance Weighting interpolation for Jarge voids in very flat areas, and an advanced Spline Method for Jarge voids in other terrains. • Application of BP2 developed targeting tools by National Grower's Associatioo for the development of Denomination of Origin for higher val u e crops The BP2 project and the directorate for Intellectual Property of the National Federation of Colombian Coffee Growers (FNC) Jed a pilot study to understand the feasibility of supporting the implementation of denomination of origin for coffee and to derive the principies for irnplementing denominations of origin for higher value crops. The rationale behind the study was to understand the relationship between environmental data and quality data of product samples collected from farms during the 2006 harvest. To this end, a field survey was designed on the basis of prior knowledge from similar studies. Technical staff of the regional FNC offices identified the participating farms with the aim of including farms that were accessible and covered the range of conditions that represent the coffee-growing environments in selected departments. A large number of farms were sampled. Each farm was geo-referenced to facilitate analysis of spatial correlation. To reduce variability within the data, product samples were processed in a mobile unit that standardizes harvest and post-harvest processes. Product quality characterization was conducted at the FNC headquarters and at tbe FNC research center CENICAFE in Chincbina. Soil samples were also obtained in each farm. We were able to show that spatial structures in the quality data are related to those found in tbe environmentaJ data and documented clear relationships between growing environment and product quaJity characteristics. • Watershed management and poverty alleviation in the Colombian Andes The relationship between water and poverty was assessed in two watersheds in the Colombian Andes. The methodology includes botb a participatory assessment of current poverty and an analysis of how bousebold poverty status has changed over the last 25 years. Taken together, the two results capture both direct and indirect linkages between water and poverty. They identify situations where win-win solutions may be possible, and also where it is likely that trade offs will be required, not only between environmental, economic growth and equity objectives at the watershed scale, but also between households' welfare objectives and the strategies that they use to achieve them. The results of this research suggest that in the two investigated watersheds, the indirect relationships between poverty and water vía employment and income linkages may be more important than direct linkages via domestic supply. This is consistent with the diversification of rural livelihoods and the importance of off-farm income in poverty reduction. Interventions to enhance domestic supply may have big impacts in a few specific communities, but would not generally contribute much to poverty alleviation. Interventions that would reduce employment in industries like dairying or mining in Fuquene or profitability in small scale agriculture in Coello, could bave significant impacts on poverty, since these ha ve been important pathways out of poverty o ver the past 25 years. 8 • International prize Zayed lnternational Prize. 2006. Scientific Achievements in the Environment (www.zayedprize.org) awarded to the authors ofthe Millennium Assessment. 4. lmportant Project Outcome Large research and development programs working on agricultura! development for the rural poor throughout the developing world were able to systematically direct their interventions to regions and peoples where impacts were the most needed and interventions the most appropriate. This was possible based on using previously developed concepts of poverty mapping, systems analyses and livelihood system analysis. Concrete outputs include development of spatial poverty characterization approaches. These were published in the journal Food Policy and on the project Web site (http://www.povertymap.net/). The integration ofthese methods and high resolution spatial data such as childhood malnutrition, major crop distributions, irrigated areas and environmental data such topography and climate, enabled the systematic targeting of development interventions. The 2003 - 2005 CIA T MTP identifies these outputs as "Output 3 of the PE4-Project Logframe at page 29: Analyses and prediction of socio-economic factors influencing land use development perforrned. Measurable indicators: Distribution of poverty and its causes identified more accurately using spatial inforrnation." A broad group of individuals and organizations utilized the outputs of our poverty mapping initiative. In 2006, our clients and partners downloaded 397 high-resolution poverty maps in forrnats suitable for analysis using geographic information system software (see http://gisweb.ciat.cgiar.org/povertymapping/). Users included officials of the govemment, non-governmental organizations, advanced research institutes, as well as university academics and students. Geographic targeting work based on spatial analysis methods that integrate poverty, demographic, agricultura) and environmental information was also used for a major priority-setting exercise ofthe Generation Challenge Program (GCP), a long term prograrn that invests substantially in research to improve the main crops produced by the poor in high risk drought-prone and marginal areas. Poverty mapping work specifically focused on influencing govemment and non-govemment agencies, as well as universities in Ecuador, Honduras, Mexico, Nigeria, Kenya, Malawi, Sri Lanka, Bangladesh and Vietnam. Areas identified for targeting research and development resources of the GCP include parts of South and Southeast Asia, sub-Saharan Africa, and Mexico and Central America. Maps and related analytical approaches were used by research and development targeting exercises and academic research. For example, Ecuador uses poverty and food security maps from the project in targeting food security resources. CIMMYT used their Mexico analysis to re-orient their breeding and variety testing programs to poor and marginal environments. International agencies used Bangladesh poverty maps to respond to flooding by identifying the coincidence of damage and the vulnerability of the population. In Kenya, the poverty maps were integrated into local information systems, and were used to target development assistance. Mapping work for the Generation Challenge Program was used to prioritize and focus the work on a reasonable number of crops and environments where impacts on reducing poverty was the most needed. Use of poverty maps was partly documented when our partners and clients downloaded high-resolution poverty maps from our Web site. The Generation Challenge Program adopted the poverty-drought analysis and database as a key element of their overall strategic planning and prioritization. The use of sorne of the studies was documented in the publication, "Where the poor are: an atlas ofpoverty", published by Columbia University. 9 5. Project Publications Articles in refereed journals Fisher, M.J. , Braz, S.P., Dos Santos, R.S.M., Urquiaga, S., Alves, B.J.R. and Boddey, R.M. (2007). Another dimension to grazing systems: Soil carbon. Tropical Grasslands 41: (In press). Gijsman, A.J., Thornton, P.K. and Hoogenboom, G. (2007) Using the WISE database to parameterize soil inputs for crop simulation models. Computers and Electronics in Agriculture 56:85-1 OO. Nelson, A., Oberthür, T. and Cook, S. (2007). Multi-scale correlations between topography and vegetation in a hillside catchment of Honduras. International Journal of Geographical Information Science 21(2): 145-174. Oberthür, T., Cock, J., Andersson, M.S., Naranjo, R.N., Castañeda, D. and Blair, M. (2007). Acquisition of low altitude digital imagery for local monitoring and management of genetic resources. Computers and Electronics in Agriculture. (In press). Ocampo, J., Coppens d'Eeckenbrugge, G., Restrepo, M., Salazar, M., Jarvis, A. and Caetano, C. (2007). Diversity of Colombian Passifloraceae: an updated list for conservation. Biota Colombiana. (In press). Otero, M.F., Rubiano, J.E., Lema, G. and Soto, V. (2006). Using similarity analyses to scale out research findings. Water International. Special Issue on Scales and Water Resources Management 31(3): l-26. Peralta, A., García, J.A. and Johnson, N. (2006). Dinámica y defmición de pobreza en los Andes colombianos: Enfoques participativos versus enfoques objetivos = Dynamics and defmitions of poverty in the Colombian Andes: Participatory and objective approaches. Desarrollo y Sociedad (58):1-48. Reuter, H.l. , Nelson, A. and Jarvis, A. (2007). An evaluation of void filling interpolation methods for SRTM data. International Journal ofGeographic Information Science. (In press). Rubiano, J., Quintero, M., Estrada. R. and Moreno, A. (2006). Multiscale Analysis for Promoting Integrated Watershed Management. Water International. Special Issue on Scales and Water Resources Management 31(3):1-38. Swallow, B., Johnson, N., Meinzen-Dick, R. and Knox, A. (2006). The challenges of inclusive cross- scale collective action in watersheds. Water International. Special Issue on Scales and Water Resources Management 31(3):1-37. Tomich, T.P., Timmer, D.W., Alegre, J., Arskoug, V., Cash, D.W., Cattaneo, A. , Cornelius, J., Ericksen, P., Joshi, L., Kasyoki, J., Legg, C., Locatelli, M., Murdiyarso, D., Palm, Ch., Porro,R. , Perazzo, A.R., Salazar-Vega, A., Van Noordwijk, M., Velarde, S., Weise, S. and White, D. (2007). Integrative science in practice: Process perspectives from ASB, the Partnership for the Tropical Forest Margins. Agriculture Ecosystems & Environment 121(3):269-286. Yeaman, S. and Jarvis, A. (2006). Regional heterogeneity and gene flow maintain variance in a quantitative trait within populations of lodgepole pine. Proceedings of the Royal Society B: Biological Sciences 273: 1587-1593. 10 Books and monographs Quintero, M., Estrada, R.D. and García, J. (2006). A Manual for ECOSAUT: A Mode/ for the Economic, Social and Environmenta/ Evaluation of Land Use. CIAT-CIP-GTZ-CONDESAN- WFCP, Centro Internacional de la Papa, Lima, Perú. 86 pp. Quintero, M., Estrada, R.O. and García, J. (2006). Modelo de optimización para evaluación ex ante de alternativas productivas y cuantificación de externalidades ambientales en cuencas andinas. ECOSAUT. CIAT-CIP-GTZ-CONDESAN-WFCP, Centro Internacional de la Papa, Lima, Perú. 76 pp. Utderach, P. (ed). (2006). Jmproving co.ffee quality or converting marginal areas. Seminar Proceedings Agricultura) Science and Resource Management in the Tropics and Subtropics ARTS, Universitiit Bonn, Germany. 150 pp. White, D., Rondón, M., Hurtado, M.P., Rivera, M., García, J. , Amézquita, E., Rodríguez, C.A. (2006). Valoración Ambiental y Socio-Económica de Plantaciones Forestales Prototipos en el Departamento de Córdoba, Colombia. CIA T- CVS. 68 pp. Book Chapters Bode, R., Liiderach, P. and Oberthür, T. (2006). Gestión de alta calidad- percepciones, lenguajes y paradigmas. In: Pohlan, J. , Soto, L.and Barrera, J. (eds.) El Cafetal del Futuro: Realidades y Visiones. Aachen, Shaker Verlag, Alemania. Pp. 161-176. Liiderach, P., Oberthür, T., Niederhauser, N., Usma, H., Collet, L. and Pohlan, J. (2006). Café Especial: Factores, dimensiones e interacciones. In: Pohlan, J., Soto, L. and Barrera, J. (eds.) El Cafetal del Futuro: Realidades y Visiones. Aachen, ShakerVerlag, Alemania. Pp. 141-160. Niederhauser, N. and Ritter, W. 2006. User interface for mobile data collection in rural development areas. In: Kempter, G. and von Hellberg, P. (Hrsg) Information nutzbar machen: Zusammenfassung der Beitrage zum Usability Day IV, Pabst Science Publishers, Lengerich, DE. Pp. 100-104. Papers presented at formal conferences and workshops with external attendance Atzmanstorfer, K., Oberthür, T., Liiderach, P., O'Brien, R., Collet, L., and Quiñones, G. (2006). Probability Modelling to Reduce Decision Uncertainty in Environmental Niche ldentification and Driving Factor Analysis: CaNaSTA Case Studies. Conference and Exhibition on Applied Geoinformatics-AGIT, Geolnformation for Development- gi4dev AgitSPEClAL. Salzburg, Austria, 07 July, 2006. Barona, E., Girón, E., Feistner, K.L., Dwyer, J.L. and Hyman, G. (2006). Método de procesamiento de imágenes modis para Colombia. XTI Simposio Internacional en Percepción Remota y Sistemas de Información Geográfica SELPER- Capítulo Colombia. Cartagena, Colombia, 24-29 September, 2006. Bolaños, S.L. (2006). Integración de Sistemas de Información Geográfica y Teledetección para Mapeo de Áreas de Café. XTI Simposio Internacional en Percepción Remota y Sistemas de Información Geográfica SELPER- Capítulo Colombia. Cartagena, Colombia, 24-29 September, 2006 Cook, S.E., Fisher, M., Diaz-Nieto, J. and Lundy, M. (2006). New Financia! lnstruments to Help Improve Agricultura) Water Management for Poor Farmers Under Conditions of Risk. World Water Week. Stockholm, Sweden, 20-26 August, 2006. Cook, S.E., Jarvis, A. and González, J.P. (2006). A new global demand for digital soil information. Global Workshop on Digital Soil Mapping for Regions and Countries with Sparse Soil Data Infrastructures. Rio de Janeiro, Brazil, 4-7 July, 2006. 11 Estrada, M., Uiderach, P., Oberthtir, T. and Pohlan, H.A.J. (2006). Análisis de las interacciones y del impacto de condiciones ambientales, agronómicas, y el manejo innovador sobre la calidad de taza del café (Coffea arabica L.). X Congreso Internacional de Manejo Integrado de Plagas y Agroecología. Tapachula, Chiapas, México, 27-29 September, 2006. Giran, E., Perea, C.J. and Hyman, G. (2006). Aplicación de mapeo en la web utilizando soluciones SIG de código abierto para la diseminación de información satelital sobre redes de alta velocidad como apoyo a la investigación agrícola y el manejo de recursos naturales. XII Simposio Internacional en Percepción Remota y Sistemas de Información Geográfica SELPER- Capítulo Colombia. Poster Presentation. Cartagena, Colombia, 24-29 September, 2006. Gonzalez, C.E., Jarvis, A. and Palacio, J.D. (2006). Biogeografia del roble común (Quercus humboldtii Bonpl.): distribución geográfica y su adaptación climática. Simposio Internacional sobre Robles y Ecosistemas Asociados. Santafé de Bogotá, Colombia, 11-12 May, 2006. Hyman, G., Kam, S. P., Legg, C., Farrow, A., Hodson, D. and Benson, T. (2006). Poverty and Food Security Mapping at Country-level: Lessons Learned from Seven Case Studies. IX lnternational Conference ofthe Global Spatial Data Infrastructure-GSDI-9. Santiago, Chile, 3-11 November, 2006. Hyman, G., Meneses, C., Barona, E., Girón, E. and Perea, C.J. (2006). Satellite imagery and information networks for monitoring climate and vegetation in Colombia. XII Simposio Internacional en Percepción Remota y Sistemas de Información Geográfica SELPER- Capitulo Colombia. Cartagena, Colombia, 24-29 September, 2006. Jarvis, A., Fisher, M., Jones, P., Cook, S.E. and Guarino, L. (2006). Agriculture, Risk and Climate Change. International Workshop on Tropical Agriculture Development Transforming Tropical Agriculture: An Assessment of Major Technological, lnstitutional, and Policy lnnovations. Brasilia, BrazH, 20-22 July, 2006. Jarvis, A., Paternina, M.J., Arcos, A., Rodríguez, H.J., Nagles, C. and Meto, N. (2006). Evaluación Rápida de la Adaptación al Medioambiente de Plantas Promisorias Medicinales. 11 Congreso Internacional de Plantas Medicinales y Aromáticas. Palmira, Colombia, 19-21 October, 2006. Johnson, N. and Peralta, A. (2006). Dynamics and definitions of poverty in the Colombian Andes: Participatory vs. objective approaches. International Forum on Water and Food. Vientiane, Lao POR, November 12 - 17, 2006. Johnson, N., Rubiano, J.E. and Peralta, A. (2006). lntroduction to Theme 2 (Water and People in Catchments) and the payment for environmental services in soil and water (PES-SW) initiative. Workshop on potential for Payment for Environmental Services (PES) approaches to contribute to equitable and sustainable management ofsoil and water in upper catchments. Nairobi, Kenya, 27-29 June, 2006. Uiderach, P., Collet, L., Oberthür, T. and Pohlan, H.A.J. (2006). Café especial y sus interacciones con factores de producción. 11 Diplomado sobre cafeticultura sustentable. Tuxtla Gutiérrez, Chiapas, 2006. Lliderach, P., Vaast, P., Oberthilr, T., Obrien, R., Lara-Estrada, L.D. and Nelson, A. (2006). Geographical Analyses to Explore lnteractions between lnherent Coffee Quality and Production Environment. XXI International Conference on Coffee Science. Montpellier, France, 11-15 September, 2006. Lane, A., Jarvis, A. and Hijmans, R.H. (2006). Crop Wild Relatives and Climate Change: predicting the loss of important genetic resources. ESSP Global Environmental Change Open Science Conference. Beijing, China, 9-12 November, 2006. Lentes, P., Peters, M., White, D., Holmann, F. and Reiber, C. (2006). "Assessing and Comparing lncome Generation ofLivestock Holders in Olancho, Honduras. An Analysis across Landscapes and Farming Systems." Poster presentation. Tropentag. Bonn, Germany, 11-13 October, 2006. Lozano, J. , Lema, G. and Hyman, G. (2006). Mapeo de suelo a nivel de fmca con métodos geoestadisticos. Estudio de caso en el Valle del Cauca. XII Simposio Internacional en Percepción Remota y Sistemas de Información Geográfica SELPER- Capítulo Colombia. Poster Presentation. Cartagena, Colombia, 24-29 September, 2006. 12 Niederhauser, N. (2006). Geo-information management for agricultura) high value product supply chains. Conference and Exhibition on Applied Geoinformatics-AGIT, Geolnfonnation for Development- gi4dev AgitSPECIAL. Salzburg, Austria, 07 July, 2006. Niederhauser, N. and Ritter, W. (2006). User Interfaces for Mobile Data Collection in Rural Development Areas. lnternational Conference Usability Day IV. Vorarlberg, Austria, 9 June, 2006. Oberthür, T., Cock, J., Niederhauser, N. and Kattnig, S. (2006). lnformation Management for Product Differentiation in Supply Cbains: The Case of Specialty Coffee. lnternational Conference on Coffee Science. Montpellier, France, 11-15 Septem ber, 2006. Pauli, N., Barrios, E., Oberthür, T. and Conacher, A. (2006). Earthwonns and other soil invertebrates in the Quesungual agroforestry system of Honduras: Distribution patterns and implications for management. IV Annual Meeting of the Conservation and Sustainab1e Management of Belowground Biodiversity project-GEFIUNEP. Xalapa-Catemaco, México, 2006. Peralta, A., García, A.J. and Johnson, N. (2006). Watershed management and poverty alleviation in the Colombian Andes. lnternational Forurn on Water and Food. Vientiane, Lao POR, November 12-17, 2006. Perea, C., Hyman,G., Giron, E. and Barona, E. (2006). Tiwiki y Mapserver: Herramientas colaborativas y mapas en línea para el monitoreo de clima y vegetación en Colombia usando información satelitaJ. XII Simposio Internacional en Percepción Remota y Sistemas de Información Geográfica SELPER- Capítulo Colombia. Poster presentation. Cartagena, Colombia, 24-29 September, 2006. Peters, M., White, D. , Fujisaka, S., Franco, L.H., Lascano, C., Muñoz, L.S., Sarria, P., Montoya, C.A., Vivas, N., Arroyave, 0., Lentes, P., Schmidt, A. and Mena, M. (2006). "Forage-based Protein Feeds for Smallholder On-farm Pig and Poultry Production and the Feed Industry". Simposium- Beyond the Cow: 101 Uses for Forages and Grasslands. American Society for Agronomy, ASA-CSSA-SSSA lntemational Annual Meetings. lndianapolis, USA, 12-16 November, 2006. Suárez, L.A., Uiderach, P., Oberthür, T. and Pohlan, H.A.J. (2006). Impacto de la iluminación y del grado de sombrío en los atributos de la calidad del café. XX Congreso Internacional de Manejo Integrado de Plagas y Agroecología. Tapachula, Chiapas, México, 27-29 September, 2006. Uribe, N., Oberthur, T. and Hyman, G. (2006). Valoración de los diferentes métodos de corrección topográfica en imágenes de satélite aplicado a la respuesta espectral del café. XII Simposio Internacional en Percepción Remota y Sistemas de Información Geográfica SELPER- Capítulo Colombia. Poster presentation. Cartagena, Colombia, 24-29 September, 2006. White, D., Clavijo, L.A., Lundy, M., Oberthür, T., Ribeiro, M.M., Hauser, M. and Damhofer, l. (2006). "Fostering site-specific market options to improve rural livelihoods and land management in Laos." Poster presentation. IDRC workshop. Community-based Natural Resources Management. Vientiane, Lao POR, March, 2006. Yamamoto, Y., Oberthür, T., Lefroy, R., Kashiwagi, J. (2006). Capability of Satellite lmagery for Land-Use Analysis. ll lnternational Conference on Sustainable Sloping Lands and Watershed Management: Linking research to strengthen upland policies and practices. Luang Prabang, Lao POR, 12-15 December, 2006. 13 Articles in international newsletters or other scientific series Cook, S.E., Turral, H. and Gichuki, F. (2006). Agricultura/ Water Productivity: Jssues, Concepts and Approaches. Basin Focal Project Working Paper no. l. 17 pp. Cook, S.E., Turra!, H. and Gichuki, F. (2006). Estimation at plot, farm and basin scale. Basin Focal Project Working Paper no. 2. 1-16 pp. Cook, S.E., Gichuki, F., Turral, H. and Fisher, M.J. (2006). Analyzing Water Poverty: Water, Agriculture and Poverty in Basins. Basin Focal Project Working Paper. no. 3. 1-18 pp. Bode, R. (2006). Knowledge management in val u e chains. The case of the specialty coffee association FAPECAFES, Ecuador. In: Services for Rural Development. Knowledge systems in rural areas. Bulletin ofthe Sector Project, GTZ 14: 41-46. Niederhauser, N. and Oberthür, T. CinfO Information management for coffee supply chains. Informa/ion Platformfor Diversification ofCoffee Lands. http://www.cinfo.it/ Segnestam, L., Simonsson, L., Rubiano, J. E. and Morales, M. (2006). Cross-level institutional processes and vulnerability to natural hazards in Honduras. Stockholm Environment lnstitute. 64 p. Yamamoto, Y., Oberthür, T. and Lefroy, R. (2006). Rainfed Agriculture in Northern Laos -Identification of Land Use Cycles in Slash-and-Burn Agriculture by Satellite Imagery. JIRCAS Working Report no. 47: 1-6. 14 6. Funded Project Proposals Project Donor Total US$ in 2006 CIAT Challenge Program on Water and Food (CPWF) Theme 2 TWMI 303,423 303,423 New opportunities for hillside farmers: Matching product quality, environment and market demand for high-value GTZ 485,541 329,296 agricultura! products Spatial Trade-Off Analyses for Site-Sensitive Development Austria 141,060 94,060 lnterventions in Upland Systems of Southeast Asia Overcoming Land Degradation to Mitigate Deforestation in UNEP/GEF 100,000 31,900 the Humid Tropic Ecoregions Development ofMeasures for Impact on Poverty and Hunger SFL 41 ,727 41,727 in fair-traded and Organic Coffee Producing Communities Agreement between CIAT and CPWF for Provision ofLeader CP 84,489 84,489 ofBasin Focal Projects Análisis Espacial para la Identificación de Amenazas para la TNC 44,750 44,750 Conservación en América del Sur Analyses of Coffee Quality and Production System Characteristics in the Nariño and Cauca Departments of FNC 30,875 30,875 Colombia Sustaining Inclusive Collective Action that Links Across Economic and Ecological Scales in Upper Watersheds CPWF 219,356 47,919 (SCALES) 1,451~21 1,008,439 15 7. Project Staff (* Left during 2006; ,¡ Arrived during 2006) Thomas Obertbür (lOO%) PhD, Geography Project Manager Nancy Johnson (100%) PhD, Economist Senior Scientist Simon E. Cook (50%) PhD, Crop Biology Senior Scientist Glenn G. Hyman (100%) PhD, Geography Senior Scientist Douglas White (75%) PhD, Agr. & Environ. Econ. Senior Research Fellow Andrew Jarvis (100%) PhD., Geography Senior Scientist Roberto Porro (50%) PhD., Anthropology Senior Scientist Arjan J. Gijsman PhD, Soil Science/Crop Modeling Assoc. Senior Scientist Jorge Rubiano* (1 00%) PhD, Geography Postdoctoral Fellow Andrew Farrow (lOO%) MSc,GIS Research Fellow Norbert Niederhauser (1 00%) DI(FH), Inf. & Com. Engineering Research Fellow Carolina González ( 100%) Lawyer and Economist Research Associate William Díaz* (100%) MSc, Admin./System Engineer. Systems Analyst 1 Lil iana Rojas (lOO%) MSc, Natural Resources Research Assistant 1 Viviana Gonzalías (100%) MSc, Sustainable Forestry Research Assistant 2 Andrés J. Pefia* (100%) MSc, Meteorology Research Assistant 2 James García (100%) MSc, Statistician Data Base Specialist Lilian Busingye ( l 00%) MSc, GIS Spatial Analysis Intem Germán Lema ( 1 00%) BSc, Industrial Engineering Statistical Consultant 2 Lilian P. Torres (100%) BSc, Business Administration Administrative Assistant l Luz A. Clavijo (100%) BSc, Geography Research Assistant 1 Germán Escobar ( 1 00%) BSc, Biology Research Assistant 1 Sandra Bolafios (lOO%) BSc, Industrial Engineering Research Assistant l Magda L. Perez* (100%) BSc, Catastral & Geodesta Engineer Research Assistant 1 Jenny L. Correa (100%) BA, Social Communication Research Assistant 1 María A. Peralta ( 100%) BSc, Economist Research Assistant 2 Maree la Estrada ( 1 00%) Agronomy Engineering Research Assistant 3 Mike H. Salazar (100%) Ecology Research Assistant 3 El izabeth Barona ( 1 00%) BSc, Systems Engineer GIS Analyst 3 Silvia E. Castafio (100%) BSc, Systems Engineer GIS Coordinator Claudia J. Perea (lOO%) BSc, Systems Engineer Systems Analyst 3 Jorge A. Cardona (lOO%) BSc, Systems Engineer Systems Analyst 3 Hermano Usma (lOO%) Agricultura! Technology Expert Research 1 Edward Guevara ( 100%) Environmental Engineering Technician 1 Marisol Calderón (100%) Architectural Drawing Office Clerk 1 Juan C. Barona* (lOO%) Topographic Engineer Office Clerk 4 Ovidio Rivera (lOO%) Systems Technology Office Clerk 4 Víctor M. Soto (100%) BSc, Business Administration GIS Expert Alexander Cuero ( 1 00%) Systems Technology GIS Expert Carlos Nagles (100%) Agricultura! Technology GIS Expert Yuviza Barona* (lOO%) Bilingual Secretary Bilingual Secretary Ana M. Guerrero (100%) Bilingual Secretary Bilingual Secretary Peter Uiderach ( 1 00%) MSc, Geography Visiting Researcher Natasha Pauli* (50%) MSc, Biology & Geography Visiting Researcher Juergen Piechaczek (100%) MSc, Agriculture Visiting Researcher Reinhild Bode (100%) MSc, Rural Development Visiting Researcher Fernando Rodríguez (50%) MSc, Business Administration Visiting Researcher Ramiro Cuero (50%) MSc, Candidate Visiting Researcher Walter Ritter..i* (50%) Dl(FH), lnf. & Com. Engineering Visiting Researcher Martín Wiesinger..i (100%) Dl(FH), lnf. & Com. Engineering Visiting Researcher Michael Gau..i (100%) Dl(FH), lnf. & Com. Engineering Visiting Researcher Carlos González (100%) BSc, Biologist Visiting Researcher Karl Atzmanstorfer* ( lOO%) BSc, Geography Visiting Researcher Scott Gebhardt..i ( 1 00%) BSc, Geography Visiting Researcher 16 Daniel Jiménez (100%) Julien Henique~• (100%) Diana Tangarife* {lOO%) Ginger Roberts~ (100%) Aske Bosselman~ (100%) Klaus Dons~ (100%) Abson Sae-Tang~* (100%) Clemens Bertschler* (100%) Marlene Stroj~ (100%) Sibylle Katinig~* {100%) Natalia Uribe* (100%) Luz A. Suárez• ( 1 00%) Liliana Ramírez* (100%) María J. Patemina~ {100%) Diego Sánchez• (lOO%) Gettsy Qujñones • (100%) Peter G. Jones Myles James Fisher Samuel Fujisaka Meike Andersson* James H. Cock Laure D. Collet Agronomic Engineering Agronomic Engineering Environmental Engineering Environmental Development Forestry Engineering Forestry Engineering Electronic Engineering lnf. & Com. Engineering Engineering Computer Science Topographic Engineering Topographlc Engineering Environmental Engineering Agronomic Engineering Economist Statistics PhD, Crop Physiology PhD, Crop Physiology PhD, Anthropology PhD, Animal Science/ Agronomy PhD, Plant Physiology MSc, Environmental Sciences 17 Visiting Researcher Visiting Researcher Visiting Researcher Visiting Researcher Visiting Researcher Visiting Researcher Visiting Researcher Visiting Researcber Visiting Researcher Visiting Researcher Undergraduate Student Undergraduate Student Undergraduate Student Undergraduate Student Undergraduate Student Undergraduate Student Consultant Consultant Consultant ConsuJtant ConsuJtant Consultant 8. Project Budget Summary Project (including Amazon Initiative, and also the Communities and Watersheds Project that is in Phase Out) Decision Support BP-2 SO URCE AMOUNTUS$ PROPORTION (%) Unrestricted Core 933 328 39% Restricted Core 0% 0% Sub-total 933,328 39% Special Projects 785,199 33% Water and Food CP 645,205 27% Total Project 2,363,732 100% Amazon Initiative SO URCE AMOUNTUS$ PROPORTJON (%) Unrestricted Core 38,433 14% Restricted Core 0% 0% Sub-total 38,433 14% Special Proiects 230 660 86% Total Pro_ject 269,092 100% Communities and Watersheds PE-3 SO URCE AMOUNTUS$ PROPORTION (%) Unrestricted Core 110,502 30% Restricted Core 0% 0% Sub-total 110,502 30% Special Projects 252,631 70% 0% Total Project 363 133 100% Communities and Watersheds in Asia PE-3 SO URCE AMOUNTUS$ PROPORTION (%) Unrestricted Core 0% Restricted Core 0% 0% Sub-total o 0% Special Proiects 308 880 100% Total Project 308,880 100% 18 Communities and Watersheds in Central America PE-3 SO URCE AMOUNTUS$ PROPORTION (%) Unrestricted Core o 0% Restricted Core o 0% o 0% Sub-total o 0% Special Proiects 14 453 100% Total Project 14,453 100% 19 THEME 1: UNDERSTANDING SPATIALAND TEMPORAL VARIABILITY OF PLANT-ENVIRONMENTAL INTERACTIONS WITHIN AN ACROSS LANDSCAPES 20 Phenotypic variation in important econo.mic species of medicinal plants and its implications for product quality management in high value chains Andy Jarvis a,b, Adriana Arcos e, Rector Julio Rodríguez d' Thomas Oberthür a, Maria Jose Paternina a, Nelson Melo e a International Centre for Tropical Agriculture CIATr.Cali, Colombia b Bioversity International, c/o CIAT, Cali, Colombia e Instituto de Investigación Alexander von Humboldt IAvH, Bogotá, Colombia d Fundación de Farmacia Natural de Colombia FUNDACOFAN, Cali, Colombia Introduction Markets are h.ighly dynamic and especially so in plants used for medicinal or aromatic purposes. For these species, the markets are characterized by being very specific in terms of their commercial niche, with strict requirements for the characteristics of the agricultural product (for example, percentage content of active ingredients to the leve) of individual compounds ). Yet many of the species used in the natural medicines market are poorly understood, often lacking basic agronomic guidelines for production and post-harvest practices, unknown environmental preferences and almost always have no information about variability of the active ingredients. Furthermore, the genotypes used are often selected locally, and so production systems are built on large varietal diversity with Iittle consideration of the implications of using different lines on product quality. This paper aims to develop a rapid means of evaluating the performance of species across large environmental gradients in order to define rapidly and cheaply the production niche for species with an emerging market, and to examine the important role of phenotypic variation in both biomass production and in the concentration of active ingredients. It is hoped that better understanding of phenotypic variation, and the processes that generate this variability, will permit producers to engage more rapidly in emerging markets, and improve the quality ofthe produce depending on dynamic market preferences. Specifically, the objectives ofthe study are to: • QuantifY the phenotypic variability in biomass and chemical composition and quality) of three important medicinal plant species brought about by different edapho-climatic conditions; • IdentifY the principie causal factors that generate the phenotypic variability; • Analyze the implications of phenotypic variability for high-value market chains of ni che medicinal plant species; and • Suggest a suitable experimental design to permit the rapid evaluation of phenotypic variability in species with emerging economic importance or undefined but significant phenotypic variability in product quantity or quality The work presented here is ongoing, and represents only a preliminary analysis of fmdings from the frrst round of field experiments. Laboratory reports of phenotypic variability in the extracted essential oH have only just been received, and so statistical analyses are not presented. 21 Methodology In order to address the objectives, we took an approach of combining participatory field-based experimentation with laboratory analyses, statistical data analysis and qualitative market studies. Species selection and planting material Through expert consultation, three species were selected based on their market potential and relative lack of agronomic knowledge. Furthermore, sub-specific genotypes with different chemical compositions were identified in two of the species and included in the study to examine within-species phenotypic variability between genotypes. Six genotypes were therefore studied: • Justicia pectoralis Jacq. • Lippia alba Mili. cítrica • Lipia alba Mili. tipica • Lippia origanoides H.B.K. cítrica • Lipppia origanoides H.B.K. tipica • Lippia origanoides H.B.K. patia Justicia pectoralis Jacq.: Commonly known as "amansa toros" in Colombia, this is a herb ofthe family Acanthaceae that can grow up to 1m tall (Estrella 1995). It is used for diverse purposes ranging from insect repellent, insecticide, expectorant, skin-healing treatment, and has been demonstrated to exhibit anti- oxidant activity against lipid peroxidation in brain tissue of rats (Pérez el al. 2002). The essential oil extracted from J pectoralis contains saponines, napthalides, cumarine, betaine, umbelliferone and lignans amongst other components (Estrella 1995). Lippia: This genus of the family Verbenaceae, includes approximately 200 species of herbs, small bushes and small trees. The majority of species are found in Latin America and tropical A frica. Lippia species are used typically to treat respiratory disorders, and provide a wide range of essential oils with pleasant fragrances . The chemical composition of the oils typically consists of limonene, ~cariofilene, p -cymene, camphor, linalol, cx-pinene and thymol (Pascual et al. 2001). Lippia alba (Mili): Of all Lippia spp., L. alba is among the most studied. Known locally in Colombia as "pronto alivio" (literally, "rapid relief'), the species is used very widely as an analgesic, anti-intlamatory, antipyretic (Klueger el al. 1997), as a sedative, a remedy for dysentery, a treatment for disorders of the skin, liver and bladder and gastrointestinal and respiratory disorders, as an anti-malarial and for the treatment ofsyphilis and gonorrhea (Pascual et al. 2001). The principal components ofthe essential oil are borneo!, camphor, 1 ,8-cineol, citronellol, geranial, linalool, myrcene, neral, piperitone, sabinene, 2- undecanone, ex- muurol, ~cariofilene, 13-cubebene, ~elemene, y-cadinene (Pascual et al. 2001). Lippia origanoides: Found to demonstrate anti-microbial activity, especially against Escherichia coli, Staphlococcus aureus MRSA, Candida albicans and Candida tropicalis (dos Santos et al. 2004). The essential oil contains greatest volume of p-cymene, y-terpinene, thymol, carvacrol, butyl hydroxy anisole, cx-terpinene, 1 ,8-cineole, p-thymy 1 aceta te, ¡3-caryophy llene (Pascual el al, 2001 ). Planting material was obtained from the field genebank of the National University of Colombia, run and coordinated by the local NGO Fundación Colombiano de Farmacia Natural (FUNDACOF A N). Planting material was generated from vegetative propagation of single mother plants for all genotypes, and kept 2 months in greenhouses in oasis foam to ensure root development. 22 Field experiments The field experiments were located on 24 farrns in the Cauca department in southem Colombia. Small experimental plots were designed to capture phenotypic variability in biomass production and quality of plant derivatives (essential oils in the case of the two Lippia species, and ethanol extract in the case of J. pectoralis). The experimental design was developed on the basis of the following criteria: • Small-scale to enable replicability for resource-poor organizations and associations • Efficient in capturing G x E interactions across the broadest gradient possible • Capable of providing data useful for identif)'ing the edaphic and climatic factors driving phenotypic variability These criteria lead to an experimental design distinct to classical agronomic tria! sizes, where the number of replications is high and the number of different tria! si tes often limited (<5). The low number of sites causes significant problems in ascertaining the factors driving variability in biomass production as the number of dependent variables (edapho-climatic) often supersedes the number of different sites (n), Jeading to problems in statistical analysis. Twenty-three farrns, widely distributed throughout the Cauca department (Figure 1), were selected as evaluation sites. They lie broadly in four regions: Santander de Quilichao, Totoro, Popayan and Timbioffambo. The farms cover an altitudinal range from 1 OOOm to 3200m, ha ve soils from highly acid to neutral, and annual rainfall varies from 1500mm to 2300mm. 23 Proyectos Plantas Medicinales LocariZadón de Fincas 0 .\ -~ -- Leyenda · ~ . - - Yios Figure l. Distribution of experimental sites in Cauca department, Colombia. Soil samples were taken at each site, and analyzed for 20 different physical and chemical characteristics (Table 1). Manual rain gauges and maximum/minimum thermometers were also installed in each site, and farmers trained in their use. Measurements were taken on a daily basis at 7am and registered on a specially designed form. These data were then used to provide 8 climatic indices for each site, sorne of which were based on a simple water balance model Spatial datasets were also mined to provide a range of climatic variables using the WorldClim climate database (Hijmans et al. 2005) and 19 associated bioclimatic variables (Bus by 1991) (Table 2). 24 Tablel. Soil characteristics for each ofthe 22 experimental sites Fann # Fann name Municip. Lat. Long Elev. pH OM 2 3 4 6 8 9 11 San Diego La Ange1ita El Parnaso Penjamo Las Acacias El Matadero La Fonaleza La Laguna El Provenir La Muyunga Loma Larga El Porvenir VillaStella Asomuripik --- 0 -- masl Wkg S.Quilichao 2.914 -76.457 1955 4.74 83.74 S.Quilichao 2.908 -76.480 1564 5.64 163.46 S.Quilichao 2.884 -76.482 1543 5.66 57.07 S.Quilichao 2.883 -76.510 1470 6.01 78.51 Popayan 2.461 -76.550 1925 4.91 68.33 Popayan 2.420 -76.559 2060 5.56 75.56 Popay8JJ 2.405 -76.559 2162 5.51 258. 14 Popayan 2.434 -76.558 2087 5.58 200.98 El Tambo 2.493 -76.795 1680 5.48 158.63 El Tambo 2.423 -76.766 1730 5.82 134.35 El Tambo 2.415 -76.766 1776 5.70 222.96 El Tambo 2.420 -76.744 1738 5.59 211.16 El Tambo 2.447 -76.746 1703 6.03 122.72 Totoro Vista Hermosa Totoro 2.511 -76.384 2678 5.99 146.3 1 2.514 -76.371 2801 6.02 187.58 p K Ca Mg Al Na N-NH4 N-N03 B Fe Cu Mn Zn CICE Sand Silt Clay Text Bray 11 mWkg ----- cmol/k.g ---- ---------------- mWkg --------------- --%-- 1.39 0.27 5.25 1.43 2.23 0.00 19.97 34.33 4.21 0.58 6.76 2.22 0.00 0.09 15.08 30.46 1.15 1.26 4.06 3.11 0.00 0.05 10.35 12.31 7.33 1.64 7.22 3.46 0.00 0.08 17.56 10.79 3.68 1.35 4.52 1.36 0.85 0.00 17.33 73.85 0.95 0.32 3.35 0.55 0.00 0.03 11 .51 13.46 9.18 0.73 10.23 2.35 0.00 0.09 14.54 39.72 1.50 0.81 6.49 2.14 0.00 0.05 15.74 31.85 0.28 0.42 0.78 0.36 0.00 0.04 15.72 10.09 0.69 0.39 6.50 1.27 0.00 0.04 17.35 23.97 5. 18 1.1 5 5.77 1.70 0.00 0.03 19.60 34.29 6.90 0.83 4.79 1.05 0.00 0.06 18.68 23.39 1.36 2.41 5.11 1.79 0.00 0.10 14.65 28.03 80.50 1.93 11.55 4.18 0.00 0.06 14.57 34.01 7.20 1.26 9.55 2.94 0.00 0.09 17.86 23.59 0.50 5.83 1.54 192.02 3.75 9.19 48.81 12.66 38.53 5 0.55 30. 11 1.29 65.34 0.85 5.06 0.64 2.58 1.1 6 1.71 0.79 1.39 0.35 7.74 0.67 2.17 1.36 6.36 0.96 3.24 0.73 2.22 9.66 60.59 14.49 24.92 11 1.28 53.44 4.54 8.47 23.57 17.67 58.76 0.95 135.83 5.89 12.40 37.35 18.73 43.92 0.75 154.38 4.97 8.09 - 0. 10 8.43 1.22 4.24 48.76 32.13 19.11 2 0. 17 29.36 3.97 13.39 50.2 1 29.38 20.41 2 0. 19 26.44 1.74 9.49 50.11 29.43 20.46 2 0.20 3.82 1.32 1.60 50.38 30.09 19.53 2 12 13 14 15 16 17 18 20 21 22 23 24 San Jose Totoro 2.557 -76.432 2363 5.84 157.Q3 25.34 1.31 10.49 4.15 0.00 0.08 19.31 32.54 2.24 2.52 1.61 3.08 2.05 7.95 0.08 17.45 10.66 8.19 44.37 32.33 23.30 2 0.14 10.1 O 4.62 8.66 53.64 25.68 20.68 JI 0.18 19.12 2.61 6.74 50.92 27.04 22.04 2 0.23 20.79 4.16 9.41 53.34 27.2 1 19.45 8 0.09 71.19 6.49 17.72 49.95 22.19 27.86 11 0.13 28.02 2.32 13.84 49.47 24.86 25.67 11 0. 19 98.73 6.31 16.03 46.04 24.60 29.36 11 El Agrado Totoro Villa Nueva Timbio La Coronita MiraValle Timbio Tambo 2.516 -76.499 2140 5.73 197.76 2.447 -76.701 1756 5.52 229.39 2.442 -76.694 1800 6.77 147.91 2.398 -76.782 1785 7.21 145.23 0.69 0.41 3.54 1.06 0.00 0.04 17.19 24.89 1.51 0.59 3.40 1.07 0.00 0.04 19.46 17.92 7.51 1.79 16.78 4.36 0.00 0.04 14.15 29.32 3.43 1.11 21.86 3.75 0.00 0.07 17. 19 14.82 25 0.81 4.31 0.71 4.95 1.70 0.89 1.20 0.42 0.10 10.30 1.39 5.05 47.11 33.53 19.36 2 0.19 16.15 1.52 5.10 47.90 29.00 23.10 2 0.10 13.77 3.44 22.97 56.42 26.04 17.54 8 0.06 5.98 7.13 26.79 53.47 28.88 17.65 8 Table 2. Climatic conditions at each experimental tria! site based on data extracted from the WorldClim climate database. P6 Pll PI P2 P3 P4 PS Min P7 p8 P9 PIO Mean Max Mean Mean Mean Farm # Farm name Municip. La t. Mean Mean lso- Temp temp Temp temp Long. Elev. 1 d ' 1 th temp of f 1 temp of temp of temp of f 2 3 4 6 8 9 11 12 13 14 15 16 17 18 20 21 22 23 24 annua lUma e~- season- warmest o annua wettest driest warmest o temp. range ahty ahty .00 coldest range art rt rt coldest -- • -- masl -- •e - - SanDiego La Angelita El Parnaso Penjamo Las Acacias El Matadero S.Quilichao 2.914 -76.457 1955 17.6 S.Quilichao 2.908 -76.480 1564 20.4 S.Quilichao 2.884 -76.482 1543 20.6 S.Quilichao 2.883 -76.510 1470 21.0 Popayan 2.461 -76.550 1925 17.3 Popayan 2.420 -76.559 2060 16.7 La Fortaleza Popayan La Laguna Popayan El Provenir El Tambo La Muyunga El Tambo Loma Larga El Tambo El Porvenir El Tambo Villa Stella El Tambo Asomuripik Totoro Vista Hermosa Totoro San Jose El Agrado Villa Nueva LaCoronita Mira Valle Totoro Totoro Timbio Tirnbio Tambo 2.405 -76.559 2162 16.1 2.434 • 76.558 2087 16.4 2.493 -76.795 1680 18.5 2.423 • 76.766 1730 18.2 2.415 -76.766 1776 17.9 2.420 -76.744 1738 18.2 2.447 -76.746 1703 18.4 2.511 -76.384 2678 12.4 2.514 -76.371 2801 11.8 2.557 -76.432 2363 15.1 2.516 -76.499 2140 16.3 2.447 -76.701 1756 18.1 2.442 -76.694 1800 18.0 2.398 • 76.782 1785 18.4 11.7 11.4 11.4 11.3 11.6 11.4 11.2 11.3 11.1 11.1 11.1 11.1 11.1 10.3 10.0 11.3 11.4 11.1 11.2 11.1 86 87 87 86 86 87 86 86 84 84 85 84 84 91 90 86 85 84 84 85 24.5 25.4 272 25.9. 25.5 22.3 22 22.1 24.1 23.3 23.2 23.3 23.3 18.0 19.9 21.1 22.0 21.8 22.9 24.2 pen period qu er qua er qua er quarter -------------- ·c-------------- 24.6 27.2 27.2 27.8 24.4 23.6 22.8 23.3 25.4 25.1 24.7 25.0 25.3 18.0 17.5 21.8 23.3 25.0 24.9 25.2 11.1 13.5 14.1 13.1 14.2 13 14.7 13.1 11.0 13.4 10.5 13.1 9.9 12.9 10.2 13.1 12.3 13.1 11.9 13.2 11.7 13 11.9 13.1 12.1 13.2 6.8 11.2 6.5 11.0 8.8 13.0 10.0 13.3 11.8 13.2 11.6 13.3 12.2 13.0 26 17.2 20.1 20.2 20.7 17.0 16.4 15.7 16.1 18.2 17.8 17.5 17.8 18.1 12.2 11.6 14.8 16.0 17.8 17.7 18.1 17.6 20.4 20.5 20.9 17.3 16.8 16.1 16.5 18.5 18.2 17.9 18.2 18.4 12.2 11.5 15.1 16.4 18.1 17.9 18.4 17.8 17.2 20.7 20.1 20.8 20.2 21.2 20.7 17.5 17.0 16.9 16.4 16.2 15.7 16.6 16.1 18.7 18.2 18.3 17.8 18.1 17.5 18.3 17.8 18.6 18.1 12.5 12.2 12.0 11.5 15.3 14.8 16.5 16.0 18.3 17.8 18.1 17.7 18.6 18.1 Pl2 Annual Precip. PIS P13 Pl4 Precip Precip Precip Season- of of ality wettest dri~t (Coef period penod Var) ---mm--- 2144 291 2039 263 2028 261 2142 273 2001 285 2210 307 2373 318 2277 309 2267 335 2119 332 2113 335 2127 331 2185 331 1652 226 1550 204 1946 258 2185 284 2184 326 2179 325 2130 341 64 56 58 72 52 68 83 75 67 58 55 58 62 77 82 70 78 61 61 56 42 38 37 36 45 43 40 41 44 48 50 48 45 35 30 39 38 45 45 49 Pl6 Pl7 PIS P19 Precip Precip Precip Preci~ of of of of wettest driest warmest coldes quarter quarter quarter quarte ----mm---- 786 682 676 721 808 875 908 880 897 896 915 894 888 613 552 717 802 883 884 910 223 225 227 260 203 246 285 263 244 210 204 210 224 255 274 236 267 222 221 205 566 624 573 563 212 253 295 272 302 252 240 251 272 381 346 561 278 260 254 252 786 682 676 721 808 875 908 880 897 896 915 894 888 255 274 717 802 883 884 910 . At each si te, 13 plantlets of each genotype were planted in sma ll plots in native soil, and two plantlets were f planted in a mollisol control soil (taken from CIA T-Palmira) in 3kg plastic bags (Figure 2). Additionally, 2 plantlets were planted in a greenhouse at CIAT in 3kg potted native soil from each of the sites as a counter control. Th is methodology is des igned to aid in separating the complex interactions between plant performance, climate and soil. Additional experiments using 13 plantlets for L. alba "t ípica" and J. pectora/is were set up in CIAT-Palmira under three different shade treatments; full sun, 25% shade and 50% shade. 1.5m 1.5m 1.5m '® ® ®® ®® • • • Gpo1Jp1 '"' ~~ 1.3m • • • • • • • • • • • , . • • • ' • • • • • • • • 0.5m SOcm between plants ·~ . ..,.,.. '-r'v ' "' ~· 1.3m • • • • • • • • • • • • • • • • • • • • • • • • ® ® ® ® ® ® 1.5m 1.5m 1.5m Figure 2. Field experiment plan for rapid evaluation trials. The relatively low number of replications is designed to maintain the cost of performing this kind of multi- site trial toa mínimum, ensure that the experiments can be made in small plots of land with local farmers, and to permita greater number of different s ites without requiring large volumes of planting material. The trials were planted in February 2006 over a 5-day period. Farmers were instructed not to apply any ferti lizers, and only to irrigate under extreme dry conditions, carefully noting the dates when water was applied. They were also provided with forms to make observations about plant pests and mortality. Two monitoring visits were made during the experiment to ensure farmers were correctly managing the experiment, and harvest took place in August 2006, precisely 6 months after planting. Total weight of green biomass was taken for the 13 plantlets in native soil, and for the 2 plantlets in contro l so il separately. Green biomass was placed in paper bags, and left in shaded conditions for 3 days to dry. This material was then sent to the Un iversity ofCauca for extraction ofessential oils and vegetative extract. Laboratory analysis of chemical composition Shade dried leaves were used in the University of Cauca to extract essentia l oils (Lippia spp.) and vegetative extract diluted in ethanol (in the case of J. pectora/is). Standard practice was used in a ll cases to provide these extracts. Oil production per unit biomass was measured for all samples. A subset of extracts from Lippia spp. were then selected for further chemical analysis based on the fo llowing criteria: 1) matching the required volume of oil (> 1 g), 2) representing a broad range of 27 environmental conditions (elevation, temperature and rainfall). L. alba essential oils were sent for gas chromatography analysis in the Industrial University of Santander, to identifY volume for 120 principal components. All J. pectoralis samples were sent to the University of Tolima for qualitative analysis of chemical composition for 12 components. Statistical analysis of phenotypic variability A range of statistical techniques were applied to explain variability in biomass. For each genotype, the average biomass per plant was used as the independent variable, with tbe 4 7 edapho-climatic variables u sed as dependent variables. Linear correlation and regression, cluster analysis, principal components analysis and multiple stepwise linear regression were all applied in the search for a high regression coefficient between the independent and dependent variables. Results and discussion Analysis of results is still ongoing, and so only preliminary results can be shown here. Performance of genotypes in terms of biomass production varied greatly between fanns, with complete mortality occurring in sorne tria! sites, whilst otber tria! sites produced almost 1 kilo per plant of green biomass during the experiments (Table 3). None of the genotypes survived at the maximum elevation (Farm 19, 3,400m), though two did survive at Farm 18 at 2800m elevation. Two farms stood out in biomass production: El Parnaso and El Porvenir, both being at mid elevations (1500 - 1700m). However, there was no correlation between biomass production and elevation (Figure 3), signifYing that other factors are more important in generating variability in production. 28 Table 3. Biomass production for each genotype in each experimental tria! site. Biomass Farm # Farm name Municip Lat Long Elev J. pectorales L. alba L lb "f . .. L. origanoides L. origanoides L. origanoides "citrica" · a a tptca "cítrica" "típica" "patia" l San Diego S.Quilichao 2.914 -76.457 1955 12.7 21.1 26.5 16.2 19.5 0.0 2 La Angelita S.Quilichao 2.908 -76.480 1564 1.8 0.0 0.0 0.0 0.0 9.7 " .) El Parnaso S.Quilichao 2.884 -76.482 1543 42.1 102.4 109.5 827.7 215.2 77.5 4 Penjamo S.Quilichao 2.883 -76.51 O 1470 14.6 0.0 84.0 4.4 42.4 101.1 6 Las Acacias Popayan 2.461 -76.550 1925 12.2 117.2 75.6 45.2 45.1 98.3 8 El Matadero Popayan 2.420 -76.559 2060 4.0 19.1 12.4 10.9 0.0 105.4 9 La Fortaleza Popayan 2.405 -76.559 2 162 5.8 87.9 67.4 1.7 2.5 12.6 11 La Laguna Popayan 2.434 -76.558 2087 24.2 174.1 107.4 66.9 120.1 196.1 12 El Provenir El Tambo 2.493 -76.795 1680 0.4 2.3 17.5 76.6 48.7 32.4 13 La Muyunga El Tambo 2.423 -76.766 1730 1.1 7.5 0.5 41.3 16.0 67.0 14 Loma Larga El Tambo 2.415 -76.766 1776 10.5 44.7 11.4 82.5 64.7 0.0 15 El Porvenir El Tambo 2.420 -76.744 1738 30.6 547.3 564.1 509.6 555.8 0.0 16 Villa Stella El Tambo 2.447 -76.746 1703 14.5 119.8 189.5 54.4 0.0 0.0 17 Asomuripik Totoro 2.51 1 -76.384 2678 2.3 0.0 35.2 0.0 0.0 0.0 18 Vista Hermosa Totoro 2.514 -76.371 2801 0.7 0.0 35.8 22.4 0.0 0.0 20 San Jose Totoro 2.557 -76.432 2363 27.5 113.5 2 10.4 0.0 15.6 0.0 21 El Agrado Totoro 2.5 16 -76.499 2 140 1.3 172.3 97.5 0.0 0.0 0.0 22 Villa Nueva Tímbío 2.447 -76.701 1756 1.9 7.3 37.5 131.3 0.0 0.0 23 La Coroníta Tímbío 2.442 -76.694 1800 5.8 156.0 132.4 118.7 0.0 0.0 24 Mira Valle Tambo 2.398 -76.782 1785 37.1 244.2 98.6 104.3 0.0 0.0 29 Justida pectoralis L. alba "Típica" 30 100 o • 'i 25 o • 'i 80 E g¡ 20 e- • • 11. :X 15 • e "' 2 E ~h~ 10 • • • • e 5 • 11. • • • • •• • o e- • oS .t : 60 e • 2 E 40 • S o . ,.ii5 .. .., 20 •• • e • • 11. .. , • • o 1000 1500 2000 2500 3000 3500 1000 1500 2000 2500 3000 3500 Elevaclon (m) Elevaclon (m) L. alba "Citrica" L. origanoides "Patia" 140 120 o • i 120 8 § 100 • .t .. 80 e le • ~ 8 60 • • g¡¡¡ 40 e • .. 20 11. -~ •• o o 'i 100 . e-oS 80 ... . ~ = 60 2 E • • • ... o 40 • u-,.m .., e 20 11. • • o 1000 1500 2000 2500 3000 3500 1000 1500 2000 2500 3000 3500 Elevaclon (m) Elevaclon (m) L. origanoides "Típica" L. origanoides "Cítrica" 180 700 o 160 • 'i 140 ~ § 120 11. : 100 e • 80 ~ g gii5 60 .., 40 e 20 . • • 11. • • ... • o o 'i soo E- 500 es 11. : 400 e • 2 E 300 ... o g iii 200 • .., e 100 11. ~ ... ...~. . o 1000 1500 2000 2500 3000 3500 1000 1500 2000 2500 3000 3500 Elevaclon (m) Elevaclon (m) Figure 3. Biomass production for each genotype across the environmental gradient Correlation analysis of biomass production against each of the 47 edapho-climatic independent variables yields no significant relations (p always > 0.05). However, separating the farms into distinct groups does yield statistically significant results. Through cluster analysis of the biomass production data, Farm 12 (El Porvenir) is highlighted as a clear outlier (Figure 4). When this farm is excluded, multiple stepwise linear regression successfuUy explains 90% of variability in biomass production for L. alba "citrica" through the following equation: Where Biomass = -2406 + 89.5 k+ 128.2 P7 + 0.3 Elevation- l. 7 "NNo3" k = Potassium content of soil; P7 = Annual range in temperature; and NN03 =Nitrate-N 30 o N 5 u 10 m b 15 e r 20 A V V S E L E L L E ~ E L L L L p S V E o S l l a l o l a a l l l a a a a e a l l f o S l n m r S 11 n l e m t l M a p A M A a p L e F J l p u a a Q a r n u g V a a o o A a J a o l r l t L o g y r a r g r r e m o r u l H N e a a V e u a l n u o t a o S S V S ~ e u g d r e + n d l a n 1'). a 9 e t e t l r e o e g 1'). l g o e S a l l l e 1'). e k m V r a l t a o t e a l l r o a o r a a z S l r S S a a finca Fieure 4. Cluster analysis ofbiomass production L. alba "tipica". These preliminary results highlight the importance of soil factors in determining biomass product ion. The control experiments of soil in situ permit the examination of how characteristics of the native soil might constrain growth through a comparison of biomass produced in control soil (but same climate) as opposed to the native soil (Figure 5). 600 S ·o 500 Cl) G) > 400 ;¡ ni e e~ s .! 3oo e Q. ni Q. -o 200 Cl) Cl) ni E 100 o iñ o Biomass production of plants in control soil against native soil - L. alba "cítrica" o Soil presents no limitsto biomass production • • 100 • 200 300 400 Characteristic in soil limiting biomass production • 500 600 Biomass of plants in control soi l (g 1 plant) 700 Figure 5. Analysis of soil constraints to biomass production for L. alba "citrica" 31 In the majority of farms the native soil appears to contain no constraints to biomass production, although three farms do exhibit sorne constraints, especially Farm 16 (Asomuripik) where the control soil produced over five times more biomass than the native soil. Preliminary analysis indicates the possibility of phosphorus being the major constraint, with Farm 16 having the lowest levels of all. Further statistical analyses should confirm this, and identifY similar constraining factors for the other genotypes. Preliminary laboratory results of chemical composition of the derivatives from the plants show extremely high variability between farms for each genotype, indicating that environment has a major role in determining the volume of active ingredients of the plant derivatives. Analysis of these results is ongoing, but will focus on explaining the variability, especially in chemical components, which are particularly sought after in terms oftheir use in the market. Conclusions The results presented are still preliminary and more rigorous statistical analysis is pending. However, sorne conclusions are already becoming clear: l. Biomass production is highly variable, and not correlated with elevation despite this typically being the variable used to describe plant adaptation 2. Variability in biomass production can be explained using more rigorous statistical methods, including multi-variate analyses, and edapho-climatic variables can explain up to 90% of this variability 3. Preliminary results from laboratory analyses also indicate very high phenotypic variability in terms of chemical composition, indicating very complex G x E interactions The implications of these findings are many. Firstly, a new means of defining plant suitability is needed, that moves away from the over-simplification of thermal zones (elevation) as the principie factor. The plants studied here exhibited considerable variability in production both at the species and sub-species level, indicating that producers must choose carefully the crops that they grow, especially when little is known about their adaptation. Farmer experimentation in such cases is recommended, and the protocol developed in this paper could serve as a rapid means of defining plant adaptation. In tbe case of species where product quality is just as or more important than quantity (medicinal and aromatic plants for example), the phenotypic variability in chemical composition of the extractions indicates that a furtber level of complexity exists in choosing the most suitable species for a given location, and the market itself should also consider more carefully the sourcing of plant material. 32 References Busby, J.R. (1991). BIOCLIM - a bioclimate prediction system. In: Margules CR, Austin MP, eds. Nature conservation: cost effective biological surveys and data analysis. CSIRO, Melbourne, Australia. Pp. 4-68 Dos Santos, F., Arimatéia, J. , Gracas, A., Lima, S. and, Reís, F. (2004). Composition and Biological Activity of Essential Oliz from Lippia origanoides H.B.K. Journal of Essential Oil Research 16: 504 - 506. Estrella, E. (1995). Plantas Medicinales Amazónicas: Realidad y Perspectivas. Tratado de Cooperación Amazónica (TCA), Secretada Pro Tempore, Lima, Perú .302 pp. Hijmans, R.J., Cameron, S.E., Parra, J.L. , Jones, PG. and Jarvis, A. (2005). Very bigh resolution interpolated climate surfaces for global land areas. International Jo urna/ Of Climatolog, 25: 1965-1978. Klueger, P.A., Daros, M.R., Silva, R.M., Farias, M.R. and De Lima T.C.M. (1997). Seuropharmacological evaluation of crude and semipurified extracts from Lippia alba Will. N.E. Br. (Verbenaceae). Abstracts. International joint Symposium. Chemistry, Biological and Pharmacological Properties of Medicinal Plants from the Americas. Poster Session, 2:823. Pascual, M.E., Slowing, C.E., Sánchez, D. and Villar, A. (2001) Lippia: Traditional uses, chemistry and pharmacology: A review. Journal of Ethnopharmaco/ogy 76:201- 214. Pérez Trueba, G., Rivero, R., Pardo, Z. and Roddguez, J. (2001). Evaluación de la Actividad Antioxidante de Justicia pectoralis Jacq. Revista Cubana de Investigación Biomédica 20: 30-3x. Stashenko, E.E., Jaramillo, B.E. and Martínez, J.R. (2003).Comparación de la composición química y de la actividad antioxidante in vitro de los metabolitos secundarios volátiles de plantas de la familia Verbenaceae. Revista Académica Colombiana de las Ciencias. Vol XXVII, No.105 Diciembre. Vale, T. G., Matos, F. J. A. and de Lima Viana, G. S. B. (1999). Behavorial effects of essential oil from Lippia alba (Mili.) N. E. Brown chemotypes. Journal ofEthnopharmacology 167:127-133. 33 The effect of clima te change on crop wild relatives Andy Jarvisa,b, Annie Lanec, and Robert J. Hijmansd aBioversity International, Regional Officefor the Americas, c/o CIAT, Cali, Colombia blnternational Centrefor Tropical Agriculture (CIAT), Cali, Colombia. e Bioversity International Headquarters, Ro me, Ita/y. dlnternational Rice Research Institute (IRRI), Los Baños, Philippines Abstract Crop wild relatives are an important source of genetic diversity for crop improvement. However, the survival of sorne of these wild plant species could be tllreatened because of climate change. We used current and projected future climate data for - 2055, and a climate-envelope species-distribution model to predict the impact of climate change on the wild relatives of peanut (Arachis) , potato (Solanum) and cowpea (Vigna). We considered three migrational scenarios for modeling the range shifts (unlimited, Iimited, and no migration). Climate change strongly affected all taxa, with an estimated 16-22% of these species predicted to go extinct and most species losing over 50% of their range size. Moreover, for many species, the suitable areas become highly fragmented. Wild peanuts were the most affected group, with 24 to 31 ( depending on the migration scenario) of 51 species projected to beco me extinct and their distribution area reduced by 85 to 94% over the next 50 years. The number of suitable areas changed by -19% to +4% and patch size decreased by 55 to 60%. For potato, 7 to 13 of 108 species were predicted to go extinct, and their range sizes were reduced by approximately 38 to 69%. The number of patches changed by -34% to 7% and patch size decreased by 20 to 37%. In terms of species extinction, Vigna was the least affected of the three groups, losing 0-2 of 48 species. The mean range size changed by -65% to 8%, with 8-41 of the 48 species losing more than 50% oftheir current geographic range. The number of Vigna patches increased by 12-115%, but the size of those patches sllrank by 51-59%. Our results suggest that there is an urgent need to identizy and conserve effectively crop wild relatives that are at risk from climate change. While increased habitat conservation will be important to conserve most species, those that are predicted to undergo strong range-size reductions should be a priority for collection and inclusion in genebanks. Key words: crop wild relatives, climate change, conservation, cowpea, distribution model, peanut, potato. Introduction Crop wild relatives (CWR) include crop ancestors as well as other more or less closely related species. ln the process of domestication, a crop goes tllrough a genetic bottleneck, ending up with mucb less genetic variation than is available in the wild species from which it was derived. This genetic uniformity can make crops more vulnerable to biotic and abiotic stresses. CWR have been used in formal crop improvement programs for o ver 100 years ( e.g. Mujeeb-Kazi and Kimber 1985, Large 1940), especially for increasing resistance to insect pests and diseases. For example, they have been used to enbance resistance against wheat curl mite (Malik et al. 2003), late blight in potato (Pavek and Corsini 2001 ), and grassy stunt disease in rice (Brar and Khush 1997). Crop wild relatives are being used to improve tolerance of stressful abiotic conditions such as tolerance to drought in wheat (Faroq and Azaro 2001) and have been tested for heat tolerance in rice (Sheehy et al. 2005). They ha ve also been used to raise the nutritional value of sorne crops, such as protein content in durum wheat (Kovacs et al. 1998), calcium content in potatoes (Bamberg and Hannema, 2003), and provitamin A in tomato (Pan et al. 2000). It is expected that the use of CWR in breeding will increase due to recent advances in molecular technologies that increase efficiency and accuracy in transferring desired traits from CWR to crops (Hajjar and Hodgkin in press). 34 Seeds of many CWR have been collected and conserved in genebanks (ex situ conservation). This has greatly facilitated their use, but the world's genebanks are conserving only a fraction of the total genetic variability that exists in CWR and only a small proportion of conserved accessions have been characterized. Moreover, genebank collections are not exposed to natural selection processes that affect natural populations. Conservation of species in situ allows new variation to arise and species to adapt to gradual changes in environmental conditions such as temperature and rainfall patterns. This has been referred to as the conservation of the evolutionary "process" in addition to the current "pattern" of biodiversity (Pressey et al. 2003). Therefore, an in situ conservation approach is needed that complements the ex situ collections to maintain a much larger reservoir of genetic diversity, and to ensure that habitats where CWR occur are protected and wild species continue to evolve in the wild. Setting in situ conservation priorities for habitats and taxa is not a trivial task, especially for CWR species, which are large in number, have a range of different biological, ecological and use characteristics and occur in multiple ecosystems. The conversion of natural vegetation to agriculture is a major threat to the survival of many species in the wild, but climate change is identified as an increasingly strong additional threat (van Vuuren el al. 2006). Climate is one of the major factors governing the distribution of wild plant species, acting directly through physiological constraints on growth and reproduction or indirectly through ecological factors such as competition for resources (Shao and Halpin 1995). The relatively modest climatic changes over the past century have had significant impacts on the distribution, abundance, phenology and physiology of a wide range of species. Many instances have been recorded of species range shifts towards the potes or upward in altitude, and progressively earlier seasonal migrations and breeding (e.g. Walther el al. 2002, Parmesan and Yohe 2003, Root et al. 2003, Parmesan 2006). Global warming has accelerated over the past 30 years (Osborn and Briffa 2005), and is predicted to be in the range of l.l-6.4°C by 2100 (IPCC 2007). Modeling studies (e.g. Thomas et al. 2004) indicate that climate change may lead to large scale extinctions. Given the potential impact of climatic change on global food production (Rosenzweig and Parry 1994, Hijmans 2003, Jones and Thornton 2003), and the demonstrated importance of crop wild relatives in breeding of novel varieties with improved adaptations to biotic and abiotic stresses, it is of paramount importance that crop wild relatives are adequately conserved. Safeguarding and using CWR to broaden the genetic base of modern crops is vital for adapting agricultura! systems to the impacts and consequences of climate change. Yet dueto climate change, these very genetic resources may themselves be under threat of extinction in the wild. Therefore, assessing the potential impact of climate change on CWRs and developing adequate conservation responses is a key activity to sustain agricu\tural production. The objective of this study was to estímate the irnpact of climate change on the distribution of wild relatives ofselected crops and to assess implications for their conservation. We selected the CWR ofpeanut (Arachis hypogea L.), potato (Solanum luberosum L.) and on African wild Vigna spp., which are related to cowpea (Vigna unguiculata (L). Walp.) and Bambara groundnut (Vigna subterranea (L.) Verde.). There are 68 species of wild peanut (Krapovickas and Gregory 1994), which occur in South America (Bolivia, Brazil, Paraguay, Argentina and Uruguay). The 187 species of wild potatoes occur from the southwestern United States through the highlands of Central and South America into Argentina, Chile, and Uruguay (Hijmans and Spooner 2001; Spooner et al. 2004). Only Vigna species occurring in sub-Saharan Africa are included in this analysis. The African taxa comprise 61 species and 63 subspecific taxa (Maxted et al. 2004). These species occur in a wide range of habitats but especially in grasslands, savannas, open woodlands and shrublands and generally at low to mid-low altitudes (Maxted et al. 2004). We used a clirnate-envelope model to quantify the impact of climatic change on the geographic distribution of these three groups, and we assessed the irnpact of likely range shifts on conservation status. Clirnate- envelope models use environmental data for the locations where a species has been found ( or not found) to infer its climatic requirements. These inferred requirements can then be used to classify the suitability of any other location (Guisan and Thuiller 2005). 35 A number of studies have applied climate-envelope-based species-distribution models to the problem of understanding the impacts of climatic change through the use of climatic data for the present and the future (Thomas et al. 2004) and the past (Ruegg el al. 2006). These methods essentially transfer a species adaptation temporally, assum ing on the one hand no more plasticity than currently observed and on the other hand zero evolution, and many overlook the possible consequence of changes in biotic interactions such as competition (Lawler et al. 2006). There is a growing body of research evaluating the suitability of applying species-distribution models to predicting range shifts and assessing extinction risk in the face of climate change (Thuiller el al. 2004, Araújo et al. 2005a, b, Araújo and Rahbek 2006, Hijmans and Graham 2006, Lawler et al. 2006). A central question in the application of species-distribution models to understanding the impacts of climate change re lates to the migrational (dispersa!) capacities of species (Pearson 2006). Species capable of migrating at high rates are more likely to survive, and indeed in sorne cases may gain geographic range thanks to greater land mass in higher latitudes, and species-energy relationships (Menendez et al. 2006). Most modeling studies account for migration by assuming it to be either unlimited or non-existent, yet the reality is likely to be somewhere in-between (Pearson 2006). Thomas et al. (2004) estimate extinction rates of 1103 species in diverse parts of the world under these two migrational scenarios, reporting extinction rates of 21-23% with unlimited migration, and 38-52% with no migration. When the migrational rate is known for a particular species, it is relatively easy to account for this in modeling (Midgley et al. 2006). Despite the uncertainties associated with species-distribution modeling applied to understanding the likely impact of climate change in species survival, the results are important because they can help select and prioritize actions to mitigate negative impacts. ln this paper we use climate-envelope models to assess the potential geographic shifts in distribution of these species. Using three migrational scenarios, we evaluate changes in potential range size, and in fragmentation ofthese climatically suitable areas. Materials and Methods CWR occurrence data Only species for which we had at least 1 O distinct localities of occurrence were included in the analysis, resulting in a study of 21 O individual species in the three groups analyzed. The wild peanut data (Jarvis et al. 2003) consisted of 2175 unique point localities for the 68 species but only 51 species had lOor more unique point localities. The wild Vigna data consisted of 7733 unique point localities for 65 species (Maxted et al. 2004). Just 51 species had 1 O or more point localities, and were included in the species distribution analysis. The wild potato data (Solanum sect. Petota) consisted of 9822 unique point 1ocalities for 187 species (Hijmans and Spooner 2001, Spooner et al. 2004). Of these, 108 species have 1 O or more point localities and were used in the modeling. Species distribution modeling Many different statistical techniques have been used in species-distribution modeling (Segurado and Araújo 2004, Elith et al. 2006). We used the Bioclim model (Busby, 1991) as implemented in DIVA-GIS (www.diva-gis.org). While this model performed relatively poorly in the comparative study of Elith et al. (2006), we use it because it represents a stable modelling approach for application to climate change research (Hijmans and Graham 2006). While it is somewhat biased towards underestimating future species ranges, it does not have occasional erratic behaviour that sorne other models display and it is easy and fast to run, both important considerations when dealing with as many species as we did. The semi-continuous output from the Bioclim model was transformed into presence/absence by assigning presence to the areas where the Bioclim scores were within the 2.5- 97.5 percentile range. 36 We ran this model to predict the current and future geographic distribution of each wild relative species in the genepools under study. The environmentaJ data consisted of climate surfaces for present and projected future conditions. For data on present climate, this study uses WorldClim climate surfaces (Hijmans et al. 2005) for their high spatial resolution ( -1 km) and global extent. Future climate data are available from a number of global climate models (GCMs) with differing greenhouse gas emission scenarios, model characteristics and spatial resolutions. There continues to be heated debate on the uncertainty in GCM predictions and the emission scenarios that drive them and the approach taken in many studies is to use severa) GCMs and emission scenarios (Stainforth et al. 2005). We did not consider variation between GCM projections. We used data from Govindasamy et al. (2003) because it had the highest spatial resolution available. They used the CCM3 model at a 50 km spatial resolution and for a concentration of C02 in the atmosphere of 600 ppm (twice that of pre-industrial conditions). This C02 concentration (including other greenhouse gasses expressed as C02 equivalents in terms of their warming potential) might occur around 2055. In arder to match the 1 km resolution ofthe current clirnate conditions, a downscaling procedure was applied to the CCM3 data by calculating the predicted change in monthly means from the CCM3 model. These change data were then downscaled to 1 km resolution using smoothing (spatial interpolation), and added to the current WorldClirn climate surfaces. For both present and future monthly climate data, the following 'bioclimatic' variables were derived: annual mean temperature, mean diurnal range, isothermality, temperature seasonality, maximum temperature of warmest month, minimum temperature of coldest month, temperature annual range, mean temperature of wettest quarter, mean temperature of driest quarter, mean temperature of warmest quarter, mean temperature of coldest quarter, annual precipitation, precipitation of wettest month, precipitation of driest month, precipitation seasonality, precipitation of wettest quarter, precipitation of driest quarter, precipitation of warmest quarter, precipitation of coldest quarter. Range me tries for quantifying threat We used the estimated current and future species ranges to calculate a number of range metrics that describe the extent and configuration of a species' range, that were subsequently used to calculate extinction risk. The range metrics were designed to capture three scenarios of migrational capacity, ranging from "unlimited" (populations can move to any site where the clirnate is suitable), to "lirnited" (populations can move only a defmed distance), to "no migration" (populations carmot move). We used 300km as the migrationallimit under the limited migration scenario, based on previous studies (Jarvis et al. 2003). All range area metrics were analysed using Arc/Info and using Are Macro Language (AML) scripts for automating the analysis. In addition to range area metrics, we calculated two metrics associated with the configuration of species distributions, with emphasis on fragmentation. For this, we used the Patch Analyst V extension to Arcview 3.2. The metrics calculated are total number of patches and average area of each patch for each ofthe three migrational scenarios. Species richness We mapped species richness under current and future conditions for each ofthe three genepools under the unlimited migration scenario and the no-migration scenario to identify the broad spatial patterns of the impact of climate change. We analyzed broad patterns to examine shifts in species richness across latitudinal and elevational gradients. Results Climatically suitable area For all migration scenarios we considered, the wild peanut species were most affected, followed by wild potatoes, with Vigna projected to be least affected (Table 1 ). Climate change reduced the size of the 37 climatically-suitable area of 80-100% of species, affecting 98-100% of the Arachis species, 80-100% of Solanum species, and 63-100% of Vigna species. We predicted 16-22% ofall species modeled to go extinct dueto complete loss of climatically suitable areas. Arachis was most affected with 24-31 species (47-61% ofspecies) facing extinction; Solanum was projected to lose 7-13 species (7-12% ofspecies) and Vigna 0-2 species (0-8% of species). Table l. Average effect of climate change on potential distribution (climatically suitable area) of species :from three crop wild relative genera. Data oore averaged for each genus and computed for three migratory scenarios: tmlimited, limited, and no migration (see Metbods). Arachis- 51 species Pocato - 108 species Vigna- 51 species Unlirrited Lirrited No Unlimited Limited No Unlimited Limited No migration migration migration rrigration migration migration migration rrigration migration Averagearea Loss (%) 85 89 94 38 52 69 -8 51 65 No. !i>ecies ~th area gains 1 o o 21 9 o 19 1 o No. species ~th area losses 2fJ 51 51 79 92 95 31 50 48 No. species \\ilh area loss = 100% 24 27 31 7 7 13 1 o 3 No. species ~th area loss > 50"/o 46 48 51 41 52 80 8 23 41 No. ~ies ~th area loss > 75% 41 42 47 23 37 48 3 6 13 Under the limited migration scenario, for Arachis species the average size ofthe suitable areas was reduced by 89%, with 48 species losing more than 50%. The amount of area lost varied significantly with the current potential area of distribution, with the highest losses occurring where currently suitab1e area was smallest (~=0.73, P0.05). Vigna species lost an average 51% of climatically suitab1e area and 23 species suffered losses of over 50%. As we found for Solanum, the re1ationship between size of current area and size of loss was not significant (~ = 0.19, P>0.05) for Vigna. The effects of climate change were much larger under the no migration scenario. Obviously no species saw gains in distributional area as species could not move into new areas. Complete loss of suitable areas (i.e. extinction) was projected for 31 Arachis species, 13 Solanum species and three Vigna species. The area projected to be suitable-in the future for Arachis species was 36,486 - 74, 867 krn2 but 1-3 species on top of the 24-31 predicted to lose all area will have suitable areas of less than 500 km2 (including A. batizocoi, A. gracilis andA. kempff-mercadoi). For Solanum, the projected average size ofthe suitable area was about 38,672- 60,000 krn2 but 7-8 species that retained an area of less than 500 krn2• Vigna species had a relatively large future distribution area of about 1.26 - 2.40 million krn2 and apart from the 0-2 species projected to go extinct (V. longifolia and V. keraudrenii), no species had less than 500 krn2 of suitable area in the future. Species with a relatively small suitable area are more vulnerable to loss of genetic diversity and possibly extinction and could be priority species for conservation. For peanuts, species with less than 1000 krn2 projected to be suitable in the future included: A. batizocoi, A. burkartii, A. decora, A glandulifera, A. gracilis, A. kempff-mercadoi, A. kretchmeri, A. matiensis, A. oteroi, A. paraguariensis, A. pseudovillosa, A. retusa, andA. subcoriacea. For Vigna the suitable area for most species remains high, although specific habitat constraints may mean the actual distributional area is considerably less than reported with the climate-suitability models. Only one species faced intermediate threat: V. angivensis with 26,000-61,000 krn2 ofpredicted future area (a loss of71-82% ofcurrent distributional area). For Solanum, species with less than 1,000 krn2 of predicted future area were S. arnezii, S. acroscopicum, S. clarum, S. lignicaule, S. limbaniense, S. piurae and S. raquialatum and S. xsambucinum. 38 It is important to note that not al! species had a reduction in their climatically-suitable area under climate change. No Arachis species gained area, but sorne 19 Vigna species gained area under the unlimited migration scenario with V. praecox gaining as much as 4.1 times its current suitable area, and four other species (V. longifolia, V. nyangensis, V. radiata, V. richardsiea) more than doubling their potential range. Under limited migration only V schlechteri gained area (0.4 times current area), and under no migration tbere were no gains. For Solanum, 21 species gained area assuming unlimited migration, and nine species under the limited migration scenario. However, only tbree species gained more than 1 0% of current area (S. hastiforme, S. maglia and S. xsetulosistylum) and none more than 30%. To aid in the prioritization of species suitable for conservation, a list of priority species based on extinction probability, percentage loss of suitable area and projected future range size is presented in Table 2. Table 2. List of priority species for conservation based on likelihood of extinction, percentage loss of range size and predicted remaining range area, assuming unlimited dispersa!. Peanut Potato Vigna Predicted extinction (no future range area) A . appressipila, A. archeri, A. benensis, A. cryptopotamica, A . douradensis, A . guaranitica, A . hatschbachii, A. helodes, A. hermannii, A. lignosa, A. marginata, A. palustris, A. setinervosa, A. simpsonii, A. stenophyl/a, A . magna, A. tuberosa, A. hoehneii, A . burkartii, A. retusa, A. glandulifera, A. paraguariensis, A. pse udovil/osa, A . decora S. ve/ardei, S. tarnii, S. xmichoacanum, S. xrechei, S. ugentii, S. chancayense, S. incamayoense No species 10 s~ecies with <10,000 km future range area (km2) A. benthamii (946S), A. cardenasii (S 163), A. correntina (3264), A. triseminata (1308), A. matiensis (802), A. batizocoi (7 17), A . oteroi(609), A. subcoriacea(30 1). A. gracilis (232) S. irosinum (S), S. paucissectum (S), S. hoopesii (4 1), S. piurae (87), S. raquialatum (146), S. longiconicum (179), S. arnezii ( 193), S. lignicaule (2SO), S. acroscopicum (422), S. xsambucinum (47S) V. monantha (16), V. virescens (38), V. keroudrenii (11 0), V. phoenix (363), V. mungo ( 1 066), V. richardsiae (28 66), V. bosseri ( 3686), V. hosei ( 4387), V. mude ni a (9S90) 1 O species with greatest % loss ofrange area (% loss) A. gracilis (99%~ A. kretschmeri (99%), A. oteroi (99%), A. matiensis (99%), A . subcoriaceae (98%), A . triseminata (97%), A. kempff- mercadoi (96%), A. major (96%), A. batizocoi (96%), A. correntina (9S%) S. irosinum (99%), S. hoopesii (97%), S. piurae (96%), S . xsambucinum (96%), S. paucissectum (9S%), S. acroscopicum (9S%), S. raquialatum (93%), S.jamesii (91%), S. arnezii (88%). S. trifidum (8S%) V. keroudrenii (98%), V. decipiens (8 S%), V. phoenix (78%), V. procera (64%), V. mungo (63%), V. angivensis (59%), V. antunesii (56%). V. gazensis (55%), V. p/atyloba (SI%), V.juncea (SO%) A number of wild relatives of peanut that have been used in breeding are projected to lose a considerable amount of suitable area: A. diogoi Hoehne (77-97% loss) andA. batizocoi Krapov. (96-9&% loss) has provided resistance to root-knot nematode; A. cardenasii Krapov. (93-99% loss) for resistance to com earworm and southem corn rootworm (Stalker and Lynch 2002); andA. paraguariensis (100% loss) andA. appressipila (1 00% loss ), ha ve been used as a so urce of resistance to early leaf spot (ICRISA T 1995). 39 Amongst Solanum species, S. demissum, which has resistance to late blight (Ross 1986), kept 33-90% of the size of its currently suitable area; S. chacoense and S. berthaultii with resistan ce to Colorado potato beetle (Plaisted et al. 1992) will lose 40-53% and 2-65% of distribution areas respectively; and S. microdontum, which can be used to breed for varieties with increased calcium content (Bamberg and Hanneman 2003) stands to have a change of -28% to 9% in its suitable area. A number of wild Vigna species that contribute to food security are under threat from climate change. For example, the tubers of V. adenantha, (50-68% loss), are eaten (Padulosi and Ng 1990); fruit and seeds of V. junceum are eaten by humans but will lose 50-80% area, and the tubers of V. stenophylla, a species that stands to gain 67% or lose 93% of area depending on the migrational scenario, are eaten by humans (Padulosi and Ng 1990). Species richness The change in the pattems of species richness under the unlimited migration scenario (Figure la) and no migration scenario (Figure 1 b) was variable in space and between the three CWR groups. There is a general pattem of Arachis species moving south-eastwards towards sorne of the cooler climates of the higher elevations of south-east Brazil under the unlimited migration scenario. Species richness at lower latitudes was Jeast affected, and the general trend of species richness across the elevational gradient shifts sorne 200m upwards (Figure 2). lndeed, species richness increased at 800-11 OOm assuming unlimited migration. For Solanum, species richness diminished most in lower elevations as areas suitable for species moved to higher altitudes. Greatest 1osses in species richness were found in mid-northem latitudes (20-25~) and far southem latitudes (20-40°S). Under the unlimited migration scenario, Vigna species moved southwards into South Africa and northwards particu1arly into Ethiopia. Greatest losses occurred in northem latitudes (0-15~), and at low elevations. Under unlimited migration species richness increased at elevations above 1500 m. 40 Figure 1. Modeled potential species richness under current and future climate scenarios and the difference between the two for ea eh of three groups of crop wild relative species. Peanuts (Arachis), Potatoes (Solanum), and Vigna (related to cowpea and bambara groundnut) species. Data shown are for (A) an unlimited migration scenario and (B) a no migration scenario. A Current Species RIChne8$ Cll e: Cl 5 l ' = · -- !7 = · -- 2t B Current Species Rlchness -· - lT e= • -- :1'1 1' 1 , (y'- Future Specie$ Richness = · -- :!7 ., - Fut\Jre Species Rlchness l . -· _ ,. 41 •. ~ · • f J ·' •, ~ ,' Change tn Specle$ Rtehne8$ - ·U = D - ·1' -· -- n '· _ .. =o - · .. , l' Figure 2. Shifts in species richness across latitudinal and elevational gradients for each of three groups of crop wi1d relative species. A, B: Arachis; C, D: Solanum; and E, F: Vigna species. 1.1 1.8 A- Peanut u ~ 1.2 i 1 ~ "' 0.8 i os .. 0.4 0.2 -6 1.2 C- Potato ! 08 ¡ e ii o.s ¡¡ .; l 0.4 .. 0.2 E- V~gna 10 ¡ j "' i . .. 2 - 10 25 20 15 10 - annRr::tww. •• •• ••. fiAw.~ (trlrriteddilperNI) - -• - · Fl.an RlchnM• (rodll.ptna9 ·15 -20 ·25 IAU..clo - o..rent Aic,..... -- .• . . . F\aft FiehnMe (w6r0d dilpetM) --•-· MAt.ftRcMMI(rockpttNt) -5 -10 -15 ·XI ·2!5 --30 IAU..clo Fragmentation of suitable area 1.60 1.40 " 1.20 j ¡ 1.00 e ~ 0.80 "' i 0.60 10 0.40 0.20 0.00 ·""' • . 00 3.50 " 3.00 j -; 2.50 1 ~ 2.00 1 1.50 l w 1.00 0.50 12.00 10.00 ¡ 8.00 i ~ 6.00 i . .00 .. 2.00 0.00 lOO -o.rent~ · · ·· · ·· F1.U•~ ( ........... I*...., - ·• - · Fuble F&:hnM& (ro~ul) 1111) 1300 11111) Bevrion !m tCXI) 19:10 :2aX) 2500 3CQ) JeOO «X11 4500 $XX) 5!5fl) Bav.tlon ... ) F - V~gna 500 tiXXI 1500 :i!XXI - a..rentRictnlas · ··• · • . ,... flcMMe (~ clti*Uf) - ·•- · F\&n~(nocllilperul) S.v.don(m) Current habitat patches of Arachis, Solanum and Vigna became smaller as a result of climate change (Table 3). Again, Arachis species were most affected, with the average patch size decreasing by 55-60%. The total number of patches, however, remained relatively stable, with a 4% gain assuming unlimited migration, and a 19% loss under the no migration scenario. The smaller the habitat patch for these species the greater the loss of area (r2 = 0.42) and the more vulnerable it was to the impacts of climate change. For Solanum, there was a 20-37% decrease in average patch size, accompanied by a +7 to -35% change in total number of patches. There was no relationship between patch size and potential loss of patch area (~ < 0.001). Vigna species gained 12-115% in the number of patches, but had decreases in average patch size of51-59%. 42 Table 3. Predicted effect of climate change on patch size and area for the potential distribution ( climatically suitable areas) of species from three crop wild relative genera. Data are averaged by genus and computed for three migratory scenarios: unlimited, Jimited, and no migration (see Methods). The nurnber ofpatches and average patch size in the future is calculated based only on species predicted to survive. Arachis - 51 s~cies Po tato - 108 s~cies Vigna - 51 species Unlimited Limited No Unlimited Liroited No Unlimited Limited No migration migration migration migration migration migration migration migration migration Curren! average patcb area (km2) 65 39 199 futurc average patch arca (km2) 26 29 27 28 32 25 84 98 81 Curren! number ofpatches 1538 1309 9461 Future nurnber ofEatches 1600 1246 1314 1397 858 891 20403 11143 10568 Potentially most at risk from fragmentation were Solanum species as the patch sizes are in general low, both at present (39 km2) and in the future (25-32km2). The number of patches for many Vigna species increased and for 1-4 species, which had sorne of the smallest mean patch sizes, patches actually increased in size under sorne or all migrational scenarios. These species were V schlechteri (+1% to 43%), Vjuncea (+ 13% to -28%), V davyi (+5 to -40%) and V. procera (+7% to -14%), the latter despite losing 64-72% of distribution area. This species is restricted to southem-central-west Africa where it is uncommon (Maxted et al. 2004). Discussion Uncertainties in predictions There are a number of sources of uncertainty about the projections presented here. One of the principal problems with the use of species distribution modeling is that when they are used to project future conditions, the results cannot be validated (Araújo and Rahbek 2006). Sorne authors have used "hindcasting" to validate modeling approaches, whereby the past is used as a key to the future (Martínez- Meyer et al, 2004; Araújo et al. 2005a). Uncertainty due to differences between modeling algorithms and climate projections can be addressed through multi-model inference (Thuiller et al. 2004) and sensitivity analysis. However we considered these to be beyond the scope of this study given the large amount of species and large geographic areas that we covered. Rather we opted for a preliminary analysis to investigate whether we would find large responses and variation between species and groups of CWR. Future work could focus on sorne of the species identified as being under threat, and do more elaborate modeling for these species. In addition, we believe that more emphasis should be given to experimental work, for example by assessing species adaptation to different climatic conditions, including conditions outside ofthe species' current range (Zavaleta, 2006). Our results might overstate the threat from climate change to sorne species because Bioclim has been shown to underestimate distributionaJ area with climate change (Hijmans and Graham 2006). However, the analysis presented here also fails to account for other impacts on species distribution, like past, present and future habitat alteration and harvesting of species from the wild. This may lead to overestimation of future suitable area as species are alfeady severe ly limited in distributional area and migrational capacity due to fragmentation of babitats. C1imate models should ideally be coup1ed with land-use projection models to understand tbe current pattem of habitat fragmentation and predicted future patterns based on projection of parameters that drive 1and use change such as population and consumption (Hannah et al. 2002). Anotber major source of uncertainty stems from assumptions made about plants' capacity to migrate, tbough it is known that many plant species have a low capacity to migrate (Pearson 2006). We found considerable variability in the results depending on the migrational scenario used, especially for 43 fragmentation measures. Another source of uncertainty is the likelihood of new climates, where climate change brings about climatic conditions not currently experienced. This occurs especially in hot lowland regions, where future temperature increases create climatic conditions with temperatures do not occur. This is the case in sorne regions where wild peanuts are currently found. Finally it is important to note that climate-envelope models do not take into account competition, habitat or other biotic interactions, and the extent to which plasticity is accounted for is aslo debatable (Parmesan 2006). The true range of a species is likely to be reduced by these factors. The model we used here does not take into account these complex interactions, which will also be affected by climate change. The significant differences in extinction rates between the groups studied are of interest. Essentially, for a species to maintain its range size in the face of climatic change the species must migrate at a rate equal or greater to the speed of horizontal displacement of climates, or be able to adapt to keep pace with a changing climate. The horizontal displacement of climate depends partly on the magnitude of the climate cbange in that particular region, and Jandscape characteristics such as topographic heterogeneity. Relatively flat areas are likely to have much faster horizontal displacements in climate than mountainous regions, where a climate could be tracked over short distances by moving uphill. Many Arachis spp. are predicted to suffer higher extinction rates due to their distribution in predominantly flat regions where the horizontal displacement of climate is fastest. Furthermore, the fate of wild Arachis may be further compounded by a migrational capacity that has been reported to be as low as 1m per year (Gregory et al. 1973) due to its geocarpic habit (burying its seed), but no empírica! data are available, and this rate seems very low given the large distribution areas of sorne species. On the other hand, Arachis species are currently found in sorne of the hottest and driest parts of Latin America. Climate change is expected to created "new" climates not currently found on the continent. The degree to which these species are adaptated to these "new" very high temperatures is not known and could be examined through experimental work. Implications for conservation Habitat fragmentation creates spatiaJ barriers to species migration and diminishes colonisation of suitable habitats. Prevention and reduction of habitat fragmentation therefore is critica! to mitigating impacts of climate change (Lavendel 2003). Migration corridors to connect landscape fragments could facilitate range shifts of mobile species. Linking corridors between habitat patches expand the potential range of species and, as indicated by this study for sorne groups, larger areas have greater buffering potential against climate change than smaller areas. Species like Arachis are highly vulnerable to climate change impacts and bave a limited capacity to migrate. Moreover, with little time to adapt their conservation may require human intervention such as translocations and collection for ex situ conservation. Strategically placed protected areas may be key in preserving genetic diversity (Lavendel 2003). Areas selected should be evaluated for tbeir potential as climate change refugues for vulnerable species. Such reserves should contain target populations large enough to ensure persistence as the larger the number of individuals of a species in a given area, the greater the probability that the species will survive (Araújo et al. 2004). Sorne crop wild relatives, including sorne wild potato and Vigna species are weedy or early colonizers and prefer disturbed environments. Thus, these species do not require pristine habitat and persist along roadsides provided that disturbances are not too frequent or intense. Road networks and other tracts of disturbed areas could become corridors for the migration ofthese species (Peters et al. 2005). These analyses can help in prioritizing species for conservation in the wild (in situ). However, in sorne cases ex situ conservation is the only feasible strategy when there is no suitable area for a species in the future. Climate change must be a fundamental consideration in conservation management, and selection of new protected areas and management plans for existing areas need to account for its projected impacts. 44 Economic implications The economic value of crop wild relatives to sustainable agriculture is large. The contributions of CRW to crop yield and qua\ity ofUS-grown or US-imported crops were calculated to be over $340 mi Ilion ayear in the early eighties (Prescott-AIIen and Prescott-AIIen 1983), and this figure is likely to have increased since then. Our study has shown that many crop wild relatives of Arachis, So/anum and Vigna that have already been important for crop breeding programs are under threat from climate change. The importance of CWR genes to peanut breeding is clearly demonstrated by the fact that nearly half of new peanut cultivars and germplasm lines registered in the joumal Crop Science between 2000 and 2005 contained CWR in their pedigrees. Breeding for resistance and tolerance in crops is essential to continue adaptation of farming systems to changed and irregular climate conditions. Conclusions The results presented in this paper provide a strong índication that climate change alone presents a significant threat to important agricultural genetic resources, specifically the wild relative species studied here, with 16-22% of species predicted to have no climatically suitable areas and perhaps go extinct by 2055. When the effects of climate change are combined with alteration of habitats and other anthropogenic impacts, the status of many of these species should be considered as highly threatened, and measures to conserve the genetic resources will need to be expanded. Conservation strategies should be complementary, combining both ex si tu germplasm conservation and in si tu conservation and management of populations. The impacts of climate change have been shown to be heterogeneous in space, and variable depending on the geographic origin, habitat and migrational capacities of the species under question. This is evident in the difference in predicted extinction rates between the three groups of CWR, with 47-61% of Arachis species facing extinction compared to just 2-6% of Vigna species predicted to be facing extinction. Although the data presented here are based on inherently uncertain model predictions, they clearly indicate that further investigation is needed to truly understand the potential impacts of climate change on plant species in the wild. In the case of wild Arachis, important uncertainty can be attached to the results presented here as little is known about the adaptation of the species in "new" climates not currently found in Latin America. Experimental studies could take advantage of ex situ collections and perform common- garden experiments of populations in different climatic environments (representing present and future climates) to gain a mechanistic understanding ofthe physiological basis of climatic adaptations. Acknowledgements This study was partly funded by a UNEP/GEF-supported project on in situ conservation of crop wild relatives in Africa, South Asia, Central Asia and the Caucuses and Central America coordinated by Bioversity Intemational (www.bioversityintemational.org). 45 References Araújo, M. B. and, Rahbek, C., 2006. How Does Climate Change Affect Biodiversity?. Science 313 , 1396-1397. Araújo, M. B., Cabeza, M., Thuiller, W., Hannah, L. and Williams, P. H. (2004). Wou1d clirnate change drive species out of reserves? An assessment of existing reserve-selection methods. Global Change Biology 10: 1618-1626. Araújo, M.B., Pearson, R.G., Thuiller, W. and Erhard, M. (2005a). Va1idation of species-c1imate impact models under climate change. Global Change Biology 11:1504-151 3. Araújo, M.S ., Whittaker, R., Ladle, R. and Markus, E. (2005b). Reducing uncertainty in projections of extinction risk from climate change. Global Ecology and Biogeography 14:529-538. Bamberg J. B. and Hanneman, R. E., 2003. Calcium rich potatoes: lt's in their genes. Agricultura! Research Magazine March 2003. Brar, D. S. and Khush, G. S. (J 997). Alien introgression in rice. Plant Molecular Biology 35:34-57. Busby, J. R., 1991 . BIOCLIM - a bioclimatic analysis and prediction system. Plant Prot. Q. 6, 8-9. Elith, J., Graham, C.H., Anderson, R.P., Dudík, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, R. , Lehmann, A. , Li, J ., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.McC., Peterson, A.T., Phillips, J ., Richardson, K. , Scachetti-Pereira, R., Schapire, E., Soberon, J ., Williams, S., Wisz, M. and Zimmermann, E. (2006). Novel methods improve prediction of species' distributions from occurrence data. Ecography 29: 129-151 . Faroq, S. and Azam, F. (200 1 ) . Co-existence of salt and drought toleran ce in Triticeae. Hereditas 135:205-21 O. Govindasamy, B., DuffY, P.B., Coquard, J. (2003). High-resolution simulations of global climate, part 2: effects of increased greenhouse cases. Climate Dynamics 21:391-404. Gregory, W. C ., Gregory, M.P., Krapovickas, A., Smith, B.W. and Yarbrough, J. (1973). A structure and genetic resources of peanut. In: Peanuts-Culture and Uses. American Peanut Research and Education Association, Stillwater, OK, USA. Pp. 47-133. Guisan, A., Thuiller, W. (2005). Predicting species distributions: offering more than simple habitat models. Ecology Letters 8:993-1003. Ha.üar, R. , Hodgkin, T. (in press). The use ofwild relatives in crop improvement: a survey of developments over the last20 years, Euphytica, online first DOI - 10.1007/s 10681-007-9363-0. Hannah, L., Midgely, G.F. and Millar D. (2002). Climate Change-Integrated Conservation Strategies. Global Ecology and Biogeography 11:485-495. Hijmans, R.J. (2003). The effect of climate change on global potato production. American Journa/ of Potato Research 80:27 1-280. Hijmans, R.J. and Graham, C. H. (2006). The ability of climate envelope models to predict the effect of climate change on species distributions. Global Change Bio/ogy 12:2272-228 1. Hijmans, R.J. and Spooner, D.M. (200 1 ). Geographic distribution of wild potato species. American Journal of Botany 88:2 101-2 112. Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G. and Jarvis, A. (2005). Very high resolution interpo lated climate surfaces for global land areas. lnternat ional Journal of Cl imatology 25: 1965-1978. ICRISAT, (1995). Annual Report, 1994. ICRISAT, Patancheru, India. Jarvis, A. , Ferguson, M.E., Williams, D.E., Guarino, L., Jones, P.O., Stalker, H.T., Valls, J .F. M., Pittman, R.N., Simpson, C.E. and Brame), P. (2003). Biogeography of wild Arachis: Assessing conservation status and setting future priorities. Crop Science 43:1 100- 11 08 . Jones, P.G. and Thomton, P. K. (2003). The potential impacts of climate change on maize production in Africa and Latin America in 2055. Global Environmental Change 13:51-59. Kovacs, M.I.P. , Howes, N.K., Clarke, J.M. and Leiste, D. (1998). Quality characteristics of dururn wheat lines deriving high protein from Triticum dicoccoides (6b) substitution. Journal of Cereal Science 27:47-5 l . Large, E.C. (1940). The Advance ofthe Fungi. Johnathan Cape, London. 488 pp. Lavendel, B. (2003). Ecological restoration in the face of climate change : obstacles and initiatives. Ecological Restoration 21:199-203. Lawler, J ., Whit, D., Nelson, R. and Blaustein, A. R. {2006). Predicting clirnate-induced range shifts: model differences and model reliability. Global Change Biology 12: 1568-1584. Malik, R. , Smith, C.M., Brown-Guedira, G.L., Harvey, T.L. and Gill, B.S. (2003). Assessment of Aegilops tauschii for resistance to biotypes of wheat curl mite (Acari : Eriophyidae). Journal of Economic Entomology 96:1 329-1333. 46 Martínez-Meyer, E., Peterson, A.T. and Hargrove, W.W. (2004). Ecological niches as stable ilistributional constraints on mammal species, with implications for pleistocene extinctions and climate change projections for biodiversity. Global &ology and Biogeography 13:305-3 14. Maxted, N., Mabuza-Dlamini, P., Moss, H., Padulosi, S., Jarvis, A. and Ouarino, L. (2004). Afiican Vigna: an ecogeographic study. IPORl, Rome, ltaly. Menendez, R., Oonzalez, A., llill, J.K., Braschler, B., Willis, S., Collinghan,Y., Fox, R., Roy, D. and Thomas, C.D. (2006). Species richness changes lag behind climate change. Proceedings of The Royal Society 8-Biologica/ Sciences 273(1593): 1465-1470. Midgley, O.F., Hughes, 0.0., Thuiller, W. and Rebelo, A.O. (2006). Migration rate limitations on climate change- induced range shifts in Cape Proteaceae. Diversity and Distributions 12:555-562. Mujeeb-Kazi, A. and Kimber, O. (1985). The production, cytology and practicality of wide hybrids in the Triticeae. Cereal Research Communications 13: 111-24 Padu1osi, S. and Ng, N.Q. (1990). Wi1d Vigna species in Africa: Their collection and potentia1 utilization. In: Ng, N.Q., Monti, L.M. (eds.), Cowpea Genetic Resources. liTA, lbadan. Nigeria. Pp. 58-77. Pan, Q., Liu, Y-S., Budai-Hadrian, 0., Sela, M., Carmel-Ooren, L., Zarnir, D. and Fluhr, R. (2000). Comparative genetics of nucleotide binding site-leucine rich repeat resistance gene homologues in the genomes of two dicotyledons: Tomato and arabidopsis. Genetics 155:309-322. Parrnesan, C. and Yohe, O. (2003). A globally coherent fingerprint of climate change impacts across natural systems. Nature 421:37-42. Parrnesan, C. (2006). Ecological and evolutionary responses to recent climate change. Annual Review of &ology and Systematics 37:637-669. Pavek, J.J. and Corsini, D.L. (2001). Utilization of potato genetic resources for variety development. American Joumal of Patato Research 78:433-441 . Pearson, 0., Wi1fiied, T., Araújo, B., Enrique, M., Lluis, B., Colin, M., Lera, M., Pedro, S., Dawson, P. and Lees, C. (2006). Model-based uncertainty in species range prediction. Journal of Biogeography 33:1704-1711 . Peters, M., Hyman, 0.0. and Jones, P.O. (2005). ldentifying areas for field conservation of forages in Latín American disturbed environments. Ecology and Society 10:1-14. Plaisted, R.L., Tingey, W.M. and Steffens, J.C. (1992). The germplasm release ofNYL235-4, a clone with resistance to the Colorado potato beetle. American Potato Journal 69:843- 846. Prescott-Allen, C. and Presscott-Allen, R. ( 1983). Genesfrom the Wild: Using Wild Genetic Resourcesfor Food and Raw Materials. Intemational lnstitute for Environment and Development, London. Pressey, R.L., Cowling, R.M. and Rouget, M. (2003). Formulating conservation targets for biodiversity pattem and process in the Cape Floristic Region, South Africa. Biologica/ Conserva/ion 112:99-127. Root, T.L., Price, J. T., Hall, K.R., Schneider, S. H., Rosenzweig, C. and Pounds, J.A. (2003). Fingerprints of global warming on wild animals and plants. Nature 421:57-60. Ross, H. (1986). Potato breeding- problems and perspectives. Advances in Plant Breeding (Supp/ements) 13: 1-132 Rosenzweig C. and Parry, M.L. (1994). Potential impact of climate change on world food supply. Nature 367:133- 138. Ruegg, K.C., Hijmans, R.J. and Moritz, C.(2006). Climate change and the origin of migratory pathways in the Swainson's thrush, Catharus ustulatus. Journal of Biogeography 33:1172-1182. Segurado, P. and Araújo, MB. (2004). An evaluation of methods for modeling species distributions. Journal of Biogeography 31: 1555- 1568. Shao, O. and Halpin, P.N., (1995). Climatic controls of eastem North American coastal tree and shrub distributions. Journal of Biogeography 22:1083-1089. Sheehy, J.E., Elmido, A., Centeno, C. and Pablico, P. (2005). Searching for new plants for climate change. Journal of Agricultura/ Meteorology 60:463-468 Spooner, D.M., Van den Berg, R.O., Rodríguez, A., Bamberg, J. , Hijmans, R.J. and Lara-Cabrera, S.I. (2004). Wild potatoes (Solanum section Petota) ofNorth and Central America. Systematic Botany Monographs 68:1-209. Stainforth, D.A., Aína, T., Christensen, C., Collins, M., Faull, N., Frame, D.J., Kettleborough, J.A., Knight, S., Martín, A., Murphy, J.M., Piani, C., Sexton, D., Smith, L.A., Spicer, R.A., Thorpe, A.J. and Allen, M.R (2005). Uncertainty in predictions of the climate response to rising \eve\s of greenhouse gases. Nature 433:403-406. Stalker, H.T. and Lynch, R.E. (2002). Registration of four insect-resistant peanut germplasm lines. Crop Science 42:312-313 . 47 Thomas, C.D., Cameron, A., Green, R.E., Bakkenes, M., Beaumont, L.J., Collingham, Y. C., Erasmus, B.F.N., Ferreira De Siqeira, M., Grainger, A., Hannah, L., Hughes, L., Huntley, B., Van Jaarsveld, A.S., Midgley, G.F., Miles, L., Ortega-Huertas, M.A., Peterson, A.T., Phillips, O.L. and Williams, S.E. (2004). Extinction risk from climate change. Nature 427:145-148. Thuiller, W., Araújo, M.B., Pearson, R.G., Whittaker, R.J., Brotons, L. and Lavorel, S. (2004). Biodiversity conservation: uncertainty in predictions of extinction risk. Nature 430:34. Van Vuuren, O.P., Sala. O.E. and Pereira, H.M. (2006). The future of vascular plant diversity under four global scenarios. Eco/ogy and Soc iety 11 :25. Waltber, G.R., Post, E., Convey, P., Menzel, A., Parmesan, C., Beebee, T.J.C., Jean- Fromentin, M., Hoegh- Guldbergand, O. and Bairlein, F. (2002). Ecological responses to recent climate change. Nature 416:389-395. Zavaleta, E. (2006). Shrub establishment under experimental global changes in a California grassland. Plant Ecology 184:53-63. 48 Another dimension to grazing systems: Soil carbon 1 M.J. Fisher0 S.P. Brazb R.S.M. Dos Santosb S. Urquiagl B.J.R. Alvesband R.M. Bodde/ °Centro Internacional de Agricultura Tropical, Cali, Colombia b Embrapa Agrobiologia, Seropédica, RJ, Brazil Abstract In 1998, Fisher et al. attempted to draw together published and anecdotal information to answer sorne of the questions raised by the fmdings of Fisher et al. ( 1994; 1995), that introduced pastures of African grasses on the eastem plains of Colombia accumulated large amounts of C in the soil. This review synthesises the work in both Colombia and Brazil over the last 7 years that answers sorne of the questions raised and speculations made by Fisher et al. ( 1998). The most important studies ha ve shown that the rate at which litter decays at the soil surface has been grossly underestimated in the past. As a consequence, net aerial primary productivity (NAPP) was shown to be 33.3-33.5 tlhalyr in well managed pastures of introduced grasses without either a legume component or N fertiliser. While data for soil C vary according to the past history and states of the pasture, well managed pastures do accumulate C in the soil to levels above that under the native grassland vegetation. Net primary productivity below ground was only slightly Iess than NAPP. Deficiencies ofN and Pare primarily responsible for the widespread degradation that occurs when introduced pastures are overgrazed and not fertilised. Heavy stocking rates profoundly change the N cycle and lead toN deficiency and hence degradation in the bulk pasture area by concentrating N recycling from faeces and urine in rest areas and watering points. Here the pasture is so damaged by trampling that it cannot take advantage ofthe increased fertility. This paper has been accepted for publication in the jouma1 Tropical Grasslands. Readers interested to receive a reprint as a pdffile should contact the senior author by e-mail m.fisher@cgiar.org. Introduction Atmospheric carbon dioxide (C02) concentrations have increased from 270 ppm befare the industrial revolution to 367 ppm in 1999 (IPCC 2001) and are continuing to in crease at an average rate of about 1.5 ppm/yr. Without action to limit global C02 emissions, the concentration of C02 in the atmosphere could reach 550 - 1000 ppm by 21 OO. C02 is one of the m a in contributors to the so-called "greenhouse effect", which the lntergovemmental Panel on Climate Change (IPCC) in its Third Assessment Report (IPCC 2001) concluded could lead to substantial warming ofthe global climate by 1.8- 5.4oC by 21 OO. In terrestrial systems, plants capture C02 from the atmosphere by photosynthesis. As dead plant material decomposes both within the soil and at the soil surface, plante can be stored or sequestered in soi1 organic matter (SOM). The store of soil organic C (S OC) is a1most 3 times that held in biota ( 1550 compared with 550 Pg) (e.g. Lal el al. 1995). Therefore, small increases in SOC could slow the increase in the concentration of C02 in the atmosphere (Sch1esinger 1995), and in so doing have considerable effects on the global climate system. African grasses introduced into savannas of the central lowlands of tropical South America can increase SOM and accumulate more e in the soil than the native grasses they replace (Fisher et al. 1994; 1995). In a review published in 1998, Fisher el al. explored the possible impact of this finding, and the role of "anagement options that can increase the potential of tropical pastures as a significant e sink to mitigate 1 Accepted for publication by Tropical Grasslands. 49 global warming." They went on to write, "Neither the extent of this fmding in the Latin American savannas nor understanding ofthe mechanisms in the soil-plant system by which it occurs are yet known." Over the 5 years to 2004, a project partly fmanced by the UK Department for lntemational Development has allowed sorne ofthese issues to be addressed. In this paper, we attempt to update Fisher et al.'s (1998) review with recent information, sorne ofwhich is unpublished or in the process ofrevision. Tbe potential of grasslands Approximately one-fifth of the world's land, 3.4 billion ha, is covered by grasslands (Hadley 1993; FAO 1993), with about 1.5 billion ha in the tropics, ofwhich Pearson and !son (1987) considered as muchas 0.7 billion hato be "improvable grasslands". Houghton (1995) estimated the mean C content ofthe soils under tropical grasslands and pastures to be 48 tlha, although it is not clear what depth of soil was used to obtain this estímate. Fisher et al. (1998) measured 1.5- 5 times that amount ofC in the soil toa depth of l m on the eastem plains of Colombia. The data on soil C stocks presented below, also to a depth of 1 m, are 2-3 times Houghton's figure. Vegetation communities Scholes and Hall ( 1995) estimated that tropical savannas, woodlands and grasslands occupy at least 11 .5% of the global land surface. In calculating the e budget for these biomes, net annual e fixation in tropical tree-grass systems is about 7.6 Pg, which is about half of the net annual C fixation attributed to tropical forests, which have closed canopies and no herbaceous !ayer. The main factors that control C fixation are water, nutrient availability and vegetation composition and structure. The total C stock in tropical grasslands, savannas and woodlands is about 135 Pg, 80% ofwhich is in soil (Scholes and Hall 1995). Many of the generalisations about savannas have ignored the South American grasslands (for example, Parton et al. 1995), so it is worth describing them briefly. The area of central Brazil known collectively as ''the Cerrados" covers 205 M ha. (In this paper, we use the terms with their common meanings. "Cerrados" refers to the lands in central Brazil and "cerrado sensu strictu" refers to the low scrubby vegetation that covers a large proportion of the Cerrados.) Fisher et al. ( 1998) asserted that most introduced pastures on the Cerrados were sown on the 24% (Haridasan 1992) that is either treeless or has only a few stunted trees and shrubs, concluding that: "lt is on these lands that most of the 50 M ha (Sano et al. 2000) of introduced pastures are found." This now appears not to be the case, in that most introduced pastures were in fact sown on lands that formerly carried cerrado sensu strictu vegetation. Any calculations of net carbon accumulations on the centrallowlands oftropical South America (Fisher and Thomas 2004) must discount figures for C accumulation by the amount of C that is lost when the cerrado sensu strictu vegetation is cleared. Since cerrado sensu strictu vegetation covers a continuum from communitíes with scant arboreal component to near open forest (Cochrane et al. 1985), it is no trivial exercise to calculate historie C stocks on the lands sown to pasture. While we draw attention to this anomaly, further discussion is outside the scope of this paper. The savarmas of Colombia and Venezuela (about 32 M ha) have not had, from historical times, a significant tree component (Moreno and Moreno 1989). Since this area is vast, any intervention in the savannas of the central lowlands of tropical South America that will increase their net primary productivity, and hence the stock of C in the soil, could be large enough to be of global significance. Net primary productivity above ground Long et al. (1989; 1992) found net primary productivity (NPP), that is accounting for mortality of above- and below-ground organs, at 5 natural grassland sites in the tropics to range from 0.14 to 1 O kg /m2/yr dry matter (DM), as much as 5 times greater than if they used the methodology of the lntemational Biological Programme (Milner and Hughes 1968). Long et al. (1992) indicated that allS sites were potential sites of net C accumulation. In the absence of fires, their terra firma sites in Kenya, Mexico 50 and Thailand accumulated 144 g/m2/yr C, and 40 g/m2/yr C with occasional fires (0.5/yr). They also found a net loss of70 g/m2/yr C with more frequent fires and drought, suggesting that the balance, in terms ofthe sites being a sink or source of C, was delicate. Their studies indicated, however, that the grass-dominated communities have the potentia1 to actas significant sinks for C. Rezende et al. ( 1999) carried out an experiment on lands c\eared from the coastal Atlantic forest in southern Bahía state in central-eastern Brazil to determine the effects of introducing a forage legume (Desmodium ovalifo/ium) in pastures of Brachiaria humidicola at different stocking rates. At monthly intervals over 3 years, amongst other measurements, they collected data on the rates of plant litter fall and decomposition by measuring litter already on the ground and the rate of litter fati over the subsequent 14 days. Doubling the stocking rate from 2 to 4 head/ha caused a highly s ignificant decrease in litter fati, but the legume treatment had little effect (Table 1). Table l. Annual means of existing litter, litter deposited in 14 days and litter decomposition parameters in pastures of B. humidicola in monoculture under three stocking rates of crossbred Brahman cattle in the period January to December 1995 at Ceplac station, ltabela. The values are means of 12 monthly evaluations, from 10 quadrats per paddock, and 3 replicate paddocks per treatment. Stocking Means of existing and Decompos Totallitter deposited deposited litter -ition in 12 months Consump Annual rate NAPP3 Existing Deposited constant Estímate Corrected2 -tion in 14 da~s k 14 da~s Anlha g/m2 g/g/d tlhalyr 2 116.6 72.5 0.0706 18.9 29.7 7.8 35.5 3 105.8 66.4 0.0734 17.3 27.5 9.1 33.3 4 73.7 49.1 0.0797 12.8 21.3 12.4 30.6 Mean 98.7 62.7 0.0746 16.3 26.2 9.8 33.1 1 Calculated from [(litter deposited in 14 days}/14] x 365. 2 Allowing for losses during the 14 days of deposition (See Rezende et al. 1999 for details). 3 Annual NAPP (net aerial primary productivity) = total deposited litter + forage consumption + change in standing biomass. The change in standing biomass during the year was -2.0, -3.3 and -3.1 tlha for the stocking rates of2, 3 and 4 animalslha, respectively. Data from Rezende et al. (1999) and Boddey et al. (2004). Their data showed that 15 - 18 tlha of litter DM were deposited annually. However, the amount of existing litter was always relatively smatl (annual means were 0.8-1.5 tlha DM), suggesting that litter decomposes rapidly, with calculated half life of22-33 days. Moreover, even this calculation must underestimate the true rate at which litter disappears, because a substantial proportion of it must have disappeared within the 14-day collection periods. To reso1ve these issues, Rezende et al. ( 1999) developed an equation to correct for the loss of litter between sampling dates. The correction gave constants for litter decomposition of 0.037 to 0.097 g/gld, equivalent to litter half Ji ves of 9-20 days. They summed these data, together with estimates of animal consumption, to calculate net aerial primary productivity (NAPP) of the pastures at 30 - 36 tlhalyr (Tab1e 1 ). These are astonishing figures, even compared with Long et al. 's ( 1989) estimates, but are consistent wíth Fisher et al. 's (1998) reasoned hypothesis that NAPP of sown Andropogon gayanus pastures on the eastern plains of Colombia was likely to be as much as 43 tlhalyr. lt is also worth poínting out that Long et al. 's data carne 51 from naturalised grasslands, which usually have much lower NAPP than sown pastures of introduced species to which sorne fertiliser has been applied (Fisher et al. 1992). Rezende el al. (1999) also experimented with Jitter bags, and a "covered litter" system, which allowed soil fauna to access the litter. This experiment showed that soil fauna had JjttJe impact on the rate at which litter disappeared. Moreover, these techniques, commonly used to estímate litter decomposition, actually underestimated true rate of litter decomposition at least tenfold, because, in contrast to litter bags, in the open field fresh Jitter is being added continuously. As this material consists of both easily degradable ("active") and recalcitrant fractions, the easily degradable fraction fuels an active rnicrobial biomass that continuously degrades the less decomposable material. They noted in conversation that, when they went to the field in the early morning at ltabela, "the litter felt warm" (R.M. Boddey, unpublished data). In this seminal paper, they concluded that their approach gives more realistic, and much higher, estimates of the NAPP oftropical grasslands and pastores than the techniques used until now. The approach by Rezende el al. (1999) used a mathematical adjustment to take account of presumed degradation Josses during the 14-day measurement interval, which may not be accepted by all. The result could be confirmed, albeit rather laboriously, using the techniques of tissue tumover developed by Hodgson and his coUeagues at the former Hill Farming Research Institute in Scotland (Hodgson 1990). The method depends on determinillg the rate of leaf emergence coupled with taking a census of the number of tiiJers and the number of Jeaves on eacb. In many cases, the number of leaves per tiller is constant (5 or 6 in many Brachiaria species), irnplying that old leaves must die at the same rateas new ones appear. Data of mean mass of newly fallen leaves, multiplied by tiller density per unit area and multiplied by tbe rate of Jeaf appearance, would provide an independent estímate of litter fall (g/m2/d). Net primary productivity below ground Althougb Long et al. (1989; 1992) broke new ground by including roots in their estirnates of NPP, they límited their measurement of roots to the surface 15 cm. At the time, this was thought reasonable, because rnost measurements of plant behaviour were dorninated by studies on sown crops (Fisher et al. 1998). However, the empbasis in the Tropical Pastores Program of the Centro Internacional de Agricultura Tropical (CIA T) for many years was to select tropical grasses with deep and abundant root systems, that can exploit nutrients and water at depth in the soil. In many environments, especially in the semi-arid tropics, deep roots can also confer ecological advantages, for example, adaptation to drougbt. Fisher et al. (1994) speculated that deep-rootedness was at Jeast part ofthe mecbanism by which C was accumulated2 by pastures of introduced grasses in the neo-tropical savannas. However, there are few measurements of the contribution of grass roots at depth to NPP. Trujillo et al. (2005) estimated the net below-ground primary productivity (NBPP) on the eastem plains of Colombia. They assessed the rate of root decomposition and calculated the annual input of soil organic carbon under native savanna vegetation (NS), a degraded pasture of B. humidicola (Bh) and well managed pastores of B. dictyoneura alone (Bd) and in mixture with the legume Arachis pintoi (Bd+Ap). Standing root biomass in Bd (8.6 tlha) was about 3 times that in NS (2.9 tlha), reflecting the low growth rates reported for tbe savanna species (Fisher et al. 1992). NBPPs of the well managed pastures were 30.0 and 31.3 tlhalyr for Bd and Bd+Ap, respectively, compared with only 12.5 tlhalyr for NS. Turnover losses of Bd and Bd+Ap were 1.5 - 2.5 times those in either NS or Bh (Tables 2 and 3). The decornposition constant of roots of Bd was lower than that of either NS or Bd + Ap, which resulted in a longer residence time for Bd roots. The amount of NBPP remaining in the soil after one year of decomposition under well managed pastores was about 2.3 times tbat under NS. 2 Soil carbon is often described as "sequestered". Since we do not k:now the exact forro ofthe newly accumulated C, we prefer to refer to itas "accumulated". Where we quote from earlier work, in wbich the authors used the term "sequester" and its derivatives, we have substituted the alternative of"accumulate" enclosed in square brackets [ ]. 52 Table 2. Mean net below-ground primary productivity calculated from ingrowth tu bes of a native savanna (NS) grassland compared with a degraded Brachiaria humidicola pasture (Bh) and well managed pastures of B. dictyoneura alone (Bd) and in association with the legume Arachis pintoi (Bd+Ap). Data from Trujillo et al. (2005). Parameter NS Bh Bd Bd+Ap t/halyr Sum of short-term tu bes 2.42 e1 5.33 b 9.74 a 9.99 a Last long-term tube 1.22 e 2.89 b 8.03 a 7.91 a Peak in 1ong-term tubes 2.88 b 3.47 b 8.63 a 8.96 a Tumo ver 5.88 e 5.11 e 11.06 a 7.66 b Last long-term tube + 7. 10 e 8.00 e 19.09 a 15.57 b Tumover 1 Within rows, treatment means followed by the same letter did not differ signífieantly (P>0.05). Table 3. Mean net below-ground primary productivity of a native savanna grassland (N S) compared with well managed pastures of Brachiaria dictyoneura alone (Bd) and in association with the legume Arachis pintoi (Bd+Ap), calculated from a compartment-flow model. Data from Trujillo et al. (2005). Parameter NS Bd Bd+Ap t/ha/yr Change in live roots 3. 14 b12 7.28 a 7.92 a Change in dead roots 1.82 b 6.32 a 5.16 a Amount of decomposition 7.50 b 16.41 a 18.26 a Tumo ver 9.32 b 22.72 a 23.42 a BNPP 12.46 b 30.00 a 31.34 a lnputs of SOC 4.99 b 10.30 a 11.58 a 1 Within rows, treatment means followed by the same Jetter did not differ significantly (P>0.05). 2 The figure in Trujillo et al. (2005) is 1.92, whieh is a typographieal error. The eorreet value is 3. 14 (W. Trujillo, personal eommunieation, 29 January, 2007). ln a complementary study, Trujillo et al. (2006) estimated macro-organic matter (> 150 ~m, MOM) in the soil under the same pastures described above. They separated the MOM into light (LF), intennediate (IF) and heavy fractions (HF), using solutions of differing specific gravity, and analysed each fraction for C and N. The C:N ratios ofthe MOM fractions decreased in the order LF > IF > HF in all pasture treatments (data not presented). While the MOM fraction in NS and Bh accounted for a small percentage (8%) of the total SOC pool, the contribution increased to 21 % under Bd+Ap and 30% under Bd. Trujillo et al. (2006) concluded that the MOM represented only a small percentage of the total soil organic carbon pool, beca use it tums over rapidly as indicated by the decrease in C:N ratios as the density fraction increased. They further pointed out that the amount and C:N ratios of the LF were most sensitive to differences in soil depth, vegetation type and pasture management. 53 Relation between root mass and pasture health Embrapa-Agrobiology have measured root mass under pastures in various states from productive to degraded. Degraded pastures in the Cerrados are commonly identi:fied as having a discontinuous grass cover, invasion of shrubby and other weeds and many termite mounds. The data seem to show that root mass increases as fertility declines up to a critica) point, after which there is a cataclysmic decline in root mass as the roots disappear and the area becomes infested with termite mounds. They speculate that the dead roots are feed for the terrnites. 12 Depth interval (cm) - 0-5 10 c:::J5-10 -10-20 ~20-30 ~30-40 - CIJ40-50 ro 8 .!: J 'l;.w>'~/150-60 -C> - 60-80 ::ti:. o - 80-100 o o 6 T""" - "O Q) ·:;. ..... 4 o o (k:: 2 o 1 7 9 20 Age of pasture (yr) Figure l. Root stocks in three B. brizantha pastures of increasing age and a 20-year-old degraded pasture of B. decumbens, Fazenda Barreirao, near Goiania. Data are means of 4 replicates. Values in parentheses above the stacked bars are the total root stocks (kglha DM) to 1 00 cm. Renato, personal communication. In severely degraded pastures, root biomass is often far lower than in productive pastures. In a chronosequence of B. brizantha pastures near Goiania, a 7-year-old pasture had a Iarger root-stock (10.4 t/ha DM to 100 cm) than either a pasture established 1 year before (5.2 t/ha DM) or a heavily utilised pasture established 9 years previously (4.6 t/ha DM, Figure 1, Renato, personal 54 communication). A 20-year-old heavily degraded B. decumbens pasture showed the lowest root-stock of only 2.9 t/ha DM. Oliveira et al. (2004) studied 3 pastures established under well controlled conditions at the Embrapa-Beef Cattle Centre near Campo Grande. The root-stocks to 40 cm were compared in a newly established pasture and two 13-year-old pastures (all B. brizantha), one ofwhich had been well fertilised including additions of N, and one grazed without fertiliser application since its establishment. The results showed that the unfertilised 13-year-old pasture had the highest stock of roots (Table 4). Table 4. Root biomass (t/ha) in soil under Brachiaria brizantha pastures at Campo Grande site. Data are means of 4 replicates. Depth (cm) Treatment 4 months Fertilised Not fertilised 0-10 3.2 8.3 12.2 10-20 0.7 1.9 2.4 20-40 0.7 1.9 3.1 Total in profile 4.6 b1 12.1 e 17.7 a 1 Means followed by the same Jetter are not significantly different (P> 0.05). A further study (R. Schunke, personal communication) at Fazenda3 Ribeirao, Chapadao do Su1 in a large grazing experiment with B. decumbens pastures regularly ferti1ised with P and K (but not N), showed that increasing stocking rate from 0.6 to 1.0 animal units/ha (1 AU = 450 kg live weight) had no significant effect on rooting density (g root DM!kg soil). A further increase to 1.4 AU/ha increased rooting density by approximately 50% (data not presented). These data suggest that, as soi l fertility declines or demand on the pasture increases, initially the grasses increase their root mass, presumably to increase their capacity to capture nutrients from the soil. We hypothesise that finally the photosynthetic apparatus of the plants is unable to support such a large root system and there is a massive shedding of roots, leading to the low values typical of severely degraded pasture. Table 5. Root biomass (t/ha) in soils under a chronosequence of recuperated Brachiaria brizantha pastures of different ages (5 months, 4, 6 and 9 years), Fazenda Barreirao, near Goiania. Data are means of 4 replicates. Depth (cm) Pasture age 5 months 4 years 6 years 9 years 0-10 13.6 13.5 34.5 8.8 10-20 1.7 2.0 2.3 1.2 20-40 2.0 1.6 2.9 1.4 Total in profile 17.3 a 1 17.1 a 39.7 b 11.4 e 1 Means followed by the same letter are not significantly different (P> 0.05). 3 Fazenda is the Brazilian equivalent ofthe US ranch or the Australian station. 55 Oliveira et al. (2004) also monitored root biomass under a range of pastures at the Fazenda Barreirao near Goiania (Table 5). A newly established pasture of B. brizantha showed a total root biomass to 40-cm depth of about 17 t/ha DM, which was very similar to a 4-year-old pasture at the same site. However, a 6-year-old pasture had more than double this root biomass (39.7 t/ha), while a 9-year-old pasture had only 11.4 t/ha. These are on-farm data and not from long-term experiments. Grazing pressure on the different pastures was not monitored nor was it necessarily the same for each pasture. Despite these limitations, the data are broadly consistent. With the ongoing decline ofpasture productivity, there may be important consequences ofthis behaviour of roots for the understanding of C accumulation under grazed pastures. While the application of generous levels of fertiliser to Brachiaria pastures almost certainly results in increased aerial productivity and allows increased stocking rates, root biomass may well be lower than in pastures that receive little or no maintenance fertiliser. This conclusion is certainly surprising, likely to be controversia!, and needs confirrnation from long-term experiments and studies ofroot turnover, i.e., NBPP. A further conclusion would be that, for under-fertilised (or as is most usual, completely unfertilised) pasture, root biomass, and presumably inputs of C derived from roots, will increase with time. Root biomass should reach a peak, followed by a large deposition of dying roots as the system collapses. Under this hypothesis, the maximum rate of C deposition will occur at this time, but subsequently C stocks will gradually decline, and, if the area is not cultivated or fertilised, the C stocks will decline to levels below those originally present under the native vegetation. The role of introduced grasses in C accumulation Fisher et al. (1994) reported that African grasses introduced into the savannas of Colombia could accumulate organic carbon at depth in the soil compared with the native savanna vegetation. They reported that SOC from O to 80 cm depth in a grass-legume and apure grass pasture exceeded that in savanna by 7.04 and 2.59 kglm2 C, respectively. More than 75% of the additional soil C was found below 20 cm, or below the plough !ayer. They concluded that "this e should therefore be less prone to oxidation, and hence loss, during any cropping phase that might be undertaken in integrated crop and pasture systems. Jndeed, such systems should be able to accommodate rotations with annual crops and still contribute to C [accumulation]." Total amounts of SOC to 80 cm depth ranged from 19.7 kglm2 C under the native savanna to 26.7 kg/m2 C under B. humidicola-A . pintoi grass-legume pastures (Fisher et al. 1994; Figure 2). These values of SOC represent the upper range for tropical soils including oxisols, which generally range from 2- 22 kglm2 C (Moraes et al. 1995 and the references therein). Tarré et al. (200 1) reported the use of stable C isotope ( 13C) analysis to determine the so urce of soil C in the 20 years following clearing of the Atlantic forest vegetation at a si te in southern Bahía state of east-central Brazil. Eleven years after clearing in 1977, B. humidicola was established and subsequently regularly fertilised with P and K. After 9 years of pasture, soil (Typic Paleudult) C levels under the original forest and under the sown pasture were similar, but that the source of the C had changed, with 9 t/ha of the original C lost from the 0-30 cm !ayer and replaced with C derived from the sown grass. Stocks of soil C to 1 00 cm ( corrected for compaction caused by grazing by calculating the C stock in a mass of soil equal to that under the forest - Neill et al. 1997) under the sown grass were not significantly different from those under the forest. Where the legume Desmodium ovalifolium had been introduced into the sward, the mean rate of accumulation of soil carbon almost doubled from 0.66 to 1.17 t/ha/yr C. Although these rates are far lower than those reported by Fisher et al. ( 1994) on the Colombian Llanos, the results confirmed the positive effect ofthe legume. However, the 13C abundance data showed that little C derived from B. humidicola was 56 deposited below 40-cm depth. This suggests that, in this biome where rainfall is fairly evenly distributed throughout the year, B. humidicola did not root deeply. This contrasts with the deeper rooting of the same grass species in the Llanos of Colombia (Rao 1998), which has a higher annual rainfall (- 2200 mm) than Bahia (-1400 mm), but also has a strong 3-4 month dry season. 10 20 E 30 ~ E. 40 (1) Cl 50 60 70 LSD = 0.05 e NS a Bh A Bh/Ap 80 +-----~----~----~----~----~ o 1 2 3 4 5 % C (modified Walkley-Biack) Figure 2. Soil organic C distribution by depth in introduced pastures ofthe grass Brachiaria humidícola alone (Bh) and with the legume Arachis pintoi (Ap) compared with native savanna pasture (NS) on a clay loam Oxisol at Carimagua on the eastem plains of Colombia. From Fisher et al. ( 1998). Until recently, very few studies from the Cerrado region of Brazil had evaluated the impact of substituting pastures of Brachiaria spp. for native vegetation. Corazza et al. ( 1999) studied the soil C stock under a B. decumbens pasture established in 1976 on an Oxisol (49- 59% clay) near Brasília. ln 1982, a single crop of soybean was planted with intensive tillage and the pasture was then resown. The stock of soi1 C was found to be 150 t/ha to a depth of IOOcm, 16.6 t/ha C more than that under the cerrado sensu strictu community, although this may be an overestimate, as no correction was made for differences in soil bulk density in the 2 profiles (Neill et al. 1997). Freitas et al. (2000) compared the stocks of C to 40-cm depth under the native vegetation and a productive and a degraded sown pasture on an Oxisol (36 - 46 % clay) near Goiania. They presented data for bulk density, but did not correct the data for equal soil mass between the 2 pastures. Recalculating the data shows that both sown pastures had less soil C tban was originally present under the cerrado sensu strictu vegetation. The correction for compaction by grazing animals shows the importance of correcting for equal mass of soil in the profiles (Table 6). 57 Table 6. earbon concentration, bulk density ande stocks (as presented and corrected for equal mass of soil) in an Oxisol (36- 46% clay) near Goiania under cerrado sensu strictu vegetation compared with productive and degraded pastures. Means of9 replicate samples per depth in each plot. Adapted from Freitas et al. (2000). Depth Carbon Soil bulk Soil carbon stock densi~ {tlhaq (cm) (glkg) (kgldñ() Preseoted 1 Corrected2 NVC Native vegetation ofthe Cerrado3 0-10 22.6 1.19 26.9 26.9 10-20 19.6 1.24 24.3 24.3 20-40 13.7 1.12 30.7 30.7 0-40 81.9 81.9 PP- Productive pasture 0-10 22.0 1.10 24.2 25.6 10-20 16.1 1.32 21.3 19.9 20-40 10.8 1.33 28.7 24.2 0-40 74.2 69.8 DP- Degraded pasture O-lO 19.0 1.10 20.9 22.4 10-20 16.4 1.23 20.2 19.9 20-40 12.2 1.23 30.0 27.3 0-40 71.1 69.6 1 C concentration x soil bulk density x depth. 2 Corrected for same mass of soil to 1 00 cm as under native vegetation (Neill et al. 1997). 3 The tenn used by Freitas et al. 2000; cerrado sensu strictu vegetation in the terminology we use bere. A third, very recent study compared the soil e stocks under an area of cerrado sensu strictu vegetation and 6 areas of different forage species on an Oxisol at the Embrapa-Cerrado Centre near Brasília (da Silva et al. 2004). The area ofnative pasture, which was regularly grazed, contaioed 99.7 tlha e toa depth of 100 cm. The carbon stocks, also to 1 00 cm, under: • Grazed Panicum maximum established 7 years before sampling, • A grazed mixed pasture of Stylosanthes guianensis with B. brizantha established 3 years before sampling, and • Two ungrazed fields of B. brizantha and Paspalum atratum established 4 years before sampling and used for seed production. All sbowed significantly increased soil e stocks, which reached 110- 113 tlha C. No significant increase in e stocks was observed under a B. decumbens pasture, nor under a mixed Jegume-Androp ogon gayanus pasture, both of which were grazed, but which had been established for only 3 years. eonsidering the short period of this study, and that soil C stocks are generally significantly reduced at the time of land clearing and pasture establishment, these data indicate a considerable potential for accumulation of soi1 C under productive pastures. 58 The so urce of the accumu/ated C The C accumulated in the soil must originate in the C fixed by the pasture, that is, it must come from the pasture's NPP. Fisher et al. (1998) pointed out that, unlike the synchronised development of short-cycle annual crops, pastures have a continuous cycle of initiation, growth and death of individual units (tillers in grasses and branches in legumes). As the volume of herbage in apasture increases, the rate of senescence and death ofthe older units also increases until finally they equal the rate of initiation of new units, leading to the so-called maximum yield that is commonly measured by agronomists and others. Unless there is sorne environmental constraint (drought, nutrient deficiencies, pests or diseases) or the grasses flower, which they rarely do in well-managed sown pastures under grazing in the central lowlands of tropical South America, primary growth probably continues at something close to the maximum measurable rate for the whole growing season. Fisher et al. (1998) estimated that yearly NAPP of A. gayanus on the Colombian eastem plains, where rainfall is abundant (- 2200mm annually with at least an 8- month growing season) and temperatures are uniformly high throughout the year, may be 4.3 kg/m2• We emphasise that these conditions are for the central lowlands of South America and may not be typical of other regions with less favourable climates. On the Colombian Llanos, Rao ( 1998) estimated standing root biomass during the growing season under grazing of introduced pure grass and grass-legume pastures at 570 and 380 g/m2, respectively, compared with 140 g/m2 for a native savanna. Root tumover in pure grass pasture was estimated to be twice and in grass-legume pasture 3 times that in native savanna. Based on these data, Fisher et al. ( 1998) concluded that, ''we can safely assume that roots tumover at about the sarne rate as above ground material, and app1y that to the maximum measured yield ofroots [to estímate NPP]". Fisher et al. ( 1998) then attempted to estímate C inputs to the soil of pastures under grazing, assuming that plant dry matter is 40% C. They concluded that total C inputs from shoots and roots in a grazed pasture on the eastern plains of Colombia were likely to be about 2.6-3.2 kg/m2/yr. We can now draw together the experimental data of Rezende et al. (1999), Boddey et al. (2004) and Trujillo et al. (2005), discussed elsewhere in this paper, to obtain revised estima tes of C inputs under grazed pastures. The data for litter fall of B. humidicola plus faeces retum under the lower 2 stocking rates in the Itabela experiment were .1-3.2 kg/m2• NPP ofroots on the eastem plains ofColombia for the B. dictyoneura pasture was 3.0 kg/m2. Summing these and converting them to C, we may safely estímate the total C inputs under reasonably well managed Brachiaria pastures to be about 2.4-2.5 kg/m2/yr, only slightly less than the lower end ofFisher et al.'s (1998) 'back-of-the-envelope' estimates. Moreover, because the present data come from widely distributed experiments, we can be confident that they apply more generally to the central lowlands of tropical South America. Are there differences among tropical grass species? All grasses that show a capacity to accumulate more C in soil than the native grasses are of African origin (Thomas and Grof 1986). Fisher et al. ( 1998) quoted data of sampling from pastures of B. decumbens in the Cerrados near Brasilia that showed little or no C accumulation (J. Duxbury, personal communication; since rendered somewhat moot by the data of da Silva et al. 2004 summarised above). However, they added that, in the Llanos of Colombia at Carimagua Research Station, SOC accumulation to 100 cm depth in 13-year- old pastures of B. decumbens and B. decumbens-Pueraria phaseoloides was 25.6 and 34.1 tlha C, respectively, greater than under native savanna vegetation. Fisher et al. ( 1998) speculated that possible explanations for differences amongst species might be related to differences in the composition of litter, which in turn would affect their rates and pattems of decomposition. For example, the C:N ratios of leaf litter ranged from 88 for B. decumbens to 130, 126 and 117 for A. gayanus, B. dictyoneura and B. humidicola, respectively (Thomas and Asakawa 1993). Data from roots indicate a range of C:N ratios of 159-224 (Thomas, Ayarza and Ce lis, personal communication), 59 and although these differences were not reflected in short-term decomposition constants (Thomas and Asakawa 1993), they may affect longer-term decomposition and conversion to recalcitrant forms of SOC. Fisher et al. (1998) continued, "The lower contribution of B. decumbens-based pastures to C [accumulation] may in part be explained by higher rates of decomposition of shoot and root litter of B. decumbens dueto lower C:N ratios. Therefore the ability to [accumulate] C in soil may be species specific as noted for tree species (Sánchez et al., 1985)." There does not appear to be any recent work that clarifies these speculations. Demands on additiona/ N supp/y Fisher et al. (1998) discussed this issue in sorne detail, but because of the lack of data, much of it was speculation. We have summarised their arguments below. Boddey et al.'s (2004) paper on detailed studies of the N cycle of grazed pastures is broadly relevant to this topic, but they addressed the C cycle only indirectly. Boddey et al. 's (2004) paper, which gives understanding ofthe processes ofpasture degradation, is discussed below in the section on grazing management. Fisher et al. (1998) drew attention to anomalous C:N ratios ofthe soil organic matter under the grasslands of the Colombian eastem plains, which at 21.5 are mucb wider than the 1 0-12 found elsewhere (Schlesinger 1995). They pointed out that litter of introduced grasses in the sown pastures has extraordinarily wide C:N ratios (75-194 for above-ground litter and 158-224 for root litter), and that after only 9 years of a sown pasture on the eastem plains of Colombia, 6 of them with the legume Arachis pintoi, the C:N ratio toa depth of 80 cm had increased to 33.2. They concluded that further work was needed to develop an "understanding ofthe processes ofbreakdown ofplant material ofhigh C:N ratios". Fisher et al. ( 1998) went on to speculate on the possible so urce of N to allow the accumulation of C at depth in the soil. By using known inputs of N from the legume component and associative fixation to be 200 kg/ha/yr N in grass-legume pastures and 40 kg/ha/yr N in pure grass pastures, they reasoned that the C:N ratios of the newly acquired SOM must be around 200, which is close to the values of the root and shoot litter. They concluded that this "suggests that the newly-acquired SOM is particulate plant material. How it gets to depth in the soil is unknown at present, but deep rootedness is a clear possibility". Boddey et al.'s (2004) study on N cycling in grazed pastures, discussed below, is broadly consistent with this conclusion, although they did not address the issue of C:N ratios of SOM. Relationship with soil moisture Fisher et al. (1998) quoted Brown and Lugo's (1982) study oftropical forest soils, in which they reported a positive relationship between the amount of soil C and moisture, which was also associated with different plant communities and soil types. They further concluded that, "Soils in wet climates exhibited greater variations in soil C content with changes in land use, in terms of both loss and recovery, than did soils in dry climates (Lugo and Sánchez, 1986)." There have been no recent studies that provide further evidence on this topic. Management options to increase C accumulation Introduction of forage legumes to improve N supp/y In pure grass pastures on the eastem plains of Colombia, the amount of C accumulation is remarkably constant at about 3 t/ha/yr, and is 2.5- 5 times this rate with a legume component (Fisher et al. 1994). The constant rate of C accumulation by pure grass pastures suggests that the process is rate-limited, and the increased rate with a legume suggests that the limitation is N. Ifthis is so, sowing a legume component in pastures will clearly increase their capacity to accumulate C. If there is a new equilibrium value for the maximum amount of C that can be accumulated in the soil, increasing the rate of accumulation will mean that the equilibrium value will be reached sooner. However, tbe new equilibrium value is unknown, so that the role of increasing the rate is uncertain. There are other options to increase the N supply to a pure-grass 60 pasture, such as application of modest amounts of N fertiliser and seeking means to increase associative N fixation (Fisher et al. 1998). The same arguments wou1d also apply to them, except that there is a C cost in producing N fertiliser. Use offire Much of the native tropical grasslands or savannas (as opposed to sown grasslands) are burned as frequently as annually, and they are rarely fertilised. Therefore, Long et al. (1992) suggested that soil C is at an equilibrium that is less than it would be if pastures were not burned and if sorne fertiliser was applied. Greenland ( 1995) hypothesised that, with these simple management options, the tropical savannas could be an even greater sink for C than is presently forecast. However, it is doubtful whether traditional sedentary or nomadic users of tropical grasslands would change their management practices without sorne strong economic incentive todo so. lt is outside the scope ofthis paper to discuss the topic further. lt is worth noting that introduced pastures are rarely burned, except by accident, at least until they have become seriously degraded and invaded by undesirable woody weeds. Increased activity of soil macro-fauna Lave! le el al. ( 1994) pointed out that pastures usually ha ve greater populations of soil invertebrates than other forms of land use and that these organisms require significant amounts of energy. An active earthworm community may consume the equivalent of 1.2 tlhalyr C (Lavelle 1996), although, given the data that we quote in this review, this is only about 5% of NPP of a well managed Brachiaria pasture. On the other hand, Martin (1991) determined that rates of carbon mineralisation in earthworm casts may be 70% less than that of the bulk soil. In the acid soils of the eastem plains of Colombia, mass of soil macro- fauna in pastures of B. decumbens was 5 times, and in B. decumbens-P. phaseoloides was 1 O times, (up to 60 g/m2, Decaens et al. 1994) that in the native savanna. Earthworms domínate the soi l fauna population and can ingest up to 10 times their body weight in soil each day, leading Fisher et al. (1998) to calculate that, within 3 years, 60 g/m2 of earthworms have the potential to pass the who1e of the soil volume to 1 m depth. Fisher et al. ( 1998) postulated that earthworms must ha ve considerable impact on the processes for moving C from the surface to depth in the soi l, although there are no data to determine whether they are the only vehicle or whether other processes are involved. lt is known that earthworm casts have substantially different properties from the bulk soil, with more water-stable aggregates due to cementing of the soil particles (Guggenberger el al. 1996). Fisher et al. (1998) concluded "We do not know for example if the benefits brought by earthworms in terms of soil improvement (e.g., macro-aggregation, nutrient cycling) and hence increased NPP of the pasture, will outweigh the costs, in terms of carbon, of supporting the activity of earthworms. Available evidence indicates differences depending on the fauna species that domínate the populations (Lavelle et al., 1994)." In this regard, Decaens et al. (1999) showed that the casts of the large aneic (surface-feeding) earthworm Martiodrilis caramaguensis can account for the accumulation of as much as 8.6 tlha/yr C. Grazing management Most studies on grazed lands have focused on aspects of animal production, herbage production and utilisation with little attention being paid to impacts on the soil resource base (Fisher et al. 1998). Fisher et al. ( 1998) concluded that results of evaluations of the effects of grazing on SOC are inconsistent, both increases and decreases being reported with increased grazing pressure, and that many factors are involved in the response of SOM to grazing (soil type, sward type, nutrient status etc.). They further asserted that there is a need to develop an indicator or sorne parameter of system state that reflects the overall result of the different factors involved in determining actual or potential C accumulation. They speculated that the concept of sward "steady-state" as described by Hodgson (1990), which can be used to optimise growth, 61 productivity and senescence in tenns of a simple measurement of sward height in temperate ryegrass pastures, might be used in tropical pasture systems. They thought that such an estímate of sward state could be linked to the concept of Spain et al. ( 1985) in tenns of the management of tropical pastures within a "grazing envelope", which ensures sustainability of production and could optimise net accumulation of SOC. They added that, "This work is in its infancy for tropical pastures and requires much more attention." From the few investigations that have been perfonned, and from a theoretical point of view, it is evident that, if pasture productivity declines, which almost invariably occurs because landowners rarely apply maintenance fertiliser and frequently overgraze their pastures, soil C stocks will be reduced. To this end, Embrapa-Agrobiology are engaged in studying soil C stocks under Brachiaria pastures of contrasting productivity compared with those under neighbouring native vegetation. It is extremely difficult rapidly to assess the productivity or "degree of degradation" of pastures. Traditional soil fertility analysis does not reveal the difference between productive pastures and even those in a fairly advanced state of decline. Lilienfein el al. (2003) made an intensive comparison of soil properties between 3 sets of areas of native vegetation, productive pastures and degraded pastures, all within an area of 100 km2• They sampled the soil solution at 5 depths (5 replicate suction cups per depth) down to 2.0 m at weekly intervals through 2 wet seasons (October- April). All samples were analysed for P, K, Al, Ca, Mg and Na as well as Fe, Mn, Zn, ~ +, N03- , total organic C and total S. There were no significant differences in the concentrations of these various elements/ions between the productive and degraded pasture with the exception of Ca, K and Mg. The authors suggested that the higher grass productivity in the productive pasture was due to the significantly higher concentrations of these 3 nutrients and concluded that maintaining constant productivity "requires the regular application ofCa, K and Mg." This conclusion contrasts strongly with the consensus amongst pasture agronomists working in the Cerrados that N and P are the 2 nutrients that principally limit pasture growth in this region (Carvalho et al. 1990; Zirnmer and Correa 1993; Macedo 1995). In Lilienfein et al.'s (2003) study, P and NH4+ concentrations in the soil solution were below the limits of detection oftheir analytical techniques (0.2 and 0.05 m giL, respectively) as were many analyses of N03- (limit 0.01 mg/L). Furthennore, there was no significant difference between the productive and degraded pastures for the measurable values of N03- in the soil solution. Oliveira et al. (2004), working on a chronosequence of pastures of 2, 4, 7 and 17 years of age, could not rank their state of productivity using traditional analysis of either soil fertility or plant tissue. While soil mineral N was not evaluated, available P (Mehlich-1) in the soi1 under all pastures was at 1.0 or less mglkg. However, these same authors had previously shown at 3 different sites in the Cerrados that, while degraded pastures did not respond to P, K, Sor micronutrient fertilisers and responded weakly toN application, most responded strongly when N was supplemented with P or full fertilisation with NPKS and micronutrients (Oiiveira et al. 2001 ). In summary, these results vindicate the belief that generally P and N deficiency are the principal causes of pasture decline (Boddey el al. 2004), and furthennore the extremely high affmity of the dense mat of roots for P, N03- and NH/ means that, even in productive pastures, the concentrations of these elements/ions in the soil solution are extremely low, as they are absorbed by the roots as soon as they become available from mineralisation of soil organic matter. The lack of simple analytical tests for evaluating the relative productivity of Brachiaria pastures on-fann was a major obstacle in the effort to detennine the effects of pasture productivity on soil C stocks at a farm level. Cattle farmers often cannot provide accurate infonnation on grazing history, as they nonnally rotate their anirnals through many paddocks during a single season. For this reason, Oliveira et al. (2004) investigated dynamic/biological indicators of the state of decline/degradation of pastures in the Cerrados. 62 Their best indicators of pasture heaJth, which should preferabty be used together for more certainty in the assessment, were: l. Rate of regrowth after cutting, 2. Rate of litter fall after clearing alllitter from the soil surface, 3. Microbial biomass C, and 4. Light-fraction ofthe soil organic matter, which is consistent with Trujillo et a/.'s (2006) conclusions. The work now in progress at Embrapa-Agrobiology will examine different chronosequences of productive and degraded pastures on neighbouring areas for their apparent productivity using these indicators, and C and N stocks to a depth of 1 m and to identify the ori~in of the soil carbon ( derived from C3 native vegetation or from the C4 Brachiaria) by evaluating the 1 C natural abundance of the C at each sampling depth. Preliminary results are now available for 2 ofthese sites: a) Fazenda Palota (soil 80% clay), near Luz in Minas Gerais state and, b) Fazenda Ribeirao (soil 18% clay), near Chapadao do Sul in Mato Grosso do Sul state. Both regrowth after cutting to a height of 5 cm and the rate of litter deposition were measured after 28 days (Table 7). The results confirmed the initial classification ofthe productivity ofthe pastures at both sites. Table 7. Forage re-growt.h and litter deposition in putative productive and degraded pastures at two sites in the Brazilian Cerrados. Means of 4 replicate plots per pasture. Si te ,/ Fazenda Palota, Luz, MG Fazenda Riberno, Cha ada:o do Sul, MS Productive astur e 264** 1 235ns 1 •• P0.05. Litter de sited on soil surface Productive Degraded asture 119 37* 27 206 114** 35 At Fazenda Palota, the area had been used for growing coffee since the 1930s and subsequently for extensive grazing, probably of Hyparrhenia rufa and Melinis minutiflora. While these grasses (also African introductions) are much less vigorous than Brachiaria spp., they are also C4 and so have high 13C abundance similar to Brachiaria spp. The productive pasture (B. brizantha) was established in 1995 with tillage and the application of lime and P fertiliser on an area that had been under B. decumbens since the 1970s. The area sampled was only lightly grazed and used as a reserve of forage in the dry season. The degraded pasture (B. decumbens) was established in an area of H. rufa and M minutiflora in 1995, immediately after a crop of limed and fertilised upland rice, but was heavily grazed and at sampling had been heavily invaded by weeds. The soils were sampled in 2003. The C stocks differed markedly between the vigorous and the degraded pasture, being 164.6 and 138.0 t/ha C ( corrected for equal mass of soil to 100 cm depth under the native vegetation, whose C stock was 63 117 .O t/ha). The distribution of the e derived from the native vegetation and from the Brachiaria pastures for the 2 pastures and the native vegetation is shown in Figure 3. o NV 0-5 pp DP NV 5-10 pp DP NV - 10-20 pp E DP o - NV (ij 20-30 pp ~ DP Q) ... NV .E 30-40 pp ~ DP ... a. Q) NV "' 40-50 pp ·c; DP (j) NV 50-60 pp DP NV 60-80 pp DP NV 80-100 pp DP Soil carbon concentration (kg/m3) 5 10 15 20 25 30 - e derived from native vegetation D e derived from Brachiaria spp. 35 Figure 3. earbon derived from e 3 native savanna vegetation and from e 4 Brachiaria spp. in a stand of cerrado sensu strictu vegetation (NV) compared with a productive B. brizantha pasture (PP) anda degraded B. decumbens pasture (DP). The soil has 80% clay and NV has 117.0 tlha e toa depth of 1 m. Data are means of 4 replicates. Fazenda Palota, near Luz in Minas Gerias state. Note that the scale ofthe abscissa is about twice that ofFigure 4. At Fazenda Ribeirao in both pasture areas, the native vegetation was originally cleared at the start of the 1980s and sown to B. decumbens. The productive pasture is on an area that was cropped to soybean in 1991-94, and then sown to B. brizantha, which is well managed. The degraded pasture was an area, which remained in B. decumbens with indifferent management. The e stocks (0-1 00 cm) under these 2 pastures were, respective1y, 62.6 and 53.1 tlha e, the 1atter under the degraded pasture being lower than that under the neighbouring native vegetation (57.1 tlha C). Distribution of the e derived from the native vegetation and the Brachiaria down the profile for the 2 pastures and the native vegetation is shown in Figure 4. 64 o 0-5 5-10 50-60 80-100 Soil carbon concentration (kg/m3) 2 4 6 8 10 12 14 16 18 - e derived from native vegetation D e derived from Brachiaria soo. Figure 4. Carbon derived from C3 native savanna vegetation and from C4 Brachiaria spp. in a stand of cerrado sensu strictu vegetation (NV) compared with a productive B. brizantha pasture (PP) and a degraded B. decumbens pasture (DP). The soil has 18% clay and NV has 57.1 tlha C toa depth of l m. Data are means of 4 replicates. Fazenda R.ibeirao, near Chapadao do Sul in Mato Grosso do Sul state. Note that the scale ofthe abscissa is about halfthat ofFigure 3. Tbree tentative conclusions can be made from these preliminary results: l . Productive, lightly grazed pastures on soils high in clay content can eventually accumulate considerable quantities of soil C, well above the stocks under the cerrado sensu strictu vegetation. 2. Under productive pastures, considerable quantities of C can be stored at depth, as deep as 1 OOcm. 3. When pasture productivity falls due to lack of maintenance fertiliser and inappropriately heavy grazing, C derived from the grass is lost and appears to be almost entirely eliminated from depths greater than 40 cm. Grazing management and pasture degradation Pasture degradation, which is widespread on the Cerrados, has a profound influence on soil C stocks. Fisher and Thomas (2004) attempted to estímate the total contribution of introduced pastures in the tropical lowlands of tropical South America. They included the Amazon, which is not relevant to this review, so comments will be limited to the Cerrados and the Orinoco basin ofColombia and Venezuela. They used the 65 figure of 0.44 M km2, recently updated to 0.50 M km2 (Sano et al. 2000), of the treeless grasslands of the central lowlands that bave undergone substantial conversion from tbe native vegetation to pastures, mainly Brachiaria spp., in the last 30 yr. By using the extensive descriptions of the Iand systems of the central lowlands of Cochrane et al. (1985), they extrapolated data for C accumulation in the soil under introduced pastures on the eastem plains of Colombia (about 3 tlba/yr C) to similar soils and topograpby elsewhere to estímate the probable change in C stocks as a result of conversion to pasture in the grasslands as a whole. They pointed out that losses of above-ground C on conversion of the fonner treeless grasslands are negligible. Fisher and Thomas (2004) addressed the issue of pasture degradation, and by using a simple model of a declining ramp function (Figure 5) to calculate mean rate of C accumulation for a number of scenarios, they calculated a "degradation index" (DI). DI is the time-averaged amount of C accumulated in pastures that degrade and are recuperated compared with well managed pastures that do not degrade, expressed as a percentage. They concluded that, if pastures were recuperated soon after they degraded, they probably accumulated as muchas 50% ofthe Cofa well managed pasture (DI=SO), and that "It requires draconian mismanagement of rapid degradation and long-delayed renovation for the [DI] to fall below 30." 2.0 ...----------------. 108 a::- 0.. ...... =: 106 Cll o .r;- =cm Cll o!!! 1 4 ~-¡;; o -CI> Cl>iU ~ 0§. 102 01e ~% Maximum sustainable productivity Pmax 1- \ \ \ \ \ \ \ \ 1 1 1 1 1 1 1 1 1 \ : 1 1 1 1 1 1 1 1 1 1 1 1 1 Yí ' 1 rc,.cl•\ 1 1 \ 1 1 u Cll o .... ú)CI> Threshold of 1 1 1.o 'CiilQráéláilciñ···,·····¡·················T··-r-· .. ·· \ 1 \ 1 ~ 0.8 \ 1 \ 1 1 1 1 \ 1 1 1 \: \: 'L~~~ 25 ha) and 62 small fanns (0.5- 5.0 ha) in Colombia; and two fanns of about 5 ha in Mexico (Figure 3). The estates in Colombia were located one in the municipality of Concordia (longitude -75.89; latitude 6.03; 1870 masl, Department of Antioquia) and one in the municipality of Piendamo (longitude -76.57, latitude 2.75, 1640 masl; department of Cauca). The small fanns were located in the municipalities of lnza (33 fanns, longitude -75.99 to -76.02, latitude 2.47 to 2.53, 1630 to 1990 masl; department of Cauca). The two Mexican farms were located in the state of Veracruz. One fann was in the community ofEl Encinal (longitude -96.82, latitude 19.21 , 890 masl) in the municipality of Totutla and the other in the community of Auxcuapan (longitude -96.98, Iatitude 19.20, 1490 masl) in the municipality of Tlaltetela. Departments in Colombia represent the sarne administrative level as the states in Mexico but communities in Mexico are one administrative leve) lower than the municipalities in Colombia. To resemble commercial fann operations as closely as possible, the sampling units in all fanns were management units (MUs). MUs are land areas that can be independently managed by the grower during all production stages, including post-harvest processing of batches of beans. Depending on the technical infrastructure, an MU can therefore be a single individual field, a group of fields or a complete small fann . All farms and their MUs were assessed by means of interviews with the growers and general field descriptions. The two estate fanns in Colombia represented intensive, shade-free coffee-production systems. In the Concordia estate slope varíes from O to 15° and the MUs have a wide range of aspects, which vary their exposure to the sun. Annual precipitation is 2300 mm, mean annual temperature is 19.3°C and the soils are Entisol-inceptisols. In the Piendamo estate the slopes range from O to 23°. Annual precipitation is 2200 mm, the annual mean temperature is l9.2°C and the soils are lnceptisols. The small fanns in Colombia varied widely in terms of shade levels, slopes, and aspect. Slopes ranged from 3 to 35° and averaged 20°. Annual precipitation ranges from 1580 to 1760 mm; and annual average temperature is about l8°C. The soils are principally Entisols-inceptisols. The Mexican sites have flat topography. In El Encinal, annual precipitation is 1200 mm with an mean annual temperature of 21 °C. Soils are Cambiosols. Axocuapan has an annual precipitation of 1800 mm and an annual average temperature of l8°C. The soils are Andosols. Selection ofbiophysica/ variables and management practices The different biophysical variables and management practices selected are shown in Table l. The estate fanns in Colombia provided the widest choice of management options. The growers identified five different management units in eacb estate tbat presented northern, western, southern and eastern aspect. In addition, one plateau MU was selected. In the MUs with different aspects, two sites were chosen on the upper and lower parts of the slope to give contrasting levels of soil fertility; growers contend that the upper slopes are less fertile than the lower slopes. 78 Table l. The different biophysical variables and management practices examined at the various case-study sites. Note that not all variables and practices were represented at all sites. Biophysical variables and management practices Aspect1 Soil Varieties S hade Fruit Canopy fertililf (#) thinning leve e Concordia 5 2 None 50% 3 levels Piendamo 5 2 Non e 50% 3 levels lnza small farms VNA56 VNA5 VA4 Non e Whole tree El Encinal Flat 1 4 VA4 Non e Whole tree Axcocuapan Flat 2 VA4 Non e Whole tree 1 Aspect (north, east, south, west and flat; in Concordia northwest instead of nortb). 2 Soil fertility leve! (fertile is lower slope position, infertile is upper slope position). 3 Number ofhorizontal strata harvested. 4 V A = variable analyzed here. 5 VNA = variable not analyzed here. Harvests (#) 2 VNA5 1 1 In each of the nine identified sites different harvesting strategies were implemented. These practices were selected after consultation with the growers and included harvesting fruits separately from different tree canopy Jevels (low, middle, high; in Concordia), fruit thinning (in Concordia and Piendamo) and harvest time (Piendamo ). The first canopy leve) included the upper orthotropic nodes and comprised leafy primary plagiotropic branches with few fruit-bearing nodes. The middle region comprised primary plagiotropic branches with a large majority of heavy fruiting nodes but with few leaves. The lower canopy region comprised plagiotropic branches that had already produced the previous years and bore secondary and tertiary branches that had few fruiting nodes. The fruit thinning consisted in removing 50% of the fruits nine weeks after the main flowering from 25 previously-labeled trees. At this time the fruits have initiated the bean filling stage and have reached about 10% of their fmal size (Arcila-Pulgarin et al. 2002). The harvest in Piendamo was divided into an early harvest on 12 May and in a late harvest on 9 June. Other management practices were implemented by the growers using their usual standards. Managing so many different factors was impossible in the small Colombian farms. Farm owners identified one MU in each small farm for the inclusion in this study. Shade Jevels were defined in each of these management units. The other agronomic management practices were very similar in aH small farms. In Mexico aspect, variety and shade levels were determined for eight management units in El Encinal and four management units in Axocuapan. The other agronomic management practices followed local comrnercial standards but were similar in aH Mexican MUs. Harvest and processing Twelve kg of ripe berries were harvested by hand during the peak of the 2006 harvest. The maturation index of Marín et al. (2003) was used as guidance. In the estate farms, berries from 50 trees for each management practice and each biophysical variable were harvested by estate workers. For comparison, berries also were harvested from 25 control trees for each different bio-physical variable that were not subjected to fruit thinning and harvest at various canopy levels. Samples in the small farms in Colombia were harvested by the farmers from trees in a previously delimitated 30 m x 30 m area within the identified MU. Before processing, damaged, green and infected berries as well as stones, Jeaves and other artefacts were removed. 79 Immediately a.fter harvest, samples from both the estates and the small farms were delivered to a mobile, truck-mounted processing unit. In the unit the berries were de-pulped and the mucilage removed using a J.M. Estrada Model 100 unit. The beans were subsequently fermented in 10 1 buckets using only the water attached to the grains. The samples were then dried using the integrated dryer of the processing unit The dryer consists of a metal closet with four shelves each containing four individual drawers, which are perforated on the bottom. The dryer thus has the capacity to process 24 samples of 1-1.5 kg at the same time. A ir, heated to 45°C by a gas bumer, is blown into the bottom of the closet and ascends through the closet drying the beans. The most recent samples are placed in the top drawers and moved down to the next lower level when new samples are added, thereby emulating the process of industrial dryers. Sarnples are dried until the parchment beans reach a humidity of l 0% to 12 %, which occurs nom1ally after 14 to 16 hours. The sarnples were then placed in sealable plastic bags and stored at 18° C until the cupping process. Samples from Mexico were harvested during the peak of the 2005/06 harvest. The samples were processed the same day according to the wet local method which included de-pulping, ferrnentation, washing, and drying in a standardized manual manner. The slightly different procedures used in Mexico and Colombia did not present a problem in the data analyses because there is no direct comparison of Mexican and Colombian san1ples. There is also no direct comparison ofthe results from the assessments of samples from the Colombian estates and small farms. Physica/ assessment ad beverage quality eva/uation The parchment beans were milled and the percentage and weight of bean and husks determined. The density of the beans was calculated a.fter humidity was measured. Thereafter beans with primary and secondary defects were quantified, and their weight and percentage recorded. Any beans with defects were then removed by hand. The defect-free beans were sieved and the bean size distribution deterrnined using standard sieves from 14/64 inch to 18/64 inch. Physical assessment data were recorded but not used in the analyses presented here. The objective of this study was to understand the impact of selected biophysical field variables and management practices on the beverage quality. All the Colombian samples 250 g ofbeans were roasted in a laboratory roaster the day before the beverage assessment. All samples were roasted for about 11 minutes with an initial temperature of 200 oc to a standard reddish-yellow color. Exact roasting time was recorded. Roasted beans were ground to the recommended intermediate particle size immediately before the beverage quality assessment using a precision grinder. Sensory beverage quality assessment was done by cupping of the coffee liquid prepared for each sample: water ( 150 mi at 97 °C) was poured on 1 O g of ground coffee in each of five cups. This produces coffee with a range of 1.1% to 1.3% soluble solids. The five cups were treated as replicates for the sensory beverage quality assessment. The sensory attributes evaluated were fragrance, aroma, acidity, aftertaste, body, flavour, sweetness, preference and final score. Fragrance is the sensation of gases released from ground coffee. Aroma is the sensation of gases released from brewed coffee. Fragrance and aroma were assigned one value. Acidity is a measure of the intensity of acidic sensation. Aftertaste is the taste that remains in the mouth after having tasted the brewed coffee. Body is the oral feeling ofviscosity. Flavour is the taste perception ofthe coffee beverage on the tongue. Sweetness is the detection of soluble sugars on the tongue tip. Preference represents the overall impression of the coffee by the cupper. Final Score is the sum of the attributes evaluated plus three times their average. The attributes were rated on a scale of 1 to 1 O with 0.5 point increments, using the cupping protocol of the Specialty Coffee Association of America (SCAA, Lingle 2001 ). A score of one implies many defects, two sorne defects, three implies a very deficient coffee, four a deficient beverage, five is a standard coffee, six is a good coffee, seven very good, eight refers to an excellent coffee, and nine to exceptional quality. 80 The estate samples were cupped by one cupper of high intemational reputation. The samples from the small farms were assessed by a national panel of severa! cuppers of which on ly the results of the most consistent cupper were included in the analyses. Cupper consistency was assessed using statistical discriminant function analyses (Hair 1992). The national panel assigned on average lower values than the intemational cupper so that the results could not be analyzed jointly with the estate samples. The Mexico samples were assessed by a panel of seven cuppers in the cupping laboratory of Café-Veracruz, A.C also according to SCAA standards. As suggested by the official Mexican norm only the attributes fragrance, aroma, aftertaste, acidity and body were assessed. Mexican cuppers used a scale that ranges from 0-15, 0-5 being low quality, 5-9 medium quality and >9 high quality. The different scales used did not presenta problem in the statistical analyses because there is no direct comparison ofMexican and Colombian samples. Acquisition of environmental information and statistical analyses Geographic location was detennined using a Trimble Pro-XR global positioning system (GPS) with Omni- ST AR real-time correction. Aspect in degrees (0 ) was measured with a compass. Hemispherical imagery to describe shade levels was taken with a NIKON Cooi-Pix E4500v 1.3 digital camera using a fish-eye lens with a field ofview of 180°. The imagery was then processed using Win-SCANOPY (Regent lnstruments 2005) software to derive illumination parameters. First the pixels of the imagery were classified in canopy and sky, the output ofthis process is a black and white image. The second step is the analyses ofthe canopy which comprises the analyses of the canopy structure and the radiation analyses. The canopy structure variable derived for the present analyses was the gap fraction. Gap fraction is the number of pixels classified as sky region divided by total number of pixels. The shade percentage is simply the numerical complement ( 100%- gap fraction). In the radiation analyses the average direct and diffuse photosynthetically active radiation (PAR) over (PPFDO) and under (PPFDU) the shade tree canopy were estimated in MJ m·2 d"1. PAR radiation is assumed to be constant over time and is a percentage of solar energy flux. This value is 0.5 1 J/m2s. Atmospheric attenuation is inversely proportional to atmospheric transmittance at zeoith and relative path length to the zenith. As cornmon in meteorological studies incident radiation was cosine corrected to account for the radiation angle of incidence with respect to the receiving surface. The preceding fonnulae are used by WinSCANOPY to compute direct radiation above the canopy for the selected sun track, which is created automatically by WinSCANOPY for a specified period of time. These calculations are a function of latitude and longitude, and the defined growing season and time zone. PAR uoder the canopy was calculated the following manner: For all sun positions on a suo track the instantaneous radiation value below canopy varíes between zero and the direct radiation value above canopy in function of the pixels value in the image under the sun tracks at the moment. Wheo the pixel is classified as sky it is assumed that all radiation is intercepted so the value below is zero. When the pixel is classified as sky, it is assumed that all radiations above canopy passes unimpeded so it is equal to the radiation leve! above canopy. These parameters include the average direct and diffuse photosyntetically active flux density over (PPFDO) and under (PPFDU) the shade tree canopy measured in MJ m·2 d"1• All field measurements, product quality assessment data and other information related to management practices were entered into Cinfo (Oberthür 2006). CINFO is an interactive, online spatial data- management system for supply chains of higher-value agricultura! products. Data were exported for statistical analyses in spreadsheet format and summary statistics were computed. These, the ANOV A analyses and the Duncan tests were done using the S+ package (Insightful 2001 ). Information of sensorial beverage quality are on a quasi-interval scale (l to 10 with increments of 0.5 giving 20 available points), which is analogous to a Likert scale. These data are usually analyzed using interval procedures in parametric statistical methods (e.g. Decazy et al. 2003; Avelino et al. 2005; Vaast et al. 2006). In a review ofthe literature, Jaccard and Wan (1996) concluded, "for many statistical tests, rather 81 severe departures (from intervalness) do not seem to affect Type I and Type li errors dramatically." They suggest, provided the scale has at least five, and preferably seven categories, the assumption of normal distribution can be assumed to be valid. Results Summary for al/ si tes Table 2 summarizes the results of the coffee beverage sensory analyses. Table 2. Descriptive statistics for all sites including the two Colombian estates (Concordia, Piendamo), the small farms oflnza in Colombia and the two Mexican farms (El Encinal, Axocuapan). Samples for all biophysical variables and management practices are included in the analyses. Mínimum ~ 1st quartile ~ Mean ~ Median "' -g 3rd Quartile J: Maximum Std devn. Mínimum )e 1st quartile 11 e Mean "' 1! Median o g 3rd Quartile 8 Maximum Std devn. Minimum 1st quartile M íí' Mean ; Median ..5 3rd Quartile Maximum Std devn. Minimum ~ 1st quartile e Mean a Median .S (.) S 3rd Quartile iij Maximum Std devn. Minimum ~ 1st quartile !!: la Mean g. Median a 3rd Quartile ~ Maximum Std devn. Aroma fragrance 4.00 7.00 7.53 7.75 8.00 9.00 0.91 5.00 7.25 7.64 7.75 8.00 9.00 0.72 2.00 5.00 6.05 6.00 7.00 8.00 1.43 8.50 9.20 9.73 9.70 10.20 11.40 0.58 8.20 9.10 9.52 9.50 9.90 11.10 0.58 1 na= not available Acidity 5.00 7.00 7.60 7.75 8.00 9.25 0.81 6.25 7.50 7.82 8.00 8.25 9.25 0.64 3.00 5.00 5.78 6.00 6.00 8.00 1.19 6.10 7.50 8.27 8.30 8.80 10.40 0.89 6.70 8.00 8.81 8.80 9.42 11.40 1.09 After- taste 3.75 6.50 7.04 7.00 8.00 9.50 1.24 5.00 7.00 7.43 7.25 8.25 10.00 0.98 4.00 5.00 5.77 5.00 7.00 8.00 1.11 na 1 na na na na na na na na na na na na na Body 7.00 7.50 7.90 8.00 8.00 9.00 0.46 6.00 7.75 7.99 8.00 8.50 9.00 0.58 3.00 5.00 5.56 5.50 6.00 9.00 1.29 4.80 5.80 6.00 6.05 6.30 7.20 0.42 4.70 5.87 6.07 6.10 6.50 7.00 0.50 82 Flavor 4.00 6.75 7.33 7.50 8.25 9.25 1.13 6.00 7.00 7.85 8.00 8.50 10.00 0.86 4.00 5.00 5.n 5.00 7.00 8.00 1.05 na na na na na na na na na na na na na na Sweet- ness 6.00 7.25 7.83 8.00 8.25 10.00 0.71 5.00 7.50 8.13 8.00 8.75 10.00 0.92 2.00 2.50 2.83 3.00 3.00 5.00 0.70 na na na na na na na na na na na na na na Prefer- ence 4.00 6.75 7.08 7.25 8.00 9.15 1.43 6.00 7.00 7.83 8.00 8.00 10.00 0.94 3.00 5.00 5.37 5.50 6.00 8.00 0.98 na na na na na na na na na na na na na na Final seo re 59.8 74.4 78.1 80.0 84.3 91.5 7.56 63.5 79.0 82.3 81.9 85.1 92.0 5.03 39.0 49.0 54.6 54.0 58.0 77.0 8.58 na na na na na na na na na na na na na na The two Colombian estates have average values between seven and eight for the sensory characteristics. Concordia tends to have higher values and also results that are less variable as indicated by the smaller ranges and lower standard deviation of the data. Concordia reaches an average final score of more than 80 points, which is remarkable. The highest final scores for the both estates were more than 90 points. The Inza fanns had relatively low values between three and six for the sensory characteristics which was expected due to the different quality preferences of the panel. The results indicate highly variable product quality coming from the 33 farms. The results from the two Mexican farms indicate that the quality of the coffee is very similar in both farms, aJthough results from Axocuapan tend to be slightly more variable. Biophysical variables On the Concordia estate the best quality coffee comes from south-facing slopes with a final score of 83.9. Berries harvested from the plateau also achieve very good results with a final score of83.4. The east-facing slopes have generally the lowest values albeit with a still acceptable final score of 80.8 (Table 3). The situation presents itself very different in the Piendamo estate where east-facing slopes score second best after the plateau site. South facing slopes perform badly compared to aJI other aspects and achieve only a final score of 72.8. This represents an astounding difference of almost eight points between the best and the least performing site. Table 3. Effects of aspect and position in the slope on coffee beverage quality on samples from the Concordia and Piendamo estates based on one-way ANOV A and t-test. Data for the same attribute followed by the same letter are not significantly different according to Duncan 's multiple range test (P 2. As stated previously, by running a general analysis, we predict areas that produce high-quality coffee based on evidence data from distinct environmental conditions and insights of interactions with coffee quality are on1y of a general nature. When analyzing niche by niche, a more detailed set of responsible factors can be obtained. Tab1e l. P values ofthe likelihood ratio chi-square for the entire area and for the two niches. Ca u ca El Tambo-Timbio Inzá 25 1 7S 0.43 0.062 0.86 98 501 so O.OS6 O.OS1 0.081 7S 1 2S 0.13 0.019 0.014 Table 2. Quality enhancing factors impacting on the final score attribute ofthe niches lnza, El Tambo- Timbio and the whole Cauca sampling area. The significance indicator e is shown in parentheses. Quality enhancing factors Entire data set lnzá El Tambo-Timbio Altitude (masl) Average annual dew point (0 C) Average annual temperature (0 C) 1750- 1800 (2.02) 1652 - 1725 (2.32) 1725 - 1798 (2.39) 11.9 - 12.2 (2.43) 12.3 - 12.6 (2.07) 17.7 - 18.1 (2.55) 18.0 - 18.4 (2.21) 12.3 - 12.8 (2.38) 17.8 - 18.9 (2.32) Average annual precipitation (mm) 1645 - 1674 (2.2) 1760 - 1934(2.31) 1587 - 1616(2.1) Table 3. Qua1ity reducing factors impacting on the final score attribute ofthe niches Inza, El Tambo- Timbio and the whole Cauca sampling area. The significance indicator e is shown in parentheses. Quality reducing factors Altitude (masl) Slope (degrees) Average annual dew point (0 C) Average annual temperature (0 C) Average annual solar radiation (Mj/m2/d) Average annual precipitation (mm) Dry months (mth/yr) En tire data set lnzá El Tambo-Timbio 1528 - 1623 (2.74) 34.5 - 40.9 (2.55) 22.4-25.6 (2.54) 21.6-27.9 (2.10) 12.8 - 13.5 (2.4) 11.5 - 11.9 (2.57) 14.3 - 14.8 (2.00) 1133 - 1587(2.78) 17.3 - 17.7 (2.47) 20.0-21.0 (2.02) 21 .8- 22.3 (2.32) 3 (2.81) For both niches, altitude, average annual temperature and average annual dew point enhance final score quality. The ranges are only slightly different between the two Cauca sites, loza having lower temperatures and higher altitudes than El Tambo-Timbio. Average annual precipitation is an important enhancing factor in loza and for the entire Cauca data set. In contrast, greater slope influences final score negatively in both niches. Dew point above and below the range identified as enhancing quality have a negative impact as does average annual temperature. The optimal annual average temperature in loza is 17.7- 18.4°C but is slightly higher (17 .8 - 18.9°C) for El Tambo-Timbio. The results demonstrate variability in the environmental factors that impact on final score and the need to assess these factors according to their niches. Variability in space Recent studies show the interactions of environmental factors and coffee quality and the correlations between quality attributes for selected study sites: Vaast et al. (2005a) reported no differences in the caffeine content of high and low quality coffees while Avelino et al. (2005) did not find any strong correlation between sensorial characteristics and caffeine, trigonelline, fat, sucrose, cholorenic acids. Vaast et al. (2005a, b) showed that there is a strong relationship between high trigonelline content of coffee beans and higher bitterness and lower acidity of the coffee beverage; Decazy et al. (2003) found a pos itive relationship between bean sucrose content and coffee acidity and quality and also that high bean fat content related to good acidity and beverage preference. Little work has been done on the spatial variability of coffee attributes and their interactions with the environment. Our analysis of Lara-Estrada's (2005) study 99 puts data on coffee quality in a spatial perspective. The correlation of the " score" response maps, for ten different quality attributes, demonstrated the variability in correlation between the responses (Table 4). Table 4. Correlation coefficients of response variable pairs (Pref. = preference, Caff = caffeine, C.A. = cholorenic acids, F.C.= fat content, Suc. = sucrose, Trigo= trigonelline ). Acidit~ Aroma Bitter Bod~ Flavor Pref. Caff C.A. F.C. Suc. Aroma 0.72 Bitter -0.22 -0.08 Body 0.69 0.76 0.17 Flavor 0.82 0.64 0.06 0 .74 Pref. 0.82 0.74 0.00 0 .82 0.90 Caff -0.13 -0.03 0.16 0.07 0.07 0.02 C.A. 0.08 0.08 0.26 0.25 0.12 0.17 0.50 F.C. -0.24 -0.21 -0.17 -0 .24 -0.24 -0.22 0.28 -0.04 Suc. 0.33 0.44 0.03 0.42 0.28 0.35 -0.10 0.35 -0.23 Tri~o 0.18 -0.04 -0.23 -0.06 0.09 0.03 -0.37 -0.57 -0.31 -{).27 A single figure averaging the correlation of a pair of variables is often not meaningful. For example, sugar content and flavor are poorly correlated (r = 0.28); only when visualizing the r coefficient on a map, does the importance ofthe spatial variability become evident (Figure 3) and highly correlated areas (with values as muchas r = 1 or r = -1) identified (Figure 3). The spatia l correlation with window sizes of 3, 5, 9 and 17 grid cells, translates into 9, 16, 65 and 94 ha for each analyzed window. These correspond to individual farms up to 9 ha in s ize, groups of farms up to 16 ha, associations up to 65 ha and micro-catchments up to 294 ha. The different resolutions give insight on the scale where correlation pattems emerge, which is valuable information for profile identifications of coffee quality and their marketing. The analysis also demonstrates to farmers' associations the strengths and weaknesses of their coffee qualities. Not only do quality responses vary in space, but environmental factors also impact on quality. A GWR analysis on the overall importance of environmental factors that impact on flavor result in flavor being s ignificantly dependent on "number of average annual dry months" (DM) and "average annual diumal temperature range" (DTR) at P=O.OS and P=O.Ol respectively. 100 N Moving Window for Correlation between Sugat Content and Flavor- Window Size 3, 5, 9, 17 N + + 3 9 o 50 100 200 --==--==-------141ometers Correlation Coefficient (r) - -1 · ~j - .0.7 - .o 4 -.04 · ~.1 [:=J.0.1- 0.1 -0.1-04 - 0 4 -07 - 0.7 · 1 Figure 3. The variability ofthe correlation between sugar content and favor at different resolutions ofmoving window. 101 Variability of Environmental Factor lmpact on Flavor DryMonltl 0 Hogh 0 M- O t- Figure 4. Variability ofthe impact on navor for two decisive environmental factors (dry months and diurna! temperature range). Bigger dots are representing a larger impact on Oavor than smaller dots. Even though these two environmental factors are significant for flavor; their contribution at each site is distinct. Figure 4 shows the variable impact of DM and DTR on the flavor quality. Bigger dots representa larger impact on flavor than smaller dots. The impact of these factors is very heterogeneous in space but there are clusters that form niches where either DM or DTR are more important. Conclusions CaNaSTA predicts niches likely to produce high final scores for coffee quality at confidence levels of P = 0.056- 0.1 for an area of 800,000 ha in Cauca using on ly quality data from 88 sites. Within pre-defined ni ches, quality classes can be predicted at P = 0.019 - 0.051 for the El Tambo-Timbio ni che ( 160,000 ha) and at P = 0.014 - 0.081 for the Inza ni che ( 16,005 ha). The ranges of the factor-enhancing or -reducing quaJity in the two niches of Cauca are very different depending on the environmental envelopes predominating in the niche. The importance and utility of SDS tools and geographical analyses for assessing the variability of environmental factors and causal quality responses is shown in a case study in Nicaragua. These are very powerful tools that allow the extrapolation of point information to give information about a broader surface. Environmental factors and their impact on quality are very heterogeneous in space. Nevertheless geographical analyses a llow the identification of niches with similar factor combination. 102 References Avelino, J., Barboza, B., Araya, J.C., Fonseca, C., Davrieux, F., Guyot, B. and Cilas C. (2005). EtTects of slope exposure, altitude and yield on coffee quality in two altitude terroirs ofCosta Rica, Orosi and Santa Maria de Dota. Journal of the Science of Food and Agricultre 85: 1869-1876. Decazy, F., Avelino, J., Guyot, B., Perriot, J.J., Pineda, C. and Citas, C. (2003). Quality of different Honduran coffees in relation to severa! environments. Sensory and Nutritive Quality of Foods 68:2356-2361 . Fotheringham, A.S., Brunsdon, C., Charlton, M. (2002). Geographically Weighted Regression: The Analysis of Spatially Varying Re/ationships. John Wiley and & Sons, West Sussex, England. Jarvis, A., Rubiano, J ., Nelson, A., Farrow, A. and Mulligan, M. (2004). Practica! use of SRTM data in the tropics - Comparison with digital elevation models generated rrom cartographic data. Working Document no. 198. CIA T, Cali, Colombia. Jones, P., and Thomton, P. (2000). MarkSim: Software to generate daily weather data for Latín America and Arrica. Agron. J. 93: . Jones, P.G., Thomton, P.K., Diaz, W. and Wilkens, P.W. (2002). MarkSim, Version l. A computer too! that generates simulated weather data for crop modeling and risk assessment. CIA T CO-RO M series, CIAT, Cali, Colombia. Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G. and Jarvis, A. (2005). Very high resolution interpolated climate surfaces for globalland areas. lnt. J Climatol. 25: Uiderach, P., Oberthur, T., Niederhauser, N., Usma, H., Collet, L. and Pohlan, H.A.J. (2006). Café especial: Factores, dimensiones e interaciones. In: El Cafetal del Futuro: Realidades y Visiones. Pohlan, J., Soto, L. And Barrera, J. (eds). Shaker Verlag, Aachen. 462pp. Lara-Estrada, L.D. (2005). Efectos de la altitud, sombra, producción y fertilización sobre la calidad del café (Coffea arabica L. var. Caturra) producido en sistemas agroforestales de la zona cafetalera nor-central de Nicaragua. Master thesis, Centro Agronómico Tropical de Investigación y Enseñanza. Costa Rica. 1 06pp. Nelson, A. (2004). The spatial analysis of socio-economic and agricultura! data across geographical scales: Examples and applications in Honduras and elsewhere, PhD Thesis, University ofLeeds, England. pp369. Linacre, E. ( 1977). A simple formula for estimating evaporation rates in various clima tes, using temperature data alone. Agríe. Meteoro!. 18:409-424. O'Brien, R. (2004). Spatial Decision Support for Selecting Tropical Crops and Forages in Uncertain Environments. PhD thesis, Department ofSpatial Sciences, Curtin University ofTechnology, Perth, 278pp. Pearl, J. (1990). Bayesian decision methods. In: Readings in Uncertainty Reasoning, Shafer, G. and Pearl, J . (eds.) Morgan Kaufinann, San Mateo, CA. Pp.345-352. Vaast, P. Cilas, C. Perriot, J; Davrieux, J; Guyot, B; Bolaños, M. (2005a). Mapping of Coffee Quality in Nicaragua According to Regions. Ecological Conditions and Farro Management. In Proceedings of the 20th lnternational Congress on Coffee Research (ASIC) Bangalore, India. Pp 842-850. Vaast, P; Van Kanten, R; Siles, P; Dzib, B; Frank, N; Harmand, J; Genard, M. (2005b). Shade: A key factor for coffee sustainability and quality. Proceedings of the 20th lnternational Congress on Coffee Research (ASIC), Bangalore, India. Pp887-896. 103 THEME3: IDENTIFYING THE CONDITIONS UNDER WIDCH ENVIRONMENTAL GOODS AND SERVICES CONTRIBUTE TO EQUITABLE AND SUSTAINABLE DEVELOPMENT 104 High-value agricultural products: Can smallholder farmers also benefit? Jonathan Hellin°, Douglas Whiteb and Rupert Bestb 8Centro Internacional del Mejoramiento de Maíz y Trigo CIMMYT, Mexico bCentro Internacional de Agricultura Tropical CIAT, Cali, Colombia Abstract A rapid ly changing agriculture and food economy threatens resource-poor smallholder farmers. Consumers throughout the world are shifting away from staple-based diets (grains and tubers) to energy-rich diets with increasing amounts of meat, especially poultry. In addition, consumers are eating perishable foods such a fresh fruits and vegetables that also require timely and careful handling. Smallholders often raise and cultivate these high-value agricultural products (HV APs). Nevertheless, their products rarely reach such consumers or obtain high prices. The ambiguous term HV APs can confuse development practitioners, industry professionals and policymakers in their efforts to support smallholders. Farm products acquire high-vaJue by way of: (a) their inherent value arising from quality, scarcity or luxury attributes (e.g. specialty coffees and organic products), or (b) value-added activities that enhance consumer convenience (e.g. processing and packaging). Sorne HV APs have both inherent value attributes and value- added activities. Marketing strategies enhance consumer awareness and their willingness to pay. For products with imperceptible attributes (e.g. organic, social or environmental benefits), marketing information is essential for authentication. While many smatlholder farmers have comparative advantages in the production of HV APs, research and development efforts may not be effective in fostering the marketing components. This paper outlines cost-effective investment strategies according to different production, market and policy contexts. Introduction Approximately two-thirds of people living on less than US$1 per day are smallholder farmers or landless workers. They face extensive changes in the global agricultural economy including: a) A two-decade decline in real prices of commodities (Robbins 2003, FAO 2004); b) The dismantling of marketing boards that had provided free or subsidized inputs such as credit and fertilizer as well as extension and training (Gitt and Carney 1999, Byerlee and Echeverria 2002); e) Increasing competition from liberalized international trade policies (Bannister and Thugge 2001); d) Different consumer preferences for foods resulting from expanding and wealthier urban populations (Delgado, el al. 1999, Weinberger and Lumpkin, 2005); and e) Increasing1y stringent food quality and safety standards such as lower pesticide residues (Hill 2000). These changing conditions can further marginalize the ability of smatlholder farmers to participate in the market economy. Many resource-poor farm and landless families are by-passed by technology-based advances in agriculture production and marketing. In addition, smallholder producers often lack the supporting institutions (e.g. market information and transport networks) to compete in the marketing sphere (Mangisoni 2000, World Bank 2004). Unable to compete in the market economy, many smallholders will continue subsistence production or migrate to urban areas. 105 Food systems can no longer be viewed simply as a way of moving basic staples from farm to local plates (001 2003). Numerous farmers now produce to the demands of niche markets and according to specific grades and standards of sophisticated supply chains. In many regions, retail systems have spurred these changes, especially the growth and increasing concentration of supermarkets. For example, the share of supermarkets in Latín America has risen from a mere 10-20% offood retail in 1990 to 50-60% by the early 2000s. Small shops and open-air markets were displaced in the process (Reardon and Berdegué 2002). In China, no supermarkets existed in the late 1980s; food retail was nearly completely controlled by the government. Yet by 2003, supermarkets had captured a 13% share ofnational food retail sales and 30% of urban food retail sales (Hu el al. 2004). Changes in the global agricultura! economy provide rural producers with new opportunities. High-vaJue agricultural products (HV APs) are seen to be an attractive development option for smallholder farmers (USAID 2004, ADB 2005). Government, non-government and industry efforts can all contribute to smallholder involvement in HV AP markets. Confusion about what HV APs are, however, creates misunderstandings amongst researchers, development practitioners and policy-makers regarding the actions needed to foster more lucrative and equitable HV AP markets for smallholder farmers at national, regional and international levels. This paper provides a conceptual framework that characterizes different types of HV APs, and identifies research, development, and policy efforts that facilitate farmers' participation in HV AP market chains. Such support ranges from training in production, processing, business development, organizational strengthening and marketing activities to institutional and policy changes. Clearing up the misconception of HV APs What is a HVAP? A good starting point is to state that HV APs are not staple commodities. They are not homogeneous products, traded on the basis of price; nor are HV APs the basic components of a diet. The first two words, 'high value', cause a strict definition to be elusive. High-value is a relative term, not only with respect to a dimension of being low-high, but also regarding the word value, which can be seen from either the consumer or producer perspective. Despite this ambiguous context, HV APs can be defmed as: a) Products that retum a higher net earning per unit of input (e.g. per hectare and/or per unit labor) than standard commodities; and/or b) Products that ha ve attributes for which consumers are willing to pay (WTP) a price prermum. HV APs often require special skills for their cultivation or collection. Fruits, vegetables, meat, eggs, milk, fish and many non-timber forest products can be considered HV APs (Gulati el al. 2006, Science Council 2005). An HV AP can also obtain value by its unique or scarce nature. Such "rare" products include spices, medicinal plants, and even crops that yield illegal drugs, e.g. coca and poppy (heroin). Scarcity value is also determined by time and place. For example, the value of fruits and vegetables increases outside the harvest season or production regions, despite the costs of transport or storage contributing to such higher prices. Product value reflects a consumer' s willingness to pay. The value of a product depends on its inherent attributes and how a consumer perceives them. lnherent attributes that differentiate products can be perceptible and imperceptible. Quality characteristics such as color, shape and size perceptibly distinguish a product. Less perceptible attributes include a product' s origin, which can be cultural and/or geographic. Examples of cultural attributes are products from ethnic groups and/or smallholders that are in contrast to products of faceless corporate agriculture. Geographic attributes refer to production from certain known and delineated areas (Umbarger et al. 2003). Other intangible attributes include product safety, for 106 example, having few pesticide residues or, to sorne, no genetically-modified organisms. Notwithstanding, a lack of safety can also generate higher values. Unsafe and risky products, such as illicit drugs, can have valued thrill-seeking characteristics. Both marketing efforts and product standards and certification help to preserve and enhance product attributes. Intangible attributes become more tangible to the consumer and enhance WTP. lnformation management along the supply chain can preserve and validate authenticity of intangible product attributes. A certification process of product management practices, such as organic or environmentally-friendly production, can ensure HV AP attributes. Socially-equitable product attributes, such as fair-trade, also have certifications. Specialized versions of commodities can also be HV APs. Many agricultura) products ha ve been changed from commodities into HV APs. The key to success of these initiatives has been the process of product differentiation accompanied with comprehensive product tracking and information systems. Tactics to authenticate product attributes and preserve their public perception will be discussed below. High-value versus va/ue-added Confusion remains between the terms high-value and value-added. In the context of HV AP markets, value- added corresponds to post-harvest activities such as processing, transformation, packaging, and marketing. These activities tend to pertain to links placed further along the supply chain and away from rural areas. distinguishes the term value-added from high-value by using examples of coffee, maize, fruit and vegetables, and animal (e.g. chicken) products. Table l. A comparison ofvalue-added and high-value products Market Commodity High-value Primary!Raw Coffee Bulk maize Fruits and vegetables Whole chicken Shade-grown or denomination coffee QPM, non-GMO or organic maize (Organic) fruits and vegetables Whole chicken Product Value-added Roasted coffee Maize flour Packaged fruits and vegetables Chicken parts Roasted denomination coffee Organic com chips Packaged organic fruits and vegetables Chicken parts (and free range) Value-added products are processed versions of primary or raw products. Post-harvest acttv1t1es of roasting coffee and packaging rice, for example, add value to the primary products. Fruits, vegetables and animal products can be considered commodity products. In many countries, large-scale industrialized agriculture produces a majority of these products. The same products can also have added-value with additional post-harvest activities. Packaging and storage, for example, enables a product to be sold at a different times ofthe year; transport permits sales in different locations. High-value products depend on their inherent attributes to obtain greater market prices. Unlike commodities, which are 'standard ' value products sold in commodity markets, high-value products typically have specific niche markets. Sorne commodities have become HV APs. Coffee is one of the most notable examples of the 'decommodification ' of a commodity. Coffee was long characterized as a homogeneous product with falling terms of trade and volatile prices. Yet, in recent years, coffee has 107 undergone growing product differentiation in final markets (e.g. shade-grown or origin denominated), with premium prices being paid (Kaplinsky and Fitter 2004). Decommidification examples exist for practically all commodities. To different degrees, differentiation has occurred with potatoes (Devaux el al. 2006), cassava (Best et al. 1994) and maize (Bender and Hill 2000). Although the inherent attributes of HV APs are acquired or managed before the harvesting of the product, post-harvest activities are often needed to maintain product identity and promote consumer awareness. Facilitating higher earnings by smallholders By addressing both farm management and post-harvest ( e.g. processing and marketing) contexts of smallholder farmers, development support can become more effective. Three of the six development support initiatives indicated below describe HV APs: ( 1) specialized management and handling (fruits, vegetables and animal products), (2) product differentiation with perceptible quality attributes (color, size, shape, smell, taste), (3) product differentiation with imperceptible quality attributes (safety, origin, producer characteristics). Related to HV APs, are initiatives that enhance (4) value-added activities (post-harvest processing and handling). In order to enhance and preserve consumer knowledge requires development efforts on (5) product and market and inforrnation. An enabling environrnent that ' levels the playing field' can be realized with development efforts that address (6) organizations, institutions and policies. Figure 1 summarizes these development support initiatives. Overlaps amongst these initiatives demonstrate that support efforts are not necessarily mutually exclusive and coordinated effort can lead to greater impacts. Effective, or at least non-detrimental, organizations institutions and polices are a necessary condition to achieve development impacts. At nearly the same leve) of importance is product and market inforrnation. In order for products with imperceptible quality attributes to obtain higher prices, product and market inforrnation is essential. The initiatives are presented in detail below. Organizations, institutions & policies Figure l. Development support initiatives for smallholder participation in markets 108 Specialized management and handling A fundamental farm management decision is whether to continue producing lower-value but known agricultural products, or change to higher-value but riskier products. While fruits, vegetables and animal products can have higher profit margins, their cultivation and raising typically requires extra inputs and managed care. Numerous factors influence technology adoption decisions, ranging from the availability of inputs (e.g. Iand, fertilizers and pesticides) to contextua) issues such as agronomic and market conditions, Iand tenure and infrastructure (Feder and Umali, 1993; Franzel et al., 2001 ). Fitting a technology into the agricultura\ system of the resource-poor farm is often required to assure technology adoption (Collinson, 1972; Zandstra, et al., 1981; Byerlee and Siddiq, 1994 ). Family labor is one of the most important production inputs of smallholder farms (Schultz, 1964; Ruthenberg, 1976; Byerlee and Collinson, 1980). Nevertheless, farmers and economists have dramatically different perceptions regarding the value of labor, especially during peak demand periods such as planting and harvest (White, et al. 2004). Three conditions of market failure lead to this disparity. First, labor availability during peak demand is often insufficient; or in other words, labor supply markets are inelastic (de Janvry, et al. 1991). Although a farmer may wish to expand production, sufficient labor cannot be acquired despite the payment of higher wages. Second, resource-poor farmers rarely have sufficient financial resources with which to purchase labor. The seasonal nature of harvest sales and the lack of credit markets leads to cash flow constraints. Three, even if funds were available to purchase labor, numerous transactions costs exacerbate difficulties of obtaining and managing temporary laborers. In remote areas where many resource-poor farms are located, a rudimentary transportation and communication infrastructure inhibits the acquisition of additional labor. Together, these conditions of market failure (inelastic labor supply, liquidity constraints and transactions costs) restrict the ability of farmers to meet temporary labor demands and obtain higher farm eamings. Va/ue-added activities Agro-enterprise and post-harvest systems can raise incomes of rural people. Such value-added activities (e.g. cooling, cleaning, sorting and packing) are labor intensive and generate value added earnings. Urban consumers are increasingly aware of and willing to pay for foods of consistent quality and safety. Compliance with conditions set under sanitary and phytosanitary regulations that accompany WTO negotiations are critical for the continued participation of smallholder farmers in intemational agricultura) markets. The handling and treatment of agricultura! products after they leave the farm gate Iargely determine the ability of agricu1tural products to meet the required conditions. Such food safety standards are non-tariff barriers not only imposed by importing countries but also by the prívate sector (Reardon and Farina, 2001 ). To participate in increasingly formal urban and export markets, producers need access to well-organized post-harvest chains that can adequately process and market agricultural products. Consequently, smallholder farmers will need to become more integrated with upstream processing oftheir produce (Minot, 2005). Product differentiation with perceptible qua/ity attributes Consumers are often willing to pay more for color, size, shape, smell, and taste. Smallholder farmers are frequently not aware of the potential opportunities to market high quality - high value products. lf smallholders are conscious of particular opportunities, they need to identity optimal production niches on their land that are most suitable for high-value crops. The differentiation away from standard mediocre quality agriculture products will most likely occur in environmental niches that provide the appropriate conditions for production and marketing of these products. The growth of the specialty coffee market is ample evidence that demand for higher valued quality coffees is increasing (Oberthür, 2004). 109 Product differentiation with imperceptible qua/ity attributes A growing emphasis on production methods (for example, organic production and fair trade) and end traits (e.g. non-GMO products), requires identity preservation (in sorne cases, segregation may be sufficient) and separate marketing channels. This marks a departure from the traditional bulk commodity focus based on blending and large volumes. Niche markets for non-GMO products may develop, similar to the present market for organic foods, which is characterized by separate identity-preserved marketing and associate premium prices. Marketing mechanisms, however, are generally not in place. Maize is typically considered a commodity. An important obstacle currently impeding the widespread cultivation of value-enhanced maize is the lack of a widely recognized price mechanism for the specialty characteristics (Meng and Ekboir 2001 ). Maize can al so be HV AP and an input for other HV APs. In Brazil, the expansion of the feed and poultry industries has induced a transformation of maize producers into commercial farmers. The growth of demand for maize feed has had major implications for the opportunity to differentiate maize, including quality protein maize (QPM). QPM has been successfully introduced into Brazil and China for use as an improved livestock feed. QPM can serve as a fortification programme within local people' s normal nutritional regime (Lauderdale, 2001 ). Many HV APs are more perishable than commodities. In order to ensure quality and consistent supply of perishable goods, supermarkets (and other retailers) are pushing the food marketing system toward more vertical coordination (Gulati et al. 2006, Boehlje 1999). Vertical coordination allows retailers to standardize quality, improve bargaining power, and achieve economies of scale. This vertical coordination has such profound impacts on smallholder farmers. Retailers traditionally procure products from traditional wholesale markets, which are aggregated from many producers. The growth in demand for HV APs has been accompanied by a shift from exclusive reliance on commodity markets toward the use of specialized/dedicated wholesalers (Reardon et al. 2006). Specialized wholesalers are usually more responsive to quality, safety, and consistency requirements than traditional wholesalers. Product and marlcet information Realizing production increases are not the only challenge facing smallholder farmers. To ever increasing degrees, management with respect to fertilizer and pesticide application, is coupled with proper harvesting, grading and processing in order to obtain higher prices. Although reputation can function in local and national markets, systems of information management are needed to ensure that producers obtain higher prices for their superior products. For example, markets for differentiated commodities exist in the deve1oped world. They required the development of mechanisms and technologies to identify specialized traits and track the movement of the produce through the supply chains. 1 The key question is whether similar markets exist in the developing world and whether smallholder producers can access these markets. Agricultural processors and traders face increasing pressures to certify the safety of production practices (such as to avoid pesticide residues in the fmal product), quality attributes, adequate supplies and on-time deliveries. To do so, they also must rely on a large number of independent small farmers. Vertical integration, contract farming, and traders' associations can address these problems by reducing the moral hazard of farmer non-compliance, which can compromise an entire supply chain. lnformed policies and a conducive regulatory environment increase the incentives for agro-processors to use the produce of small-scale farmers as inputs, and improve their capacity to meet the product attributes required in a rapidly modemizing agricultura) marketplace. (Minot, 2005) 1 See www.identifvpreserved.com for examples from the United States and http://web.aces.uiuc.edu/valueJ for specific examples from Illinois 110 Organizations, institutions and policies The growth in high-value agriculture and the increasingly vertically-integrated chains associated with HV APs has raised questions about govemance and equity of supply chains (re-goveming markets). Close linkages are needed between farmers, processors, traders, and retailers to coordinate supply and demand (Gulati et al. 2006). This integration focuses on increasing profits and product value and is usually led by the more powerful members of the chain, in particular, the retail sector. For smallholder farmers who manage to enter such lucrative markets, many find it difficult to remain due to an inability to withstand losses stemming from unexpected production and/or price decreases and cost of cultivation involved. The challenge for farmers is that HV APs are typically associated with high transaction costs (Pingali et al. 2005). If development practitioners wish to contribute to poverty reduction with HVAPs, a market-literate approach is required. Such an approach aims to promote the growth and improved performance (for example, competitiveness, productivity, employment, value addition, linkage coordination) of chains in ways that benefit poor small-scale producers (Hellin et al. 2005). The value chain analysis school (for example, Kaplinsky and Morris 2001 ), analyzes the segments of a supply chain, and in particular the relations between the segments of the chain. New institutional economic analysis refers to this process as organizational and institutional analysis of govemance and coordination of the supply chain (Reardon 2005). An analysis of HV AP supply chains can reveal where inefficiencies in the chain exist and by bringing different stakeholders together, these chains can be made to work more effectively and efficiently through participatory approaches for example, the Leaming Alliances (Lundy, et al. 2005). A key issue that policy makers and development practitioners have focused on is farmer organization as a tool for collectíve action to bring about greater equity in HV AP supply chains (Hellin et al. 2006). Collective action and producer organizations are therefore not surprisingly one of the foci of the pro-poor market approach (e.g. DflD 2005). Another cause (and consequence) of increased vertical coordination in supply chains is the growth of private quality and safety standards as well as prívate enforcement of public standards. In particular, strict requirements will be in force in two areas: food safety, driven by recent food crises in Euro pe, and the traceability of the identity and origin of tbe product (V ellema 2004). Smallholder farmer are unlikely, unless they are very well organized, to meet many of these quality standards (Shepherd 2005). Farmer organisation is not, however, a panacea. Research in Chile (Berdegué, 2002) and Meso-America (Hellin et al. 2006) demonstrates tbat only under certain market conditions do small farmers to engage in collective action in order to access markets. Such efforts imply significant costs and risks, but only occasionally have the potential to benefit farmers. Producer organisations are more effective amongst producers of HV APs as opposed to producers of commodities. There is far less incentive for producers of commodities to organize themselves as the transaction costs associated with market access are relatively low: there are so many buyers and sellers that farmer organizations would have little impact on, for example, prices. Strategies are, therefore, needed that enable producers to diversify or upgrade production, and to compete more effectively in markets where they have advantages. Making HV AP markets work for the poor involves strengthening competitiveness in the enterprises, supply chains and wider business environments on which rural producers depend. A key element is the provision of business developrnent services (BDS). Once an enterprise has been established, there is an on-going need for it and other supply chains actors to access services of different types, both market and technical, that will allow it to grow and maintain its competitiveness (Best et al. 2005). The range of BDS includes: input supplies (seeds, livestock, and fertil izers); market information (prices, trends, buyers, suppliers); transport services; and support for product development and diversification. 111 Much debate surrounds identifying the role of the prívate and public sector in the provision of BDS (for example, Miehlbradt and McVay 2005). Traditionally, BDS to small enterprises (including agricultura! supply chains) have been delivered with the support of donor and government subsidies. Critics suggest that this distorts market prices and undermines the provision of BDS by the prívate sector. The neo-liberal market paradigm in the provision of BDS delivery signifies a shift in thinking from subsidized supply-led BDS provision to market-determined demand-driven services. In many cases, however, the prívate sector has preven incapable of replacing previous state services due to high transaction costs, dispersed clientele and low ( or non-existent) profits (Lundy et al. 2002). Contract farming is an intermediate institutional arrangement, a form of ' vertical coordination ' as well as a sourcing mechanism, that líes between commodity market purchases and commodity production on owned or rented land (Echánove and Steffen 2005). Under contract-farming arrangements, landowners or tenants have obligations with agribusiness marketing and/or processing firms, which specify prices, timing, quality and quantity/acreage of the produce to be delivered . Contract farming allows firms to participate in, and exert control over, the production process without owning or operating the farms (Key and Runsten 1999). In the developing world, contract farming has spread rapidly since the 1980s, especially in relation to HVAPs. Conclusions The global agricultura! economy is changing, with significant implications for the vast number of smallholder farmers. Gulati, et al. (2006) posit that to significantly improve their incomes per capita over the next twenty years, farmers in Asia must either be part of the shift to HV APs or increase the share of income they get from non-agricultura! sources. Furthermore, unless smallholders become verticaJly integrated with processors and retailers, they will increasingly have difficulties in participating in high-value markets (Gulati, et al. 2006). 112 References Asian Development Bank (ADB), Asian Development Bank lnstitute, Department for lntemational Development. (2005). Joint workshop on making markets work better for the poor (MMW4P). ADB Headquarters, Manila, Philippines: 15-16 February. http://www.dfid.gov.uk/news/files/trade news/adb-workshop.asp Bannister, G.J. and K. Thugge (200 1). lntemational Trade and Poverty Alleviation. Finance and Development. 38(4) http://www .im f.org/extemal/pubs/ftlfandd/200 11 12/bann iste.htm Bender, K. and Hill, L. (2000). Producer Alternatives in Growing Specialty Corn and Soybeans. Department of Agricultura! and Consumer Economics, College of Agricultura), Consumer and Environmental Sciences, University of Illinois at Urbana-Champaign. Berdegué, J. (2002). Leaming to beat Cochrane's treadmill: Public policy, markets and social leaming in Chile's small-scale agriculture. ln: Leeuwis, C. and Pyburn, R. (eds.). Whee/barrowsfu/1 offrogs: Socia/learning in rural resource management. Pp. 333-348. Best, R., Henry, G., Gottret, M.V. (1994). El impacto de la industria de la yuca seca en la costa Atlántica de Colombia. [The lmpact of dry cassava industry in the Atlantic Coast of Colombia]. In: El desarrollo rural en América Latina hacia el Siglo XXI memorias del seminario-taller internacional. Pontificia Universidad Javeriana, Santafé de Bogotá, D.C. (Colombia). 2:3-17. Best, R., Ferris, S. and Schiavone, A. (2005). Building linkages and enhancing trust between small-scale rural prodicers, buyers in growing markets and suppliers of critica! inputs. ln: Almond, F.R. and Hainsworth, S. D. (eds.). Beyond agricu/ture - making markets workfor the poor. Proceedings of an intemational seminar, 28 February - 1 March, 2005. Westminster, London, VK. Crop Post-Harvest Programme CPHP), Natural Resources lnternational Limited and Practica) Action. Pp. 19-50. Boehlje, M. (1999). Structural changes in the agricultura! industries: how do we measure, analyze and understand them? American Journal of Agricultura/ Economics 81:1028-1041. Byerlee, D., Collinson, M.P. (1980). Planning Technologies Appropriate to Farmers: Concepts and Procedure. ClMMYT, Mexico. Byerlee, D., Siddiq, A. (1994). Has the Oreen Revolution been sustained? The quantitative impact of the seed- fertilizer revolution in Pakistan revisted. World Development 22:1345-1361 . Byerlee, D. and Echeverria, R. (eds.) 2002. Agricultura/ research po/icy in an era of privatiza/ion. Wallingford, UK: CABI Publishing, Wallingford, VK. Collinson, M.P. (1972). Farm Management in Peasant Agriculture. Praeger, New York. de Janvry, A., Fafchamps, M., Sadoulet, E. (199 1). Peasant household behavior with missing markets: sorne paradoxes explained. The Economic Journa/1 O 1: 1400-1417. Delgado, C., Rosegrant, M., Stienfeld, H., Ehui, S., and C. Courbois. (1999). Livestock to 2020: the Next Food Revo/ution. lnternational Food Policy Research Institute. Washington D.C. DflD (2004). What is pro-poor growth and why do we need to know? Pro-Poor Growth Briefing Note. DflD, London. DflD (2005). Making Market Systems Work Beller for the Poor (M4P) - An introduction to the concept. Discussion paper prepared for the ADB-DFID ' leaming event' ADB Headquarters, Manila. February 2005 Devaux, A., G. Thiele, G. López, C. Velasco. (Eds.), (2006) Papa Andina: Innovación para el desarrollo en los Andes. 2002-2006. CIP-Papa Andina, Lima, Republic of Peru, pp.83, 92-9060-286-4. http://papandina.cip.cgiar.org/fileadmin/documentpool/ Administrativo/lnforme/Fase/06-1 O-JI- Compendio 2002 2006.pdf Echánove, F. and Steffen, C. (2005). Agribusiness and farmers in Mexico: the importance of contractual relations The Geographica/ Journa/171: 166-176. Evenson, R.E. and Gollin, D. (2003). Crop Variety lmprovement and lts Effect on Productivity: The lmpact on lnternational Agricultura/ Research. : CABI Publishing, Wallingford, VK. FAO (Food and Agriculture Organization of the United Nations). (2004). The State of Food fnsecurity in the World 2004. Annual report by Economic and Social Department, FAO, Rome, Italy Feder,G. and Umali, D.L. ( 1993). The adoption of agricultura! innovations: a review. Technologica/ Forecasting and Social Change 43:215-239. Franzel, S., Coe, R., Cooper, P., Place, F., Scherr, S.J. (2001). Assessing the adoption potential of agroforestry practices in sub-Saharan Africa. Agricultura/ Systems 69:37-62. Gill, G. and Camey, D. (1999). Competitive agriculture technology funds in developing countries. Natural Resource Perspectives. (41) London: Overseas Development lnstitute. URL: http://www.odi.org.uk/nrp/4 J.html Gulati A., Minot, N., Delgado, C. and Bora, S. (2006). Growth in high-value agriculture in Asia and the emergence of verticallinks with farmers. Submitted to MTID Discussion Paper series 113 Hellin, J., Griffith, A. and Albu, M. (2005). Mapping the market: market-literacy for agricultura! research and policy to tackle rural poverty in Africa. In : Almond, F.R. and Hainsworth, S.D. (eds.). Beyond agriculture - making markets workfor the poor. Proceedings of an international seminar, 28 February - 1 March, 2005. Westminster, London, UK. Crop Post-Harvest Programme CPHP), Natural Resources Intemational Limited and Practica! Action. Pp. l 09-148. Hellin, J., Lundy, M. and Meijer, M. (2007). Farmer organization, collective action and market access in Meso-America. (In press.) Hill, A. 2000. Qua/ity Control Proceduresfor Pesticide Residues Analysis. Guidelines for Residues Monitoring in the European Union. Second edition, 1999/2000. Docurnent No. SANC0/31 03/2000. On behalf of the European Commission following discussions at the 2nd EU Workshop on Coordinated Analytical Quality Control, Athens, Greece, 15-17 November, 1999. http://ec.europa.eu/food/fs/sfu/ph ps/pest/gualcontrol en.pdf Kaplinsky, R. and Fitter, R (2004). Technology and globalizations: who gains when commodities are de- commodified? lnternationa/ Journa/ ofTechnology and Globalisations 1:5-28 Key, N. and Runsten, D. (1999). Contract farming, smallholders, and rural development in Latin America: The organization of agroprocessing firms and the scale of outgrower production. Wor/d Development 27:381-401. Lauderdale, J. (2001). CIMMYT technology improves nutritionaJ quality ofmaize. In: Pingali, P.L. (ed.). CIMMYT 1999-2000 World Maize Facts and Trends. Meeting World Maize Needs: Technologica/ Opportunities and Prioritiesfor the Pub/ic Sector. CIMMYT, Mexico, D.F. Lundy, M., Gottret, M.V. and Ashby, J. (2005). Leaming alliances: An approach for building multistakeholder innovation systems. ILAC Brief8. Lundy, M., Ostertag, C.and Best, R. (2002). Value adding, agroenterprise and poverty reduction: A territorial approach for rural business development. Paper presented at Globalization of Agricultura! Research conference, 25-27 Feb., Turrialba, Costa Rica. Mangisoni, J.H. (2000). Economic efficiency and investment potential in the smallholder crop sector in Malawi. lnternational Journal ofSocial Economics 27:968-979. Meng, E. and Ekboir, J. (2001). Current and future trends in maize production and trade. In: Pingali, P.L. (ed.). 2001. CIMMYT 1999-2000 World Maize Facts and Trends. Meeting Wor/d Maize Needs: Technological Opportunities and Prioritiesfor the Public Sector. CIMMYT, Mexico, D.F. Miehlbradt, A.O. and McVay, M. (2005). From BDS to making markets work for the poor. lntemational Labour Organisation, Geneva, Switzerland. Minot, N. The Development ofPost-harvest Systems and Agro-industry as a Strategy to Raise the lncome ofthe Rural Poor. http://www.ifuri.org/themes/mp20.htm Oberthur, T. (2004). Science, business and growers team up to help de-commodifying smallholder agriculture. Coffee Faces. (11) October. Virmax, UK http://www.virmax.com/site/pdfVirmax/coffeeFaces11 .pdf Overseas Development lnstitute (ODI) (2003). Food policy old and new. ODI Briefing Paper November 2003 (unnumbered) ODI, London, UK. Pingali, P., Khwaja, Y and Meijer, M. (2005). Commercializing small farms: Reducing transaction costs. ESA Working Paper No. 05-08. October 2005.URL: htop://www.fao.org/es/esa Agricultura! and Development Economics Division. F AO, Rome. Reardon, T. and Farina, E. (2001). The rise of prívate food quality and safety standards: illustrations from Brazil. lnternational Food and Agribusiness Management Review. 4:413-421. Reardon, T and Berdegué, J. (2002). The Rapid R.ise ofSupermarkets in Latin America: Challenges and Opportunities for Development. Development Policy Review 20:371-388 . Reardon, T. (2005). Retail Companies as lntegrators of Value Chains in Developing Countries: Di.ffusion, Procurement System Change, and Trade and Development Effects. GTZ Reardon, T. and Flores, L. (2006). Customized Competitiveness. Strategies for Horticultura! Exporters: Central America Focus with Lessons from and for other Regions. Food Policy 31:. Reardon, T., Berdegué, J. Balsevich, F., Flores, L. and Hemández, R. (2006). Supermarkets and small horticultura! product farmers in Central America. In: Swinnen, J. (ed.) Global supply chains, standards, and poor farmers. CABI Publishing, Wallingford, UK. Robbins, P. (2003). Sto/en Fruit: The Tropical Commodities Disaster. Zed Books. London and New York. Ruthenberg, H. (1976). Farming Systems in the Tropics. Oxford University Press, Oxford. Schultz, T.W. (1964). Transforming Traditional Agriculture. Vale University Press, New Haven. Science Council ofthe CGIAR. (2005). URL: http://www.sciencecouncil.cgiar.org/activities/spps/pubs/Priorities%200ct'l/o2005.pdf 114 Shepherd, A. (2005). The i.mplications of supermarket development for horticultura) farmers and traditional marketing systems in Asia. F AO, Rome, ltaly. Stirling, C.M., Harris, D. and Witcombe, J.R. (2006). Managing an agricultura) research prograrnme for poverty alleviation in developing countries: an institute without walls. Experimental Agriculture 42:127-146. Umbarger, W. J., Fuez, D. M., Calkins, C.R. and Sitz, B.M. (2003). Country-of-origin labeling ofbeefproducts: U.S. consumers' perceptions. Presented at the 2003 FAMPS Conference, Emerging Roles for Food Labels: lnform, Protect, Persuade. Available online: www.rcalf.com/COOL/COOL%20Beef%20Products%20-%20Consumers%20Perception%20Survey.pdf USAID (2004). USAID Agricu/ture Strategy: Linking Producers to Markets. U.S. Agency for lntemational Development .Washington, D.C. Available online: www.usaid.gov/our work/agriculturelag strategy 9 04 508.pdf USDA (2004). Fruit and Vegetable Consumption Looking Ahead to 2020. Economic Research Service. Agriculture Information Bulletin 792-7 October. Vellema, S. (2004). Production chains as configurations for development: perspectives for innovation, market development and collaboration. lntemational Development Discussion Paper no 2006-1. Wageningen UR. Weinberger, K. and T. Lumpkin. (2005). High value agricultural products in Asia and the Pacific for small-holder farmers: Trends, opportunities and research priorities. Workshop How Can the Poor Benefit from the Growing Markets for High Va/ue Agricultura/ Products. October 3-5. CIAT, Cali, Colombia. http://www.zef.de/modulelregister/media/3ced weinberger lumpkin 2006.pdf World Bank (2005). Directions in Development: Agricultura/ Growth for the Poor: An agenda for deve/opment. Washington, D.C: World Bank. World Bank (2004). Agriculture lnvestment Sourcebook. Smal/holder Dairy Production. http://www-esd.worldbank.org/ais/index.cfin?Page=mdisp&m=04&p=2 Zandstra, H.G., Price, E.C., Litsinger, J.A. and Morris, R.A. (1981). A Methodologyfor On-Farm CroppingSystems Research. IRRI, Los Baños. 115 Watershed management and poverty alleviation in the Colombian Andes Nancy Johnson8 , James García\ Adriana Moreno\ Sara Granadosc, Harvey Rodriquezb, Jorge A. Rubianob, Alexandra Peralta8 , Jorge E. Rubiano8 , Marcela Quintero8 , Rubén Estradad and Esther Mwangie a Centro Imternacional de Agricultura Tropical CIA T, Cali, Colombia b Semillas de Agua, Cajamarca, Colombia eSCALES Project (CJAT), Cali, Colombia d Concorcio para el Desarollo Sostenible del Ecorregión Andina CONDESAN, Lima, Peru elnternational Food Policy Research Institute IFPRI, Washington, USA Abstract Watersheds, especially in the developing world, are increasingly being managed for both sustainability and equity objectives. How complementary are these objectives? In the context of a watershed, the actual and potentiallinkages between land and water management and poverty are complex and likely to be very site specific and scale dependent. This study uses multiple methods to analyze the importance of water in the livelihoods of the poor at multiple scales. At the household scale, the Stages of Progress method (SOP), a participatory approach for assessing poverty dynamics, is used to assess how poverty has changed over time in communities and to identifY the assets, including land and water, and livelihood strategies associated with poverty andlor progress. The results are then interpreted in the broader economic and environmental context of the watershed, and the implications for potential poverty-environment tradeoffs are identified. In the case of Colombia, the results show that while many households do improve their welfare status by investing in improved water access, opportunities to lift households out of povetry via better water access alone are limited. Off-farm labor, mainly in mining and commercial dairy, were the major pathways out ofpoverty in the region. Since activities also contrinutors to environmental problems in the watersheds, the potential exists for poverty- environment tradeoffs. The results demonstrate potential of multi-scale analysis to contribute to the design of interventions that minimize tradeoffs between environrnental and social welfare objectives. Introduction Watersheds, especially in the developing world, are increasingly being managed for both environmental and rural development objectives. The actual and potential linkages between land and water management and poverty are, however, complex and may be site specific and scale dependent (Swallow et al. 2006). Sorting out these relationships to arrive at policy- and management-relevant conclusions requires methods of poverty assessment that allow for complex relationships but are not too complex or costly to implement, since site specificity may limit the extent to which results can be extrapolated from one site to another. This paper uses the stages of progress methodology (Krishna 2002) to assess the relationship between water and poverty in two watersheds in the Colombian Andes. The methodology includes both a participatory assessment of current poverty and an analysis of how household poverty status has changed over the last 25 years. Taken together, the two results capture both direct and indirect linkages between water and poverty. They identif)' situations where win-win solutions may be possible, and also where it is likely that trade offs will be required, not only between environmental, economic growth and equity objectives at the watershed scale, but also between households' welfare objectives and the strategies that they use to achieve them. 116 The paper is organized as follows. The next section summarizes the literature on poverty and watersheds, from the perspective of watershed management and diversification of livelihoods. We then describe the study watersheds, followed by the presentation of the SOP methodology. We then present and analyze the results, and conclude with a discussion ofthe implications for policy and practice, using examples from the watersheds. Watersheds: The " missing middle" in livelihoods research The complex relationship between watershed management and poverty alleviation rests on the fact that watershed management is not just about managing water, but is al so about managing land, forests and other resources. The livelihood strategies of the poor interact with these resources in a range of ways. Even though they may not "own" them, the poor, even the landless poor, are often able to access and use resources, which means that changes in the rules that govem resource use can directly affect them. In addition, livelihood strategies of rural households are increasingly diverse. Even in rural areas, households do not depend exclusively on agriculture or extraction of natural resources. Off-farm income from sale of labor or cornmercialization of products and services is important for household welfare of the rural poor. Therefore, the impacts of environmental, industrial, transportation and other policies that often are considered as coming under the ambit of watershed management may have significant implications for the their welfare. Poverty and watershed management The term watershed management refers to the use and management of inter-dependent resources- cropland, pastures, forests, wetlands, as well as water--within hump-backed ''watersheds" that dispose water into streams and bowl-shaped 'catchments" that collect water into a common outlet (Swallow et al. 2001b, Kerr and Chung 2001). Because watersheds integrate diverse resources, environmental services, uses and users, an integrated approach to management that considers the goods and services they provide is essential. Often these goods and services, for example, soil stabilization, water catchment protection, agricultura) land, pastures, forests, or wetlands have significant impacts on livelihoods of the rural poor who live both in upstream and downstream areas of a watershed. The interconnectedness of upstream and downstream sections of watersheds has been extensively investigated by biophysical scientists studying hydrological cycles, stream flows, soil nutrient and sedimentation flows. In contrast, a similar unified analyses of livelihoods, poverty and Jand use along different points ofthe watershed are Jacking. Many social and economic studies have focused on individual elements within a watershed such as agriculturalland or forests, but not on the watershed as a system that is both biophysical and social, with two-way linkages between upstream and downstream (Swallow et al. 2006). Researchers are increasingly recognizing the watershed as the "missing middle" for research on land use and livelihoods (SwaJJow and Meinzen-Dick 2001), where land use and environmental services have impacts beyond the individual farms yet not at the regional or globallevel. Sorne evidence exists that there may be a synergistic relationship between rural development and watershed management. A recent global review of watershed management projects implemented in the last 25 years indicates that projects that focused on individual farms rather than the entire watershed tended to be site- specific, isolated and dispersed, serving farmers superficially but reducing the potential of broader impact (Perez and Tschinkel 2003). In contrast, better integrated projects had scope for increasing ecological benefits while at the same time contributing to the economic development of poor rural communities (Grewel et al. 1999). For example a watershed management project in the Himalayan foothills in the northem states of India included various ecological restoration and conservation efforts as well as water development and breed improvement. lt resulted in increased employment, increased maize and wheat yields and reduced drudgery of women, among other benefits. 117 A better understanding of tbe adaptive capacities of the rural poor to manage water supply and quality in light of their multiple needs and the multiple functions of watersheds is necessary in integrated watershed management programs (Fereres and Kassam 2003). lt provides an opportunity for addressing natural resource management as well as important socio-economic problems. Understanding livelihoods and their implications for watershed resource use A livelihood comprises a household's 'means to a living', or its strategy to attain a level of material well- being, which can be reflected in income, consumption, satisfaction, health, longevity and other measures (Bezemer and Lerman 2004). For example, household livelihood strategies may involve different members of the family seeking and fmding different sources of food, fuel, animal fodder, cash and support in different ways in different places (on farm, through commonly shared resources, or markets) at different times of the year (Cbambers 1995). Livelihood strategies are complex and diverse, contributing to household incomes in ways that are not always irnmediately evident toan externa! observer. Components of livelihoods tbat enable households to make a living include assests (natural, physical, human, social and financia!) and the activities by which individuals gain to access these assets. The way that households combine these assets, the trade-offs they may make between assets at different times and the welfare outcomes are tbe subject matter of poverty and livelihoods analyses. However, De Haan and Zoomers (2005) encourage analysts to look beyond material aims and motives, and also to consider tbe structures and processes that may enhance or limit activities and assets. Property rights and power relations often influence the range of activities that individuals and groups can engage in. Similarly, geographical settings, seasonality and distance to markets all influence the set of opportunities and outcomes available to the poor (Zoomers 1999). It has long been established that the rural poor engage in multiple livelihoods strategies. In eastem Zambia for example, Peterson (1999, cited in Gladwin et al. 2001) shows that while agriculture forms the basis of households' incomes, it is not the only source. Women are engaged in altemative cash-generating activities including beer brewing, sale of snack foods, vegetable production and processing and selling, performing ganyu (piece work), and petty trading. Men also are engaged in informal income-generating activities, including renting out oxen for plowing, made brick making, construction, selling firewood and various forms of petty trading. These diverse activities are generally classified into those that generate incomes away from one's family farm ( otf-farm) and those non-farm activities that involve the processing of raw goods into secondary and tertiary products (such as flour milling, cheese production) or use labor for the production of services (Barrett and Reardon 2000). In general these portfolios can be measured, often by studying the allocation of resources, mainly labor, to different activities or in terms of household income shares. Although Africa is largely viewed as a continent of subsistence farmers, non-farm sources are growing in importance and account for up to 45% of household incomes (see Bryceson and Jama) 1997, Reardon 1997, Little et al. 2001 cited in Barrett el al. 2001 ). For a detailed distinction of rural livelihoods strategies and their different income retums 2000b), however non-farm incomes overshadow those based entirely on agriculture. Multiple strategies, multiple motivations and effects on rural welfare There are many reasons why individuals attempt to diversify their strategies (Barret, Reardon and Webb 2001). First, "push factors", which include risk reduction, response to declining retums to assets, and high transactions costs, pusb households towards self-provision. For example, African women farmers combine farm and non-farm income-eaming activities to reduce risk and food insecurity where drought and pest attack reduce food security (Gladwin et al. 2001 ). Following the introduction of structural adjustment policies, small-holder farmers in Malawi resorted to micro-enterprise and closer integration with markets, which allowed them to adapt to increasing prices of organic fertilizer, loss in soil fertility and declining maize yields (Orr and Mwale 2001 ). 118 Second, "pull factors" include realization of complementarities between activities such as crop-livestock integration, specialization dueto advantages offered by superior technology, skills or endowments. These push and pull factors do not operate at the household or individual leve! alone, but are often linked to structural-level constraints/opportunities. Weak or incomplete financia! systems can create incentives for individuals to select activities that will smooth income and consumption in the face of climatic uncertainty, labor and land constraints. Similarly, proximity to urban areas can create opportunities for diversification in production- and expenditure-linked activities. ln sum livelihood diversification can serve the function of risk mitigation (Tumer et al. 2003) orbe a coping/survival strategy (Wood 2003, Orr and Mwale 2001, Reardon et al. 2001) or even a pathway out ofpoverty (Ellis 2000). But what are the effects on incomes and welfare of livelihood diversification on the rural poor? Does livelihood diversification, which is the consequence of individuals and households pursuing multiple livelihoods, have beneficia] and unambiguous effects on rural livelihoods? A growing number of studies in different parts ofthe world suggest that indeed it does. In Nepal, farmers in the central and westem mid-hill regions of are increasingly participating in off-farm wage labor to the point where it contributes significantly to farm income (Shivakoti and Thapa 2005). In Ethiopia, Block and Webb (200 1) show that income diversification is positively linked to improved welfare, including nutrition . In Tanzania income shares in peri-urban areas rise with per cap ita food consumption (Lanjouw et al. 2001 ), while Barrett et al. (2001a) report a strong, positive association between greater income diversification and wealth in Cote d'lvoire and in Kenya. However, the gains of diversification are tempered by access to public services such as education, communication and transportation, which determine participation in nonfarm activities. ln Uganda greater diversification increased income inequality. Poor, uneducated, women, recent migrants, and others lacking social ties enjoyed less access to new income-generating opportunities compared with educated males with strong social networks in the community (Canagarajah et al??? 2001). Moreover, the impacts of increased off-farm activities on land management practices are yet to be more carefully assessed. In the Honduran hillsides, for example, soil conservation is negatively associated with off-farm work; rather than use their off-farm income to boost their farming, farmers tend to practice minimally extensive farming because farming is now relegated to food security alone (Morera and Gladwin 2006). The authors suggest that off- farm work should be treated as one amongst many strategies as opposed to the main development strategy in the effort to resol ve poverty. The analysis of"off-farm" incomes is incomplete without including products and services that are accessed from areas that are off-farm but not necessarily in secondary or tertiary industries, typically on the commons. Such areas include common pool resources that may be held and managed under various property regimes such as common property or state property. These include forests, fisheries, pastures, swamps, roadsides that can be exploited for fodder, building material, food, game meat, fish, medicine etc. A cross section of studies in different settings shows that the commons often have a positive contribution on the livelihoods of the poorer and marginalized members of communities. In certain parts of India, income derived from CPRs contributes about 12% to the household incomes of poorer households (Beck and Nesmith 2001), and 15-25% in other parts (Jodha 1986, 1995). Reddy and Chakravarty (1999) found that forest income was associated with small reductions in income inequality for a sample of northem lndian farmers. In Zimbabwe, the share of household income attributed to CPRs is close to 40% for the poorest income quintile (Cavendish 2001 ), thus contributing to redistributive equity. In Malawi, Fisher (2004) found that 30% derive incomes from forests, which results in 12% reductions in measured income inequality. Here, asset-poor households were more reliant both on both low- and high-retum forest activities, compared with the better off. In southeastem Ghana the poorest households rely on sources to meet 20% oftheir food requirements during the lean season, compared to providing 2% and 8% ofwealthy and middle-income households needs respectively (Dei 1992). Taken together, these studies show that products from the commons can improve the welfare of the rural poor by supplementing incomes and 119 acting as safety nets when other opportunjties are lacking, or even as ways to reduce poverty where eamings are high enough. In considering the livelihoods strategies and their contributions to rural welfare it is evident that off-fann incomes are critica), but that the magnitude of benefits vary depending on prior distribution of assets such as education, social networks and on structural factors such as market access and also on gender~ all of which are critica) entry barriers. The category of off-fann income can also be expanded to include incomes derived from resources on the commons. These are important safety nets for the poor, do help to mitigate income inequalities and can under certain conditions contribute high eamings. Description of the study si tes Lake Fuquene Watershed The Fuquene Lake and Coello River watersheds are typical of the socio-environmental situation in the Andes (Ramirez and Cisneros 2006). Fuquene Lake watershed (Fuquene), which encompasses the valleys of Ubate and Chiquinquira in the state of Cundinamarca, Colombia (Figure 1 ), has an area of 187,200 ha and a population of 229,000 (Rubiano 2005), about 59% of which is rural (DANE 2005). The altitude ranges from 2300-3300 mas), with an annual rainfall range of700 and 1500 mm. Fuquene is located about two hours from the Colombian capital, Bogotá, on a good all-weather road. The area of influence of the watershed in eludes 17 municipalities in the states of Cundinamarca and Boyaca 4 • For the municipalities in the watershed, the 2003 life condition index, a measure ofwelfare, ranges between "very low" and "high" (Sarmiento et al, 2006), reflecting the socioeconomic heterogeneity in the zone. 4 The municipalities that belong to the Fuquene watershed are Cannen de Carupa, Ubate, Tusa, Sutatausa, Cucunuba, Suesca, ViJiapinzon, Lenguazaque, Gacheta, Fuquene, Susa y Simijaca in Cundinamarca and San Miguel de Serna, Raquira, Caldas, Chiquinquira y Saboya in Boyaca. 120 ·-Fuquene Wát"'hed = + + '\ ·-· ·· + ;.._._......;--......;'-1~------~.~------------+~-----------.~ 1 Figure l. Fuquene watershed, Colombia The largest land use in the watershed is pasture (59%), followed by agriculture (26%), forest (4%), paramo (2%) and lake (2%) (Rubiano et al. 2006). Land degradation is important, with 13,000 hectares classified as severely eroded and 40,000 as moderately eroded. In the past, major investments were made in soil conservation activities, however aside from stabilizing fragile areas, the impact of these investments on productivity has not yet been rigorously assessed. Conservation tillage has been widely promoted, however adoption has been limited until recently when payment for environmental services (PES)-type schemes have begin to promote it. The principal economic activities in the watershed are livestock, agriculture and mining. Livestock are mainly raised in the lower part ofthe watershed, in highly productive, high-input, medium- and Iarge-scale dairy farms. Crops are grown in the upper and middle parts of the watershed. Land ownership is generally by smallholders, however in the higher areas appropriate for potato cultivation, much of the land is rented out to large-scale producers who are willing and able to take the risks associated with a crop like potato whose production costs are high and whose market price is highly variable. Because of the potential profitability of potato, significant cultivation occurs in the paramos, which are ecologically fragile and play a key role in maintenance of ecosystem function, especially supply and regulation of water tlow (Rangel, 2006). Destruction of páramo is a concern for communities in the upper areas as well as downstream. Agricultura! production on these ecologically important areas is restricted by law, however there are problems with implementation. By law, municipalities are mandated to buy land in upper watersheds to protect water sources, and this is beginning to happen in sorne areas. 121 The middle part of the watershed is mainly occupied by small fanns producing grains and legumes. Decades ago, these were productive family fanns, however a combination of massive soil degradation, changing rainfall patterns, and the opening of markets to externa! competition significantly reduced the relative profitability of grain fanning. Sorne large-scale soil conservation and reforestation project were implemented, but with little effect. As a result, the middle part ofthe watershed has experienced large-scale out-migration ofyoung people, and those who stay depend mainly on off-fann work and transfer payments. The lake, located at the bottom of the watershed, is at the center of environmental controversy and is the issue driving change in the watershed2• In the past, government policy was to drain the lake to make more land available for agriculture (Mayorga 2003). The policy was reversed, however the surface area of the lake continues to decline due to of eutriphication and an explosion of water hyacinth caused by in part to agro-chemical run offfrom both crop and dairy fanns and untreated domestic waste water. The national government and the regional environmental authority have placed high priority on resolving the problems of Fuquene, and after large-scale flooding downstream of the lake in the spring of 2006 was attributed to reduced capacity of the lake, the President called for a Consejo Nacional de Política Económica y Social (Conpes), a special initiative to address an urgent problem of national significance. (DNP, 2007)0ne option is to focus on a technical solution such as dredging the lake, however this is likely to be very costly and unsustainable unless the underlying causes of watershed degradation are addressed. Addressing them is complicated by the fact that any policy changes could have implications for fann profitability and, by extension, land values. Cities downstream of the lake rely on it for drinking water supplies, however they have not yet perceived the threat and mobilized in response to it. Mining is also an important economic activity in sorne municipalities in the Fuquene watershed, and has increased in recent years due to the record high prices of coa) (team Scales Projec, 2005a, 2005b). The industry consists of both large-scale, regulated mines and small, informal operations, the latter having increased dramatically with the increases in price. Mining is an important source of income and employment and draws migrants from other regions; however it is also a majar contributor to water pollution (team Scales Projec, 2005a, 2005b). The environmental authority for the Fuquene watershed, the Corporacion Autonoma Regional de Cundinamarca (CAR)3 ,is in charge of allocating water permits and collecting water charges, and has participated in conservation activities. However, there is widespread discontent with the fact that the CAR has yet to develop and implementa management plan that addresses the issues facing the watershed. Local municipal govemments have sorne responsibility for resolving water conflicts and for undertaking conservation activities. While sorne are more active than others, they are limited in what they can achieve given their purely local scope. Coello River Watershed The Coello River watershed, located the state ofTolima in the central Andean Cordillera (Figure 2)4 has an area of 190,000 ha, ranging from 280 to 5300 mas l. Annual rainfall ranges from below 1 000 mm to more than 3970mm. The watershed includes ecosystems ranging from dry forest to páramo to snow-capped peaks, and is home to national parks and private reserves (team Scales Projec, 2005a, 2005b). The population ofthe watershed was 622,395 in 2005, including the city of fbagué (pop. 425,770). Only 16% of the population is rural; excluding the municipality of !bague, urbanization rates are still above 50%. The life condition index for municipalities in the Coello watershed ranges from medium low to medium high, a 2 See http://www.livinglakes.org/fuguene/. 3 See http://www.car.gov.co 4 The municipalities that make up the Coello River watershed are lbague, San Luis, Rovira, Cajamarca - Anaime, Espinal, Flandes, Valle del San Juan y Coello. 122 slightly narrower range than for Fuquene. While indicators are generally higher in !bague, in the rural municipalities there are problems with basic coverage of primary and secondary schools (Sarmiento el al, 2005). - - 1 - ..... -- 1 - r-- - \ _ - - ' -· - ' - - ------ Figure 2. Coello watershed, Colombia Principal economic activities in Coello include agriculture and livestock. The upper part ofthe watershed is mainly forested, however land there is increasingly being converted for livestock, coffee and horticultura! crops. In the middle altitude areas, sugar cane and fruit trees are common, and the lower part of the watershed includes 30,000 ha of large-scale, irrigated rice, cotton, and sorghum. Landownership is highly unequal (team Scales Projec, 2005a, 2005b), however small-scale agriculture is more viable in Coello than in the middle part of Fuquene because the land is less degraded. The Panamerican highway passes through the watershed, generating economic activity but at a cost of soil erosion and air pollution. In the upper parts of the watershed, there is presence of actors in Colombia's armed conflict, something which is not true in Fuquene. The security situation is much more serious in Coello, and there is significant interna! displacement. Water has not traditionally been scarce in Coello, however there is growing awareness that inappropriate land use in the upper watershed combined with growing demand for irrigation, domestic water and hydroelectric power in the lower areas are rapidly leading to a situation that is not sustainable. Water quality is also an issue as contamination is increasing due to agrochemical use, and domestic and industrial waste. 123 The environmental authority in Coello is the Corporacion Autonoma de Tolima (CorTolima (http://www.cortolima.gov.co), and as in Fuquene the local govemments also have important responsibilities. There is a wider range of actors in Coello than in Fuquene, including environmental NGOs, universities, the irrigation district, and agricuJtural producer associations. Poverty dynamics and the stages of progress (SOP) methodology While national-level poverty rates are often slow to change, for many of the rural poor, poverty is not a static situation. It changes according to various influences such as seasonality, climate variability, household-level shocks (such as illness and death) and lifecycle changes, and public policies. In addition, the group of poor people is itself constantly changing, individuals and households can escape from poverty but they can also descend further into poverty. There is therefore a need to track the movement into and out of poverty and to explain it. A longitudinal analysis of household welfare can foster a better understanding of the conditions that keep people in poverty and those that move them out in order to identify general pattems and to assist policy targeting (eg Sen 2003, Barrett, Carter and Little 2006). lt provides us with better insights into the processes that lead to pattems of disadvantage and inequality, but also different ways by which the poor may improve their welfare. In both cases, public policy can be tailored to include protection for the most vulnerable without pulling back those that are escaping. The stages of progress (SOP) methodology (http://www.pubpol.duke.edu/krishnalmethods.htm) were developed to assess the dynamics of poverty and the causes behind them. SOP is a participatory methodology that relies on community defmition of poverty at a household sea le. The poverty leve! of each household in the community is assessed, and explanations sought of the causes behind changes in poverty leve) over time. The method takes its name from the stages or steps that each household passes through as it makes its way out of poverty. To defme the stages, the group must first cometo agreement on a defmition of the "poorest family in the community." Once this is done, the group successively answers the question "What would this family do with a little more money?" until they reach the point at which the household is considered prosperous. Because they are defmed locally and in reference toa particular poor family, they vary by community and reflect the specific conditions and values ofthe community. Once the stages are identified, the group then assigns each family in the community- based on a census which must be obtained or constructed- to the stage where they currently are and the stage where they were at sorne point in the past, usually 1 O, 20 or 25 years ago. Families are then categorized as follow: A - Poor in the past, poor now B - Poor in the past, not poor now C - Not poor in the past, poor now O- Not poor in the past, not poor now For a randomly-selected sub-sample of families, the community then identifies the reasons for any change in stage. The final stage in the methodology is to conduct follow-up interview with a sample of families to confirm the results ofthe community analysis and to gather more information on specific issues. In the case of this study, interviews included questions on water use, conflicts, and management at the household and community scale. Using this methodology, studies in India, Kenya and Uganda reveal that factors associated with decline are distinct from those associated with escape. Generally, decline is associated with ill-health and high cost of healthcare, while escape from poverty is most commonly associated with income diversification through trade and employment (Krishna, 2006, 2004; Krishna et al. 2006; Krishna et al. 2004). However there are variations. In Rajasthan, high-interest prívate debt, and large social and customary expenses constitute major additional reasons for households declining into poverty. In Andhra Pradesh, drought, membership in a scheduled tribe and large family size were additional factors that contributed to decline into poverty. 124 Irrigation and access to non-traditional crops were further reasons associated with escape from poverty in Andhra Pradesh. While land and the uneconomic subdivisions are seen as a primary concern among declining households, access to land was found to be critica) for those that escaped poverty in Uganda. A quantitative and qualitative analysis of poverty across two time periods in Uganda identified asset ownership and access as major factors that influence poverty transitions and persistence. Education, land and cattle are the primary assets identified, while increasing household size and high dependency rates were seen as key factors in decline. Non-agricultura! activities provide an important escape route in rural areas, however, this is determined by levels ofhuman capital. Data and results from two Colombian watersheds For this study, the SOP methodology was applied in 13 communities (veredas) in six municipalities in the Fuquene watershed, and ten communities in five municipalities in the Coello watershed during the frrst half of 2005 (Table 1 ). Sites were selected in the upper, middle, and lower parts of the watershed, on the basis of prevalen ce of poverty and the expected intensity of water conflicts. Si te selection was based on available secondary data and on extensive interviews with key informants in the region. Information gathered in each community consisted of quantitative data from the SOP methodology-including movement in and out of poverty and their main causes-as well as qualitative data from interviews with households and key informants, and from observations by project staff in the field. Table l. Sites where the SOP methodology was applied S tate Municipality Community # ofhouseholds 1 Tausa Ladera Grande 53 Rasgata Bajo 41 Chipaquin 94 2Sutatausa Palacio 59 PeHas de Cajon 69 Gacha 128 Fuquene 3 Guacheta La Isla 92 La Puntica 90 Centro y Guata 182 4Fuquene Chinzaque 39 Nemoga 119 5Cucunuba Chapala 86 6 Carmen de Carupa Apartadero 43 La Leona-APACRAQ 1 Cajarnarca El Rosal La Alsalcia Minidistrito La Leona Coello COOCRA Coello 21bague San Cristoba1- Honduras 3 Espinal Dindalito 60 4 Coello-COCORA Potrerillo Chaguala Adentro 5Rovira LaOcera 125 The importance of water in the definition of poverty and stages of progress The poorest families in the communities were identified as landless day laborers who lacked quality housing, health care and other services, and who were unable to send their children to school. Non-material dimensions of poverty were also mentioned. In both watersheds, non-participation in community activities was also considered to be an indicator of poverty. Half of the communities in Coello included this in their defirution of the poorest family. In Fuquene, participation in community activities was considered to be a component of well being. The number of stages that communities defined ranged from 7 to 24, and the number of stages below the poverty line ranged from 3 to 1 O. As expected, shortages of the basic necessities such as food, education, clothing and housing were the most common early stages in nearly all communities. As household welfare increased, the items mentioned in the stages began to diverge with sorne communities focusing more on agriculture-related investments, others on services and others on durable goods (Table 2). Table 2. Frequency of categories of items that appear as stages below the poverty line, in order of importance. DescriEtion Order Frequenc~ Food 1 20 Education 2 18 Clothing 3 12 Housing 4 15 Small anima1s 5 14 Land 6 8 Services 7 8 Appliances 8 8 Health 9 4 Crops 10 4 Other 11 2 Transportation 12 2 S a vi ngs/In vestment 13 1 Recreation 14 2 Sorne communities were much more demanding than others in terms of what a household must have in order not to be considered poor. Often the communities that were better off on the basis of objective well- being measures were more demanding. For example, sorne communities required only food, education and shelter, while in others households were considered poor until they had land, television sets or the means to engage in recreational travel. This led to the more demanding communities being classified by SOP as having higher levels of poverty than other communities, which by objective measures were likely to be poorer5. As such, these results will not necessarily be consistent with national poverty statistics, nor will they be comparable across communities. lt is important to keep in mind that the purpose of a subjective poverty assessment such as the SOP is not to identifY uniform indicators, but rather to classifY families in a way that local people understand and that will allow them to carry out the analysis of the dynamics of poverty. The fact that the poverty measures may not be consistent does not invalidate the results regarding livelihood strategies that are associated with poverty and/or progress . .s For a detailed analysis ofthe SOP poverty lines, their components and their relationship to other poverty measures, see Peralta et al. 2006. 126 Access to water is clearly an important element of well being since 23 of the 25 communities mentioned access to water as a stage, either directly or as part of sanitation services. In only one community (Chinzaque in Fuquene) was agricultura! use of water specifically mentioned. While acquiring water usually means getting a household connection to a potable water system, domestic water is often used for productive, income-generating activities as well. In Fuquene, for example, small-scale cheese processors must ha ve access to potable water in arder to get certification to sell their products. In eleven communities (44%), water was included in categories below the poverty line while in 56% it was either not mentioned or was in categories above the poverty line. One reason that water is in categories above the poverty line is that in many cases households have good access to water from natural sources such as wells or springs or from shared taps. Thus, a home connection is somewhat of a luxury. Where water is in categories above the poverty line, improving access would improve livelihoods but would not reduce poverty. In three communities (12%), water was the included in the stage just below the poverty line, which means that improving access could literally get households out of poverty. Ranked from most to least demanding, the three communities where water is the Jast stage befare getting out of poverty occupied 5th, 15th and 20th places. What this means in tenns of how many poor households could actually be helped by better access to water in these two watershed depends on how many still do not have it. According to the data, 13% ofhouseholds are without water in communities where access to water is included in categories below the poverty Jine. Of these 4% are at the limit where getting access to water would get them out of poverty. Of the remaining households, only 8% are without water however by community standards they are not considered poor. The remaining 78% ofhouseholds already have water. Thus it seems that improving household access to water would not be an effective way to address poverty in these watersheds. There are specific cases of households and communities that would benefit significantly from improved access to water, but in general this is not the case. As Table 2 suggests, interventions to improve access to food, education, clothing, housing, small animals or Iand would be better targeting towards helping the poor. Role of water in the dynamics of poverty According to the results of the categorization of families (Table 3) poverty declined in nearly all communities over the last 25 years. In 1980, roughly 70% of families in both watersheds lived in poverty. Bet\.veen 1980 and 2005, 30% of the families in Fuquene escaped poverty (Category B), while only 3 percent fell into poverty. In Coello, the results are even more dramatic; 59% of families got out of poverty while only 3 percent became poor. These results clearly show that people in rural communities perceived important advances in their quality of life in recent decades. Nonetheless, over 40% of families in Fuquene and 10% offamilies in Coello continued to be poor in 2005. 127 Table 3. Changes in poverty status from 1980 to 2005 (% of families per category). CatA CatB Cate CatO CatE Vereda N Poor- Poor- Not poor- Not poor- New poor not poor poor not poor arrival Fuquene Ladera Grande 53 66 8 17 4 6 Rasgata Bajo 41 24 22 2 5 46 Chipaquin 32 34 41 13 13 o Palacio 59 37 47 o 2 14 Pei'ias de Cajón 69 17 61 o o 22 Gacha 81 38 26 6 28 1 La Isla 92 40 30 3 25 1 La Puntita 90 39 32 o 4 24 Centro y Guata 82 90 1 o 1 7 Chinzaque 39 23 46 .., .) 28 o Nemogá 119 13 29 4 46 8 Chápala 86 83 13 o o 5 Apartadero 43 30 70 o o o TOTAL 886 42 30 3 14 JO Coe/lo Apacra 13 o 31 o 69 o El Rosal 13 o 77 8 15 o La Alsalcia 14 21 57 o 21 o Minidistrito La Leona 11 45 55 o o o Cocora 18 6 22 o 44 28 Dindalito 26 15 46 8 31 o San Cristóbal-Honduras 19 5 47 5 42 o Potrerillo 31 o 100 o o o Chaguala 14 29 57 7 7 o LaOcera 16 6 75 o 19 o TOTAL 175 11 59 .., .) 24 3 For each family in the survey, up to three causes were identified to explain the change in poverty category between 1980 and 2005. A total of 25 causes were identified (Table 4). Among the causes that were mentioned in first place for each family, the most important was off-fann employment (20%), followed by inheritance (17.2%), help from family and friends (9.4%), day labor (7.8%) and help from the govemment (7.5%). These results are consistent with diversification ofrurallivelihoods. 128 Table 4. Principal causes of change in poverty status(% of families) Mentioned Mentioned Mentioned Cause as first as second as third All cause cause cause (N=778) {n=36Q {n=284} {n=l33} Government help 7.5 4.9 14.3 7.7 Help offamily and friends 9.4 16.2 8.3 11.7 Unexpected loss 0.3 0.0 0.8 0.3 U nexpected benefit 0.3 0.0 0.0 0.1 Educationltraining 1.9 3.5 8.3 3.6 Off fann employment 20.0 20.1 8.3 17.0 Day labor 7.8 9.9 5.3 8.1 Small or low quality landholding 0.8 0.7 0.8 0.8 Credit 0.6 2.5 3.0 1.7 lllness/accident 3.6 1.8 2.3 2.7 Large family 1.4 0.7 0.8 l. O Small family 0.0 1.1 3.0 0.9 Newly established family 3.9 1.1 0.8 2.3 Agriculture 6.1 7.7 11.3 7.6 Livestock 1.7 3.2 8.3 3.3 Good money management 0.6 2.1 0.8 1.2 Bad habits 3.3 2.1 0.0 2.3 Legal or family problems 3.0 1.1 1.5 2.1 lnheritance 17.2 6.3 6.8 11.4 Savings/investment 3.3 10.6 9.8 7.1 Old age 0.8 0.4 0.0 0.5 Pension 4.4 2.1 2.3 3.2 Community work/collective action 0.6 1.4 3.0 1.3 Fishing 0.8 0.7 0.8 0.8 Migration 0.8 0.0 0.0 0.4 Total tOO 100 100 100 While not related to one's own farrn, off-farrn employment in these watersheds usually refers to steady work that is agriculture- or natural-resource-based. The dairy sector in Fuquene generated significant employment, ranging from administrators who manage farrns for absentee landlords to milkers, mainly women and often heads of households. Small and medium scale agro-enterprises based on production of value-added dairy products, such as cheese and yogurt, were al so common in Fuquene. The sale of services such as machinery rental to farrners and in the commercialization and/or transportation of agricultura) products such as fruit, coffee or livestock were common in both watersheds. Mining was also an important source of off-farrn employment in sorne municipalities in Fuquene as was the collection and sale of sand and other materials from the river in Coello. Among secondary causes, off-farrn jobs and family help continue to be important, however other causes such as agricultura! production and savings and investments also were mentioned. Govemment assistance and agriculture were the most important third causes. 129 The results are similar for Fuquene and Coello, with the important exception that agricultura! production was much more important in Coello than in Fuquene (Tables 5 and 6). ln Coello, agriculture is the most frequently mentioned first cause, followed by off-farrn labor and day labor as the second and third causes for changes in poverty status. Table 5. Principal causes of change in poverty status, Fuquene (% offamilies) Mentioned Mentioned Mentioned % ofall as first as second as third Causes causes cause cause cause mentioned {n=25Q {n= l 87) {n=71} Government help 6.4 4.8 7.0 5.9 Help of family and friends 8.4 15.0 9.9 11 .0 Unexpected loss 0.0 0.0 1.4 0.2 Education/training 1.2 2.1 4.2 2.0 Off-farm employment 23.9 23.5 11.3 22.0 Day labor 7.2 8.0 4.2 7.1 Small or low quality landholding 0.4 1.1 1.4 0.8 Credit 0.0 0.0 4.2 0.6 lllness/accident 4.0 1.6 4.2 3.1 Large family 1.6 0.0 1.4 1.0 Small family 0.0 1.6 4.2 1.2 Newly established farnily 5.6 1.6 1.4 3.5 Agriculture 2.4 4.3 5.6 3.5 Livestock 0.8 3.2 8.5 2.8 Good money management 0.4 1.6 0.0 0.8 Bad habits 3.6 2.7 0.0 2.8 Legal or farnily problems 2.0 0.5 0.0 1.2 lnheritance 18.7 9.1 11.3 14.1 Savings/investrnent 3.6 14.4 14.1 9.0 Old age 1.2 0.5 0.0 0.8 Pension 6.0 3.2 4.2 4.7 Community work/collective action 0.4 0.0 0.0 0.2 Fishing 1.2 1.1 1.4 1.2 Migration 1.2 0.0 0.6 Total 100.0 100.0 100.0 100.0 130 Table 6. Principal causes ofchange in poverty status, Coello (% offamilies). Mentioned Mentioned Mentioned % ofall as first as second as third Causes causes cause cause cause mentioned {n= 110} {n=97} {n=61} Government help 10.0 5.2 5.2 7.1 Help of fami ly and friends 11.8 18.6 18.6 15.8 Unexpected loss 0.9 0.0 0.0 0.4 Unexpected benefit 0.9 0.0 0.0 0.4 Education/training 3.6 6.2 6.2 5.1 Off-farm employment 10.9 12.4 12.4 11.8 Day labor 9. 1 13.4 13.4 11.6 Small or low quality landholding 1.8 0.0 0.0 0.7 Credit 1.8 7.2 7.2 5.0 Illness/accident 2.7 2. 1 2. 1 2.3 Large families 0.9 2.1 2.1 1.6 Agricu lture 14.5 14.4 14.4 14.5 Livestock 3.6 3.1 3.1 3.3 Good money management 0.9 3. 1 3.1 2.2 Bad habits 2.7 1.0 1.0 1.7 Legal or family problems 5.5 2.1 2.1 3.5 Inheritance 13.6 1.0 1.0 6.2 Savings/investment 2.7 4.1 4.1 3.6 Pension 0.9 0.0 0.0 0.4 Community worklcollective action 0.9 4. 1 4.1 2.8 Total 100.0 100.0 100.0 100.0 As expected, in the majority of cases specific causes are associated with eitber progress or poverty. Regressing cause dummy variables (!=cause was mentioned for the househo1d, O=no) on a dummy for whether the household was poor in 2005, we can see the contribution of each cause to poverty (Table 7). According to the results (Table 8) the probability of being poor in 2005 was reduced for families with stable jobs (-37%), agricultura) production6 (-35%), govemment help (-30%), pensions (-27%), help from farnily and friends (-23%), livestock (-23%), savings and investment (-22%), or inheritance (-17%). The probability of being poor in 2005 was increased by fam ily problems (+62%), health problems or accidents (+37%), or starting a new family (+25%). 6 In Coello, no agricultura! households were ranked as poor in 2005. 131 Table 7. Impact of major causes of change in poverty status on wbetber bousehold was poor in 2005 (n=359). Cause Coef. Std. Err. z P>lzl 95% Conf. lnterval Fúquene 0.8294 0.2606 3.18 0.001 0.3185 1.3402 Govemment help -0.7866 0.2522 -3.12 0.002 -1.2809 -0.2923 Help from family and friends -0.8534 0.2246 -3.8 o -1.2935 -0.4133 Education -1.2576 0.4380 -2.87 0.004 -2.1161 -0.3992 Off-farm employment -1 .2735 0.1983 -6.42 o -1.6621 -0.8849 Health/accident 0.9861 0.4466 2.21 0.027 0.1108 1.8615 New families 0.6794 0.4182 1.62 0.104 -0.1402 1.4990 Agriculture -1.8283 0.3797 -4.82 o -2.5725 -1 .0842 Livestock -1.0400 0.4285 -2.43 0.015 -1.8799 -0.2001 Family problems 1.7996 0.7036 2.56 0.011 0.4207 3.1786 lnheritance -0.5756 0.2077 -2.77 0.006 -0.9827 -0.1684 Savings and investment -0.8336 0.2399 -3.48 0.001 -1.3037 -0.3635 Pension -1.3996 0.4045 -3.46 0.001 -2.1925 -0.6067 Constant 0.3137 0.2703 1.16 0.246 -0.2162 0.8435 LR chi square (13) 205.8, log likelibood = -128.48344, pseudo R2 = 0.4447. Table 8. Influence of cause on probability of being poor Cause Offfarm employment Agriculture Government help Pension Education Help from family and friends Livestock Savings and investment lnheritance Fúquene New farnilies Health/accident Family problems Delta P(Poor2= 1) without cause (i) - P(Poor2= 1) with cause (i) 0.3650 0.3503 0.3041 0.2666 0.2569 0.2343 0.2310 0.2182 0.1673 -0.2368 ..0.2517 -0.3706 ..0.6243 Mean 0.6953 0.1662 0.2521 0.0776 0.3934 0.0582 0.0499 0.1634 0.0720 0.0443 0.2465 0.1773 0.0693 In spite of being an important cause, day labor did not significantly affect the probability of being poor in either watershed. Where families depend on day labor as a primary livelihood strategy (that is where it is given as their first cause), it seems to be associated with poverty. However, when day labor is a complementary livelihood option, it can contribute to progress.Accidents and illness were relatively rare arnong families in both watersheds, indicating tbat families are not as vulnerable as might be expected to externa) shocks. Ratber, their challenge is to find and take advantage of opportunities. In CoeJJo, failure to progress is due almost exclusively to bad habits or family problems7• This contrasts with SOP results from 7 In Coello, there were cases of families who had fled the zone dueto political violence, but, because they were no longer in the community, they were not considered in the analysis. 132 other countries, as mentioned in section 4. Though consistent with the relatively higher living standards in Colombia, such explanations should always interpreted with caution since they may constitute a superficial explanations for deeper problems that would not be recognized by the community. Lack of importance placed on illness suggests that indirect links between poverty and water quality via health are not important in these watersheds. While water was not mentioned specifically as a cause, severa! of the strategies mentioned have obvious links with the environmental issues of the watersheds. For example, dairy farms and mines are important sources of employment in Fuquene, but they also contribute to problems of water contamination. If measures to improve environmental outcomes were to reduce the profitability ofthese industries and reduce their demand for labor, poverty could worsen in Fuquene. A similar argument could be made for small- scale agriculture, especially in Coello. Degradation ofthe upper watersheds for potato cultivation and extensive ranching appears to present less of a poverty-environment tradeoff since the day labor that is generated either by potato cultivation or ranching does not contribute significantly to poverty alleviation the way that labor does in other sectors. Discussion These results suggests that in these two watershed, the indirect relationships between poverty and water via employment and income linkages may be more important than direct linkages via domestic supply. This is consistent with the diversification of rural livelihoods and the importance of off-farm income in poverty reduction. lnterventions to enhance domestic supply may have big impacts in a few specific communities, but would not generally contribute much to poverty alleviation. Interventions that would reduce employment in industries like dairying or mining in Fuquene or profitability in small scale agriculture in Coello, could have significant impacts on poverty, since these have been important pathways out ofpoverty over the past 25 years. lnterventions aimed at reducing the destructive impacts of agriculture and ranching in the upper watersheds would appear to be a better options for irnproving the environment without adverse effects on poverty since the day labor generated in potato cultivation is not associated with poverty reduction. In Fuquene, a payment for environmental services scheme based on promoting conservation agriculture among potato farrners is an example of such an intervention (http://www.condesan.org/Andeanlprojects.htm). Colombian legislation allows for stakeholder participation in watershed management decisions. While it is increasingly recognized that stakeholder participation is an important part of integrated water resources management (IWRM), effective participation presumes a good understanding of the issues, especially the socio-economic and biophysicallinkages within watershed systems. Results from studies such as these can contribute to improving the community knowledge base, and therefore to helping stakeholder groups better identify the issues that are important to them, and their potential allies in reaching their goals. Such interests may cross the deep sectoral, socio-economic and cultural divides that exist in Andean watersheds. Even though these results suggests that the poor are not a homogenous block of people whose interests are necessarily opposed to those of better-off groups, they do show that one thing the poor do have in common is that they tend not to participate in community Jevel processes. Participation is considered to be a component of wellbeing, and in many communities the poor are identified as being those who do not participate. Therefore, in addition to technical and institutional interventions to change Jand use, efforts should also focus to build wiUingness and capacity of the poor to participate in community activities. Ln both Fuquene and Coello, local NGOs are Jeading such a process. Using the results from this study and those from biophysical studies could be used to build awareness of the upstream-downstream linkages and their implications for equitable and sustainable watershed management. 133 References Barrett, C., Bezuneh, M., and Aboud, A. (2000). The Response of Income Diversification to Macro and Micro Policy Shocks in Cote d ' Ivoire and Kenya. Barrett, C.B. and Reardon, T. (2000). Asset, Activity, and lncome Diversifica/ion Among African Agricu/turalists: Some Practica/ /ssues. Project report to USAID BASIS CRSP, University of Wisconsin-Madison Land Tenure Center. Barrett, C., Reardon, T.and Webb, P. (2001b) Nonfarm income diversification and household livelihood strategies in rural Africa: Concepts, dynamics, and policy implications. Food Po/icy 26:315-31 Barrett, C. B., Bezuneh, M. and Aboud, A. (200 1 ). Income diversification, poverty traps and policy shocks in Cote d' Ivoire and Kenya. Food Po/icy. 26:367-384. Beck, T. and Nesmith, C. (2001). Building on poor people's capacities: The case of cornmon property resources in India and West Africa. Wor/d Development 29:119-133 . Bezemer, D.J. and Lerman, Z. (2004). Rural livelihaods in Armenia. Post-communist &onomies 16:333-348. Block S. and Webb P. (2001). The dynamics of livelihood diversification in post-famine Ethiopia. Food Po/icy 26:333-350. Bryceson, D. F . and Jama!, V. (eds.) 1997. Farewe/1 to Farms: De-agrarianisation and Emp/oyment in Africa. Ashgate, Aldershot. Canagarajah, S., Newman, C. and Bhattamishra, R. (2001). Non-farm income, gender, and inequality: evidence from rural Ghana and Uganda. Food Po/icy 26:405-420. Cavendish, W.(2000). Empirical regularities in the poverty-environment relationship of rural households: Evidence from Zimbabwe. Mimeo. Chambers (1995). See page 2. Chambers, R. (1995). Poverty and livelihoods: whose reality counts?. Environment and Urbanization, Vol. 7, No. 1, 173-204 (1995) de Haan, L. and Zoomers, A. (2005). Exploring the frontier of livelihoods research. Developmenl and Change 36:27-47. DANE (2005). Population projections from 1993 to 2050. URL:http://www.dane.gov.co Dei, G. (1992). A forest beyond the trees: Tree cutting in rural Ghana. Human &ology 20:57-88. DNP (2007). CONPES aprobó plan de salvamento para la laguna de Fúquene. Boletín de Prensa, 18 de enero de 2007. URL: http://www.dnp.gov.co/archivos/documentos/GCRP _ Comunicados_Prensa/ene _ 18 _ 07 _ fuquene.pdf Ellis, F. (2000). The determinants of rural livelihood diversification in developing countries. Journa/ of Agricultura/ Economics 51:289-302. Fereres, E. and Kassam, A. (2003). Water and the CGIAR: A strategic framework. Water /nternationa/28:122-129 Fisher, M. (2004). Household welfare and forest dependence in Southem Malawi. Environment and Deve/opment &onomics 9:135-154. Gladwin, C.H., Thomson, A.M., Peterson, J.S. and Anderson, A.S. (2001 ). Addressing food security in Africa via multiple livelihood strategies ofwomen farmers. Food Policy 26:177-207. Grewel, S.S., Dogra, A. S. and Jain, T.C. (200 1 ). Poverty alleviation and resource conservation through integrated watershed management in a fragile foothill ecosystem. ln: Stott, D.E., Mohtar, R.H. and Steinhardt, G.C. (eds). Sustaining the Global Farm. Selected papers from the 10th lntemational Soil Conservation Organization Meeting held 24-29 May, 1999 at Purdue University and the USDA-ARS National Soil Erosion Research Laboratory. URL: http://topsoi l.nserl. purdue.edu/nserlweb/isco99/pd ti'iscod isc/Sustaining«>lo20the%20G lobal%20 F armiP 197- Grewel.pdf Jodha, N.S. (1986) Common property resources and the rural poor in dry regions of lndia. &onomic and Política/ Weekly. 21(27) : 169-181. Jodha, N.S. (1995). Studying cornmon property resources: Biography of a research project. &onomic and Po/itical Week/y 30(11):556-559. Kerr, J. and Chung, K. (200 1 ) . Evaluating watershed management projects. Water Po/icy 3:537-554. Krishna, A. (2004). Escaping poverty and becorning poor: Who gains, who loses, and why?, Wor/d Development 32:121- 36. Krishna, A., Kristjanson, P., Radeny, M. and Nindo, W. (2004a). Escaping poverty and becoming poor in 20 Kenyan villages. Journal of Human Deve/opment 5:21 1-26. Krishna, A ., Lumonya, D., Markiewicz, M, Kafuko, A, Wegoye, J. and Mugumya, F (2004b). Escaping poverty and becoming poor in 36 villages of central and westem Uganda. Working Paper. URL: http://www.pubpol.duke.edu/krishna/documentsl. 134 Krishna, A., Kapila, M., Porwal, M. and Singh, Y. (2005a). Why growth is not enough: Household poverty dynamics in Northeast Gujarat, Inclia. Journal of Development Studies 41:1163-192. Krishna, A., Kristjanson, P., Kuan, J., Quilca, G., Radeny, M. and Sánchez-Urrelo, A. (2005b). Fixing the hole at the bottom of the bucket: Household poverty dynamics in forty communities of the Peruvian Andes. Working Paper. URL: http://www.pubpol.duke.edu/k:rishna/documents/. Lanjouw, J., Quizon, P. and Sparrow, R. (2001). Non-agricultura! eamings in peri-urban areas ofTanzania: Evidence from household survey data. Food Policy 26:385-403 Mayorga, F. (2003). Concepto sobre la propiedad de la Laguna de Fuquene solicitado por la CAR. Bogotá. 36 pp. Moreral , M.C. and Gladwin C.H. (2006). Does off-farm work discourage soil conservation? Incentives and disincentives throughout two Honduran hillside communities. Human Ecology 34:355-378. Orr and Mwale (2001). See page 3. Orr A.; Mwale 8 .(2001). Adapting to Adjustment: Smallholder Livelihood Strategies in Southem Malawi. World Development, Volume 29, Number 8, August 2001 , pp. 1325- 1 343(19). Perez, C. and Tschinkel, H. (2003). lmproving watershed management in developing countries: A framework for prioritizing sites and practices. AgREN Network Paper No. 129, July 2003. ODI, London. 15pp. Ramírez, M. C., and Cisneros, H. (eds) (2006). Andean System of Basins: Watershed Profiles - Enhancing Agricultura/ Water Productivity Through Strategic Research. Technical Report No. 1, Challenge Program on Water and Food. P.O. Box 2075, Colombo Sri Lanka. Rangel, O. (2006). Presentation CORPOCUENCAS Intemational Seminar: Management and Conservation of Andean Ecosystems. Cali, 4-6 December, 2006. Reardon, T. ( 1997). Using evidence of household income diversification to inform study of the rural nonfarm labor market in Africa. World Development 25:735-747. Reardon, T., Berdegue, J. and Escobar, G. (200 1 ) . Rural nonfarm employment and in comes in Latín America: Overview and policy implications. World Development 29:395-409. Reddy, S.R.C. and Chakravarty ,S.P. (1999). Forest dependence and income distribution in a subsistence economy: Evidence from India. World Development21:l141-l 149. Rodríguez, H. and Rubiano, J. (2005). Aspectos bioflsicos y socio- económicos de la cuenca del río Coello en el departamento del Tolima, Colombia. Corporación Semillas de Agua: Proyecto de Conservación y Desarrollo en la Cuenca del Río Anaime, Cajamarca. Rubiano, J. , Quintero, M., Estrada. R. and Moreno, A. (2006). mu1tiscale ana1ysis for promoting integrated watershed management. Water lnternational31: (in press). Swallow, B.M., Garrity, O.P. and van Noordwijk, M. (200 1 ). The effects of scales, flows and fi1ters on property rights and collective action in watershed management. Water Policy 3:457-474. Swallow, J. and Meinzen-Dick (200 1 ). Working with people for watershed management. Water Policy 3:449-455. Swallow, B., Johnson, N., Meinzen-Dick, R. and Knox, A. (2006). The challenges of inclusive cross-scale collective action in watersheds. Water International31: (In press) Zoomers, A. ( 1999). Linking Livelihood Strategies lo Development. KIT -CEO LA, Amsterdam. 108 pp. Zoomers, A. ( 1999) Livelihood Strategies and Development lnterventions in the Southem Andes of Bolivia: Contrasting Yiews on Development, Cuadernos del Cedla no. 3 (pdf format). Shivakoti, G.P. and Thapa, S.B. (2005). Farmers' perceptions of participation and institutional effectiveness in the management ofmid-hill watersheds in Nepal Environment and Development Economics 10:665-687. Tumer, B., Kasperson, R., Matson, P., McCarthy, J., Corell, R., Christensen, L., Eckley, N., Kasperson, J., Luers, A., Martello, M., Polsky, C., Pulsipher, A. and Shiller, A. (2003). A framework for vulnerability analysis in sustainability science. Proceedings ofthe National Academy ofSciences 100:8074-8079. Wood, G. (2003). Staying secure, staying poor: the "Faustian bargain." Wor/d Development 31:455-471. Sarmiento, A.; Cifuentes, A.; González, C. and Coronado, J. (eds.) (2006). "Los municipios colombianos hacia los objetivos de desarrollo del milenio: salud, educación y reducción de la pobreza". Bogotá, DNP, PDH, UNDP y GTZ. Team Scales Project (2005a). Documentos de losTalleres Realizados en las Veredas de las Cuencas de la Laguna de Fúquene y del Río Coel/o. Mimeo. Team Scales Project (2005b). Resumen de Características de las Veredas de las Cuencas de la Laguna de Fúquene y del Río Coe/lo. Mimeo. 135 Dynamics and definition of poverty in the Colombian Andes: Participative vs. objective approaches Alexandra Peralta0 , James A. García C.b and Nancy Johnson° 8Challenge Programfor Water and Food (CPWF) - Centro Internacional de Agricultura Tropical C/AT, Cali, Colombia bCentro Internacional de Agricultura Tropical CIAT, Cali, Colombia Abstract The objective of this study is to examine the consistency of results of a participatory poverty assessment methodology applied in two Colombian watersheds with those from more objective approaches. The results suggest that there is a set of elements that are considered basic to both types of poverty assessment; however at the same time there are others that depend on household and community preferences. Moreover, the results indicate that the concept of poverty is context-specific that is a household that is considered poor in one community m ay not be considered poor in another. The results of the participatory methodology are useful to identifY who the poor are, wby they are poor, and provide a better understanding ofthe nature and dynamics of poverty. However, it may not be appropriate to generalize on the basis of the results of such methodologies since they may differ in both nature and magnitude from the results off objective poverty measurements. Keywords: Poverty, rural analysis, participatory methods, Colombia. Introduction Although reduction of poverty is one of the major challenges of the Millennium Development Goals, there is still no clear defmition of what poverty actually is. The standard definitions of US$1 or US$2 per day, unsatisfied basic needs (UBN), life-condition index (LCI) and so on, do not recognize that poverty is complex and includes both material and non-material dimensions. In response to these limitations, participatory methods have been developed that allow local perceptions to be included in definitions of poverty. These methods, usually applied at the local scale, provide a leve) of detail that goes beyond the objective quantitative measures to show the dynamics of poverty. Therefore, they can be very useful in the design of interventions to reduce the problems of exclusion and poverty. However, because they are based on local perceptions, the results of participatory assessments from different places are not necessarily comparable. It may therefore not be appropriate to use information from such assessments to design regional interventions. Similarly, the generalization of conclusions based on such local results may not be appropriate. This document attempts to validate the interna) and externa! consistency of data on poverty obtained using a participatory method based on local community perceptions for two watersheds in the Colombian Andes. The principal research questions were: Are the definitions and results obtained consistent with objective meas u res of poverty? Are the results consistent with the condition ofthe study communities? 136 The results seek to contribute to a better understanding not only of what poverty is, but also how to use and interpret different types of poverty data. It was of particular interest to identizy the conditions under which different types of infonnation on poverty do or do not give consistent results. The next section summarizes sorne of the issues raised by different types of poverty measurements, and considers the methods currently used in Colombia. We then describe the study watersheds, and go on to present the findings of the poverty analysis. We also examine the community-level data on which these findings were based, specifically the community-level definitions of poverty and poverty line (described below). We then summarize and discuss our conclusions. Empirical aspects and measurements of poverty in Colombia Objective and subjective approaches lo poverty, the methods used and their advantages and disadvantages Poverty is a multi-dimensional phenomena as the wide range of concepts in the literature illustrates. In concept, poverty can be objective or subjective, for example the need for people to have adequate diets contrasted with peoples' preference for specific foods independent of their nutritional value. lt can be absolute or relative, based on the lack of income or of earning capacity. lt can be chronic as a consequence of structural problem in an economy or temporary when caused by a short-tenn economic malfunction . Finally, poverty can be related to inequality, vulnerability and exclusion within a society. The use of these different concepts detennines the way that poverty is measured and also influences the nature and implementation of policies and programs to alleviate it (Lok-Dessallien 1995). Objective meas u res define poverty trom the perspective of a group of nonnative criteria that define what is required to overcome poverty. In general, objective assessment emphasizes quantitative data, which are simple to record and to compare. In contrast, subjective measures consider poverty fi'om the point of view of individual preferences and the value individuals place on goods and services. Subjective assessment mainly uses qualitative criteria to describe the intrinsic characteristics of what is being measured, although it is necessary to translate them into sorne quantitative fonn to be able to aggregate them (Lok-Dessallien 1995). Objective measures of poverty can be either non-monetary or monetary, and, in tu m, the latter can be absolute or relative (Figure 1 ). SUBJECTIVE - MEA SURES POVERTY NON- ~ - MONETARY MEA SURES MEA SURES OBJECTIVE RELATIVE - MEA SURES t-- - MEA SURES - MONETARY 1--- MEASURES ABSOLUTE - MEASURES Figure l . Diverse measurements ofpoverty. Source: MERPD (2006). 137 Now we describe the characteristics of the methodologies commonly u sed for the study of poverty, and discuss their advantages and disadvantages. In doing this, we adapted the methodologies developed by the Mission for the Design of a Strategy for the Reduction of Poverty and Inequality (MDERPD 2006). The commonly-used indicators of subsistence are US$1 income per person per day (PPD) for indigent peoples. US$2 PPD is used to define poverty in purchasing power parity (PPP) in Colombia. "although somehow illustrative for effects of international comparisons, [these indicators] are not conceptually bound to the idea of poverty as a lack of basic necessities, this is beca use the val u es of the poverty line do not represent a specific basket of goods and services that a society considers indispensable to live a worthy life, according to its culture, geographical conditions and socioeconomic level. Also, clearly, one or two dollars PPP is not enough to cover the necessities that a country of medium income [such as] Colombia could consideras basic" (MDERPD, 2006, p. 21 ). 1 The measure of absolute poverty called the poverty line is based on the value of a mínimum basket of representative goods such as food, clothing, transport, etc. For any reference population this corresponds to the poorest 25%. The monetary value of the poverty line changes as the contents of the basket and/or the sources of information are modified (MDERPD 2006). The most commonly used non-monetary objective indicators are the human development index (HDI), the index of unmet basic necessities (UBN) and the condition of life index (CLI). Another index used in Colombia is the system for the selection of beneficiaries of social programs (el sistema de seleccion de beneficiarios para programas socials) SISBEN, which we shall not consider further because of its similarity to the CLI. The HDI is not exactly an indicator of poverty because in addition to a per capita income index it also includes an index of life expectancy at birth and a compound education index. Actually, the concept of human development is wider and more complex than captured by the HDI because it only three of the multiplicity of elements of human development. For example, participation in society was highlighted in the Human Development Report 2004 as an important part of human development~ however this variable is not included in the HDI (MDERPD 2006). "The indicator of unsatisfied basic necess1t1es [UBN] is sometimes [u sed] as an alternative measurement of poverty~ it considers as poor those households or people that have at least one unsatisfied necessity out of the five defined (poverty for [UBN]) and as in misery those households or people that have at least two unsatisfied necessities (misery for [UBN])" (MDERPD 2006, p. 15). The criteria in the UBN are inadequate housing, housing without services, overcrowding in a household, non-attendance at school and high economic dependence. The UBN determines which ofthe criteria are not met and ofthose which is the most common in households in a given location. However, it hides how many people are in poverty and misery, because it uses the household as the unit of evaluation, ignoring the number of people that comprise it. Moreover, important aspects, such as nutrition, are not included in the UBN (MDERPD 2006). Finally, the CLI, developed by the Social Mission of the National Department of Planning (NDP) of the Colombian govemment "tries to give a more integrated and more informative coverage of the satisfaction of basic needs and quality of Iife than the UBN." (MDERPD, 2006, p. 17). lt incorporates indicators of physical goods, present and potential human capital, and composition of the household, each variable included in these indicators has an assigned weight. The CLI is an indicator that is not used to determine 1 In 2004 U$1 a day PPP was equivalent to 24,137 Colombian pesos (COP) a month, while U$2 a day PPP to COP48,274 a month (MDERPD, 2006, p.20). The minimum legal monthly salary for 2004 was COP324,000. 138 which homes are poor and which are not, but it allows comparisons to be made at a given time and place. However, the weights that it assigns to the variables contradict what the society might consider desirable, for example, for garbage disposal it assigns a score of 2.59 to people who throw their garbage in a river and 1.59 to those who burn or bury it (MDERPD, 2006). The subjective measurements of poverty that are used in Colombia are obtained from the answers to the questions of the survey of quality of life (SQL) about the perception of poverty by households. The following is a sample of such questions: l. Do yo u consider yourself poor? 2. Presently, how the life conditions ofyour household are? 3. What do you consider should be the monthly mínimum income that your household requires to satisty its necessities appropriately? 4. What leve! of income would you classity as excellent? Bad? These questions are not very clear about what is being referred to when speaking of poor in 1, of life conditions in 2 or ofminimum income required by the household in 3. For that reason the results are not the most appropriate to determine strategies for programs of poverty relief. For example, in accordance with the calculations of the MERPD based on the question 1 in SQL 2003, 36% of the households in the fifth quintile in urban areas are considered poor (MDERPD, 2006). There are participatory methodologies available, which have been used in Colombia and in other developing countries to generate profiles of poverty based on the local perceptions of communities. The World Development Report 2000/2001 gathered experiences of qualitative participatory studies on poverty carried out in 50 countries. A comparative study of 23 countries was made, with the objective of including the perceptions of the poor in political proposals for poverty reduction. As a result of this investigation, the World Bank published a methodological guide that created participatory profiles ofwell-being to be used in work groups with representative members ofthe communities. For Colombia, we have the development of regional pro files of poverty (Ranvborg 1999) that determine the levels of well-being of the households in a given community. Workshops convened of representative members of the community classity households into the different levels of well-being. In this way average levels ofwell-being of all households in the community are also determined2• We applied Krishna's (2002) methodology in the present study. This methodology uses community participation to determine qualitative poverty lines and defines stages of progress within the community. The objective is to determine what stage households were in 25 years ago and what stage are they in now. This provides information on the dynamics of poverty and what were the causes of any changes. The method has been applied in India, Kenya, Uganda, Peru and the United States (Krishna 2004a, Krishna et al. 2004b, Krishna et al. 2004c, Krishna et al. 2005a, Krishna et al. 2005b, Krishna et al. 2006). This methodology attempts to understand the dynamics of poverty at a disaggregated leve!, in both the communities and households. Although economic growth stagnated during the last decades in sorne developing countries, there were interna! dynamics that allowed the poor to improve their situation or caused it to worsen. The knowledge of how sorne people escaped from poverty can be useful to help those did not (Krishna 2004a, Krishna et al. 2004b)3. Commonly-used objective measurements of poverty apply a series of normative criteria to determine which are the best for people. Studies carried out based on this information generally use nations as the unit of 2 For more detailed information see http://www l. worldbank.orglprem/poverty/voices/index.htm 3 For more detailed information see http://www.pubpol.duke.edu/krishna/index.html 139 measurement. They require the availability of data collected o ver a substantial period of time and moreover that the data collected at different times are both comparable and complete. If the data do not fulfil these requirements, the studies will give questionable results and simply should not be carried out. In addition, this type of study lacks those elements that allow understanding of people' s perceptions about poverty and the strategies that they develop to confront it. Moreover, they are expensive in that they require that time- series data are available. For all these reasons, the poverty dynamics method is a superior alternative (Krishna 2004a, Krishna et al. 2004b ). Stages of progress methodology and poverty line The stages ofprogress methodology (SOP) was applied in 23 communities (veredas) ofthe municipalities ofthe watersheds ofFúquene Lake in Cundinamarca (Fúquene) and in the Coello River (Coello) in Tolima between March and July of 2005. Tbe communities were selected in consultation with the mayors of the municipalities and were based on inforrnation of the relevant indicators in the municipal Outlines of Territorial Classification (EOT). The sites chosen were poor villages with water sources, near the paramo and in the hillsides. Workshops were held with the participation of 886 households Fúquene and 175 in Coello. The nurnbers of participants in Coello was limited because of security problems. Although the sample in Coello was not representative from a statistical viewpoint, we believe that it was essentially representative ofthe rural villages in the watershed. Prior to the workshops, we agreed the concept of what constituted a household and what constituted a community. A household is the group of people that shares food "from the same pot or kitchen." A community is a compound group of 25 to 60 households with similar geographical and environmental conditions as well as similar access to markets. Furtherrnore, the households in the community must know each other and bave remained in the community for at least the last 25 years. The research tearn made an initial visit to each community and had discussions with local officials and other key people such as the presidents of meetings of communal action and community leaders. The characteristics of each community were described, including history, socio-economic characteristics, infrastructure and migratory movements of households. Where immigrants were more than 25% of population, the community was deleted from consideration. Next the steps that were undertaken were (Tearn ofthe project Escalas CPWF 2005a): l. Convene representative groups: A representative group was convened in each community, including members of different households, members of the poorest households and those that were not land owners or who were not entitled to use land. The invitees were 50 or more years old and had lived for more than 25 years in the community. In addition, youths born in the community and sorne representatives of the local authorities were also invited. The Tearn sought to have gender equality, trying to have equal numbers ofmen and women. 11. Explain the objectives: Members of the Team and of the institutions involved were introduced to the communities and explained the objectives and the process of inforrnation gathering. To avoid any possibility of encouraging false expectations among the participants, the Tearn emphasized that this was entirely a scientific exercise. The Tearn also explained clearly what the timetable ofthe workshop would be so that people would stay until the end ofthe process. 111. Collectively define stages of progress and determine the poverty lines: The objective was to arrive at a common understanding of poverty within each community and what it means to be considered poor. To avoid any prejudicial atmosphere, alternative words were used to define poverty. The following questions were asked in each community: On what does an extremely poor household spend any money that it obtains? What is the first expense that they cover? lf the 140 household obtains a little more money, what do they spend it on? And later, what would be the third expense? The fourth? And so on. In this way, people defined the stages without referring to money as such, for example, by asking, "lf that poor family was a little better off or if they obtained stable work, what would they do? What material or non-material things would the household have that the very poorest do not? Each stage was described as specifically as possible for the workshop participants, for example education would be either elementary school or high school and for housing both the size and the material were specified. Based on this information the poverty lines were described in each community. In each case, the participants were asked, "At what stage would a household no longer be considered poor?". The poverty line was identified by means of progressive questions, such as, "If a household has food but does not have clothes, does it continue to be poor?" until consensus was established among the participants. IV. Identify events that occurred 25 years ago, and determine the current situation and the situation 25 years ago: To establish what a 25-year timeframe means, each community was asked to identify historical events that all could easily identify. Three events were eh osen: The Popayan earthquake in 1983; The Nevada del Ruiz eruption and lahar (mud flow) that overwhelmed Armero in 1985; and The assault on the Palace of Justice by the arrned forces in 1985. This 25-year time frame was chosen to emphasize the changes that take place within a generation. In each community, the workshop participants evaluated the list of households in their community and estimated the stage they were at 25 years befare and agreed the stage that they currently are at. V. Group households in four categories (A-D below) and to choose a random sample of households in each category: Using the results of the previous step, each household was allocated to one of five groups: • Category A. Households that have remained poor (that were below the poverty line all the time during the last 25 years). • Category B. Households that have escaped from poverty (they were below the poverty line 25 years ago but at the moment they are above it). • Category C. Households that have fallen into poverty (they were above the poverty line the 25 years ago but at the moment they are below it). • Category D. Households that have remained non poor (they have been above the poverty line all the time in the last 25 years). • Category E. Households who were not in the community 25 years ago. The next steps in the workshops were to establish the reasons why sorne households escaped from and why sorne others fell into poverty. The Team visited those households to enquire more deeply into the causes. Because we shall not be referring to this information in this paper, we shall not present detail about these two final steps. 141 Description of the study sites The SOP methodology was applied in thirteen communities in six municipalities ofthe Fuquene Lake watershed (Fuquene) and in ten communities in five municipalities ofthe Coello River watershed (Coello) (Table 1). Table l.Watersheds, municipalities and communities where the stages or progress (SOP) methodology was applied. So urce: Progress Stages Project Scales, CPWF. Watershed Municipality Settlement Tausa 1 Ladera Grande 2 Rasgata Bajo 3 Chipaquin 2 Sutatausa 4 Palacio 5 Pellas de Cajón 6 Gacha Fúquene 3 Guachetá 7 La Isla 8 La Puntica 9 Centro y Guata 4 Fúquene 10 Chinzaque 11 Nemogá 5 Cucunubá 12 Chápala 6 Carmen de Carupa 13 Apartadero 1 La Leona-APACRAQ 2 El Rosal 3 La Alsalcia Cajarnarca 4 Minidistrito La Leona 5 Coello COOCRA 6 San Cristóbal-Honduras 2 lbagué Coello 3 Espinal 7 Dindalito 8 Potrerillo 9 Chaguala Adentro 4 Coello-COCORA 5 Ro vira 10 La Ocera Fuquene Lake Watershed Fuquene Lake watershed (Fuquene) is located in the Eastem Cordillera ofthe Andes north ofBogota, in the U bate and Chiquinquira valleys, with an altitudinal range of 2300 to 3300 masl (Figure 2). lt includes 17 municipalities in the departments ofCundinamarca and Boyaca. The population ofthe municipalities ofthe watershed in 2005 was of 232,000, 59% of which is rural (DANE popu lation projections). The CLI for 2003 (Sarmiento et al. 2006) in these municipalities varies from low to high. The percentage of households with UBN for 2002 varíes from 7 to 47%, which emphasizes the heterogeneity of living conditions, which is characteristic of the Andean basins. Moreover, the distribution of rural land in the watershed is inequitable with an average Gini4 index for rural property of0.59 in 2002. 4 The Gini index measures the extent to which the distribution of income arnong househo1ds within a country deviates from a perfectly equal distribution. A value ofO represents perfect equality, a va1ue of 1 perfect inequa1ity. http://hdr.undp.org/reports/g1obaV2003/indicator/indic 126 1 l.html 142 -= - - - - + + f i~------~--------~-~------~~1 Figure 2. Map of Fuquene Lake watershed. Source: Ramirez and Cisneros (2006). There are problems with legal title to property in the communities we surveyed. Moreover, according to the data in the Esquema de Ordenamiento Territorial (EOT) for the municipalities, unemployment in the communities reaches 70% or more, which causes migration to nearby urban centers. The communities have low levels of education (Team Project Scales CPWF, 2005b). The main economic activities in the region are cattle, agriculture and mining. Cattle are the most important economic activity, mainly dairying for the production of milk; the area produces a large proportion of the national total. The most important crops are potatoes, wheat, peas and maize. Agricultura! activity has quite a severe environmental impact. Cultivation is increasingly moving to steeper slopes, increasing soil loss by erosion, and to the paramo and forested areas, with the destruction of both ecosystems. Finally, mining is for coa!, as well as stone quarries and river sand as construction materials. All three have negative environmental impacts by causing both water and air contamination (Team Project Escalas, 2005b, Ramirez and Cisneros, 2006). In the communities, there are also landholders who do very little fanning, instead working off-farm because of high costs of production and advanced land degradation. Dairy farmers produce milk either for household consumption or for sale to milk companies. There are sorne sheep for sale as meat and family vegetable gardens support household food security. The Regional Autonomous Corporation (Corporación Autónoma Regional, CAR) promoted acacia and eucalypt plantations in sorne areas, mainly in the hillsides, to stop soil erosion (Team Project Scales, CPWF, 2005b). In the upland areas, land is commonly rented for potato production. Poor access to education and medical services occasionally forces people to move to other communities or to nearby urban centers. Public services are inaccessible and poor quality, especially access to water 1 water supply, although the latter is not always a problem (Team Project Scales CPWF, 2005b). This is a very important watershed because of its biophysical and socioeconomic characteristics. It is representative of the basins of the Andes in terms of the environmental problems, the disparities between the amount available and the demand for water, the topography, the inequalities in the distribution of 143 resources and the social conflicts. The results of the analyses of this basin can be extrapolated to similar basins in the Andes. Coe/lo River watershed The Coello River watershed (Coello) is located in the north central section ofthe Tolima department, on the eastem slope of the Central Cordillera and includes an important portion of Tolima (Figure 3). The total number of inhabitants of the area (including fbagué) in 2005 was of 622,000 people of which 16% were rural (projections DANE). More than 56% of the populations of most Coe11o municipalities are urban. The índices of life conditions range from medium-high to medium-low and health from medium to medium- low. The average years of education range from high to medium-low with most ofthe high in lbagué (the department's capital). In the other municipalities there are problems of coverage of and access to educational centers, both elementary secondary schools (Sarmiento et al. 2006). ' 1 / - - Olr41t".MI"10 Dll TOIIMA -- \ - ..._- Figure 3. Map of Coello River watershed. Source: Project Sea les, CPWF (2005). In the communities we studied the most important economic activities were agriculture, livestock breeding and mining. The main perrnanent crops in the region are coffee, sugar cane for production of panela, and fruits (mainly mango, followed by lemon and guanabana). The annual crops ofthe region are cotton, rice, sorghum, soybean and sesame. Mining is mainly stone quarries and river sand as materials for construction. Access to public services in rural areas is good, with almost 100% coverage by reticulated electricity. In contrast reticulated the water supply has problems of quality and continuity, and the sewage service has severe coverage problems (Rodríguez and Rubiano 2005). There is migration to nearby urban centers of lbague, Armenia, Bogota, Cajamarca, El Espinal and Rovira, with the objective of seeking work. The communities are farrners either from the zone or who immigrated from the departments of Quindio, Cundinamarca, Tolima and Boyaca. In the communities of Coello- 144 Cocora, Dindalito, Gualanday and Chaguala Adentro there are city immigrants. Property ownership is characterized by inequality, absence of communalland, predominance of a few large holdings and many small ones (Rodríguez and Rubiano 2005, Team Project S cales, 2005b ). The main environmental problem is due to the Panamerican highway in that people living near it are affected by severe respiratory diseases. Furthermore, tunnel construction in La Linea8 causes soil erosion and sedimentation. As well, there is the pollution by residual waters from towns, inadequate livestock and agricultura! management, including agrochemical contamination on the riverine plain, deforestation to increase the areas cropped to rice, illegal crops, commercial forestry plantations and mining. lt is important to note that there is increasing conflict in access to water between smalllandholders in the upper catchment and large-scale rice growers and the hydropower company in the lower catchment (Rodríguez and Rubiano 2005, Team Project S cales 2005b ). Results Results ofthe application ofthe SOP methodology The results the SOP study show how households in the sampled communities perceive their livelihoods in 2005 compared with 25 years ago (Table 2). The column A + C is the percentage of poor households in 2005, while B + D is the percentage who ar not poor. At the present 45% of the households in Fuquene are poor, with the highest numbers in Ladera Grande, Centro y Guata and Chapala, where 83%, 90% and 83% respectively of households are poor. In contrast, only 13% of households in Coello are poor and 84% are not poor. lt is noteworthy that between 55% and 100% ofthe households in the ten communities surveyed are classified as not poor. 8 The colloquial name for the Panamerican highway between !bague on the eastem and Pereira on tbe westem sides of the Central Cordillera. 145 Table 2. Results ofthe stages ofprogress (SOP) methodology, Fuquene Lake and Coello River watersheds, 2005. Source: Project Scales CPWF (2005), DNP-UNDP (2006). Municipality Community CatA 1 CatB Cate CatO CatE A+C B+D CLI % % % % % % % 20032 Ladera Grande 66 8 17 4 6 83 11 Tausa Rasgata Bajo 24 22 2 5 46 27 27 M Chipaquin 34 41 l3 l3 o 47 53 -o Sutatausa Palacio 37 47 o 2 14 37 49 M Q) Pefias de Cajón 17 61 o o 22 17 61 ..e ~ Gacha 38 26 6 28 1 44 54 E ~ Guachetá La Isla 40 30 3 25 1 43 55 M ~ Q) La Puntita 39 32 o 4 24 39 37 e: Q) Centro y Guata 90 1 o 1 7 90 2 :S CT Fúquene Chinzaque 23 46 3 28 o 26 74 M · :S u.. Nemogá 13 29 4 46 8 17 75 Cucunubá CháEala 83 13 o o 5 83 13 LM Carmen de CaruEa AEartadero 30 70 o o o 30 70 LM Total 42 30 3 14 10 45 44 La Leona - Apacra o 31 o 69 o o 100 Cajamarca El Rosal o 77 8 15 o 8 92 M La AJsalcia 21 57 o 21 o 21 79 -o Q) Minidistrito La Leona 45 55 o o o 45 55 ..e V) Cocora 6 22 o 44 28 6 67 ..... Q) .... lb agué MH ~ San Cristóbal-Honduras 5 47 5 42 o 11 89 ~ .2 Espinal Dindalito 15 46 8 31 o 23 77 MH d) o Coello Potreri1lo o 100 o o o o 100 LM u Chaguala 29 57 7 7 o 36 64 Ro vira La Ocera 6 75 o 19 o 6 94 LM Total 11 59 3 24 3 14 83 Grand Total 37 35 3 16 9 40 50 Categories (2005-25 years ago): A poor- poor, B poor-not poor, C not poor-poor, D not poor- not poor, E recent immigrants. 2 The condition of life index (CLI) is aggregated by municipalities for rural and urban zones. L = low, M = medium, H = high. Comparing categories A and B shows how household poverty has evolved in the last 25 years. ln Fúquene as a whole, 42% continue to be poor while 30% have escaped from poverty. However, we saw extreme and intennediate cases in the communities we studied. Ladera Grande, Centro y Guata and Chapala have stagnated, with between 66% and 90% of households that have always been poor. In contrast, in Peñas de Cajon and Apartadero 61% and 70% of households have escaped from poverty. In Coello as a whole only 1 1% of households continue to be poor and 59% ha ve escaped poverty, an important advance. More than halfthe households in nine ofthe thirteen communities improved their situation; the outstanding ones were Potrerillo, El Rosal and La Ocera with more than 70% of their households escaping from poverty in 25 years. The SOP results for current levels of poverty contrast with the CLI assessments made by the NDP for the year 2002. For Fuquene, where there was a high incidence of poverty according to SOP but the levels of CLI were medium. For Coello, with a smaller incidence of poverty in each community and where there was 146 greater progress according to the SOP evaluation, there were municipalities with low to medium CLI 9• Sorne differences are noteworthy, such as Chápala in Fúquene. According to SOP most ofthe population is poor, but the UBN of 6.7 in 2002 for Cucunuba municipality, where the Chapala community is located, indicates the opposite. Although worrying, these comparisons are not conclusive because the data are not disaggregated for urban and rural communities. It is important to keep in mind that the SOP evaluation depends on the poverty line defined by each community and that these poverty lines can differ, making comparisons across communities difficult. For example, a household considered to be poor in one community according to its poverty line may be above it in another community with a different poverty line. This leads us to analyze the poverty lines of the different communities: the elements they in elude and the number of stages there are to break out of poverty, together with the level of difficulty to overcome each ofthem. At this stage it is important emphasize the utility of information obtained by the SOP methodology, especially compared with the indicators calculated by official state organizations. The participatory method used here obtains information of both current and former living conditions in the areas under consideration in an opportune way. In contrast, obtaining information at the rural level from official sources is difficult, because official data are calculated from samples that are not representative when disaggregating departmental and municipal data into urban and rural components. Moreover, collecting data it is time- consuming and expensive. Definition and variability of the stages During the participatory process of the SOP, each community set its own poverty lines and defined the stages of progress out of poverty. The communities defined 7 to 24 stages in total and, in turn, decided which ofthem it was necessary to overcome to escape from poverty. They variously settled on from three to as many as ten stages. The stages of progress of the communities within the same municipality were different in both the actual stages that comprise the poverty line as well as in the number of the stages and their arder. This difference was especially notable in Sutatausa, Guachetá, Fúquene, Cajamarca, Ibagué and Coello (Table 3). There were differences in the definition of the poverty line between communitíes, which affect the particular community' s perception that, to escape from poverty, both the stages and the order in which they must be achieved are important. 9 The CLI in 2003 for Fuquene was 76.6 and for Coello was 73.4. 147 Table 4. Categories of Poverty Line in order of importance and frequency of mention. Source: Calculated from data in Project Scales CPWF (2005). Description lndex Order Frequency Food 0.84 1 20 Education 0.55 2 18 Clothing 0.41 3 12 Housing 0.41 4 15 Small animals 0.35 5 14 Land 0.23 6 8 Utilities 0.22 7 8 Appliances 0.20 8 8 Health 0.12 9 4 Cultivations 0.10 10 4 Other 0.08 11 2 Vehicles 0.06 12 2 S a vi ngs/i nvestment 0.04 13 1 Recreation 0.03 14 2 The index 1 measures the relative importance of each category in accord with the position that each community gave it in the stages required to overcome poverty. Moreover, one can deduce how difficult it might be to escape from poverty in a specific community by summing the relative weights given to each of the stages mentioned by that community. Those communities that mentioned larger number of stages to overcome to escape from poverty may be thought to be "more demanding" and are discussed in more detail below (Figure 4). ~ 3 .50 .e &. 3.00 ~ ª 2.50 :3 ~ 2 .00 Cll ~ 1 .50 "' 1! 1 .00 "' ·e:; ~ 0 .50 o .§ 0 .00 8 :;:; .5 1 (,) .!9 ., ~ ~ c.:> ,.., o ~ !l .m ~ ~ ..2! o ~ ! ~ &! ~ (,) o <'2 -' c.:> i c.:> !l o !l ., (,) ~ o ~ ~ •!i -' i :E 1 1 ; ~ ~ ~ i .. 18 t .;. ! [ i i :g u !l o: (,) o (,) ~ ~ !l ~ ¡¡¡ -8 0.. z 1 (,) :::1: 1 1 (,) ~ (,) -' ·!'! !l (,) ~ V eredas Figure 4. Leve( of demand for the poverty lines for each community. Source: Based on data in the Project Scales CPWF (2005). X axis= Communities, Y axjs = Index ofrelative importance ofthe poverty line. 149 In Peñas de Cajon, Apartadero, Potrerillo and La Ocera a large proportion of households escaped from poverty over the last 25 years. They mentioned three to five stages, including food, education, clothing, health, small anima\s, housing, land and crops. Not only do these communities mention fewer stages compared with the more demanding ones, but the stages are less complex. In contrast, the definitions of poverty in those communities where a large percentage of households have remained poor for the last 25 years implied more stages to overcome and more complex stages. In Fuquene, the households of Chapa! a and Centro y Guata needed to overcome 1 O and 8 stages respectively. In Chapala the initial stages included access to secondary education, a plot of land and housing with walls, roof and bathroom. In Centro y Guata the initial stages included improved crops by means of constructing a reservoir, while in La Puntica the poverty line has eight stages of which the seventh is entertainment by making trips. In those communities there are many stages to overcome, implying making investment to acquire and improve land and housing, and being able to access secondary education or entertainment. Ifthe households that are considered poor in "more demanding" (poorer) communities like Centro y Guata and Chápala moved to sorne ofthe " less demanding" communities, like Peñas de Cajon or Apartadero, they would no longer be considered as poor. In Chapala, youth migration has led to a population that is dominantly adu\t with few households having children under twelve. As a result the community mentioned high school education as a component in the poverty line. In Chapala, La Puntica and Centro y Guata agriculture is not a productive activity, only garden plots for self-consumption. In Chapala, the main activity is coal mining with emigration of mainly young people to work the mines. In Peñas de Cajon, most households are involved with coal mining. Good incomes from mining (up to COP1 ,500,000/month in 2005) has influenced the perceptions ofthe inhabitants ofthe communities, so that they consider entertainment and the possession of appliances inside the poverty line. Although miners' wages are good, it is not clear why the poverty lines for these communities include fewer elements than elsewhere. In contrast, in Apartadero, Potrerillo and La Ocera, most households are devoted to the agriculture and cattle, and these communities mentioned land and crops. The definition of poverty lines of each community is based on the local perceptions, which depend on social, economic and cultural factors that seem quite relative. We detennined the most (or least) common categories in the defmitions of poverty line in the communities using multiple correspondence analyses. This showed that services, food, educatíon, and housing were the most common, with savings/investment and other the least common. We therefore eliminated the latter categories because of their low discrimination capacity. We then used a conglomerate analysis on the reduced data set to generate the first two principal components to cluster communities in accordance with the elements that comprise the poverty line each (Figure 5). 150 3 2 o N -1 e: o "' e: Q) E i:5 -2 Grupos de Veredas Centro y Guata o Apartadero o ~aquin Palacio O La Isla O o Coello e La F\.o1tta o Olaguala Adentro o Gacha -3 -2 -1 o Dirrension 1 Sal Oistobai-Hondoxas o o o Olohzaque La Ocera 8 Rosal Pl:llrerillo D o Nerrogá Grupo La Leona Ladera Grande D o R!llas de D D D Cajón ~lito Mhidistrito D 4 La Leona o 3 o o Rasgata o Olapala Bajo La AlsaiCia o 2 o 1 1 2 Figure 5. Cluster analysis ofthe elements ofthe poverty line by Community. Source: Based on the information of the project Scales CPWF (2005). There are four groups, the poverty lines of which include one or more variables that was not mentioned by the other three groups. Group 1 includes land and appliances, Group 2 small animals and clothing, Group 3 vehicles and recreation and Group 4 crops. No pattem was detected for watershed or municipality. The communities of Group 1 are dominated by small holdings, which explains the inclusion of land as an important component ofthe poverty line. In contrast, Group 3 did not include landas the communities have large holdings for dairy cattle. The poverty lines in all communities include certain basic e lements such as food, education, housing and public services. Once the poverty line is overcome, communities tend to emphasize improved access to, and quality and quantity of, these basic elements. The conformation of the four Groups shows that there are sorne items that are fundamental to one group, but are unímportant to the others. There are not enough elements to allow us to determine why a group includes a particular element. Nevertheless, it is likely that factors that influence those activities that generate income such as land and sorne institutional factors, influence people's perceptions. For the communities ofGroup 2, the main source of income is off-farm work as most work in construction, tourism or in flowers production. Households involved with cattle find it easier to satisfy sorne food needs (milk and cheese) than those growing crops. In the case of Group 4, agriculture is the most important source of income, which explains why crops are included asan element ofthe poverty line. 151 In both Groups 2 and 4, households participated in both cooperative activities and women Colombian lnstitute ofFamily Welfare, (Instituto Colombiano de Bienestar Familiar ICBF), there is no explanation for the inclusion ofthe elements small animals, clothing and crops. The elements included in the poverty line in studies in other countries (Krishna 2004, Krishna et al. 2004a, Krishna et al. 2004b, Krishna et al. 2005a, Krishna et al. 2005b) do not differ in any important way from those in our study (Table 5). The poverty lines in each are made up of four to seven stages. However, sorne elements that are part of the poverty line for the Colombian communities are not included in the poverty lines of other countries, but we cannot say whether because these elements are not below the poverty line in these countries or because the communities do simply not consider them. For example, appliances, health, recreation and services do not appear for Peru, Kenya, India or U ganda. A possible explanation is that the number of communities and households in Colombia were smaller than in the other countries. lf this is so, we confront two possibilities if we enlarge the study uní verse in Colombia. We may find more differences in the stages defmed by the communities or, altematively, we may find more commonalities in the group of stages that comprise the poverty line that would enable us to generalize more widely to Colombian rural communities. 152 Table 5. Stages of prog ress for Colombia and other countries. Sources: Source: Krishna 2004a, Krishna et al. 2004b, Krishna et al. 2004c, Krishna et al. 2005a, Krishna et al. 2005b. Stage Coello-Fúquene K en ya Perú Uganda India 1 Food Food Food Food Food 2 Education Clothing Clothing Clothing School for children Home Basic home School for 3 Clothing improvement improvement children Clothing (root) Elementary Repair existing 4 Housing school for Smaller animals house (roof) Pay debts children Elementary Buy small 5 Small animals Buy chickens school for NA' children animals 6 Land Buy sheep Buy land Buy smallland NA Smaller Buy a bicycle 7 Utilities Buy local livestock for NA livestock (sheep, alpaca, transportation llama) Home 8 Appliances improvement Buy more land Buy more land NA (fum iture) High school for Home Buy a 9 Health permanent NA children lmprovement house 10 Cultivations Buy land Bigger animals Start a small NA bussiness 11 Other Buy livestock High school and Buy a car or NA higher educaton start a bussines 12 Vehicles Buy land SmaJI business NA Savings 1 Build a Buy a house in 13 pennanent NA investment house the city 14 Recreation lnvest in a NA bussiness 1 Information not available. Tbe stages of poverty highlighted in blue are those included in the poverty line created by the comrnunities in each ofthe case studies. There are e lements common to the poverty line in all countries (Table 5). Food is in frrst place in a ll cases, with clothing, housing, education, small animals, land or cro ps, generally included, although in different order. Although the results from the different countries are not strictly comparable, from the perspective of the social, economic and cultural contexts, there is a group of material assets that are part of the necessary stages to overcome poverty in rural areas in a ll countries. Most importantly, these are not presently considered by any other type of measurement. We can compare the e lements included in the poverty line with those of the tradit ional measurements of poverty such as the UBN and the CLI (Table 6). It is noteworthy that food, considered the most important in the poverty line is not is not included in either of the other two indicators. Moreover, there is a group of assets, such as small animals, land and crops that were included by the communities in this study that are not included in the so-called objective measurements of poverty, like the UBN and the CLI. ln the case of 153 rural communities, assets such as these con tribute to food security of the household and the survival of its members. Table 6. Elements included in the poverty line compared with the UBN and the CLI indicators. Source: MERPD, PNUD (2006). Poverty line l . Food 2. Education 3. Clothing 4. Housing 5. Small animals 6. Land 7. Utilities 8. Appliances 9. Health 10. Cultivation 11. Other 12. Vehicles 13. Savings/investment 14. Recreation UBN Non-school assistance: A household with children from 7 to JI years old who do not go to school. Inadequate housing: House with earth floor or flimsy wall material. House without utilities, a home with no water in urban areas and no connection to sewerage system or septic tank Critica/ overcrowding: Number of people per room more than 3. High economic dependency: A household with more than 3 dependent members, where the head ofthe family has a maximum of3 years' basic education. CLI Education and human capital Proportion of children from 5 to 11 years old in school Average schooling leve! for a 1 2-year- old or older Proportion of youths from 12 to 18 years old that go to high school or university Maximum schooling leve! ofthe head ofthe family Quality ofthe house Main floor material. Main material in the walls. Water source. Oil for cooking. Garbage collection. Siz.e and composition ofthe household Overcrowding in the house. Proportion of 6-year-old or younger children. If in the UBN and the CLI for rural areas included these types of assets, they may be more relevant to the real world situation in rural Colombia, in that households with these assets would have more favourable poverty scores than those tbat lack them. On the otber hand, access to education, is a most important element, in the poverty lineas well as in both the UBN and the CLI. The latter two are very specific about access to the education. In the UBN attendance at school of children between 7 and 11 years old is the criterion. Tbe CLI takes a more sweeping view considering not only school attendance by children of school age, but as well tbe years of education of the household's head and of the other members of the farnily, that is it measures tbe accumulated human capital ofthe household. 154 In the poverty line determined by the communities in this study, both elementary and secondary education for the chíldren was of great importance, as well as the capacity of the household to buy the supplies that are required to attend school. This seems reasonable, because parents generally consider that their chíldren must first be fed, then have school supplies, and only then attend school. They also considered access to technical training for adults when this was identified as a means to overcome poverty. Neither the household size nor the dependence rate were elements of the poverty line. In contrast, both the UBN and the CLI take account of the rate of economic dependence, together with overcrowding and the size and composition of the household. For the poverty line, only in El Rosal community in Fúquene was the second stage investment in family planning, although this is not the only community where women participate in programs of the ICBF. In the other cases the size of the household is not mentioned as a factor that has an impact in poverty, although the maintenance of more people does requires more effort of the household. Finally, most ofthe elements of the UBN and the CLI are related with the material attributes of quality of the housing, or with the accumulation of human capital in the household. Neither considers the availability of food nor the holding of other types of material assets, which contribute to the livelihoods of the rural poor. Moreover, holding other assets is also indirectly related to adequate food availability; which can mark the difference between being or not being poor. Conclusions Existing methodologies for the measurement of poverty have to face the challenge of trying to express, in a synthetic way, a complex and multidimensional phenomenon. The so-called objective methodologies (UBN and CLI, amongst others) require standard procedures to allow comparisons between the data that are obtained at different times, and they favour the use of quantitative data to fulfil this objective. The standards determined by these measurements are absolute, and on occasions they are out of context. The standardized elements lack explanatory power for the sake of simplicity and comparability. They also exclude sorne elements that are considered important by society, and in doing so generate contradictions and ambiguities in the definitions ofpoverty. In contrast, the subjective and participative methodologies we used here allow poverty to be understood in a given social, economic and cultural context. The context can be quite relative, particularly in conditions of heterogeneity and inequality, like in the watersheds we studied. But it is precisely because of these features that the capacity to make comparisons among communities is lost, as much in the temporal as in the spatial environment, due to their specificity to the place where they were detennined. Therefore, the objective and participatory methodologies are complementary, which can lead to defmition of better standards that include those elements that are important to people, and in this way allow more effective policies and poverty relief programs. On the other hand, subjective methods provide opportune information on the livelihoods in an area or region, when it is not possible to have up-to-date objective standards on a regular basis. It is worth emphasizing again that objective data obtained by the official organizations takes time and is costly, particularly for rural areas. The results of the poverty lines defmed by the methodology such as the stages of progress used here, include elements that are basic from the point of view of the objective measurement of poverty (food, housing, health, services, education). As such they are not in principie inconsistent with the objective standards. However, the elements that are basic to the objective measurements were not the only ones considered to be necessary in the community for a household to be classified as not poor. The additional 155 elements perceived as important depend on the preferences of the households of a particular community and define the differences of the stages between communities. The state of poverty from the perspective of the communities we studied depends on the context in which individuals form their conception of it, not only from the point of view of what is understood as being poor, but also by considering how complicated it can be to overcome poverty. For this reason there is a lot of variability regarding the location ofthe poverty line. One has to be very careful in considering the elements related to individual preferences and with their order, because it is not possible to make comparisons ofthe state of poverty between the different communities. This is because poverty lines differ as much in their composition as in the order of their elements. The results we present show that we know little about what causes the defmitions of the poverty lines to differ, because we found no pattem within the stages and their characteristics among the communities. We noted elements that could be associated with the possession of material assets, but in fact, they were not enough. Therefore, to deepen in the social and cultural characteristics of the communities it is necessary to understand, for example, why elements like recreation and owning a television are part of what a group of individuals can consideras essential stages to overcome poverty. The comparison of the stages obtained within Fúquene and Coello with the stages set by the communities of other developing countries, suggests that there is more work to do. We need: to carry out the stages of progress exercise with more households and with other poor communities in Colombia to determine ifthere is sorne commonality in the stages, and what characteristics of the different communities influence the poverty lines that each of them builds. Furthermore, the question arises whether we could make sorne generalizations of what is needed to overcome poverty. For example, what material assets are needed and what kind of human capital is required for rural communities to escape poverty. These generalizations are needed to focus the govemment's poverty alleviation programs in meaningful ways. We can conclude that the stages of progress methodology provides important information on the elements necessary to allow a household to overcome poverty. In the case of the rural areas we studied here, possession of material assets such as small animals, land and crops that con tribute to food security would improve the situation of many people. Equally, the possibility to send children to school and to provide the requirements to increase the levels of school attendance are high-priorities in the programs of poverty relief in the areas we studied. Participative methodologies like the stages of progress, can be very useful to identizy how the dynamics of progress out of poverty in an area have been, since the definition of the poverty lines describes the evolution of households' welfare over time. We must caution, however, that defining poverty is complicated by the specificity ofthe context ofwhat it means to be poor. The methodology of stages of progress can be u sed to focus on the poorest popu lation in a given area. However, we recommend that not only the categories be used but also the type of stages that the communities have not yet overcome. This will provide a sound description of the differences between the communities in terms of elements that are lacking. For example, it is not the same thing that a household is considered to be below the poverty line because it does not have the economic capacity to make recreational trips, compared with another household that is also below the poverty line because it does not have adequate housing. 156 References Boltvinik, J. (1997) Poverty measurement methods: An overview. Available at h!tJ>://www.undp.org/povertv/publications/pov red/. Team Project Scales CPWF (2005a). Metodología Etapas del Progreso Para el Análisis de la Dinámica de la Pobreza, adaptado de Kuan, J. Stages of Progress Methodfor Poverty Dynamics Analysis. Mimeo. Team Project Scales CPWF (2005b). Documentos de los Talleres realizados en las veredas de las cuencas de la Laguna de Fúquene y del Rio Coe/lo. Mimeo. Team Project Scales CPWF (2005c). Resumen de características de las veredas de las cuencas de la Laguna de Fúquene y del Río Coello. Mimeo. Krishna, A (2004). Escaping poverty and becoming poor: Who gains, who loses, and why?". World Development 32: 121- 136. Krishna, A, Kristjanson, P, Radeny, M and Nindo W (2004a). Escaping poverty and becoming poor in 20 Kenyan villages. Journal of Human Development 5:211-226. Krishna, A, Lumonya, D, Markiewicz, M, Kafuko, A, Wegoye, and y Mugumya, F (2004b). Escaping poverty and becoming poor in 36 villages of central and westem Uganda. Working Paper. Available at h!tJ>://www.pubool.duke.edu/krishna/documents/. Krishna, A, Kapila, M, Porwal, M and Singh, V (2005a). Why growth is not enough: household poverty dynamics in northeast Gujarat, India". Jo urna/ of Development Studies 41:1163-1 192. Krishna, A, Kristjanson, P, Kuan, J, Quilca, G, Radeny, M and Sánchez-Urrelo, A (2005b). Fixing the hole at the bottom of the bucket: Household poverty dynamics in forty communities of the Peruvian Andes. Working PaperA vailable at h!tJ>://www.pubpol.duke.edu/krishna/documents/. Krishna, A, Gibson-Davis, C, Ciasen, L, Markiewicz, M and Perez, N (2006). Escaping poverty and becoming poor in thirteen communities in rural North Carolina. Working Paper, Terry Standford Institute of Public Policy, Duke University. Available at h!tJ>://www.pubpol.duke.edu/krishnaldocuments/. Lok-Dessallien, R. (1995) Review ofpoverty concepts and indicators. Available at h!tJ>://www.undp.org/poverty/publications/pov red/. MDERPD (2006). Metodología de medición y magnitud de la pobreza en Colombia: texto preliminar sometido a ajustes eventuales. Bogotá. MDERPD. Ramirez, M.C., and Hector Cisneros (eds) 2006. Andean System of Basins: Watershed Profi/es - Enhancing Agricultura/ Water Productivity Through Strategic Research. Technical Report No. 1, Challenge Program on Water and Food. Colombo Sri Lanka. Ranvborg, H ( 1999). Desarrollo de Perfiles Regionales de Pobreza basados en Percepciones Locales. Cali. CIA T. Rodríguez, H and Rubiano, J (2005). Aspectos Bíofisicos y Socioeconomicos de la Cuenca del Rio Coello en el Depatamento del Tolima. Corporacion Semillas de Agua: Proyecto de Conservación y Desarrollo en la Cuenca del río Anaime, Cajamarca, Colombia. Sarmiento, A, Cifuentes, A, González, C and Coronado, J (eds.) (2006). Los Municipios Colombianos hacía los Objetivos de Desarrollo del Milenio: Salud, Educación y Reducción de la Pobreza. DNP, PDH, UNDP andGTZ, Bogotá, Colombia. 157 Agricultural water productivity: Issues, concepts and approaches Basin Focal Project Working Paper no. 2a Francis Gichuki, Simon Cook and Hugh Turra! Cha/lenge Program on Water and Food Introduction The world has fmite water resources, which are under increasing stress as the human population and water demand per capita both increase. These problems are not new but are now becoming more widespread and their impacts more devastating. This has provided additional ímpetus for the search for solutions to problems arising from the mismatch between demand and supply in terms of water quantity, quality and timing. Increasing water productivity has been identified as one of the global challenges that requires urgent attention. This document examines issues, concepts and approaches to assessing water productivity in agriculture. Section 2 presents a set of concepts and issues for improving our understanding of the complexities associated with assessing and improving water productivity. Section 3 presents approaches in assessing agricultura! water productivity and highlights the challenges in quantifying and valuing inputs and outputs. Section 4 presents the rationale for increasing water productivity from the global, basin, irrigation system and farm leve! perspectives. Concepts and issues Water scarcity: a driver for increasing water productivity Although globally there are adequate water resources to meet the needs of the current and future world population, locally there are many areas experiencing water scarcity (see Box 1 ). Box 1: Symptoms of growing water scarcity_ -· _ Declining dry season river flows have become a common phenomenon in many rivers. The ecological implications of decline in dry season river flows include (a) insufficient quantity of water to flush sediments leading to siltation problems; (b) insufficient quantity of water to dilute water pollutants leading to permissible pollution levels being exceeded; (e) reduced flows into estuaries result in a rapid advance of a satine front, which extends the extent of the estuary, damages the aquatic ecosystem and threatens bio-diversity. When such ecological changes occur, livelihoods of downstream communities are adversely affected leading to conflicts. Groundwater depletion. Studies in the Indo-Gangetic plains of northem India show that groundwater tables are dropping by 0.5 to 0.7 meters ayear and that 25 per cent of India's grain harvest is threatened by unsustainable use of groundwater. Postel (1999) estimated that globally groundwater is overused by 200 km3/year. The most serious over pumping occurs in [ndia, the United States, Mexico, the Mediterranean countries and China. Over-depletion of groundwater in the coastal areas of Gujarat state in lndia has lowered the groundwater levels to a point where salt water intrusion is contaminating the supplies of drinking water (Poste!, 1993). Water pollution arising from agricultura! and non-agricultura) sources is a major contributor to water scarcity. Water erosion transports soil particles and nutrients and deposits them in water bodies contributin to loss of reservoir ca aci and eutro hication of water bodies. lntensification of ero and 158 livestock production systems is a major contributor of pesticides and fertilizer pollutants of surface and groundwater resources. Discharge of untreated or partially-treated waste water into river systems aggravates the problem with sorne rivers becoming virtual sewers. For example, Yamuna river, which passes through New Delhi, receives 200 million litres of untreated sewage per day and has coliform counts as high as 25 million per 100 milliliters (Ciarke, 1993). Inequitable allocation of water resources, combined with poor enforcement of permitted water withdrawals leads to conflicts amon uses and users ofwater. Water scarcity exists when the demand for water exceeds the supply and it can be classified based on the context as: (a) physical water scarcity in which water availability is limited by natural availability; (b) economic water scarcity when human and financial resources constraints availability of water; (e) managerial water scarcity where availability is constrained by management limitations; (d) institutional water scarcity where water availability is constrained by institutional short-comings; and (e) political water scarcity where política! forces bar people from accessing available water resources (Molle and Molinga 2003). These types of scarcity can occur concomitantly, increasing both the severity and impacts of water scarcity. Molden et al. (2003) estimated that by 2020 approximately 75% ofthe world's population willlive in areas experiencing physical or economic water scarcity. Most ofthese areas happen to be where most ofthe poor and food insecure people live. Meeting their food needs with locally produced food presents enormous challenge. Hence, the need is to increase water productivity of agricultura! production systems that the poor people in water scarce areas depend on. Production function and technical and allocative efficiency Agricultura! production involves the combination of inputs to produce agricultura! outputs. For each agricultura! production system a generic production function (input-output relationship) can be derived: o= f(Il , 12, 13, ... In)(l) Where O is the output and Il, 12, D, and In are the production factors (land, labor, water, capital, energy and other inputs used in the production). As production resources become scarce, producers seek ways to enhance the productivity of the resources and of the entire production system. Understanding the production function is a pre-condition for identifying opportunities for improving the performance of a production system. lncreases in productivity can be achieved by two approaches: (a) by increasing technical efficiency through more efficient utilization of production inputs; and (b) increasing allocative efficiency by producing outputs with the highest returns. Here below, we illustrate how these two approaches can be used in identifying opportunities for enhancing productivity. The leve! of output produced when production resources are used most efficiently defines the technical efficient limit. For example, in Figure 1 the line A-B defmes the limits oftecbnical efficiency. Points below the curve are technically inefficient because the same leve! ofyield could be attained with less water. Points above the line are not technically feasible. The single factor production function presented in Figure 2 denotes the production possibilities for a given leve! of technical efficiency and of other production inputs. From Figure 2, we note that the output per unit input decreases at higher levels of input. In the case of crop production, a decrease in crop yield at higher levels of water input is mainly attributed to inhibited uptake of oxygen by the roots under water-saturated soil conditions. 159 8 ~--------------------------~~----------------~ ~ 6 +-----------------------~~~~~~~~~~~--~ ca ~ 32 -~ 4 +-----------------~~~~~~~~~~~~--~~~ >. Ql .1::1 ca ~ 2 +-----------~~~~~~~~~~~~~~~~=---~ o 1 00 200 300 400 500 600 700 800 900 1000 11 00 Annual rainfall (rrm) Figure l. Maize yield as a function of annual rainfall in Ewaso N giro Basin Kenya ca .r. e, .X ..¿ a; ·;:. E o o 8000 6000 4000 2000 1 ___ _j 1 1 1 1 1 1 0 ~--~~~~----~~ ----~----~ o 200 400 600 800 1000 Wate r supply, mm Figure 2. Crop-water production function for irrigated com in Brazil. At point A, a yield of 521 O kglha was achieved with the application of 600mm water. The technically efficient water application for thls yield is 300m m. Data from Frizzone et al. ( 1997). An analysis of a single factor production function enables us to assess opportunities for maximizing retums from the use of this factor. Let us take a case where the only way of increasing crop yield is by increasing the water input. To optimize the production system, one must understand how output increases with increase in water input. The contribution of water to the production process can be described on both average and marginal (incremental) basis as shown below at different levels ofwater input. 160 Output Average Product of Water == __ ;:;___ Water Input Change in Output Marginal Product of Water= __ ___::::..__ _ ___.!__ _ Change in Water I11put (2) (3) At low levels of input, marginal product is higher than average product and the average product is increasing. The average product equals the marginal product when the average product reaches its maximum. At high Ievels of inputs marginal product decreases and product concept can be used to aid farmers in deciding what is the optimal level of a given input to apply. How much production input is allocated to competing uses? The production resources available to farmers are Iimited and have competing uses. The farmers therefore have to choose the most desirable mix of agricultura! outputs that they can produce with the resources at their disposal and with their state of technical know-how. They know that sorne outputs can only be produced if they forego others in keeping with the opportunity cost principie (see Figure 3). Take a case of a farmer who has to allocate a given quantity of water W to producing two crops, product A and product B. The production possibility curve for this enterprise shows how much of product A and product B he/she can produce for a given level of water input. Increasing the output ofproduct A can on ly be achieved by reducing product B for that leve) ofwater input and production technology. Increasing the amount of water input will enable him to produce more of product A and product B at a higher production possibility curve. m ..... u ~ 'O e ll. A Production possibility 1 - + - - 1 X 1 • _j_ _ _ 1 1 1 1 curve (or frontier) y 1 1 _, __ 1 1 1 1 • e Product A Figure 3. Product possibility curve (PPC), which depicts all maximum output possibilities for two (or more) goods with a given a set of inputs (resources, labor, etc.). The PPC assumes that all inputs are used efficiently. Points A, B and C all represent points at which production of both product A and product B is most efficient. At point X resources are not being used efficiently in the production of both products, while point Y cannot be attained with the given inouts. 161 The concept of allocative efficiency is used here to illustrate how a farrner could make a decision on what and how much to produce wíth a given leve! of input. The slope of the production possibility curve presented in Figure 3 is the rate at which crop A is substituted for crop B and is called the marginal rate of transformation (MRT). The farrner maximizes his returns ifthe marginal rate oftransfonnation of crop Ato crop Bis equal to the price ratio ofthe two crops. PA MRT AB=p B (4) What combination of input levels minimizes production costs? A given leve! of crop yield can be attained using different combination of production inputs. Hence farrners are confronted with the challenge of substituting one production input for another. The concept of marginal rate of technical substitution illustrates how this could be done. Isoquants are curves that show all possible combinatíon of inputs that yield the same leve! of output. An isoquant map is a combination of severa! isoquants in a single graph that describe how levels of output vary with different combination of input levels. lsoquant maps are used to assess input substitution as the slope of the isoquant indicates how the quantity of input can be traded off against the quantity of other inputs, holding the output constant. This slope is called the marginal rate to technical substitution (MRTS). As more and more of one input replaces the other, its productivity decreases as that of the other input beco mes more productive. For example, the marginal rate oftechnical substitution (MRTS) of fertilizer for water in a crop production system is the amount by which the input of water can be reduced when one extra input of fertilizer is used, holding the crop output constant. MRTS tells the farrners the nature of the trade-offs involved in adding fertilizer and reducing the leve! of water input. The decision on how much water and how much fertilizer to use depends on the relative cost of these inputs. Optimum production is achieved when the following conditions ho1d MRTS = MPF = PF fW MP. P. w w (5) Where PF and Pw are the unit prices of ferti1izer and water respectively and MPF and MPw are marginal products of fertilizer and water respectively. MPFIPF is the additional output that results from spending an additional dollar on fertilizer. Similarly MP../Pw is the additiona1 output that resu1ts from spending an additiona1 dollar on water. The above equation therefore te11s us that the farrner would minimize cost by choosing the quantities of inputs so that the last dollar's worth of any input added to the production process yields the same amount of extra output. We illustrate this point with a nurnerical example. Suppose that the unit costs of fertilizer and water are $1 O and $2 respectively. Jf an additional unit of water would increase output by 20 units, then the additional output per do llar of water input is 20/$2 = 1 O units per do llar and an additional unit of ferti lizer increases output by 4 units. Because a do llar spent on water is five (20/4) times more productive than a dollar spent on fertilizer, the farrner wants to use more water than fertilizer. Jf he reduces fertilizer and increases water, the marginal product of fertilizer will rise and the marginal product of water will fa11. Eventually, a point will be reached where the production of an additional unit of output costs is the same regardless of which additional input is used. At that point the farmer minimizes his costs. Producers versus social net benefits Farmers seeking to maximize benefits will use water and other inputs at levels where the incremental value generated is at least equal to the incremental cost. Farmers choose cropping patterns that maximize net benefits over time subject to their resource endowment, relative input and output prices and market opportunities. In places where water is scarce relative to available land, farrners will choose crops that maximize net retums to their Jimited water supplies. The way that they manage the water resources will 162 depend on the cost and availability of water together with the technologies and management practices available for improving water productivity. From society's perspective, the goal for managing public water resources can be described as maximizing the present value of net benefits over time. Net benefits include farm-level net returns minus the cost of water delivery, while accounting for the opportunity cost and extemalities. Opportunity costs are the incremental values water might have in alternative uses, for example the value that it might generate for a downstream user. In addition to the spatial dimension (is the water worth more somewhere eJse?), opportunity costs can also have a temporal dimension (would the water be worth more at sorne time in the future?) In these examples, opportunity costs must be discounted by the cost of conveying the water to the altemative site or the losses during storage. Society's objectives in increasing water productivity are to meet the ever-increasing demands on a finite water resource. They fall into the following categories: • Food security for all; • Poverty alleviation; • Employment creation- jobslm3; • Equity and • Meeting environmental demands (for example, flows needed to maintain wetlands). Extemalities are the off-farrn effects of water use that irnpose costs or benefits on other water users. Positive extemalities are any benefits that accrue, for example the generation of usable runoff to a desirable wetland area. Negative extemalities are short- and long-term damage caused by runoff and deep percolation, for example, waterlogging and salinization. Performance assessment Rationale for performance assessment Under conditions of increasing scarcity of resources, performance measures can play an important role in identifying opportunities to improve performance. Performance measures of similar (sub-) systems in different geographic locations or those tracking the performance of a particular (sub-) system over time can provide answers to strategic questions, sucb as: "What types of systems get the most retum from lirnited water and land resources?" At the same time, they provide a cost-effective means of tracking performance in individual systems and valuable information that can be used by: • Planners to evaluate how efficiently and effectively land, water, labour and capital resources are being used; • Agricultura! producers and managers of water systems to identify long-term trends in performance for use in setting reasonable overall objectives and to measure progress; • Researchers to compare systems and identify factors that lead to better performance; and. • Policy makers and development facilitators (donor agencies, prívate sector and NGOs) to assess the impact of their interventions so that they can be designed to be more effective. • One ofthe challenges to improve the performance ofwater in agricultura) systems is to answer the questions: • How well should a system be performing; • How well is it performing; and • How can its performance be improved in a cost-effective way? • Such performance analysis implies the need for: • Performance standards that will be used for comparison; 163 • Tools and methods for assessing performance and any shortfalls in performance that there may be; and • The ability to analyze critically the performance data and the determinants of those performance shortfalls that are modifiable. Challenges of evaluating highly di verse production systems A river basin comprises of a mosaic of highly diverse agricultura! production systems whose outputs include crop, tree, livestock and fish products and a vast multitude of ecological goods and services. The bio-physical, socio-economic and institutional settings under which these production systems operate and the multiple and sometimes conflicts goals of the key actors present additional challenges. The decisions made by agricultural producers and managers of water systems determine the levels of technical and allocative efficiency of the water resources available in the basin. Their decisions are influenced by the policy and regulatory instruments and by the level of complementary interventions such as infrastructural development. We therefore consider increasing water productivity to be a shared responsibility, however here we focus on the perspective ofthe agricultura) producers. They have multiple objectives upon which they assess the performance of their production systems, namely the productivity, profitability, stability, diversity and time-dispersion (see Box 2 for brief definitions of each). The relative importance of each of these objectives depends on whether their production system opera tes at subsistence leve! or is partially commercial or fully commercial. Box 2. Multiple objectives of an agricultura! producer. Productivity: Ratio of output to input that serves as a measure of the relative suitability of a farming system oran activity within the system. Profitability: Net benefit accruing from the farming system. Stability: The absence or minimization of season-to-season or year-to-year fluctuations in the leve! an CIJ ::J C) e :¡::; C1> c.. C1> -o 1 e o z C1> CIJ ::J Economic .... C1> - benefits ~ CIJ C1> CIJ ::J C) e :¡::; C1> c.. C1> Cl Benefits Figure 5. Water use and associated benefits. 168 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 An agricultura! production system may be perceived as providing benefits of primary and secondary goods and services as shown in Table l. Many secondary goods produced in agricultura! systems are complementary to one another. For example, crop production systems may sequester carbon in sorne circumstances but in other circumstances they may release carbon. In most cases, the improved function of a system comes from complementarities in space and time. The difficulty is to compare parallel benefits from a range of water users. Table l. Sorne agricultura\ oroduction svstems and their benefits. Agricultural Benefits production system Primary goods Secondary goods Serví ces Soil cover to reduce Crop production Harvestable yield Crop residue for erosion, enbancing agro-livestock feed biodiversity, carbon seques tra tion Livestock production Meat, milk and eggs Draft power, manure, leather Bio-diversity, water Tree production Timber, fuel wood Food catchment protection, carbon sequestration Fish production Meat Manure Bio-diversity of aquatic ecos stems We now focus on agricultura! output, which can be evaluated for physical, nutritional or monetary benefit, within the bounds of the scale and time period being considered. Difficulties arise when outputs are difficult to value or when output quantities are expressed in different units. Crop production, for example, may produce a range of significant outcomes, including grain yield, fodder for livestock and organic matter for soil quality improvement. The first is measured in kg of grainlha, the second in kg of gain in animal live weightlha and the third in kg of organic matter accumulatedlha. It is little wonder then that past assessments of agricultura! benefits have focused mainly on primary benefits of a given production systems and assessed the benefits as shown in Table 2. The table shows that even when focusing on primary output, there are still many ways in which the output can be quantified. This presents an enormous challenge when comparing total water productivity of different agricultura! production systems. 169 Table 2. Agricultura) production systems and their primary goods and indicators Agricultural Primary lndicator production system output Physical/ nutritional Economic • Total above ground biomass (marketable and non-marketable) at standarclized water content (kg) Value of Crop production Harvestable • Total harvested product at the water output or yield content at which it is consumed (kg) gross margin • Nutritional content (kcal, or grams of of output protein, vitamin or micro-nutrient content • Live weight of animal (kg) • Meat (kg) Value of Meat, milk • Milk (kg or liters) output or Livestock production and eggs • Eggs (number or kg) gross margm • Nutritional content (kcal, or grarns of of output protein, vitamin or micro-nutrient content Value of Tree production Timber, fuel • Timber (kg) output or wood • Fuel wood (kg) gross margin ofou ut • Meat (kg) Value of Fish production Meat • Nutritional content (kcal, or grarns of output or protein, vitamin or micro-nutrient gross margin content) of output Current approaches to assessing the agricultura) water-use benefits generally ignore secondary goods and services. In sorne cases these could be important. For example, in rainfed farming systems, grain is only one output of value to the farmer; others are green and dry fodder (grazing during early crop growth and straw and stubble after harvest). In pastoral systems, the value of green biomass varies at different stages of growth so that it is usual to convert green and dry biomass into digestible dry matter to account for this variability. Additionally, the value of a product may vary according to its position within often-complex farming systems. These benefits often have relevance for broader water management goals and should be acknowledged in any attempt to define water productivity. Hence, the need to develope methodologies that take the following into consideration: Non-grain benefits of water use in crop production such as the use of crop residues as fodder and/or mulch. Benefits from by-products of livestock and fish production and their role as food supplements for livestock and fish production systems oras inputs to enhance soil fertility. Benefits from ecosystem goods and services (biodiversity, ecosystem integrity, habitat maintenance) and socio-cultural benefits, such as aesthetics and cultural importance, derived from hydrologic flows in agriculturaJ water use systems. 170 Indicators of water productivity Water productivity is a very robust measure that can be applied at different scaJes to suit the needs of different stakeholders. This is achieved by defining the inputs of water and outputs in units appropriate to the users' indicator needs. The numerator (output derived from water use) can be defmed in the following ways: • Pbysical output, which can be total biomass or harvestable product; • Economic output (the cash value of output) either gross benefit or net benefit. The water input can be specified as volume (m3) or as the value of water expressed as the highest opportunity cost in altemative uses ofthe water. The combination of the different numerator and denominator parameters yield a wide range of water productivity indicators as illustrated in Table 3. Table 3. Range of water productivity indicators and the units that can be u sed. Output 2arameter Physical measures Physical/ economic Economic Water input parameters measures measures (m3 or $ value) Biomass Harvestable Gross Net Gross Net (kg) yield value value val u e value (kg) ~$2 1$2 ~$2 ($) Gross inflow kg/m3 kg/ m3 $/m3 $/m3 ($/$) ($/$) Net inflow kg/ m3 kg/m3 $/m3 $/m3 ($/$) ($/$) A vailable water kg/m3 kg/ m3 $/m3 $/m3 ($/$) ($/$) Depleted water kg/m3 kg/ m3 $/m3 $/m3 ($/ ) ($/$) Beneficially depleted water kg/ m3 kg/m3 $/m3 $/m3 ($/$) ($/$) Process de2leted 2rocess water kg/ m3 kgLm3 $/m3 $/m3 ~$/$~ ~$/$~ We now consider bow the different indicators could be used to assess water productivity for a cropping system at different scales: • Crop scale: is of interest to crop physiologists to assess how efficient a particular crop or cultivar of a crop is in converting water into biomass. At this scale the output can be quantified either as total biomass or crop yield (harvestable produce). The water input tbat is relevant for this assessment is the water used in transpiration, which here we call process depleted water. • Field scale beneficia) use: is of interest to the farmer, agronomist and water specialist to assess bow efficiently a particular cropping system converts water into beneficial output. At this scale the output can be quantified as total biomass or crop yield and the water inputs are the amount ofwater that was used in process depletion (transpiration). 171 • Field/farm scale beneficial and non-beneficial use: is of interest to the fanner, agronomist and water specialist to assess the opportunities of saving water lost th.rough non-beneficia! use. At this scale the output can be quantified as total biomass, crop yield (kg), crop value ($) while the water input is the amount ofwater depleted from the system through (a) evaporation, (b) flows to sinks that are not recoverable, (e) pollution to Jevels that render it unfit for use and ( d) incorporation into the product. • Irrigation system scale: is of interest to the irrigation system manager in assessing how productively the water available to the irrigation system is being used. At this scale the manager takes into consideration both the amount of water depleted and that which is recaptured for re-use downstream. At this scale the output can be quantified in physical and economic tenns and the water can be accounted for in either volume or in value tenns. • Sub-basin scale: is of interest to planners and river-basin managers in assessing options for increasing water productivity at this scale. The output can be quantified as either biomass or harvestable produce in kg or their cash value in $. The water input becomes the net inflow, which is difficult to value in monetary tenns and therefore generally assessed as volume. It is particularly useful in assessing the opportunities for investing in water infrastructure. • Basin sea le: is of interest to river-basin managers and planners in assessing the productivity of the renewable water that enters the basin, mainly as rainfall. The output includes all the water benefits derived by water as it m oves across the basin landscape and could even include the value of near- shore marine life. Aggregating multiple outputs water productivity One of the advantages of the water productivity concept is that it allows us to assess the productivity of multiple-use systems such fish production in irrigation canals, or where crop residue is an important source of livestock feed. Under such conditions there are multiple benefits arising from using the same quantity of water. To calculate water productivity of depleted water in a multiple-use system we could use the formula: Where Yij is the amount of output for production systemj on field i (kglha), Wij is the amount ofwater depleted (m3/ha), Aij is the production area, p is the number of production system and n is the number of fields (7) Molden et al. ( 1998) proposed an approach for standardizing crop benefit by using the standardized gross value of production indicator within an area, computed as: Where A, is the area cropped with crop i, Y; is yield of crop i, P; is local price of crop i, (8) Pb is the local price ofbase crop (the main locally-grown, intemationally-traded crop) and 172 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 P w is the value of the base crop traded at world prices and N is the number of crops grown. Where data exist, SGVP for other agricultura) produce (fish, livestock and trees) may be assessed using a similar procedure. The equation can also be expanded, at least in principie, to include the value of other goods and services. These might include value society places on other ecosystem services generated by the hydrologic flow ofwater through the system such as biodiversity, ecosystem integrity, habitat maintenance, aesthetics, etc. Rationale for increasing water productivity Global imperatives The global population, which reached 6 billion in 1999 and is expected to reach 7.8 billion in 2025, is putting enormous pressure on the finite renewable water resources as the demand for food and other water- dependent goods and services increases. Irrigated agriculture, which accounts for 72% of global and 90% of developing countries' water withdrawal will have to increase its productivity to mitigate the growing water crisis (Caí and Rosegrant 2003). Other agricultura) water uses will also have a role to play. It is estimated that increases of 30 and 60% in water productivity from rainfed and irrigated agriculture, respectively will be required to meet the demands for food security. To achieve sustainable agricultura) growth necessary for food security, Kofi Annan, Secretary General of the United Nations in his Millenium Report to the General Asembly, April, 2000 and repeated in his report to the Millenium Conference in October, 2000 called for a "Blue Revolution in agriculture that focuses on increasing productivity per unit ofwater - 'more crop per drop" (Annan 2000). This formed the basis for setting global target for reducing water use in agriculture that is stated as (CPWF 2002): "maintaining the leve/ of global diversions of water to agriculture at the leve/ of the year 2000, while increasingfood production, to achieve internationally-adopted targets for decreasing malnourishment and rural poverty by the year 2015, particular/y in rural and peri-urban areas in river basins with low average incomes and high physical, economic or environmental water scarcity or water stress, with a specific focus on low-income groups within these areas " Basin-level rationale At the basin leve!, the rationale for increasing water productivity líes in the need to: • Increase water availability to users and uses that are disadvantaged. For example the need to in crease water productivity in the upper reaches of rivers so as to reduce water depletion and hence increase water availability in downstream reaches; • Reduce overall water demand and develop additional water resources (dam development, groundwater exploitation and water transfers from regions with excess water to regions that experience water scarcity); and • lncrease total basin leve) water benefits through more productive use of the available water resources. Severa! basins are exploring options for enhancing water productivity to achieve various social, economic and environmental goals. For example, the Yellow River basin, which is currently experiencing severe water shortages in the dry seasons 11, has set a target of reducing water use in agriculture by 4 billion m3 by 20 1 O so as to meet the needs of urban development. 11 In 1972, the river dried up failing to reach the outlet for 15 days. Since 1985, the tower reach has run dry each year, with the dry period and the length of the dry section getting increasingly longer. In 1996, the river dried for 133 days and in the drougbt year of 1997 for 226 days. 173 System leve/ rationale At the level ofthe irrigation system, increases in water productivity may be required to: • Secure water for downstream farmers who experience water shortages; • Reduce operation and maintenance costs associated with desilting and water outtake including the costs of pumping; • Make water available for expansion of the irrigated perimeter where the cost of saving water through increasing water productivity is less than the cost of developing additional water resources; and • Comply with water pennit and pollution regulations to ensure adequate provision of safe water for non-agricultura! users. Farm leve/ rationale At the farm level, increases in water productivity are required to: • Reduce water costs (costs ofpumping, delivering water or water fees); • Reduce loss of land productivity associated with soil erosion, waterlogging and salinization; • Expand irrigated areas with the same amount of irrigation water available; and • Increase agricultura! output, food security and profitability. 174 References Annan, K. A. (2000). We the Peop/es: The Role of the United Nations in the 21'1 Century. Millenium Report of the Secretary General to the General Assembly. United Nations, New York. URL: htlJ?://www.un.org/millennium/sg/report/full.htm Chapter V Sustaining Our Future, p. 61. Cai, X. and Rosegrant, M. (2003). World water productivity: Current situation and future options. Ln: Kijne, J.W., Barker, R. and Molden, D. (Eds.) Water Productivity in Agriculture: Limits and Opportunities for lmprovement. CABI, Wallingford. Pp. 163-1 78. Clarke, R. (1993). Water: The lnternational Crisis. Earthscan Publications, London. CPWF. (2002). Challenge Program on Water and Food Proposal. Frizzone, J.A., Coelho, R.O., Dourado-Neto, D., and Soliana, R. (1997). Linear programming model to optimize the water resource use in irrigation projects: An application to the Senator Nilo Coelho Project. Sci. agríe. (Piracicaba, Braz.) 54:136-148. URL: htfJ?://www.scielo.br/scielo.php?script=sci arttext&pid=SO 103-901619970003000 16&lng=en&nrm=iso Accessed 20 May, 2006 Guerra, L.C., Bhuiyan, S.I., Tuong, T.P., Barker, R. (1998). Producing More Rice with Less Water from lrrigated Systems. lntemational Rice Research Institute, Manila (Philippines), 19 pp. Keller, A., Keller, J. and Seckler, D. 1996. lntegrated Water Resource Systems: Theory and Policy Imp/ications. Research Report 3. International Water Management lnstitute, Colombo, Sri Lanka. McConnell, D.J. and Dillon, J.L. ( 1997). Farm Management for Asia: A Systems Approach. F AO Fann Systems Management Series 13. FAO, Rome. Molden, D. (1997). Accountingfor Water Use and Productivity. SWIM Paper l. International Irrigation Management Institute, Colombo, Sri Lanka. 26pp. Molden, D., Sakthivadivel, R., Perry, C.J. de Fraiture, C. and Kloezen, W.H. (1998). Jndicators for Comparing Performance Indicators of lrrigated Agricultura/ Systems. Research Report 20. Intemational Water Management Institute, Colombo, Sri Lanka. Molden, D., Murray-Rust, H., Sakthivadivel, R., Makin, l. (2003). A water productivity framework for understanding and action. In: Kijne, J.W., Barker, R., Molden, D. (Eds.) Water Productivity in Agricu/ture: Limits and Opportunities for lmprovement. CABI Publishing, Wallingford, UK, pp. l- 18. Molle, F. and Mollinga, P. (2003). Water poverty indicators: conceptual problems and policy issues. Water Policy 5:529-544 NRM3 (2000). Natural Resources Monitoring, Modelling and Management Database, Laikipia Research Program, Nanyuki, Kenya Poste( , S.L. (1993). Water and agriculture. ln: Gleick, P.H. (Ed.) Water in Crisis : A Guide to the World's Fresh Water Resources. Oxford University Press, Oxford. Pp. 56-66. Postel, S.L. (1999). Pillar ofSand: Can the lrrigation Mirac/e Last? W.W. Norton & Co., New York. 175 Water productivity: Measuring and mapping in benchmark basins Basin Focal Project Working Paper no. 2 Francis Gichuki, Simon Cook and Hugh Turral Chal/enge Program on Water and Food Basic concept: Water productivity for agricultural production systems The concept ofwater productivity (WP) is offered by Kijne et al. (2003) as a robust measure ofthe ability of agricultura! systems to convert water into food. While it was used primarily to evaluate the function of irrigation systems -as 'crop per drop' - it seems useful to extend the concept to include other types of livelihood support, such as mixed cropping, pasture, fisheries or forests. The basic concepts and rationale for estimation are described more fully in the first working paper ofthis series. The purpose of this paper is to present ideas of metbods of estimating WP at a range of scales, and for different agricultura! systems. Water productivity ofnon-agricultural systems is not considered. A third paper in this series describes how estimates are used to defme actionable goals of agricultura! water management for poverty alleviation. For now, we assume two basic uses of WP estimates: firstly, WP provides a diagnostic too! to identify low or high water use efficiency in farming systems or sub-systems; secondly, WP provides robust insight into the opportunities for re-distribution of water within basins towards a goal of increased basin-scale and global water productivity. In practice, measurement of WP over Jarge areas requires approximations and assumptions that can introduce important errors. The subsidiary purpose oftbis paper is to enable developers to make judgments about how acceptable these errors are and what altematives there may be to resolve the technical problems. Basic expression: Productivity is a measure of system performance expressed as a ratio of output to input. For agricultura! systems, WP is a measure of output of a given system in relation to the water it consumes. Assessment may be required for the whole system or parts of it, defined in t ime and space. Units: WP = Agricultura/ Benefit Water Use It is normal to represent WP in units of kglm3• If production is meas u red in kglha, water use is estimated as mm of water applied or received as rainfall, convertible simply to m3/ha (!mm = 10m3/ha). Altemative notations include food (kcal/m3) or monetary value ($/m\ Defining the system for which Water productivity is to be assessed: Water productivity is estimated for an agricultura! system or sub-system, defined with in a given area and time period. The simplest purpose of WP is to enable rapid comparisons between water use systems in space and time; a WP of 1.5 kg/m3 might be considered 'good' whereas one of 0.5 kg/m3 might be thought ' bad'. For this purpose, it is preferable to restrict the concept to component parts of a system, rather than try to estímate overall productivity for the en tire system. 176 Systems are defmed by plot, field, sub-basin and basin. Estimates of WP for single activities are called partial WPs. WP of Jarger areas containing complexes of multiple Jand use requires integration of partía! WPs for each activity contained within them. Changes in water use in the hydrologic pathway will have impacts both upstream and downstream, so it is necessary to analyze the impacts of different interventions in a way that intemalizes hydrologic feedback in terms ofwater quantity and water quality. The best way todo this is to integrate the production system, the hydrology and economics within one modeling framework. This can vary from simple spreadsheets, through suites of hydrologic, allocation and production models; to fully integrated hydrologic and economic models. The precise requirernents and solutions will vary according to the basin context and data availability. Defining the time periodfor esfimation The time period over which WP is estimated is detennined by the cycle of agricultura} production that drives the system. Nonnally, this would include at least one complete crop cycle, extended over a complete year to account for productive and non-productive water use. Assessment may be extended over severa! years to derive estimates of average, mínimum or maximum water productivíty within each season. Complex agricultura! systems may require assessment over severa! years to include all productive and consumptive phases. The value of product may vary according to its position within the farming system it is used, often in quite complex ways. For example, livestock systems in semi-arid regions have developed to cope with fluctuations in water availability in different seasons, so assessment in any one season may not represent productivity ofthe whole system. Cropping systems provide interna) benefits in addition to yield, such as fodder or soil nutrition, which may significantly influence water productivity in subsequent years. Forest products may provide small but critically important gap-filling products. The fluctuation over time of drivers of productivity such as climate or markets introduces a further source of estimation error. This is because the condition of WP will reflect the state of these drivers at the time of assessment, which may, or may not, be representative ofthe average situation. Defining the are a for esfimation The flfSt step is to defme the boundaries ofthe system for which WP is to be estimated. This is determined by the definition of production system (field-by-field, farm-scale, multiple administrative units) and the area for which water consumption can be defined (plot, field, sub-basin or basin). There is a trade-off between accuracy of measurement over small areas and representation of a larger hydrologic system. Measurement of partial WP for a single crop at field or plot level is simplest. However, such an assessment wiU represent only part of the benefits generated within a farming system. Additional activities within the farming system such as livestock, trees or fish may need to be included to represent essential benefits, but will also introduce major uncertainties. Yield data exist for many countries as secondary statistics, expressed crop-by-crop according to administrative boundaries. In these cases, manipulation in GIS is required to make the data spatially coherent with water-use data. Techniques of proportional areal estimation are described in standard GIS texts (see Davis, 2003 for a review of methods in relation to poverty mapping). The effect of spatial scale on variation in water storage should also be considered. In rainfed areas, WP will vary spatially according to varying water storage capacity (Bouman, 2006), such that definition of a particular production system can be over- or under-represented within areas with a high or low storage capacity. Systems in which groundwater flows make a significant contribution to production (either vía re- 177 emergent or extracted groundwater) need to be defined over an area large enough for this source of variation to be identified. Production: Estimating the enumerator The beneficial outcome of agriculture can be expressed in a range of forms, as yield (kg, Mg, t) or food equivalent (kcal.); income ($) or other agreed measure of well being derived from the goods and services coming from the agricultura! system. The common forms of evaluation are listed in Table l. Table l. Possible forms ofthe numerator for estimating water productivity. Parameter Physical water productivity at field, farm or system level. Economic water productivity at farm level. Economic water productivity at basin scale. Macro-economic water productivity at regional or nacional scale. Physical measurement of productivity Nwnerator Yield (kg) of total biomass, or above ground biomass, or grain, or fodder. Gross value of product, or net value of product, or net benefit of irrigated production compared with rainfed production. Any of the above valuations including those derived from raising livestock, fish or agro-forestry. Monetary value of all direct and indirect economic benefits minus the associated costs, for all uses of water in the domain of interest. The simplest option is to consider the water productivity of a principal crop, in kg or t, for an area of known extent for which there exist data on agricultura! water use. Primary yield data may be generated from direct measurement or by crop survey. More commonly, crop production data will come from secondary statistics -as total tonnage for a given administrative area (convertible to t/ha if the area dedicated to each crop is known). In sorne cases, global or national level statistics can be manipulated to provide useful insight (e.g. Falkenmark and Rockstrom, 2004). This will enable partial water productivity of principal crops to be estimated for large areas. Raw production in kg may be converted to nutritional value (see Rockstrom et al. 2003). Remote sensing provides a third option to estímate production over large areas. Wide area estimation of grain yield from the normalized di.fference vegetation index (NDVI) has been used successfully for basin or national scale estimation ofbiomass and crop yield for severa} years. Its accuracy varíes widely, depending on the ability for an episodic estimation of reflectance to estímate biomass and final yield. Under good conditions, remotely sensed imagery correlates about 70% with final grain yield. Local calibration seems essential since in sorne areas correlation between NDVI-based estimates and actual yield can be extremely low. Remote sensing has also been used to assess total feed value ofpastures. In all cases, the accuracy ofremote sensing techniques is highly dependent on the availability oftimely imagery. 178 Remote sensing tends to be most successful in arid and semi-arid regions in which cloud-free imagery is available for the whole growing season. Ahmad et al. (2005) propose a combined remote sensing approach to estímate water productivity at a regional scale, using a variety of scales of imagery (Landsat at 28.5 m pixel to Moderate Resolution Imaging Spectroradiometer (MODIS) at lkm (thermaJ) and 500m (visible, near- and medium- infrared wavebands)). Ground truth, crop nistories, classification, biomass development and yield will be required to understand the relationship between net primary productivity and yield and better assess harvest index as a function of crop condition. Representative areas for survey can be selected from a preliminary analysis of satellite images, and local knowledge. Economic measures of agricultura! productivity The simplest measure of economic productivity at a field scale is gross margin (GM) for a single product during a single phase ofthe crop rotation. The system may require estimates ofGM from severa) seasons to cover all phases of a farming system. For areas that contain different production systems, a composite measure is standardized gross value of product (SGVP) in which the price of the product is converted to the equivalent price of a standard crop, such as rice, then converted to the world market price. Expressed in a formula: "" . (local price}v . SGVP= LJ (Area x Yzeld x . orld marlcet przce) Eochcrop base przce The utility of SGVP may be questioned since it includes no estímate of costs, and therefore attributes average total benefit of all farming inputs to water. The full range of economic benefits from agricultura) production extends far beyond the simple measure of local production, to include indirect and broader impacts (Hussain et al. 2005). Multipliers of economy- wide farm 1 non-farm multipliers vary widely. Estimates in India suggest a multiplier as low as 1.2 for local schemes up to about 3 for the country as a whole. Multipliers tend to be Jarger in developed economies, estimated as high as 6 for Australia (Hill and Tollefeson, 1996). Hussain et al. (2005) point out that the most meaningful measure is ofmarginal value, that is, the additional value that is created when water is added (or lost when water is not available). WP assessment is more directly linked to problems in water-scarce or water-costly situations than in systems which are supplied with plentiful, low value water. WP is most meaningful as an indicator as water resources become increasingly scarce. The range ofproductivity-based indicators is summarized in Table 2. Table 2. Productivity indicators. Water productivity-based indicators Indicators Average product per unit of water Average gross value of product per unit of water Average gross margins per unit o f water Average gross net value of product per unit of water Val u e of marginal productivity of water. Note: CommonJy used denominators for calculating water productivity based indicators are amount of water diverted/supplied, water applied, gross inflow ofwater (rainfall plus irrigation), and crop evapotranspiration (Et). 179 Estimates of marginal value may be necessary where assessment is needed to identify 'optimal' distribution amongst contrasting users. Oweis and Hachum (2003) cite the benefit of supplemental irrigation in these terms and demonstrate marginal WP of up to 2.5 kg/m3• The concept of marginal value is reasonably standard in resource-based economies but data on which to evaluate it is difficult to find beyond research stations. It may be possible to derive crop production functions that estimate the contribution of water to productivity (see, for example, estimates for Rechna Doab in Pakistan; Ahmad et al. 2005). Assessment to emphasize pro-poor outcomes might also weight assessment to account for the increased value of benefits in low in come groups. This argument is made on the basis that • Income has diminishing marginal utility in purely economic terms; • If the intention is equitable income distribution, a dollar generated on behalf of a low income earner is worth more than one generated for a richer person and • On a one-person, one-vote principie the per person benefit counts more than the per dollar benefit. On the basis of this analysis, the relative value of fisheries and forests can be much greater than initial analysis suggests because they are often of great importance to the landless poor and marginalized people. Non-economic measures of agricultura! productivity Non-economic benefits from water by agriculture can be significant factors to include in assessment oftotal WP. They can, however, be difficult to evaluate- each type demanding complex methodologies to assess complex benefits. Environmental benefits can include direct product (including protection of fish resources), indirect benefits (e.g. impact on carbon stocks) or environmental flows - namely, flows deemed necessary for proper function of basin processes (Smakthin et al. 2004). The potentials for payment for ecosystem services (PES) appear to be increasing as more effort is put into practical evaluation and implementation (Farber et al. 2002; Kozanayi, 2002). As explained in Paper 1 of this series PES exemplifies the benefit of building social capital through management of common water resources. The final non-economic benefit that we mention here is the political capital that accrues through agreements to share or trade water resources or their products (such as hydro-electric power). Analysis by Wolf et al. (2003) of reported events involving trans-boundary basins indicate these predominantly construct political capital. Valuation systems are inextricably Iinked to the attitudes people have towards water, ranging from prívate, depleting uses to common, observable (non-depleting) attributes (Groenfeldt, 2003; Tumer et al. 2004). Difficulties in quantification arise when outputs are difficult to value or when output quantities are expressed in different units. Sorne issues that may require specific attention include: • Assessment of WP in complex livestock-based farming systems. This would need to include exchange ofplant and animal products around the system (see, for example, Peden et al. 2002). • Forest and agroforestry systems, which may provide ecosystem services, and be of unusual importance culturally or beca use of biodiversity considerations. • WP of fisheries and other aquatic systems, for which both output and consumption may be very difficu lt to quantify, yet provide essentiallivelihood support to the world's poorest people. One suggestion is to adopt a broadly-based indicator of water wealth that portrays the in come per m3 on a per capita basis. Per capita income, however, does not estímate the total support provided by water so ultimately does not relate to the problem that more food will need to be produced for more people witb less water. For example in the Lakes region of Kenya and Uganda rainfed agriculture supports a very high density of people, each with a small per capita wealtb. Total WP is far higher tban per capita WP. Another approach may be to take account of the number of people supported by a given water resource and the level 180 to which they are supported, using standard measures of livelihood support, such as Human Development Index (see Maxwell, 1999) or Basic-Needs Index (see Davis, 2003). Estimating the denominator: Water consumed A key distinction when computing WP is to differentiate between water input to an agricultura} system and water depleted by it. WP is estimated from the amount of water directly consumed by the agricultura! system (that is, evaporation and transpiration), not simply the amount ofwater supplied. This distinction is increasingly important as we move upscale from field to farm to basin, because water that is taken into a system, but not consumed, is available downstream and hence is excluded from calculation (see Molden, 1997). In measuring depleted water, we account for flows not used by the crop and returned to the hydrologic system. Quality of downstream water is potentially an important factor. Activities that damage water quality effectively reduce or even remove water that would otherwise be available to downstream users. Estimating water input can be confounded by not being able to define contributions from shallow or deep groundwater, although if available, water table modeling can assist. Another problem is not knowing the extent of run-on to rainfed lands from surrounding catchment areas. lt is also possible that there will be varying amounts of soil water carried over between seasons, depending on the year and the timing of rainfall. For example, we would expect all soil moisture in the root zone to be depleted every year in the Karkheh basin, with its strong pattern of winter rainfall and very high rates of potential Et in the dry summer. Water balance The basic expression of water balance is (input - output), accounting for change in water stored in the system: Q¡o = Qout + ~S (1) Where: Q;n includes rain, groundwater and surface-supplied irrigation and run-on, Qout includes runoff, drainage and evapotransp.iration, and ~S is change in soil water content. At the field scale, the key term is evapotransiration, considered as: Et = P + 1 + G + Q - ~S (2) Where: Et = evapotranspiration, that is evaporation from soil and water surfaces plus crop transpiration P = rainfall 1 = irrigation inflow G = net groundwater flow Q = runon (positive) or runoff (negative) ~S = changes in soil water content within the root zone Sorne components may not be relevant and be removed to simplify evaluation (e.g., no irrigation in rainfed farming, no run-on (incoming overland flows) or no capillary rise from high water table). Using both actual Et and net water supply as denominators can help define the context and options for management. 181 Direct and indirect measurement of Et Et of crops is routinely inferred for large areas from more easily measured climatic variables (for details see Linacre, 1977; Allen et al. 1998). Quantitative estimates of consumptive water use by crops over large areas is possible using the Surface Energy Balance Algorithm (SEBAL) method. This determines Et as a residual ofthe energy balance using routinely available weather data in conjunction with satellite-sensed thermal radiation. Remote sensing offers the chance to represent land use and its spatial variation accurately, to determine actual Et (Eta) and possibly to fill gaps that there may be in the coverage of rainfall data. Eta is obviously a better measure of water consumption by agriculture than potential Et (Etp), which assumes water is freely available and that the crop canopy remains fully developed and active. However there are a number of challenges to be addressed: • SEBAL relies on cloud-free imagery of cropped areas, • Sub-pixel disaggregation of land use (between crops and between cropped and fallow land), when using lkm or SOOm pixel (MODIS or Advanced Very High Resolution Radiometer (A VHRR)) data, • Corresponding sub-pixel disaggregation and attribution of Eta to each land use, or aJtematively to land use defined by higher resolution imagery (Landsat at 28.5m) and • The SEBAL procedure needs improved calibration for rainfed, pasture and forest land covers. New research is providing sorne insight on the estimation errors. Estimating consumptive water use by simulation modelling lt may be possible to represent the effect of climate variation on rainfed-crop WP by coupling a weather generator with crop simulation models. This has been done for large areas using the MarkSim procedure (Jones et al. 2002) coupled to the Decision Support System for Agrotechnology Transfer models (DSSA T, Hoogenboom et al. 2004) by Díaz-Nieto et al. (2006). Results can be spatialized in GIS using exhaustive pararneterization of model inputs. An altemative approach is to establish the spatial distribution of a small number of ' typical' soil profiles for which more exhaustive modeling results exist (Pracilio et al. 2001). The purpose ofthis would be to identify theoretical benchmarks of crop WP from which may be identified intrinsic factors liable to reduce water productivity. Allocation scenarios can be simulated by changing the balance of land under rainfed and irrigated conditions, or by adjusting water supply inputs through: • Rainwater harvesting, • Soil water conservation practices, • Supplemental irrigation, • Changing surface or groundwater allocations and • Conjunctive use policy to balance demands for surface and groundwater. The estimation of Et from water input data can be complicated by not being able to define contributions from high water table (although water table mapping will assist, if available) and not knowing the extent of run-on to rainfed lands from surrounding catchment areas. The soil water storage term is normally assumed to make an insignificant contribution to seasonal water use. For example, we would expect all soil moisture in the root zone to be depleted every year in regions with strong pattems of winter rainfall and very high rates of potential Et in summer such as the Karkheh 182 and upper Volta basins. However, it is possible that in sorne situations there might be carry-over of soil water between seasons in regions such as the Mekong basin, depending on the timing of rainfall between years. More complex 2- and 3-dimensional modeling may be necessary to understand the consequences of land- use change on water availability and consumptive water use. Where the system is govemed by surface water supply with Jimited groundwater, a simple node-link model like the Stockholm Environment lnstitute's water evaluation and planning (WEAP) system may be adequate to represent water budgets. If the system is dominated by rainfed agriculture, then a modellike the USDA soil and water assessment tool (SWA T), which integrates land use and hydrology may be preferred, although there may be problems in representing groundwater and surface water diversions. Higher-dimensional hydrologic models such as TOPOG (Dawes and Hatton, 1993) may be used to represent water balance within spatially-variable landscapes. More complex process-based models (such as the Danish DHI Water & Environment MIKE- SHE model) integrate all process, but present very serious challenges in calibration and parameterization, due to extensive data requirements, often related to soil characteristics. There are intermediate solutions, such as the New South Wales Department of Land and Water Conservation integrated quality and quantity model (IQQM), which is basically a node-link model with more advanced hydrology options for catchment yield, ungauged inflows and storage (see Hameed and O'Neill, 2005). A long history of development and application of such models can be found in the literature. However, the data requirements may be daunting. A major lesson seems to be 'proceed with caution', since propagation of error within data-hungry models can render complex results meaningless. lfthere is significant groundwater use, and saJinity is an important factor, then the integrating model should be a groundwater model, incorporating a salt transport module (e.g. the USGS modular three-dimensional groundwater flow model MODFLOW with the MT3D module). Creating groundwater models is a very time and data intensive exercise, and is usually limited to well-defined areas. Jt is highly unlikely that groundwater models can be built and calibrated at whole basin scale. The recent publication by the International Water Management lnstitute (IWMI) on the Zayendeh Rud basin in Iran (Murray-Rust and Selemi, 2002) provides a good example of the integration of models at different scales using simple spreadsheets as Iinks, although it is a little superficial on the integration of groundwater Water accounting The problem of estimating water consumption becomes more difficult for large, heterogeneous areas that contain complex mosaics of land uses. Discrepancy of meaning between WP of different uses can obstruct comparison of different water users within a single area. To simplify this, the method of water accounting may help track different flow paths ofwater depletion (Molden, 1997). Water accounting tracks the movement of water volumes within a field, an irrigation system or a basin according to four basic designations (Figure 1 ): • Water inflow (positive) • Change in storage (positive or negative) • Depleted water, that is water used or removed in either process (e.g. growing and processing food) or non-process ( e.g. depletion by evaporation from soil surfaces and ditches, or deep percolation to groundwater that is non-recoverable) or non-beneficia! (transpiration by weeds; washing motor vebicles). 183 • Outflow that is either committed as direct input into sorne downstream application or is available for use downstream (utilizable) or not (non-utilizable, as in the case of satine or contaminated water). Volumes in each category are measured (e.g. irrigation inflows), inferred by modeling (drainage outflow) or inferred from other data (e.g. use of remete sensing to estímate Et). 1 1 1 1_ - - - - - - - - - - 1 r--, --- 1 __ __ . ____ Uncommitted ~ ··~"·/~~ o 'E 5 Removal from f Addition to r--~ L . • Figure l. Water accounting framework (Molden et al. , 2001 ). WP of rainfed cropping systems Rockstrom et al. (2003) provide tables of consumptive water use for a range of tropical and temperate crops, based largely on observations from the 1950s to 1970s to compute WP from published values for crop water use efficiency (see also Rockstrom et al. 1999). Rockstrom et al. ( 1999) observe a wide range of WP around the universal average of about 0.7 kg/m3 of green water. (non-irrigation water use by agriculture). Within-field variation in yield is even greater (hence WP), suggesting substantial scope for improvement of WP. Water use efficiency of dryland crops has been studied for over 90 years. Yield data are widely available and in them attention is generally focused on estimating the denominator WP, the water consumed. At its most basic, consumptive water use (Et) is expressed as growing season rainfall and soil water changes: Et = P growing season + ~S (3) No account is taken of losses/transfers of water by runoff, nor of losses through deep percolation beyond the rootzone, both ofwhich wiU reduce WP at any given point. In dryland systems, changes in soil water at the beginning and end of growing season may be assumed to be insignificant, so that water consumption is simply estimated as rainfall during the growing season. 184 The compilation of results of water use by dryland cereal crops by Sadras and Angus (2006, Figure 2 below), shows several interesting features: • An intercept of about 60mm (range 80-11 O mm according to site characteristics), attributed to evaporative loss, • An overall maximum conversion efficiency of about 22 kggra;nlmmwa1er (equivalent to WP of 2.2 kggrain/m3, somewhat higher than the es ti mates of WUE collated by Rockstrom et al, 1999) and • A large spread of data below the maximum line, demonstrating the potential gains that could be achieved by better agronomic management. • Clina Loess Plateau • Mediterranean Basin 6 n=691 o • North Amerk:an Great Plains o o oe o SE Australia o • • 5 - 22 x (water use- 60} _4 • C'D • :E • • e-3 • 32 Q) }:: o o 2 .... • • • • o 1 • • • o o 100 200 300 400 500 &X) V\éter use (rmV Figure 2. Yield ofwheat as function ofthe amount ofwater evaporated and transpired (Source: Sadras and Angus, 2006) WP of irrigated crops Molden et. al. (200 l) analyzed WP of two irrigated systems - Chishtian in the Indus basin in Pakistan and Bhakra in the Ganges basin in India. They showed that there are marked differences in yields, and hence WP, with the system in India reporting higber values (Table 3). They attributed the higher productivity of the Indian system to higher land productivity and deficit irrigation. 185 Table 3. Units and their values for indicators of agricultura! water productivity. lndicator value Indicator of agricultura! WP Units Bhakra Chishtain Cropped area 103ha 2945.0 103.8 Wheat yield ton/ha 2.3 1.4 Rice yield ton/ha 3.0 2.1 SGVP US$/ha 782.7 413.3 Wheat yield per unit Et k:g/m3 1.1 0.6 SGVP per gross inflow SGVP / Gross inflow 0.12 0.06 SGVP per available water for irrigation SGVP /A W irrigation 0.15 0.06 SGVP ~er ~rocess consum~tion SGVP/ETa 0.17 0.07 A procedure has been developed by scientists at IWMI for determining the water productivity of irrigated crops as follows: l. Map irrigated areas and crop types within the surface water 1 groundwater system • Identify conjunctive use areas with the irrigation system • Map high water table areas (secondary data) • Obtain crop yield data through appropriate combinations of secondary (administrative or hydraulic district) data or from primary crop survey. • Obtain data on straw and green fodder production and utilization from irrigated crops, (usually frorn primary survey). • Determine livestock holdings and fodder use (by survey) 2. Overlay irrigation networks, and determine flow data for prirnary, secondary and possibly tertiary canals. • Select units for investigation, where sufficient water supply data exists • Estímate gross irrigation inflows 3. Obtain and spatially interpolate rainfall data. Using secondary data, determine typical values of e:ffective rainfall (that retained in the root zone oras surface storage in the case of rice) 4. Obtain canal flow data and determine seasonal surface water supply. Where flow data are not generally available at lower levels of the distribution network, it is possible to develop and apply disaggregation techniques to estímate the net local supplies from canal head flows (Ahmad and Bastiaansen, 2003) 5. Survey groundwater pump locations, capacities and average operating hours to determine groundwater supplies. • Where necessary, apply more advanced procedures to estimate net groundwater contribution (see PhD thesis by Ahmad, 2002), using remote sensing and soil-plant-water models. 6. Estimate Eta using SEBAL for each crop season (Droogers and Kite, 2001). Disaggregate Eta by cropping system. 7. Calculate land productivity (LP) in terms ofGVP and gross margin. 186 8. Calculate water productivities (WP), with respect to total supply and Eta: • Physical production (kg) • Gross value (SGVP) • Gross margin 9. Identify innovative water use practices where WP is low but LP is high and vice versa. 10. Calculate water productivity at larger scales of irrigation system and basin, using the depleted and process fractions of water supply (Molden el al. 200 1) 11. Determine system and basin average WP across all agricultura! uses. Livestock systems Peden el al. (2002, Figure 3) illustrates the complexity of accounting for water use in livestock systems in A frica. Figure 3. Framework for improving water productivity of livestock (From Peden et al. , 2002). Multiple benefits of livestock systems include meat, milk, hide/wool, draught power, and drought protection. These benefits will be realized within complex farming system at different times (e.g. draught power is required for ploughing; the sale of animals to help buffer incomes is expected to occur only occasionally) and space (animals are moved large distances between grazing and between grazing, fattening areas and markets). Evaluating the transferred benefits within such systems is consequently difficult, and would require an estímate of net gain for the area for which water consumption is estimated. Consumption of water occurs both directly through stock watering and making downstream water non- utilizable through pollution and indirectly through the production of feed as crop, sown pasture or as rangeland. In rainfed farming systems, grain is on ly one output ofvalue to the farmer- others include green fodder and dty fodder (straw and stubble). In pastoral systems, the value of green biomass is optimal ata certain stage of growth and it is common to convert estimates of green and dty biomass into estimates of 187 digestible dry matter (DDM). It may be possible to combine estimates of grain, green fodder and straw according to DDM basis, such that total production is expressed as: Production (kgDDM) = Grain (kgDDM) + Green fodder (kgDDM) + Dry fodder (kgDDM) Hill and Donald (2003) present a commercialized method of using satellite remote-sensing to quantify digestible pasture feed over large areas. It may be necessary to estímate marginal value to evaluate WP fully, since it is uncertain whether the water consumed by pasture would be more productive if used elsewhere? Certainly the low stocking densities of rangeland will present low benefit per m3 Et. The marginal value of pasture or rangeland vegetation is realized only when the feed resource is accessed by animals. Peden et al. (2002) examine options for alternative routes of water in livestock systems. Fish production systems There are few accurate assessments of the economic val u e of fisheries for most parts of Africa, Asia and Latín America (LARS2, 2003; Neiland, 2003). Furthermore, the special contribution of fisheries to food security and livelihoods is poorly represented in official statistics. Va/uation of benefits from jisheries Methods to evaluate the full range of benefits from fisheries and aquatic resources are summarized by Bené and Neiland (2003) according to the following categories. These are specifically intended to address the complex issues associated with changes to fisheries and aquatic resources: l. Conventional economic analysis • Economic efficiency analysis. Seeks actions which maximize social welfare in comparison with costs. • Total economic value. Acknowledges use and non-use values (see Figure 4). 2. Economic impact analysis: Assesses effect on specific variables 3. Socio-economic analysis: Distributional analysis of winners and losers from changes 4. Livelihood analysis: Broader analysis ofmultiple attributes that support sustainable livelihoods ~ Use Value Aquatic resources r- ..... Non-use Value Direct use value Indirect use value Option value Bequest value Existence value ) e.g. Harvesting of fish, aquatic and timber resources e .g. Ecological support function of inundated forest for ftsheries e.g. Harvesting opportunities on a later occasion e.g. Harvesting opportunities by future generations e.g. Knowledge of continued existence of the aquatic resources Figure 4. Total econornic value and valuation methods (from Barbier et al. , 1997) 188 Demand for water by fisheries Baran et al. (2001) related fish catches to water levels in the Tonle Sap river in the Mekong basin, calculating a loss of between 2500 and 5000 t of 'Dai' fish catch for each drop of 1 m in the average October levels ofthe river. The assessment ofthe denominator (water required to achieve a given outcome) for fisheries and aquatic resources is more complex than consumptive uses by crops, not least because - given the complex life cycles of both fish and their feed - severa! years' measurement are required to determine the actual requirement. Welcomme (2001), working in the Niger, concluded that at least 14 years are required to eva\uate the impact of \ow flows on fish stocks. See Dugan et al. (2005) for more details. Severa! issues are relevant to the demand for water by fisheries: • River fisheries require substantial non-consumptive volumes of water to provide suitable environments for growth and breeding. • There are opportunities for dual-use of irrigation infrastructure for fisheries (integrated agriculture- aquaculture - IAA), which can provide substantial supplements to both incomes and food security (Renwick, 2001). • Development of rice-fish culture can significantly increase income and soil fertility in deep flooded paddy rice, without increasing water consumption. However, these gains have to be offset against increased cost of buying more water, especially if it is priced at its real cost to encourage more efficient water use. Direct consumption of water by aquaculture occurs through evaporation and seepage. The forrner can be estimated from data of pan evaporation over the time required to produce a given weight of fish. For example, Brummett (2002) showed that integrated aquaculture in Malawi produced up to 264g/m2 of footprint, approximating to evaporative consumption of between 0.2 and 0.7 m3 of water for a 100 day production cycle ( 1.3 to 0.3 8 kg/m3). This figure needs to be evaluated in relation to the price of fish products, which are norrnally readily available for traded products, and other costs of production that can be quite high. lmpacts of intensive aquaculture on downstream water quality should also be considered. Tree systems Benefits from tree-based systems include both timber and non-timber products. Timber products are evaluated as m3/ha, or $/ha. Non-timber forest products comprise a wide array of animal and plant products, often of particular value to the poorest people that live in or on the margin of forests. Like fish resources, these tend to be under-valued by conventíonal economic analyses. lt is possible that evaluation techniques similar to those used for fisheries could be used. With respect to consumptive water use, a huge literature exists of forest hydrology, including sorne more recent evaluations that question the hydrologic benefits routinely attributed to forests (see Bruijnzeel et al. 2006, Mulligan and Burke, 2005). S pace does not permit review of the methods of evaluating the scale of positive and negative hydrologic effects of forests. Multi-scale estimation of WP We envisage WP will be estimated at 3 scales 1: • Whole basin estimation of WP: Coarse-resolution estimation of WP using readily-available productivity data at national scale or adrninistrative district level. Whole-basin estimates of well- 1 Given the complexity of this subject, we propose it as the first of a series of technical papers to be developed during the life ofthe Basin Focal Projects ofthe Water and Food Challenge Program. 189 being derived from agriculture may be approximated from readily available data of crop production, rural population, the extent to which basic needs are met, the proportion of the population engaged in agriculture, etc., and compared with water consumption using Molden's (1997) water accounting methods. Wbile useful for broad scale estimation, this will not provide the detailed insight necessary for further analysis. • Small-area estimation of WP: Detailed estimation of WP is envisaged from small case study areas, where these exist. These will produce high-resolution crop, fisheries and livestock productivity/income data, which can be combined with detailed estimates of water balance. Such studies will provide valuable detailed insight into variations of WP within farming systems, and of the hazards and limitations that constrain increases in WP. Where reasonable, these estimates could be used to represent the WP of farming systems. • Aggregated estimates of partial WP: Having estimated partía! WP for contrasting disaggregated agricultura) systems, it is necessary to consider how to estímate WP for aggregated farming systems at the sub-basin leve! or, to provide a detailed picture over whole basin or sub-basins. The following approaches are offered for consideration: • Classification into sub-units: Sub-divide the area into n parcels of i classes, defined according to Jand use, agroclimatic zone or other classification. Aggregate individual estimates of WP for n land uses in a single measure for a larger area as l:(WP;.A;). This approach may work well for small or moderate areas, for which reasonably secure estimates exist of each WP. For larger areas, the approach is likely to prove difficult because: • Many areas are likely to have missing data; • Different land uses may use contrasting valuation systems; • Estimation in sorne land uses will be seriously affected by seasonal variation and • Spatial resolution will vary between different land uses, hence apparently equivalent land units will contain different degrees ofuncertainty. • 'Hot-spot' approach: Broad estimates of WP are offered, for regional comparison, supplemented by more detailed information about specific areas known to be of particular interest. • An integrated approach, whereby coarse-resolution estimates of WP are used as a ' background' for overlays of more detailed estimates, resolved according to known variations of land use, agro- ecological zone, terrain position, erosion intensity or other factor that has a known (if approximate) associatjon with WP. Coarse resolution and detailed estimates of WP may be combined using a probabilistic Bayesian Network approach (see Lacave and Diez, 2000). 190 References: Ahmad, M.D. (2002). Estimation of net groundwater use in irrigated river basins using geo-information techniques: A case study in Rechna Doab, Pakistan. Ph.D. Thesis, Wageningen University, the Netherlands. http://www.gcw.nJ/dissertations/3334/dis3334.pdf. Ahmad, M.D. and Bastiaanssen, W.G.M. (2003). Retrieving soil moisture storage in the unsaturated zone from satellite imagery and bi-annual phreatic surface fluctuations. Irriga/ion and Drainage Systems 17: 141-161 . Ahmad, M.D., Bastiaanssen, W.G.M. and Feddes, R.A.. (2005). A new technique to estímate net groundwater use across large irrigated areas by combining remote sensing and water balance approaches, Rechna Doab, Pakístan. Hydrogeo/ogy Jo urna/ 13:653-664 Allen, R.G., Pereira, L.S., Raes, D. and Smith, M. ( 1998). Crop Evapotranspiration - Guide/inesfor Computing Crop Water Requirements. FAO Irrigation and Drainage Paper 56. FAO, Rome. ISBN 92-5-104219-5 Baran, E., van Zalinge, N. and Bun, N. P. (2001). Analysis ofthe Cambodian Bagnet ("Dai") flshery data. ICLARM, Penang, Ma1aysia, Mekong River Commission Secretariat and Department of Fisheries, Phnom Penh, Cambodia. Barbier, E. B., Acreman, M. C. and Knowler, D. (1997). Economic valuation ofwet/ands: a guidefor po/icy makers andp/anners. RamsarConvention Bureau, Gland, Switzerland. ISBN 2-940073-21-X Béné, C. and Neiland, A. E. (2003). Contribution of ínland fisheries to rural livelíhoods in A frica: Empírica! evidence from the Lake Chad Basín areas. In: Welcomme, R. and Petr, T. (eds.) Proceedings of the Second lnternational Symposium on the Management of Large Rivers for Fisheries, Vo/ume JI. pp. 67-77. FAO Regional Office for Asia and the Pacific, Bangkok, Thailand. RAP Publication 2004/l7. Bouman, B. (2006). A systems approach to the analysis of crop water productivity. Agricultura/ Systems. In review. Bruijnzeel, L.A., Burkard, R., Carvajal, A., Frumau, A., Kohler, L., Mulligan, M., Schellekens, J., Schmid, S. and Tobón, C. (2006). DflD Project R7991 Final Technical Report. 31 January 2006. Brummett, R. (2002). Realizing the potential of integrated aquaculture: Evidence from Malawi. In: Uphoff, N. (ed.) Agro-eco/ogical lnnovations- Increasing Food Production with Participatory Development. Earthscan Publications, Sterling, Virginia, pp. 115- 124. Davis, J. (2003). Choosing a Method for Poverty Mapping. FAO, Rome. ISBN 92-5-104920-3. URL: ftp: //fto.fao.org/docrep/fao/005/y4597e/y4597eOO.pdf. Dawes, W. and Hatton, T.J. (1993). TOPOG-IRM. l. Model Description. CSIRO Technical Memorandum 93/5. CSIRO, Australia. Díaz Nieto, J., Cook, S., Jones, P. and Skees, J. (2006). Developing a method of site-specific drought insurance for bean farmers in Honduras. Research Journa/ on Microflnance and Deve/opment (In press) Droogers, P. and Kite, G. (200 1). Estimating Water Productivity at Different Spatia/ Sea/es using Simula/ion Mode/ing. Research Report 53. Intemational Water Managernent Institute (IWMI), Colombo, Sri Lanka. Dugan, P., Dey, M.M. and Sugunan. V. V. (2005). Fisheries and water productivity in tropical river basins: Enhancing food security and livelihoods by managing water for fish. Agricultura/ Water Management. In press. Falkenmark, M. and Rockstrom, J. (2004). Ba/ancing Water For Humans And Nature: A New Approach In Ecohydro/ogy. Earthscan, London, UK. Farber, S.C., Costanza, R. and Wilson, M.A. 2002). Economic and ecological concepts for valuing ecosystem services. Eco/ogica/ Economics 41:375-392 Groenfeldt, D. (2003). The future of índigenous values: Cultural relativism in the face of economic development. Futures 35:917-929. Hameed, T. and O'Neill, R. (2005). River Management Decision Modelling In Iqqm. Presented at MODSIM 2005 Conference, Brisbane. URL: http://www.mssanz.org.au/ modsim05/paperslhameed.pdf Hill, M.J. and Donald, G.E. (2003). Estimating spatio-temporal pattems of agricultura! productivity in fragrnented landscapes using A VHRR NDVl time series. Remole Sensing Environment 84:367-384. Hill, H. and Tollefeson, L. (1996). lnstitutional questions and social challenges. In: Pereira, L.S., Feddes, R.A., Gilley, J.R. and Lesaffre, B. (eds.) Sustainability of Jrrigated Agricu/ture. pp. 47-59. NATO ASI Series. Kluwer, Dordrecht. Hoogenboom, G., Jones, J.W., Porter, C.H., Wilkens, P.W., Boote, K.J., Batchelor, W.D., Hunt, L.A. and Tsuji, G.Y. (eds.) (2003). Decision Support System for Agrotechno/ogy Transfer Version 4.0. Vo/ume 1: Overview. University ofHawaii, Honolulu. Hussain, l., Turra!, H., Molden, D. and Mobin-ud-Din, A. (2005). Measuring and enhancing the value of agricultura! water in irrigated river basins. lrrigation Science Journa/ (Submitted.) 191 Jones, P.G., Thomton, P.K., Díaz, W., Wilkens, P.W. and Jones, A.L. (2002). MarkSim: a computer tool that generates simulated weather data for crop modeling and risk Assessment: Version J. Centro Internacional de Agricultura Tropical (CIA T), Cali, Colombia. Kjine, J.W., Barker, R. and Molden, D. (eds.) (2003). Water Productivity in Agriculture. CABI, Wallingford. Kozanayi, W. (2002). The Use of Informal lnstitutions and Social Networks to Access and Manage Water Resources in a Semi-Arid Area in Zimbabwe. URL: http://www.ictn.trieste.it/-eee/files/WP2%20%20Kozanayi.odf Lacave, C., and Diez, F. J. (2000). A Review of Exp/anation Methods for Bayesian Networks. Technical Report IA-2000-01, Dept. Inteligencia Artificial, UNAD, Madrid. LARS2. (2003). Summary conclusions. In: Statement and recommended actions. ln: Welcomme, R. and Petr, T . (eds.) Proceedings of the Second lnternational Symposium on the Management of Large Rivers for Fisheries, Volume l . pp. Vll-X. FAO Regional Office for Asia and the Pacific, Bangkok, Thailand. RAP Publication 2004/16. Linacre, E.T. ( 1977). A simple formula for estimating evaporation rate in various climates, using temperature data alone. Agrícultural Meteorology 18:409-424. Maxwell, S. 1999. The Meaning and Measurement of Poverty. ODI Poverty Briefing, ODI, UK. URL: http://www.odi.org.uk/publications/briefmglpov3 .html Molden, D. ( 1997). Accounting for Water Use and Productivity. SWIM Paper l. System-Wide lnitiative for Water Management. Intemational Water Management lnstitute (IWMJ}, Colombo. Molden, D., Sakthivadivel , R. and Habib, Z. (2001). Basin-Level Use and Productivity of Water: Examples from South Asia. IWMI Research Report 49. Intemational Water Management lnstitute (IWMI), Colombo. Mulligan, M. and Burke, S. (2005). Global C/oud Forests and Environmenta/ Change in a Hydro/ogica/ Context. DflD FRP Project ZF0216. Murray-Rust, H. and Salemi, H.R. (2002). Water Resources Development and Water Utiliza/ion in the Zayandeh Rud Basin, Iran. - IAERl-IWMl Research Report. Neiland, A. (ed.) (2003). Valuation of tropical river fisheries: A global review. Report for the Worldfish Center, Penang, Malaysia. Oweis, T. , and Hachum, A. (2003). Improving water productivity in the dry areas of West Asia and North Africa. In: Kijne, W.J., Barker, R. and Molden, D. (eds.) Water Productivity in Agriculture: Limits and Opportunities for Improvement. pp. 179-197. CABI, Wallingford. Peden, D., Tadesse, G . and Mammo, M. (2002). lmproving the Water Productivity of Livestock: An Opportunity for Poverty Reduction. URL: http ://www.ilri.org/publicationslcdrom/integratedwater/ iwmi/Documents/Papers/Don.htm. Accessed 18 May, 2006. Praci lio, G., Asseng, S., Cook, S.E., Hodgson, G ., Wong, M.T.F., Adams, M. and Hatton, T. J. (2001 ). Estimating spatially variable deep drainage across a central-eastem wheatbelt catchment, Westem Australia. Australian Journa/ of Agricultura/ Research 54:789- 802. Renwick, M.E. (2001). Valuing Water in Jrrigated Agricu/ture and Reservoir Fisheries: A Multi-use lrrigation System in Sri Lanka. Research Report 51 , Intemational Water Management lnstitute (IWMl), Colombo, Sri Lanka. Rockstrom, J., Barron, J. and Fox, P. (2003). Water productivity in rainfed agriculture: Challenges and opportunities for smallholders farmers in drought-prone tropical ecosystems. In: Kjine, J.W., Barker, R. and Molden, D. ( eds.) Water Productivity in Agriculture. pp. 145-162. CABI, Wallingford. Rockstr6m, J., Gordon, L. , Folke, C., Falkenmark, M. and Engwall, M. (1999). Linkages among water vapor flows, food production, and terrestrial ecosystem services. Conservation Ecology 3:5. [onJine] URL: h ttp ://www .con seco l.orglvo 13/iss2/art5/ Sadras, V.O. and Angus, J .F. 2006. Benchmarking water use efficiency of rainfed wheat in dry environments. Austra/ian Journa/ of Agricultura[ Research 57(8) (In press). Smakthin, V., Revenga, C. and Doll, P. (2004). Taking into account environmental water requirements. In : Global- Sea/e Water Resources Assessments. Comprehensive Assessment Research Report 2. lntemational Water Management lnstitute (IWMI), Colombo, Sri Lanka. Turner, K., Georgiou, S., Clark, R., Brouwer, R. and Burke, J. (2004). Economic Va/uation of Water Resources in Agriculture: From the Sectoral to a Functional Perspective of Natural Resource Management. FAO Water Reports - 27. lSBN: 9251051909. Welcomme, R.L. (2001). Inland Fisheries: Eco/ogy and Management. Fishing News Books. Blackwell Science, Oxford, UK. Wolf, A.T., Yoffe, S.B and Giordano, M. (2003). lntemational waters: identifying basins at risk. Water Policy 5:29-{)0 192 Analyzing water poverty: water, agriculture and poverty in basins Basin Focal Project Working Paper no. 3 Cook, S.E., Gichuki, F., Turral, H. and Fisher, M.J. Challenge Program on Water and Food. Introduction: Why Analyze Poverty in Basins? The purpose of analyzing poverty data within basins is to identify: (a) How many people are affected adversely by poor availability or access to water, to what extent, and how this can be modified through improved agricultura! water management, and (b) Where low water productivity appears to induce poverty, and to what extent this is modifiable. The analysable relationship between water, agriculture and poverty can be explained as follows: The total gain that people derive from water used by agriculture is a product ofthe amount ofwater they take and the productivity they achieve per volume consumed. In sorne cases the total gain is limited by water availability, through either problems of availability or distribution. Otherwise the total gain is limited by water productivity, and poverty will be related to other problems such as land degradation, lack of land tenure, poor access to markets, inadequate labour or capital. All the above factors can be estimated at a range of scales within basins and used to analyze the existence of water-related causes of poverty. Such analysis is required within each basin to answer the following types of questions: • How many people are affected by drought, and to what degree? • How many could benefit from improved governance and infrastructure within a given irrigation domain? • What are the consequences of soil erosion within an area on water productivity in the basin as a whole? How will this impact on livelihoods? • What are the potential gains in water productivity within the basin? Analysis is required to identify the incidence and depth of poverty associated with attributes of agricultura! water management, and to provide a richer understanding of the nature of poverty and the degree to which it can be alleviated through improved agricultura! water management. This is a necessary step for devising evidence-based, targeted interventions. According to Black and Hall (2003) the water poor comprise 12: • Those whose livelihood base is persistently threatened by severe drought or flood; • Herders, fishers and farmers whose livelihood depends on environmental services of water that are not dependable beca use of upstream factors beyond their control; 12 Black and Hall (2003) also include those who are impoverished by Iack ofprovision ofwater for drinking and sanitation, which are potentially indirect consequences of poor agrciultural water rnanagement. 193 • Those whose livelihood base is subject to erosion, degradation, or confiscation (e.g. for construction of major infrastructure) without due compensation; • Fanners who cannot improve agricultura( productivity because ofthe high risk and uncertainties of markets and rainfall, which could be reduced by a little water at the right time; • Subsistence fanners who are constrained from higher value products such as fruit, vegetables or meat because of lack of access to water; and • Those living in areas with high levels of water-associated disease (bilharzia, malaria, trachoma, cholera, typhoid, etc.) without means ofprotection. Water-related poverty occurs because people are either denied dependable water resources or because they lack the capacity to use them, because they have insufficient land, degraded land, poor access to market, capital or other factor known to constrain development. Improvements in agricultura! water management that offer poverty alleviation include: • Provision ofwater resources to people who require it to sustain food production; • In creases in water productivity at the field or fann level through removal of production constraints; • Protection of envirorunental flows to increase dependability of supply; and • Protection from water-related health hazards. An important factor to consider is the collective vs. individual gains. Water is distributed within basins by surface and groundwater flows and irrigation infrastructure. Options to modify distribution include modification of the irrigation infrastructure and conservation of the soil and water resources. Other aspects of the water balance may be modified by changes in vegetation type as in converting native savanna to more productive sown pasture and/or increasing its growth rate as by correcting nutrient deficiencies. Net gain occurs when the benefit acquired by recípíents exceeds the losses suffered by altemative users, where these exist. Section 2 of this working paper clarifies the concepts that link poverty and agricultura( water management. Section 3 outlines a methodology to analyze poverty at basin scale to determine the effects of agricultura( water management. Conceptual relationship between water, agriculture and poverty Concepts thatlink water, agriculture and poverty The linkage between water, agriculture and poverty is complex and non-linear; not all poor people lack adequate water resources. On the other hand, not all people who live in dry areas are poor. Water resource endowment alone does not explain the state of poverty within basins; it is a necessary, but of itself not sufficient basis for explanation (Castillo et al. 2006). The purpose of analysis is to determine to what extent agricultura( water management explains poverty, in relation to other factors, and to what extent it can be improved. Water is used in a variety ofboth productive and consumptive activities and contributes to rural and urban Iivelihoods in many different ways. Lack of access to drinking water is itself an indicator of poverty, but the role of water in human well-being is far more complex than simply access to drinking water. Food crop production, fishing, agro-processing, and health can all influence and are influenced by the quantity and quality of available water. Rural upper catchments largely contribute to livelihoods by providing valued primarily ecosystem services to downstream urban, agricultura!, and industrial users. As the principal water user, agriculture offers important, if complex, opportunities for improvement of Iivelihoods for both consumers and ' producers' ofwater. Many ofthese issues are detailed by Castillo et al. (2006). 194 The complexity can be simplified. For example, Figure 1 shows the complex pathways of water within a livestock management system in sub-Saharan Africa (Peden et al. 2002). It seems reasonable to simplify this concept and see it as flow through three systems: a hydrologic system; an agricultural system and a livelihood support system. The well-being that people derive from water therefore depends on the interaction between: (a) The water system, that determines availability and reliability; (b) The agricultural system that converts the water into livelihood support, through food, income or other attribute - this is defined by water productivity; and (e) The livelihood system that modifies access according to social relations, institutions or organizations (AIIison and Ellis, 2001 ). Water availability Oround &sol! water recha¡ge Water produetivity of system Livelihood support O rain & other foods v-'m3 . Animal feed IC""'5I Plant production wood & Flbre Wlld blodlverslty Meat mílk, eggs, kcal/m3 hldes, pOMJr Manure, Wealth Eco- Exportad water aervices DiSchar¡¡e Floods Degraded water Figure l. Framework for improving water productivity of livestock. (Modified from Peden et al., 2002.) The constraint that water places on well-being is attributable to two factors: its availability to people (as individuals or groups) and the agricultural systern in which they use it. People will derive well-being through the interaction between the resource and the agricultural production systern. The objective of analysis, therefore, is to determine evidence for relationships within a three-variable system in which poverty (which we define for now as the lack of ' water wealth ' ), is a function of the water availability and the water productivity of the agricultural water rnanagernent system that enables people to derive livelihood frorn it. We represent this in Figure 2 195 .············ Water availabiJjty Water wealth ~----------- Water productivity Figure 2. Representation of the components of water wealth. The remainder of Section 2 provides information about the functional significance of different agricultura] systems to poverty alleviation, indicating the ways in which productivity can be modified through agricultural water management. Significance of agricultura[ water management to poverty alleviation While water is the component that we seek to change, the reality is that change is effected through the agricultural system on which most of the world's poor depend. Nash (2005) reported that 63 percent of global population (and 73 percent of poor) live in rural areas. He further noted that even with rapid urbanization, more than 50% of the poor will rernain in rural areas by 2035. For these rural poor, agriculture 13 is their main source of livelihood. For most poor countries, agriculture is a major economic sectors accounting for 30-60 percent of GDP. Agriculture growth contributes to poverty reduction through four main pathways: • Household food self sufficiency for subsistence farmers, fishers and pastoralists; • Low food prices, particularly for the urban poor for whom the cost of food accounts for a large percentage of their in come; • Employment opportunities and high wages; and • Foreign exchange earnings that make it possible for governments to import goods at prices the poor can afford. Water is one ofthe main factors that constrain their agricu ltural output, income and profitability. According to UNDP ( 1997), about half of the poorest people in the world eam their Iivelihood in areas where water constrains agricultural production. Importance of crop production to the poor Crop production is the main agricultura! sub-sector in most countries. In most developing countries, crop production is carried out by smallholder farmers and is generally labor intensive. Low producer prices, low yields and high cost of inputs constraints its potential for getting farmers out of poverty. Agricultural 13 In this paper, agriculture includes crop, forestry (plantation and tree crops), livestock and fisheries production. 196 laborers are amongst the most poorly-paid workers and are generally under-employed due to the seasonal nature of labor demand for crop production. Nevertheless, for lack of altematives, it continues to be tbe main source of livelihood for most poor people. Analysis of major farming systems in Sub-Saharan Africa illustrates the importance of crop production to the poor, as follows (IF AD, 2002): • Cereal-root crop mixed and irrigated systems have low incidences of poverty mainly attributed to the higher yields and favorable prices for produce; • Agropastoral and forest-based crop production systems have the highest incidences of poverty mainly attributed to low yield and remoteness of the farms; • Mixed maize and agropastoral farming systems have high potential for poverty reduction mainly due to the high demand and tbe potential for yield increases; and • Approximately 43 percent of the population in the region depends on maize mixed ( 16% ), cereal- root crop (15%) and root crop (12%) farming systems, which occupy lO, 13 and 12% ofthe total land area in the region. Crop production is constrained by biophysical, technological, socio-economic and institutional factors, water being one of the most critica! factors. lt is generally observed tbat crop yields are low in areas: • Where soil nutrients are depleted by erosion and/or nutrient leaching; • Witb low, erratic and unreliable rainfall ; • That are water-logged and/or have high levels of salinity. In contrast, crop yields are high: • E ven in semi-arid areas so long as runoff is minimized and rainfall is well distributed; and • In adequately irrigated land as compared to that receiving inadequate irrigation or none at all. Importance offorest (tree) products to the poor Forest products include timber products (sawn wood, building material, wood-based fibres, fue! wood and charcoal) and non-timber products (food-stuffs and medicine). Harvesting and processing these products is labour-intensive and is a significant source of employment. Forest and trees can also provide valuable ecosystem services. Forest products are important sources of cash income and employment at certain times of the year and for certain groups of people. Forest values related to subsistence use, environmental services and other indirect benefits are generally not accounted for. World Bank forest strategy and policy document highlights the critica! role that forests play: (a) 1.6 billion rural people are dependent upon forests to sorne extent; (b) l billion out of 1.2 billion extremely poor depend on forest resources for part of their livelihoods; (e) 350 million people are híghly dependent on forests; (d) 60 million indigenous people are almost wholly dependent on forests; (e) Production of wood and manufactured forest products contri bu te more than US$450 billion to the world market economy; (f) The annual value of intemationally traded forest products totals US$150-200 billion; and (g) Globally, forest based industries provide about 47 million full time jobs (Nash, 2005). 197 Forests make a major contribution in the provisioning of water services, particularly to the poor. Forests buffer the effect of rainfall on stream flows reducing flood peaks and increasing dry season flows. This service is very beneficia! to the poor who have limited water storage facilities and who live in flood plains. Importance of fish to the poor For poor communities with good access to aquatic resources, fisheries offer complementary livelihood strategies as illustrated below: • High national importance as evidenced by (a) global trade of US$55-66 billion annually, with 50 per cent of the trade from developing countries; (b) national in come from license fees; and (e) multiplier effect to the national economy; and • Benefits to the poor as evidenced by: (a) the livelihood support for 30 mi Ilion poor fishers and their families, employing an additional 150 million people in developing countries in associated sectors, e.g. marketing, boat-building; (b) main source offood security for 400 million poor people; and (e) high potential source of altemative employment for rural poor through aquaculture. The poor engaged in fisheries comprise artisanal fishers and aquaculturalists. Artisanal fishers comprise approximately 8 million people of which more than half are engaged in sea fishing activities. They generally use un-motorized boats without decks and their activities are centered around landing sites from where they set out each day for sea or lake fishing. The following factors contribute to making them one of the weakest livelihood groups: • Lack of control over the water resource on which their fisheries depend and of capital. • Dangerous working conditions, which lead to high mortality rates. A proftle of artisanal fishers in Benin indicated that malaria is endemic and diarrhoea and respiratory infections are common, especially in the rainy season when people are weakened by food shortages (F AO, 2000); • Lack of investment capital and low profit margins hence low labour productivity. The southern Lake Volta fishing communities reported that increasing cost of production and transportation is eroding their profit margins. Their profit margins fell by 10-25% o ver a 1 O year period (Pittaluga et al. 2003); • Seasonal variation in fish availability and consequently uncertainty of income and food availability. The southem Lake Volta fishing communities reported that fishing contributes on average 70% of the household revenue and during the lean season (November-May) a large proportion of the families are unable to meet daily food needs (Pittaluga et al. 2003); • Poor fisheries management, which can result in over-fishing and eventualloss offish stock and • Water pollution that degrades the aquatic ecosystem and reduces productivity. Importance oflivestock to the poor Keeping livestock is one ofthe main livelihood strategies ofthe poor and food insecure and directly affects the livelihood of approximately 987 mili ion poor people (Table 2). Livestock contributes to the livelihoods of the poor in many and diverse ways. The relative importance of these different ways varíes between households, time of year and prevailing biophysical and socio-economic conditions. The main ones include: • Livestock is the main source of income for the poor in semi-arid and arid areas. Most ofthe income in semi-arid areas is derived from small animals- goats and poultry. In Mali, it is estimated that 78% of the cash income on small-scale mixed farms comes from livestock (Sissoko el. al. 1992 quoted by Livestock in Development, 1999). A study in Bangladesh reported that 40% of tbe landless households owned cattle and a Pakistan study reported more than 50% of the landless 198 1 1 1 1 1 1 1 1 households' income was derived from livestock (Subrahmanyam and Rao 1995, Kurosaki 1995, both quoted by Livestock in Development, 1999). The landless supported their 1ivestock production through the use of crop residues, other waste material and feed resource found in communaJ land and roadsides. • Livestock is a cherished way of storing cash as livestock can be accumulated in good times and sold off when the need arises. A study in Lesotho reported that in investing in livestock earned farmers the equivalent of 10% interest rate while a bank account lost 10% due to inflation (Swallow and Brokken 1987, quoted by Livestock in Development, 1999). This is also a good strategy in remote areas where banking services are not readily available. • In mixed crop-livestock farming systems, livestock plays a key role in enhancing productivity of the farming system of the poor households through provision of draught power and manure. Draft animal power reduces drudgery associated with land preparation and transportation of water and other farm inputs and outputs, while manure returns soil nutrients for maintaining soil fertility. • Livestock enhances livelihood security through diversification of the farming system. A diversified farming systems buffers the poor against shocks associated with drought, floods, diseases and market failures. A study in Nepal reported that the introduction of dairy buffalo into a village reduced the period of food deficit from eight to two months in a year. It also contributed to a reduction of the proportion of people in the village with inadequate food intake from 50 to 18% (Thomas-Slayter and Bhatt 1994 quoted by Livestock in Development, 1999). • Livestock production enhances nutrition status as it is a source of high-quality protein and energy as well as essential micronutrients. Assessing the contribution of water-related benefits through livestock is complicated by the multiple pathways, stores and products that link water to benefits through multiple uses (feed, direct consumption) and multiple benefits (meat, mi1k, eggs, hides, power etc). Quantifying the flow of water through intermediate benefits shown in Figure 1 can be extremely difficult, as can assessing the benefits of the products. Additionally, there are numerous constraints and livestock related hazards that obstruct the poor from benefiting from livestock. The main ones include: • Regulations restricting the keeping of livestock production in urban areas; • High cost of (jvestock and lack of credit facilities; • High risk of diseases and drought; • A vailability of feed resources; and • Low prices offered for livestock products. Table 1. Distribution of the poor livestock keepers by agroecological zones. So urce: Livestock in Development (1999). Agroecological zone Arid and semi-arid Temperate (including tropical highlands) Humid, sub-humid and sub- tropical Total Category of livestock producers Extensive Poor rainfed Landless graziers mixed graziers livestock 87 107 194 199 (millions) 336 158 192 686 kee rs 107 107 Analyzing the Linkage Between Poverty and Agricultural Water Management in Basins This section lays out a framework in which to analyze the linkage between poverty and agricultura) water management within basins. The analytical framework has three components: (a) Measurement of poverty within the basin. This process includes mapping at best resolution feasible to improve analysis against biophysical and socio-economic attributes. (b) Analysis of poverty variation against measurable attributes of agricultura) water management within basins, and (e) Modeling the current and future status of agricultura) water management in basins with respect to poverty alleviation. Measuring poverty within basins Concepts of poverty Lok-Dessalien (1998) provides an exhaustive review of poverty concepts and indicators. She argues that the distinction between different poverty indicators was important because poverty measurement and subsequent policy and program implications depend on the facets of poverty being addressed. For example, to address both temporary and chronic poverty, two sets of policies and programs would be required along with their indicators for establishing baseline conditions and monitoring progress. Likewise the definition of poverty determined the appropriate poverty measures, policies and programs to address it and corresponding indicators. She highlights the following poverty concepts: • Absolute poverty refers to inability to meet (food, shelter, education and health) needs that enable a person to enjoy a mínimum acceptable standard of living. The needs define the required goods and services and the value of these goods and services u sed to define the minimum income needed to acquire them - the income poverty line. • Relative poverty focuses on the inequality and uses income quintiles or deciles to compares the Jowest and upper segment ofthe population. • Objective poverty involves normative judgment of what constitutes poverty and what is required to lift people out oftheir impoverished state. • Subjective poverty puts emphasis on individual utility in terms of how much people value goods and services. The subjective poverty approach has led to the development of participatory poverty assessment methodologies. • Physiological deprivations concept is linked to the basic needs concept. Under this concept, people are poor because they lack income, food, clothing and shelter. Poverty-reduction strategies emerging from this approach focus on increasing income/consumption of the poor and their attainment of the acceptable levels of basic needs. • Sociological deprivation perspective argues that people are poor because of the underlying structural inequalities, inherent disadvantages and other factors that constrain access of the poor to credit, water, common property resources and information. These structural inequalities thereby hamper them from using the resources at their disposal to climb out of poverty. Hence, poverty is not just low consumption but also the lack of opportunities to lead valuable and valued lives. Poverty assessment Poverty measurement exists for all countries. The nature and source of poverty data affects its ease of analysis and is detailed in Appendix II. Analysis against water-related variables is constrained by two further factors: date of sampling and spatial resolution. 200 The date of sampling may cause problems of analysis if the data are out of date and do not provide an accurate current view or are asynchronous with the data with which they are to be compared. Within trans- boundary river basins, data of different age (and probably different character) will need to be combined in a single data-set. Effects of droughts or flooding are likely to be partially date-variable, and so may not be visible if poverty is measured at a time when effects were minimal. Conversely, poverty measurement immediately after a crisis may over-emphasize an effect which is only of short-medium term. The spatial resolution and definition of location may limit the quality of analysis against other factors that are highly variable spatially. Significant effects of localized biophysical factors, such floods, land-use potential, and groundwater availability or socio-economic factors, such as access to markets will be obscured if the resolution of poverty meas u res is at a regional leve!. Poverty Mapping Poverty maps (normally of absolute poverty measures) improve analysis with respect to water-related attributes within the basin, which are difficult to understand without acknowledging spatial variability. Poverty mapping has been developed in many countries and used to: (a) Target public interventions by identifying where the neediest population live, (b) Target emergency response and food a id programs, and (e) Jmprove transparency ofpublic decision making. Henninger and Snel (2002) highlight the value of spatial analysis to provide basin information on which to decide where, how, when to intervene. A very important factor with respect to basin analysis is that maps provide a common data-framework on which to model socio-economic, agricultura! and hydrological processes. Since many hydrological processes can only be represented effectively in spatial form, GIS provides a logicaJ analytical platform to which other analyses relate. Davis (2003) provides a comprehensive review of techniques of mapping poverty and food security, pointing out that at the time of writing there existed no gold standard of poverty mapping because of the wide array of applications. He groups the different methods according to: • Small-area estimation of o Household leve! data o Community leve! data (Bigman et al. 2000) • Weighted basic-needs index, using o Principal components analysis o Factor analysis o Ordinary least-squares • Combined qualitative-plus-secondary data (detailed in section 3.3) o Primarily based on qualitative assessment, (WFP, 2006) o Primarily based on secondary data • Extrapolated participatory approach o Calibrated participatory assessment (Ravnborg 1999, Kristjanson et al. 2005) • Direct measurement o Household survey data o Census data 201 Analyzing poverty variation with respect to agricultura! water management The general process of analysis comprises the following: • Defmition of the hypothetical 'model' that links poverty vanatlon with agricultura! water management within the basin, on the basis of a theoretical and contextua! understanding of the problem and an awareness of the data that are likely to be available. This step should also outline the method of analysis. • Acquisition of poverty data for the basin, where possible using spatial analysis to improve the resolution and reliability of the data using methods of small-area estimation. Detail is provided above about the rationale and methodology of poverty mapping. • In consultation with collaborators, assemble data of candidate explanatory variables. • Analyze the general and site-specific relationship between the two variables. Coudouel et al. (2002) provides many useful suggestions (and cautions) to guide analysis, which commonly employs regression of poverty measures against 'explanatory' variables. Conventional Poverty Analysis Coudouel el al. (2002) provide a comprehensive review of conventional methods of poverty analysis. Analysis is intended to identi:fY correlates against a range of poverty measures (e.g. income, consumption, inequality) that may help understand the general nature of poverty. Analysis normally uses a form of regression analysis to identify poverty etfects for specific groups, by disaggregating data according to geographical region, age, gender, employment or other factor contained in the data. Vulnerability analysis presents a special type of analysis, which looks ata measured decline in well-being that results from specific shocks. Analysis is confounded by geographical variables that are not accounted for in the regression model. This could be reduced by including dummy variables of location, or map-derived variables of access to water, drougbt or flood incidence, market access etc., wbere these are considered to be candidate variables. Analysis of poverty maps: Wby map? Tbe effort to map poverty and its explanatory variables can be justified by the following: (a) lmproved data resolution by interpolation and small area estimation techniques (b) lmproved coincidence of socio-economic and bydrologic data on a common GIS data platform. These variables are generated through different sampling and estimation techniques. (e) Visual representation of geographical pattems. Experience in generation and use of poverty maps has demonstrated the potential value of looking at both spatial and temporal dimensions of poverty (Table 2). Poverty mapping at high spatial resolution has identified pockets of poverty amidst areas of prosperity. More importantly, it has enabled explanation of variations in poverty incidence by comparison with other spatial attributes such as drought incidence. This last feature can be crucial because the effects of agricultura) water management are unlikely to be constant throughout an area and will require geographically localized analysis. Severa) other examples of spatial analysis of poverty against factors which include water-related attributes can be found in Hyman et al. (2005). Trend or change analysis, that is, comparing data from two of more sampling periods, enables the dynamics of poverty to be assessed, for example where people have moved out of poverty or ha ve been hit by natural or human-induced shocks. Figure 3 from Farrow et al. (2005) shows how poverty in Ecuador changed over the period 1991 to 200 l . Areas in the Andes improved, in the face of drought stress (see Farrow et al. 202 2005), as a result of concerted action by organizations to ameliorate problems of isolation. Conversely, areas on the Pacific coast deteriorated as a consequence of El Niño damage to income-generating plantation crops. Pinpointing these variations enables us to interpret poverty within the basin-specific bio-physical, socio-economic and institutional settings and therefore get a better understanding of the causes and appropriate interventions. Hazards that confront the unwary analyst include: • Assumed correlation between measured and non-measured well-being variables, e.g. consumption vs. income measures (Coudouel et al. 2002). • Use of the same inforrnation in explanatory and dependent variables, for example, if land quality is used in small-area estimation of poverty, it should not be used in analysis as an 'explanatory' variable. • Complex variance structures may be hidden within data covering large areas: Geostatistical analysis by Farrow et al. (2005) revealed complex non-random patterns in poverty data that, if undetected, would have reduced the value of 'conventional' regression techniques. Analysis by Leclerc (2002) at department, municipality or village leve! shows that the leve! of disaggregation can significantly modify the advice coming from analysis. • Non-stationarity of models: Conventional analysis of poverty may unjustifiably assume stationarity - leading to significant error (Coudouel et al. 2002). Analysis by Farrow et al. (2005) show strong geographically variation, with both positive and negative regression coefficients for the same explanatory variable at different locations. This may cause particular difficulties when ascertaining the geographical variation of significance of household leve! influences such as gender and age, which are unlikely to be analyzed at basin scale. Table 2. Explanatory variables &om poveny mapping in seven case studies. Source: Hyman and Imminck (2003). Mexico lndigenous groups Education Accessibility Population density Ecuador Accessibility Water availability El Niño Land tenure Nigeria Rainfall Vegetation (more analysis needed) Sri Laoka Access to land and water 203 Malawi Educational attairunent Non-agricultura] activities Dependency ratio Keoya Soil resources Rainfall and climate NDVI (vegetation growth rate) Access to education Accessibility to towns Baogladesb Educational attainment Availability of infrastructure Land tenure Flood-prone lands Soil suitability for rice cultivation Modelling the effects of changes in agricultura[ water management on poverty The converse process is to predict poverty distribution on the basis of variation of attributes that represent the effects of agricultura! water management. Davis (2003) provides a useful characterization of the two principal methods as used to map poverty: • Combined qualitative-plus-secondary data (detailed below) o Primarily based on qualitative assessment o Primarily based on secondary data Prirnarily gualitative information approach: This method uses qualitative information such as a land-use map as a basis for ' first-cut' categorization of explanatory factors. Davis (2003) describes two examples. The ftrst has been used successfully by the World Food Program Vulnerability and Mapping system (VAM) to target emergency aid. The method of Seaman et al. (2000) could be modified to map impacts of water-related interventions as follows: (a) Define agricultura! system zones for each basin (Dixon et al. 2001). System zones define areas containing similar combinations of agricultura! activities. (b) In each zone, defme major categories of livelihood support. (e) For each of these categories, determine information of the impacts of livelihood support of attributes of water-availability and water productivity known to be significant from analysis. (d) Use the above as a baseline from which to estimate the possible impacts of changes in water availability and /or productivity resulting from detailed study of individual factors. A second variant may be useful to focus on particular groups who are vulnerable within basins (e.g. fishers). Ln this method, mutually-exclusive, livelihood-strategy groups are defined by workshops of experts, following which the impact of changes in water availability and productivity would be estimated from institutional attributes. Water poverty index The water poverty index (WPI, Sullivan et al. 2003), subsequently modified to the water wealth index (WWI) attempts to define poverty that includes al! factors relevant to the livelihood support provided to the poor by water resources in five dimensions: • Per capita resource availability • Access to water; • Capacity to benefit; • Water uses; and • Environmental impact. Maps have been produced at national scale and sub-national scale. The WPI has also been applied to analyse community-level characteristics, but the feasibility of more detailed mapping may be limited. This concept has undoubtedly broadened the scope of examination, but the rigid definition of relative weights reduces its value as an analytical tool, especially since sorne of the factors that could be used to help explain poverty variation are used within the index itself. WPIIWWI may provide greater value as a diagnostic indicator for subsequent analysis. Falkenmark water stress index The Fatkenmark water stress index provides easily quantifiable measures that assume no direct association between poverty and water (Falkenmark and Widstrand 1992). This was modified by Ohlsson and Appelgren (1998) to include measures of social capital - that is, the ability and willingness of people to engage in activities that provide collective benefit - that seem likely to modify the ability to cope with 204 stress. Useful as a broadscale indicator of the imperative for action, this makes no distinction between impact and condition. Vu/nerability Severa) defmitions on vulnerability exist. IFAD defines vulnerability as the probability of an acute decline in access to food or consumption, which leads to " inability to meet minimum survival needs". This definition captures two main elements that need to be present simultaneously: the exposure to risk, and the inability to cope with it. Kasperson et al. 2001 defmed vulnerability as "the differential susceptibility to loss from a given insult''. They argued that vulnerability has three dimensions: • Exposure is a roeasure of the probability that a certain risk will occur. lt is related to both the presence ofthe risk in a given location and to people being in that location. • Sensitivity is influenced by both socioeconomic and ecological conditions, which together determines the degree to which a group will be affected by environmental stress. For example, people in poorer health condition are more sensitive to a health-affecting environroental stress than people in good health. • Resilience is the extent to which an individual or a community utilize coping and adaptation strategies to help them retain their basic properties under stress, recover from damage, and enact change to prevent future damage. The risk of severe poverty increases with membership in certain identifiable social and age groups that also suffer a hígher risk of perpetuating poverty into the next generation. The strongest predictor of poverty is inability to perform or lack of access to paid work or lack of access to productive resources. The groups mostly affected by poverty are: • Children, youth, and families with many children: The young face the highest risk of poverty and moreover the risk of poverty increases with the number of children in a family. Families with three or more children have a higher rate of poverty than those with fewer children, significantly affecting their long-term life prospects; • Single parent families, particularly female-beaded households in rural areas; • Families with unemployed members; • Agricultura) families, particularly in areas of low productivity; • Pensioners • Homeless families; and • Abused, neglected or abandoned children Primarily secondary data: The other hybrid method described by Davis (2003) is based on ' indicators ' and is typified by the famine early warning system (FEWS) promoted by the USAID. The method (modified form Davis 2003) comprises the following steps: (a) Determination ofthe principal water-related 'drivers' for which information exists over the basin; (b) Selection and transformation of indicators over the basin; (e) Weighting of indicators, based on analysis, expert judgment; (d) Ranking according to summed scores ofindicators; and (e) Mapping of indicator scores. 205 Comparison of analytical and modeling activities The analytical and modeling components contrast in the ways in which poverty is inferred to be related to agricultura! water management: • Identify the incidence of poverty and infer how much variation is associated with water management. This is called backward chaining: (jrom analysis of Y, infer the influence of X) and is useful to help explain the causes ofpoverty. An example is the analysis ofpoverty in relation to the intensity of drought in Ecuador (see Farrow et al. 2005). • From information about biophysical and socio-economic characteristics in the basin infer, from modeling, the impact on poverty. This is forward chaining: (from X, infer the likely status of Y). This is useful to represent targetable problems on the basis of prior understanding. The approach is used to predict poverty effects of vulnerability (see the Vulnerability and Analysis Mapping reports ofthe World Food Program, WFP, 2006). It is helpful to clarify the complementary use ofthese two processes. The frrst process helps to understand the causes of poverty that are related to agricultura! water management, and to quantify, as far as the data allow, the relationship between the two. The result is a model that can be used- within limits of plausibility - to link explicitly the relationship between poverty and other measurable attributes in the basin. The second process takes the best current understanding of causes to portray where water-related poverty, and changes to it, is likely to occur within the basin, given data about the basin. Summary: The complete analysis and representation of agricultura/ water management-related poverty in basins depends on four assumptions. The first is that poverty can be measured in sufficient detail to identify tbe effects of variations in agricultura! water management, should these exist. In sorne areas, data are likely to be of insufficient spatial resolution to compare with short-range hydrologic features. Temporal resolution may be insufficient to define the effect of changes or extreme events. The second is that agricultura! water management is a major controlling factor on poverty. Logically, a wealth of evidence exists from case studies to support this view: agricultural production systems support the poor and agricultura! production systems are influenced significantly by water availability and use. lt seems crucial to separate the constraint of water availability on agriculture from the effects of other factors that influence the benefit people acquire from it. The third assumption is that analysis of data from basins will reveal significant effects of agricultura! water management on poverty within different parts ofthe basin. Many factors may confound such analysis: lack ofhigh-resolution poverty measurement; spatial and temporal confusion of different data; poor quality data of 'explanatory' variables. The fourth assumption is that the basin system can be modeled to represent the current condition of people living in the basin, together with the likely impact of changes targeted by the analysis. Analysis will be required to provide maps and tabular data to support assessment of the state of water-related poverty, and the degree to which it is modifiable. 206 Appendix 1: Types and sources of poverty data Data for poverty analysis can be obtained from two major sources: service records and surveys. Service record data sets are the data collected by various government organizations such as: health service data on nutrition status, disease incidence, inpatient and outpatient visits; education data on school enrolment and performance; agricultura! statistics on agricultura! produce, prices and wages. Sorne of the limitation ofthese data sets include: limited coverage, questionable quality, not generally disaggregated and in most cases they are in a raw form requiring a lot oftime to pre-process. Housebold surveys are the basis for poverty data collection. Survey data contain information collected based on the needs of the study objectives and includes both quantitative and qualitative data. A poverty- study survey will in most cases yield a more comprehensive dataset than wil\ other types of survey. Quantitative methods tend to define poverty in externa! terms such as need deprivation and focus on measurable and observable parameters. Qualitative methods tend to use an interactive process to understand both the constituents and their sources of well being. The surveys include: • Living Standard Measurement Survey (LSMS) a large, multi-topic household survey comprising three sections: household, community and prices • Integrated Survey (IS) is similar to LSMS and was designed to collect data for analyzing impacts of structural adjustment on household. • Priority Survey (PS) coiJects data for identif)ring and monitoring population groups most affected by structural adjustment policies. • Household Income and Expenditure Surveys (HIES) core data sets include: household characterístics (size, structure, composition and activities of its members); household income (both individual and collective, in-kind, in-cash, paid and self employment); and househo\d expenditure (purchased goods and services, consumption of self-production). Depending on the coverage, the survey may also include: consumer prices, income distribution, inequality, poverty, savings, taxation, elasticity of demand for goods and services, and nutritional data. • Demographic and Health Surveys (DHS) that focuses on maternal and child health, fertility and fam ily planning, but also include education, occupation and knowledge data. • Consumer Price Surveys (CPS) carried out to assess the comparative costs at current prices of the same basic basket of goods and services over time. • Labour Force and Employment Surveys (LFES) that focus on the relationships between poverty and occupation and livelihoods. • Food Consumption and Nutrition Surveys (FCNS) that collect information on: (a) type and severity of nutritional deprivation; (b) consumption and production of certain foods; and (e) consumption expenditure and effects of subsidy programs. • Agricultura) Surveys (AS) that cover: (a) comprehensive statistics of agricultura! land, crops cultivated, irrigation, number and types of livestock; (b) benchmarks for improvements in crop and livestock production; (e) agricultural structure attributes such as size and distribution of holdings, extent of various forms oftenancy, agricultura! resources, production facilities and practices; (d) agricultura! machinery and inputs; and (e) food. 207 • Other specialized household surveys such as gender equity, education and literacy, housing, access to markets, schools and hospitals. Poverty indicators and their deriva/ion Single indicators Poverty line: The poverty line defines who are poor and who are not poor by establishing individual or household incomes or expenditures levels below whích they are considered poor. Poverty lines are usually established using one of the following methods: • The food energy method determines the consumption of a bundle of food items required to reach a mínimum agreed caloric intake. Regression of caloric intakes with income or expenditure levels then determines the income at which the mínimum energy intake is realized. This becomes the poverty line income or expenditure level. The method implicitly takes into account non-food expenditures. • The US$1 per day per person poverty line was set up to facilitate comparison of poverty situations of different countries or of different areas in the same country in a uniform manner. The US$1 is converted to local currency using purchasing power parity (PPP) índices, which are derived from the costs in constant US dollars of a national average consumption bundle. The index therefore does not reflect the composition or the relative prices of typical consumption items of poor households but rather national averages compared to a world average. lntemational comparisons are thus made on a fairly inadequate basis, and their results should be used with much caution. Poverty incidence of a given area is computed as the percentage of the total population that is poor. This measure is intuitively understandable but fails to indicate the depth of poverty for different groups of people. Poverty gap: Poverty gap captures the depth of poverty by assessing the income shortfall if all the poor had to have incomes equal to the poverty line. The poverty gap is defined as the income transfer required to lift the in comes of all poor exactly up to the poverty line and is expressed as the percentage of total income that needs to be redistributed. Composite Indicators Poverty index: Severa! poverty índices, (HPI, HDI, GDI, WPI) that combine different poverty indicators have been proposed (see table below). 208 Component indicators Longevity Knowledge HDI Lífe expectancy at birth Adult literacy rate: Combined enrollment rate GDI Female and maJe life expectancy at Female and male adult literacy birth rate: HPI Percentage of people not expected to live to forty Human Development index Female and male combined enrollment ratio Adult literacy rate Standard of living Adjusted income 1 capita Female and male earned income share Percentage of population without access to safe water; Percentage of the population without access to health services Percentage of undemourished children under five From a human development perspective, poverty is defined as deprivation of capabilities and opportunities essential for human development, which include material welfare, education, health, freedom of choice, and participation (Dzenovska, 2001 ). According to the 1997 Human Development Report, human poverty is the denial of choices and opportunities most basic to human development- to lead a long, healthy, creative life and to enjoya decent standard of living, freedom, dignity, self-esteem and the respect of others. Lack of any or all of these capabilities and opportunities constitutes poverty beca use provision of these capabilities is a desired end of human development process. Hence, from measures of human development, poverty is pinned on two primary indicators - the human development index (HDI) and the human poverty index (HPI). HDI measures the conditions of human development and its fluctuations while HPI measures deprivation ofhuman development and its fluctuations. HDI encompasses: life expectancy at birth; literacy rates and combined primary, secondary and tertiary enrollment; and adjusted income. HPI for developing countries encompasses indicators for the percentage of people not expected to survive the age of 40; percentage adult literacy; and economic provisioning in terms of the percentage of people who do not have access to health services, safe water, and the number of malnourished children under the age of 5. These índices do not measure all aspect of development and poverty and are aimed at identi:tying potential problem areas and for comparing general trends among countries, regions or population groups at a very aggregated leve!. These índices therefore fail to capture the distribution of the deprivation across income groups, social and ethnic groups and regions. They also fail to provide information needed to interpret the data more broadly. For example these índices fail to illuminate the political and historical context, which may be imperative for understanding the trends of development or deprivation as well as devising policy solutions. Human development approach challenges the assumption that economic growth is the primary vehicle for poverty reduction and asks what kind of economic growth is conducive to poverty reduction for all with a focus on quality and equity rather than quantity of economic growth. Technology-driven economic growth tends to favor the rich and increases income inequality among countries and within countries. 209 References: Allison, E.H. and Ellis, F. (200 1 ). The livelihoods approach and management of small-scale fisheries. Marine Policy 25:377-388. Bigman, D., Dercon, S., Guillaume, D. and Lambotte, M. (2000). Community targeting for poverty reduction in Burkina Faso. The World Bank Economic Review 14: 167-194. Black, M. and Hall, A. (2003). Pro-Poor Water Governance. Asían Development Bank. Castillo, G., Namara, R., Hussein, M.H., Ravnborg, H., and Smith, L. (2006). Improving Agricultura/ Water Management for Poverty Reduction: Challenges and Pathways. Comprehensive Assessment for Water in Agriculture. Draft report. IWMI, Colombo, Sri Lanka. Coudouel A., Hentschel J. S., and Wodon Q.T. (2002): Poverty Measurement and Analysis. PRSP Sourcebook, World Bank, Washington. VRL: http://povlibrary.worldbank.org/files/5467 chapl.pdf. Accessed 19 June, 2006 Da vis, B. (2003). Choosing a Method for Poverty Mapping. FAO, Rome. 48pp. Dixon, J., Gulliver, A. and Gibbon, D. (2001). Farming Systems and Poverty: lmproving Farmers' Live/ihoods in a Changing World. FAO Rome. 412 pp. Dzenovska, D. (2001): Thematic review ofNHDRs: Poverty. UNDP report. Falkenmark, M. and Widstrand, C.. (1992). Population and Water Resources: A Delicate Balance. Population Bulletin. Population Reference Bureau F AO (2000). The State of Food Insecurity in the Wor/d- 2000. F AO, Rome. 31 pp. URL: http://www.fao.org/documents/show cdr.asp?url file=/docrep/x8200e/x8200eOO.htm. Accessed 19 June, 2006 Farrow, A., Larrea, C., Hyman, G. and Lema, G. (2005). Exploring the spatial variation of food poverty in Ecuador. Food Policy. 30:51 0-531. Henninger, N. and Snel, M. (2002). Where Are the Poor? Experiences with the Development and Use of Poverty Maps. World Resources Institute and UNEP/GRIDArendal, Washington. VRL: http://www .povertymap.net/publicationslwherearethepoor/where are the poor.pdf. Accessed 19 June, 2006. Hyman, G. and Imminck, M. (2003). Resu/ts ofCGIARIFAO/UNEP Poverty Mapping Case studies. Presentation to ESRI Users' Conference, San Diego, 7-11 July, 2003. Hyman, G., Larrea, C. and Farrow, A. (2005). Methods, results and po1icy implications of poverty and food security mapping assessments. Food Po/icy 30:453-460 IF AD. (2002). Assessment of rural poverty: Eastern and Southern A frica. Jnternational Fund for Agricultura( Development, Rome. VRL: http://www.ifad.org/poverty/region/pf/ index.htm. Accessed 22 June, 2006. Kasperson, R.E., Kasperson, J.X., Dow, K., Ezcurra, E., Liverman, D.M., Mitchell, J.K., Ratick, S.J., O'Riordan, T. and Timmerman, P. (200 1 ). lntroduction: Global environmental risk and society. ln: Kasperson, J. X. and Kasperson, R.E. (Eds.). Global Environmenta/ Risk. United Nations University Press, New York. Krisganson, P., Radeny, M., Baltenweck, 1., Opgutu, J. and Notenbaert, A. (2005). Livelihood mapping and poverty correlates ata meso-leve! in Kenya. Food Policy 30:568-583. Leclerc, G. (2002). Poverty targeting with local and c/assic indicators: An attempt to integra/e mu/tip/e perspectives. Land Use Project, CIA T, Cali, Colombia. Livestock in Development. ( 1999). Livestock in Poverty-Focused Developmenl. Livestock in Development, Crewkeme, United Kingdom. 96pp. URL: http://www.theidlgroup.com/downloadsllivestock.pdf. Accessed 19 June, 2006. Lok-Dessallien, R. ( 1998). Review of Poverty Concepts and Indicators. SEPED Poverty Reduction Series. URL: http://www .undp.org/poverty/publications/pov red!Review of Poverty Concepts.pdf. Accessed 19 June, 2006. Nash, J. (2005). Sustainable development strategies in agriculture and rural development: A paper presented during the WTO Symposium on Trade and Sustainable Development held in Geneva, 10- ll October, 2005. VRL: http://www.wto.org/Englishltratop e/envir e/sym oct05 e!nash session2 e.ppt. Accessed 19 June, 2006. Ohlsson, L. and Appelgren, B. ( 1998). Water and social resourcescarcity: Alternative socially based approaches to assessment and management ofwater scarcity. URL: http://www.thewatemage.com/SoicalResourceScarcity.htm. Accessed 19 June, 2006. 2 10 Peden, D., Tadesse, G. and Mammo, M. (2002). lmproving the Water Productivity of Livestock: An Opportunity for Poverty Reduction. URL: http://www.ilri.org/publications/cdrom/integratedwater/iwmi/Documents/Papers/Don.htm. Accessed 18 May, 2006. Pittaluga., F., Braimah, L.l., Bortey, A., Wadzah, N., Cromwell, A., Dacosta, M., Seghieri, C. and Salvati, N. (2003). Poverty Projile of Riverine Communities of Southern Lake Vo/ta., Sustainable Fisheries Livelíhoods Programme in West Africa, Cotonou, Bénin. 70 pp. URL: http://www.stlp.org/ftp/dload/frl8.pdf. Accessed 19 June, 2006. Ravnborg, H. (1999). Poverty and soil management- evidence of relationships from three Hounduran watersheds. In : Workshop: Assessing the Impact of Agricultura/ Research on Poverty. Costa Rica, September 1999. Seaman, J., Clarke, P., Boudreau, T. and Holt, J. (2000). The Household Economy Approach: A Resource Manual for Practitioners. Save the Children Fund, London. Sullivan, C.A., Meigh, J.R. , Giacomello, A.M., Fediw, T., Lawrence, P., Samad, M., Mlote, S., Callan, J.A., Schulze, R.E., Dlamini, D.J.M., Cosgrove, W., Delli Priscoli, J., Gleick, P., Smout, 1., Cobbing, J., Calow, R., Hunt, C., Hussain, A., Acreman, M.C., King, J., Malomo, S., Tate, E.L., O'Regan, D., Mi lner S. and Steyl, l. (2003). The Water Poverty Index: Development and application ata community scale. Natural Resources Forum 27:189-199. UNDP. (1997). Human Development Report 1997. Oxford university Press, Oxford. 134 pp. URL: http:/lhdr.undp.org/reports/global/1997/en/. Accessed 19 June, 2006. WFP. (2006). Vulnerability Analysis and Mapping. World Food Programme. URL: http://www.wfp.org/operations/vaml. Accessed 1 O June 2006. 211 Environmental and socio-economic evaJuation of prototype forest plantations in Cordoba department, Colombia Douglas White0 , Marco Rondon°, Maria del Pilar Hurtado0 , Mariela Rivera0 , James Garcia0 , Edgar Amezquita0 and Carlos Andres Rodríguez Plazas6 3Centro Internacional de Agricultura Tropical CIAT, Cali, Colombia bCorporación Autónoma Regional de los Valles del Sinú y San Jorge CVS, Montería, Colombia Introduction This study identifies the economic and environmental benefits of different land uses. The analysis concentrates on reforestation, cattle raising and a silvopastoral system. Córdoba department has high potential for forestry. According to the forest zoning of the Ministry of Agriculture, Córdoba has 946,400 ha suitable for forestry. Similarly the department has a total of 1,404,400 ha suitable for protected forestry or for restoration (Cruz and Franco 2006). While agricultura! and forest systems generate multiple environmental benefits, priority is given to the environmental benefit of carbon sequestration owing to two factors. First, it is possible to estímate the benefit (quantities of carbon sequestered in accordance with the growth of trees). Second, there is a growing market for carbon with clients prepared to pay (for example, the BioCarbon program ofthe World Bank). In contrast, the benefits in soil, water and biodiversity are difficult to identify given the relation between the supply and the benefit. Moreover, there is little market for these services within Cordoba. lnvestment portfolios are presented based on this information ofthe environmental and economic benefits. Methods As well as timber products, reforestation produces environmental benefits such as carbon sequestration, soil protection and improvement, conservation of biodiversity and regulation of stream flows. Carbon sequestered in biomass was calculated for seven forest species. A time horizon of 30 years was used for the financia! analysis of the systems. This strategy allows accounting for the growth of the forest species that have different harvest íntervals. Spreadsheets were created to represent growth behavior and financial aspects of the production systems such as inputs ( e.g. costs of labor, administration, capital and fixed costs). These spreadsheets were based on previous work of CVS (plantations) and Corpoica (cattle production and silvopastoral systems). In addition, the analysis takes account ofthe different sources of income such as sale of milk, meat, wood and other income such as Forest Incentive Certificates (Certificados de Incentivo Forestal, CIFs) and Certified Emission Reductions (CERs). The financia! analysis a llows adjustment of parameters to see the effect of different scenarios in the performance of the farm and/or plantation. The scenarios compare management strategies (for example, stocking rates of 0.8 or 1.5 head/ha), financia! context (for example, interest rates), national programs (CfFs) and intemational programs (CERs). 212 Forest plantations To facilitate comparisons between the seven forest species, the economic analysis reduces the number of variables that affect yield of biomass and wood. Biomass yield of trees depends on both environmental factors and management strategies. The majority of the species used in the economic study are long cycle, with the main exploitation occuring at the end of the time horizon, in this case 30 years. These species included: Roble (Pink Tecoma), Teca (Teak), Ceiba, Caoba (Mahogony) and Cedro (Spanish Cedar). While shorter periods could be used for the shorter-cycle species, the use of the same time horizon allows comparison of species performance (Figure 1 ). Planting distance ofthe chosen species, with the exception of Acacia, is 3m x 3m giving 1111 treeslha. The planting distance of Acacia is based on a plantation in Tierralta (2222 trees/ha). Acacia and Tambolero (a Jesser known native species) are species with more than one cycle during the analysis horizon; their rapid growth allows harvest each ten years giving three cycles during the analysis. The plantations are assumed to be thinned twice, removing half the trees before the fmal harvest. Sixty five percent of the tree is assumed to be lumber and the remaining 35% is branches and felling losses. ~r-7=======~--------------------------------------------, .. ..e 500 400 ~300 --+-Teca ---ceiba --Tarrbolero -Acacia o 5 10 15 20 25 30 Ano Figure l. Estimated yields oftrees (m3) over 30 years Carbon content is related to biomass. Carbon is estimated from wood density, yield and the length of the rotation (Table 1). Wood densities range from 0.39 t/m3 for Ceiba to 0.6 t/m3 for Teca. Furthermore, the volume of timber per ha takes account of the planting density and the level of thinning. 2 13 Table l. Wood density, yield and harvest cycle ofthe forest species. Yield Common name Scientific name Worddensity (m3/halyr) Native Caoba Swietenia macrophylla 0.43a 10- 18 Cedro Cedrela odorata 0.42" 11-25 Pochota quinata Ceiba roja 1 Pachira quinata 0.3g> 15 - 20 Ceiba tolúa Bombacopsis quina/a Bombachosis quinata Roble Tabebuia rosea 0.54b 8- 15 Tambolero, frijolito Schizolobium parahybum 0.4d> 13- 15 Introduced Acacia Acacia mangium 0.3 22 Teca Tectona grandis 0.4-0.6 15- 17 Adapted from: CONIF (2001) In the Consejo Regional de Competitividad (2002). Notes: Harvest frequency r 25-30 20-30 20-30 20-30 12-30 10 20 • Aróstegui, A. 1982 Recopilación y análisis de estudios tecnológicos de maderas peruanas. Documento de trabajo No. 2. Proyecto PNUD/FAO/PER/81/002. Fortalecimiento de los programas de desarrollo forestal en selva central, Lima. In: Baker et al. (2004). b Proyectos Andinos de Desarrollo Tecnológico (P ADT) en el área de los recursos forestales tropicales. 1981. In: Baker et al. (2004). Wood products of the plantations also vary according to the species and the management strategy. A conversion factor of 0.65 was used to convert biomass to sawn lumber in the estimation of wood production. The estimates of tons of C02 equivalentlha was calculated based on the estimated volumelha. This calculation takes account of the carbon content of wood of 0.46 giving a conversion factor of 1.6 and C to C02 of 3.67. Owing to the important quantities of carbon fixed in roots of the trees (Silver et al. 2004 ), its contribution was calculated at 0.2 above ground biomass. For Cedro, Ceiba and Caoba, the ratio of biomass and C02 is 1.3. For the dense wood of Roble, the ratio is 1.5 and for the fast-growing Acacia, 1.1. Estimates of C02 are not always easy to calculate. For sorne species, the density of word is not constant throughout the life of the tree. In Teca, for example, it varíes between 0.4 and 0.6 depending on tree age. Studies of physico-mechanical properties provide this information (V alero 2000). For this study, the ratio biomass to C02 varíes from l.36 to 1.86. As well as the age factor, wood density can vary according to geography, especially in Amazonia (Baker et al. 2004). The uncertainty of these assumptions draws attention to the need to refrne the relation between biomass and C02 as a priority for forestry research. 214 The perfonnance of storage of C02/ha (Figure 2) is similar to yield m3 of the tree species. However, there are small differences owing to the species characteristics. For example, assuming equal growth (15 m3/ha) of Roble and Tambolero, Roble has higher carbon storage because of its higher density. Moreover, it is notable that the density of Teca increases with age. Nor is the increase a straight line relationship, but an upward curve. Over the time horizon of 30 years, Ce iba has the highest average storage of carbon (353 tons C02 equivalent/ha). Cedro is about 328 followed by Teca 249, Caoba 233, Roble 226, Acacia 101 and Tambolero 69. 700 --+--Teca - c e1ba 600 500 _ _..._ Roble ~ ~Tambolero 400 --Acacia 1l -g 9 N 300 o u 200 100 o 5 10 15 20 25 30 Año Figure 2. Tonnes ofC02/ha in the forest species Cattle production The average annual inputs ofthe traditional system is $11 7,430 1 which includes the cost ofposts, barbed wire, animal health and livestock (Table 2). The labor cost for maintenance and management has an annual average cost of $280,420. Inputs to the intensive system are greater, $215,836, and differ inasmuch as the cost of livestock, use of supplements, fertilizer and seed for improved pastores. The average annual labor cost of the intensive system is $282,520. The average annual indirect costs for both systems are $38,652 and $59,169, made up oftransport and administration. 1 2006 Colombian pesos. 2 15 Table 2. Parameters for traditional and intensified cattle production. Concepto Stocking rate Milk production (litre/yr) Meat production (kg/yr) lnputs ($) Labor($) Indirect costs ($) Silvopastoral system Tradicional 0.8 600 130 117,430 280,420 38,6S2 Intensificado l. S 1000 300 21S,836 282,S20 S9,169 Corpoica's silvopastoral system uses dua1-purpose catt1e production with a stocking rate of 0.8 animals/ha (Cajas 2006, Convenio CVS-Corpoica-CIA T 2006, Corpoica 2006). This silvopastoral system stores an average of 119 tons co2 equivalent. Prices The analysis is based on actual data and estimates of the prices of the products, which are assumed to be at the farm gate. The value of meat is $2700/kilogram and mi !k $600/liter. The prices of forestry products are more diverse. As high-value species, Teca and Caoba command higher prices. In contrast, Acacia is a species oflower value (Table 3). Table 3. Prices offirewood and lumber (,000 Colombian pesos) Prices/values (units) Caoba Cedro Ce iba Roble Tambolero Teca Acacia Product of pruning or harvest ISO 1SO ISO ISO ISO ISO IOO (frrewood) Thinning I, third class lumber 3SO 2SO 200 2SO 2SO 300 S* (m3) Thinning 2, millable lumber (m3) 800 300 2SO 300 300 800 I70 Final harvest, millable lumber 800 360 360 360 360 1,0I8 180 m3 *Potes per unít. Sources: Intervíews; CVS; Fedemaderas (2006). The price of carbon from organic sources is about US$4 on the internationa1 market (Lasso 200S). From other sources such as the energy sector, the international price varies between $1S and $23. This low price reflects the uncertainty of sequestration in the long-term and the lack of precision in measurement of biological carbon. Moreover, it is unlikely that the plantation owner would receive this amount of money as multiple transaction costs, such as paperwork, local and international negotiations and monitoring of carbon storage diminish the cash return. For the analysis, it was assumed that the value of a ton of C02 equivalent has a value of $9,200 Colombían pesos (about US$4). 216 Costs The revised land use (forestry, cattle production and the silvopastoral system) produces different economic benefits. These benefits may be interpreted in different ways since they depend on land productivity, management strategy, fixed costs, operating costs and the products themselves. This study concentrates on wood, meat, milk and carbonas products. The assumptions ofthe analysis are detailed below together with the results. Like any investment, inputs are necessary to generate outputs. Forestry plantations need a variety of expenditures over the 30 years of the time horizon. The inputs for plantations include the cost of planting ($1 ,121 ,000/ha) for planting material and fertilizer. Fencing the land to protect the trees costs $731 ,000/ha, based on 163 lineal m. This includes the costs of posts, barbed wíre and so on. The labor costs ($15,677,000/ha) includes planting (land preparation, digging the holes and planting), maintenance (weeding and pruning), thinning and final harvest. The labor cost for protective fencing is $143,000/ha over 30 years. The high variation in land prices because of the availability of access to infrastructure (roads, ports and so on), proximity to the centers of consumption, topography, quality in terms of fertility and availability of water etc.) affects the financia! results and makes them difficult to compare. For these reasons, the cost of land was not taken into account in the analysis, assuming that whatever cost might be used affects the different plantations in the comparison equally and that, in general terms, the trends of the results will not change. Costs of technical assistance are estimated at $558/ha over 30 years and include the costs of professional foresters, technicians and auditing. fndirect costs ($31 , 136) are calculated as a percentage of the other costs. These costs include tools (5% of labor costs) and transport of inputs (15% of the inputs). The silvopastoral system has an average cost of establishment and maintenance of $614,070. Results: The investment portfolios The land uses (that is cattle production, forestry plantations and the silvopastoral system) have different costs and incomes during the thirty years. A discount rate is applied because the value of income (and costs) in the future ha ve less importance than those in the present. The fmancial results of the systems are sensitive to the assumptions and need interpretation. Three measures are used to analyze the economic benefits {Table 4): net present value (NPV), interna! rate of return (lRR) and cost-benefit ratio (CBR). Plantations of Teca, Caoba and Tambolero give better NPV s but their arder changes according to the discount rate. With arate of 15%, the silvopastoral system gives a better NPV than Teca. In contrast, Roble gives a lower IRR while Tambolero gives the best result by this measure because of its more frequent retums during the thirty years. The lRR is the rate of interest received on the investment over the time horizon of the analysis. For this reason, it is not a measure by itself of profitability. This result implies that the system is resistant to high interest rates. Profits can be less but still have a high fRR, such is shown by the cattle production system. 217 Table 4. Summary offinancial indicators ofplantations, catlle production and silvopastoral systems with CIFs and CERs (,000 Colombian pesos). Financia! Caoba Cedro Ce iba Roble Tambo- Teca Acacia Traditional Intensified Silvo- results le ro can le can le pastoral NPV 5% 71 ,083 43,763 43,227 22,268 52,341 70,62 1 20,655 3,619 12,051 23,371 NPV IO% 24,468 15,084 16,013 6,379 26,372 22,413 8,339 1,818 6,635 12,258 NPVI5% 9,644 5,952 6,786 1,539 15,074 7,695 3,234 977 4,075 7,091 IRR 31.6% 29.9% 31.3% 19.1% 75.4% 26.5% 22.9% 33.6% 64.4% 51.5% CBR 12.5 8.1 7.6 4.9 4.6 12.9 2.6 1.6 2.5 5.2 The margin ofpreference ofthe systems, in terms ofNPV, change according to the rate applied (Figure 5). With arate of 5%, for example, Caoba and Teca have higher NPVs than the other systems. This scenario implies that the investor has the facilities to wait for profits in the long-term. Nevertheless, the result is different when a rate of 15% is applied. The differences in the NPVs are reduced and the arder of preference ofthe options changes. With a discount rate of 15%, Tambolero shows a higher NPV because of the three harvests. Sale of wood that takes place in the future (for example Teca) is not so attractive as retums generated at various intervals during the time horizon of the anaJysis. This financia] situation is typical for many smaJJ producers who are not easily able to wait for retums. W,OOO .----------------------------------------------, o C> C> X 60,000 g 40,000 ~ ~ 20,000 DVPN 5% • VPN 10"/o • VPN 15% Figure 5. Net present vaJue (VNP) of plantations, traditional and intensive cattle production systems (ganadería) and the silvopastoral (silvopastoril) system without externa! backing (CER or CIF). In contrast, Table 5 shows annual retums of the forestry systems. Teca and Caoba have the best average and maximum (in the finaJ year) annual retum. Acacia, which is a plantation of 2200 seedlings/ha needs more investment in inputs and labor. Therefore Acacia shows lower retum in its first year 218 Table 5. Mean, maximum and mínimum annual retum without discount. Caoba Cedro Ce iba Roble Tambolero Teca Acacia Mean 8,184 5,058 4,699 2,786 4,126 8,463 1,867 Maximum 191 ,344 120, 114 103,482 70,025 31,6 18 198,533 22,345 Minimum -1 ,547 -1,460 -1,542 -1 ,63 1 -1,678 -1 ,792 -2,914 Financial support affects the financia! performance of forestry plantations. Table 6 and Figure 5 give summaries of the economic benefits if CIFs and CERs are not available: profitability of all plantations is reduced. The returns of Roble become negative under the assumption of a discount rate of 15%. Moreover, with 15%, intensive cattle production and tbe silvopastoral system are more profitable than Cedro, Ceiba, Roble and acacia. With a discount rate of 1 0%, intensive cattle production gives better retums than either Cedro, Roble and Acacia. The silvopastoral system is better than Teca with a discount rate of 15%. At almost all discount rates, traditional cattle raising provides the lowest cash retum. Table 6. Summary offmancial indicators ofplantations, catlle production and silvopastoral systems without CIFs and CERs (,000 Colombian pesos). Financia! Caoba Cedro Ceiba Roble Tambo- Teca Acacia Traditional lntensified Silvo- results Iero cattle cattle pastoral NPV 5% 67,281 39,017 38,462 18,536 47,431 67,572 17,360 3,619 12,051 22,316 NPV10% 21,650 1,683 12,426 3,612 23,002 20,464 5,924 1,818 6,635 11 ,515 NPV15% 7,311 3,199 3,863 -753 12,482 6,260 1,265 977 4,075 6,531 !RR 24.0% 20.1% 21.3% 13.6% 47.0% 23.0% 17.5% 33.6% 64.4% 47.1% CBR 12.2 7.7 7.3 4.6 4.4 12.6 2.4 1.6 2.5 4.7 The fmancial support of CIFs and CERs add between 2% and 98% more to the retums of forest products, witb the CERs providing more benefits to the investor for all plantations. CIFs provide more support only for acacia under the assumption of a discount rate of 15% because CIF's are a benefit that is obtained at the start of the investment, and are not discounted like the CERs, which are obtained at the end of the time horizon. CERs make up between 0 .5% (Teca, NPV with 5%) to 81% (Roble, NPV with 15%) oftotal returns of plantations (Figure 6). The first example, Teca, shows that the value of wood dominates the returns. In contrast, Roble wood does not have much value because of its slow growth and its low price in the local market. In this case, returns depend on the CERs, but they sbould consider intemational markets such as Nortb America. Tropical timbers of tbe genus Tabebuia have considerable hold in today's markets and can compensate for the effect of the local market. 219 100.0"/o ,----------------------;:::::===~ 50.0"/o Caoba Cedro Ce iba Roble Tarrilolero Teca DVPN 5% • VPN 10"/o • VPN 15% Acacia Figure 6. Contribution of Certified Emmission Reductions (CER) to the net present value of (VPN) of the forest species. CIF's provide benefits to the investor who depends on the species and the assumptions of the discount rates (Figure 7). Owing to its slow growth and low retums from the sale of wood, Roble receives the major part of its retum (65%) from CIFs. Moreover, because it is a native species, it has greater support (of the CIF). The retums oftambolero depend little on CIFs because it is a fast-growing species that produces high-value timber. 100.0"/o ~--------------------------------------------r=====~---, Caoba Cedro Ce iba Roble Tambolero DVPN 5% • VPN 10"/o • VPN 1 Teca Acacia Figure 7. Contribution of Forestry Incentive Certificates (CIF) to the net present value of (VPN) of the forest species. In summary, the selection of a species for plantations or other land-uses depends on the preferences of the investor. In addition to financial factors, other non-financial factors must be taken into account, such as the promotion of environmental andlor social benefits. Nevertheless, the financial analysis gives an information base to forecast retums. The analysis also shows the effects of govemment policies such as ClFs and the potenrial of environmental markets such as that for carbon (CERs) to intluence land-use. 220 In the silvopastoral production system, the saJe of products (milk and meat) provides 98% ofthe system's retums. The average net annual in come ís $2,268,878/halyear. When the val u e of the CERs is included, the system gives a mean profit of $2,321 ,965. The benefits of si\vopastoral systems tally with others such as that of CIPA V (Murgueito 2002). Conclusions A forestry context includes topics of public and prívate incentives. This study' s analysis shows the economic benefits of forest plantations. The majority of the species can provide profits without the support of the public sector (CIFs) or of the carbon market (CERs). Nevertheless the decisions to reforest bring more than econom ic benefits. The market for environmental services offers a powerful incentive for the conservation and restoration of tropical forests and new income opportunities for rural people. Nevertheless, it is still not clear which producers, consumers and types of forest resources will be the real beneficiaries of these markets. Nor is it clear what are the most effective conditions for the creation of markets for environmentaJ services to achieve the objectives offorest policy. The majority ofthe markets are still incipient and their development demands concerted action of governments. The implicatíons for the future of decisions that will be taken in the next years require a good review ofthe impacts ofthe efficacy, efficiency and equity ofthe markets. The Colombian forest sector is characterized by a confluence of production from natural forests and forest plantations, a diversity of species with great potential, but scarcely developed. The Colombian balance of trade of wood and manufactured timber products is increasing exports. Colombia is taking advantage of the immense opportunities that intemational commerce of forest products offers. Because of its closeness to the sea and maritirne ports, Cordoba can have good access to the demand for forest products concentrated in developed countries (Martínez y González 2005). Moreover, the warm humid environments of Cordoba provide the possibility to produce trees and timber rapidly and profitably. Plantations can al so be u sed in silvopastoral systems. A topic worthy of further research is the identification ofsynergies instead of competition between the production of trees and pastures (Andrade et al. 1999). Reforestation in given circumstances can be a good business, viable not only for companies that have considerable resources but for smaJlholders wishing to reforest their Iands. 22 1 References Andrade H., Ibrahim, I.M., Jiménez, F., Finegan, B. and Kass, D. (1999). Dinámica productiva de sistemas silvopastoriles con Acacia mangium y Eucalyptus deglupta en el trópico húmedo. Agroforestería en las Américas. http://www.fao.org/WAIRDOCS/LEAD/X6337S/X6337SOO.HTM Baker, T.R., Phillips, O.L., Malhi, Y., Almeida, S., Arroyo, L., Fiore, Erwin, A. T., Killen, T.J., Laurance, S.G., Laurance, W.F., Lewis, S.L., Lloyd, J., Monteagudo, A., Neill, D.A., Patino, S., Pitrnan, N.C.A., Natalito, J. , Silva, M. and Vásquez Martínez, R. (2004). Variation in wood density determines spatial patterns in Amazonian forest biomass. Global Change Biology 10:545- 562. Cruz Ospina, T. and Franco María, J. (2006). Plan de Desarrollo Forestal Córdoba (PDFC). Cadena Forestal Madera y Mueble del departamento de Córdoba, Corporación Nacional de Investigación y Fomento Forestal (CONIF), Corporación Autónoma Regional de los Valles del Sinú y del San Jorge (CVS), Secretaria de Desarrollo Económico y Agroindustrial - Gobernación de Córdoba, Cámara de Comercio de Montería. Primera edición, Enero. Montería, Córdoba. 136pp. Fedemaderas. 2006. Precios en Agosto. URL: http://www.fedemaderas.org!Archivos/ BOLETIN%20FINALPRECIOS%20DE%20LA%20MADERA%20AG OST0%202006.xls Lasso, M.A. (2005) ¿Campeona en el Mercado del Carbono? Reportaje. Directora Editorial, TierraAmérica. URL: http://www.tierramerica.net/2004/1120/articulo.shtml Martínez Covaleda, H.J. and González Duitama, E.D. (2005). Características y Estructura del Sector Forestal- Madera-Muebles en Colombia: una Mirada Global de su Estructura y Dinamica 1991-2005. Ministerio de Agricultura y Desarrollo Rural. Observatorio Agrocadenas Colombia. Documento de Trabajo No. 95. URL: http://www .agrocadenas. gov .co Murgueitio, E. (2002). Sistemas Agroforestales para la Producción Ganadera en Colombia. ln: Intensificación de la Ganadería en Centroamérica. FAO. URL: http://www.fao.org/WAIRDOCS/LEAD/x6366s/x6366sl3.htm Silver, W.L., Kueppers, L.M., Lugo, A.E., Ostertag, R. and Matzek, Y.V. (2004). Carbon sequestration and plant cornmunity dynamics following reforestation oftropical pasture. Ecologica/ App/ications 14:1115-1127. Valero, S.W. (2002). Relación entre anatomía y propiedades físico-mecánicas de la especie Tectona grandis proveniente de los llanos occidentales de Venezuela. Revista Forestal Venezolana 1:46. 222 Impacts and indicators of impact of fair trade, fair trade organic, specialty coffee Samuel Fujisaka, Thomas Oberthür, Raul Rosales, Hermann Usma, and German Escobar Centro Internacional de Agricultura Tropical C/A T Introduction The intent of the research reported in this paper is perhaps best explained by a recent email from Michael Dupee, Vice President, Corporate Social Responsibility, Green Mountain Coffee Roasters (GMCR): "For many years, Green Mountain Coffee Roasters has supported a variety of collaboratively developed environmental, economic, and social initiatives in communities where we purchase coffee. Our goal has been todo good in the world, and while we believe that this has been the case, we have been chatlenged to measure the effectiveness and impact of these prograrns. We have often asked ourselves questions Jike, "Have these programs had the in tended impact, and if so, how do we know?" "Can these programs be more effective, and if so how?" "What does success loo k like, and how do we measure it?" " In 2005, we decided to focus our efforts on four bottom line goals: reducing poverty, hunger, and waste, and increasing responsible energy usage in the communities we touch. Over the past few months, Green Mountain Coffee Roasters, the Sustainable Food Lab, CIAT (International Center for Agricultura! Research), ForesTrade and other stakeholders, have undertaken a study and fieldwork focused on developing Key Performance Indicators (KPI' s) related to poverty and hunger in coffee- growing communities. Our hope is that these indicators will ultirnately enable Green Mountain Coffee Roasters to be a better partner to these communities by being able to jointly develop goals for, and measure the impact of, the social programs and business practices we have developed with our partners throughout all of our supply chains. We also hope that these results may be of value to others in the specialty coffee industry, and beyond" (Dupee 2006, personal communication) There are severa\ approaches to indicators of poverty, ranging from single criteria to long lists of welfare variables to criteria elicited from the poor themselves. The World Bank has used single indicators (e.g., income of Jess than US$2.00 a day) based on census estimates to establish baselines and to measure change for large numbers of people. Although widely criticized, the approach has served as a litmus characterization at the district, national, and regional leve!. Other single specific measures include our use of childhood stunting as the poverty measure for our global work on both crop biofortification and development of drought tolerant cultivars (Hyman, Fujisaka, Jones 2006). Large researcher-determined lists of indicators have been used to measure impact. Data elicited through large sample surveys include income sources and amounts, self-employment, seasonal and occasional labor, land and livestock holdings, costs of farm/enterprise inputs including family labor, housing and housing construction materials, holdings and value of different assets from radios to domestic appliances and motorcycles, levels of education, literacy, health measures, debt and savings, and access to water, power, health care, education, roads, and markets. Such data can be subjected to econometric analysis with results that are not always convincing but they are almost always expensive to come by. A somewhat more recent approach has been to measure changes in terms of human, social, physical, financia!, and natural capital. One such approach is the Department for lnternational Development (DflD) 223 Sustainable Livelihoods Framework 1• The approach is attractive in that gains and losses are estimated for the different capitals and that the capitals are recognized as convertible from one form to another ( e.g., trees [natural capital] are cut down to provide funding [financia! capital] for schooling [human capital]), which in turn can lead to better jobs [more financia! capital] and to better housing [physical capital]). The difficulty ofthis and the long-list approach is in monetizing or standardizing values across indicators. More technical work on crop improvement or innovations in management can be analyzed in terms of changes in enterprise budgets and benefit-cost ratios. Such analysis is quantitative and focused, but deals neither with whole farm budgets nor necessarily directly with poverty. On the other hand, for farm farnilies largely reliant on a single crop ( e.g., coffee, rice), simple enterprise budget analysis may provide a good portrait of welfare improvement. In reaction against measures thought to be locally inappropriate, researchers have used different forms of indicators elicited from the poor themselves. One approach asks people about their classificatory systems regarding poverty and wealth. Once categories are defmed, definitions of each category are elicited; and these latter serve as locally-appropriate indicators of poverty. In village x, for example, housing material may be irrelevant to local defmitions of either poor or rich; and the number of educated sons may be all important. The recommendation would be to use the locally agreed-upon indicators. The approach used in this study is similar to the one described above. Key to the work was the eliciting of livelihood circumstances in good compared with bad years and of the use or allocation of good-year resources. As shown below, the approach generated a set of indicators relevant to coffee growers; albeit given the heterogeneity of groups, each is not necessarily applicable to all groups. Methods Coffee producers were interviewed in three areas of Guatemala (Huehuetenango, Coban, and Barbarena) and in two areas of Mexico (Jatelnango in Chiapas and Huatusco in Veracruz). Three to five communities were visited in each area, with fieldwork facilitated by representatives of respective local organizations selling coffee to GMCR. The research tearn included Fujisaka, Oberthür, Rosales and Usma in Guatemala; and Fujisaka, Rosales, and Escobar in Mexico. 1 The word ' livelihood' can be used in many different ways. The following definition captures the broad notion of livelihoods: 'A livelihood comprises tbe capabilities, assets (including both material and social resources) and activities required for a means of living. A livelihood is sustainable when it can cope with and recover from stresses and shocks and maintain or enhance its capabilities and assets both now and in the future, while not undermining the natural resource base.' (Adapted from Chambers and Conway 1992) DflD stresses the importance to livelihoods of capital assets and distinguishes five categories of such assets: natural, social, physical, human and financia!. It also stresses the need to maintain an 'outcome focus', thinking about how development activity impacts u pon people's livelihoods, not only about immediate project outputs. DflD is operationalizing livelihoods approaches in many different contexts. Broadly speaking one can aim to promote sustainable Jivelihoods through direct support to assets (providing people with better access to the assets) or contribute to the more effective functioning of the structures and processes (policies, markets, social relations, etc.) that intluences not only access to assets, but also determines which livelibood strategies are open to poor people. The idea is that if people ha ve better access to assets they will have more ability to influence structures and processes so that these become more responsive to their needs. Ata higher organizational leve!, DfiD has identified three types of activity that can contribute to poverty elimination: (a) enabling actions to support the policies and the context for poverty reduction, (b) inclusive actions that broadly improve opportunities and services generally, and (e) focused actions that are targeted directly at the needs ofpoor people. 224 Groups of small-holder coffee producers were interviewed using participatory methods in all communities except Barbarena, Guatemala (where three large-bolders were interviewed). Elicited and prioritized in the group interviews were: • Livelihood activities or resources used in good and bad years, with "good" and "bad" defmed in terms of coffee production and price. The elicited responses identified the different enterprises, activities, and income sources that producers relied upon and their relative importance in both good and bad years. As a set of potential impact indicators, projects or programs such as fair trade payments and organic certification and premiums would strive to reduce "bad" year outcomes and to increase years approximating what farmers described as "good" years. • Allocation ofresources gained in "good" years. Farmers' real or desired investments in good years pro vide insights into desired outcomes that, if and when met, can serve as indicators of impact. • Co.ffee production and co.ffee-related problems. Solution of prioritized problems related to coffee production, processing, and marketing would have clear, positive impacts on the lives of farmers. Work on increasing benefit-cost ratios via problem solution and improvement of returns to factors of production within the coffee enterprise could be achieved through technical programs (not encountered in the course ofthe research). • Rough estimation of costs-benefits. Farmer provided very rough estimates of yield, production costs, prices paid for conventional specialty vs. organic/fair trade specialty coffee. Detailed enterprise budgets would have been desirable but were not possible to elicit given the limited time in the field and the number of researchers. • Community problems. Groups identified and prioritized community leve! problems. Programs seeking to ensure that price premiums benefit local producers could easily invest in community rather than individual needs. In each of the above areas, except for the estimates of cost-benefits, respondents quantitatively priorítized the elicited items by distributing counters (1 00 beans or kernels of maize) relative to perceived irnportance. Interviews, conducted by the researcbers working in pairs, required up to 2-1 /2 hours; were preceded by an explanation of objectives and methods, possible outcomes, and a request to continue. Interviews ended with questions and concems of the respondents. Group participation was universally lively and enthusiastic. Where strong, people would conduct heated discussions in their indigenous languages before turning back to provide decisions in Spanish. 8oth males and females participated. Each and every one was encouraged to participate. The participatory work sessions were usually liberally sprinkled with hurnorous banter and jokes. Travel time was generally long and uncomfortable, albeit with incredible scenery. The team interviewed and visited the farms of three wealthy large-holder coffee producers in the area around Barbarena, Guatemala. Findings Livelihood activities and resources used in good and bad years As expected, coffee production was the most important livelíhood source, fo llowed by hiring out as farm laborers (Table 1 a). Other important sources of livelihoods were encountered in sorne, but not all communities: remittances (important in 10 of 14 communities), production of maize and beans (11 of 14 comrnunities), and use of forest products (10 of 14 communities). Other sources were sales of borne products (foods, crafts, other in 6 communities); chickens and pigs in five communities; government or aid assistance in three communities; cardamom production in the three communities in Coban, Guatemala; coffee seedlíng or flower nurseries in three ofthe five Huatusco communities; and cattle in one community. 225 Other crops such as sugar cane, fruit, nuts, and vegetables also contributed to livelihoods in severa! communities. The good year-bad year contrast for livelihood activities or sources was strong (Tables 1 a, 1 b ). Overall, in response to bad years, farmers relied less on coffee; had to hire out (work off-farm) more; re lied more on remittances; and were able to hire less labor to work on their own farms (as is shown in the next section). Working as migrant laborers meant neglecting coffee plantations, parcels of maize and beans, and cardamom production (in the one Coban community in which respondents greatly increased hiring out in bad years). Producers also tended to rely somewhat more on forest products and on small livestock (chicken and pigs) in bad years. 226 Table la. Livelihood activities in good (G) and bad (B) coffee years. Huehuetanago Coban Jatelnango Huatusco ~ a~ "' o "' ! ' a ~ 1 ~ ~ o Q) "() "' 1 'f o t:; "' Q o "' <-o "' "' Q) "' ·- ca .... 0.. r:n .:: en Q) Q) 1 Q):z Q) ¡:¡:¡ "' ) >..J:J .... ·¡: N "' en .~ (:. t: f- Q)"' Q) E o -~ ..e S Q) ... t:; o ::S 5 Q) ' o ::So ::S ·- "' :.2 8 ~ ::S Q) Q) ::su ::S o o !::1 ::S 0.. 5::Z:: :a~ ;z o. .... zo. z-o o o >< (1;1 o "O a ú3 e gp Q) u 0.. u u ::S u :.a Q) o ....J "O a:¡ u o.. r:n ~ e -Livelihood G B G B G B G B G B G B G B G B G B G B G B G B G B G B Coffee 55 20 50 20 37 20 36 9 35 13 50 27 36 8 53 9 34 16 46 6 45 9 37 8 24 14 27 11 Laborer 12 23 18 27 42 13 3 37 2 8 5 9 27 26 15 34 5 26 14 35 22 39 14 21 21 19 8 21 Maize-beans 22 18 12 4 11 17 13 8 5 19 3 10 6 5 o o 17 6 5 5 5 o o o o o 4 4 Remittances 3 21 8 15 10 50 o o o o o o 8 29 5 JI 7 15 o o 13 46 15 56 16 33 33 29 Forest prod o o o o o o 14 21 12 4 4 9 4 6 4 6 o o 2 3 4 o JO 15 10 15 2 5 Home prod o o o o o o 3 2 4 7 o o o o 6 13 8 6 6 15 o o o o o o 1 7 Chick/pigs o o o o o o 4 8 2 14 o o o o 10 12 8 6 o o o o 4 o o o o o Assistance o o o o o o o o o o o o o o o o 15 15 o o 6 6 o o o o 6 8 Cardamom o o o o o o 23 9 19 19 34 36 o o o 1 ol o o o o o o o o o o o o Nurseries o o o o o o o o o o o o o o o o o o 7 14 o o o o 21 10 3 5 Cattle o o o o o o o o o o o o 5 6 o o o o o o o o o o o o o o Other 8 18 \S 26 o o 4 6 21 16 4 9 14 20 7 15 6 10 20 22 5 o 20 o 8 9 16 10 Others en S en en en ~ti Q)tl Q) en Q) en en Q) ~ Q) a .... ~ la .... Q) en - en ::S~~ o :0 "' ::S a ::S ::s- ~ 5 a ;z c..D"' .... ..D ·- oz oz o;z~ "2 $:!:E S O·- fl C<3 S o ~ .<;:: ~ .<;:: ~ 2 Q) ~.<;::o Q) Q) .<;::a ~ti;! ¡:.¡..~u Q) a:¡ ~a:¡ 1:10 1:102 b02 bl) ¡:.¡.. 1:10 1:102b0 2a:¡G.>U Q) Q) ::S Q) > ¡:.¡.. > > > ~¡:.¡.. ~u.. r:n > ~u..> 227 Table lb. Means, good and bad coffee year livelihoods Good Bad Coffee 40 14 Laborer (hire out) 15 24 Remittances 8 22 Maize-beans 7 7 Forest products 5 6 Home products 2 4 Chickens/pigs 2 3 Assistance 2 2 Cardamom 5 5 Nurseries 2 2 Cattle o o Other 11 12 Impact indicators would have to include: • Contribution of coffee to family livelihoods2; • Need to work as laborers off-farrn; • Ability to remain on-farrn tocare for different whole farrn enterprises; and • Need to rely upon remittances. Uselallocation of "good year" resources Respondents ' majar priorities for the allocation of good-year resources were investrnent in the coffee plantation, improved diet and food intake, housing improvement or expansion, and health or medicine. Slightly more secondary priorities were clothing, school goods, increased hiring in, savings and/or debt payment, and less hiring out (Table 2). 2 It is important here to note that most of the business models employed in the source communities had most of the elements of a typical commodity model. The contribution of coffee to livelihoods may become more important if different " specialty coffee" business models such as direct relationship models are models that transparently transmit information and benefits between supply chain actors. A full supply chain analyses would be required in order to obtain the necessary insights into the workings of the currently used business models. 228 Table 2. Use/investment of"good year" resources Use Plantation Food Housing Health C1othing Education More hire in Save/pay debt Less hire out Land M u les Vehicle Other Huehue 9 o o 25 16 o 28 25 o 25 24 25 o o 2 o o o 14 o 7 o o o o o 5 29 7 4 17 o o o 20 7 o o 11 Coban 18 o 10 16 19 17 4 13 13 7 8 o 10 14 7 26 o 6 o o 3 o o o 8 1 o 12 25 9 8 23 o 6 o 15 o 2 Chiapas o 32 16 18 10 13 14 o 7 15 lL 8 25 o o 4 15 o o 3 o o o o 2 7 Huatusco 38 20 12 12 16 12 5 8 5 10 8 9 o o 11 o o 22 4 o o o o o 1 7 20 26 16 o 19 9 11 13 8 17 6 lO o 15 9 6 o o 3 o 8 o o 4 o o a "' 3 c. ::S u 42 18 6 16 4 o o 7 o 2 o 2 13 16 14 13 12 10 8 7 6 5 3 3 o 4 Variability by location of the responses was important in terms of indicator development. Communities varied considerably in what they did with resources in good years. Not one of the categories was elícited across all communities visited; and what was a highest priority in one community might not be mentíoned by another. For example: • Although plantation improvement was the highest priority in five communities, it was not even mentioned in four others. • Although health and education were highest príorities in two Huehuetenago, Guatemala, communities, these were low priorities in two of the Coban, Guatemala, communities. Health was not mentioned at all by respondents in Nueva Colombia, Chiapas, Mexico. • In the two Guatemalan communities in which health and educatíon were highest priorities, one also prioritízed food and clothing, items not roentioned by the other set of respondents. • The ability to hire out less in good years was a strong priority in sorne communities, but was not mentioned at all in others. • Sorne communities were land-scarce; others were land-rich. People with plantations on steep slopes with poor access needed mules, while others with easy access did not. • Only respondents in two of the Huatusco, Mexico, and communities saved for buying a pick-up truck, a priority not mentioned in any ofthe other areas. The key lesson is that, to be useful and appropriate, impact indicators will need to vary according to (very) local circumstances. Baselines and change, however, can be measured in terms of the set of priorities elicited: those shown in Table 2 also indicate that the category "other" had relatively few entries across the board. 229 Problems related to coffee Price received was the major problem in l l of the 15 communities (Table 3) and the greatest problem overall (albeit two communities did not mention prices). Pests and diseases, cost and timely availability of hired labor, and cost oftransport followed as problems. Other probJems were excess rain, cost offertilizers, lack of a local collection and storage fac ility, and lack of means for wet processing. Drought was a problem in three communities; hurricanes in the three cornmunities in Chiapas. Lack of capital, poor harvest logistics, and the need to rehabilitate older plantations were also seen as problems in severa! cases. Again, for each category elicited, sorne groups of respondents failed to mention it. The two communities that did not mention the selling price for coffee were among the three in Coban, Guatemala, who also relied on the (lucrative) production and saJe of cardamom. Table 3. Coffee problems H uehuetenango Coban o o e:: e:: ... X ·~ 'O ¡;¡ ·:; .€ Q) o o.. tiS § ::r:: :E tiS en (.) :E ~ Q) ... ~ en ... Q) Q) ~ ~ r:n Q) o. 05 ... ¡..... en < (/) (/) en ~ U.l Q) 1 Q) -e:: 1 E e:: Cll e:: en tiS .5 > o :.E tiS o ~ (/) Q) :E e: :a (.) ::l Q) ~ ::l u :.2 ;z: Problem a::l u Price 23 34 50 30 o 22 o Pests/ diseases 11 22 16 - 20 15 25 Hired labor 31 o o 10 o o 25 Transport cost o o 18 6 o o 25 Excess rain o 26 o o o o o Fertilizer cost 14 o 16 24 o o o Collection point o o o o o o 25 Wet processing o 18 o 30 4 o o Drought 15 o o o o o o Others 6 o o o 76 1 63 2 o 1 Lack of capital, harvest logistics. 2 Poor soils, certification difficulties, low temperatures. 3 Hurricanes 4 Hurricanes s Hurricanes 6 Need to rehabilitate old coffee plantations 7 Need to rehabilitate old coffee plantations, lack of capital 8 Lack of capital Chiapas Huatusco Cll ·¡:¡ Q) Cll < ::l - > o.. u 01) tiS Q) tiS > ::l Q) z .....;¡ ::l ;z: 17 19 24 52 38 24 32 47 27 5 o 10 8 o 6 11 13 11 6 16 13 7 o 4 12 13 9 14 14 14 3 25 15 o 4 9 20 19 - 8 o 5 25 8 7 o o o 9 o o o 6 5 8 o 7 7 o 15 o 8 5 4 o 14 o o o o o 5 o o o o o o 6 o 1 26 3 22 4 18 S 6 37 6 31 7 14 8 1 21 • Price received by farrners is a strong and obvious indicator. In the present study, however, rnany of those interviewed were producing conventional coffees; and others were in transition to organic certification. 230 • Technical assistance to address pest and disease problems is needed in several areas. Benefits to producers could be channeled through locally adapted integrated pest management programs. Losses to pests and diseases would have to be monitored starting as a baseline activity. 3 • Other indicators can include: Is coffee the only cash crop? If so, is coffee more important compared to the others? If not, or if coffee is more important than other cash crops, what are the marketing channels available? Are niche-marketing channels generating substantially more income than other channels? Community leve/ problems Groups complained about the lack of, or distance to, elementary schools and/or the costs involved in sending students to secondary schools in 13 of 15 communities (Table 4). "Schools" represented the most pressing problem among those named in 8 communities. "HeaJth" (i.e., lack of a health post, presence of a health post but lack of doctors, and/or presence of health post and doctors but lack of medicines) was aJso a problem in all but one community and was the highest priority in three. Six groups were concerned about the lack of potable water or the poor state of delivery systems. The lack of or poor conditions of access roads were problems for seven communities. Roads were good and travel times short in the areas visited in Huatusco; and none ofthe groups visited there mentioned roads as a problem. Long travel times and poorer roads in parts of Coban, Guatemala, and Chiapas, Mexico, were reflected in groups from these areas identifying roads as a problem. People in five communities wanted a community meeting center. Lack of latrines was a problem in four communities. Lack of or faulty electrification was a major problem in a11 three communities in Chiapas, a problem not mentioned elsewhere. Poor or inadequate nutrition was a problem in two communities in Guatemala. Petty crime was a problem in one community in Chiapas. Most common among the category "other'' was the lack of a recreational area, that is, a football field. 3 A functioning extension system was only found in the Sierra de las Minas region where Forestrade is providing outstanding business and agronomic support services to the producers. Unfortunately the small difference between organic-fairtrade coffees and the currently strong market price for conventional coffees place this business model under pressures. CoincidentaUy the Sierras de las Minas region provides ample opportunity to develop business models that are based on a number of symbolic product qualities su eh as environmental service provision, not just certification schemes. 23 1 Table 4. Community problems Huehuetenango Coban Chiapas o e: g c e: e; o o 5h e: ~ - .J:: V) Q) ~ c Q) "O' Q) :::1 E- u :.a z ~ ::l Problem ¡:o u z Schools 23 20 45 22 50 22 11 38 o Health 17 7 24 24 50 20 24 17 34 Water 54 15 o o o 6 o o 17 Roads 6 16 o lO o 18 24 16 o Com center o 19 17 o o o 27 o o Latrines o 15 o 19 o 19 o o o Electricity o o o o o o o 20 22 Nutrition o o o 25 o o 14 o o Crime o o o o o o o o 7 Other o 8 14 1 o o 15 2 o 9 20 3 1 Recreational areas (14) 2 Transport (1 0), access to improved varieties (3), wet processing (2) 3 Flooding from adjacent river (20) 4 Recreational areas (6), wet processing (7) 5 Recreational areas (17), ecotourism center (17) c Q) :::1 z 11 14 16 23 5 5 13 o o 13 4 Huatusco ~ ~ ..... Q) :.a ..... ·¡: o Q) ·¡: () () .... .... :; o ..... c- O. o 0.8 >. Q.CI > Ql Ql .8 E.- e~~ E .7 -;;~ i:i'o Cllll .6 ·= ~ E ~~~ .5 ... -o - - • Coastal artisanal fishing -- Rice wheat --Highland mixed --Rainfed mixed -- Dry rainfed - - . Pastoral - - . Sparse arid J!..! ~~ ni )( .4 Cll--~~~ o ni e .S! .3 o- .2 :en~ o Q. .1 o ... a. u.O o 20 40 60 80 100 X= failed seasons In 100 yr East Asia and Pacific Cl e ·¡:¡ 1.0 e -- ' - -- Lowland rice Cll ·~ .... 0.9 Cll>. Q.CI > G> Cll 0.8 E.- Cll .: 0.7 t;; ~ i:i'o m• 0.6 ·=: e~ 0.5 J!.!! '5:! 0.4 1 ., )( ' Cll - -., 0.3 .. o c.!! .. o- 0.2 :e "' •• o '• Q. 0.1 o ., ct . ' 0.0 .. . . . . ' ' ' . ' - - . Tree crop mixed - - . Root tuber -- Upland intensive mixed -- Highland extensive mixed - - . Pastoral - - . Sparse arid ' . ,. •• • o 20 40 60 80 100 X= falled seasons in 100 yr Figure 3B. The proportion of area within each farrning system experiencing at least a given number of failed seasons in a 100 year period. Systems represented by so lid fines are among the 14 systems of tbe world with more than 2.5 million stunted children. 248 Combination of poverty, crops and drought indicators Poverty and drought are a combined problem in Latin America and the Caribbean, Sub Saharan Africa, South Asia and East Asia (Table 1). These regions are discussed in greater detail below. Farming systems in Eastern Europe and Central Asia, and in the Middle East and North Africa generally have fewer poor and less cultivated areas susceptible to drought. While these regions do suffer from drought combined with poverty, they are relatively less important in the context both ofpopulation and cultivated area. Poverty and drought in the farming systems of the Latín America and Caribbean region are shown in Table 2. The maize-beans system in Mexico and Central America stands out, with 2.8 million stunted cbildren and global and regional drought rankings of 14 and 4, respectively. The second system in the ranking is coastal plantation mixed, a system tbat follows mucb of the coast of northern South America, Central America and Mexico. This system has high numbers of urban population related to the port cities on the coast. The third system in the list is the irrigated system, found in northern Mexico and along the Peruvian coast. Tbis system also has high urban population, including one ofthe region's largest cities, Lima, Peru. The fourth system in the list, dryland mixed, is often considered to be particularly drought prone, but it ranks in the middle third of the global ranking of farming systems according to drought. The remaining systems in Latin America have less than 600,000 stunted children and global drought rankings greater than 24. Overall Latín America and the Caribbean conform to the accepted view that the region is less poor than Africa and Asia and suffers less from drought. Table 2. Top 1 O systems in Latín America by stunted children and both regional and global failed seasons rankings. Farming system Maize-beans Plantation mixed lrrigated Dryland mixed High altitude mixed Forest based Intensive highland mixed lntensive mixed Extensive mixed Cereal-! ivestock Region CA Coastal LAC LAC Central Andes LAC N Andes LAC Cerrados-t tan os Campos Millions 2.8 1.7 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.2 Failed 15 9 38 22 42 31 37 11 19 16 Failed 4 1 10 7 11 8 9 2 6 5 Sub Saharan Africa and the Middle East and North Africa suffer more from poverty and drought: each of the poorest top 10 systems has more than 2 million stunted children (Table 3). Four ofthese systems are in the top 1 O globally in the crop drought rankings. The cereal root crop and maize mixed systems both ha ve 6.3 mil! ion stunted children. These two systems, spanning the southern portion of the Sahel and a large part of East Africa, have high rural populations. The root crop system has a high number of stunted children (5 milüon), even though drought intensity is relatively low. The other notable system in this region is agro- pastoral millet sorghum, a Sabe! system with more than three million stunted children and bjgb drougbt intensity. 249 Table 3. Top 1 O systems in Sub S ah aran A frica and the Middle East and North Africa, by stunted children and both regional and global Farming system Region Millions Failed Failed stunted global regional Cereal root crop SS A frica 6.3 3 Maizemixed SS Africa 6.3 8 3 Root crop SS Africa 5.0 lO 4 Forest based SS Africa 3.2 18 6 Pastoral SSAfrica 3.2 27 9 Agro-pastoral millet -sorghum SS Africa 3.1 6 2 Highland temperate núxed SS Africa 2.8 21 7 Highland perennial SS Africa 2.6 26 8 Sparse arid ME&NA 2.4 56 5 Tree crop SS Africa 2.3 13 5 Areas of high drought risk in Asia have even higher numbers of stunted children (Table 4). Five of tbese systems each bave more than 10 million stunted children. Five of the top six drought systems are also the top five systems with stunted children. The rainfed mixed system in South Asia stands out, with the second highest stunting value and the highest global drought ranking. The rice-wheat system in South Asia has the highest number of stunted children and the fourth highest drought ranking. The lowland rice system in East Asia has the second highest drought index, but less than half as many stunted children compared to the South Asian rice-wheat system. These three Asian systems are marked by large populations with large cultivated areas. The upland intensive mixed system ofEast Asia has somewhat lower poverty and drought figures compared to the top two Soutb Asian systems, but these are still among the top ran.kings of the 63 systems. Fifth in the list in Table 4, tbe rice system in South Asia has a high number of stunted children (fifth largest), with a drought ranking of 7 out of 63 systems. The highland mixed and dry rainfed systems have less than 10 million stunted children, with drought rankings of 24 and 14 respectively. The Asian systems rank very high for poverty and drought throughout their top 1 O. Table 4. Top 1 O systems in Asia by stunted children and both regional and global failed seasons rankings. Farming system Region Millions Failed Failed stunted global regional Rice-wheat South Asia 28.3 4 2 Rainfed mixed South Asia 24.5 1 Upland intensive mixed E Asia & Pacific 15.4 5 2 Lowland rice E Asia & Pacific 13.4 2 1 Rice South Asia 11.7 7 3 Highland mixed South Asia 5.2 24 5 Sparse (forest) E Asia & Pacific 4.4 36 6 Dry rainfed South Asia 3.6 14 4 Tree crop mixed E Asia & Pacific 3.1 25 4 Temperate mixed E Asia & Pacific 2.6 23 3 250 Poverty and drought are more severe in the farming systems of Asia and Africa, with notably less severity in Latín America. Table 5 shows the top 15 farming systems ranked by the absolute number of stunted children. These systems each have more than 2.5 million stunted children. We chose 2.5 million as a threshold for inclusion in this table because these systems rely more heavily on staple crops. The values just below 2.5 mi Ilion in Table 1 are mostly livestock-based systems. Below these, the number of stunted children begins to decrease substantially. Table 5. Fifteen farming systems with over 2.5 miiJion stunted children, with global (fsg) and regional (fsr) farming systems rankings according to potential drought irnpact index. Crops appearing first time in the list are highlighted and in italics. System Stunting Crops fsg fsr SA rice wheat 28.3 a millet. wheat. maize 4 2 SA rainfed rnixed 24.5 EA upland intensive mixed 15.4 5 2 EA lowland rice 13.4 rice, maize, wheat, sweet potato, groundnut 2 1 SA rice 11.7 rice, pulses ( chickpea) 7 3 SSA cereal-root 6.3 sorghurn, rnillet, maize, 3 groundnut, cass~ SSA maize rnixed 6.3 maize, cassava, sorghurn, pulses, groundnut, 8 3 millet, bean, sweet potato SA highland mixed 5.2 rice, maize, wheat, potato , groundnut, 24 5 pulses (chickpea) SSA root 5.0 maize, cassava, rice, sweet potato, cowpea, lO 4 sorghum, groundnut, bean SA dry rainfed 3.6 Sorghum, millet, chickpea, groundnut, bean 14 4 SSA agro-pastoral millet sorghum 3.1 millet, sorghum, pulses groundnut, maize 6 2 LA maize beans 2.8 maize, bean, sorghum 15 4 SSA high temperate mixed 2.8 maize, wheat, sorghum, barl millet, 21 7 pulses EA temperate rnixed 2.6 maize, wheat, potato, groundnut, rni llet 23 3 EA hi~hland extensive rnixed 2.5 rice, maize, wbeat, Eotato, ~roundnut, Eulses 28 5 The main crops of the farming systems with high levels of poverty and drought are also shown in Table 5. Each of these crops covers at least 5 percent of the total cultivated area in each respective farming system (Table 6). This list suggests that poor farmers in drought-prone areas rely largely on 12 crops: • Rice • Mi\let • Cassava • Wheat • Sorghum • Sweet patato • Chickpea • Groundnut • Bean • Maize • Cowpea • Barley The drought ranking of the 15 systems shown in Table 5 are all within or near the top third of the 63 farming systems globally. Nine of these 15 systems are in the top ten in terms of their drought ranking. Only the East Asia temperate mixed system (drought rank=23) and the East Asia highland extensive mixed system (drought rank=28) did not fall into the top third ofthe 63 systems. 251 Table 6. The proportional area of each crop in the agricultural systems with more than 2.5 mili ion stunted children. Areas shaded in gray have more than 5% of the area in the ir respective system. BANP is the combined category of bananas and plantain. OPUL is the combined category of cowpea, chickpea, lentils and other pulses Farming Syslems Re ion BANP% BARL% BEAN% CASS% GROU% MAIZ% MILL% OPUL% POTA% RICE% SORG% SWPY% WHEAo/. Maize-beans LAC 0.017 0.008 0.161 0.002 0.007 0.668 0.009 0.005 0.009 0.098 1 0.000 0.016 (Mesoamerica) ----' Cercal-rool crop mixed SSA 0.007 0.001 0.032 0.052 0.093 0.125 0.224 0.126 J 0.004 0.045 j 0255 ~3~ 0.002 Maize mixed SSA 0.000 0.004 f:' 0.091 0.063 0.461 0.059 0.073 0.023 0.024 0.075 0.056 1 0.016 Root crop SSA 0.004 0.000 060 0.222 0.074 0.275 r 0.030 0.027 0.002 1 0.117 0.074 0.117 0.000 Agro-pastoral SSA 0.003 0.002 0.010 0.018 1 0.125 0.065 0.377 0.146 1 0.001 0.009 1 0.238 1 0.006 0.001 milletlsorghum Highland temperate SSA 0.000 0.166 0.040 0.009 o;J 0.251 0.062 0.051 1 0.008 0.003 1 0.168 1 0.016 mixed Rice-wheat SA 0.002 0.009 0.053 0.000 0.0 11 0.101 0.109 0.118 0.023 1 0.428 1 0.038 0.001 0. 106 Rainfed mixed SA 0.002 0.002 1 0.109 0.001 r 0.071 0.066 0.170 0.150 0.001 0.205 0.166 1 0.00 1 1 0.056 Rice SA 0.015 0.001 0.0 11 0.0 11 0.020 0.009 0.038 0.100 0.02 1 0.722 0.023 0.003 0.026 r-- High1and mixed SA 0.009 0.023 0.026 0.007 0.007 0.196 0.081 0.117 0.021 0.281 0.013 0.002 Dryrainfed SA 0.001 0.000 go~ 0.064 0.021 0.321 0.132 O.OOJ 0.020 0.358 0.000 0.023 Upland intcnsive mixed EAP 0.012 0.009 0.035 0.045 0.267 0.017 0.02 1 0.062 0.249 0.005 0.080 0.137 1 0.060 1 Lowland rice EAP 0.012 0.012 0.043 0.04 1 0.166 0.004 0.009 0.026 0.379 0.004 1 0.097 0.148 1 0.080 ·---Tempcrate mixed EAP 0.000 0.010 0.037 0.005 0.070 0.532 0.051 0.026 0.002 0.041 0.049 0.093 Highland extensive EAP 0.004 0.008 0.041 0.012 1 0.056 0.144 0.022 1 0.055 0.094 0.427 1 0.002 0.015 mixed 252 Conclusion This assessment of poverty, crops and drought suggests that 15 farming systems should be given high priority for agricultura! research and development {Table 5 and Figure 4). These systems account for substantial populations of the poor, including over 70 percent of stunted children in the world. The 14 systems have large areas of cultivated Jands susceptible to drought. Land use and the agricultura! economy in these 14 systems rely largely on just 12 crops. ·~· Figure 4. Priority systems, with over 70 percent of a ll stunted children and substantial drought. With few exceptions, the poorest, most drought susceptible systems have diverse environments and farmers have developed effective mechanisms to cope with risk. These farmers cope through diversity of livelihoods, including livestock. Judicious employment of irnproved crops may weLI be successful if the varieties can fit into such diverse and risky systems. The databases used and developed in this study have great potential for research and priority-setting for developing country agriculture. Our assessment used criteria that would help us focus on a reasonable number of regions and crops that could be g iven priority for investment in agricultura! research and development. These criteria cou ld be easily modified to reflect other priorities than those developed in the study to date. The results of this study are limited by the aggregate scale of the analysis. Future work could in elude other poverty indicators and poverty analysis at finer geographic resolutions within the farming systems. In addition, further work shou ld develop crop-specific drought models in a rder to provide more detailed information to crop improvement programs. This analysis excluded assessment of factors within farming systems such as variety adoption and the potential to use agricultura) technology. Nor did we make economic assessments to estímate the potential impact of focusing on the priority crops and systems identified in the study. While further research could complement the resu lts obtained here, this study 253 provides an initial assessment of the previously by-passed poor that face high drought risk, and the principal crops on which they depend. Acknowledgments We thank Elizabeth Barona and German Lema for their assistance in developing the spatial database used in this analysis. 254 References Balk, D., Storeygard, A., Levy, M., Gaskell, J., Sharma, M. and Flor, R. (2005). Child hunger in the developing world: An analysis of environmental and social correlates. Food Policy. 30:568-583. CIESIN (Center for lnternational Earth Science lnformation Network), Columbia University and Centro Internacional de Agricultura Tropical (CIAT). (2004). Gridded Population of the World (GPW), Version 3. Columbia University, Palisades, NY .. URL: http://beta.sedac.ciesin.colurnbia.edu/gpw. Accessed on 1 November 2006. ClESIN (Center for lntemational Eartb Science lnformation Network), Columbia Uníversity. (2005). Global subnational infant mortality rates [dataset]. Columbia University, Palisades, NY. URL: http://www.ciesin.columbia.edu/povmap/ds g lobal.htrnl. Accessed on 1 November 2006. CIESIN (Center for lntemational Earth Science lnformation Network), Columbia University. (2006). Where the Poor Are: An Atlas ofPoverty. Columbia University, Palisades, NY:. URL: http://www.ciesin.colurnbia.edu/povmap/. Accessed on 1 November 2006. CGJAR (Consultative Group on Internacional Agricultura] Research). (2000). Charting CGJAR's future: a new vision for 201 O. URL: http://www.Rimisp.lc/cg20 tob/doc4.htrnl# ftnl. Accessed on 15 August 2005. Dixon, J. and Gulliver, A. with Gibbon, D. (2001). Farming Systems and Poverty: Jmproving Farmers ' Live/ihoods in a Changing World. FAO and World Bank, Rome and Washington DC. Evenson R. and Gollin, D. (2003). Assessing the impact ofthe Oreen Revolution, 1960-2000. Science 300:758-762. Food and Agriculture Organization. (2003). Chronic Undernutrilion Among Children: An Jndicator of Poverty. Poster and Unpublisbed data set. Rome: F AO. Food and Agriculture Organization. (2005). The State of Food /nsecurity in the World. Rome: Food and Agriculture Organization. Food and Agriculture Organization. (2006). Agro-MAPS: A g lobal spatial database of agricultura! land-use statistics aggregated by sub-national administrative districts. URL http://www.fao.org/landandwater/agll/agromaps/interactive/index.jsp. FAO, Rome. Freebairn, D. (1995). Did tbe Oreen Revolution concentrate incomes? A quantítative study ofresearch reports. World Development. 23(2):265-279. Gryseels, G., De Wít, C.T., McCalla, A., Monyo, J. , Kassam, A., Crasswell, E., Collinson, M. (1992). Setting agricultura! research priorities for the COlAR. Agricultura/ Systems 40:59-103. Jones, P.G. (1987). Current availability and deficíencies data relevant to agro-ecological studies in the geographical area covered in IARCS. In: Agricultura/ environments: Characterisation, c/assification and mapping. Bunting, A.H. (ed.) CAB lntemational, WaUingford, Great Britain. Pp.69-82. Jones, P.G.; and Thomton, P .K. (2000). MarkSim: Software to generate daily weather data for Latin America and Africa. Agronomy Journa/ 93:445-453. Jones, P.G., Thomton, P.K. , Diaz, W. and Wilkens, P.W. (2002). MarkSim, Version 1. A computer too! that generates simulated weather data for crop modelling and risk assessment. Centro Internacional de Agricultura tropical (CIA T) CD-ROM series. Cati, Colombia: ClA T. CD-ROM + Guide. 87pp. Keig, G., and McAlpine, J. R. (1974). A computer system for analysis of soil moisture regimes from simple climatic data. Tech. Memo 74/4. Division of Land Research, Comrnonwealth Scientific and Industrial Research Organisation, Canberra, Australia. Linacre, E. (1977). A simple formula for estimating evaporation rates in various clirnates, using temperature data alone. Agricultura/ Meteorology 18:409-424. Shah, M. and Strong, M. ( 1999). Food in the 21 51 Century: from Science to Sustainable Agriculture. World Bank, Washington DC. You, L. and Wood, S. (2006). An entropy approach to spatial disaggregation of agricultura! production. Agricultura/ Systems 90:329-347. World Bank. (2004). Drarnatic decline in g lobal poverty, but progress uneven. URL http:/ /web. worldbank.org!WBSITE/EXTERNAUTOPICS/EXTPOVER TY /- O,contentMDK:20 195240--pagePK: 148956-piPK:21 6618- theSitePK:336992,00.htrnl. Accessed on 7 September 2006. WHO (World Health Organization). (2000). Global Database on Child Growth and Malnutrition: Forecast ofTrends. Document WHO/NHD/00.3. Geneva: WHO. 255 How cost-effective is biofortification in combating micronutrient malnutrition? An ex-ante assessment J. V. Meenakshi, Nancy Johnson, Víctor M. Manyong, Hugo De Groote, Josyline Javelosa, David Yanggen, Firdousi Naher, Carolina Gonzalez, James Garcia and Erika Meng Micronutrient malnutrition and tbe potential of biofortification The magnitude of micronutrient malnutrition is increasingly taking centre stage in policy discussions relating to food and nutrition security. lt is recognized that food security needs to refer not merely to adequate energy intakes, but also to ensure sufficient intakes of essential micronutrients. Estimates of the number of people affected by micronutrient malnutrition are high, with up to 5 billion people suffering from iron deficiency and about a quarter of all pre-school children (about 140 million) from vitamin A deficiency (United Nations 2005; pp. 14 and 19). Estimates ofthose at risk of inadequate zinc intakes are put at between a quarter anda third ofthe developing country population (Hotz and Brown 2004). Public health interventions to address this problem include fortification (of flour with iron, for example) and supplementation (twice-yearly vitaminA capsules for pre-school children). Yet few governments bave the resources to fund such programs on a permanent basis. Biofortification, which uses plant breeding techniques to enhance the micronutrient content ofstaple foods, is a new, complementary, approach. The premise of biofortification is that diets of undernourished people are based primarily on a staple, as poor people lack the purchasing power for a more diverse diet containing sufficient quantities of micronutrient-rich foods. The objective of biofortification is to enhance the micronutrient content of these staple food crops, through plant breeding techniques, thus resulting in higher micronutrient intakes by vulnerable people. Unlike commercial fortification, which necessitates the purchase of the fortified food, biofortification particularly targets rural areas where produced food stays within the community and is consumed either on-farrn, or locally. Further, repeat purchases or treatments are not necessary; for most crops, a one-time investment of disseminating varieties with the nutrient-dense trait becomes self- sustaining. Research has shown that it is feasible to breed at least sorne staple food crops to have higher micronutrient levels (Bouis 2000). The proof of concept that biofortified crops can have an impact on public health is starting to emerge from efficacy studies wbere trials are conducted with human subjects under a controlled setting. For example, there is evidence from a 9-month feeding trial in the Pbilippines, that regular consumption of rice containing an additional 2.6 ppm of iron was efficacious in improving body iron stores among women with iron-poor diets (Haas et al. 2005). Similarly, a feeding tri al of school children in South A frica indícates that consumption of orange-fleshed sweetpotato, high in beta-carotene, led to improvements in their vitamin A status (van Jaarsveld et al. 2005). Given this, the question becomes whether it is also economically efficient, and it is this question that this paper attempts to answer. Biofortification is a long-terrn strategy requiring a significant up-front investment in agricultura! research and development. Its success will also depend on the current diets of the target populations, how much of the staples they eat, in what forrns and witb what other foods. Thus, its economics are quite different from those of interventions such as fortification of flour or sugar, or the distribution of vitamin capsules. Recognizing this, in this study we estimate the cost effectiveness of biofortification for a sample of crops and countries throughout the developing world. Because biofortified crops are still being developed, the analysis is ex ante in nature. In determining cost-effectiveness, we use the disability-adjusted life years (DALYs) framework, which captures both morbidity and mortality outcomes in a single measure (see below). Relatively underutilized 256 in the economics literature as a metric for welfare, the use of DAL Ys obviates the need for the monetization of health benefits, a contentious issue that has been the subject of long debate with little satisfactory resolution. lnstead, benefits can be quantified directly using DAL Ys saved, and costs per DAL Y saved offer a consistent way of ranking a range of altemative health interventions- be they water and sanitation projects, or biofortification, the case considered here. In particular, this paper presents a synthesis of the evidence from severa! countries and crops that are targeted under HarvestPius, a program that is engaged in biofortification research. The target nutrients are: provitamin A 1 in cassava, maize and sweetpotato; and iron and zinc in beans, rice, and wheat. The choice oftarget countries (eleven in al!) is based on a number offactors, including the magnitude ofmicronutrient deficiencies, the importance of the target crop in the diet, and the availability of reliable data. To understand how sensitive results may be to the specifics of cropping patterns and diets, we include two East African, one Central African and one West African country. Similarly three South Asia and one Southeast Asian country are featured in the analysis, as are three Latín American countries. 2 This is thus the first paper to provide a comprehensive overview of the evidence to support biofortification, one that spans across crops, countries and micronutrients. The results provide evidence on whether biofortification can be a useful approach to combating micronutrient malnutrition, as well as identify the conditions under which is it most likely to be successful. The basic reports on which this synthesis has been generated are listed in the references. Given that the analysis is ex ante in nature for many crops, we consider both pessimistic and optimistic scenarios, which also enables a check on the robustness ofthe results to changes in assumptions. Quantifying micronutrient malnutrition The disability-adjusted life years (DALYs) framework The first step in assessing the cost-effectiveness of any intervention, including biofortification, is to determine the magnitude of the problem that the intervention is trying to address. One strand of literature has focused on the productivity losses that occur as a consequence of malnutrition (see for example, see Horton 1999 and Horton and Ross 2003). Other studies have examined the impact of malnutrition individually on mortality outcomes, cognitive development or child growth (for example, Gillespie 1998; a good review ofthe issues is contained in Alderman et al. 2004). An increasingly popular measure for quantifying the magnitude of ill health is the disability-adjusted life year or DALY, first detailed in the seminal contribution by Murray and Lopez (1996). lt is also important to mention the contribution of Zimmerman and Qaim (2004) who first used this framework in the context of biofortification. DAL Y s lost enable the addition of morbidity and mortality outcomes and are an annual measure of the burden of disease. Also, they provide a way to "add up" the burden of temporary illness (such as diarrhea) with more permanent ones (such as blindness) to give a single index. Thus, DALYs lost are the sum of years of life lost (YLL) and the years lived with disability (YLD). YLL represents the number of years lost because of the preventable death of an individual; while the YLD represents the number of years spent in ill-health beca use of a preventable disease or condition: DAL Ys Lost = YLL + YLD 1 There is a distinction between provitamin A and vitamin A: plants contain provitamins A, such as beta carotene, which are precursors to the vitaminA that is formed in the Ji ver. 2 1n the case ofBrazil, the estimates refer not to the entire country, but only to one region---the northeast-where poverty and undemutrition levels are high. 257 A public health intervention is expected to reduce the number of DALYs lost, and this difference is a measure of its benefit. Thus YLL saved represents the years of life saved because a death has been prevented and YLD saved or averted refers to the years of life spent in perfect health, because a non-fatal outcome or disability has been cured or prevented. The DALYs saved are then a direct metric for analyzing benefits of an intervention, and do not necessarily have to be monetized to ensure comparability across interventions. Unlike most agricultura! technologies, biofortification does not lead to a shift of the supply function; hence changes in economic surplus are not relevant. lnstead, it is the supply of dietary sources of iron that is increased, and ít ís the ímpact ofthis shift on public health that is captured here. DAL Y s saved also ha ve the appeal of being consistent with "specific egalitarianism" whereby everyone-irrespective of income--is presumed entitled to be free of ill-health. F or this reason, cost-effectiveness meas u res expressed in terms of DAL Y s saved are increasingly being used by agencies such as the World Bank and the WHO in priority ranking exercises (World Bank 1993). The use of disability weights, ranging from zero to unity, enables the incorporation of the severity of the disability, with higher weights implying greater disability (and 1 representing death). Further, since sorne outcomes affect only certain target groups (young children, or pregnant women for example), disaggregatíon by gender and age-specific target groups is needed. Finally, since many of the adverse outcomes are permanent and may last the rest of an affected individual's life span, a conversion to an annualized meas u re is necessary. Thus, more formally, the DAL Y burden may be written as: where ~ is the total number of people in target group j, M¡ is the mortaJity rate associated with the given disease, L; is the average remaining life expectancy, l ;¡ is the incidence rate oftemporary disease i that is of interest, D ;¡ is the corresponding disability weight, dy is the duration ofthe disease. (For permanent disabilities d.¡ equals the remaining life expectancy L1 .), and r represents the discount rate that captures time preferences. That is, the use of the discount rate implies that health gains today count more than health gains in the future. In adapting this framework to the present exercise, we made a few modifications to the original model. The first is that we exclude the age-weighting term that assigns a higher weight to the disabilities of the young than for those who are older. This is because a form of age-weighting is implicit above, as permanent outcomes that affect young children add up to more DALYs lost than those that affect adults. Also, unlike the original exercise, where the estimated life expectancy is interpreted as the maximum possible in a biological sense, here, we use a country-specific figure. We justify this on the grounds that the amelioration of a given micronutrient deficiency alone is not expected to change the average life expectancy in a country. Of greater irnportance, perhaps, is that the adaptation of this approach to the specific context of micronutrient malnutrition necessitated sorne additional modifications in terms of the level of disaggregation incorporated in determining the functional consequences of vitamin A, iron and zinc deficiencies. Nutritionists gave expert opinionon the detail of specific outcomes that may be attributed to 258 each of these deficiencies. In doing so, the approach was conservative, for example, adverse functional outcomes are proven only for clinicaJ manifestations3 of vitamin A deficiency (V AD), and these are all that are incorporated. We considered moderate and severe, but not mild, anemia in calculating the burden of iron deficiency, moreover anemia has multiple causes, of which iron-deficiency is only one. Similarly, we included only those outcomes for which there is evidence from meta-analyses. Where only an association is noted, (as, for example in the case ofvitamin A where studies suggest that its deficiency is associated with diarrhea, acute respiratory infection, stunting and maternal mortality) these are excluded from the analysis. The estimated used here may be construed as a lower bound in attributing disease and adverse functional outcomes to micronutríent deficiencies. Stein et al. (2005) give a more detailed description ofthis topic. Burden of vitaminA deficiency Yitamin A deficiency leads to vision impairment, including night blindness, corneal scarring, and in the most severe cases, blindness. In addítion, it also causes increased mortality for children under the age of six, and in the incidence and poor recovery from measles. An estimated 3% of the mortality rate of young children is attributed to vitamin A deficiency. One-fifth of the incidence of comeal scarring and meas les is due to V AD; and all of night blindness (both among children and pregnant and Jactating women) is due to insufficient intakes of vitam in A. While S te in et al. (2005) provides additional details, these form the basis of the calculation of DAL Y s lost due to V AD (Table 1 ). Table l. Burden ofvitamin A deficiency, by country. Country Ethiopia Kenya Uganda D.R Congo Nigeria Northeast Brazil Total DALYs lost (millions) 0.39 0.12 0.16 0.39 0.80 0.05 YLL as percent of DAL Y s lost 73 71 73 98 98 90 DAL Y s as percent of population 0.5 0.4 0.6 0.8 0.6 0.1 Yirtually all of the DALYs lost above- lost either due to mortality or morbidity-are accounted for by children under six, underscoring the disproportionate impact of the burden of vitamin A deficiency on young children. The DALYs lost from VAD are high in Africa, corresponding to between 0.4 and 0.8 percent of the population in these countries. Thus, annually, 121 thousand DAL Y s are lost to V AD in Kenya ea eh year, while in Nigeria, the figure is nearly 800 thousand DAL Ys lost. Put another way, between 0.5 and 1 percent of the national product is lost due to V A D each year in these countries. 4 In contrast, the magnitude of V AD is not as high in Latín America as it is in most regions of Africa. However, there are pockets with high levels of deprivation as is found in the northeastem regions of Brazil. In this part ofBrazil, VAD Jeads to the loss of the equivalent of 0.1 percent of the nationa l income each year. Note once again that these are conservative estimates of the loss due to V AD, largely because these take into account only clinical 3 Clinical manifestations include cornea! scarring, and problems with vision. Subclinical vitaminA deficiency is far more prevalent and insidious as it is nota disease in itself and is in sorne sense asymptomatic, but renders an individual more susceptible to infections. 4 That is, had this proportion ofthe population been healthy, they would have been able to contribute to the national income, and the average gross national product provides an approximation ofthis loss to the economy. 259 manifestations of V AD, and only for those outcomes for which definitive causality has been shown in the literature. The bulk-over 70o/o-Of all DALYs lost are dueto years of life lost dueto premature mortality . This explains why, for instan ce, the burden of V AD is higher in U ganda than in Kenya, countries with approximately similar population sizes. The nurnber of deaths of children under 6 (per 1000 live births) in Uganda (152) is higher than in Kenya (114), while life expectancies are approximately the same. Burden of iron deficiency lron deficiency leads to impaired physical activity for all age groups and impaired mental development for children under the age of six. In addition, it is estimated that 5% of all maternal mortality is caused by iron deficiency. A mother's death, in tum, implies still-bom children, and other child deaths dueto the absence ofbreast-feeding and care that she would have provided had she lived (for complete references see Stein et al. 2005). To estímate the DAL Y burden, we use published data on the prevalence of anemia. However, sin ce not all anemia is due to íron deficiency, we assume that approximately 50% of it is due to insufficient dietary intak.es of iron (this percentage can vary by country). This percentage was determined based on expert opinion from nutritionists. These form the basis for the figures detailed in Table 2. Table 2. Burden ofiron deficiency, by country. Percent share of Country Total DAL Ys lost YLDs of children DAL Y s as percent (millions) under 5 to total of population DALYs Bangladesh 0.49 66 0.4 India 4.00 66 0.4 Pakistan 0.52 53 0.3 Philippines 0.07 35 0.1 Northeast Brazil 0.20 66 0.4 Honduras 0.02 41 0.3 Nicaragua 0.03 53 0.5 In quantitative terms, the burden is, as expected, the highest in the populous countries of India, Bangladesh and Brazil. Normalized for population size, the burden of iron deficiency ranges between 0.1 percent ofthe total population in the Philippines to 0.5 percent in Nicaragua. Much ofthis burden arises from disability- especially among children aged five and under-with a share between 35 to 66 percent. These figures also illustrate the advantage of using the DAL Y methodology over methods that, for example, are based on the number of deaths caused. The use of the DAL Y method, which can "add" mortality and disability outcomes, indicates for example that in northeast Brazil, the burden of iron deficiency is higher than that of vitamin A deficiency- whereas the use of the 'number of deaths caused' criterion would indica te V AD to be a far greater problem than iron deficiency. Burden of zinc deficiency There is evidence from meta-analyses to show that zinc deficiency is implicated in adverse functional outcomes associated with diarrhea, pneumonia and stunting among children. Further, sorne diarrhea and pneumonia cases can also be fatal. Thus nearly one-fifth of diarrhea incidence, nearly two-fifths of 260 pneumonia and 4% of mortaJity of children under 5 can be attributed to zinc deficiency. The evidence in Table 3 suggests that 0.1% of the population in the Philippines, and 0.3 to 0.4 % of the population in South Asia suffer the consequences of zinc deficiency on an annual basis. The bulk of the burden is contributed by infants under the age of one; further, it is the mortality component of the DALYs lost that drives these results. Table 3. Burden of zinc deficiency, by country. Country Bangladesh India Pakistan Philippines Northeast Brazil Honduras Nicaragua Total DAL Ys lost (millions) 0.44 2.83 0.64 0.08 0.10 0.01 0.01 Percent share of DALYs of children under l m total DAL Y s 71 70 77 71 66 70 74 DAL Y s as percent of population 0.4 0.3 0.4 0 .1 0.2 0.2 0.2 Thus the burden of micronutrient deficiencies, both in tenns of the numbers of people affected, and its economic cost (even when valued at national GDPs), is extremely high.5 The nex.t section examines whether biofortification can lead toa substantial reduction in the burden of micronutrient malnutrition . Analyzing the reduction in burden of micronutrient deficiencies The extent to which a food-based intervention such as biofortification can help ameliorate micronutrient deficiencies depends on a number offactors; severa! steps are involved in achieving impact on health. First, once plant breeders have developed biofortified varieties, these have to be adopted by fanners. Conditional on adoption, once biofortified crops are produced, these have to be consumed by target groups in a form that minimizes processing losses of nutrients. Finally, enhanced micronutrient intakes have to translated into improved health outcomes and result in a reduced DAL Y burden. As with any new technology or public health intervention, at each of these steps, outcomes are uncertain. One way to account for these uncertainties is to specify probability distributions and compute the expected value of benefits. However, for many of the outcomes discussed here, such probabilíties are difficult to assign. Instead, we used a scenario ana1ysis where a range of plausible outcomes may be specified at each step, and benefits computed under the collective best and worst case scenarios. These assumptions are elaborated below. Likely coverage rates, by region The coverage rate, or the share of biofortified staples in production and consumption, is a key detenninant of the magnitude of impact. The more farmers produce, and thereby the more that target households consume biofortified staples, the greater the reduction in the prevalence of insufficient intakes. The biofortification strategy is to have micronutrient-dense trait in a mainstream staple foodstuff so that there will be a multiplicity ofbiofortified varieties available for each crop. 5 A direct comparison with WHO estimates ofthe DALY burden ofmicronutrient deficiencies is not feasible because of differences in methodology; however, the order of magnitude of their estima tes is similar to those presented bere (WH02006). 261 In this paper, we make assumptions on likely coverage, from both producer and consumer perspective, based on experience with the spread and diffusion of other modem varieties in these countries.6 For crops where the micronutrient trait is visible, such as with provitamin A, consumer acceptance also needs to be factored in. For thjs reason, we assume lower coverage rates for maize, sweetpotato and cassava than for high mineral rice and wheat. Experience suggests that farmer adoption with cereals in Asia, which has well- developed seed systems in place, coverage rates are likely to be high. As a conservative estímate, we assume a 30% coverage under a pessimistic scenario, and a 60% coverage in the optimistic scenario. In Africa, which does not have such well-developed seed systems, we use much lower coverage rates, with a pessimistic assumption of 20% and an optimistic assumption of 40% for all crops. In Latín America, coverage rates are assumed to range between 25 and 30%; in northeast Brazil, however, where coverage of new varieties of cassava has always been low, we assume 10-25% coverage for this crop (see Evenson and Gollin 2003 for summary of adoption data for maize, cassava and beans). Farmers in this region typically cultivate traditional varieties and do not receive much govemment support for agriculture (Gonzalez et al. 2005). Consumption of staples by target populations, and processing losses Clearly, the higher the level of consumptíon of a given staple food (including how many people consume the staple and how frequently), the greater the impact of a given increment in micronutrient intake will be. Thus, with a 400 gram daily intake of a given food, a 1 O ppm in crease in micronutrient content will translate into a 4 mg increase in micronutrient intake, whereas a 200 gram intake will translate only into halfthis amount. Obtaining data on food consumption and micronutrient intakes is difficult. For example, information on food intakes by crop for each age and gender-specific target group is scanty. Ideally, such consumption estimates should be based on individual-leve! dietary recall data, but unfortunately such data sets are rarely, if ever, nationally-representative. Where food composition tables and unit record data are available from dietary recall surveys, these have been used to derive micronutrient intakes. Where nationally- representative data sets are available, these tend to report food consumption at the household leve!, and not the individual leve!. When such data were available, as for example with Bangladesh and India, we used consumer equivalent units to derive food consumption at the individual leve!. In Latín America, we used regression techniques to identify consumer equivalence. For many countries in Africa, food consumptjon surveys are dated, and are based on smaller sample sizes. Therefore in these cases, we have used the most recent information available, and validated these figures through qualitative surveys. Additional details are contained in the individual reports listed in the references. Table 4 details the consumption figures used in each case (for ease of presentation the table reports data for only one target group: children under 6 years of age, but the calculations include all other relevant target groups). These range from approximately 100 grams of sweetpotato in Uganda, to about 225 grams of cassava (fresh roots) in the DR of Congo. For maize in East Africa, consumption levels are lower-they range from 70 grams in Ethiopia to 120 grams in Kenya. 6 For simplicity, we do not take into account any trade effects, or ofthe possibility ofbiofortified food aid 262 Table 4. Average staple crop intakes by children under 6, and assumptions on processing losses, by nutrient and country. Consumption Processing Josses Processing losses N utrient/Crop/Country among children Pessimistic Optimistic <6 years {grams/da;t} (%) (%) Provitamin A Cassava (fresh weight) DRofCongo 225 90 70 Nigería 176 90 70 Northeast Brazil 122 64 54 Maize Ethiopia 71 90 50 Kenya 120 50 40 Sweetpotato Uganda 96 25 18 Iron and Zinc Beans Honduras 56 5 o Nicaragua 45 5 o Northeast Brazil 57 5 o Rice Bangladesh 140 o o India 118 o o Philippines 121 o o Wheat India 87 20 10 Pakistan 94 20 10 For beans, consumption levels are much lower, approximately 45-55 grams per day for children under six in Latin Arnerica. Consurnption levels of rice among children, the staple food in much of Asia, are higber, and are between 120 and 140 grams per day. The corresponding consumption levels for adults are about 2-3 times as mucb. Finally, wheat consumption among young children is about 90 grams per day. Related to the level of consumption are processing Josses between the harvest and the plate. This is particularly important in the case of provitamin A. For example, sweetpotato and cassava are commonly sundried, which can result in the complete degradation of provitamin A. Other processes such as fermentation (to make gari in Nigeria or injera in Ethiopia, for example) can also reduce provitamin A content. Table 5 outlines the key parameters used for processing losses. 263 Table 5. Micronutrient content ofbiofortified crops under pessimistic and optimistic scenarios (parts per million). Crop/Scenario VitaminA lron Zinc Cassava* Pessimistic JO Optimistic 20 Maize* Pessimistic 10 Optimistic 20 Sweetpotato* 32 Beans Baseline 40 30 Pessimistic 80 40 Optimistic 100 50 Rice Baseline 3 13 Pessimistic 6 24 Optimistic 8 35 Wheat Baseline 38 31 Pessimistic 46 37 Optimistic 61 55 * Note: These crops currently have no betacarotene, the baseline is thus zero. Source: HarvestPius crop leaders and plant breeding coordinator. On the basis of qualitative surveys, it appears that processing losses are greatest in cassava in Africa, where between 70 and 90% of provitamin A may be lost in cooking (Manyong et al. 2005). In northeast Brazil, processing cassava into farinha (flour) gives losses between 54 and 64%. Maize-based foods are prepared by different methods so that processing losses differ on a country by country basis. In Ethiopia cooking maize to make injera (pancake-like bread) may lose as much as 90%, while in Kenya making ugali (a porridge) loses less, about 50%. Sweetpotato is usually boiled in cooking so that post-harvest losses of beta-carotene are relatively low, 18-25%. Note that there are no processing losses for rice, which is consumed in boiled fonn. Micronutrient content is expressed in milled form, thus milling losses are not relevant. Expected increases in micronutrient content Relatively low levels of consumption or high processing losses may be offset by a high micronutrient concentration in the biofortified crops. Since biofortification is still in the research phase for most crops, target micronutrient concentrations are based on the best-estimates from plant breeders, based on data from gennplasm screening. These are typically (but not always) higher than the mínimum incremental breeding targets that have been determined by nutritionists as being necessary for demonstrating health (biochemical) impact. 264 Current levels of beta-carotene in the widely-consumed varieties of bean, maize and sweetpotato ar currently ni!. For cassava and maize, breeders hope that under a pessimistic scenario, it will be possible to breed varieties containing 1 O ppm; and under an optimistic scenario this could be as high as 20 ppm (Table 5). The case of sweetpotato is different. Breeders bave already identified varieties that are higb in beta-carotene content, and these are being disseminated in east and southem Africa on a pilot basis. Tbe average beta-carotene content ofthese orange-flesbed sweetpotato varieties is approximately 32 ppm. Thus unlike the case witb cassava and maize, wbere varieties bigb in beta-carotene bave yet to be developed, there is more certainty about the tecbnical parameters that underlie the DAL Y analysis for sweetpotato. The expected increase in iron concentration ranges between 3 and 5 ppm for milled rice, 8 and 23 ppm for wheat and 40 and 60 ppm for beans. The increase in zinc concentrations is likely to range between 11 and 22 ppm for rice, 6 and 24 ppm for wheat and 1 O and 20 ppm for beans. It is important to note that these increases are all expected to be acbieved using conventional breeding techniques; none of the scenarios pertain to wbat might be acbieved using transgenic methods. For example, conventional breeding methods cannot enhance the provitamin A of rice because there is no known naturally-occurring genetic variation for this trait that breeders could exploit. We did not consider the transgenic provitamin-A-dense "golden" rice in this study. Dose response Tbe impact of any food-based intervention depends on the dose-response to increased nutrient intakes. ldeally, this entails determining a biologically-based relationship between enhanced micronutrient intakes and nutritional outcomes. Many such relationships are based on step functions, where the response is measured to a nutritional supplement, whicb usually translates into intakes above the recommended dietary allowance (RDA). Theoretically, bowever, the relationsbip is a continuous one. Here we use an inverse hyperbolic function to capture this continuum, as proposed originally by Zimmerman and Qaim (2004) (Figure 1) and elaborated by Stein et al. (2005). Adverse health outcome \ ~ ::; !:; ¡¡-\ ;: rr \ a :111 \ ¡;· ~ ¡¡- c:r ' rr s· \ O' :.., ' . ~S ~ :n :., ¡~ :e ¡ ' :. Cl ¡: ' ,¡ ~ : ...... : ...... ¡ B -- : -- A Micronutrient intake Figure l. Modeling the impact of increased intakes on bealth outcomes. Source: Zimmerman and Qaim (2004). 265 An increase in intakes from biofortification results in a reduction in the burden of deficiency of a magnitude given by the ratio ofthe areas A and A+B. A hyperbola that intersects the horizontal axis at the ROA fixes this functional formas 1/x - 1/RDA. 7 Note that the use of this function implies that the greater the distan ce between current intakes and the RDA, the greater the impact of a given increment in dietary intakes. This is in line with well-established principies in nutrition, which suggest that individuaJs with poor initial nutritional status have a higher biological response to an intervention than those with better initial nutritionaJ status. The bioavailability and absorption of the additional micronutrients that are available through the consumption of biofortified staples is also important. For the purposes of this paper, we assume that the diets of target populations are characterized by low bioavailability of the relevant micronutrient, and that this situation will prevail, as diets continue to be based on cereal/root crops. Thus, in computing the deficit in intakes we use the ROAs corresponding to low bioavailability for iron and zinc. Also, to ensure general comparability for all cases we consider in this paper, we use the same RDA for all countries. 8 These ranges in assumptions and parameters are u sed to derive the impact of biofortification in reducing the DAL Y burden of vitamin A, iron and zinc deficiencies under both pessimistic and optimistic scenarios. A summary of data sources used in making these computations is listed in Appendix Table l. Impact on V AD The percentage reduction in the burden of V AD ranges between 3 and 30% in the case of cassava, and between 1 and 32% in the case of maize (Table 6). In the case of sweetpotato, between two-fifths and two- thirds of the burden of V AD may be eliminated through the successful dissemination of orange-fleshed varieties. The reason for the much greater impact of orange-tleshed varieties of sweetpotato is not difficult to discern. A child consuming 100 grams of OFSP with 32 ppm beta-carotene would obtain nearly haJf the recommended dietary intake of 440 RE (assuming 18% losses, and bio-conversion factor of 1: 12) from this one food alone. In contrast, a child consuming a much higher 200 grams of cassava with 1 O ppm beta- carotene, with 90% losses from processing, would obtain less than 4% of the recommended dietary intake of vitamin A. Similarly, a child consuming 120 grams of maize with 10 ppm beta-carotene, with 50% retention ofthe nutrient, would obtain slight1y more than 10% ofthe RDA. 7 ldeally the point of intersection with the horizontal axis should be a value greater than the RDA, as the RDA represents the leve! at which the requirements of most-but not all- individuals in the population are met. However, since the requirements of 97.5% of healthy individuals would be met at the RDA, and because a higher number can be determined only somewhat arbitrarily, we use the RDA. Note further that the use ofthe estimated average requirement is not appropriate here, as the focus is not on determining prevalence rates ofinadequate micronutrient intakes. 8 For example, for countries such as the Philippines, where diets contain more meat products tban the other countries considered in this study, a higher bioavailability figure may be more appropriate; indeed, the RDA figures commonly used in the Philippines are lower than those used here. 266 Table 6. Reduction (percent) in DAL Y burden of micronutrient deficiency through biofortification under pessimistic and optimistic scenarios, by nutrient crop and country. N utrient/Crop/Country Pessimistic Optimistic VitaminA Cassava DRCongo 3 32 Nigeria 3 28 Northeast Brazil 4 \9 Maize Ethiopia 1 17 Kenya 8 32 Sweetpotato Uganda 28 64 Iron Beans Honduras 4 22 Nicaragua 3 16 Northeast Brazil 9 36 Rice Bangladesh 8 21 India 5 15 Philippines 4 11 Wheat India 7 39 Pakistan 6 28 1 Zinc Beans Honduras 3 15 Nicaragua 2 11 Northeast Brazil 5 20 Rice Bangladesh 15 46 India 20 56 Philippines 11 39 Wheat India 9 48 Pakistan 5 33 1 In Pakistan, average iron intakes for young children are believed to be sufficient; bence the DAL Y calculations refer to the impact only of improved intakes among older children and adults. 267 The percentage reduction in DAL Ys lost is lower in Ethiopia than in neighboring Kenya because of much higher processing losses of beta-carotene (particularly under the pessimistic scenario) and the lower consumption levels of maize in Ethiopia. Indeed, under the pessimistic scenario, there would only be a 1% reduction in the burden of VAD in Ethiopia. In northeast Brazil, up to one-fifth of the burden of V AD is eliminated by biofortified cassava in the optimistic scenario. lmpact on iron deficiency Even though consumption levels of beans is low (50-60 grams/day), the influence of the incremental iron content is high, higher than for the other crops. The expected decrease in the burden of iron deficiency ranges from 3-22 percent in Central America and between 9-33 percent in northeast Brazil. For rice, the reduction in the DAL Y burden of iron deficiency is 4-8 percent in the pessimistic scenario and 11-21 percent in the optimistic scenario. Even thougb the expected increments are more modest (certainly as compared to beans), consumption is twice or more that of beans. Furtherrnore, anemia is more prevalent in South Asia is higher than in Central America. 9 lmpact on zinc deficiency The reduction in the DAL Y burden of zinc deficiency from biofortified beans is 3-20 percent ín Latín America. Biofortification ofrice in Banglasesh reduces the burden 15-46% ofwheat in Pakistan by 5-33%. This is not surprising, given that the incremental zinc density as well consumption is much higber for wheat and rice than for beans. Cost-effectiveness of biofortification The figures discussed above suggest that biofortification can lead to reductions in the burdens of micronutrient deficiencies, even though the reductions are modest in sorne cases ofthe pessimistic scenario. The next question is what are the costs of achieving these reductions,and how do they compare with the cost of other interventions. As noted earlier, costs per DAL Y saved provide a consistent way of ranking altemative interventions. The costs of biofortifcation involve research and development, adaptive breeding and dissemination and maintenance. lnvestment in basic research and development is incurred in the initial years of any planned intervention. Once promising parent lines are identified, there is a phase of adaptive breeding, where these traits are bred into popular varieties that are cultivated in target countries, a process that can take up to five years. Once dissemination takes place, sorne costs are incurred to maintain the trait over time. Thus, the bulk of the investment is upfront. The key components of the costs estimated ín this exercise are summarized in Table 7. 9 Note that tbe figures for India cited in another paper (Stein et al., forthcoming) are somewhat different; this is because a different methodology, one that utilizes unit record data to compute a distribution of intakes, has been used in computing the reduction in DAL Y burden. 268 Table 7. Key costs, by category, nutrient and country (US $ per year). Dissemination R&D costs R&D costs Adaptive and Nutrient/Crop/Country (years 1-4) (years 5-10) breeding maintenance costs costs (high assumEtion2 Provitamin A Cassava DRCongo 120,000 36,000 400,000 200,000 Nigeria 240,000 72,000 600,000 185,000 Northeast Brazil 677,000 206,000 500,000 100,000 Maize Ethiopia 145,000 57,000 300,000 100,000 Kenya 145,000 57,000 300,000 60,000 Sweetpotato Uganda 778,000 243,000 368,000 147,000 Iron and zinc Beans Honduras 35,000 14,000 70,000 20,000 Nicaragua 70,000 28,000 70,000 20,000 Northeast Brazil 448,000 181,000 700,000 200,000 Rice Bangladesh 320,000 110,000 100,000 100,000 India 1,600,000 552,000 800,000 200,000 Philippines 320,000 110,000 50,000 200,000 Wheat India 1,400,000 600,000 800,000 200,000 Pakistan 600,000 300,000 600,000 200,000 Source: HavestP1us budgets and country-specific expert opinion. The researcb and development costs used for this exercise are derived from HarvestPlus budgets, apportioned to countries taking into account both plant breeders' estimates of geographical allocations as wel l as production sbares. For example, breeding efforts for cassava are focused on countries both in Africa and Latin America, with equal emphasis on both; thus half the research and development costs are allocated to eacb region. Within a region, approximate production shares are used to a llocate costs, so that northeast Brazil accounts for two-thirds the costs for Latín America. With regard to iron and zinc in Asia, we attribute all ofthe research and development costs to India but add to this 20% for Bangladesh and 20% for the Philippines. While this accounts for 140% of total costs in these three countries, the error is on the side of overstated costs. Further, we do not attempt to disaggregate research development costs for iron and zinc and use the entire crop budget in each case. Again, while this may be double-counting, there is no natural way to separate tbese cost, apart from assuming a 50% sb.are to each mineral, as screening and breeding for both nutrients occur simultaneously. 269 We estimated costs of adaptive breeding for each country using the opinions of experts within each so that they are country-specific. The estimates show that adaptive breeding can cost up $400,000 to $600,000 per year for about 5 years for cassava and $800,000 and $100,000 per year for rice in India and Bangladesh respectively. Similarly, we used expert opinion and current budget levels in each country to estímate country-specific costs of dissemination and maintenance breeding. For the last category of costs we use pessimistic and optimistic scenarios to capture possible differences in the ease of dissemination of biofortified foods. In all cases we estimatedd incremental costs of incorporating nutrient-dense traits into plant varieties that are currently under development. Moreover, the estimates are for using conventional breeding techniques, so that regulatory costs associated with transgenic crops do not apply here. Costs and benefits are discounted at 3%, a figure commonly used in the health economics literature. All calculations assume a 30-year time horizon, with dissemination starting in year 1 O, and ceiling adoption levels (be they under the pessimistic or optimistic scenarios) achieved in year 20. The resulting cost per DAL Y saved estimates are presented in Table 8. The World Development Report for 1993 (World Bank 1993), which reviews many public health interventions, suggests that those costing less than US $150 ($196 in 2004 dollars) per DAL Y saved are highly cost-effective. 10 10 To cite the report: "Govemments need to ... rnove forward with ... promising public health initiatives. Severa! activities stand out because they are highly cost-effective: the cost of gaining one DAL Y can be remarkably low- sometimes less than $25 and often between $50 and $150" (World Bank 1993, p. 8 ). 270 Table 8. Cost per DAL Y saved with biofortificaiton under pessimistic and optimistic scenarios, by nutrient, crop and country Nutrient/Crop/Country VitaminA Cassava DRCongo Nigeria NE Brazil Maize Ethiopia Kenya Sweetpotato Uganda lron Beans Honduras Nicaragua Northeast Brazil Rice Bangladesh India Philippines Wheat India Pakistan Zinc Beans Honduras Nicaragua Northeast Brazil Rice Bangladesh India Philippines Wheat India Pakistan Cost per DAL Y saved (US$) Optirnistic Pessimistic 3.70 45.80 3.00 35.40 84.00 433 .60 5.00 98.00 9.50 44.10 4.20 9.70 19.80 ll3.60 23.30 139.00 12.50 56.30 3.40 10.50 2.90 9.60 49.60 197.30 1.10 6.60 6.50 33.30 48.50 422.80 208.40 1880.30 95.10 799.00 1.70 6.10 1.10 3.30 7.00 46.30 1.20 7.10 4.50 33.90 Provitamin A-dense cassava, maize and sweetpotato In the optimistic scenario, the costs per DAL Y saved for provitamin A-dense staples are all less than $1 O per DAL Y saved for all crops and countries with the exception of northeast Brazil. ln the pessimistic scenario, costs per DALY saved are stillless than $10 for sweetpotato. For cassava, they are between $35 and $46 per DAL Y saved in A frica, and greater than $400 in northeast Brazil. With maize, biofortification costs $44 per DAL Y saved in Kenya and $98 in Ethiopia (recall that this latter figure assumes only a 10% retention of betacarotene). Nevertheless, even in the pessimistic scenario, all but the northeast Brazil figures correspond toa highly cost-effective intervention. 271 An important question is how the costs per DAL Y saved with biofortification compare with those associated with other vitamin A interventions, by fortification and supplementation. The literature in this area is limited; we use the figures cited in the influential 1994 World Bank report, which were in tum drawn from Levin et al. ( 1993 ). They report that supplementation costs approximately US $ 9.3 per DAL Y saved (in 1994 dollars, corresponding to about $12 in 2004 terrns) and fortification costs are about $29 per DAL Y saved, equal to almost $37 dollars in 2004 terrns. These figures are not directly comparable with costs per DAL Y saved computed here for biofortification, primarily on account of the differences in the assumed rate of coverage. The calculations for supplementation and fortification asswne a 75% coverage rate, whereas with biofortification, coverage rates even under an optimistic scenario are assumed to be more conservative (50% in Africa). Despite these differences, costs of biofortíficatíon are much lower than those of fortification and supplementation in the optimistic case, the only exception is the northeast Brazil. Even in the pessimistic scenario, biofortification is relatively more cost-effective than the other two vitamin A interventions in the case of sweetpotato. We also ran a separate exercise, retaining all the other ' pessimistic' assumptions on increase in micronutríent content and processing losses, but enhancing coverage rates to 75%. With these higher coverage rates, biofortification costs are lower than those of fortification in all cases except in northeast Brazil, and comparable or lower than those of supplementation in the DR Congo, Nigeria and Kenya. However, we do not report these results in Table 8, as we believe that the 75% coverage assumption is unrealistically high (even for the altemative interventions). !ron-dense beans, rice and wheat With iron, as well, costs per DAL Y saved fall under the "highly" cost-effective category in the optimistic scenario. For rice in South Asia, they are particularly low at approximately $3 per DAL Y saved, and are somewhat higher in the Philippines at about $50 per DAL Y saved. E ven in the pessimistic scenario, the costs for rice are about $1 O per DAL Y saved in South Asia. Costs of iron biofortification are similarly extremely low for wheat in South Asia, as little as $1 per DALY saved. With high-iron beans in Latín America, costs are between $12 and $23 per DALY gained in the optimistic scenarío, and go up to $139 per DAL Y saved in the pessimistic scenario. Compared with iron supplementation, which is reported to cost about $17 per DALY saved, biofortification enjoys an advantage in the optimistic scenario in all regions and with all crops. Even in the pessimistic scenario, biofortification costs are lower in South Asia with wheat and rice, and are comparable in Latín Ameríca. As above, the fortification calculations assume a 75% coverage. To examine the impact of coverage rates, we re-computed the costs in the pessimistic scenario, assuming a higher 75% coverage, but retaining the pessimistic increment in iron content for the three crops. The results of this exercise suggest that the costs of biofortification would be comparable or lower than those of other forrns of supplementation in all regions and with all crops. Similarly, costs of iron fortification have been estimated at about $6 per DAL Y saved 11 Biofortification enjoys a relative advantage even here in South Asia certainly in the optimistic scenarío, and even in the pessimistic scenario if the higher 75% coverage rates are assumed. Only with high-iron beans in Latín America does fortification appear more attractive than biofortification. Once again, we do not report these numbers in Table 8, as the 75% coverage rates appear somewhat unrealistic. In making these comparisons it is important to note other differences in methodology as well. For example, Levin et al. (1993) use a $0.50 per capita cost for supplementation, and $0.20 per capita cost for 11 These figures are in 2004 dollars, and are converted from the $12.80 per DAL Y saved for supplementation and $4.40 per DALY saved for fortitication reported by Levin et al. (1993). 272 fortification to derive the costs per DAL Y gained. At least in the context of provitamin A, these costs have been questioned by sorne as being too low (Fiedler el al. 2000). Zinc-dense beans, rice and wheat Once again, in South Asia, biofortification is extremely cost-effective, with cost per DAL Y saved lower than $1 O even in the pessimistic scenario for both wheat and rice. Costs per DAL Y saved with beans in Latín America are higher, but still cost-effective in the optimistic scenario. lt is only in the pessimistic scenario that costs per DAL Y saved exceed $196 by a wide margin. A comparison ofthe zinc biofortification numbers with that of other zinc-based interventions is difficult as these are primarily used therapeutically (as for example administration of zinc supplements). 12 Discussion and Conclusions This paper presents, for the fLTSt time, evidence from a large number of countries and crops that biofortification can have important impacts on the burden of micronutrient malnutrition. Moreover, it does so in a cost-effective manner, with most costs per DAL Y saved falling in the "highly" cost-effective category. Also, in all cases but one, benefit-cost ratios of biofortification exceed one, that is, benefits far outweigh costs. These results are encouraging for biofortification, especially since the underlying cost assumptions are often estimated on the high side, as for example, with the ' double counting' of costs for increasing the concentration of two minerals in the same crop. Depending on the context and the scenario, and subject to the caveats noted in the text above, biofortification also appears to be more cost-effective than supplementation or fortification. ln South Asia, biofortification enjoys a clear advantage, given that the high proportion of the rural populations in these countries, and the relatively well-functioning seed systems there. Similarly, in Africa, biofortification enjoys an advantage over supplementation in many cases, both in the optimistic scenario, as well in the pessimistic scenario when higher coverage rates are assumed. Relative to other interventions, the only instances where biofortification may not enjoy an advantage are in Latín America, with cassava in northeast Brazil, or with zinc in Latín America especially in the pessimistic scenario. In interpreting these results, it is important to bear in mind that these figures are based on national averages, because this is the way the data of prevalence of deficiencies are reported. However, we caution that data of national averages can mask considerable variation within any one country, especially where consumption data are concemed. Because of this the impact of biofortification in areas of high staple consumption and micronutrient deficiency can be greater than implied by considering national data. The analysis here considers the impact ofthe consumption ofa single biofortified staple. In actuality, diets often consist of more than one staple ( cassava and beans, rice and wheat, or maize and beans, for example ). ln these situations, the consumption of more than one biofortified staple is likely to have an enhanced impact (for example, if vitamin A improves iron absorption). Capturing the impact of an intervention with multiple micronutrients by taking into account their interactions is an area for further research. 12 As an additional exercise, we also compute benefit-cost ratios, as these are commonly reportee!. Ratios that exceed unity are indicative of a worthwhile investment. These require benefits to be monetized; that is, a do llar value needs to be assigned to the DAL Ys saved. Needless to say this valuation is problematic: if GDP perca pita is used to value benefits, this tends to favor high-income countries. We use a somewhat-arbitrary US$1 000 per DAL Y saved for all countries. The results in Appendix A suggest that benefit cost ratios are all high, and well exceed unity in a11 cases. The only exception is the case with zinc in Nicaragua in the pessimistic scenario, where the value ofthe benefits appears too low to justify costs. The use of an altemative figure, say of US $500 per DAL Y saved, does not alter the substance ofthe results: biofortification continues to be cost-effective, except that with this lower valuation of benefits, biofortification ofbeans with zinc in Latín America is also no longer feasible. 273 The challenges to implementing biofortification should not be underestimated. Attention will need to be paid to community awareness, dissemination, and behavior change interventions, many of which are commonly used in health and nutrition programs but are foreign to agriculture. This will be especially true when the micronutrient trait is visible, for example, in the case of the color change due to increased provitamin A content. The results of this analysis suggest that the pay offs from línkíng both agriculture and public health approaches, which often function independently, can be very high. In sum, the analysis above suggests that biofortification is a viable strategy, and an important complement to the existing set of interventions to combat micronutrient malnutrition. 274 Appendix Table l. Benefit-cost ratios ofbiofortification under pessimistic and optimistic scenarios, by nutrient and country. Nutrient/Crop/Country VitaminA Cassava DRCongo Nigeria NE Brazil Maize Ethiopia Kenya Sweetpotato Uganda Iron Beans Honduras Nicaragua Northeast Brazil Rice Bangladesh India Philippines Wheat India Pakistan Zinc Beans Honduras Nicaragua Northeast Brazil Rice Bangladesh India Philippines Wheat India Pakistan Benetit-cost ratios Optimistic Pessimistic 271 33 1 12 198 105 239 39 43 63 296 347 20 931 153 16 4 8 590 894 144 807 222 275 22 28 2 10 23 103 9 7 18 95 104 5 153 30 2 <1 1 163 308 22 141 29 References Alderman, Harold, Jere Behnnan and John Hoddinott (2004). Hunger and Malnutrition, Background paper for the Copenhagen Consensus. Bouis, Howarth E., editor, (2000). Special Issue on lmproving Human Nutrition through Agriculture, Food and Nutrition Bulletin, volume 2 1, no. 4. Evenson, Robert and D. Gollin (2003). Cop Variety Improvement and Its Effect on Productivity: The Impact of Intemational Agricultura! Research (Cabi Publishing). Fiedler JL, Dado DR, Maglalang H, Juban N, Capistrano M, Magpantay MV (2000). "Cost analysis as a vitamin A program design and evaluation too!: a case study of the Philippines," Social Science and Medicine S 1 (July):223-42. Haas, Jere D, John L Beard, Laura E Murray-Kolb, Angelita M del Mundo, Angelina Felix and Glenn B Gregario (2005). "lron-biofortified rice improves the iron stores of nonanemic Filipino women" Joumal of Nutrition, vol. 135, pp. 2823-2830 Hotz, Christine and Kenneth Brown (2004). Assessment of the risk of zinc deficiency in populations and options for its control: Intemational Zinc Nutrition Consultative Group Technical Document l. Gillespie, Stuart (1998). Major issues in the control ofiron deficiency, Micronutrient Initiative. Gonzales, Carolina, 1 Kruze, L Sequiera, C Fukuda, R Olivera and N Johnson (2005). "Findings of the qualitative survey on cassava and beans in Paraiba, Brazil," mimeo. Horton, Susan (1999). "Opportunities for investment in nutrition in low-income Asia" Asian Development Review, vol. 17; pp. 246-273. Horton, S usan and Jay Ross (2003), "The economics of iron deficiency" Food Policy, vol. 28, pp. Sl -75. Manyong, Victor, A.S. Bamire, J.P. Banea 1.0. Sanusi, D.O. Awotide, A.G.O. Dixo and 1.0. Akinyele (2005), "lmpact and policy analysis ofbiofortified cassava-based diets in West and Central Afiica" mimeo. Murray and Lopez (1996), The Global Burden ofDisease (Cambridge University Press). Stein, A. P. Nestel, J. V. Meenakshi, M. Qaim, H.P.S. Sachdev and Zulfiqar A. Bhutta, ( forthcoming), "Piant breeding to control zinc deficiency in India: How cost-effective is biofortification?" Public Health Nutrition. Stein, A., J. V. Meenakshi, M. Qaim, P. Nestel, H.P.S. Sachdev and Z. Bhutta (2005), " Analyzing the Health Benefits of Biofortified Staple Crops by means of the Disability-Adjusted Life Years Approach: A Handbook Focusing on lron, Zinc and VitaminA," HarvestPlus Technical Monograph 4, Washington, D.C. and Cali, 2005 United Nations Standing Committee on Nutrition (2005), Fifth report on the world nutrition situation: nutrition for improved development outcomes (Geneva). van Jaarsveld, Paul J, Mieke Faber, Sherry A Tanumihardjo, Penelope Nestel, Carl J. Lombard and Ambrose J Spinnler Benade, 2005. "P-carotene-rich orange-fleshed sweetpotato improves the vitamin A status of primary school children assessed with modified-relative-dose-response test," American Joumal of Clinical Nutrition, vol. 81, pp. 1080-7. World Bank (1993), World Development Report 1993, (Washington D.C.) Yanggen, David, Health and Economic lmpact Analysis oftbe Introduction ofOrange-Fleshed Sweetpotato (OFSP) in Uganda Using Disability Adjusted Life Years (DALY) Analysis. World Healtb Organization (2006), The Global Burden ofDisease Project Revised Estimates for 2002. Accessed from http://www.who.int/healthinfo/bodgbd2002revised/en/index.htrnl Zimmerman, Roukayatou and Matin Qaim (2004). "Potential health benefits of Golden Rice: a Philippine case study" Food Policy vol. 29, 147-168 . 276 Spatial trade-off analyses for site-sensitive development interventions in upland systems of Southeast Asia 1 Annual Progress Report: CIA T Laos February 2007 General lnformation Project Title Spatial trade-off analyses for site-sensitive development interventions in upland systems of Southeast Asia StartDate December 2004 End Date December 2007 Budget €117,000 (US$141 ,000) for2006 CGIAR Centre CIA T (Centro Internacional de Agricultura Tropical) Project Leader Douglas White Partner: Department of Sustainable Agricultura\ Systems BOKU Department of Economics and Social Sciences Department of Landscape, Spatia\ and lnfrastructure Sciences Research for Development Forum Project Leader Michael Hauser Other Partoers • National University ofLaos (NUoL) • National Agriculture and Forestry Research Institute (NAFRI) • Oudomxay Community Initiative Support Project (OCISP) • lnternational Fund for Agricultura\ Development (IFAD) • CIA T-Asia Regional Office Narrative Summary Background The project has completed its second year with significant advances with respect to the three research-for- development (R4D) themes of: (1) spatial analysis of changing natural resource endowments, (2) participatory market chain analysis, and (3) collaborative livelihood analysis and opportunity identífication support. Línks to other projects and development organizations, both government and non-govemment, continue to deepen and broaden. Project goals To contribute to equitable and sustainable improvement of livelihoods of rural communities in northern Laos with síte-specific identification and implementation of intervention strategies, and ultimately in similar upland systems elsewhere that rely on shifting cultivation. 1 This article is essentially the report subrnitted to the donor, the Austrian Government but without the budget details. 277 Deveiopment objectives • To identify, design, share, adapt, and implement intervention strategies with local communities and rural development partners that are sensitive to site and market conditions in order to optimize social, economic and ecologic benefits. Scientific objectives • To develop and apply a robust and cost-effective method that identifies and quantifies agronomic production risks that arise from the spatially-variable suitability of land resources to intensification and diversification. • To analyze product supply chains for their ability to provide equitable distribution of economic benefits. • To develop and implement effective methods to co-identify livelihood strategies with households and communities. Previous achievements (2005) • Collaborative research plan refined with a major IF AD development project in the province: the Oudomxay Community lnitiative Support Project (OCISP), an eight-year integrated rural development project. 2 • Spatial analyses of income generation by NTFP products commenced to discem pattems and their causes. Data provided by [f AD. • Links formalized with the East-West Center and National University of Laos (NUoL) project: Understanding Dynamic Resource Management Systems and Land Cover Transitions in Montane Southeast Asia. 3 A small financia! grant from USAID (US$ $16,000) enabled a comparison with a similar R4D effort in the neighboring Luang Namtha province. • Research activities conducted with the Sma/lscale Agro-enterprise Development for the Uplands of Vietnam and Laos Project- SADU (funded by the Swiss Agency for Development Cooperation- SDC). Achievements (2006) • Collaborative Livelihood Analysis and Opportuníty-identification Support (C-Laos) approach developed, refined and implemented with 4 communities, 12 OCSIP extension staff, and 3 scientists from the National Agriculture and Forestry Research lnstitute (NAFRI). • Market supply chain of paper mulberry, an important non-timber forest product (NTFP), co- analyzed with 42 diverse participants (farmers, traders, govemment staff, manufacturers, exporters, and development organization staft). • Village and local trader action plans developed. • Implementation of action plans started. • Identification of the pattem and magnitude of spatial and temporal land cover changes in Oudomxay province using spatial analysis from time series of Landsat 7 ETM+ images and ground reference data from field work. Predominant land cover classes include permanent agriculture, fallow and forest. 2 The OCISP project is financed from three different sources: an fFAD loan project of 10.8 million SOR (- US$l5m), a grant of€1.7 million from the Luxembourg govemment, US$1.7 mili ion grant from the World Food Program. The Austria-funded project contributes to the 5th component ofOCISP effort by improve farming systems and natural resources management for increased farm production and income. 3 http://www.eastwestcenter.org/res-pr-detail.asp?resproj ID= 196 278 • Research support from 4 BOKU students (1 PhD, 3 MSc). o Maria Miguel Ribeiro: Supply chains of non-timber forest products and Leaming alliances in Oudomxay, Northem Laos o Kendra Leek: Natural resources and rurallivelihood strategies in Oudomxay o Antonia Fay Forster: Classifying fallow period lengths using Normalized Difference Vegetation Index (NDVI) o Kathrin Leitner: Women 's views on paper mulberry in Lao PDR: Local knowledge, perceived potentials • Poster presentation atan IDRC (Canada)-NUoL workshop in Vientiane, Laos: Comrnunity-based Natural Resource Management Research and Capacity Building. Planned activities (2007) The third year of the project will conclude field research activities at mid-year. Emphasis will shift to consolidate, interpret, and synthesize results. Concerted efforts are planned to discuss, share and publish results with other CIA T project efforts and similar R4D projects in Laos. Details of planned activities are in the section on Project implementation below. Management issues No personnel issues affected the project and financial resources are ample. A portion of unspent funds is being used to cover researcher salaries given that core fund availability has markedly decreased since project inception. Research site is in Oudomxay and CIA T-Asia office (in Vientiane) pro vides management support. Research process Main results and milestones are outlined below in project implementation. Project advances in Laos have influenced curricula at BOKU (Department of Sustainable Agricultural Systems) and fostered discussion with another CIAT-BOKU project in Africa The follow aspects ofthe project addresses the ADA goals: 1) Sustainable increase in agricultura/ productivity: • NTFPs for household food and economic security • Livelihoods analysis of resources and opportunities 2) Conserva/ion and efficient use of natural resources and biodiversity: • Sustainable management strategies for NTFPs • Spatial analysis ofthe productive capacities ofnatural resources 3) Development of sustainable production and marketing systems: • NTFPs production potential and altemative options • Market chain analysis for equitable distribution ofbenefits • Leaming alliances for market identification and innovation 4) Strengthening institutions and improving policy development: • Leaming alliances for institutional analysis, public-private partnerships and empowerment. • Collaboration with IFAD project and government ministries for project short and long tenn impacts. 279 Capacity-building and training achievements: • With the Learning alliance workshops, interaction and communication between stakeholders of the supply chain of paper mulberry bark resulted in: o Diffusion oftacit knowledge along market chain o Development of an action plan made by participants o Experimentation with new knowledge on production, post-harvest and market requirements. • Farmer and govemment extension training in: o Paper mulberry bark harvest and post-harvest techniques. • 42 participants: farmers, traders, govemment staff, manufacturers, exporters, and development organization staff o Livelihood analysis and opportunity identification. • 12 OCSIP extension staff • 3 NAFRl scientists Problems encountered and lessons learned related to capacity building: • Although farmers are interested in producing other crops, farmers prioritize the staple crop rice. Consequently, participation in the training on paper mulberry bark production and post-harvest techniques was lower than expected. Lesson: For many rural Lao, identifying livelihood options is new. Diversifying from rice cultivation, a centuries-old tradition, requires process of review and analysis with farmers. • Lack of leadership in one of the villages has led to lower farmer participation in training and other type of events. Lesson: Leaders, or champions, play an important role in fostering village interest and participation. Again, facilitating livelihood change is a process that may time extra time and effort. • Training local stakeholders may not benefit women farmers if govemment extension staff cannot communicate in Khamu language. Lesson: Extra effort is required to facilitate equitable benefits. Village men can help translate Lao to Khamu. • Provincial and district extension staff are accustomed to using standard questionnaire formats. Lesson: Training sessions required discussion of development processes and the dual role of extensionists (solution providers and opportunity identifiers/facilitators). For survey instruments, the C-Laos approach concentrated on developing ways to foster mutual leaming instead of extractive data gathering. The perspective of having conversations instead of conducting interviews addressed the process and content of extensionist-farmer interactions. • Efforts to encourage conversational interactions, a review of farmer responses, and critica! thought require extra time and discussion with extension staff. Lesson: Despite a difference in culture and educational backgrounds, extensionists appreciated learning different ways to interact with fanners. A new training technique to the extensionists, role plays as farmers, víllage leaders and extensionists, was effective in encouraging participation, discussion and debate. 280 Dissemination of outputs • Development of project websites: http://gisweb.ciat.cgiar.org/Austria-Laos and http:/ /www. wiso.boku.ac.at/laos.htm 1 • Two posters presentations at the KEF Commission for Development Studies MDG+S workshop: A critical look at the role ofresearch in achieving the MDGs (November 2005). • Market-led options to improve farmer livelihoods in Oudomxay, Laos • Building a learning alliance in Lao POR • The project presented one poster at the IDRC workshop: Community-based Natural Resource Management Research and Capacity Building (March 2006) • Fostering site-specific market options to improve rural livelihoods and land management in Laos • Project results summarized in CIA T annual reports. • Project brochure (4 pages) developed to disseminate at meetings and discussions. Project implementation The effects of language, weather, and the agricultural calendar had the most impact on project implementation. Oiscussions and interviews with farmers often require a double translation (Khamu-Lao- English). Subtle and abstract concepts need careful explanation. In order to avoid leading a dialogue and question process, the use of analogies and personal experience needs to be employed carefully. Field activities are difficult to realize during the rainy season. Roads to villages become dangerously slick when wet. All fieldwork needs to be coordinated to avoid busy periods of the agricultura] cycle (e.g. planting, harvest). • Livelihood analysis: Livelihood transitions in the uplands of Lao POR are rapid and wide ranging. The aim of this research is to: (1) explore livelihoods transitions in different bio-physical, economic and social contexts, (2) examine the potential for extensionists to faci litate community discussion about tradeoffs associated with livelihood transitions. o Tradeo:ffs include changes to: • Household food and economic security, • Resource impacts (soi~ forest, water), and • Cultural traditions at the household (e.g. gender) and community level (e.g. social equity). o The C-Laos approach at the household level is an adaptation of the community-level work of 2006. Extension staff will implement the dialogues with households (n= 54) in four communities. Steps include: o Recognising live\ihoods strengths (goal instead of problem-oriented approach) • Envisioning future livelihood goals • Oesígning Iivelihood development strategies • Drawing and implementing livelihood actíons plans • Conducting participatory monitoring and evaluation o A fourth MSc student will conduct research activities in Oudomxay to: • Detennine the Iikely effects of new livelihood strategies on household and community institutions/traditions. • Identify development policies that foster equitable outcomes. • Adapt researcb with NUoL and NAFRI experience and bui ld capacity. o A synthesis will be conducted of 36 enterprise budgets compíled by lF ADIOCSrP of predominant agriculture and livestock systems. o Tradeoff (economic-social-environmentaJ impacts) and scenario analysis of livelihood options coordinated with results of spatial analysis (June). 281 • Market analysis o Refinement of paper mulberry supply chain analysis. o Implementation of action plans for villages and local traders. o Conceptualization and participation in Southeast Asia NTFP workshop (SNV, FAO) in Vientiane {April). o Co-development ofsummary NTFP information (with SNV, NAFRI) for development workers and policymakers regarding optimal conditions and requirements of • Agro-environments • Management • Post-barvest and marketing • Spatial analysis o Detailed map ofland coverchanges from study years 2001 to 2003. o Summary maps ofhuman pressure on land use created. o Trends ofhuman pressure on land use estimated. o Results discussed with villages and development projects (IF AD). o Conference poster presentation Analysis ofTime Series of Image Data. Bozen, ltaly June 2007) http:/ /las.physik. un i-o ldenburg.de/timeseries-workshop/ • Synthesis of project results and implications discussed with project partners. • Co-organization, participation, and publication of synthesis and Jessons-Jearned workshops at two sea) es: o CIA T projects in Southeast Asia. o Other major research-for-development projects in Laos (NAFRI, SDC, SNV, SIDA, IDRC, IFAD, etc). 282 Assessing the potential impact of the consortium for improved agriculture- based livelihoods in Central Africa (CIALCA): Spatial targeting of research activities Andrew FarroW', Lilian Busingyea, Paul Bagenzea With contributions and cornments by Steffen Abeleb, Piet van Astenb, Guy Blommec, Dieudooné Katun~ Musaled, Speciose Kantengwad, Valéry Kaserekaa, Chris Leggb, Jean-Paul Lodi Lamaa, Pascal Sangingaa, Kai Sonderb, Bemard Vanlauwea a/nternational Center Jor Tropical Agriculture (Centro Internacional de Agricultura Tropical - CIAT) b International Institute ofTropical Agriculture (liT A) c/nternational Networkfor the /mprovement of Banana and Plantain (IN/BAP) dConsortium for Improved Agricultura/ Livelihoods in Central Africa (CIALCA) Introduction Consortium for improved agriculture-based livelihoods in Central A frica In 2005, the Directorate General for Development Cooperation (DGDC - Belgium) approved three proposals for projects working with the national agricultura) research systems of Rwanda, Burundi, and the Democratic Republic of the Congo (DRC) and in similar geographic regions. The three projects are led by TITA, Bioversity-INIBAP, and TSBF-CIAT, focussing on banana-based systems in the frrst two cases and on sustainable natural resource management and marketing, coupled with resilient legume germplasm in the third project. In order to exploit these similarities in project objectives, structure and sites, the three COlAR institutes and their NARS partners proposed to operate as a consortium. Using the different capacities of the participating institutes, the consortium aims to enhance research synergies, while avoiding needless duplication ofresearch activities. The setup ofthe Consortium for lmproved Agriculture-based Livelihoods in Central Africa (CIALCA) was endorsed by the Director-Generals of INERA, ISAR, ISABU, IRAZ1 in Kigali, on 15-16 September 2005. CIALCA mandate areas CIALCA has chosen to work in ten geographical areas in Burundi, DRC and Rwanda. These ten 'Mandate Areas' (Figure 1) have been chosen to represent the diversity in agro-ecological characteristics, in demographic profile and in access to markets that are encountered in the three countries (Table 1). The mandate areas eh osen also reflect the areas where bananas and legumes are an integral part of the farming system (Figure 2). 1 l' Institut de l'Environnement et de Recherches Agricoles (INERA- DRC), lnstitut des Sciences Agronomiques du Rwanda (ISAR), l'lnstitut des Sciences Agronomiques du Burundi (ISABU), lnstitut de Recherche Agronomique et Zootechnique (IRAZ) - regional mandate based in Burundi. 283 N A o Location of Mandate Areas in Central Africa KJNSH.A.SA • / ~ d ~. KIGAU BUJUMBURA Lubumbasm • 4 - ICibii,~..Qn''' 0:J- Gitoro•e -6 - Qt~J• -7 - ICIJIIII.ICJ,_,JO D•· ICII'IIIdo O 9 - Hord-ICw• -•topen - ro - s.d-«Jw •o•t•JI~"' - H · U-tore Figure l. Mandate areas in Central Afríca - Bean and Banana Procluction in Rwanda and Burundi :·. ·.· .. : .. . . . .. . . .. .. . . : · .. :·. . . ..... .. . , ·.·: . . . . ·... . . ·. , ..... . . . .. .. • Gtits O Mllndatt Artas O 01untry Boundorles Lahs S.an Production SOOHa Banana Prodlction W Principal Arta Figure 2. Bean and Banana productíon in Rwanda and Burundi (data unavaílabte for banana production in DRC). 284 Table l. Characteristics of the 1 O mandate areas. Approximate Biophysical Description/reason M a in N ame Country political eh a racteristics for interest markets boundaries Bas-Congo Cataractes, Lowlands, close to Kinshasa, DRC Lowlands Mbanza-Lukaya district Kinshasa Ngungu Buruncli Cibitoke Banana good access Ruzizi plain 1 province/ district Ruz izi plain to Bujumbura, Bujumbura Rwanda lowlands (&OOm) Kigali, Gisenyi- Gisenyi, kibuye Highlands, high Bananas, beans with Gisenyi, Kibuye Rwanda provinces ppt, relatively good growth potential Goma, young soils Ruhengeri, Kibuye High population Kigali, Gitarama Rwanda Gitarama Granite plateau, density good market Gitarama, province poor soils access, poor soils on Ruhango granitic plateau Banana, representative of Gitega Burundi Gitega province Poor acidic soils, central area, high Gitega, central plateau, population density, Bujurnbura with input use common 900-1200mm, Kigali- Kigali, kibungu 900 - 1700m Bananas, beans, Kibungu Rwanda provinces mid-altitude, old potentially good Kigali weathered soils access to Kigali on schist 900-1200mm, Banana area, similar 900 - 1700m to Kibungu, Kirundo, Kirundo Burundi Kirundo province mid-altitude, old intemational NGOs potentially to weathered soils are there Kigali on schist Beni, Butembo, Ngweshe, Banana-bean area, Kasindi, Kivu Kabare, Masisi, high population, high Kampa\a, montagneux• DRC Butembo, Highlands rainfall, potential Bukavu, Kamanyola good access to urban Goma, (territoires) centres Kigali, Cyangugu, Gisenyi 800-IOOOmm U mutara with drought Recently settled, U mutara Rwanda province stress, potentiaUy good Kigali potentially good access to Kigali NRM base • Kivu montagneux has been split into Sud-Kivu montagneux and Nord-Kívu montagneux. 285 Representativity and scaling-out A number of locations have been chosen as candidates for action si tes. At each location a participatory rural assessment (PRA) was carried out to determine the important characteristics of the communities, such as the major farming systems, their engagement with markets, and to gauge the presence and strength of local organisations. As part of the characterization of mandate areas we ha ve al so analyzed the representativity of each PRA site within the mandate site as a whole in terms of the major variables described above. In addition an exploratory scaling-out exercise has been carried out for each PRA site in order to judge poten ti al diffusion areas and beneficiaries of the technologies developed by the CIALCA projects. Methods and Materials Data Each mandate area was characterized in terms of certain key variables, these included the agro-climatology of the area, the population density of the area, and the access to markets. Population density was provided by the Global Rural Urban mapping project (GRUMP 2005) datasets for Africa. This dataset was chosen to ensure consistency across the three countries. GRUMP is inferior to sorne publicly-available data for Rwanda (the GRUMP data are older and the spatial resolution is poorer) but it is the only reliable source of data for DRC. Different markets were chosen for characterizing the mandate areas (Table 1 ); these were based on local knowledge of current trade flows as well as for the identification of potentially new markets. Accessibility was then calculated to all of these markets. The access is modelled in a geographical information system (GIS) using a set of rules and data, which results in a value in minutes toa pre-determined market (Farrow and Nelson 2001). The model takes into account road location and quality as well as barriers such as international borders and constraints to movement, like slope. The values generated were validated by the project team coordinators in Rwanda, Burundi and DRC. Each market was modelled individually, assuming a transport mode of medium-sized trucks able to carry agricultura! goods (both input and output) in bulk2• This Jinks with CIALCA's research on collective action for marketing outputs and bargaining power for the purchase of inputs. Due to lack of digital soils data, the characterisation of the agro-climatology was limited to the annual precipitation (Hijmans et al. 2005) anda calculation ofthe length of growing season (Thornton et al. 2006). An indication of the soils can be derived by the underlying geology. For Rwanda geological maps are available in scanned format (Selvaradjou et al. 2005) and can be used for those mandate areas located in that country. No data were available for Burundi or DRC. The results of characterising the mandate areas were combined to form development domains similar to those used for priority setting by the Association for Strengthening Agricultura! Research in East and Central Africa (ASARECA). The agro-ecological potential !ayer was the same as that used by IFPRI- ASARECA. Slight modifications were made to the ASARECA development domains, especially in the choice of markets and in the threshold values used. Assessing representativity of P RA sites While it will be the results from the PRA that decide which sites wíll be chosen as the locations for interventions3, it was deemed important by CIALCA that the PRA si tes themselves are broadly representative of the mandate area. This has slightly different connotatíons in each mandate area but in general the researchers were keen to avoid outliers in terms of the major biophysical and socioeconomic 2 Where no roads or tracks are mapped, the model uses values for walking. 3 Half ofthe PRA sites in each mandate area will be chosen as action sites depending on a number offactors including the strength of community organisations, and current engagement with markets. 286 variables used to characterise the mandate areas. A samp\ing framework is traditionally devised to ensure representativity and to enable inferences to be drawn about the population at large. However given the small number of intervention sites, our analysis of representativity is visual and intuitive rather than statistically rigorous. We compare the values of the characteristics at the location of the village where the PRA was undertaken with the values of the mandate areas as a whole. All of the PRA sites are plotted on the histogram of observations. A histogram can be produced for each variable where each observation refers toa lkm2 cell. The interpretation of the annotated histogram is improved4 when the variable is weighted according to the population density as shown below (Figure 3). Annual Rainfall 1000 ~----------------~ 1600 >o 1400 u 1200 ; 1000 :S 800 r soo 11. 400 ,,.. • Frequency • PRAs~e 1 o. - - 200 o n,~ '>~ .k.~ _m~ ,(\~ n,~ ,.~ _v,~ _m~ ''" ... +---w-'.1~~[ ol.olooq-...,-¡.----4 ... ~ ""' ""- o\'1' ~ ,ro ~ ""(j"' ~ - ... - - - - - - mm (a) (b) Figure 3. Histograms ofannual rainfall in Bas-Congo mandate area; (a) unweighted, and; (b) weighted by population density. Sca/ing-out as a guide to potential impact - The interventions that will be carried out as part ofthe CIALCA projects in the action sites will be analyzed and the best practices will be scaled-up via development partners and the national agricultura! research and extension systems. Another aspect of"going to scale" is the horizontal replication across the three countries and beyond (Cook and Fujisaka 2004). The potential areas ofthis scaling-out are determined in part by the interventions themselves, which are constrained by various externa! factors, notably the edapho-climatic conditions. the receptiveness ofthe livelihood systems and access to input and output markets. We have used two methods of assessing the potential for the scaling-out of interventions in the action sites. The first and most fully developed is the use of the Homologue software (CIA T 2004) to search for environments similar to those encountered in each mandate area. The second method has resulted in the creation of development domains based on key variables related to agricultura! livelihoods (e.g .• Pender et al. 2004). 4 The PRA site should be representative ofthe population ofthe mandate area rather than the area per se. 287 Homologue Homologue produces raster grids of the probability of encountering an environment similar to the input location. For each mandate area we chose the position of each PRA as the input location. These individual grids were then merged to form a probability cloud, whose val u es varied between 1 ( 1 00% chance of encountering a similar edapho-climatic environment) andO. The probability 'cloud' was then used to give the population in the probability band between 50% and 100%. No account was taken ofthe actual diffusion mechanism, that is whether the diffusion was a natural process or was aided by NARS and NGOs. ASARECA Development Domains For scaling-out purposes we have decided to use the developrnent dornains developed for ASARECA (2005). These domains have been created using many more markets than those used in the characterization of the CIALCA mandate areas. The domains have been produced for eastem and central Africa, so the results are only comparable for a selection of countries. Each PRA in the mandate area is located in a particular domain, and each domain has a total population in eastem Africa. Therefore for each mandate area the scaling-out population is the total population of all the domains represented by the PRA sites. Results Assessing representativity of P RA sites The PRA si tes in Bas-Congo represent well the mandate area in all of the variables although they tend to be located in the lower and drier parts of the region. Differences between the weighted and un-weighted distributions are obvious for elevations above 800m and for annual rainfall totals above 1500mm. These sparsely-populated. higher and wetter areas in Madimba territoire tend to have poorer access to Kinshasa and to Mbanza~Ngungu. All except two of the PRA sites are within 8 hours of the capital Kinshasa. In Gitarama the single PRA site is broadly representative of the mandate area although a location a little higher and wetter would have been more typical. More importantly the PRA site is located on the granitic plateau that characterises the south-central part of the Mandate area and for which reason the mandate area was chosen. The focus of research in the Gitega mandate area is banana-based systems; as such, the higher elevation parts of the district were not sampled. The differences between the weighted and un-weighted histograms are small in this mandate are due to the lack of variability in the distribution of the (high) rural population. Also in Burundi, the PRA sites in Kirundo represent the mandate area well. The sites offer sufficient variation to capture differences in key biophysical and demographic variables, although the driest areas were not captured. In the Kibuye-Gisenyi mandate area the two PRA sites are in the lower and drier locations close to Lake Kivu; this is where the urban population is located and where market access is good. Nevertheless the PRA sites do not represent well the whole mandate area as a whole, this is perhaps because the emphasis in this mandate area is bananas, which are not well suited to the higher areas of the mandate area. In terms of geology the PRA sites are located on the granitic/volcanic-derived soils in the north and on schist-derived soils in the south, representing well the mandate area. In contrast the PRA sites in Kigali-Kibungo represent the mandate area well. Despite similar altitudes they are well spread in each of the other variables and are located on the granite-derived soils in Bugasera and on schist-derived soils in Kibungo, representing well the range ofthe underlying geology in the mandate area. The PRA sites in the Rusizi plain and Nord-Kivu montagneux mandate area represent well the mandate areas in all of the variables, although in the latter case they tend to be located in the lower and drier parts of 288 the region, which is not surprising given the mountainous terrain. The PRA sites in sud-kivu montagneux represent well the mandate area, especially when the westem portion of the area (which is included for administrative rather than biophysical reasons) is ignored. Finally in Umutara the four PRA sites represent well the granite-derived soíls in the north and the schist-derived soils in the west and south, representing well the mandate area. There is perhaps a gap in the coverage for the lower elevations, which are areas with longer growing seasons and higher precipitation. These areas are, however, almost mutually exclusive with the lower elevations in the east of the mandate area associated with lower rainfall and shorter growing seasons. We have shown that access to markets in Burundi and Rwanda appears to be more widespread than in DRC, due to the number and quality of the feeder roads in these two small countries. The quality of the digital data is also probably a factor with very good datasets available for both Burundi and Rwanda. Potential impact of CIALCA interventions We summed the total populations that inhabit those areas in Africa that have similar biophysical conditions to those encountered in each of the CIALCA mandate areas (Table 2). The population in this table could be considered the potential population that could be positively impacted by the interventions proposed in the three CIALCA projects. The results show that the Bas-Congo mandate area and the Gitega mandate areas ha ve homologue environments with large populations; in the latter case this in eludes almost half of Burundi as well as the higher density areas of eastem DRC. Despite the fact that there are eight PRA sites in the Sud-Kivu montagneux mandate area, the population in similar environments is quite small. Table 2. Population in homologue environments for each CIALCA mandate area. N ame Bas-Congo Rusizi Gisenyi Gitega Gitararna Kigali Kirundo Nord-Kivu Sud-Kivu U mutara Algeria o o o o o o o o o o Angola 1519540 o o 1723620 o o o o o o Burundi 668126 1153700 59911 4972120 1120200 915970 2104300 28645 249612 158455 Cameroon 35876 o o o o o o 256198 o o CAR 8432 o o o o o o 22128 o o Comoros 41583 o o o o o o o o o Congo 1750460 o o o o o o 32048 o o DRC 12983800 1244920 3021390 1331850 116444 4001 1276570 6928020 1642020 o Ethiopia o 731651 43134 5154090 680552 1481410 867383 295808 295540 274822 Gabon 150812 o o o o o o o o o Ghana 502816 o o o o o o o o o lvory 107498 o o o o o o o o o K en ya o 51924 421689 o 42866 43724 36865 2299330 o 55196 Madagascar o 5095 160 332220 435472 o o o 116420 1350000 o Malawi 2206 95508 o 262678 o o o 45987 o o Mozambique o 484495 o 293501 o 19979 13322 53368 o o Nigeria 5759550 o o o o o o o o o Zimbabwe o 98373 o 2678680 o o o o o o Rwanda o 2703 150 2904360 1226640 4583970 3571420 4687340 3377510 660407 2509720 South Afri o 63 1952 o 4387470 640283 1215580 1158240 o 43449 240925 Sudan o 121 o 10 o o o o o o Swaziland o 11 9135 o 595835 o 88446 87171 o o o Tanzania 1445380 866905 665303 1157340 1207100 1709000 1794390 453330 o 815034 Togo 55291 o o o o o o o o o Uganda o 213048 53 1940 3991 54832 156219 151654 3571960 7214 195280 Zambia o 17499 o 235754 o o o o o o Total 25,03 1,37013,507,541 7,979,947 24,459,05 1 8,446,247 9,205,749 12,177,235 17,480,752 4,248,242 4,249,432 289 An altemative strategy for gauging the potential population who might benefit from the technologies tested in the CIALCA projects is to analyze the development domains in which each PRA site is located, and extrapolate to the population of the same domains in the countries of eastem and southem Africa. The result ofthis extrapolation (Table 3) gives a far larger potential population than the results ofthe analysis of homologue environments. This is most evident in the Sud-Kivu Montagneux mandate area, wbich in this case has the largest potential scaling-out population. Table 3. Population in development domains ofthe PRA sites in each mandate area. N ame Bas-Congo Rusizi Gisenyi Gitega Gitarama Kigali Kirundo Nord-Kivu Sud-Kivu U mutara Burundi 3652910 2173050 5089160 4848160 1784100 4101446 152990 2284940 454968 Ethiopia 25817000 25075460 15487910 33077400 2479110 35077720 15487910 12494807 27782117 14869530 K en ya 11901490 10949010 16374860 15138510 6218460 16517975 16374860 4939037 12902587 2701355 Madagascar 8859280 12011990 3156820 2652740 1671280 10745486 3156820 5197738 14242758 8658940 Rwanda 2487108 4763287 7005370 3093455 4640750 3169464 7005370 129219 4892457 233870 Sudan 10715580 22918420 7645530 2323631 5874270 14342136 7645530 14965450 29784040 12025820 Tanzania 15864820 14127053 7194633 9543220 891803 21259736 7194633 10519240 20973033 14123440 Uganda 14017360 5889020 13207940 15767220 2300640 15713171 13207940 1928310 7339010 4296850 Zaire 31592510 37912270 12607700 11946190 6151890 37683040 12607700 12058610 43347200 29336850 Eritrea 747565 2779658 1257985 269180 989499 986556 1257985 2378141 3846719 707079 Total 125,655,623 138,599,218 89,027,908 98,659,706 33,001,802 159,596,730 83,938,748 64,763,542 167,394,861 87,408,702 Conclusions and Discussion The case of Umutara highlights the difficulty of assuring representativity in multivariate space, especially when limited to at most eight sites across areas of up to 20,000 km2• One solution is to reduce the dimensions ofthis multivariate space by classifying and merging the variables to create domains that reflect the objectives or research questions tackled by the project, program or organization. The CIALCA mandate areas ha ve been assessed in terrns of the development domains, which are based on access to markets, agro ecological potential and populatíon densíty. The markets used in the creation ofthese domains are specific to the particular mandate areas. This makes it difficult to use the same development domains for scaling-out purposes anda more general set of domains, such as those used by ASARECA was necessary. The analysis of spatially-variable characteristics has been used in combination with the participatory rural assessments which were carried out in each mandate area. A críticism of the methodology used by CIALCA could be that the characterisation should precede the choice of sites for the PRAs. There are, however, practica! reasons why this approach was chosen, not least because the characterization consumes time, which is itself a limited resource in research projects like CIALCA. Nevertheless the characterization process has confirmed the soundness of the sites chosen for PRAs and ultimately for action sites and interventions. The assessment of areas for the scaling-out of interventions might normally be confined to an ex ante economic analysis. We feel that a combination of economic impact assessment and spatial analysis can deliver a realistic range of impacts. Both the strategies of assessing the potential scaling-out populations have flaws. The use of homologue environments ignores the socio-economic conditions that ensure the success of the intervention, whíle the development domains are too broad for anything other than policy recommendations. Neither of these considers tbe precise mechanisms for scaling-out and neither considers the effects of distance, for instance sorne homologue environrnents occur in southem Africa). The scaling- out assessments undertaken in this study can also be improved by the consideration of the welfare or nutritional levels of tbe potential beneficiaries (see, for example, the section Strategic approaches to 290 targeting technology genera/ion: Assessing the coincidence of poverty and drought-prone crop production by Hyman et al. in this Report). A response to this evaluation would be the further development ofthe CaNaSTA (Crop Niche Selection in Tropical Agriculture) too! (O'Brien, 2004) for a range of technologies allied with research on socio- ecological niches (Ojiem 2006) and impact pathways (Douthwaite et al. 2003). 291 References ASARECA (2005). Fighting poverty, reducing hunger and enhancing resources through regional collective action in agricultura) research for development. ASARECA (Association for Strengtheni.ng Agricultural Researcb in Eastern and Central Africa) Strategic Plan 2005-2015, August 2005, Entebbe, Uganda. 94 pp. CIA T (2004). Homologue: A Computer System for ldentifying Similar Environments throughout the Tropical World. Software available at: http://gisweb.ciat.cgiar.org/homologue/ Cook, S.E. and Fujisaka, S. (2004). Spatial Dimension of Scaling Up and Out. In: Pachio, D. and Fujisaka, S. (eds). 2004. Scaling up and out: achieving widespread impact through agricultura) research. CIAT Publication No. 340. Cali, Colombja, pp 53-63. Douthwaite, B., Kuby, T., van de Fliert, E. and Schulz, S. (2003. lmpact pathway evaluation:an approacb for achieving and attributing impact in complex systems. Agricultura/ Systems 78:243 - 265. Farrow, A. and Nelson, A. (2001). Accessibility Modelling in ArcView 3: An extension for computing travel time and market catchment information. Software manual, CIA T, Cali, Colombia. A vailable at: www.ciat.cgiar.org/access/pdf/ciat access.pdf GRUMP (2005). Global Urban-Rural Mapping Project (GRUMP). Dataset available at http://beta.sedac.ciesin.columbia.edu/gpw/# Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., and Jarvis, A. (2005). Very high resolution interpolated climate surfaces for globalland areas. International Journa/ ofCiimatology 25:1965-1978 O'Brien, R. (2004). Spatial decision support for selecting tropical crops and forages in uncertain environments. PhD thesis. Curtin University of Technology, Perth, WA, AU. (Published). Available at: http://adt.curtin.edu.au/theses/avai lable/adt-WCU20050526.144234/ Ojiem. J.O. (2006). Exploring socio-ecological niches for legumes in western Kenya smallholder farming systems. PhD thesis. Wageningen Universiteit en Researchcentrum (WUR), Netherlands. (Published). Available at: http :/ /librarv. wur .n J/wdalabstracts/ab4063 .htm 1 Pender, J. , Jagger, P., Nkonya, E. and Sserunkuma, D. (2004). Development pathways and land management in Uganda. World Development 32:767-792. Selvaradjou, S-K., Montanella, L., Spaargaren, 0., and Dent, D. (2005). European Digital Archive of Soil Maps (EuDASM): Soil Maps of Africa. EUR 2 1657 EN. Office of Official Publications of the European Communities, Luxembourg. 386pp. Thomton, P.K., Jones, P.O., Owiyo, T., Kruska, R.L., Herrero, M., Kristjanson, P., Notenbaert, A., Bekele, N. and Omolo, A., with contributions from Orindi, V. , Otiende, B., Ochieng, A., Bbadwal, S., Anantram, K., Nair, S., Kumar, V. and Kulkar, U. (2006). Mapping Clima/e Vulnerabílity and Poverty in Africa. Report to the Department for Intemational Development, ILRI, Nairobi, Kenya. 17 1 pp. 292 THEME5: OBTAINING, DEVELOPING, AND MANAGING DATA AND INFORMATION 293 An evaluation of void-filling interpolation methods for SRTM data H.I.Reutef, A.Nelson° and A.Jarvish a Institute for Environment and Sustainability, Ispra, Ita/y. b lnternational Center for Tropical Agriculture (CIAT) and Jnternational Plan/ Genetic Resources Institute (IPGRI), Cali, Colombia. Abstract The Digital Elevation Model that has been derived from the February 2000 Shuttle Radar Topography Miss ion (SRTM) has been one of the most important publicly-available new spatial datasets in recent years. However, the 'fmished' grade version of the data (also referred to as Version 2) still contains data voids (sorne 836,000 km2) - and other anomalies- that prevent immediate use in many applications. These voids can be filled using a range of interpolation algorithms in conjunction with other sources of elevation data, but there is little guidance on the most appropriate void filling method. This paper describes: (i) a method to fill voids using a variety of interpolators, (ii) a method to determine the most appropriate void-filling algorithms using a classification of the voids based on their size and a typology of their surrounding terrain; and (iii) the classification ofthe most appropriate algorithm for each ofthe 3,339,913 voids in the SRTM data. Based on a sample of 1,304 artificial but realistic voids across six terrain types and eight void-size classes, we found that the choice of void-filling algorithm is dependent on both the size and terrain type of the void. Contrary to sorne previous fmdings, the best methods can be generaJised as: Kriging or Inverse Distance Weighting interpolation for small- and medium-size voíds in relatively flat low-lying areas; Spline interpolation for small and medium sized voids in high altitude and dissected terrain; Triangular Irregular Network or lnverse Distance Weighting interpolation for large voids in very flat areas, and an advanced Spline Method (ANUDEM) for large voids in other terrains. This paper has been accepted for publication in the International Journal of Geographic Informa/ion Science. Readers interested to receive a reprint as a pdf file should contact the senior author by e-mail a.jarvis@cgiar.org. Key words: DEM, interpolation methods, void filling, DEM fusion Introduction A digital elevation model (DEM, or more correctly a land surface model - LSM) is one of the most useful sources of information for spatial modelling and monitoring, with applications as diverse as: environment and earth science, e.g. catchment dynamics and the prediction of soil properties; engineering, e.g. highway construction and wind turbine location optimisation; military, e.g. land surface visualisation, and entertainment, e.g. landscape simulation in computer games (Hengl and Evans 2007). The extraction of land surface parameters - whether they are based on ' bare earth ' models such as DEMs derived from contour lines and spot heights, or 'surface cover' models derived from remote sensing sources that include tree top canopies and buildings for example - is becoming more common and more attractive dueto the increasing availability of high quality and high resolution DEM data (Gamache 2004). One of the most widely used DEM data sources is the elevation infonnation provided by the shuttle radar topography miss ion (SRTM) (Coltefli et al. 1996, Farr and Kobrick 2000, Gamache 2004), but as with most other DEM sources, the SRTM data requires signillcant levels of pre-processing to ensure that there are no spurious artefacts in the data that would cause problems in later analysis such as pits, spikes and patches of no data (Dowding et al. 2004, Gamache 2004, Chaplot et al. 2006, Fisher and Tate 2006). In the case ofthe SRTM data, these patches of no data are pervasive (USGS 2006b) and must be filled or interpolated, preferably 294 with auxiliary sources of DEM data. This paper describes a procedure to determine the most appropriate interpolation methods (with and without auxiliary DEM data) for no data patches of different sizes and in different terrain types. The rationale for this paper stems from a statement by Fisher and Tate (2006) that no single interpolation method exists for the most accurate interpolation of terrain data. Such a procedure is necessary for developing a high quality global DEM derived from the SRTM data where all no data areas have been filled using the best performing interpolation algorithm available. The shuttle radar topography mission (SRTM) The 11 day SRTM flew in February 2000, and has provided publicly-available e1evation surface data for approximately 80% of the world's land surface area (from 60°N to 56°S), wíth a post spacing of 1 are second (often quoted as 30 metres resolution) in the USA, anda degraded 3 are second (often quoted as 90 metres resolution) product for the rest of the world. lt is a snapshoot of the reflective surface of the earth during the time period of the mission, and is about 100 times more detailed than other ex.isting freely- available global e1evation data, such as GTOP030 (USGS 1996) and GLOBE (Hastings and Dunbar 1998). The SRTM elevation data is derived from X-band and C-band interferometric synthetic aperture radar (InSAR) (Wemer 2001, USGS 2006b). This paper deals with the better-known and widely-available C- band product, which we shall refer to as SRTM elevation data. As with all DEMs derived from remote sensing sources, the SRTM elevation data include trees, buildings and other objects on the earth surface and therefore the dataset is a surface elevation model (Rodríguez et al. 2005, 2006). Severa! products have been derived from the SRTM data. Firstly, the raw data were processed by a suite of programs at JPL (Farr and Kobrick, 2000), and was made available primarily for research purposes. Thís was termed "unfinished" data. Further processing generated DEMs in full DTED compliance leve!, and these were termed "finished" data (Slater et al. 2006). For both datasets: elevations outside the USA are degraded either by (i) averaging or (ii) by thinning (i.e. taking one sample out of the nine available posts). The horizontal datum of the SRTM data is WGS84, whilst the vertical datum is EGM96 which has implications for certain applications. The C-band product has significant areas of missing data due to the nature of radar data and the interferometric process used to create the DEM (Figure 1 ). The reasons for the missing data are geometric artefacts, specular reflection of water, phase unwrapping artefacts and voids due to complex dielectric constant (see Kervyn 2001 for further information). For example, the InSAR instrument used to generate the SRTM elevation data had an incidence angle of between 30° and 60°, making it difficult to generate images for terrain slopes corresponding to that range of angles (Gamache 2004, Eineder 2005). For the purpose ofthis paper, we defme any areas ofmissing data that ex.ist in the SRTM data as voids. The number of remaining voids with different sizes in the SRTM data are a considerable problem for many uses and applications, including hydrological modelling, terrain índices, land surface characterisation, digital soil mapping and many other geomorphometric models, and thus these voids need to be filled to create a seamless DEM (MacMillan et al. 2000). The "finished" version of the SRTM data (described more fully in below) provided by United States Geological Survey (USGS) and National Aeronautics and Space Administration (NASA) still contains 3,399,913 voids accounting for 803,166 km2 (an area comparable to Pakistan or somewhat larger than Texas), and in extreme cases, such as Nepal, they constitute 9.6% of the country area with sorne 32,688 voids totalling an area of 13,740 km2. Figure 1 shows the proportion of each 1 x 1 degree SRTM tile that is composed of void areas. Figure 2 shows two extreme examples of regions where there are many voids, Libya (upper) and Nepal (lower). Of the 210 countries covered by the SRTM data, two countries have void areas larger than 10% of their country size, nine countries more than 5% and 14 more than 2%. In total, 44 countries have 1% or more of 295 their area covered by voids. The void size 1 fi-equency distribution of all voids in the SRTM dataset is shown in Figure 3. 1 - -- - 0 0-S S - 10 10 - 25 >25 Figure l. The global distribution of voids in the SRTM data, represented by the proportion of void area in each 1 x 1 degree SRTM tile. Note the clustering ofvoids over mountainous and desert areas and the northem extent ofthe SRTM data (60°N). 296 o ' ' . 1000 2000 3000 4000 .sooo 6000 7000 8000 ¡· , ' 1 J 1 t - ·k . , . ' • ' . Figure 2. Voids (in black) overlaid on the SRTM30 elevation data for two extreme cases, Libya (upper) and Nepal (lower). The 1 x 1 degree SRTM tite boundaries have also been superimposed to express scale. The key shows elevation in meters above mean sea leve!. 297 .... : .. . .. 10 100 1000 10000 100000 1000000 10000000 Figure 3. Log log plot ofthe void size against the frequency distribution for the global void dataset (n= 3,339,913). Voids occur for different reasons in different terrain types. A void dueto shadowing will more likely occur in mountainous areas, whereas a void due to complex dielectric constant is more likely to occur in desert areas like the Sahara. Void frequency with respect to elevation has been demonstrated to have a bimodal distribution with peaks of the distribution occurring in flat areas and in steeply sloping areas (Gamache 2004, Falomi el al. 2005). This distribution is clearly seen in Figure l. Since this study focuses on methods of void filling the SRTM elevation data, we will not discuss the accuracy or the errors in the SR TM data, though it is worth mentioning that the sensor error is stated to be +/- 16m (USGS, 2006b). Further details on SRTM accuracy are available in the literature (Toutin 2002, Rabus et al. 2003, Gamache 2004, Falomi el al. 2005). Fisher and Tate (2006) provide a thorough review of the causes and consequences of error in DEMs. They classify errors into three different groups: (i) gross errors (e.g. system malfunctions), (ii) systematic errors, which might be described by a functional relationship (Thapa and Bossler 1992, p836 in Fisher and Tate 2006), and (iii) random errors with or without spatial dependence that arise for different reasons. Voids are one type of systematic error, which can be overcome with specific algorithms. Voidfilling methods Interpolation methods are widely used in the generation of DEMs. However, void-filling (VF) methods contain a special subset of interpolation algorithms with certain restrictions, and also other methods such as the fill and feather approach (Dowding et al. 2004). All interpolation algorithms for VF DEMs use the elevation data surrounding the void in the interpolation process. If auxiliary sources of elevation (for 298 example, ASTER DEMs, GTOP030, digitised topographic maps and land survey measurements) are available, then sorne of these algorithms can incorporate this information to improve the accuracy of the interpolation. However, there are often severe differences between the DEM and the auxiliary data that need to be addressed before the VF algorithms can use auxiliary data. These differences can occur in: (i) the spatial resolution, (ii) the vertical datum, (iii) horizontal and vertical shifts, (iv) first or second order trends, (v) production errors, (vi) the type of surface model (SRTM is a surface model, whereas a DEM based on topographic data is abare earth model) and, (vii) the spatiaJ distribution of errors (see for example Hutchinson 1989, Kaab 2005, Fisher and Tate 2006). VF algorithrns can be categorised into surface (Katzil and Doytsher 2000), volumetric (Vedera et al. 2003) or example based methods (Sharf et al. 2004). A\ternatively, Katzil and Doytsher (2000) divide the algorithms into polynomials (such as linear estimation, lD and 20 polynomials of the third order, cubic splines, or iterative spline algorithms) and non-polynomial approaches (such as kriging, inverse distance weighting, fill and feather approaches or moving average). Several authors have evaluated the quality of different algorithms to fill in voids for radar data as well as other DEM sources. Katzil and Doytsher (2000) tested linear estimation, kriging and cubic spline for elevation, but the evaluation was not performed on real voids. Instead a method called cross-validation was applied (removing one point and then comparing the elevation of the generated surface against the elevation of the point), which showed no significant differences between methods. Dowding et al. (2004) used a fill and feather (FF) approach (i.e. they used an auxiliary elevation dataset to patch the void area, and then smoothed the transition zone between both datasets) to incorporate auxiliary information into a VF algorithm. They selected seven voids with sizes ranging from 36 to 2,541 pixels and then compared the results both visually and against a reference DEM and ground control points. The resu\ts showed differences between O m and 22 m for the area of seven different voids against a reference DEM. Kuuskivi et al. (2005) used seven real world voids across different terrain types to evaluate the performance of a commercial fiJI and feather algorithm that used high quality auxiliary DEM data against three freeware programmes: 3DEM (Visualization Software LLC, 2004); VTBuilder (Virtual Terrain Project, 2004); and BLACKART (TerrainMap.com, 2004). The study clearly demonstrated the large differences in results that can occur when using different VF algorithms and the potential improvements that can be achieved with good quality auxiliary information. Grohman et al. (2006) presented a geostatistical algorithm (inverse distance weighting - IDW) together with a linear adjustment of the elevation height called Delta Surface Fill (DSF) and compared it against a FF approach using five artificially created voids in void-prone terrain types from the SRTM data. The authors concluded that the performance of the DSF algorithm produced better results based on visual interpretation and reduced the standard deviation ofthe error surface. Severa! other studies have presented algorithms that are capable of filling void areas, but did not provide statistical results. For example Hofer et al. (2006) tested an advanced cubic spline method which keeps certain error bounds on nine voids and evaluated the VF results graphically. Almansa et al. (2006) evaJuated four different artificial voids at different locations by comparing three different methods. Another example is Kaab (2005) whereby a simple replacement of SRTM by ASTER data was made without any further evaluation. There are severa! observations that can be drawn from these studies. Firstly, each study typically compares three or four algorithms at most, whereas a GIS may contain many more suitable algorithms (e.g. IDW, Kriging, ANUDEM, Spline, and Trend Estimation). Secondly, sorne of the studies contain aJgorithms that are not easily reproducible within a GIS or image anaJysis system because their description is too vague or dueto commercial interests. We argue that if an improved global DEM is to be produced from the SRTM data, then the VF algorithms should be accessible and repeatable. Thirdly, the studies presented results based on a handful ofvoids which may not be representative ofthe 3,339,913 voids in the SRTM data, and are not sufficient for robust statistical analysis. A much larger sample of real world voids is required before we can suggest one algorithm over another with any degree of confidence. Fourthly, occasionally these 299 voids were artificial and hence may not be representative of real world voids. We realise that unless a high quality auxiliary DEM is available, then it is impossible to assess the veracity of the results from VF as there are no ground truth data. However, it is possible to create artificial voids that are representative of real voids in terms of size, shape and terrain location. Fifthly, terrain can have a large influence on VF results. Katzil and Doytsher (2000) acknowledged that terrain has an influence on the VF process; however, they could not suggest a recommended method across three terrain types (mountains, hilly and planar). Grohman et al. (2006) recognised the relief type for their five voids, which lead to decreasing average standard deviation from rugged to moderate to flat terrain. Again, the sample of voids should be sufficiently large across a range of terrain types in order to assess the influence of terrain ruggedness on algorithm choice and performance. Finally, voíd size is critica!. It ís much harder to fill a large voíd accurately than a small one. Grohman et al. (2006) recognised the importance of void size, but did not pro vide further insight. Again, the sample of voids should in elude sufficient numbers of voids of sizes that are representative of voids found in each terrain type. In conclusion, there has been no thorough evaluation ofthe many (GIS ready) VF algorithms for DEM data using a sufficient number of voids of varying size across different terrain types in order to determine the most appropriate VF method(s). Research objectives The objectives of this study are to: (i) describe void characteristics in terms of size and terrain unit; (íi) determine whích VF algorithm performs best without any auxiliary information on an exhaustive dataset, with respect to terrain unit and voíd size; (iii) determine if low-grade auxiliary information can improve the VF algorithm performance; (iv) determine which VF algorithm performs best with respect to terrain unit and void size using auxiliary informatíon, and; (v) provide a global void dataset stating the best VF algorithm that should be used based on the results from (ii) and (iv). Study area and data As shown in Figure 1, data voids occur in all regions of the SRTM data, but after consideration of the spatial distribution of voids and terrain units and computational limitations, we limited our sample of voids to Africa, an area of approximately 29,800,000 km2 and containing 1,168,136 voids. This provided a sufficient number of voids across a wide range of sizes and terrain units from which to perform the sampling. SRTM data preparation SRTM data are available in different formats from different distributors (a testament to its usefulness and popularity), and here we use the "finished" 3-arc second averaged SRTM data set that is available from the USGS EROS data server (USGS 2006a). The pre-processing and editing ofthis data is described by USGS (2006b), but the essential details are that spikes and pits in the data with surrounding elevation differences greater than 100 m were removed, voids smaller then 16 pixels were filled with a nearest neighbour interpolation while larger voids were left as were, and water bodies and coastlines were depicted as described by USGS (2006c). The data are available in one x one degree tiles in 16 bit integer BIL format. 3,250 tites were downloaded from ftp://e0sro01 u.ecs.nasa.gov/srtm/version2/SRTM3/Africa and were converted to ESRI™ grid format and mosaicked together in ArcGIS 9.1c to create a continental DEM for Africa with extents 39<>N-35°S and 30°W-60°E. Since these data have been edited for small voids, coastlines and water bodies, we assume than any remaining land area that contains no elevation information is a void. For each ofthe 1,168,136 voids we stratified the voids based on the naturallogarithm ofthe number ofvoid pixels and grouped them into eight size classes with the following number of pixels (numbers in parentheses are mínimum, average and maximum) per class; [A] (1 ,10,25), [B] (26,50,80), 300 [C] (81,120,140), [D] (141,240,400) [E] (401,600,800) [F] (801,1100,2500), [G] (2501,4000,8000) and [H] (8001,10000,1267052). Auxiliary DEM data As stated in above, sorne VF methods incorporate auxiliary DEM information in arder to improve the accuracy ofthe results. In this study we used the SRTM30 (Gamache 2004, USGS 2006d) and GTOP030 DEMs (see the relevant sections on the SRTM and on the STRM data preparation above for more details on these two datasets), which were stored in ESRI™ grid format, as these are the only DEM datasets available for all of Africa. Terrain typology The terrain typology is based on the SRTM30 data. A half-degree resolution, 15-class terrain typology was derived from this DEM based on a combination of the average SRTM30 elevation within each half degree pixel and the relief roughness of the SRTM30 data within the pixel, defined as the range of SRTM30 elevation values in the pixel divided by half the pixel length connecting the centre of each grid pixel (Meybeck et al. 2001). For simp1ification, we aggregated these 15 classes into six majar terrain units (TU), which have similar land surface characteristics. The relief classes (1) plains, (2) mid-altitude plains and (3) high-altitude plains were grouped into PLAINS; (4) lowlands, (6) platfonns, (7) low plateaux, and (8) mid- altitude plateaux were grouped into PLATEAUX; (9) high plateaux and (1 O) very high plateaux were grouped into HIGH PLATEAUX; (5) rugged lowlands and (11) hills were grouped into HILLS, (12) low mountains, (13) mid-altitude mountains were grouped into MOUNTAINS, and; (14) high mountains, and (15) very high mountains were grouped into IDGH MOUNTAINS. Figure 4 shows the original 15 terrain classes for Africa and their grouping into six majar terrain units. 301 The sizes of TU are quite different across Arrica. Void size is also significantly different between TU. Tberefore we used that as additional stratification factor, which allowed us to assess differences between TU (Figure 5). The highest percentage of void area per total TU area is reached in PLAINS, which can be attributed to the dunes in desert areas, followed by voids in IDGH MOUNTAINS. The PLAINS cover between 30~ 50% of all large void areas (void-size groups G and H), with a percentage around 20 % for all other size classes. HlGH MOUNTAINS on the other hand covers shows an increase rrom 20% to 30% over all size classes (except theH size class). A decreasing percentage of voids can be observed for PLATEAUX, MOUNTAINS and Hll.LS with increasing void size, whereas the HIGH PLATEAUX show a strong increase in percentage ofvoids with increasing size class (Figure 5). 1 HIGH Pl.Al'EAUS t NGH MOUNTMCS lnlgh plaiiiU l'lgl\ ll!iOIÑiiN "---'WIY hlgll,..._. '- -..r, ~ fi'IOUIIliiN Figure 4. A 15 class terrain typology for Africa, based on the methodology proposed by Meybeck et al. (2000). 302 101M ~ ~ ~ j ea. ¡ j ~ ]; ' o ..,. - ~ ~ 115 "" A 8 e o E F G H s;;;z.c&asses Figure 5. Percentage ofvoids per terrain unit and void size class for the global dataset. Void selection The voids were sampled using the following procedure. We first identified the terrain unit of each of the 1, 168, 136 voids. For the few occasions where a void !ay in the boundary between two terrain units, it was assigned to the terrain unit in which the greater area of the void !ay. No TU was assigned in the very few cases where a void was evenly distributed over two or more terrain units. In the cases where the half degree resolution TU map did not extend over coastal voids, the closest TU p ixel (almost always PLAINS) was assigned to the void. Secondly, the previously discussed size class was assigned to each void. The distribution of all the voids by terrain unit and s ize class is shown in Table l. The distribution by terrain units agrees with the fi ndings of Falorni el al. (2005), but we have found no corresponding study for the distribution ofvoid sizes and void size by TU. Table l. Number of global voids by terrain unit and void size (A-H). A B e D E F G H Su m PLAINS 458,851 27,968 7.414 3,746 1,810 1,453 552 437 502,231 PLATEAUX 765,745 54,739 13.700 5.425 1,831 965 223 67 842,695 H. PLATEAUX 24,056 1.082 278 171 107 128 91 85 25,998 HILLS 195,055 13,097 3,286 1,157 339 157 21 5 213,117 MOUNTAINS 789,587 58,258 15,673 6,319 2,125 1,108 228 51 873 ,349 H. MOUNTAINS 680,952 53,497 15,825 7,235 2,869 1,888 523 154 762,943 Su m 2,914,246 208,641 56,176 24,053 9,081 5,699 1,638 799 3,220,333* *The total number ofvoids in this table does not sum to 3,399,913 dueto sorne voids not being assigned terrain units. The vast majority ofthese unassigned voids are in coastal areas and should be classed as PLAINS. 303 Within each terrain unit we randomly selected 15 voids based on their size distribution (i .e. we selected more small voids than large ones in a given terrain unit if small voids occurred more frequently). We then duplicated each of these voids a number of times, again in relation to their size distribution, and manually relocated them to neighbouring locations within the same terrain unit, ensuring that the void was relocated to a similar landscape and that it did not overlap with existing voids. For example, a small void (size class A) in TU PLAIN would be duplicated 15 times and moved to 15 neighbouring locations in TU PLAIN to create 15 voids for analysis. Duplication of real void areas en sures that the shape, size and orientation of the artificial voids are representa ti ve of real voids. However, it does risk the creation of artificial voids in areas of the terrain that are not representa ti ve of the terrain of the real void. To ensure that we relocated the voids to comparable terrain, we employed (i) visual inspection ofthe terrain and (ii) computed the mean and standard deviation ofthe elevation in a buffer zone surrounding the real void and compared it the mean and standard deviation of the elevation in buffer zones surrounding the relocated void. No significant changes could be observed in the standard deviation of the surrounding elevation between the groups of relocated and the real voids across the original 15 terrain types (based on a t test), except for rugged lowlands and low plateaux. No differences were observed when the results were aggregated into the six TU (results not shown). Therefore we assume that the visual relocation was an acceptable approach for generating realistic artificial voids. In this way we created 1,304 artificial but realistic voids based on real void characteristics distributed across six terrain units and eight size classes (Table 2). These voids were stored as polygon coverages in Arclnfo. Table 2. Number of artificial voids by terrain unit and void size (A-H). A B e o E F G H Su m PLAINS 45 45 60 60 45 15 15 15 300 PLATEAUX 65 60 75 &> 60 20 28 12 400 H.PLATEAUX 30 40 40 40 35 10 5 o 200 HILLS 15 20 25 20 20 o o o 100 MOUNTAINS 30 40 39 40 30 lO 14 6 209 H. MOUNTAINS 15 15 20 30 5 5 5 o 95 Su m 200 220 259 270 195 60 67 33 1,304 Data pre-processingfor the artificial voids A working area for each single void was created by enlarging the maximum extent of the void by 100 pixels in all directions and extracting the underlying DEM data within this buffer zone. The number of pixels for this buffer was based on empírica! testing of the algorithm under different TU (resu1ts not shown). We then 'punched out' the void from this buffered area and used the remaining ' ring' of DEM data to create (i) elevation spot heights or points, with one point for each DEM pixel, and, (ii) contours at 1 O m intervals. There were occasions where it was not possible to extract contours from the buffered region, for example where the terrain was extremely flat. ln these cases we employed an iterative process to decrease the contour interval until a contour !ayer could be created (the lower lirnit was 1 m). Where this process did not result in a contour !ayer, we extended the buffer up to a maximum threshold of 0.1 degrees. These contours were stored as line or point coverages in Arclnfo. Auxiliary spot height elevation data was extracted from the SRTM30 and GTOP030 at 30 are second spacing unless either one of two restrictions were met. The fLrSt restriction is the size of the void (coarse auxiliary data will not help in the interpolation of small voids); and the second considered the shape of the void (e.g. a long, thin void is better filled with on ly the surrounding DEM data). 304 SRTM data have a high absolute accuracy in contrast to GTOP030 and to account for such differences we adjusted the elevation values in the GTOP030 DEM as follows. For each void, the auxiliary DEM was re- sampled to the resolution of the SRTM dataset and the void area was punched out from this re-sampled auxiliary dataset. The difference in elevation between both datasets was used to raise or lower the elevation values for the original resolution auxiliary dataset. We thereby accounted for sorne differences between datasets (i, ii, vii) as outlined in the section on the SRTM above. We did not make any adjustments for other errors ( e.g. geometrical location or trends in the auxiliary data). The spot heights were stored as point coverages in Arclnfo. The SRTM elevation from the void stamped area constitutes the 'truth' !ayer against which the results from the VF aJgorithms were compared. Essentially, this is equivalent to withholding pixels from the interpolation process and then comparing the results from the VF algorithms against them afterwards (see al so Y ang and Hodler 2000). Thus, for each of the 1,304 voids we created a truth DEM, a set of contours and spot heights buffering an artificial void area, and where applicable, two auxiliary spot height datasets. Void fllling methods The procedure for applying and evaluating the VF algorithms is outlined in Figure 6 and described in detail in the following sections. Unless otherwise stated, all processing was carried out in Arclnfo Workstation 9.1 using Are macro language (AML) routines and standard Arclnfo interpolation functions from the Are and Grid environments. The results of the VF methods were projected from geographic projection (latitude/longitude) into MoUweide Equal Area projection where the longitude ofthe projection centre was the longitude ofthe centroid ofthe void, and both vertical and horizontal units were in metres. 305 DEM + CREA TE VOID AREA ¡ BUFFER VOID WITH THRESHOLD / generate contour lines/points from DEM L VF - algorithm L. KRIGING, JDW L. ANUDEM L. SPLINE, TRENO L. FILL and FEATHER L. MOVJNG WJNOOW l Export to / test for void size and void shape for use of aux. data resample aux. data to OEM resolution (mask void} + compare + adjust helght + generate point data merge into original DEM single VF DEM-patches Figure 6. Flow diagram ofthe VF algorithm assessment methodology. Voidfilling methods The following eight VF algorithms were implemented: i) Kriging (KR), ii) Spline (SP), iii) Trend (TR), iv) lnverse Distance Weighting (TDW), v) Moving Window Average (MW), vi) Fill and Feather (FF), vii) Triangulated Irregular Nehvork (TIN) and viii) ANUDEM (ANU). The geostatistical methods (KR, SP, TR and TDW) require severa! input parameters, but we used default values wherever possible, since it was too complex and too time consuming to adjust the parameters for each method and each void. One might criticise this approach since adjusting the parameters could improve the interpolations on a case by case basis. However since the VF algorithms are all well-known and long- standing implementations, we assume that the default values provide reasonable results under most conditions. 306 The implementation for KR is based on McBratney and Webster (1986), using a spherical semi-variogram model with an automatically fitted function. We used a tenth-order linear polynomial regression (based on manual testing, results not shown) for TR. The SP method follows Franke ( 1982) and Mitas and Mitasova ( 1988) and performs a two-dimensional minimum-curvature spline interpolation resulting in a smooth surface that passes exactly through the input points. In this case we used regularised splines, which yield a smooth surface. We did not test tension splines as in Mitasova and Hofierka ( 1993), even though it is available asan option in Arclnfo. IDW was implemented following Watson and Philip (1985) based on the 12-nearest neighbouring points. MW interpolates the void area by computing elevation values in the void pixels next to the void boundary based on the local average of the neighbouring elevation pixels. This process continues inwards until a ll void pixels are filled. For FF, we re-sampled the auxiliary information to the resolution of the truth DEM, buffered it inwards, filled in tbe void using this re-sampled auxiliary DEM, and closed any remaining void pixels applying the MW method with a 3 x 3 pixel window. More advanced approaches are possible, which alter the surface of the original DEM (Dowding el al. 2004 ), however these were not implemented he re. MW and FF are complementary in that MW is used where there is no auxiliary DEM, and FF is used where there is. For TIN, which is by definition a triangular network of mass points with 3D-coordinates connected by edges to form a triangular tessellation, the weed tolerance (the minimal tolerance between two data points ata line) was set to 0.0001 of the maximum extent of the input data set, whereas the proximal tolerance (minimal distance between single data points) was set to the machine precision ofthe host computer. The ANU approach (Hutchinson 1989, Hutchinson and Dowling 1991) is implemented in Arclnfo as TOPOGRJD, and creates a hydrologically correct DEM using a multi-resolution iterative finite difference interpolation (extended spline) method, which ensures that ridges are maintained, streams are enforced and spurious sinks are removed. The ANU approach contains three parameters that are used for the smoothing of the input data and the removing of sinks (ESRJ 2000). The TOLI tolerance was initially set to 5 m (half the height difference between contours); however, if the maximum elevation difference observed in the data preparation dataset was below the contour interval, it was set to half that value. Values for the horizontal and vertical standard errors were set to one and zero respectively. The authors limited the number of implementations to the above described algorithms, though even more advanced algorithms (Soile 1991, Hofer el al. 2006) look advantageous. lf needed, more advanced algorithm e.g. conditioned simulations (Holmes el al. 2000) could be implemented in the processing chain. Evaluation For each void we applied the seven VF algorithms three times, once with no auxiliary data, and once each with the SRTM30 and GTOP030 information, resulting in 21 DEMs, plus the reference DEM. A visual example ofthe results from each VF algorithm for a set of voids is shown in Figure 7. Figure 7a shows the void to be filled, whereas the other examples (b to i) present the different VF results. The effect of coarse resolution auxiliary data is clearly visible in Figure 7i with the Fill and Feather approach. In Figure 7h the extrapolation by moving window shows sorne limitation as extrapolation from the borders of the void occurs. The geostatistical algorithms in Figures 7d, e and f show only slightly different visual appearances, e.g. the representation of the peak on the left side of the large voids. Finally, the TIN and ANUDEM approaches show slightly different elevation surfaces, with TIN creating a ridge in the major void (Figure 7c), which is not visible in the hydrologically correct ANUDEM surface (Figure 7b). 307 Figure 7. Examples ofthe VF aJgorithms applied toa test area (a) in the SRTM DEM. The methods are (b) TOPOGRJD, (e) TTN, (d) IDW, (e) Spline, (f) Kriging, (g) Trend, (h) Moving Average, and (i) Fill and Feather. We compared the elevation of the reference DEM (ZRer) to the 21 DEM (zoEM) by computing the root mean square error (RMSE), Pearson's correlation coefficient (p), the average difference (J..l), the sum difference (y) and the standard deviation of the difference between both surfaces (a). Additionally, we computed the area for each void. We chose an evaluation based on the total area ofthe void similar to Grohman et al. (2006) in contrast toa selected number ofGCP (Dowding et al. 2004). The RSME was computed between the reference elevation and the 21 elevation surfaces. Each void-filled DEM patch was ranked from 1 (lowest RMSE) to 7 - or 21 for a comparison across all variations - (highest RMSE). The distribution of the ranking relative to terrain and void size was assessed by summarising the ranking results by: (i) terrain unit, (ii) size class, (iii) terrain unit and size class and (iv) auxiliary datasets. Fisher and Tate (2006) argue that the RMSE is not necessarily a good estimator ofthe error, recommending the mean error (ME) and the error standard deviation (S), where n is the number of pixels in the void: 308 We perfonned the comparative analysis based on RMSE, ME and S and observed consistent best VF results for different TU/void size classes for RSME and S. ME showed diverse rankings of different algorithms. For the remaining course ofthis paper we provide results for RSME only, and discuss the S and ME results where appropriate. We are aware that a global statistic was used to compare the filled voids against the truth surface, rather than evaluation methods that take the spatial pattem of errors into account or which identify the different factors which led to that error. Results Evaluation of the void filling algorithms with respect to terrain unit Table 3 shows the statistical summary of the ranking results when the void is classified by its terrain unit, irrespective ofthe size ofthe void. The Table 3 shows the mean and standard deviation (in parentheses) of the ranking of each VF method for each terrain unit, with and without an auxiliary DEM. The geostatistical method KR and the mechanical method SP (e.g. no assessment ofthe uncertainty ofthe model is possible) are consistentJy the "best" methods, with KR perfonning better in tlatter areas (PLAINS, PLA TEAUX AND HIGH PLATEAUX) and SP performing better in mountainous terrains (HILLS, MOUNT AINS AND HIGH MOUNTAINS). Differences between VF methods in RMSE can triple (e.g. see first row differences between KR (6.04 and SP 17.40), which agrees with the Fisher and Tate (2006) statement that no single interpolation method exists that is the most accurate for the interpolation ofterrain data. Still, the "best" method in Table 3 sometimes contains groups of two different VF methods, almost similar in RSME results. An example is the VF for PLAINS without any auxiliary DEM, where KR (6.04) and IDW (6.44) show similar results (similar results = ± 1 RMSE difference). On the contrary SP, MW and ANU are the most variable in tenns of performance. On the other hand, for certain TU, e.g. in HIGH MOUNT AINS, the algorithm SP ( 4.06) shows the " best" performance, with results of all other algorithms being quite different. Related to that last observation, the standard deviation, which is larger than for all other VF methods in that TU, suggests that there were severa! voids where SP perfonned poorly. Therefore the SD is a good indicator ofthe general applicability ofthe VF algorithm for a given TU. In this case, it would be advisable to check further to identify cases that are not well handled (e.g. it could be an effect ofthe void size) and to rerun the analysis. Generally, the use of auxiliary information of the GTOP030 dataset increased the RMSE for all VF methods except for small improvements in TR. This might be attributed to the different types of errors not accounted for in our methodology (see also the section on the SRTM above). 309 The use of SRTMJO as an auxiliary dataset decreased the RMSE/S for most case, and decreased the standard deviation indicating less variation in the leve! of ímprovement. This ís not surprising since SRTM30 is an up-scaled and void-filled derivative of the SRTM data. One could argue that the use of SRTM30 should be preferred, however in practice this means that we are down-scaling data (SRTMJO to SRTM) from a previous up-scaling (SRTM to SRTM30)! This circularity may be acceptable in certain cases if we know how the SRTM30 data were generated in the area of the particular void we are filling. This problem is discussed further below. Table 3.Mean and standard deviation (in brackets) ofthe RMSE ranking for each method by terrain unit. Best results are in bold. No auxiUary DEM KR• SP TR row MW TIN ANU PLAINS 6.04 (3.82) 17.40 (4.15) 9.79 (5.60) 6.44 (3.76) 10.6 (5.06) 8.52 (4.66) 10.00 (5. 18) PLATEAUX 5.71 (4.12) 9.79 (6.75) 16.40 (3.76) 10.50 (3.83) 10.70 (4.21 ) 6.74 (3.91) 7.02 (4.37) H. PLATEAUX 5.18 (3.56) 6.23 (6.24) 17.50 (2.06) 11.50 (3.07) 11.10 (3.35) 6.88 (3.53) 7.01 (4.04) HILLS 5.79 (4.39) 5.36 (5.62) 17.50 (2.18) 11 .50 (3.35) 10.70 (3.95) 7.01 (3.41) 8.36 (4.17) MOUNTAINS 5.95 (3.82) 5.16 (5.77) 17.10 (3.35) 12.20 (2.99) 11.30 (3.29) 7.14 (3.83) 7.78 (3.81) H. MOUNT AINS 5.97 (3.50) 4.06 (4.93) 17.50 (2.28) 12.90 (2.46) 12.60 (2. 79) 6.46 (3.3 1) 8.24 (3.95) GTOP030 KR SP TR IDW FF TIN ANU PLAINS 7.25 (4.72) 17.50 (4.02) 9.74 (5.51) 6.44 (3.76) 15.70 (6.48) 10.40 (5.30) 11.8 (5.39) PLATEAUX 6.00 (4.24) 10.20 (6.98) 16.30 (3.69) 10.50 (3.82) 18.60 (4.37) 7. 75 (4.54) 7. 72 (4.74) H. PLATEAUX 5.83 (4.27) 6.86 (6.83) 17.20 (2.23) 11.50 (3.06) 19.60 (2.46) 8.20 (4.21) 7.94 (4.46) HILLS 5.83 (4.31 ) 5.71 (5.90) 17.40 (2.17) 11.50 (3.36) 19.30 (3.40) 7.25 (3.54) 8.44 (4.28) MOUNTAINS 6.39 (4.33) 6.90 (7.26) 17.00 (3.15) 12.20 (2.98) 17.90 ( 4.79) 8.35 ( 4.59) 8.54 ( 4.03) H. MOUNT AJNS 6.14 (3.83) 4.47 (5.77) 17.40 (2.33) 12.90 (2.46) 18.60 (3.83) 6.91 (3.86) 8.53 (4.05) SRTM30 KR SP TR TDW FF TIN ANU PLAINS 5.67 (3.79} 16.70 ( 4.30) 9.65 (5.49) 6.44 (3. 76) 10.00 (6.38) 7.97 (4.98) 9.97 (5.3 1) PLATEAUX 5.28 (3.93) 9.34 (6.39) 16.20 (3.72) 10.50 (3.82) 15.50 (5.58) 6.65 (3.99) 6.69 (4.49) H.PLATEAUX 5.22 (3.48) 5.94 (6.03) 17.30 (2.20) 11.50 (3.06) 17.20 (3.78) 7.43 (3.81) 7.36 (4.1 8) HILLS 5.68 (4.23) 5.39(5.51) 17.40(2.14) 11.50 (3.36) 16.30(4.49) 6.88 (3.45) 8.30 (4.30) MOUNTAINS 5.49 (3.86) 5.02 (5.49) 17.00 (3.22) 12.20 (2.98) 15.50 (4.74) 6.97 (4.05) 7.50 (4.03) H. MOUNTAINS 5.63 (3.30) 3.86 (4.49) 17.50 (2.18) 12.90 (2.46) 16.70 (3.99) 6.38 (3.29) 8.00 (4.07) Evaluation of void filling methods with respect to void size Table 4 sbows the statistical summary of the ranking results when tbe void is classified by size. Again, the table sbows the mean and standard deviation (in parentheses) of the ranking of each VF method for each size class, with and without an auxiliary DEM. In tbis classification of voids, KR is consistently the best method for small and medium sized voids, regardless of the use of auxiliary data. If auxiliary data are u sed, KR performed best up to void-size class F compared with up to void-size class D without auxiliary information. The reason behind this is probably that the KR delivers an average elevation surface, which closer resembles the reference DEM, in contrast to the TIN dataset In TIN the relationship between triangles and their adjacent neighbours are handled more stringently, more closely resembling the input datase t. For Jarge and very large voids, the inclusion of an auxiliary DEM has an obvious effect on the performance of the algorithms. TIN is better where there is no auxiliary information, IDW is better where GTOP030 is used and TIN or ANU are best when SRTM30 is used. One might speculate on the differences between the 310 1 1 aux.iliary datasets. One possible explanation is that TIN/ANU requires 'good quality' auxiliary information, whereas IDW as a geostatistical algorithm that generates an "average surface" as mentioned earlier. 1 Evaluation of void filling methods with respect to terrain unit and void size Table 5 shows the best method for the final classification based on both terrain unit and void size, resulting 1 in 48 possib\e void typologies, although as can be seen from Table 5a, six of these typologies contain no voids (HIGH PLATEAUX in particular is arare terrain unit) and the number of voids per typology varies from 80 to 3. Tables 5a to Se show the best perfonning VF algorithm for each typology, again using no 1 auxiliary DEM (Table 5a), GTOP030 (Table 5b) and SRTM30 (Table Se). The table cells are shaded to help interpretation. 1 Table 4. Mean and standard deviation (in brackets) ofthe RSME ranking for each method by void size class. Best results are in bold. 1 No auxiliary DEM KR* SP TR IDW MW TIN ANU A 6.20 (3.84) 16.60 (4.74) 9.26 (5.97) 6.94 (3.89) 10.60 (5.45) 9.01 (4.89) 10.00 (5.45) 1 B 5.56 (3.99) 14.40 (6.74) 13.00 (5.23) 7.66 (4.52) 10.70 (4.71) 7.94 (4.40) 9.16 (4.82) e 5.74 (4.05) 9.77 (6.74) 16.80 (3.30) 10.70 (3.53) 10.70 (3.89) 6.27 (3.48) 6.70(4.18) D 5.74 (4.28) 4.49 (4.95) 17.70 (1.35) 12.00 (2.99) 11 .00 (3.70) 6.89 (3.26) 8.15 (4.10) E 5.52 (3.88) 5.33 (5.73) 17.30 (2. 77) 12.00 (3.16) 11.40 (3.39) 7.15 (3.81) 7.80 (4.05) 1 F 6.62 (3.50) 5.86 (5.97) 16.70 (4.29) 11 .60 (2.95) 9.95 (2.84) 6.67 (3.80) 6.41 (3.23) G 5.68 (3.36) 4.22 (5.64) 17.60 (1.72) 12.80 (2.48) 12.70 (2.98) 6.74 (3.40) 9.07 (3.85) H 7.20 (3.64) 6.36 (6.53) 17.40 (3.24) 13 .00 (2.37) 12.60 {2.32) 6.00(3.01) 6.16 (3.14) 1 GTOP030 KR SP TR IDW FF TIN ANU A 6.53 (4.15) 16.80 (4.69) 9.16 (5.93) 6.94 (3.89) 14.80 (7.07) 9.65 (5.35) 10.60 (5.56) 1 B 6.89 (5.03) 14.40 (6.66) 13.00 (5.11) 7.66 (4.51) 16.30 (5.94) 10.10 {5.12) 11 .00 (5.43) e 6.14 (4.25) 10.30 (7.05) 16.60 (3.22) 1 o. 70 (3.53) 19.60 (2.78) 7.55 (4.44) 7.62 (4.68) D 5.75 (4.20) 4.81 (5.28) 17.60 (1.37) 11.90 (2.99) 19.50 (2.54) 7.13 (3.40) 8.22 (4.21) 1 E 5.83 (4.19) 5.64 (6.08) 17.20 (2.82) 12.00 (3.16) 18.00 (4.85) 7.77 (4. 14) 8.23(4.19) F 8.31 (4.81) 12.00 (8.39) 16.20 (3.62) 11 .60 (2.92) 19.80 ( 1.63) 10.70 (5.32) 8.91 (4.31) G 5.57 (3.26) 4.11 (5.41) 17.50 (1.79) 12.80 (2.48) 17.90 (4.26) 7.00 (3.65) 9.20 (3.87) H 7.73 (4.44) 7.66 (8.02) 16.90 (3.30) 13.00 (2.37) 20.00 ( 1.66) 7.43 (4.65) 7.10 (3.99) 1 SRTM30 KR SP TR IDW FF TIN ANU 1 A 6.26 (3.84) 16.5 (4.74) 9.21 (5.94) 6.94 (3.89) 10.50 (6.51) 8.85 (5.06) 10.10 (5.47) B 4.96 (3.81) 13.40 (6.37) 12.80 (5.21) 7.66 (4.51) 11.20 (6.83) 7.35 (4.65) 8.90 (5.00) e 5.21 (3.84) 9.20 (6.29) 16.60 (3.24) 10.70 (3.53) 16.00 (5.00) 6.15 (3.59) 6.32 (4.29) D 5.64 (4.12) 4.51 (4.82) 17.70 (1.30) 11 .90 (3 .00) 16.80 (3.87) 6.79 (3.30) 8.09 (4.22) 1 E 5.53 {3.85) 5.21 (5.63) 17.20 (2.81) 12.00 (3.16) 16.60 (4.36) 7.43 (3.93) 7.96 (4.09) F 5.65 (3.80) 6.65 (6.48) 16.30 (3.89) 11.60 (2.92) 14.70 (4.01) 6.46 (4.38) 5.95 {3.46) G 5.17 {3.31) 3.15 (3.50) 17.50 {1.75) 12.80 (2.48) 16.20 (5.12) 6.38 (3.49) 8.64 (4.30) 1 H 6.10 (3.28) 5.73 (5.66) 17.20 (2.99) 13.00 (2.37) 16.00 (3.49) 5.76 (2.89) 5.40 (3.08) *Methods are Kriging (KR), Spline (SP), Trend (TR), Inverse Distance Weighting (IDW), Moving 1 Window Average (MW), Fill and Feather (FF), Triangulated Irregular Network {TIN), and ANUDEM (ANU). 1 311 1 The frrst observation is that KR outperfonns any other VF algorithm for very small voids (Class A), except in High Mountains, where SP is better. Secondly, for very large voids in mountainous terrain (arguably the most difficult voids to interpolate), ANU and TIN are the best, with ANU being superior when SRTM30 is used. Other noticeable trends include IDW in planar areas for large voids with no DEM or GTOP030, and SP for medium to large voids (C1asses C to F) for all terrain units except Plains, and especially for Mountains and High Mountains. A test of the SD of the RSME for the different methods showed a similar distribution across all VF methods, void sizes and terrain units, which allows us to be quite confident in the results presented here. One exception is however for the largest void class (H) in the terrain unit PLAINS and PLA TEAUX for the VF method SP. The reason for that exception might be attributed to the relatively low number of investigated voids in that class. Table 5. Best method results in tenns of average rank by terrain unit and void size using (a) no auxiliary DEM, (b) GTOP030 auxiliary data, (e) SRTM30 auxiliary data, and (d) across all data methods. (a) No auxiliary Terrain unit PLAINS PLATEAUX H. PLATEAUX HlLLS MOUNTAINS H. MOUNT AINS (b) GTOP030 Terrain unit PLAINS PLATEAUX H.PLATEAUX HlLLS MOUNTAINS H. MOUNT AINS (e) SRTM30 Terrain unit PLAINS PLATEAUX H. PLATEAUX HlLLS MOUNTAINS H. MOUNTAINS (d) All cases** Terrain unit PLAINS PLATEAUX H. PLATEAUX HILLS MOUNTAJNS H. MOUNTAINS A s e KR5.70 *KR 6.1 KR5.92 KR3.33 KR4.83 KR5.30 KR 4.98 KR 4.48r--""'~~ SP 3.65 SP4. SP2.40 KR 4.07 K.R 4.05 KR 4.97 SP 3.84 KR 4.40 SP 2.35 A B e SP 1.55 SP 2.40 D KR6.11-KR5.70 KR 5.82 KR 4.98 KR 4.48 KR 3.36 SP 3.65 SP 4.75r----==-==- KR4.83 KR3.90 KR4.05 KR 5.30 KR 4.95 SP 3.84 SP 2.40 KR 4.40 SP 2.35 A B e KR 6.11··· KR 5.70 KR 5.77 KR 4.98 KR 4.48 KR 3.60 SP 3.65 SP 4.15 D KR6.80 SP 5.29 KR 4.83 KR 4.02 KR 4.05lii~II!~ .. M..,.I111111111 KR 5.30 KR 4.95 SP 3.84 SP 1.55 SP 2.40 KR 4.40 SP 2.35 SP 2.40 A B e D Void size E Void size E Void size E E F H SP4 w 520 -\'\l ' S. n \'\l ~~() F F KR4.00 KR4.95 F G H K.R 7.00 G H * Methods are Moving Window A~e (MW), Kriging ~ Spline {§f), lnverse Distance Weighting ~. Triangulated lrreguJar Network ~. and ANUDEM (1111). **No auxiliary DEM (no aux), SRTM30 (sr30) and GTOP030 (gt30). 312 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Looking across tables 5a to Se, the use of coarse resolution auxiliary DEM data has little or no impact on the results for void size classes A through D, but the possible inclusion of such auxiliary infonnation becomes important for the larger void classes (E through H). lf a VF algorithm is to be recommended for these void sizes for global application, then we must differentiate between the three choices for auxiliary infonnation. Table Sd shows the best perfonning VF algorithm and the auxiliary infonnation that was used foral! terrain units for size classes E through H. Looking only at the auxiliary DEM results, we can see that SRTM30 is the preferred auxiliary DEM for very large voids (size classes G and H), whilst there is no one recommended auxiliary DEM for classes E and F. Looking at the VF method and auxiliary infonnation together, KR always uses SRTM30, and ANU always uses SRTM30, except for the PLAINS, where a variety of methods is recommended based on the void size class. A mixture of VF algorithm is recommended for medium size voids (E-F) based on the TU (IDW for Plains, TIN for HILLS, and SP/ANU for the remaining TU). lt is interesting to see that certain algorithms deliver the best results without auxiliary infonnation (e.g. recommending SP without any auxiliary infonnation for size class F in high plateaux). Further investigation is required to determine why the errors without auxiliary infonnation are less than with VF algorithms that use auxiliary infonnation. One reason might be that we a have not sufficiently compensated for the errors in the dataset (cf. the section on the SRTM above) during the preparation of input data for the VF and this may bias the results. Another observation is that most of the recommended VF algorithms create a smooth surface which stays within the elevation range of the input data. This means that mountain ridges for example can not be represented properly, even if the RMSE/ME proves that the approximation is best for the VF patch, and that the local noise structure of the surrounding area is smoothed out. Finally, GTOP030 is derived from a range of topographic sources, and the variation in quality of these sources across Africa is likely to have an impact on these results. Eva/uation of void filling methods with respect lo auxi/iary information To determine if the inclusion of low quality auxiliary infonnation has any improvement on the VF algorithms, for each void we computed the percentage difference between the RSME with no auxiliary infonnation and (i) RMSE with SRTM30 and (ii) RMSE with GTOP030. The results are summarised by void size class and terrain (Table 6) foral! void sizes E through H. Percentages lower than l 00 indicate that the inclusion of auxiliary information resulted in an improvement and vice versa. A percentage equal to 100 means that the first restriction of the area threshold ( cf. the section on data pre-processing for the artificial voids) has been met, however due to the second restriction (e.g. the shape ofthe void) the auxiliary data have not been used. As expected, SRTM30 improves the VF results more overall than GTOP30, but surprisingly both auxiliary datasets also degrade the VF results in severa\ cases (e.g. in mid-sized voids and in sorne TU - Table 6). This suggests that the area threshold should depend on the TU. For example, for VF algorithm KR in MOUNTAINS using GTOP030, an increase in accuracy can only be observed if the void area is larger than void-size class F. Below that void-size class, GTOP030 did not improve the VF results and even decreases accuracy. For GTOP030, the area threshold should be only the largest class (H) for PLAINS, LOW PLATEAUX, HILLS and HIGH MOUNTAIN. For HIGH PLATEAUX no recommendation can be given (or even not to use coarse scale auxiliary infonnation), and for MOUNTAINS the area threshold should be class G. SRTM30 not surprisingly outperforms GTOP030. Generally, the largest improvements for SRTM30 occur in large and very large void sizes (G and H). Similarly to GTOP030, the area thresholds and when to use auxiliary informatjon vary. For PLAIN, LOW PLATEAU and MOUNTAINS the void size class G is 313 recommended; for HIGH PLATEAUX again no recommendation can be given and for HILLS and HIGH MOUNTAIN the area threshold should be set to H. These results indicate that such coarse resolution auxiliary data is generally only applicable to extremely Iarge voids, and highlights the need to use higher resolution auxiliary datasets in filling voids in the SRTM data, rather than the SRTM30 and GTOP030 datasets used here. Table 6. Average reduction in RMSE (in%) when auxiliary DEMs are used. Methods are Kriging (KR), Spline (SP), Triangulated Irregular Network (TIN), and ANUDEM (ANU). Terrain Unit VF GTOP030 SRTM30 Method E F G H E F G H KR 143 118 161 192 109 100 91 72 SP 141 114 103 35 99 87 59 16 PLAINS TIN 171 144 218 279 l07 lOO 99 89 ANU 183 157 210 273 ll4 103 99 96 KR 105 103 119 107 99 94 95 87 SP 110 134 157 84 99 105 96 58 PLATEAUX TIN 111 120 130 121 101 105 94 82 ANU 107 115 126 120 100 100 94 81 KR 121 104 HIGH SP 138 100 PLATEAUX TIN 144 120 ANU 128 112 KR lOO 100 110 100 lOO lOO 102 97 SP 100 100 162 78 100 100 118 71 HILLS TIN 100 lOO 112 106 100 100 101 94 ANU 100 100 JOS 104 100 100 103 91 KR 100 127 81 100 93 72 MOUNTAINS SP 100 161 137 100 101 84 TIN 100 155 94 100 114 77 ANU 100 149 85 lOO 120 71 KR 112 116 121 86 93 95 100 61 HIGH SP 239 161 207 62 134 107 140 37 MOUNTAINS TIN 122 137 235 91 95 100 180 64 ANU 114 118 205 90 95 97 162 63 Application ofthe 'best ' VF methods to the global SRTM data One of the objectives of this paper has been to provide a worldwide data base of voids, in which each void has an assigned "best" method based on terrain unit and void size. This has been performed based on the results in Table 5, and the database will be provided to the intemational user community at http://srtm.jrc.it/ and http://srtm.csi.cgiar.org/. The Intemational Center for Tropical Agriculture (CIA T), through the Consortium for Spatial lnformation (CSI), has been providing ready-to-use searnless (i.e. void filled) 314 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 SRTM elevation data since 2003. These derived data have been gradually improved over three versions through the use ofbetter interpolation algorithms (currently ANUDEM) and auxiliary DEMs. The database of "best'' VF methods could be used to create a fourth version of the seamless SRTM elevation data by identifying the voids where there is no high resolution auxiliary DEM informatioo available (which is currently the vast majority of voids), and by then applying the recommended VF algorithm to the remaining voids. Where high resolution DEMs are available we recommend the ANUDEM procedure. Conclusions General conclusions We assume that each void occurs dueto a technical reason, which can be partly attributed to terrain, land use and other reasons. In the course ofthis analysis, void areas in the SRTM data set have been quantified in terms of their terrain and size, providing a statistically sound and extensive evaluation of different VF algorithms over a wide range ofterrain units and void sizes (objective (i)). Different VF algorithms have been implemented in a GIS, and used to analyse performance using RMSEIS on 1,304 relocated voids. Based on these results a decision table has been created, which provides an answer to an important question: which VF method can be recommended for a void of a given size in a given terrain unit? Contrary to sorne previous findings, the best methods can be generalised as: Kriging or Inverse Distance Weighting interpolation for small- and medium-size voids in relatively flat low-lying areas; Spline interpolation for small and medium sized voids in high altitude and dissected terrain; Triangular Irregular Network or Inverse Distance Weighting interpolation for large voids in very flat areas, andan advanced spline method (ANUDEM) for large voids in other terrains (objectives ii and iv). We have shown that coarse resolution auxiliary infonnation was only helpful if the void areas exceeded a certain size threshold (objective (iii)). Differences in decrease of RMSE could be observed between use of the SRTM30 and the GTOP030 DEMs. Finally, we have created a database that can be used to select a VF algorithm and auxiliary DEM to fill each ofthe 3,339,913 voids in the SRTM data (objective (v)). Further work lssues with the SRTM30 data. In this paper we tested only coarser scale resolution data as auxiliary information for the VF process. The VF using the SRTM30 obviously creates a better result than using GTOP030 auxiliary information (Table 3). Still , SRTM30 is a seamless DEM based on the SRTM data product and is therefore also influenced by the voids. Most voids in the underlying SRTM data were interpolated in a 1 O x 1 O averaging process that re-sampled the data from 3 are seconds to 30 are seconds. Voids that were too large to be interpolated in this manner were filled using GTOP030. Since the SRTM data do not have global coverage, GTOP030 data from areas above 60 i.Aa M+4..iioe -·· • C.l ... •fl • eauud.~ ~ L.OU·•C~ _j ,..._.,... ... ~4 ~ ·--, _ Cil4et , .. 2742 ... ,..,...,... .......... .......... ,., fiftc.a• •.u" m.w~o4.. Figure 4. Cinto online farm map interface. - --· - · -· -· ... _ ..,...,,_...,......, ..... "-. ~· . . .......... ., (1) (2) (3} The coffee product track runs fi-om the field where coffee is grown to consumers all over the world while the pbysical state of coffee changes fi-om the green, fruit pulp-covered coffee bean to a wide variety of roasted, blended, ground and packed coffee of different flavours for different uses. Tbe main product tracking nodes through which coffee runs are production in the field, (on-farrn) post-harvest processing (fruit pulp peeling, fermentation, and washing, drying), further processing in coffee cooperatives (sorting, quality determination, peeling and others), change of ownership and transportation through exporters to importers to roasters (roasting, blending, grinding), and fmaJJy the marketing process with the end-point sale toa customer (general public or food industry). To link the final product to geographical origin and processing infonnation a set of product tracking codes are applied. These codes are directly attached to the product and participants can obtain infonnation about the product from the central data storage and monitoring system (CintO database). CintO distinguisbes three different code systems (Figure 5), the management unit code, the coffee-lot code (also named cupping code) and the product code. Tbe management unit code is used for linking all production and processing data as well as geographic information about the production zones to a specific environment or site in time, defined as the management unit. The co:ffee-lot code links the management unit code with the harvest date and lot quantity enabling the tracking of quality data and other harvest-dependent parameters. The product code is mainly for roaster and end customer use to identify the origin of the packed and ready-to-sell product. Due to the fact that many coffee roasters use a mixture of coffee from different regions and producers, this code is needed to link severa! coffee-lots with the help of a new code that appears on the coffee pack. 330 Fanns: Post-Harvest Processlng • Oepulpillg • Orylng Cupping 1 Quúty • C4Jpplng Standards (SCM. CoE. .. . ) Management-Unit..Code: [Example: M-G078F6] ).> Generated on management-unit registmtiQn in database. ).> Enables to link site-information in ~se. Code users: producers, coffee coope@!!O.flS Lot name 1 Cupping Code: [Example: LOT ·1234/ C-1 A2B] ).> Generated on harvest registration in data base. ).> Links harvest date, quantity toa certain management-unit. Code users: coffee coope¡atio.!J§, coffee quality laboratories, exporters. importers, roasters Product..Code: [Example: P-G0006F] ).> Generated from roaster on registering ~od.-g[Qduc,Y; in 1 data base. ).> Links the different coffee-lot codes 1 ................... ~...... to a certain ~nd-Qrod~. 1 { P· 002E45 \ Code users: roasters, traders, consumers : H·XX3 •• ......................... ./ L-~========~~============~~ ------------------ -· Figure 5. Basic code system enabling product tracking along the coffee supply chain. The management unit code has to be a simple code, easy to read and process. Furthennore the code is unique system-wide and comes with a simple error detection mechanism. The checksum method was selected to provide an easy to implement the error detection mecbanism (for details see Tanenbaum and Goodman 1999). On registering a new field in tbe CintO system, a new system-wide unique number is generated. Afterwards a checksum bit is added to this number and the whole expression is converted to a hexadecimal number. The hexadecimal format is shorter than purely numeric formats and the alpbanumeric presentation is easier to read. The coffee-lot code (cupping code) follows a different scheme. ft is designed to facilitate objective coffee quality determination through "blind cupping" and determination of physical characteristics. lt consists of one number (between 1-9) followed by a character (A-Z). This number-character-combination is repeated depending on CinfD's demand for codes. The preftX "C-" is used to identify this codeas coffee-lot code. There are two possible ways of creating the product code. The roaster can either select the corresponding coffee-lot codes for all coffees used for sorne blend and receives a product code from the CintO system, or the roaster chooses a bar or other code unique for that particular batch and links it with the coffee-lot codes from the CintO database (Niederhauser, 2004). Wagner and Glassheimer (2003) explain a similar but more generic approach for a food tracking system. The code is presented in hexadecimal numbers. Electronic sean devices and RFID technology are not currently available for the CintO pilot project phase, however current hexadecimal codes are compatible 331 with these technologies and have the advantage of incorporating an error detection system, which is essential for use with sean devices. Geographical analyses and data export. If specific product characteristics (e.g. coffee quality) are linked in real time to existing spatial models (altitude model, climate, political division) the combined data can be used to discover production niches for particular types of coffee. Since computing power and software technology for real time applications like the generation of dynamic web content is still limited, only basic models can be run directly on CinfO. For more sophisticated analyses or data interchange with other systems, data stored in the central CinfO database can be made available as a downloadable file in different formats: • CSV-file (comma separated value, readable with any spreadsheet program); • Shape-file for GIS software (Fig 6); • Google-Earth® format (Fig. 7); and • A special file format for Expector- a software program for statistical analyses . ., Unhtlcd ArcMdp Artlnfo r:,_ 'e~ ~ Be {lit l!M ~ :;.-. Ioois l!ti'>Ciow ~ D ~ g 5 .., ~ ""lu~.240.!68=,_.::::;.;:=3c,· ~ IAyero =. !i!l - • - !i!I OirtH.... E¡ . llnalypsT .... • · c.toQr·T-, .~Tools "· """·l .... ·•011·-'""' · ·~~-• • -Anllvtl '""' · ·LiiNr~l-• • 5potio1 Anllytt Tools • . 5potiolst-.Tools ~ o "..:.'--------------..J...!.J ~ O • A • @jAN~ i]¡;¡¡-::j B I U A_• "- • .!,• ~· Fieure 6. GIS analvses on CINFO data. 332 - ·- ......... ~-· E!J • ot -·-........ ..., .,JS la• • ,.....,.. ... ~ ......,__. ,.. E!!• lA ,_ ............ .. El • .. ,_,._ ..._....., ,,,. e!• u . .._ .... eJ~...aw nu ,_ u ... ............. eJ•• , ____ __ e~:7' __ .. , ·---~ eJ:':-_,. ,.., ·---•c..w. ......,...,,... E!!• • ,...._.,. .. ............. ,.,. El• .. , __ ",.._ 16,:-::-..... .u Figure 7. CiniD data on Google-Earth® For example, an intelligent feedback (two-way dialogue) system provides the farmer with a quick overview of bis productivity status in terms of yield and quality compared with others with similar conditions, thus increasing his knowledge about his potential opportunities for profitable differentiation. Complex automated data-analysis processes produce easily understandable and powerful reports in different presentation. An example is demonstrated in figure 8, a farm map that indicates the productivity and quality status of different management units. Discussion System uptake The development of the system and its presentation to specialty-coffee supply-chain actors has spurred considerable interest in the adaptation of the whole system or of specific modules. Producers' associations use the system to monitor and document their production better in order to address variation in product quality systematically. Large associations representing producers at a national scale are currently considering the option to research with the project team the feasibility of implementing denomination of origin concepts based on improved information management. Specialty coffee exporters are using components ofthe system to (a) control product quality upstream and (b) provide their downstream buyers traceability of the product to the field. Large specialty coffee roasters are considering using principies developed under CintO to design sourcing strategies. 333 Jnformation asymmetry e high quality e low quallty @) profitable @) no profit or loss Figure 8. Aerial view image offarm management units and their rating in production and quality. The CinfD system is set up and functional. Both producers and users have found it a useful tool for managing their businesses with special emphasis on improved management of the supply chain for differentiated coffee products. Nevertheless, it is still in the early stages of development and several areas could be improved and various deficiencies need to be addressed: One of the principal aims of Cinto is to facilitate harmonious dialogue along the whole extent ofthe supply chain including the end user. 334 Up to now the roasters and buyers provide the information related to end users' opinions, preferences and whims. This is not ideal as the filtered information can cause asymmetries anda Iack oftransparency in the information flow along the supply chain. For example, it may not be in a roaster's best interest to pass on the information that a particular coffee from a few special farms was extremely highly rated by a particular group of end customers as this would possibly raise the farm-gate price of the coffee and reduce his margins. Or on the other hand if the roaster does not obtain information directly from the end customer the whole supply chain may never know that they have a particular niche product for that group of consumers and everyone loses out. ln the future, in the same manner as cupping schemes were set up in this project to determine the quality of the coffee from individual farms, it m ay be necessary to establish coffee-cupping sessions for end customers. Harmonizing communication The experience gained in this pilot phase indicates that there are differing views about what constitutes coffee quality at the different nodes along the supply chain. To give an example, sorne farmers say coffee is of good quality when it does not smell like chemicals whilst others associate good quality with sophisticated management. On the consumer side, coffee quality is often associated with a distinct cupping qualíty proftle coupled with excellent product presentation. To make the information flow along the supply chain it is crucial to know what infonnation is needed ata certain node, and how to present or deliver this information in respect to the knowledge and needs of the supply-chain member at this node. Sophisticated information engineering is needed to model and transform informatíon to suit the varied needs of supply chain members without distorting that information and losing its intrinsic value. Information generation Up to now, members of the pilot scheme team have supervised much of the data collection at the farm level. Furthermore, much ofthe data have been tabulated by hand for later digitalization. lfthe pilot scheme is to become a standard for the whole sector, then this close supervision will be impossible, and data collection will surely have to be in the hands of the chain actors themselves, who will ha ve to use digital data-capture systems. lt is only when the producers themselves, the roasters and taster and all the others along the supply chain perceive the benefits of collecting and supplying data to CinfD that they will enthusiastically and consistently feed the system with reliable information. The initial results already indicate tangible benefits for all members of the supply chain and hence the future potential of CinfD. Experience with more enterprising producer groups suggests that this is possible. Groups we have visited have established theír own cupping and physical qualíty-testing laboratories and in preliminary trials with portable computing devices such as PDAs were promising. Specialized software for these systems will need to be developed to facilitate data capture, but it does seem highly probable that producer organizations will accept and adopt these technologies. Already groups such as the San Roque group in Oparapa, Huila have seen the advantages of carefully evaluating and grading their coffee so as to sell into specialized markets at higher prices. Furthermore, they have been able to associate certain growing conditions and management practices with particular qualities that are sought after and have been able to focus production goals on these particular niche markets. 335 References Antle, J. (1987). Econometric Estimation of Producers' Risk Attitudes. American Journal of Agricultura/ &onomics 69:509-522. ASA (1999). Symposium 1998 ASA meeting Baltimore. Crop Science 39:1547-1630. Beurskens F. AgriBiz and Associates, [nc. (2003). Market facilitation of grain marketing: The end user's perspective. Proceedings: Product Differentiation and Marlret Segmenta/ion in Grains and Oi/seeds: Implications for Industry in Transition. Economic Research Service, USDA and the Farm Foundation. Washington, DC, 27-28 January,2003. Christopber, M. (1992). Logistics and Supp/y Chain Managementfor Reducing Costs and Jmproving Services. Pitman, London. P.l3. Clause, R (2003). Supply chains, quality assurance and traceability using ISO 9000-2000 in agriculture. Proceedings: Product Differentiation and Market Segmentation in Grains and Oi/seeds: Implications for Industry in Transition. Sponsored by Economic Research Service, USDA and The Farrn Foundation, Washington, OC January 27-28, 2003. URL: http://www.farmfoundation.org/projects/documents!RegClause.pdf#search=%22traceability%201S0%22 Cooper, M.C., Lambert, D.M. and Pagh, J.D. (1997). Supply chain management: More than a new name for logistics joumal: Jnternational Journal of Logistics Management 8: l-14. Daviron, B. and Ponte, S. (2005). The Coffee Paradox. Zed Books. London and New York. 295pp. Dent, D. and Young, A. (1981). Soil Survey and Land Evaluation. George Allen and Unwin, Norwich, United Kingdom. 278pp. Dickson, P.R and Ginter, J.L. (1987). Market segmentation, product differentiation and marketing strategy. Journal of Marketing 57: 1-1 O Evans, L.T. and Fischer, R.A. (1999). Yield potentiaJ: its definition, measurement, and significance. Crop Science 39:1544-1551 Go1dsmith, P.D. and Bender, K. (2003). Ten conversations about identity preservation. Journal ofChain and Network Science. 4:111 -123. Gunasekaran, A., and Ngai, E.W.T. (2004). lnformation systems in supply chain integration and management. European Journa/ ofOperational Research 159:269-295. Henrich, J. and McElreath, R. (2002). Are peasants risk averse decision-makers? Curren! Anthropo/ogy 43:1 72-181 . lDTechEx (2005). RFID case study: Agri-Tracabilité Québec (ATQ). URL: http://www.idtechex.com/products/enlarticles/00000290.asp. Accessed 17 May, 2006. lnformation Society and Media DG (2004). Farm to fork traceability for the fruit sector. URL: http://istresults.cordis.lulindex.cfm/sectionlnews/tpVarticle/BrowsingType/Features!ID/73181. Accessed 17 May, 2006. lnformation Society and Media DG (2005). Keeping track of food as packing crates get smart. URL: http :/ / istresults. cord is.l ulindex.cfm/sectionlnews/tpVarticle/BrowsingType/F eatures/ID/77349. Accessed 17 May, 2006. Infonnation Society Technologies (2005). GeoTraceAgri final project report (GT A). URL: http://www.geotraceagri.net/ Accessed 16 May, 2006. Juan Valdez (2006). http://www.juanvaldez.com/newjuan.htm# Kaerkkaeinen, M. & Ala-Risku T. (2005) Tracking based material flow transparency for small and medium sized enterprises. PhD thesis, Helsinki University of Technology, Department of Industrial Engineering and Management. Pp.1- 18 Klirkkiiinen, M., and Ala-Risku, T., (2005). Tracking based material flow transparency for small and medium sized enterprises. Working paper, unpublished. © 2005 by authors. Ketkar, V.N., Whitman, L. & Malzahn D. (2003). Ontology-based product tracking system. URL: citeseer.ifi.unizh.ch/ketkar03ontologybased.html. Accessed 16 May, 2006. Kingwell, R.S. (1994). Risk attitude and dryland farm management. Agricultura/ Systems 45:1 91- 203. Lambert, D.M, Cooper, M.C. and Pagh, J. (1998). Supply chain management: implementation issues and research opportunities. International Journal of Logistics Management 9: 1-20 Lin, W. (2002). Estimating the costs ofsegregation for non-biotech maize and soybeans. In: Santaniello, V.,. Evenson, R.E and Zilberman, D. (eds.) Market Development for Genetically Modified Foods. CABI Publishing, Wallingford, UK. 336 1 1 1 1 Niederhauser, N. (2004). Internet Based Product Tracking and Information System for Coffee and Other High Value Crops. Thesis. lnformation and Communication Engineering. Vorarlberg University of Applied Sciences, Austria. 76pp. Opara, L.U. (2003). Traceability in agriculture and food supply cbain: A review of basic concepts, technological implications, and future prospects. Food, Agriculture and Environment. l: 1 O 1-106 Smytb, S. and Phillips, P. (2002). The battle between GM canola and public, prívate and collective interests: Defming and documenting the costs and benefits of identity preservation, segregation and traceability, Proceedings of the Intemational Consortium on Agricultural Biotechnology Research Meeting; Ravello, Italy. ICABR. Sonka, S. (2003). Forces driving industrialization of agriculture: implications for the grain industry in the United States. Proceedings: Product Differentiation and Marlcet Segmentation in Grains and Oi/seeds: Jmplications for Industry in Transition. Economic Research Service, USDA and the Farm Foundation. Washington, DC, 27-28 January,2003. Tanenbaum, A.S. and J. Goodman. (1999). Computerarchitektur. Prentice Hall. 4th Edition. Munich. Pp. 87-91 The Board of Trustees of the University of Illinois (2004). "Track and Trace" agricultural grains with a whole new Ievel of sophistication using RFID system approach. URL: http://www.otm.uiuc.edu/attachments/RFID- GC.pdf. Accessed 17 May, 2006. Trienekens, J.H., Beulens, A.J.M., Hagen, J.M. and Omta, S.W.F. (2003). Innovation through intemational supply chain development: A researcb agenda. Jnternational Food and Agribusiness Management Review 6. Van Roekel, J., Kopicki, R , Broekmans, C.J.E. & Boselie, D.M. (2002). Building Agri Supply Chains: Issues and Guidelines. URL: http://www.kc-acc.org/pdf/acc0395.pdf. Accessed 16 May, 2006. Wagner, G.L and Glassbeim, E. (2003). Traceability of Agricultura/ Products -An Action Program to Develop an Efficient and Comprehensive Traceability System Using the Latest Data Collection and Transfer Technology. United States Department of Agriculture, Washington. Williamson, E., Harrison, D.K. and Jordan, M. (2004). Information systems development within supply chain management lnternational Journal of Information Management 24:375-385. Woods, E.J.(2004). Supply-Chain Management: Understanding the concept and its implications in developing countries. School ofNatural and Rural Systems Management, University ofQueensland Gatton. 18-26 World Bank (2006). http://econ.worldbank.org/programslrural dev/topic/23807/ 337 Satellite imagery and information networks for monitoring climate and vegetation in Colombia Glenn Hyman, Carlos Meneses, Elizabeth Barona, Ernesto Giren and Claudia Perea Centro Internacional de Agricultura Tropical Key words: MODIS, Colombia, climate, vegetation, NDVI, remote sensing. Abstract This paper describes a proposal for establishing a network of researchers and analysts for monitoring weather and vegetation for Colombian agriculture. Opportunities for using satellite images and other data products are evaluated. The paper suggests how such a network could be put together. Sorne preliminary pilot studies have been conducted to assess the feasibility ofthe proposed project. Introduction Weather and vegetation monitoringfor agriculture andfood security Agricultura! and environmental scientists, market analysts, farmers and others monitor weather and vegetation change throughout the growing season. The information can be used by the Ministry of Agriculture to plan extension activities. Market analysts use the data to estímate shortfalls in production or likely effects on prices of the coming harvest. Researchers use weather and vegetation information to understand better how climate affects crop growth and yield, and other factors related to agroecosystem health. Scientists and professionals in developed countries have made great progress in developing weather and vegetation monitoring systems for agriculture. Sorne of the demand for these systems has come from market analysts who want to know how reduced or increased harvest might affect farm and food prices in different parts of a country or the world. One example of such a system is the United S tates Department of Agriculture's Crop Explorer, which provides data for users throughout the world (Foreign Agricultural Service, 2006). Other uses of weather and vegetation data are food security professionals. The Famine Early Warning System developed for Africa and Central America provides information from satellites for countries to plan their food aid programs in the context of expected harvests due to weather conditions (FEWSNET, 2006). Weather and vegetation monitoringfor Colombia As with many countries, networks of weather stations cannot cover the full range of agricultura! environments throughout the country. A satellite-based weather and vegetation motútoring system that could provide data for places without ground stations would be of great benefit to agricultural areas that lack monitoring infrastructure. Food security monitoring systems Like FEWSNET are not yet available for Colombia. Although it is possible that Colombia could become a partner in FEWSNET in the future, it is unlikely to happen soon. Colombia is of lower priority for food security monitoring sin ce the country has relatively less drought than other countries. Sorne information from Crop Explorer is available for Colombia, but often researchers need access to the raw data. Another problem for Colombia is the lack of use of satellite imagery by the research and development community. Sometimes raw data is inaccessible to people outside ofthe country in which it was produced. Shipping data by Internet or mail courier may present additional problems to 338 acquiring satellite data. Language barriers, lack of training and other factors may al so con tribute to less use of satell ite imagery. How might sorne of these limitations be overcome? How could we facilitate increased use of remate sensing data for monitoring climate and vegetation in Colombia? Proposal A pro posa/ for a c/imate and vegetation monitoring network for Colombia We propose to develop a network that would improve decision-making for Colombian farmers by providing researchers and analysts with useful near-real time satellite data on weather and vegetation. The project would support the larger goal of providing information that supports íncreased productivity and food security in Colombian agriculture. Data for monitoring vegetation and weather Govemrnents and data providers are increasing the number of sateUites and imagery products that are available to users of these data throughout the world. Countries with a long history of satellite imagery prograrns, like the United States, Russia and France, are making more of their products available to users. More recently countries such as China, Brazll and India have made new imagery products available. The availability ofthese data is creatíng new opportunities for monitoring weather and vegetation. Two data products for monitoring weather and vegetation would be appropriate for Colombia. MODIS imagery and the Tropical Rainfall Measuring Mission (TRMM) data sets provide data on vegetation vigor and rainfall respectively. Vegetation índices can be developed from the MODIS imagery to estímate the vigor or water status of plants. The MODIS images are available at 250, 500 and 1000 meter spatial resolutions. Other variables available from MODIS include reflectance, temperature, aerosols and others. Imagery for a given MODIS satellite image product is taken every 16 days. The TRMM data is available every three hours from the NASA web site (NASA, 2006). Rainfall is reported in millimeters for 3-hour or greater time periods for the entire globe. One advantage of this data set is for areas where no ground rain gages are available. TRMM is therefore a viable altemative for countries that have poor coverage of weather stations or have otherwise been unable to maintain stations. The MODIS and TRMM data have been processed for Colombia as a pilot study at the Intemational Center for Tropical Agriculture (CIA T). With the assistance of the United States Geological Survey (USGS), we were able to process fully vegetation índices from MODIS images. Processing requires selecting high quality imagery, stitching together image tiles, re-projecting the data toa standard Colombian coordinate system and converting the images to formats appropriate for digital image processing and geographic information systems software. Many of the processing algorithms are provided by the USGS. Figures 1 a and 1 b shows January 2004 images of the enhanced vegetation index (EVI) and the normalized difference vegetation index (NDVI) for Colombia. Without the assistance of USGS scientists it would have been very difficult for us to put together these images. But with better-documented programs, the processing algorithms could be applied by a wide range ofpotential users. One difficulty in developing these images is in acquiring the raw data itself. We acquired these data by speciaJ arrangement with EDC. Standard download times are too slow to acquire the images over the Internet. They were therefore sent to Colombia by international mail courier on DVDs. We discuss potential solutions to acquiring the data more efficiently later in this document. 339 We have also acquired TRMM data for Colombia and the surrounding regions (Figure 2). These data sets were deve!oped by colleagues at the USOS. Each digital map shows the l 0-day rainfall accumulation in millimeters. Pixels are 15 arcminutes, about 27 km a side near the equator. The rainfall maps have adequate temporal and spatial resolution for vegetation and climate monitoring. A ..... •. -- . ....... ---- ....... -~-­. ....... --~· ..... ------ mrtrnosaic - i MODextract.log -o TmpMosaic.hdf - s '1 l O O 1 1 1 1 O O O' This instruction creates a mosaic from the data NDVl, EVI, red retlectance, NIR reflectance, blue reflectance and M1R retlectance. The values 1 in the instruction parameters indicate the data selected for extraction, while zeros indicate the data that will be skipped in the process. The output data are stored in a new HDF file, not by individual tiles, but as a mosaic for the entire area under consideration. MOD13Q1 .004 2004.01.01 2004.12.25 09,09 10, 07 08 09 11 , 07 08 09 #input dataset and version -avg size 125MB #start date #end data #horizontal, vertical grids (09,09) #horizontal , vertical grids (10,07) (10,08) (10,09) #horizontal , vertical grids (11,07) (11 ,08) (11 ,09) Fieure 2. Code used to generate the log file for the HDF format. The MODIS Reprojection Too] (MRT) is used to re-project the data SIN (Sinusoidal) to one of the more common projections, in the case of Colombia to UTM MAGNA-SIRGAS. It is necessary to use a new script to generate each one of the GeoTiff files with the selected parameters. 355 The command line scripts generate GeoTiff files corresponding to EVI, NDVI, NDVI Quality, EVI Quality, red reflectance MODIS Band #1, NIR retlectance MODIS Band #2, bJue reflectance MODIS Band # 3, and MIR reflectance MODIS Band #7. Note that the MODIS EVI and NDVI data have additional qualitative inforrnation that can also be extracted from the images such as: 0-1 VI Quality (MODLAND Mandatory QA Bits) 2-5 VI Usefulness Index (Indice de utilidad) 6-7 Aerosol Quality 8 Atmosphere (Adjacency Correction) 9 Atmosphere (BRDF Correction) 1 O Mixed Clouds 11-12 Land/Water Mask 13 Snow/ice 14 Shadow 15 Compositing Method These qualitative data may be extracted using the MODIS Land Data Operational Product Evaluation (LDOPE) tools, using SDS commands with 35 parameters for the command line instruction. The processes, especially the MRT Too!, can cause problems by exceeding the size limit on files. To avoid this, one must consider the study areas to be processed by reducing the number of ti les and the extraction parameters. This is especially important for spatial inforrnation with resolutions less than 500m. Results Each HDF-EOS file was re-projected using scripts as described above to generate output files in GeoTiff format with the same names as the input files (but different extensions). The output files correspond to each sixteen day period between 1 January and 31 December 2004. The SOS commands in the LDOPE too! generated masks ofcloud and the amount ofaerosol, which can be combined with maps ofEVI and NDVI. These combined processes were carried out in the ESR1 too! of ArcGIS (Figure 3). 356 Figure 3. Example of an EVI map of Colombia and the same image overlain with a mask of cloudiness corresponding to the first 16 days of January, 2004. One of the objectives of this work is to distribute the images through a website where users are not only able to access data but as well visualize them in a web mapping tool like MapServer. For thls application, the problem of large size of the image files must be eliminated, since this prevents them from being displayed in the common web navigators. The solution was to convert GeoTiff data to JPEG2000 using the GOAL library of MapServer (an OpenSource tool) with kakadu software, which creates images with better resolution and smaller size. All the data were integrated wíthin a collaborative too! and made available to different users through the website http://gismap.ciat.cgiar.org/valle. While this work is in its preliminary stages, it is possible to access sorne images for the Cauca valley. The images available are for the first sixteen days of each month and are available to whoever wishes to access them. Conclusions Having a methodology to process MOOIS images has added greatly to the value of the different research projects in CJAT's Land-use project as well as being able to make the data available to whoever needs them. Without the belp ofthe USGS it would have been difficult to generate the maps and mosaics and to extract information on vegetation índices. Our use of the USGS programs often required modifications, depending on the resolution of the data. the size of the data files to be processed and the inherent limitations ofthe platform used. lt was not possible to give more detail in this paper of the scripts we used. Future use of these scripts would benefit from additional documentation on the types of problems that can be encountered. We found that the method has the advantage that it can be replicated easily and at present we are generating data for all of South America. Moreover, thanks to the experience gained in this exercise, CIA T was able to obtain in native format data for the whole of South America for the years 2000 to 2005 (500 OVOs in HDF-EOS format) and for the whole A frica for the years 2000 to 2006 (more than two terabytes of image data). 357 Acknowledgments Each of the processes was evaluated and carried out by Kelly L. Feistner (USGS) under the supervision of John Dwyer (USGS). This research was carried out as part of a program ofvisiting scientists from UNEP- GRID at the EROS Data Center, Sioux Falls SD. We thank Michelle Anthony and Ashbindu Singh for making the scientific exchange between UNEP and CIA T possible. 358 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 References GSFC/NASA (2005). Especificaciones Técnicas MODIS. URL: http://modis.gsfc.nasa.gov/about/specifications.php Huete, A., Justice, C and Leewen, W. (1999). MODIS vegetation index (MOD 13). Algorithm theoretical basis documentA TBD 13. URL: http://modis.gsfc.nasa.gov/data/atbd/atbd mod 13.pdf U.S. Geological Swvey, 2006. MODIS Reprojection Tool v3 .3a Software. URL: http://edcdaac.usgs.gov/landdaac/tools/modis/index.asp U.S. Geological Swvey, 2006. MODIS LDOPE Tools Release 1.4 Software. URL: http://edcdaac.usgs.gov/landdaac/tools/ldope/index.asp U.S. Geological Swvey, 2006. MODIS Swath Reprojection Tool v2.1 Software. URL: http :/ /edcdaac. usgs. gov /landdaac/too ls/mrtswath/index.asp 359 Identifying Critical Issues to Promote Technical Change and Enhance the Efficiency and Competitiveness of the Beef Sector in Costa Rica Federico Holmanna,b, Libardo Rivas0 , Edwin Pérezb, Paul Schuetzb, Cristina Castroc, and Julio Rodriguezc 8Centro Internacional de Agricultura Tropical CIAT, Cali, Colombia b Internationa/ Livestock Research Institute ILRJ, el- CIA T, Ca/i, Colombia cCorporacion de Fomento Ganadero CORFOGA, San José, Costa Rica This inter-institutional study ILRI-CIA T-CORFOGA was concluded in 2006. During this year, surveys were executed in cattle auctions, slaughterhouses, butcheries, and supermarkets. Based on the information collected, a value-chain analysis was performed in order to detennine costs, risks, and profits along the different segments ofthe beefvalue chain ofCosta Rica. The study is now being edited for publication. Highlights • Costa Rica's livestock and beefindustry performs very unsatisfactorily. • At the farm leve!, beef production systems generate an annual gross income of $3 7/ha in the dual- purpose system to $125/ha in the fattening system. Such gross incomes are extremely low if we take into account that the commercial value of beef fannland ranges between $1 ,000/ha and $2,000/ha. As a result, the gross income cannot recover the opportunity cost ofthe capital invested in the land, making this beef activity uncompetitive. Objectives This study aims to: a) Describe the economic agents of the meat chain in Costa Rica as well as its commercial and legal relationships; b) Identify the inter-relationships between links, technological levels, efficiency indicators, installed capacity (scale), and leve! of occupation; e) Characterize and estimate cost and price structures and the generation of vaJue along the different links of the chain; d) Identity critica) costs that can be modified through technological interventions, policies, or other actions; and e) Determine biological and economic risk factors throughout the chain. A methodology that identifies and determines the costs and benefits in each segment was developed to estímate the generation of val u e in monetary terms throughout the meat chain. Results Costa Rica's meat sector has clearly suffered a downward trend since the mid-80s, with an annual decrease in production of 0.1 % over the past 20 years despite the reduction of the herd inventory, which decreased from 2.3 million head in 1985 to only 1.1 million in 2004. Government investment in the sector fell from 360 5% of the national budget in the early 1990s to only 1.5% at the beginning of the present decade. Total fann credit of both public and prívate sectors has suffered a marked decline. In 1990 it represented 15% of total loans (4% in livestock production) and in 2002 these had fallen to only 5% (1.7% corresponding to livestock credit). Productivity indicators reflect the poor dynamics ofCosta Rica's 1ivestock sector. The annual gross earning per unit area was estimated at US$44/hectare for cattle ranches, at $126/hectare for dual-purpose farms (including income from sale ofmilk), and at $135/hectare for farms where development and cattle fattening activities were carried out. Gross income at these levels is extremely low considering the commercial value ofland on beeffarms, whicb ranges between US$1,000 and $2,000/hectare. The biological inefficiencies of low calving percentages and poor liveweight gains, combined with the high cost of land, hinder the recovery of the opportunity cost for capital invested in the land, making meat-related activities relatively uncompetitive. Because of its low productivity, and based on the assumption that the only cost in cash is for labor, the cattle-raising system pays family labor wages only 60% the legal mínimum. Therefore it is imperative that the public and prívate sectors join efforts througbout the supply cbain to increase the productivity and efficiency ofthis primary sector by facilitating the adoption of improved technologies. Auctions yield a relatively good profit~ however, when analyzed on a calendar-day basis, they are not so attractive because of the Jow use of installed capacity (Table 1 ). A strategy that could pro ve useful to improve the efficiency of Costa Rica's auction system would be to integrate the different events in order to share fixed operational costs. Administrative and operational staff could rotate among existing auctions since their dates are different. This scheme would help reduce fixed costs and the commission charged, without affecting profits but improving efficiency in this link ofthe chain. 361 Table l. Operational characteristics of auctions: type of animals bought or sold, operating costs, and in come. Auction Average !ndicator 2 3 4 5 Y ear of establishment 1997 1993 1984 2001 1993 1994 Commission col!ected (%) 4 3.5 3.8 3.5 4.0 3.8 Jnstalled capacity (# animalslday) 900 500 600 500 800 660 Average no. transactions per 500 390 300 290 750 446 event (# heads) Capacity used (%) 55 78 50 58 94 68 Weekly operation (# of days) 1 2 1 2 1.4 Real capacity used (%) 9.2 26 8.3 9.7 31.3 15.8 Categories of animals bought or sold at the auction (%) • Culled cows 10 60 35 6 30 28 • Weaned calves 25 15 15 20 15 18 • Weaned female calves 20 5 5 9 10 10 • Young bulls for finishing 30 5 10 25 20 18 • Heifers for slaughter 10 10 30 33 20 21 • Finished males 5 5 5 7 5 5 Most frequent distance from the 25 40 30 60 50 4 1 auction to the farm (km) Labor at the auction (# people) • Auction day 32 25 29 16 34 27 • Day without auction 9 9 6 4 12 8 Monthly operational costs1 ($) • Labor 7,440 6,200 5,790 3,500 11 ,363 6,859 • Services 220 240 200 290 240 250 Gross monthly incomé ($) 22,733 36, 151 14,679 12,076 76,048 32,337 Net income per event ($) 3,474 3,433 1,997 1,934 7,437 3,655 Net income per animal bought or 6.94 8.80 6.66 6.67 9.92 7.80 sold per event ($) Net income per animal bought or 0.99 2.51 0.95 0.95 2.83 1.65 sold per calendar day ($) 1 Estímate based on an average cost of US$550/permanent worker, including social benefit costs for days without auction and US$25/day for transitory workers on auction days. 2 Estímate based on the proportion of animals, according to category, that arrive at the auction, number of animals bought or so1d per event, 2005 sale price, and commission collected by each auction. The industrial sector (municipal and industrial slaughterhouses) shows a low occupation of installed capacity, which results in high operational costs and very low labor efficiency. Estimates of total operational costs of slaughter range between US$32 and $66 per animal (Table 2). If the estimated unit costs are compared with the rates collected for slaughtered beef (US$15-$23), municipal slaughterhouses would appear to operate at a loss although industrial slaughterhouses do make a very low margin of profit thanks to sale ofthe by-products (hides and víscera). 362 Table 2. Operational characteristics of several municipal and industrial slaugbterhouses of Costa Rica. Municipal slaughterhouses Industrial Variable slaugbterhouse 2 3 Volume slaughtered (head/month) 45 150 650 7,635 Days of operation per month (#) 17 13 26 26 Capacity of dai1y slaughter (head) 7 50 85 500 Capacity currently used (heads 38 23 29 59 Initiation of operations (year) 1985 2002 1974 1964 Annual proportion of post-slaughter <0.1 <0.1 <0.1 <0.1 rejections (%animals) Origin of cattle slaughtered (%) Small producer 4 NA Medium producer 50 12 NA Large producer 54 NA Butcber's shops 100 50 30 NA S u permarkets NA Others NA Agent ofthe chain that assumes the Cattle Cattle Cattle Cattle owner post-slaughter risks of confiscation owner owner owner Avai1ability of insurance policy (Y es, Y es Y es Y es Y es No) Permanent employees (#)1 3 16 33 757 Productivity of labor (# of animals 15 9.4 19.7 76.3 slaughtered per worker) Operational costs ($/month) Labor 1,650 8,800 18,150 416,350 Electricity 140 1,070 2,525 64,080 Cost of slaughter ($/head)) 39.80 65.8 31.8 62.9 Cost of maquila ($/head) 20 23 20 15 1 Of tbe total number of employees, about 100 work in slaughter-related activities. Tbe retail sector (butchers and superrnarkets) present the best performance in terms of efficiency and profitability. The gross profit margin, expressed as the fraction oftbe fmal price paid by the consumer that remains in hands of the butcher as retribution for hislher work, varies widely from 3% to 40%, with an average of 32% (Table 3). lf these rates of profit are compared with those of otber altemative retail businesses (approximately 8%), then this type of activity generates excellent margins of profit at very low risk. 363 Table 3. Month ly operational costs, break-even point, and profits of urban and rural butcher' s shops in Costa Rica (U S$). Butcher's Shop Variable Urban Urban U rban Urban Urban Rural Rural Average neighbor neighbor ne ighbor market market market market hood rhood h ood Qlace Qlace Qlace .Qlace Workers 22 13 3 24 5 2 4 10.4 1 Labor cost 7150 12100 1650 13200 2750 1100 2200 5735 Energy cost 787 886 886 591 303 290 394 591 Lease of local e 3937 3937 350 295 300 280 1871 Cost of insurance 4000 policy Operational cost 157 158 160 158 157 150 140 154 Beef sales (kQ/month) Total sales of 12031 17144 6633 14299 3505 1840 3014 8351 (kg/montn Breakeven 6495 25980 3464 30310 8660 4243 3810 11852 (kg 12990 43300 4619 43300 12371 7072 6350 18635 Operational cost 3 kg meat sold g) 4500 4500 800 5000 1400 1200 1200 2800 Average cost of ~esse_p carcass 0.93 0.40 1.44 0.33 0.28 0.26 0.47 0.45 vtscera Average sale price 3.06 3.05 3.06 3.06 3.06 3.06 3.06 3.06 kg_~eat for pom 3.99 3.45 4.50 3.39 3.34 3.32 3.53 3.51 Average 6sale price consumerof dressed 4.63 4.63 4.63 4.63 4.63 4.63 4.63 4.63 carcass plus Net earnings per meat sold 0.64 1.18 0.13 1.24 1.29 1.31 1.10 1.12 Net earnings per meat sold 16.0 34.2 2.9 36.6 38.6 39.5 31.2 31.9 a. lncludes all species. 1 Assuming an average cost per month of US$550 per worker, including social benefit costs. 2 No. kg beefthat should be sold monthly to cover operational costs ofbutcher's shop. 3 Calculated by dividing total operational cost ofbutcher's shop by kg meat of all species sold monthly. 4 Calculated on the basis ofthe sale price of one 276-kg carcass at $6 11 by the slaughterhouse to the butcher' s shop plus 16 kg víscera at $35 for a total of$646 divided by 211 kg salable meat (267 kg carcass multiplied by 78% salable meat minus 6% fluid loss). The survey did not ask for this value, but it was estimated on the basis of carcass sales of slaughterhouses. 5 Calculated on the basis ofthe sum of operational cost per kg beefsold plus average cost ofpurchasing kg meat from the slaughterhouse. 6 Estimate based on Table 4 for a young bull and does not reflect the differences ofprices that exist between butcher's shops; as a result, this is an approx.imate indicative. 364 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 The value generated throughout the chain. The percentage of the final retail price of young bulls is distributed as follows (Figure 1): rancher (19%), auctioneer (1%), fattener (34%), transporters (6%), slaughterhouse (7%), and retailers (33%). The distribution of the value generated along the meat chain is completely inequitable and is not consistent with the risk faced by the different actors forming the chain. The inequity observed in the distribution of the added value reflects a clearly dominant position in the market of several actors of the chain, which allows them to capture a very high fraction of the benefits. The generation of value a long the chain ranges from US$0.28/animal per day for the rancher to US$46/animal per day for the butcher. The highest proportion of added value is concentrated at the final end of the chain. The butcher or supermarket obtains 164 times greater value from the same animal in the same time unit than the rancher but faces a lower risk because his/her raw materials, equipment, and infrastructure are usually covered by insurance policies. 6% 1% 20% 33% • Fattener Butcher's shop O Breeder O Slaughterhouse • Transporters O Auction Figure l. Value generated throughout the chain as percentage of the final value of a fat young bull at retailer price. The competitiveness of this meat chain is the sum of the efficiency and productivity of all the links that form it. A weak and rather poor demand for beef at the fmal link of the chain hinders the adoption of technology in the primary link of the chain, so it becomes a vicious cycle that generates low productivity and competitiveness. The low demand for beef implies reduced levels of slaughtering, which impedes the full use of the installed capacity of slaughterhouses and processing plants. This, in tum, hinders the generation of economies of scale and causes high unit costs that reduce the competitiveness of meat products in both domestic and foreign markets. To promote technical change and enhance the efficiency and competitiveness of the value chain of Costa Rica' s beef sector, we recommend the following: 365 (a) Leam from other chains, for example the poultry chain, by identifying actions that could improve the meat chain; (b) Milk breeding cows when a mil k market exists as a mechanism to in crease family income beca use wages are currently below the legal mínimum; (e) Promote the creation of livestock funds as a mechanism to create social capital, reduce transaction costs, and improve chain productivity and profitability; (d) Promote massive adoption offorage species with an emphasis on summer feeding to reduce weight losses of the national herd, improve farm profits, and promote modemization through the adoption of improved technologies; and (e) Establish a standard systems for beef cuts based on quality and price, allowing the differentiation of o:ffers for different consumer groups, among others. 366 1 1 1 1 1 1 1 1 1 1 1 New systems of agricultura) production and environmental services: an economic evaluation in the Altillanura 1 of Colombia Libardo Rivas, Federico Holmann y James García Centro Internacional de Agricultura Tropical CIAT Abstract In the general context, the Ministry of Agriculture and Rural Development (MADR) of Colombia proposes a mega project The Rebirth of the Altillanura of the Orinoco with the main objective to generate environmental services associated with the fixation or sequestration of carbon to mitigate the effects of progressive global warming. In this mega project, large-scale sowing of plant species that fix carbon in the aerial parts (foliage) and in the roots is proposed as a commercial product in intemational markets. The proposal calls for establishment oftrees on 6.3 million hectares over a period of20 years. It is expected at the end of this process 5 million people will be settled in the region and 1.5 mi Ilion jobs created. The total cost of the agricultural component of the project is estimated as US$15 billion dollars, aside from the investment required for physical and social infrastructure and public services. This initiative provides the setting for the natíonal policy for productive forestry development, which counts Forestry Incentive Certificates (ClF) among the main instruments to stimulate the sector. This policy intends to stimulate employment and the national offer of forestry products at the same time as generating environmental benefits associated with the control of eros ion, conservation of water resources, carbon fixation, reduction of tree felling and pressure on natural wood lands. Within the global framework, this study evaluates new farming system models with an economic, social and environmental focus. The new models include diverse components of grazing, agriculture and forestry for the production of food and primary products and, in addition, environmental products such as the sequestration of carbon. Introduction An analysis of new systems of agricultural production in the Altillanura of Colombia, which included various technological components developed by CLA T and its collaborating institutions, started in 2005 was finished in 2006. Given the imminent signing and entry into force of the free-trade agreement (ITA) with the United States and the possible effects of this agreement on prices and national agricultural production, various altemative scenarios where added to the study. These included the effects of falling agricultural prices and increases in productivity on net incomes, production and land use in different farm models in the Altillanura. Outstanding aspects • The incorporation of new technological elements into the traditional ranching systems of the eastem plains is a powerful tool to encourage regional agriculture and at the same time to conserve and improve the productive capacity of the soils. • Crop-pasture rotations within a strategy to create arable layers2 significantly improve the performance of current production systems in terms of productivity, profitability, cash flow and generation of employment. 1 There are basically two broad ecosystems on the eastem plains ofColombia, the poorly-drained and the well-drained savannas. The Meta River, a tributary ofthe Orinoco, runs east from the Andes and divides the poorly-drained savannas to the north from the well-drained savannas to the soutb. The term Alti//anura refers to the southem, well- drained savannas. To avoid a contrived translation, we continue to cal! them the Altillanura. 367 • An outstanding aspect of the intensification of productive systems of the region is the increase in their capacity to generate employment and contribute to social objectives such as improved equity and poverty reduction contrasted with traditional systems, which bave low capacity to offer employment. • The objective of this ex ante evaluation is to generate relevant information to support investment decisions in the public and prívate sector. This will permit implementation of sustainable and competitive development of the region with high economic, social and environmental impact. Techniques based on linear programming were used to address the basic economic problem of efficient allocation of scarce resources between multiple altemative uses. • The inadequate regional road system impedes the transport of machinery, agricultural inputs and products and makes transport more expensive. The scarcity of agricultural machinery and the difficult to rent it during seeding and harvest present a serious bottleneck, especially for less well-off producers. Methods The theoretical model has the following form: Maximize Z = CX subiect to" AX ~ b · X1 ,X2······Xn ~ 0 · where Z is the obiective function which in the ' J • ' ' J , present case is defmed as the net total benefit resulting from the implementation of diverse production options at the farm scale. Vector C contains the coefficients of net retum for each of the generated products in vector X . The modified model used in the study increases the evaluation period to nineteen years and limits the analysis to the altematives of cattle ranching, agriculture, forestry and carbon sequestration by different vegetative covers. The following variables are included: l. Decision altematives, also known as activities, which are under the direct control of the producer and comprise the production plan of a particular farm amongst which are crop-pasture rotations, purchase of inputs and sale of products, obtaining credit and the use of cash flow over time; and 2. Interna! (endogenous) variables and restrictions, whicb include all those variables resulting from the intemal functioning ofthe model and the economic, technical and environmental restrictions. Among productive activities that are reasonable to develop in the Orinoco region of Colombia are the following: l. Distinct altemative uses of land resources and how their efficient use can generate social economic and environmental benefits; 2. Potential uses of land in ranching, crops and reforestation or natural woodland; 3. The altematives that generate commercial products such as meat, milk, timber and environmental services sucb as the sequestration of carbon by pasture and woodland; and 4. Benefits derived from economic social and environmental land-use at the level of farm, region and country. 2 The arable !ayer strategy aims to use management techniques such as stubble mulching and minimum till to create and maintain a !ayer at the soil surface that is physically and chemically apt for cropping. URL: www.ciat.cgiar.org/riceweb/memorias/d molina.pdf. 368 For the evaluation of ranching we consider two cattle systems: l. Dual-purpose, wh.ich includes the production of meat and rnilk, emphasizing milk production by improving the productive capacity of cows through the incorporation of genes from milk breeds and feeding the herd a high quality diel 2. Breeding beef cattle, which is the first phase in cattle production directed towards the production of store-condition cattle for farms that specialize in cattle fattening. Forage on offer for the cattle can come from the following alternatives: l. Pastures alone either native savanna pasture, improved Brachiaria and the association B. decumbens-legume; and 2. Pastures in rotations with crops within which we consider several rotations. Rotation 1 starts with a period of seven years of native savanna followed by a cycle four years of biennial crops rice- soybean and maize-soybean in rotation, finishing with a pasture association B. decumbens-D. ovalifolium that remains productive for eight years. Rotations 2 and 3 are similar, both starting and finishing with a cropping cycle of six years each and including an intermediate phase of improved pasture which for rotation 2 is improved Brachiaria and for rotation 3 a pasture association of B. decumbens-D. ovalifolium. The pastures alone are evaluated for a period of nineteen years with renovation in the eighth and fifteenth years. The forestry component consists of a plantation of Caribbean pine, which produces timber and environmental services in the form of carbon sequestration. To improve the soil physically and chemically, an improved grass pasture is sown initially, which remains in production for four years before establishing the pine plantation. For the analysis, progressive incorporation of new technological components in the current grazing systems is simulated by constructing diverse sequential scenarios. lt starts with the grazing model, which can be either breeding or dual-purpose, based on extensive use of pastures alone. In the subsequent phase the model adds a component of rotations between pastures and crops in a process oriented towards progressive soil improvement through the creation of arable layers. The incorporation of trees and sale of environmental services in the form of carbon sequestration is represented in the following stage of the transformation pathway towards productive systems. Finally, to evaluate the impact of economic policy on production systems and on land-use, we constructed scenarios in which the production systems, besides including new technological components, are assisted by promotional policies such as Forestry Incentive Certificates or through schemes of advance payment for environmental services. The model considers a farm of five hundred hectares operating with average regional costs and with operating capital that can vary in the range US$5000 to US$300,000. Results The results show, amongst other possibilities for the region, that: l. lncorporation of new technological components into traditional ranching systems of the upland eastem plains gives a significant increase in farm net retum, in employment, production and productivity. In the traditional ranches, as rnight be expected, the use of improved pastures is controlled by the availability of fmancial capital. The incorporation of crop-pasture rotations excludes native savannas from the optimum solution and, on improving cash flow, facilitates the expansion of improved pastures in farms with fewer econom.Ic resources. 369 2. Breeding, compared with dual-purpose systems, represents an inferior leve! of technological development, and for this reason the introduction of improved technologies has a relatively greater economic effect. 3. Establishment of forestry plantations for sale of timber and the capture of carbon is more likely to be adopted on farms currently devoted to breeding. The simulation exercise showed that trees could enter these systems at all levels of available capital, and as such is a promising alternative to breeding, which is generally relegated to areas that are isolated and far from markets. The latter emphasizes the need for state investment in roads and transport infrastructure with parallel development of complementary services, especially those for processing, managing and commercializing forest products. 4. Technical progress significantly improves net farm in come, the objective function of the model, but especially in those with less available capital. For example, the implementation of crop-pasture rotations in breeding systems increased by 1.8 times the net income on farms with high available capital but by six-fold in those with greater financiallimitations. 5. The intensification of productive systems increases the capacity to generate employment, which constitutes a relevant impact in achieving the social aims of equity and poverty reduction. 6. The CIF forestry promotion policy has a greater impact in production units with better endowment of operating capital in that it permits them to increase their land under forest. When operating capital falls below US$20,000, the impact is nil. 7. The price of carbon in the international market at present is low and according to sorne experts will remain stagnant un ti! 20 12. On-fann research and ex ante economic studies have demonstrated the feasibility of new technological options. Nevertheless, the process of adoption of new technologies in the region still does not have the necessary dynamism that permits an observable impact in production, productivity, employment and prices. Agriculture in the Altillanura is risky since it confronts numerous technical, economic, physical infrastructure and social restrictions such as documented in the rapid appraisals carried out by CIA T in 2004 and 2005. Many of the producers interviewed see the entry into force of the FTA as a menace that would increase risks to regional agricultura! production. Nevertbeless, this bilateral agreement also offers great possibilities for the production of meat, milk and its products, fruit and forestry, activities that are well-suited to the resource endowment of the Altillanura. In a scenario in which grain prices fall more than t 10%, the contribution to net in come by crops in the rotation would become negative. In spite of this, rotation 1 (native savanna-crops- improved pasture) would continue to be profitable owing to its strong beef cattle componen t. Research to improve the yield of current crops and the search for new options of highly-productive, adapted crops to establish rotations with pastures, are alternatives to confront the economic risks of prices and the objectives of the FT A. lt is expected that a period of growth supported by policies of investment in physical and social infrastructure, state programs of credit and compensation to those sectors affected negatively by the bilateral treaty, grain production will continue to be economically viable in the uplands. For the advantages of participation in a broad and high-value market and to take maximum advantage of the natural resource endowment of the Altillanura, it is necessary that the nation underwrite integral programs of development that, besides promoting new alternative technologies at the farm scale, appües appropriate policies to overcome the restrictions that limit technical advance. 370 Reference Rivas, L., Holmann, F. and García, J. (2006). Nuevos Sistemas de Producción Agropecuaria y Servicios Ambientales: Una Evaluación Económica en la Alti/lanura Colombiana. Centro Internacional de Agricultura Tropical (CIAn and International Livestock Research Institute. (ILRI), Cali, Colombia. Working Document No. 204. 60pp. 371 Communities and Watershed project During the wind-down phase ofthe Communities and Watershed project, focus has been on the impacts of climate and land-use change on watersheds. The goal of the team during this time will be to contribute to the understanding of how agro-ecosystems are responding to climate change and land-use impacts with a focus on water resources. The strategy for research on water resources and adaptation to climate and land-use change is to compare responses of agro-ecosystems components such as soils, water and vegetation under different climatological and land management practices, using the watershed as the unit of analysis. This will be accomplished through a combination of CIA T-based research activities and the participation of local, regional and national partners to make research results relevant, applied and used for decision making. Strategic partners in this process include national organizations such as IDEAM, Ministry of Education; regional environmental agencies such as CVC and CRQ; local partners such as municipalities, water districts and farrner communities; universities such as Universidad del Valle - CINARA, University of British Columbia and NGOs such as CARE, ECOPAR and Agua Bolivia. Our current special projects and those under development in research and education are complementary in this context. They contribute to the scientific knowledge base on impact and implication of climate and land-use change, and the uptake of this information by users. We see the educational and institutional components as critica! to information management and to the assimilation of research results for development and decision-making. Continued research is on water availability and hydrologic response as well as the options available for water-efficient technologies relevant to local conditions. ln addition to the monitoring of established networks, we undertake, baseline surveys on the use of water, land, and resources and analysis in pilot watershed. Through these activities we make available to water-user associations data on water quality). The pilot watershed in which we work in Colombia are Garrapatas and El Dovio in the Valle del Cauca, Barbas watershed the Central Cordillera. We also work in Tiquipaya watershed in Bolivia 372 Protocol for the characterization of carbon and water in high mountain ecosystems Clara Roa~ Sandra Brown~ Maria Cecilia Roa~ Jorge Alonso Beltran~ Luz Dary Yepei, Jorge Luis Ceballoi, Fernando Salazarb y Cesar Buitragob °Centro Internacional de Agricultura Tropical, Cali, Colombia b Instituto de Hidrología, Meteorología y Estudios Ambientales, Bogotá, Colombia Partners: CIAT, IDEAM, UBC, World Bank and GEF A protocol for the monitoring of carbon and water cycles was developed to acquire a scientific information base about the processes that affect the cycles, their interactions, dynamics, variability, the practices that optimize storage, the effects of land use, vulnerability to perturbations and potential impacts. The conceptual model focused on the pools and flow paths and the potential impacts of climate and land use change on the compartments ofthe cycles in high elevation ecosystems (Figure 1 ). P. Pmii:Jbllon E• EvapcnUon Qo- Glatlal ftOIOI N•Fog ET• EvliPOtrii'IIPII1IIIOn G• PtrtoUIIDn (OfOIIIdWaltll) Q• ~ernc.t now C fiO ltlman COIIIUnQilO domesllc AL.= VIJ1ua1Mter- mlk AA= \1111Ual WOI!er- lbod A e = VitUII w¡ter- meat Q G l ' - Solla Figure l. Pools and flows of water in páramo witb human intervention. Anthropogenic activities such as ploughing, burning, forest harvesting and grazing have impacts on the water and carbon pools and modify the natural cycles through changed vegetative Jand cover, introduced animals, and human consumption ofwater and biomass. An algorithm was developed incorporating the selection of sub-watersheds and ecosystems to monitor and researcb relevant questions, the collection of secondary data, design of the monitoring network, installation of the monitoring program, data collection, systematization and analysis of information (Figure 2). Each block in the chart represents a flow of activities, analysis and decisions taken. Norms and criteria, 373 procedures, formats for data capture and instructions for the measurement of variables are associated with the blocks. Blocks S.IKtlon of rnearch qu .. tlona Collectlon of aecondary data through 1 partlclplltory proc .. a Dealgn of the monltortng networtl Monltorlng program Syatematlzlltlon, analy .. a of lnformatlon Algorfthm v .. l Figure 2. Algorithm ofthe protocol. The selection of the sub-watershed and ecosystems to monitor aims to account for the relative importance of the high elevation ecosystems and their variability, and logistics for implementing monitoring. The research questions are defined based on the characteristics of the selected micro-watersheds for monitoring and for comparison between ecosystems. The collection of secondary data is done in a participatory process, which includes: • Identification and involvement of interest groups; • Creation of a Jearning alliance; • Validation of indicators at the locallevel; • Assignment of responsibility for data compilation; and • Creation of metadata and quality control. For each selected sub-watershed and the re levant research questions, the monitoring network is designed in accordance with priorit ization of variables (Figure 3). Variables are prioritized based on a weighted index 374 that includes: contribution toa flow or compa.rtment within the cycles and scientific understanding ofthese processes, complexity of measurement, and equipment cost(s). Data collection includes information on biophysical, socio-economic and geographic variables. Figure 3. Schematic monitoring design for non-intervened páramo The monitoring program provides the detail for the installation of equipment and procedures for the collection of data corresponding to each selected variable. The systematization of data refers to the collection and transfer of information to a central database, and the analysis and synthesis of monitored variables to consolidate information at the project scale, for the carbon and water cycles. 375 Youth Bolivia: alliance for water-science and the future CGIAR- CIDA Canada Linkage Fund Sandra Brownn and Elena Cordero °Centro Internacional de Agricultura Tropical, Cali, Colombia Partners: Communities and Watersheds, CIAT Colombia; Institute f or Resources andEnvironment, UBC Canada and CGIAB, Agua Bolivia Achievements y activities 2006-07 Leadership The second leadership workshop consisted of two parallel programs: one for fac ilitators and one for young people. The objective of the facilitators ' workshop was to build capacity among teachers, Asiritic (the irrigation users' association) members and adult facilitators, to understand and encourage the development of leadership abilities in youth, to fac ilitate leadership action, exchange experiences, and generate new ideas. Teachers working in the high altitude sections of the Andes in general had little experience in oral expression, and members of Asiritic tended to lead youth involvement rather than facilitate processes. The workshop focused on methodological tools for young people to work through the steps of a research project including: problem trees, tlow charts, stair diagrams, chronograms, activity lists, questionnaires, strength and weakness analysis, and participatory appraisal. The youth workshop focused on creation of self awareness, affiliation within the group, self-confidence, goal setting and teamwork (Figure 4). The exercises reinforced public speaking skills, leadership roles and responsibilities, group participation, team work, and the process of designing and conducting a research project. The workshop aided in the development of individual skills, the generation and expression of new ideas, integrating youth from the cordillera and valley, and further involving female youth within the group. 376 Water rights, customs and use The irrigation systems ofMachu Mit'a, Lagun Mayu and Chankas and their use were mapped and analyzed to understand further the history of the traditional irrigation systems, water allocation within individual communities, and water use and customs with respect to irrigation. Youth "regained" knowledge of the irrigation systems, their use and potential limitations within their own communities (Figure 5). This sub- project permitted the youth to reflect on the socio-economic and environmental challenges associated with the timely distribution of water of good quality. Water consumption Water consumption in the cordillera (>4,000 m) is very low dueto socioeconomic conditions, climate and cultural practices. Despite all families having access to water either from the piped network or prívate springs, they do not have flush toilets or showers and clothes are always washed in small streams. The animals drink water from natural sources: streams, khochas or bofedales. By contrast in the valley (urban and rural Tiquipaya), water consumption is significantly higher (Figure 6). Consumo Domestico de Agua 60 50 40 Uhablda 30 20 10 o • Tlquipaya Urbano • Tiquipaya Rural e Cordillera 377 Food security Tbe production of vegetables in protected environments was undertaken as an alternative for food security in the Cordillera (4100-4200 m). A tunnel type greenhouse (7.5 x 4 m) appropriate for regions witb strong winds was constructed communally at the school in Titiri (Figure 7). Organic seed beds (2m x 4 m) with agro-film cover were also constructed with the local youth in Titiri and Totora. The first seeding included lettuce, onion, cabbage, carrots, beets, achojcha, locoto, tomato, and aromatic herbs. The greenhouses provide the possibility for the youth to generate small amounts of income and expand lettuce production targeted for market, taking advantage of the good water quality in the CordiiJera (poor water quality used in the valley to irrigate lettuce has been associated with human health risks). Social forum The "Social Summit for the Integration of Communities" was held December 6 to 9, 2006. Exhibitions included water resources, gender, sovereignty and integration with the participation of indigenous, women, youth and small producer groups. As part of the exposition, the youth project along with Agua Sostenible (Sustainable Water) presented their activities related water use and water customs. Soils and water holding capacity The water-holding capacity of soils in high elevation ecosystems is being investigated in relation to downstream availability of water. Steps in the process include: delineating watersheds, inventory of khochas, characterization of khochas, wetland coring, wetland age determination, and determining soil organic matter and downstream flow. The watershed delineation and inventory exercises use detailed (0.7 m resolution) imagery. Youths, a GIS student and indigenous farmers from La Cumbre worked together to delineate natural boundaries and canal systems (Figure 8). Khochas are being inventoried and characterized using the imagery together with field verification. Relic wetlands were cored and samples taken for 14C dating and organic matter composition (Figure 9). Downstream water flow in the canal system was analyzed using salt tracer/conductivity measurements and flow rates were determined from the time required for small reservoirs to fill. 378 379 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Role of Andean wetlands in water availability for downstream users, Barbas watershed Colombia Maria Cecilia Roa Centro Internacional de Agricultura Tropical, Cali, Colombia Partners 1 Donors: Municipio de Filandia, Quindio, CRQ, UBC, IDRC, USAID, IFC Small, high-elevation headwater catchments are the preferred source of water for a large portion of the Andean population, due to less impact of human activities on the water quality and the advantage of elevation for water distribution. However, the lack of process-based water balances has implications for water allocation and the protection of ecosystems that play an important role in water flow regulation. Decisions about water allocation for human use impact ecosystems and decisions about upland ecosystem management impact the availability (quantity and quality) ofwater for human use and for the maintenance of downstream ecosystems. In view of the increasing water demand, there is an urgent need to use a more scientific approach to provide evaluations of water supply and demand. Water demand for irrigation in the productive Cauca valley is large and increasing, as well as increased demand due to urban growth. In many headwater areas, forest.s are being converted to grazing land, small wetlands are being drained. However, the effects of these land-use changes on hydrology are not well documented. The headwater catchments of the Barbas River in Quindio department were selected as a case study because wetlands have been relatively preserved from drainage or have been restored. Moreover, there is interest by the regional environmental agency, Corporación Autónoma Regional del Quindío (CRQ), to understand the contribution of these wetlands to the annual water balance. CRQ also wishes to acquire rnanagement tools that can be incorporated into the watershed management plan particularly for the conservation of key water source areas. The catchment of the Barbas River has a strategic importance for the provision of water to rural and urban communities both up and downstrearn (Figure 1 0). Its headwaters are located in an area classified as very humid in terms of annual precipitation (2,800 mm). Considering that the highest altitude of the catchment is only about 2,300 meters, the amount of rain in the headwaters is higher than reportes at other climate stations at similar altitudes in the Central Cordillera. Age of Barro Blanc wetland (C14) 6000 5000 • 4000 • • ~ "' 3000 ~ 2000 - 1000 •• o o 100 200 300 400 Depth (cm) 380 The terrain is undulating and can be described as hummocky. There are many small flat and concave areas between steeper hills, which with the high precipitation have turned into wetlands. Individual wetlands in the upper Barbas average less than one hectare with a total area of 57 hectares (CRQ, 2001 ). A detailed land-use map was made using a Quickbird satellite image with resolution of 0.6 meters. Subsequent field survey showed that the wetland are a in the headwaters of Rio Barbas covers 7.2 hectares (Figure 1 0). The water storage capacity of wetlands in each catchment was estimated using the land use map. The Quickbird images of three selected catchments was printed at l: l ,000 scale for use in the field to mark polygons of land use, which were later digitized in ArcGIS. For each of the selected wetlands we described the area, water source, water outflow, relative location in the catchment, type of protection, dominant vegetation, duration of saturation and surrounding land use. Assuming that each monitored wetland was representative for their catchment, we calculated the spatial distribution for all the wetlands for each. We then calculated for each hydroperiod (wet season and dry season) the dynamics ofwater storage in wetlands throughout the year. ln collaboration with the University of Geneva, Switzerland, we assessed the age of the wetlands using carbon 14 dating. The data show that the organic matter accumulation over the impermeable clay !ayer has occurred over the Iast 5,000 years with a rate of accumulation of 0.6 mm of organic matter per year. This suggests that these ecosystems are highly vulnerable to changes in precipitation parteros or to temperature. Analysis of the water balance was undertaken at the wetland scale to understand the response of water storage to precipitation, surface runoff, sub-surface runoff and evapotranspiration. We used water leve! recorders both inside and in the surface outflow of the three representative wetlands. Together with the wetland profiles and the porosity ofthe wetland soils we made an analysis ofthe wetland water storage and its fluctuations throughout the year. The following field measurements are on-going: Water level in the wetland: An Odyssey0 capacitance water leve! recorder was installed in each of the three monitored wetlands to record continuously changes in water leve! (Figure 13). These automated measurements are complemented with manual measurements. Six pipes were installed in each wetland to measure water leve! changes spatially and temporally. These measurements are taken every 2 weeks starting from 30 September, 2004 using a wooden ruler that is inserted into the bottom of each pipe. The water leve! is recorded from the mark that water leaves on the ruler. 381 Wetland volume: Each wetland has a particular topography, which detennines the water storage capacity. For each wetland, a 3-dimensional model was generated using Surfer© and depth measurements taken with an extended Russian peat corer. These measurements were taken across transects at fixed distances. The depths ofthe organic matter layer ofthe wetland soil were entered into the program together with their relative location. Water storage capacity: Water content of each wetland soil was measured using a number of soil samples that were analyzed for water content and soil porosity . These data were then extrapolated to the wetland volume calculated with Surfer. Using the 30 model, the equivalent water content ofthe calculated wetland volume, and the measurements of water depth and wetland area taken every 2 weeks, and the wetland water storage calcu lated. This infonnation shows the response of the wetland to seasonal variations in precipitation, or hydroperiod (Figure 14). Once the wetland water balances are completed, the water balance will be done at the catchment scale and particular attention will be paid to dry seasons when water scarcity is experienced downstream by the municipality of Filandia, both in rural and urban areas. The comparison of the wetland outflow to the catchment hydrograph will show the influence ofwetlands on base flow. 382 Reference Corporación Regional del Quindío - CRQ, 200 l . Estudio hidrológico de los humedales cuenca superior Río Barbas, Municipio de Filandia, CRQ, Annenia, Quindio. 383 Youth, leadership and research: Improving education for development 1 Clara Roa, Enna Diaz, Jorge Alonso Beltran and Maria Cecilia Roa Centro Internacional de Agricultura Tropical, Cali, Colombia Part11ers 1 Domw rs: INTEP, CERES El Dovio and CERES Barragan, municipalidades Genova, Versal/es and El Dovio, Kellogg General objective Improve tertiary education in rural areas by he lping in the processes of natural resources research that contributse to sustainable management and improvement in the population 's quality o f life. Specific objectives • Create a systematic methodology for research in natural resources management in the Regiona l Centers of Higher Education (Centros Regionales de Educación Superior CERES) in El Dovio and Barragán. • Guide research in natural resources management and water resources in the chosen watersheds. • What kind trustworthy data of th is research o ver 1- 1/2 years. Project advances The project is working in three s ites, where they have incidents in the rura l youth population of the Colombian Andes: Versalles and El Dovio, Va lle de l Cauca (CERES El Dovio) and Genova, Quindio (CERES Barragán). Selection ofwatersheds to monitor and the research themes Watersheds of interest for research have been defined through workshops where criteria for the selection of these watersheds has been evaluated and they ha ve defined the research themes of interest focused on water resources and climate change. These workshops were conducted with the participation of the CERES students, the municipa lity and other local organizations (Table 1 ). Figura l. Taller de inducc ión para e l manejo de programas de los diferentes equipos 1 This art icle is not BP2 research but is a contribution from the now-defunct project PE3. J!< l Table 1. Micro watershed selection criteria Criteria S1111tegic importance (benefits to user populations, water volume, dependen! economic activities) Existen ce and % of strategic ecosystem Productive activitieslimpactslland use Viability of comparative study Community and institutional participation CERES-Sena students Available information Total Points: High 3 Medium 2 Low Weight 7 6 5 4 3 2 Micro watersheds San Gris Rojo Juan 2 3 2 2 3 2 3 2 2 3 3 3 3 2 2 3 1 2 3 2 Table one shows the priority selected in Genova for the watershed should in the area. Classification San Gris Rojo Juan 14 21 14 12 18 12 15 10 10 12 12 12 9 6 6 6 2 2 2 3 2 70 Tl 58 The research themes chosen for the work in micro watersheds of the three municipalities are related to the following themes: Water resource and demand What variations in the water resource can be identified in the micro watersheds and how do they affect bene.ficiaries? What is the difference between the wooded micro watersheds and high moor with and without intervention with respect to regulation of the water resource? What is the micro watershed water resource during the year in the higher parts? What is the water consumption in the high, medium and low parts? Characterization of the vegetation of representative ecosystems (wet/ands and cloud forests). Geo-referencing the location of water sources land-use homesteads in micro watersheds and sub watersheds Water qualitv What is the difference between the wooded micro watersheds and high moor with and without intervention with respect to water quality? Human interventions How to make people aware of and conscientious about and the high and the low parts for protection of water? Starting with the results ofwater consumption, what are the practices in water use in the different zones (high medium and low) and according to the productive activities and in what form is it possible to make the population conscientious about it? 385 Environmental standards What other current regulations regarding the supply of water and protection of its sources? Does the population know the legislation that concerns it? Utilization? Benefits? Is it profitable for a countryman to protect a water source of a mínimum of 1 ha by applying an incentive of a reduction in land tax? Sustainability of CERES Is CERES sustainable with its current resources? What measures need to be taken in case it is not sustainable? Choice of micro watersheds Once the watersheds to monitor and the research questions were determined, field visits will made for the selection ofmicro watersheds that would be unable to answer the questions. Definition of the methodo/ogy for use by the CERES students The economic, training and supervisory help that CIAT will provide the CERES students who participate in the project was defined, as well as the help (advice, follow-up and academic validation) they will receive from the teachers and the university. Equally the research methodology that must be followed, including the presentation of reports, was defined. Youth invo/vement in each research theme Twenty four young people were selected from the CERES (9 from CERES Barragán in Genova and 4 from CERES in El Dovio and Versalles) to start the research process taking questions that can be resolved initially. These young people make presentations on the specific theme of a preliminary project demonstrating the objectives in the methodology to be used for the research. Training workshops were carried out however the themes involved in the research. Purchase and instal/ation ofthe moniforing equipment Monitoring equipment was installed in each of the micro watersheds (Table 2 and Figures 2 and 3). 386 Table 2. Equipment installed in the project. (MWS = micro watersheds). Equipment Leve) meter 1.5m Ruler level meter I .Om Leve) meter 2.0m Ruler leve) meter I.Om Pluviometer Climate station Pluviometer Temperature Relative humidity Solar radiation Pan evaporation Max/min temp Anemometer Manual rainguage Rio Gris Rio San Juan Versalles El Dovio Below . El Espejo MWS MWS La Genova El Ret1ro El Tapón high Below MWS wetland woodland Genova Juntas Patuma Esperanza intake moor 2 2 387 3 2 3 2 Total 12 8 2 2 6 Figure 2. Location ofthe measuring equipment installed in the micro watershed Patuma at Versalles W ~ Rainguage Odyssey Ruler 388 Pluviometer in the middle part and Odyssey in the creek below Figure 3. Monitoring in the micro watershed El Retiro, in the Río Gris watershed. Achievements ofthe project • The CERES students presented corrected written proposals for the development of the research questions set out above. • Fieldwork related to each ofthe proposals was initiated and continues. • Data re being collected from each of the monitoring units installed in the micro watersheds. 389