PEOPLE AND AGROECOSYSTEMS RESEARCH FOR DEVELOPMENT CHALLENGE (PA RDC) ANNUAL REPORT 2008 Outcome Line: Agroecosystems Resilience 1 TABLE OF CONTENTS ANNUAL REPORT 2008............................................................................................................. 4 PEOPLE AND AGROECOSYSTEMS RESEARCH FOR DEVELOPMENT CHALLENGE (PA RDC).................................................................................................................................... 4 Outcome Line: Agroecosystems Resilience ............................................................................ 4 1. Outcome Line Logframe ..................................................................................................... 5 2. Outcome 2008 - Targeting of high-value crops to environmental niches through a supplychain framework ................................................................................................................... 10 3. Achievement of Output Targets for 2008 .......................................................................... 12 4. Research Highlights 2008................................................................................................. 14 4.1 Technology and impact targeting: .......................................................................... 14 4.2 Equitable and sustainable use of ecosystem services: Ecosystem services and poverty alleviation in the Andes/Amazon..................................................................... 14 4.3 Climate change: The changing geography of agricultural suitability.................... 14 5. Description of one project outcome.................................................................................. 15 Targeting of high-value crops to environmental niches through a supply-chain framework ..................................................................................................................... 15 6. Publications ...................................................................................................................... 17 Articles in refereed journals.......................................................................................... 17 Books and monographs................................................................................................. 18 Book chapters................................................................................................................ 19 Papers presented at formal conferences and workshop with external attendance ........ 21 Articles in international newsletters or other scientific series ...................................... 24 7. Funded project proposals ................................................................................................. 26 8. Project Staff (*Left during 2008)...................................................................................... 28 9. Abstracts, papers and reports ........................................................................................... 29 9.1 Technology and Impact Targeting (1657 kb) ....................................................... 30 Participatory Impact Pathways Analysis: a practical method for project planning and evaluation ....................................................................................................... 30 Participatory Impact Pathways Analysis: A Practical Application of Program Theory in Research-for-Development .................................................................. 39 Selecting sites to prove the concept of integrated agricultural research for development.......................................................................................................... 65 Presión de la Sigatoka Negra y Distribución Espacial de Genotipos de Banano y Plátano: Resultados de 19 Años de Pruebas con Musáceas.................................. 87 Identifying candidate sites for crop biofortification in Latin America ................. 97 9.2 Sustainable and Equitable use of Ecosystem Services (415 kb) ........................ 119 Challenges to Managing Ecosystems Sustainably for Poverty Alleviation: Securing Well-Being in the Andes/Amazon. Situation Analysis prepared for the ESPA Program. Amazon Initiative Consortium, Belém, Brazil......................... 119 Paying for avoided deforestation in the Brazilian Amazon: From cost assessment to scheme design ................................................................................................. 123 Is soil carbon sequestration part of the bundle of ecosystem services provided by conservation agriculture in the Andes?............................................................... 149 2 Ex ante Analysis of Legumes: The Dilemma of Using Legumes as Forage for Animal Nutrition during the Dry Season or as Green Manure for Soil Improvement ....................................................................................................... 156 9.3 Climate change and risk (804 kb) ........................................................................ 177 The Impact of Climate Change in coffee-growing regions ................................ 177 Global Impacts and Implications of Climate Change on Banana Production Systems ............................................................................................................... 181 Helping small-holder farmers to manage drought risk through insurance: A case study of dry bean production in Honduras.......................................................... 194 Risk sharing through insurance: Weather indices for designing micro-insurance products for poor small-holder farmers in the tropics ........................................ 214 Identifying the Role of Crop Production in Land Cover Change in Brazil, 19902006..................................................................................................................... 236 A Framework for Assessing the Impact of Agricultural Drought in Developing Countries ............................................................................................................. 238 Are Crop Wild Relatives a Useful Source for Genetic Traits Related to Abiotic Resistance in the Context of Climate Change?................................................... 239 3 ANNUAL REPORT 2008 PEOPLE AND AGROECOSYSTEMS RESEARCH FOR DEVELOPMENT CHALLENGE (PA RDC) Outcome Line: Agroecosystems Resilience Introduction The tropical world is characterized by considerable variation, at all scales from community to the region. Institutions at all levels from village to region tend to be numerous, and at varying levels of effectiveness, inclusiveness and governance. Small farmers’ livelihoods range from nearsubsistence to small scale commercial (although pure subsistence is less common than is sometimes thought), and households may seek or have opportunities to emerge from poverty in ways that differ according to composition, agroecological situation and socioeconomic circumstances. Development and research practitioners need tools that enable them to work at different scales, and to discriminate effectively among rural populations and environments. Outcomes tailored to specific social and biophysical contexts are needed to achieve widespread impact under these conditions. Many of the most appropriate tools will be interdisciplinary in nature, and in general need to be derived through iterative interdisciplinary research processes. Agricultural science practice cannot be successful if it is disconnected from development practice, and some of these research processes need to be embedded in development (research for development) in order to yield robust and international public goods. This project (outcome line) is new to CIAT’s portfolio of projects for 2008. It was established in late 2007, taking on components of the Markets, Institutions and Livelihoods project. The project is among the smaller ones of CIAT, and currently consists of two outputs: 1. Institutional arrangements and mechanisms for targeting, increasing and evaluating impacts 2. Policy guidelines, tools and innovations for adaptation to risk, high stress and vulnerability. The emphasis of the project is on process-based research which supports other research activities within CIAT and with external partners (including CPs), with a thematic focus on generating better understanding of water-related processes and issues surrounding climatic risk. A common theme throughout the project is that of impact mapping, both geographically and institutionally. Outputs from this project will increase the effectiveness of other projects of CIAT, as well as the wider R&D community. Output 1 specifically develops knowledge on how impact occurs in complex institutional, economic, environmental and geographic settings, and develops methodologies for monitoring and evaluating impacts. Output 2 focuses on the significant risks facing rural communities (especially from climate variability and change) through impacts on agricultural production and the natural resource base, and develops tools and methodologies for assessing and adapting to these risks from the local to the regional scale. This output specifically 4 looks at the challenges of climate variability and change to rural communities, providing policyrelevant insights of impacts and potential adaptation mechanisms. Cross-cutting between both outputs is the use of spatial analysis for characterizing the problems associated with rural development and for supporting ex-ante and ex-post impact assessments and supporting research decisions during the life of projects. This builds on CIAT’s core competency in spatial analysis, and an important component of the project’s strategy is one of service provision within CIAT and to key external partners. The project operates through close collaboration with other projects within CIAT (both germplasm and natural resources) and with external partners, especially Challenge Programs. The project leads Theme 2 of the Water and Food Challenge Program, co-coordinates the Andes Basin Focal Project of the Water for Food Challenge Program, plays a coordination role in the Lake Kivu pilot site of the SSA-CP, and supports both Harvest Plus and GCP through geographic analyses of ex ante impact. Gender analysis will be applied systematically in the work described here. 1. Outcome Line Logframe The logframe for 2008-2010 was still based on the old RDC structure of an integrated logframe with 5 products (or broad outputs). The logframe for 2009-2011 is the first which is actually structured around the Agroecosystems Resilience Outcome Line structure. 5 Targets Output Outcome Impact Greater incorporation of the interests of the poor in the design and implementation of R&D projects R&D investments have larger impacts, of which a larger share goes to the poorest beneficiaries Institutional arrangements and mechanisms for targeting, increasing and evaluating impacts Agricultural and environmental research organizations, development and environmental organizations, civil society groups, policy makers at regional, national and local scales An approach for strengthening and weaving effective networks for influence and pro-poor impact put into use in at least one R4D program Complex R4D research programs and projects, eg. CPWF, SSA CP, PABRA, EULACIAS Project, KS-inResearch Project, Cambio Andino Project; ERI project collaborators in Eastern and Southern Africa Complex R4D research projects and programs use network methods developed to monitor, evaluate and strengthen the networks that they build and foster More efficient use of research-for-development funds to foster innovation; higher quality ‘learning selection’ in projects and programs using the tools; improved relevance and impacts of agricultural innovations systems through better expression of userdemands (see above) Methodological framework for testing and evaluating innovation platforms (multistakeholder partnerships between private-public-CSOs) and other forms of partnerships for facilitating small holder participation in high value market chains National agriculture research and extension systems; SSACP, civil society organizations; decentralized local Governments and local institutions; rural service providers Increased capacities of organizations / institutions to develop and promote integrated agro-enterprise development solutions for wealth creation Effective multi-stakeholder partnerships with skills in innovative approaches for linking farmers to markets, improved performance of the research for development, better delivery of quality services, accelerated uptake of agricultural innovations and feedback to research and development priorities OUTPUT 1 Output Targets 2009 Intended User 6 Targets Output Targets 2010 Output Targets 2011 OUTPUT 2 Output Extrapolation domain analysis comprising biophysical and social parameters developed for supporting technology transfer Intended User Outcome Policy-makers (public, private & donor), farmer organizations, NGOs, researchers in CIAT and partner organizations Researchers and development practitioners using extrapolation domain analysis to identify the geographic targets for specific agricultural technologies, practices or policies Appropriate agricultural and natural resource technologies, practices and policies being used by rural communities, contributing to reduced poverty and sustainable natural resources Water-poverty interactions assessed in the Andes through expert knowledge and Bayesian network analysis CPWF, Organizations working on pro-poor development, conservation organizations, managers of water systems Improved understanding of water-poverty interactions leading to improved targeting of programs, interventions and benefits towards the rural poor in the Andes basin Targeted R+D reduces poverty associated with water-related processes in the Andes Institutional priorities and arrangements identified with respect to water, poverty and agricultural production in the Andes Organizations in the Andes that address issues of agriculture, natural resources and economic development, CPFW. Enhanced understanding of multiple objectives at basin scales leading to complementarities and tradeoffs. Discussion amongst stakeholder groups to negotiate preferable and equitable policies and projects. Improved soil, water and agricultural productivity contribute to human welfare and ecosystem resilience. Policy guidelines, tools and innovations for adaptation to risk, high stress and vulnerability. Policy-makers (public, private & donor), farmer organizations, NGO’s, researchers in CIAT and partner organizations Improved conceptual and empirical understanding of how policy enables effective research and development interventions Impact R&D efforts lead to effective, equitable and sustainable development in the tropics. 7 Targets Output Output Targets 2009 Socio-economic and agronomic vulnerability hotspots identified under current climate variability and future climate change Policy-makers (public, private & donor), farmer organizations, NGOs, researchers in CIAT and partner organizations Tools developed and applied for the identification of development policies and associated investments that support the implementation of profitable and resilient land uses Standard protocol for valuation of ecosystem services (soil and water) developed and tested in at least 2 pilot sites Policy-makers (public, private & donor), farmer organizations, private sector, NGOs, researchers in CIAT and partner organizations Ecosystem service payment schemes launched in two pilot sites, contributing to sustainable land-use systems Ecosystem service payment schemes established in two pilot sites, improved soil and water use and increased farm productivity. Poverty assessments and cropspecific drought maps for priority areas of the Generation Challenge Program Agricultural scientists of the Generation Challenge Programs and others working on drought. Agricultural scientists will be able to more efficiently target drought tolerant varieties to poor farmers in drought-prone environments Drought tolerant varieties lead to improved productivity and better livelihoods for those living in marginal environments A set of instruments (seasonal forecasting, insurance, policy), agricultural technologies and practices for coping and adapting to climate change identified and promoted in pilot sites Policy-makers (public, private & donor), farmer organizations, NGOs, researchers in CIAT and partner organizations Innovations contributing to enhanced resilience in agricultural systems to climate variability and change Less vulnerability of rural communities, especially in marginal areas, to climate variability and change Assessment of drought phenotyping trial sites to provide information for future field trial planning and dissemination of drought tolerant genotypes. Plant breeders in CIAT Improved targeting of research activities leads to development of better varieties at lower cost Drought tolerant varieties lead to improved productivity and better livelihoods for those living in marginal environments Output Targets 2010 Intended User and partner organizations, GCP, NARS Outcome Impact Improved efficiency of development interventions in increasing the adaptive capacity of agricultural systems to climate variability and change 8 Targets Output Targets 2011 Output Intended User Outcome Impact Breeding strategy recommendations to confront global climate change made for at least 3 crops on a global scale Plant breeders in CIAT and partner organizations, GCP, NARS Crop improvement programs have a 20-year vision of demand for new germplasm and use it to develop crop improvement programs Farming communities have adapted germplasm at their disposal to confront future challenges from climate change Community-based risk experimental methods developed to evaluate change scenarios at the local level in the context of global change Policy-makers (public, private & donor), farmer organizations, NGOs, researchers in CIAT and partner organizations Methodologies available for evaluating climatic risk from the community perspective Communities less exposed to climatic risk through adoption of appropriate resilient technologies and practices Weather insurance schemes based on sound climatological and agronomic science in place in at least two sites in two different countries Policy-makers (public, private & donor), farmer organizations, NGOs Methodologies developed by CIAT are adopted by partner organizations and used in the development of weather insurance schemes Reduced climatic riskexposure of rural communities leads to reduced poverty and more stable livelihoods An assessment of the potential of payment for environmental services generated from agriculture to both improve the environment and rural livelihoods Agricultural extension, Organizations working on pro-poor development, conservation organizations, managers of downstream water systems (irrigation and potable water) Where appropriate, farmers will receive additional incentives to adopt soil and water conserving practices Upland agriculture is more productive and sustainable and downstream water supplies are improved 9 2. Outcome 2008 - Targeting of high-value crops to environmental niches through a supply-chain framework CIAT research on the targeting of high-value crop options to environmental niches in Latin America and Africa has generated a number of methodologies and tools which are now being used widely by public and private organizations engaged in high-value supply chains. This outcome refers to Output 2 of the BP-2 project in CIAT’s 2007-2009 MTP, which aimed to generate “Frameworks and tools for evaluating and targeting technology and/or management alternatives in agriculture and NRM R&D”. This outcome was achieved largely through a project in Colombia and Ecuador on diversification options in hillside landscapes, and through a number of off-shoot projects including also Central America which were subsequently developed. The overriding principle of this work focused on the development of generic tools and methodologies for identifying niches for high-value crops, and the application of such methodologies in coffee, honey, medicinal plant and high-value forage supply chains. A total of 52 community-based organizations, public institutions, and private companies were involved. The CinfO system was developed and made operational, which allows the two-way flow of information between producers, exporters, and roasters, and even the consumer. During the course of the project, some 2,000 farms were integrated into the Cinfo system. Today more than 4,000 farms are registered in CinfO and this number continues to increase as other farmer organizations and secondary level organizations adopt the tool. Homologue software was developed to find homologous environments for technological transfer, where a particular variety or management regime may be well adapted. This was used to identify specific niches for high-quality coffee production, and results were validated in the field. Homologue has now been distributed to over 100 organizations across the globe, including NARS, ARIs, and sub-national research and development organizations. Canasta software was developed to combine expert knowledge with formal scientific knowledge in order to predict potential adaptation zones for a species. It was used to identify specific niches for high-quality production of coffee in Colombia and Central America, and through the CinfO system this information is fed back to the farmer in order to provide options for increasing farm income. It is now being used on a range of different crops in different continents, including for the generation of a denomination of origin for coffee in Colombia and Nicaragua. Today, Homologue and Canasta are being used by external partners across Latin America, Africa and Asia for identifying environmental niches for a wide range of crops and species, including many underutilized crop species. The analysis of the environmental drivers of coffee quality brought a number of important insights, which provided essential components for projects such as the subsequent work with the Federación Nacional de Cafeteros de Colombia on denomination of origin. A number of other spin-off projects have been generated, and continue to be implemented in the region with new institutions which have become interested in the environmental niche concept for stimulating rural development around high-value supply chains. The evidence for this outcome is available in a range of reports and scientific publications which are using the principles, concepts and tools of identifying environmental niches for high value 10 products. An impact study of the output from the 2007 MTP is pending to fully quantify adoption, uptake and changes in farm income derived directly or indirectly from this outcome. 11 3. Achievement of Output Targets for 2008 We successfully and fully achieved all output targets for 2008: TARGETS 2008 Fully Achieved X PRODUCT 1 • A method for tracking change, improving learning, accountability, relevance and impacts of agricultural innovation systems tested in at least two countries in Africa and Asia EXPLANATION The method developed is Participatory Impact Pathways Analysis which has been published in the Canadian Journal of Program Evaluation and has been adopted by CPWF, the CIPCIAT Project Cambio Andino and the EULACIAS project where it is the basis of the Co-Innovation Dynamics work package. Evidence: Rotondo, Emma, Rodrigo Paz, Graham Thiele. 2008. EVALUATION OF OUTCOMES AND IMPACTS OF PARTICIPATORY METHODOLOGIES. Andean Change project approach. WORKSHOP ON RETHINKING IMPACT, Capturing the Complexity of Poverty and Change, Cali-Colombia, March 26-28. Downloaded from: http://www.prgaprogram.org/riw/files/papers/Paper%20Camb io%20Andino%20PRGA%20Workshop%20vf.doc.on 2nd March, 2009 EULACIAS web site http://www.eulacias.org/posters_presentations.html X • A set of good practices derived from Colombia and Kenya for strengthening the participation of the poor in land and water management institutions. Johnson, N., 2008, Sustaining inclusive Collective Action that Links across Economic and Ecological Scales in upper watersheds (SCALES), Report produced by CIAT for the CPWF, 41pp, CIAT, Colombia. X • Two studies published assessing levels and dimensions of social capital and approaches that are critical for promoting pro-poor market linkages, farmer experimentation, social inclusion, and investment in natural resource management in Eastern Africa. PRODUCT 2 • Three sets of frameworks, methodology and tools to target staple crops and higher value products to environmental and socioeconomic niches developed and Report on the SCALES project delivered to the CPWF: Two publications: Susan Kaaria, Jemimah Njuki, Annet Abenakyo, Robert Delve, Pascal Sanginga. Assessment of the Enabling Rural Innovation (ERI) approach: Case studies from Malawi and Uganda. Natural Resources Forum 32 (2008) 53–63. Kaaria, Susan K.; Njuki, Jemimah; Abenakyo, Annet; Delve, Robert J.; Sanginga, Pascal C. 2008. Enabling rural innovation: Empowering farmers to take advantage of market opportunities and improve livelihoods. In: Sanginga, Pascal C.; Waters-Bayer, Ann; Kaaria, Susan K.; Njuki, Jemimah; Wettasinha, Chesha (eds.). Innovation Africa: Enriching farmers´ livelihoods. Earthscan, London, GB ; Sterling, VA, USA. p. 167-185. X CANASTA, HOMOLOGUE and empirical statistical methods developed for identifying niches for 23 underutilised crop species globally, banana, plus a range of other crops: Salazar M. and Jarvis A., 2008, Mapping of GeoEnvironmental Niche Suitability (G-ENS) For Neglected and Underutilized Plant Species (NUS), Report 12 tested for at least 15 crops (General spatial analysis tools, as well as CIAT’s Canasta and Homologue software tools, adapted to a range of crops; concepts expanded to Africa) PRODUCT 3 • A methodology and two prediction models to target higher value products to environmental niches developed and tested with at least 5 crops in LAC. to the Global Facilitation Unit (GFU), CIAT, Cali, Colombia. X Detailed niche identification methodologies and prediction models developed for coffee in Colombia, Peru and Nicaragua (reports available), including methodologies for defining denominations of origin: Oberthur, t. et al, 2008, Strengthening the Implementation of Denominations of Origin for Coffee in the Huila, Tolima, Santander, Santander Norte, César and Magdalena Departments of Colombia: Relationships between Environmental Factors and Inherent Quality Characteristics of Green and Roasted Coffee Beans, Report to the National Federation of Coffee Growers, 162pp., CIAT, Colombia. Niche identification methods and tools applied also to some high-value underutilized species: Salazar M. and Jarvis A., 2008, Mapping of GeoEnvironmental Niche Suitability (G-ENS) For Neglected and Underutilized Plant Species (NUS), Report to the Global Facilitation Unit (GFU), CIAT, Cali, Colombia. Method for measuring and analysing social capital and its relationship with market access and other variables are described for three Southern African countries in: PRODUCT 5 • • • Standard protocol to examine how farmer linkages to markets affect investments in NRM (currently in use in Malawi, Uganda, Zimbabwe, Mozambique). X Njuki, J.M., M.T. Mapila, S. Zingore and R. Delve. 2008. The dynamics of social capital in influencing use of soil management options in the Chinyanja Triangle of southern Africa. Ecology and Society 13 (2): 9. Comprehensive assessment of the state of ecosystem services and its link with poverty in the Andes/Amazon region. X Baseline spatial datasets on climate, climate risk and natural resources (vegetation) developed. X Report produced for DFID, and published on the web: ESPA-AA 2008: Challenges to Managing Ecosystems Sustainably for Poverty Alleviation: Securing Well-Being in the Andes/Amazon. Situation Analysis prepared for the ESPA Program. Amazon Initiative Consortium, Belém, Brazil. Available from http://www.ecosystemsandpoverty.org/wpcontent/uploads/2008/05/espa-aa-final-report-_smallversion_.pdf Datasets developed and being used internally, and being offered externally when bandwidth permits: http://srtm.csi.cgiar.org http://www.worldclim.org http://www.diva-gis.org Climate risk data has been generated and has been disseminated to national partners. 13 4. Research Highlights 2008 Following the three broad areas of work we engage in, here follows three research highlights: 4.1 Technology and impact targeting: 4.2 Equitable and sustainable use of ecosystem services: Ecosystem services and poverty alleviation in the Andes/Amazon We completed a strategic analysis of the entry points for both research and development in the Amazon region with regard to ecosystem services and poverty alleviation. The report aims to guide research and capacity-building priorities related to ecosystem services and poverty alleviation in the Amazon basin and eastern Andes. It is the result of extensive engagement with stakeholders in the region, combined with novel analysis of secondary data on poverty and ecosystem services such as water provision, biodiversity, and soil quality. The report presents a list of priority research challenges for the region, concluding that it is far more cost effective to prevent future degradation through incentive-based schemes that empower local communities rather than force people to comply authoritatively. Commissioned by the Ecosystems Services for Poverty Alleviation Programme (ESPA), a UK-based initiative of DFID, NERC, and ESRC to promote multi-disciplinary research in sustainable ecosystem management, this study is valuable to direct environmental-management policy at all levels. The full report is available at: http://www.ecosystemsandpoverty.org/wp-content/uploads/2008/05/espa-aa-final-report-_smallversion_.pdf. 4.3 Climate change: The changing geography of agricultural suitability There have now been a number of global and regional studies on the impacts and potential implications of climate change on agricultural productions of major crops, with some studies examining the significance of these changes to food security. Whilst a significant percentage of food intake per capita is accounted for by the world’s ten biggest crops, food and nutritional security depends on a much wider range of crops, some of which are consumed on farm and others cultivated as cash crops. Unfortunately, mechanistic-based models (like DSSAT) are only available for a handful of crops, which goes to explaining the concentration of research on major staples. We used a simpler approach to modeling the impacts of climate change on agriculture using the Ecocrop niche-based model. Under 2 different scenarios, and 18 downscaled GCM models we map the changing geographies of crop suitability to 2020 and 2050 for 50 crops. The crops studied included staples, cash-crops and traditional crops that contribute heavily at the local scale to food and nutritional security. Using agricultural production and export data from FAOSTAT, we analyzed the impacts within the context of food and nutritional security for tropical countries. The analysis shows that a great deal of opportunities exist in agriculture as a result of climate change if farmers have the access and information to change varieties and, when necessary, their crops. When the crops are grown for cash, this is easy. However, when the crops are of large cultural importance and highly traditional, adaptation measures are made significantly more difficult. We used this approach to identify hotspots of both opportunity, and of significant challenges where fundamental changes in the agricultural system may be required. The results of this research were presented in numerous international fora. 14 5. Description of one project outcome. Targeting of high-value crops to environmental niches through a supply-chain framework CIAT research on the targeting of high-value crop options to environmental niches in Latin America and Africa has generated a number of methodologies and tools which are now being used widely by public and private organizations engaged in high-value supply chains. This outcome refers to Output 2 of the BP-2 project in CIAT’s 2007-2009 MTP, which aimed to generate “Frameworks and tools for evaluating and targeting technology and/or management alternatives in agriculture and NRM R&D”. This outcome was achieved largely through a project in Colombia and Ecuador on diversification options in hillside landscapes, and through a number of off-shoot projects including also Central America which were subsequently developed. The overriding principle of this work focused on the development of generic tools and methodologies for identifying niches for high-value crops, and the application of such methodologies in coffee, honey, medicinal plant and high-value forage supply chains. A total of 52 community-based organizations, public institutions, and private companies were involved. The CinfO system was developed and made operational, which allows the two-way flow of information between producers, exporters, and roasters, and even the consumer. During the course of the project, some 2,000 farms were integrated into the Cinfo system. Today more than 4,000 farms are registered in CinfO and this number continues to increase as other farmer organizations and secondary level organizations adopt the tool. Homologue software was developed to find homologous environments for technological transfer, where a particular variety or management regime may be well adapted. This was used to identify specific niches for high-quality coffee production, and results were validated in the field. Homologue has now been distributed to over 100 organizations across the globe, including NARS, ARIs, and sub-national research and development organizations. Canasta software was developed to combine expert knowledge with formal scientific knowledge in order to predict potential adaptation zones for a species. It was used to identify specific niches for high-quality production of coffee in Colombia and Central America, and through the CinfO system this information is fed back to the farmer in order to provide options for increasing farm income. It is now being used on a range of different crops in different continents, including for the generation of a denomination of origin for coffee in Colombia and Nicaragua. Today, Homologue and Canasta are being used by external partners across Latin America, Africa and Asia for identifying environmental niches for a wide range of crops and species, including many underutilized crop species. The analysis of the environmental drivers of coffee quality brought a number of important insights, which provided essential components for projects such as the subsequent work with the Federación Nacional de Cafeteros de Colombia on denomination of origin. A number of other spin-off projects have been generated, and continue to be implemented in the region with new 15 institutions which have become interested in the environmental niche concept for stimulating rural development around high-value supply chains. The evidence for this outcome is available in a range of reports and scientific publications which are using the principles, concepts and tools of identifying environmental niches for high value products. An impact study of the output from the 2007 MTP is pending to fully quantify adoption, uptake and changes in farm income derived directly or indirectly from this outcome. 16 6. Publications Articles in refereed journals Abello, J.F.; Kelemu, S.; Garcia, C. 2008. Agrobacterium-mediated transformation of the endophytic fungus Acremonium implicatum associated with Brachiaria grasses. Mycological Research 112(3):407-413 Bode, R.; Arévalo, D.; Victoria, P. 2008. Knowledge management and communication to address information access and power asymmetries for resource-poor producers in value chains. Knowledge management and knowledge sharing in Latin America and Caribbean". Knowledge Sharing for Development Journal 4(1): 5-20 Börner, J.; Wunder, S. 2008. Paying for avoided deforestation in the Brazilian Amazon:From cost assessment to scheme design. International Forestry Review 10: 496-511. Carvajal, A.; Mayorga, O.; Douthwaite, B. 2008. Forming a Community of Practice to Strengthen the Capacities of Learning and Knowledge Sharing Centers in Latin America and the Caribbean—A D-Group Case Study. KM4D Journal 4(1):71-81 Diaz-Nieto, J.; Cook, S.; Jones, P.; Laderach, P. (submitted). Helping small-holder farmers to manage drought risk through insurance: A case study of dry bean production in Honduras, World Development. Diaz-Nieto, J. Cook, S. Lundy, M. Fischer, M. Laderach, P. (submitted). Risk sharing through insurance: Weather indices for designing micro-insurance products for poor small-holder farmers in the tropics. Journal of International Development Douthwaite, B.; Alvarez, B.S.; Cook, S.; Davies, R.; George, P.; Howell, J.; Mackay, R.; Rubiano, J. 2008. Participatory Impact Pathways Analysis: A Practical Application of Program Theory in Research-for-Development. Canadian Journal of Program Evaluation 22(2):127–159 Gotschi, E., Njuki, J., Delve, R. 2008. Gender equity and social capital in smallholder farmer groups in central Mozambique. Development in Practice 18(4-5): 650-657. Hyman, G.; Fujisaka, S.; Jones, P.; Wood, S.; De Vicente, M.C.; Dixon, J. 2008. Strategic approaches to targeting technology generation: Assessing the coincidence of poverty and droght-prone crop production. Agricultural Systems 98:50-61. Jarvis, A.; Lane, A.; Hijmans, R.J. 2008. The effect of climate change on crop wild relatives. Agriculture, Ecosystems & Environment 126:13-23. Labarta, R.; White, D.; Swinton, S. 2007. “Does Charcoal Production Slow Agricultural Expansion into the Peruvian Amazon Rainforest?” World Developmen 36(3):527-540 17 Lehner, B.; Verdin, K.; Jarvis, A. 2008. New Global Hydrography Derived from Spaceborne Elevation Data. Eos Trans AGU 2008 89(10):94-95. Maxted, N.; Dulloo, E.; Ford-Lloyd, B.; Iriondo, J.M., Jarvis, A.2008.Gap analysis: a tool for complementary genetic conservation assessment. Diversity & Distribution 14(6): 1018-1030. Niederhauser, N.; Oberthür, T.; Kattning, S.; Cock, J.H. 2008. Information and its management for differentiation of agricultural products: The example of specialty coffee. Computers and Electronics in Agriculture 61(2):241-253. Njuki, J.; Mapila, M.; Kaaria, S.; Magombo, T. 2008. Using community indicators for evaluating research and development programmes: experiences from Malawi. Development in Practice 18(4): 633-642. Njuki, J.; Mapila, M.T.; Zingore, S.; Delve, R.J. 2008. The dynamics of social capital in influencing use of soil management options in the Chinyanja Triangle of southern Africa. Ecology and Society 13(2):1-16 Salas, M.; Camacho, K.; Staiger-Rivas, S.; Villa, C.; Ferguson J.; Cummings S. 2008. Knowledge Sharing and Knowledge Management in Latin America and the Caribbean. KM4Dev Journal 3(2):2-4 Watts, J.; Horton, D.E.; Douthwaite, B.; La Rovere, R.; Thiele, G.; Prasad, S.; Staver, C. 2008. Transforming impact assessment: Beginning the quiet revolution of institutional learning and change. Experimental Agriculture 44:21-35. Books and monographs Estrada, R.D.; Holmann, F.J. 2008. Competitividad de los pequeños productores de leche frente a los tratados de Libre Comercio en Nicaragua, Costa Rica y Colombia. Centro Internacional de Agricultura Tropical (CIAT); International Livestock Research Institute (ILRI), Cali, CO. 74 p. (Documento de trabajo no. 207) Estrada, M. 2008. Evaluación de Estrategias de Manejo Especifico por Sitio para el Mejoramiento de la Calidad de Taza de Cafe (Coffea arabica L.). Tesis (M.Sc.). Universidad Nacional de Colombia, Maestria en Ciencias Agrarias. Medellín, Antioquia, CO. 81p. Monserrate, F. 2008. Análisis del Proceso de Biofortificación de Variedades de Fríjol (Phaseolus vulgaris L.) Andino de Comercial "Calima "en Colombia. Tesis (Ingeniero Agrónomo). Universidad Nacional de Colombia, Facultad de Agronomía, Bogota, DC, CO. 91 p. Laderach, P.; Jarvis, A.; Ramírez, J.; Fisher, M.J. 2008. Predictions of land use changes under progressive climate change in coffee growing regions of the AdapCC project: Final report Chiapas, Mexico, Cali, Colombia: October 2008[on line]. Centro Internacional de Agricultura Tropical (CIAT), Cali, CO. 65 p. 18 Laderach, P.; Jarvis, A.; Ramírez, J.; Fisher, M.J. 2008. Predictions of land use changes under progressive climate change in coffee growing regions of the AdapCC project: Final report Nicaragua, Cali, Colombia: October 2008 [on line]. Centro Internacional de Agricultura Tropical (CIAT), Cali, CO. 62 p. Laderach, P.; Jarvis, A.; Ramírez, J.; Fisher, M.J. 2008. Predictions of land use changes under progressive climate change in coffee growing regions of the AdapCC project: Final report Piura, Peru, Cali, Colombia: October 2008 [on line]. Centro Internacional de Agricultura Tropical (CIAT), Cali, CO. 66 p. Laderach, P.; Jarvis, A.; Ramírez, J.; Fisher, M.J. 2008. Predictions of land use changes under progressive climate change in coffee growing regions of the AdapCC project : Final report Veracruz, Mexico, Cali, Colombia: October 2008 [on line]. Centro Internacional de Agricultura Tropical (CIAT), Cali, CO. 66 p. Pauli, N. 2008. Environmental influences on the spatial and temporal distribution of soil macrofauna in smallholder agroforestry system of western Honduras. Thesis (PhD). The University of Western Australia, School of Earth and Geographical Sciences Doctor of Philosophy, Australia. 333 p. Piechazek, J. 2008. Implications of Quality-Based Agri-Food Supply Chains on Agri-Social Systems: The Case of Smallholder Coffee Growers in South Colombia. Thesis (PhD). Hohen Landwirtschaftlichen Fakultät der Rheinischen-Wilhelms-Universität Bonn. 300 p. Quintero, M.; Holmann, F.; Estrada, R.D. 2008. Ex ante Analysis of Legumes: The Dilemma of Using Legumes as Forage for Animal Nutrition during the Dry Season or as Green Manure for Soil Improvement. Documento de Trabajo. CIAT, Cali, Colombia. Rodríguez, F. 2008. Caficultores de Pequeña Producción Asociados Ingresando en Mercados de Alto Valor. Tesis (M.Sc.). Universidad del Valle, Facultad de Ciencias de la Administración, Santiago de Cali, Valle del Cauca, CO. 269p. Book chapters Börner, J.; Hohnwald, M.; Vosti, S.A. 2008. Critical analysis of options to manage ecosystem services in the Amazon/Andes Region. In: Coelho, A.B.; Teixeira, E.C; Braga, M.J. (eds.). A Situation Analysis to Identify Challenges to Sustainable Management of Ecosystems to Maximise Poverty Alleviation: Securing Biostability in theAmazon/Andes (ESPA-AA). Recursos Naturais e Crescimento Econômico. Vicosa Federal University, Vicosa, MG, Brazil. p. 1-29 Börner, J.; Hohnwald, M.; Vosti, S.A. 2008. From natural resource to pro-poor ecosystem service management in the Amazon: How to make the right choices? [abstract] . In: Tielkes, Eric. (ed.). Competition for resources in a changing world: New drive for rural development: Book of abstracts, Tropentag 2008. University of Hohenheim, Centre for Agriculture in the Tropics and Subtropics, Hohengeim, DE. p. 505. 19 Cook, S.E.; Jarvis, A.; Gonzalez, J.P. 2008. A New Global Demand for Digital Soil Information. In: Hartemink, A.E.; McBratney, A.; Mendonça-Santos, de Lourdes (eds.). Digital Soil Mapping with Limited Data. Edited by M. Springer Netherlands. p. 31-41. Dulloo, M.E.; Labokas, J.; Iriondo, J.M.; Maxted, N.; Lane, A.; Laguna, E.; Jarvis, A.; Kell, S.P. 2008 Genetic Reserve Location and Design. 2008. In: Iriondo, J.M.; Maxted, N.; Dulloo, M.E. (eds.). Conserving Plant Genetic Diversity in Protected Areas, CAB International, London, p. 23-64. Gonzalez, J.P.; Jarvis,A.; Cook, S.E.; Oberthür, T.;, Rincon-Romero, M.; Bagnell, J.A.; Dias, Bernardine M. 2008. Digital Soil Mapping of Soil Properties in Honduras Using Readily Available Biophysical Datasets and Gaussian Processes. In: Hartemink, A.E,; McBratney, A.; Mendonça-Santos, de Lourdes (eds.). Springer Netherlands. Digital Soil Mapping with Limited Data. p. 367-380. Hijmans, R.J.; Jarvis, A.; Guarino, L. 2008. Climate envelope modeling: Inferring the range of species. In: Gibbs, J.P., Hunter, Jr M.L.; Sterling, E.J. (eds.). Problem-solving in conservation biology and wildlife management (2nd edition).Blackwell, UK. p. 244-254. Kaaria, S.K.; Sanginga, P.; Njuki, J.; Delve, R.; Chitsike, C.; Best, R. 2008. Enabling rural innovation in Africa. In: Scoones, Ian; Thompson, John. Farmer First Revisited; Innovation for Agricultural Research and Development. Practical Action Publications, London. Meijer, M.; Rodriguez, I.; Lundy, M.; Hellin, J. 2008. Supermarkets and small farmers - the case of fresh vegetables in Honduras. In: McCullough, Ellen; Pingali, Prabhu; Stamoulis, Kostas (eds.). The Transformation of Global Agrifood Systems: Supply Chains, Globalization and Smallholder Farmers. Earthscan Publications Ltd. London, UK. 416 pp. Mulligan, M.; Rubiano, J.; White, D.; Hyman, G.; Saravia, M. 2008. Participatory modeling and knowledge integration - Basin Focal Project (BFP Andes): concepts and advances. In: Humphreys, E.; Bayot, R.S.; van Brakel, M.; Gichuki, F.; Svendsen, M.; Wester, P.; HuberLee, A.; Cook, S.; Douthwaite, B.; Hoanh, C.T.; Johnson, N.; Nguyen-Khoa, S.; Vidal, A.; MacIntyre, I.; MacIntyre, R. (eds.). Fighting Poverty Through Sustainable Water Use: Volumes I, II, III and IV. Proceedings of the CGIAR Challenge Program on Water and Food 2nd International Forum on Water and Food, Addis Ababa, Ethiopia, November 10—14, 2008. The CGIAR Challenge Program on Water and Food, Colombo. 183p. Njuki, J.; Kaaria, S.K.; Sanginga,P.; Kaganzi, E.; Magombo, T. 2008. Linking smallholder farmers to markets through community agro-enterprise development: periences from Uganda and Malawi. In: Scoones, Ian; Thompson, John.: Farmer First Revisited; Innovation for Agricultural Research and Development. Practical Action Publications, London. Pérez, S.A.; Tegbaru, A.; Kantengwa, S.; Farrow, A. 2008. Village Information and Communication Centres in Rwanda. In: Sanginga, P.; Waters-Bayer, A.; Kaaria, S.; Njuki, J.; Wettasinha, C. (eds). Innovation Africa: enriching farmers’ livelihoods. Earthscan, London. 20 Papers presented at formal conferences and workshop with external attendance Börner, J.; Wunder, S. 2008. The potential of payments for forest environmental services in the Brazilian Amazon: Insights from a macro-scale spatial analysis. In: 10th Biennial International Society for Ecological Economics (ISEE), 7-11 August, Nairobi, Kenya. Börner, J.; Porro, R.; Jarvis, A. 2008. Securing Social and Biostability in the Andes/Amazon: A Pan Amazonian Situation Analysis of Ecosystem Services and Wellbeing. In: International Scientific Conference Amazon in Perspective: Integrated Science for a Sustainable Future, November 17-20, 2008, Manaus, Brazil. At: www.lbaconferencia.org Börner, J.; Mendoza, A. 2008. Secondary forest valuation on family farms with different technology access in the Eastern Brazilian Amazon: Can conservation incentives compete with opportunity costs in slash-and-burn agriculture?. In: International Scientific Conference Amazon in Perspective: Integrated Science for a Sustainable Future, November 17-20, 2008, Manaus, Brazil. At: www.lbaconferencia.org Castro, A.; Rivera, M.; Ferreira, O.; Pavon, J.; García, E.; Amezquita, E.; Ayarza, M.; Barrios, E.; Rondon, M.; Pauli, N.; Baltodano, M.E.; Mendoza, B.; Welchez, L.A.; Cook, S.; Rubiano, J.; Johnson, N.; Rao, I. 2008. Is the Quesungual System an option for smallholders in dry hillside agroecosystems?. In: Proceedings of the CGIAR Challenge Program on Water and Food 2nd International Forum on Water and Food, November 10-14, Addis Ababa, Ethiopia. Available at www.waterandfood.org Cook, S.; Harrington, L.; Huber-Lee, A. 2008. Water and food in river basins in Africa, Asia, and Latin America: A comparative analysis. In: Humphreys, E.; Bayot, R.S.; van Brakel, M.; Gichuki, F.; Svendsen, M.; Wester, P.; Huber-Lee, A.; Cook, S.; Douthwaite, B.; Hoanh, C.T.; Johnson, N.; Nguyen-Khoa, S.; Vidal, A.; MacIntyre, I.; MacIntyre, R. (eds). Fighting Poverty Through Sustainable Water Use: Volumes I, II, III and IV. In Proceedings of the CGIAR Challenge Program on Water and Food 2nd International Forum on Water and Food, Addis Ababa, Ethiopia, November 10-14, 2008. The CGIAR Challenge Program on Water and Food, Colombo. 183 p. Cook, S. E. 2008. The Basin Focal Projects of the CPWF. Keynote Paper. In: 2nd International Forum on Water and Food, Addis Ababa, Ethiopia, November 10-14, 2008. The CGIAR Challenge Program on Water and Food, Colombo. 183 p. Cook, S.; Fisher, M.; Woolley, J. 2008. Meeting the Food and Water Crises. Invited Paper In: Syngenta Science Matters. Ascot, UK. 8 September 2008. Cook, S.; Fisher, M.; Woolley, J . 2008. Water agriculture and poverty- the CGIAR Challenge Program on Water and Food. In: Special Sessions. XIIIe World Water Congress. Montpellier, Sep 2008. 21 Farrow, A. 2008. What is Vulnerability to Food Insecurity and why do we want to map it?. In: Mapping for Food Security Workshop at Joint Research Centre of the European Commission, Ispra Italy, 5th November 2008. Hyman, G.; Geerts, S.; Shrestha, N; Raes, D. 2008. Environmental assessment for phenotyping network. Poster In: Annual Research Meeting of the Generation Challenge Program. 16-20 September. Bangkok, Thailand. Available from http://www.generationcp.org/UserFiles/File/ARM-2008-Posters_Theme4/4.17_Hyman_Environmental%20assessment%20for%20phenotyping%20network.pdf. Jarvis, A. 2008. Impacto de cambio climático en Colombia: Implicaciones para la agricultura y el manejo de los recursos naturales. In: VIII Encuentro de Estudiantes de Ingeniería Ambiental, Sanitaria y Áreas Afines, Universidad Nacional de Colombia, Palmira (Colombia) 9-11 Oct 2008. Jarvis, A. 2008. Agua, alimentación, pobreza y el potencial de los servicios ecosistémicos: del mundo a la región a lo local. In: Cumbre Interamericana del Agua y la Tierra, CORFIAGUA, Medellín, Colombia, Oct 2008. Jarvis, A. 2008. Modelling distribution impact with relation to agricultural biodiversity. In: FAO Expert Meeting on Agricultural Biodiversity, Rome (Italy), 28-29 Feb 2008. Jarvis, A. 2008. Modelling distribution impact in relation to agricultural biodiversity. In: FAO Expert Meeting: Climate Change Adaptation and Mitigation, Rome (Italy), 5-7 March 2008. Jarvis, A.; Gamboa, D.E. 2008. Cambio climático, agricultura y agua en los Andes: los tres mitos y medio. In: Foro Andino del Agua y la Alimentación, Challenge Program on Water & Food, Bogotá, Colombia, 29-31 Ene 2008. Jarvis, A.; Ramirez, J.; Guevara, E.; Zapata, E. 2008. Global impacts and implications of climate change on banana production systems. In: 18 International Meeting ACORBAT, Guayaquil, Ecuador, 10-14, Nov 2008. 18 p. Jarvis, A.; Ramirez, J.; Zapata, E.; Guevara, E. 2008. Use of GBIF data for conserving and adapting agricultural biodiversity in the face of climate change. In: 15th meeting of the Global Biodiversity Information Facility. Arusha, Tanzania, 1-7 Nov 2008. Laderach, P.; Jarvis, A.; Ramírez, J. 2008. The impact of climate change in coffee-growing regions. In: Taller de adaptación al cambio climático en las comunidades cafetaleras de la Sierra Madre de Chiapas, 16-28 de noviembre del 2008, Tuxtla Gutiérrez, Chiapas. Mexico. p. 1-5. Mulligan, M.; Rubiano, J.; White, D.; Hyman, G.; Saravia, M. 2008. Participatory modeling and knowledge integration - Basin Focal Project (BFP Andes): concepts and advances. In: Humphreys, E.; Bayot, R.S.; van Brakel, M.; Gichuki, F.; Svendsen, M.; Wester, P.; HuberLee, A.; Cook, S.; Douthwaite, B.; Hoanh, C.T.; Johnson, N.; Nguyen-Khoa, S.; Vidal, A.; 22 MacIntyre, I.; MacIntyre, R. (eds.). Fighting Poverty Through Sustainable Water Use: Volumes I, II, III and IV. Proceedings of the CGIAR Challenge Program on Water and Food 2nd International Forum on Water and Food, Addis Ababa, Ethiopia, November 10—14, 2008. The CGIAR Challenge Program on Water and Food, Colombo. 183p. Quintero, M.; Comerford, N.; Estrada, R.D. 2008. Is soil carbon sequestration part of the bundle of ecosystem services provided by conservation agriculture in the Andes?. In: “Second international forum on Water and Food (IFWF2)”, Addis Ababa, November 3-17. Ramirez, J; Jarvis, A; Van den Bergh, I. 2008. Presión de la Sigatoka Negra y Distribución Espacial de Genotipos de Banano y Plátano: Resultados de 19 Años de Pruebas con Musáceas. In: 18 International Meeting ACORBAT, Guayaquil, Ecuador, 10-14, Nov 2008. 92p. Rubiano, J.; Cook, S.; Douthwaite, B. 2008. Adapting to change–how to accelerate impact. . In: Humphreys, E.; Bayot, R.S.; van Brakel, M.; Gichuki, F.; Svendsen, M.; Wester, P.; HuberLee, A.; Cook, S.; Douthwaite, B.; Hoanh, C.T.; Johnson, N.; Nguyen-Khoa, S.; Vidal, A.; MacIntyre, I.; MacIntyre, R. (eds). Fighting Poverty Through Sustainable Water Use: Volumes I, II, III and IV. Proceedings of the CGIAR Challenge Program on Water and Food 2nd International Forum on Water and Food, Addis Ababa, Ethiopia, November 10-14, 2008. The CGIAR Challenge Program on Water and Food, Colombo. 183p. Rubiano, J.; Peralta, A.; Johnson, N. 2008. Scaling-up in watershed management research projects. In: Humphreys, E.; Bayot, R.S.; van Brakel, M.; Gichuki, F.; Svendsen, M.; Wester, P.; Huber-Lee, A.; Cook, S.; Douthwaite, B.; Hoanh, C.T.; Johnson, N.; Nguyen-Khoa, S.; Vidal, A.; MacIntyre, I.; MacIntyre, R. (eds). Fighting Poverty Through Sustainable Water Use: Volumes I, II, III and IV. Proceedings of the CGIAR Challenge Program on Water and Food 2nd International Forum on Water and Food, Addis Ababa, Ethiopia, November 10— 14, 2008. The CGIAR Challenge Program on Water and Food, Colombo. 183p. Rubiano, J.; Soto, V.; Rajasekharan, M.; Cook, S.; Douthwaite, B.; Idupulapati,. Rao.2008. Extrapolation Domain Analysis – A method to estimate potential global impacts of research projects Poster In: Challenge Program on Water and Food 2nd International Forum on Water and Food, Addis Ababa, Ethiopia, November 10—14, 2008. The CGIAR Challenge Program on Water and Food, Colombo. Rudebje, P. R.; Baidu-forson, J.; Van schagen, B.; Jarvis, A.; Staver, C.; Hodgkin, T. 2008. Agrobiodiversity and climate change: what do students need to know? In: 2nd ANAFE International Symposium: Mainstreaming climate change into Agricultural Education: Tools, Experiences and challenges. 28th July-1st August, 2008, University of Malawi. Schepp, K.; Laderach, P. 2008. Adaptación para los pequeños productores de café al cambio climático: Presentación de los resultados intermediarios y experiencias del proyecto piloto AdapCC: Una cooperación pública-privada entre Cafédirect y la GTZ. In: International workshop SIAASE Adaptation to climate change: The role of ecosystem services, 3-5 23 November 2008, CATIE, Costa Rica. Centro Agronómico Tropical de Investigación y Enseñanza (CATIE), Turrialba, CR. p. 1-2 Tovar, C.; Wood, S.; Hyman, G. 2008. From attractiveness to feasibility: assessing national capacity to adapt, deliver and adopt GCP technologies. In: Annual Research Meeting of the Generation Challenge Program. 16-20 September. Bangkok, Thailand. Available from http://www.generationcp.org/UserFiles/File/ARM-2008-Posters_Theme4/4.15_Tovar_Indicators%20of%20attractiveness%20and%20feasibility%20of%20GCP%20technologies_Rcd16Sep.pdf. van Zonneveld, M.; Jarvis, A.; Dvorak, W.; Koskela, J.; Vinceti, B.; Snook, L. 2008. Impact of climate change on distribution and performance of tropical pine species in Central America and Southeast Asia. p. 257. In: International Conference on Adaptation of Forests and Forest Management to Changing Climate with Emphasis on Forest Health: A Review of Science, Policies and Practices. Umea (Sweden), 25-28 Aug 2008. Swedish University Agricultural Science (Sweden). van Zonneveld, M.; Leibing, C.; Dvorak, W.; Jarvis, A 2008. Conservación de poblaciones naturales de pinos centroamericanos en el contexto del cambio climático. In: Seminario Internacional Bosques Tropicales y Desarrollo, Jardín Botánico de Medellín Joaquín Antonio Uribe, Medellín, Colombia, 8-21 Nov 2008. Jardín Botánico de Medellín, Colombia. Waddington, S.; Dixon, J.; Li, X.; Hyman, G.; de Vicente, C. 2008. Assessing production constraints and opportunities for GCP priority food crops and farming systems. Poster In: Annual Research Meeting of the Generation Challenge Program. 16-20 September. Bangkok, Thailand. Available from http://www.generationcp.org/UserFiles/File/ARM-2008-Posters_Theme-4/4.3_StephenWaddington_Assessing%20production%20constraints%20and%20opportunities%20for%20GCP%20 priority%20food%20crops%20and%20farming%20systems.pdf Walker, T. M.; Maredia, Kelley, T.; La Rovere, R.; Templeton, D.; Thiele, G.; Douthwaite, B. 2008. Strategic Guidance for Ex-Post Impact Assessment of Agricultural Research. In: Standing Panel on Impact Assessment, CGIAR Science Council. Rome, Italy. Articles in international newsletters or other scientific series Douthwaite, B.; Alvarez, B.S.; Thiele, G.; Makay, R. 2008. Participatory Impact Pathways Analysis: A practical method for project planning and evaluation. ILAC Brief 17, Bioversity, Rome Kaaria, S.K.; Njuki J. M.; Abenakyo, A.; Delve, R.; Sanginga, P.2008. Enabling rural innovation: empowering farmers to take advantage of market opportunities and improve livelihoods. In: Sanginga, P., Waters-Bayer, A., Kaaria S., Njuki, J., and Wettasinha C. (eds). Innovation Africa: enriching farmers’ livelihoods, Earthscan, London Njuki J. M.; Kaaria, S.K.; Sanginga, P.; Murithi, F.M.; Njunie, M.; Lewa, K.K. 2008. Building capacity for participatory monitoring and evaluation: integrating stakeholders’ perspectives. 24 In: Sanginga, P.; Waters-Bayer, A.; Kaaria S.; Njuki, J.; Wettasinha C. (eds). Innovation Africa: enriching farmers’ livelihoods, Earthscan, London Sanginga, P.; Waters-Bayer, A.; Kaaria, S.; Njuki, J.; Wettasinha C. (eds) 2008. Innovation Africa: enriching farmers’ livelihoods, Earthscan, London. 384 p, Thornton, P.; Jones, P.; Farrow, A.; Alagarswamy, G.;, Andresen, J. 2008. Crop Yield Response to Climate Change in East Africa: Comparing Highlands and Lowlands. In: Mountainous Regions: Laboratories for Adaptation. Germany: IHDP UPDA. Issue 2. p 23-26 Wunder, S.; Börner, J.; Tito, M.R.; Pereira, L. 2008. Pagamentos por serviços ambientais: Perspectivas para a Amazônia. Ministério do Meio Ambiente, Série Estudos 10, Brasília, Brazil. 131 p. 25 7. Funded project proposals Project Mainstreaming impact group support to the ILAC learning laboratory meeting and follow up monitoring and evaluation Environmental assessment for phenotyping network. Develop an ongoing system for monitoring habitat change in South America based on MODIS satellite imagery and NDVI. Scoping study for the competitive grant scheme for collecting threatened genetic diversity of crops focusing on wild relatives. Getting the focus right: Food crops and smallholder constraints. Mainstreaming Impact Group Logistical Support to the ILAC Learning Collective action for the rehabilitation of global public goods in the CGIAR genetic resources system: Phase 2 (GPG2), Basin Focal Project: Andean System of Basins To conduct a spatial analysis on biodiversity in East Africa Provision of cross-site research support in participatory monitoring and evaluation to the Sub-Saharan Africa Challenge Programme Manejo Integral de Cuencas Hidrográficas, Agua y Saneamiento (MARENA-PIMCHAS) Elaborar mapas de adaptabilidad de Café bajo la influencia de cambios climáticos para Perú, Nicaragua y México Greenlash in the Atlantic Forests of South America: Is there a relationship between regional deforestation and rainfall changes? Identificación y Validación de Sistemas Productivos Orgánicos Exitosos con Potencial de Mercado, en los Países del Cono Sur The Borlaug Leadership Enhancement in Agriculture Program (LEAP) Engaging Nationally Recruited Staff to Strengthen Research Capacities for Monitoring & Evaluation and Impact Assessment at Task Force Level Gap Analysis of CGIAR Genebank Collections Total Donor Total Budget Total CIAT in 2008 BIOVERSITY 21,850 21,850 GCP 160,674 50,640 TNC 53,201 53,201 GCDT 50,000 50,000 GCP 12,000 12,000 BIOVERSITY 14,950 14,950 BIOVERSITY 10,000 10,000 CPWF(KCL) 248,870 140,260 EP 11,000 11,000 FARA 120,000 120,000 CARE 33,500 33,500 GTZCAFEDIRECT 26,520 26,520 TNC 21,840 21,840 INIA-CHILE 24,860 8,230 IOWA. 10,452 10,452 FARA 56,000 - BIOVERSITY 39,000 914,717 39,000 623,443 26 Actual Expenditures 2008 Outcome Line PA-2: Risk & Climate Change Linking Farmers to Markets SOURCE HQ + LAC Unrestricted Core 539,005 Restricted Core C.E Sub-total Core Total US$ (%) 539,005 9% 0 0% Asia + Africa 539,005 0 539,005 9% 1,440,581 226,470 1,667,051 28% 190,569 3% 1,695,181 29% 1,136,752 19% Restricted Special Projects Generation Challenge Program 190,569 Sub Sahara Africa Water and Food Challenge Program 1,695,181 1,136,752 Sub Total Restricted 2,767,903 1,921,651 4,689,553 79% Direct Expenditures 3,306,907 1,921,651 5,228,558 89% Non Research Cost 425,730 247,393 673,123 11% 3,732,637 2,169,044 5,901,681 100% Total Expenditures 27 8. Project Staff (*Left during 2008) Internationally recruited Andrew Farrow Andrew Jarvis Boru Douthwite Chiuri Wanjiku Douglas White Glenn Hyman Jan Borner Jemimah Njuki Nancy Johnson* Norbert Niederhauser* Peter Laderach Roger Kirkby Simon Cook Simone Staiger Laure Collet MSc, GIS PhD, Geography PhD, Sociologist PhD, Social Scientist PhD, Agr. & Environ. Economist PhD, Geography PhD, Agricultural Science PhD, Sociologist PhD, Economist DI(FH),Inf. & Com. Engineering PhD, Agronomist PhD, Agronomist PhD, Social Scientist MSc, Communications MSc, Environmental Science Research Fellow, Kampala, Uganda Outcome Line Leader, Senior Scientist Senior Scientist Senior Scientist, Kampala, Uganda Senior Research Fellow Senior Scientist Associate Researcher, Brazil Senior Research Fellow, Harare, Zimbabwe Senior Scientist Research Fellow Postdoctoral Research Fellow PA RDC Leader Senior Scientist Leader, ICT-KM Project Research Fellow Nationally recruited Alexander Cuero* Ana Mercedes Hernandez* Ana Milena Guerrero Andrea Carvajal Carlos A. Nagles Carolina Gonzalez Clara Roa* Claudia J. Perea* Edward Guevara Elizabeth Barona Enna Bernarda Diaz Germán Lema* Hernan José Usma* James Garcia Jorge A. Cardona Katherine Tehelen Lilian Patricia Torres Liliana Rojas Marcela Quintero Maria Cecilia Roa Marisol Calderón* Natalia Uribe Ovidio Rivera Silvia Elena Castaño Victor Soto* Wilson Celemin* SystemsTechnology Sociologist Bilingual Secretary MSC, Rural Development Agricultural Technology Lawyer and Economist MSc, Sanitation and Water Resources BSc, Systems Engineer Environmental Engineering BSc, System Engineer MSc, Ecologist - Soils BSc, Industrial Engineering Agricultural Technology MSc, Statistician BSc, Systems Engineer Industrial Engineering BSc, Business Administration MSc, Natural Resources MSs, Ecologist - Soils PhD, Water Resources Architectural Drawing Topography Engineering Systems Technology BSc, Systems Engineer BSc, Business Administration Student Administration GIS Expert Research Assistant 2 Bilingual Secretary Communication Assistant 2 GIS Expert Research Associate 2 Research Assistant 2 Systems Analyst 3 Technician 1 Programmer 1 Research Assistant 2 Statistical Consultant 2 Expert Research I Statistical Consultant Systems Technician Administrative Assistant 3 Administrative Assistant 1 Research Assistant 1 Research Assistant 1 Assistant 1 Office Clerk 1 Research Assistant 3 Office Clerk 2 GIS Coordinator GIS Expert Office Clerk 3 28 9. Abstracts, papers and reports Rather than maintain the logframe structure for the detailed abstracts, papers and reports section, we have reordered into three broad categories which better represent the nature of the Agroecosystems Resilience Outcome Line. These are 1) technology and impact targeting, 2) sustainable and equitable use of ecosystem services, and 3) Climate change and risk. 29 9.1 Technology and Impact Targeting Participatory Impact Pathways Analysis: a practical method for project planning and evaluation Douthwaite, B.a, Alvarez, S.a, Thiele, G.b, Mackay, R.c a Centro Internacional de Agricultura Tropical (CIAT) Cali, Colombia. Centro Internacional de la Papa (CIP), Lima, Perú c Concordia University, Canada b Abstract Participatory Impact Pathways Analysis (PIPA) is a practical planning, and monitoring and evaluation approach developed for use with complex projects in the water and food sectors. PIPA begins with a participatory workshop where stakeholders make explicit their assumptions about how their project will achieve an impact. Participants construct problem trees, carry out a visioning exercise and draw network maps to help them clarify their ‘impact pathways’. These are then articulated in two logic models. The outcomes logic model describes the project’s medium term objectives in the form of hypotheses: which actors need to change, what are those changes and which strategies are needed to realise these changes. The impact logic model describes how, by helping to achieve the expected outcomes, the project will impact on people’s livelihoods. Participants derive outcome targets and milestones which are regularly revisited and revised as part of project monitoring and evaluation (M&E). PIPA goes beyond the traditional use of logic models and logframes by engaging stakeholders in a structured participatory process, promoting learning and providing a framework for ‘action research’ on processes of change. Introduction Project evaluation is currently used to: 1) communicate to donors the expected and actual impacts of the project; 2) show compliance with the agreed work plan, and negotiate changes to it; and 3) provide systematic information to support learning and decision making during the implementation of the project. Participatory Impact Pathways Analysis1 (PIPA) improves evaluation by allowing managers and staff to formalize their project’s impact pathways and to monitor progress, encouraging reflection, learning and adjustment along the way. Impact pathways are the detailed assumptions and hypotheses about how a project is expected to achieve its goal. They describe how individuals and organisations should act differently, strategies to bring this about, and how such change might impact on peoples’ livelihoods. Evaluators generally agree that it is good practice to first formalize a project’s impact pathways, and then evaluate the project against this ‘logic model” (e.g. Chen, 2005). In the CGIAR planning system, logic models are called ‘logical frameworks’, or ‘logframes’ for short. PIPA goes beyond the traditional use of logframes by: 1) involving key stakeholders in a joint process; 2) emphasizing the stakeholder networks needed to achieve impact; 3) providing the information 1 The Participatory Impact Pathways Analysis Wiki contains more information about PIPA: http://impactpathways.pbwiki.com 30 managers need both to learn and to report to their donors; and 4) establishing a research framework to examine the critical processes of change that projects seek to initiate and sustain. Development and use of PIPA PIPA grew out of ILAC funded work by the International Center for Tropical Agriculture (CIAT – Spanish acronym) on innovation histories (Douthwaite and Ashby, 2005) and work to evaluate impact pathways in an integrated weed management project in Nigeria (Douthwaite et al., 2003 and 2007). It was first used in a workshop in January 2006 when seven project teams, funded by the Challenge Program on Water and Food (CPWF), met for three days to co-construct their respective impact pathways in order to help the CPWF better understand the types of impacts its teams were envisioning. To date, staff from 44 CPWF projects have constructed their impact pathways in seven workshops. During 2008, PIPA will continue to be used for project planning and M&E by CPWF; by an EUfunded project in Latin America2, and by the International Potato Center (CIP - Spanish acronym) for ex-post evaluation purposes in the Andean Change Project. PIPA will also be used for ILAC’s own learning-based evaluation. PIPA is an umbrella term to describe both the participatory construction of impact pathways and their subsequent use. This brief focuses on the participatory monitoring and evaluation of progress along impact pathways. The use of impact pathways for ex-ante impact assessment is described in Douthwaite (et al., in press). Used ex-post PIPA involves using the PIPA workshop format to reconstruct impact pathways. More information on all aspects of PIPA, including an on-line manual, can be found at http://impactpathways.pbwiki.com. PIPA is similar in its philosophy to ‘outcome mapping’ (Earl et al. 2001). A main difference is that PIPA stretches participants to predict how project outcomes can lead to social, economic and environmental impacts. The PIPA workshop At the heart of PIPA is a participatory workshop in which project implementers and key stakeholders construct project impact pathways. Those who have contributed to a traditional logframe know that completing the required formats is tedious in groups and is often dominated by one or two people. Our experience is that when people are not constrained, at the outset, to fill in logframe boxes, they have tremendous energy for exploring collective ideas about how a project should work, or has worked. Therefore, in the PIPA workshop, participants only attempt to create a logic model once the underlying impact pathways have been discussed and agreed. The PIPA workshop is useful when two or more project teams in the same program wish to integrate better. At least two people for each project should attend; preferably this should include the project leader. The workshop also works well when one project team wishes to build common understanding and commitment with its stakeholders. In this case, two or more representatives from each important stakeholder group should attend. The ideal group size is four to six and the ideal number of groups is three to six. We have facilitated workshops with nine 2 EULACIAS – The European-Latin American Project on Co-Innovation in Agricultural Ecosystems 31 projects but this leaves little time for individual presentations and plenary, and participants tend to be overwhelmed by too much information. Day 1: Developing a cause-and-effect logic Participants spend most of Day 1 developing a problem tree for their project. Most people easily grasp the cause–effect logic of the problem tree, which begins with the identification of problems the project could potentially address and ends with problems that the project will directly address. When working with several projects from the same program, presentations of various problem trees help participants better understand each others’ aims, a prerequisite for successful programmatic integration. Figure 1: Presenting a problem tree in the Volta Basin Impact Pathways Workshop Day 2: Developing a network perspective Problem trees are seductively simple; they can lure people into thinking that solving a limited set of discrete problems begins a domino-like cascade which automatically achieves impact. Participants generally point this danger out themselves on Day 1. Day 2, therefore, is about balancing cause-effect logic with a network perspective, in which impact results from interactions between actors in an ‘innovation system’. These interactions can be modelled by drawing network maps showing important relationships between actors. To connect Day 1 with Day 2, participants construct a vision of success in which they imagine what the following classes of stakeholders will do differently after the project: 1. The users of project outputs, or ‘next users’; 2. Groups with whom the next users work; 3. Politically-important people and organizations who can help facilitate the project; 4. The project implementers themselves. 32 Next, participants draw a ‘now’ network map, showing current key relationships between stakeholders, and a ‘future’ network map showing how stakeholders should link together to achieve the vision. Participants then devise strategies to bring about the main changes. The influence and attitude of actors is explicitly considered during these exercises (see Figure 2(ii)) based on work by Schiffer (2007). Figure 2: Drawing network maps in a PIPA workshop (i) Drawing a network map (ii) Placement of influence towers and drawing of ‘smiley’ faces to indicate stakeholder attitude to the project Day 3: Developing the outcomes logic model and an M&E plan In the final part of the workshop, participants distil and integrate their cause-effect descriptions from the problem tree with the network view of project impact pathways into an outcomes logic model. This model describes in table format (see Table 1) how stakeholders (i.e. next users, end users, politically-important actors and project implementers) should act differently if the project is to achieve its vision. Each row describes changes in a particular actor’s knowledge, attitude, skills (KAS) and practice, and strategies to bring these changes about. The strategies include developing project outputs with next users and end users who subsequently employ them. The resulting changes are outcomes, hence the name of the model, which borrows in part from Bennett’s hierarchy (Bennett and Rockwell, 2000; Templeton, 2005) 33 Table 1: The outcomes logic model Actor (or group of actors who are expected to change in the same way) Change in practice required to achieve the project’s vision Change in KAS1 required to support this change Project strategies2 to bring about these changes in KAS and practice? 1 Knowledge, Attitude and Skills Project strategies include developing project outputs (knowledge, technology, etc.) with stakeholders, capacity building, communication, political lobbying, etc. 2 The outcomes logic model is the foundation for monitoring and evaluation because it provides the outcome hypotheses, in the form of predictions, which M&E sets out to test. The predictions are that, if key assumptions are met, the envisaged project strategies will help bring about desired changes in KAS and practice of respective actors. M&E requires that the predictions made in the outcomes logic model be made SMART (specific, measurable, attributable, realistic and time bound) so that project staff and stakeholders can know whether or not predictions are being realized. Hence, the next step in developing an M&E plan is to identify outcome targets, and milestones towards achieving them (see Table 2). Participants begin by prioritizing changes listed in the outcomes logic model in terms of what the project will actually do. Table 2: Format used for identifying outcome targets The key changes in KAS and practice that the project is responsible for Assumptions1 SMART outcome targets Means of verification? By whom? In what form? 1 Assumptions are conditions that are beyond the control of the project but which affect project success. For example, a key assumption for a project working to improve product quality (e.g. fish, rice etc.) is that farmers will receive a higher price for better quality. Moving from outcomes to impact After the workshop, participants may wish to formalize how changes described in the outcomes logic model help change the livelihoods of end users (for example when PIPA is being used for ex-ante impact assessment). In this case, we (the facilitators) use workshop outputs to construct a first draft of an impact logic model that shows the underlying cause-effect sequence of outputs, adoption, outcomes and long-term impact. We also draft a narrative explaining the underlying logic, assumptions and networks involved. These narratives have drawn on the ‘learning selection change’ theory (see http://boru.pbwiki.com/Learning+Selection+Change+Model). An 34 example of an impact logic model is shown in Figure 3, and the narrative describing it can be found at http://boru.pbwiki.com/f/PN06%20Impact%20Narrative-4.DOC. Figure 3: Example of an impact logic model for the CPWF Strategic Innovations in Dryland Farming Project Monitoring and evaluation After the workshop, participants complete their M&E plan with key staff and stakeholders. If M&E is to contribute to project learning, stakeholders should reflect on the validity of the impact hypotheses periodically, not just at the end of the project. We suggest that projects hold a reflection and adjustment workshop with their key stakeholders once a year with a smaller meeting in between. We use the graphic in Figure 4 to explain to participants how the reflection process works. The numbers below relate to the graphic. 1. During the PIPA workshop, participants develop a shared view of where they want to be in two years’ time, and describe impact pathways to achieve that vision. The project then implements strategies, which lead to changes in KAS and practice of the participants involved. 2. A workshop is held six months later to reflect on progress. The vision is changed to some extent, based on what has been learnt, the outcome hypotheses are revised when necessary and corresponding changes are made to project activities and strategies. New milestones are set for the next workshop. 3. The process continues. The project never achieves its vision (visions are generally used to motivate and stretch), but it does make real improvements. 35 1 Improvement Improvement 2 Vision Adjusted vision Impact pathways Future without intervention Adjusted impact pathways Reflection Impact Pathways Workshop 1 2 Time (years) 3 1 0 Improvement 0 2 3 Time (years) Adjusted visions Actual Improvements Periodic Reflections 0 1 2 3 Time (years) Figure 4: Reflecting on progress along impact pathways (based on Flood, 1999) These reflections are the culmination of one set of experiential learning cycles and the beginning of others. If the reflections are well documented, they can be analyzed at the end of the project to provide insights into how interventions do, or do not, achieve developmental outcomes in different contexts. PIPA M&E thus provides a framework for carrying out action research3. The quality of the research depends on the facilitation of the reflections, the data used and the documentation of the process. PIPA M&E is not prescriptive about the data used in the reflections, but does encourage researchers to gather data using multiple methods. It also recommends ways of introducing thematic and gender perspectives into the design of datagathering methods and reflection processes. One data-gathering method we have promoted in the EULACIAS project is the ‘most significant change’ approach, in particular for picking up unexpected consequences (see Davis and Dart, 2005). 3 See Douthwaite et al (2007) for a published example of evaluation of a project’s progress along its impact pathways 36 We have used PIPA-generated impact hypotheses as a basis for ex-ante impact assessment and are currently undertaking an impact assessment project to revisit them ex-post. More information on PIPA can be found at http://impactpathways.pbwiki.com. Conclusions Participatory Impact Pathways Analysis (PIPA) is a relatively young and experimental approach that involves the participatory generation of impact pathways and their subsequent use. Although this brief focuses on monitoring and evaluation, PIPA is also used for ex-ante and ex-post impact assessment. We encourage readers to experiment with PIPA and contribute to its development. More information on all aspects of PIPA, including an on-line manual, can be found at http://impactpathways.pbwiki.com. References Bennett, C. F. and Rockwell C. (2000) ‘Targeting outcomes of program (TOP)’, available at citnews.unl.edu/TOP/english/ Chen, H. T. (2005) Practical Program Evaluation: Assessing and Improving Planning, Implementation, and Effectiveness, Sage Publications, ,CA Davies, R. and Dart, J. (2005) ‘The most significant change technique: a guide to its use’, available at www.mande.co.uk/docs/MSCGuide.htm Douthwaite, B., Alvarez, B. S., Cook, S., Davies, R., George, P., Howell, J., Mackay, R. and Rubiano, J. (in press) ‘The impact pathways approach: a practical application of program theory in research-for-development’, Canadian Journal of Program Evaluation, 22(2), Fall 2007. Douthwaite, B. and Ashby, J. (2005) ‘Innovation histories: a method for learning from experience’, ILAC Brief 5, IPGRI, Rome, available at www.cgiarilac.org/downloads/Briefs/Brief5Proof2.pdf Douthwaite, B., Schulz, S., Olanrewaju, A., Ellis-Jones, J. (2007) ‘Impact pathway evaluation of an integrated Striga hermonthica control project in northern Nigeria’, Agricultural Systems vol 92, pp201-222 Earl, S., Carden, F. and Smutylo, T. (2001) Outcome Mapping: Building Learning and Reflection into Development Programs, International Development Research Centre, Ottawa, Canada Flood, R. L. (1999) Rethinking the Fifth Discipline, Routledge, London and New York Schiffer, E. (2007) ‘The power mapping tool: a method for the empirical research of power relations’, IFPRI Discussion Paper 00703 37 Templeton, D. (2005) ‘Outcomes: evaluating agricultural research projects to achieve and to measure impact’, in I. Metcalfe, B. Holloway, J. McWilliam, and N. Inall (eds) Research Management in Agriculture: A Manual for the Twenty First Century, University of New England, Armidale, Australia 38 Participatory Impact Pathways Analysis: A Practical Application of Program Theory in Research-for-Development Douthwaite, B.a, Alvarez, S.a, Cook, S.b, Davies, R.c, George, P.d, Howell, J.e, Mackay, R.f, Rubiano, J.g a Centrro Internacional de Agricultura Tropical (CIAT) Cali, Colombia. Basin Focal Projects, CPWF and CIAT c Independent M&E Specialist d CPWF Program Manager, IWMI, Colombo, Sri Lanka e M&E Specialist, Living Resources, UK f Concordia University, Canada g National University of Colombia, Palmira, Colombia b Abstract The Challenge Program on Water and Food pursues impacts on food security and poverty alleviation through the efforts of some 50 research-for-development projects. These involve almost 200 organizations working in nine river basins around the world. An approach was developed to enhance the developmental impact of the program through better impact assessment, provide a framework for monitoring and evaluation, permit stakeholders to derive strategic and programmatic lessons for future initiatives, and provide information that can be used to inform public awareness efforts. The approach makes explicit a project’s program theory by describing its impact pathways in terms of a logic model and network maps. A narrative combines the logic model and the network maps into a single explanatory account and adds to overall plausibility by explaining the steps in the logic model and the key risks and assumptions. Participatory Impact Pathways Analysis is based on concepts related to program theory drawn from the fields of evaluation, organizational learning, and social network analysis. Background Participatory Impact Pathways Analysis (PIPA) described in this paper was developed within the context of a large and complex, five-year, research-for-development (R4D) program–the Challenge Program on Water and Food (CPWF, http://forum.waterandfood.org/). The key dimensions of impact pursued by CPWF are (i) food security, (ii) poverty alleviation, (iii) improved health, and (iv) environmental security. The program is geographically extensive, covering the Limpopo, Nile, Yellow, São Francisco, Karkheh, Mekong, Nile and Volta river basins, and the Andean system of basins. It currently funds 51 projects that are implemented by 198 different institutions including the Consultative Group for International Agricultural Research (CGIAR) Centres4, advanced research institutes (ARIs), NGOs, community-based organizations (CBOs), and national agricultural research and extension organizations. The partnerships and the research are coordinated by basin coordinators (one for each basin) and five theme leaders. There are three systems level research themes–crop water productivity improvement, water and people in catchments, and aquatic ecosystems and fisheries; one basin 4 The CGIAR System comprises of 15 international agricultural research centres carrying out research-fordevelopment. For more information see www.cgiar.org 39 level theme–integrated water basin management systems; and one global scale theme–global and national water and food systems. The first five-year phase of the program began in 2004 and operates with a budget of approximately US$66m for the five -year period. The CPWF is “impact-oriented” which means the performance of the program and its projects is being evaluated not just on the delivery of research outputs, but on how those outputs are used, by whom, and to what effect (Ryder-Smith, 2002). The CPWF will be judged successful if it can demonstrate that the research it has supported has in a meaningful way “increased the productivity of water for food and livelihoods, in a manner that is environmentally sustainable and socially acceptable”5 in and beyond the river basins in which it works.” If the CPWF and its constituent projects are to be successful they must be managed for impact, that is, projects must plan and manage to achieve development outcomes, not just to deliver the outputs listed in their project documents (Ryder-Smith, 2002). Managing to achieve developmental outcomes is more challenging than managing for outputs because, while projects can largely control whether they deliver their outputs, many factors in addition to research contribute to achieving developmental outcomes (Mayne, 2004; Hartwich and Springer-Heinze, 2004). A second challenge facing the CPWF is securing adequate funding streams for long enough to achieve measurable developmental outcomes. It can take 10 years to move from basic research to useful technologies and then another 10 years to see wide-scale impacts (Collinson and Tollens, 1994). The CPWF generally commissions projects on a 3 to 5 year basis. Hence the CPWF needs an ex-ante impact assessment approach that can plausibly demonstrate to donors how project outputs will lead to development outcomes and widespread impacts after the end of the projects that developed them. The ever increasing challenges facing the CPWF are those faced by all medium and large-scale R4D programs. This paper reports efforts to date by the CPWF’s informal Impact Group (the authors of this paper) to develop Participatory Impact Pathways Analysis (PIPA) to meet these challenges, specifically to: 1. Present the logic that explains how project activities and outputs are hypothezised to contribute to a sequence of outcomes and impacts. 2. Facilitate development of shared understanding of, and agreement with, the project logic among project team members. 3. Provide the basis of a plausible ex-ante impact assessment methodology for the CPWF that will also provide a solid foundation for later ex-post impact assessment 4. Provide the basis for monitoring and evaluation that fosters learning and change in the CPWF. 5. Clarify and communicate the research-for-development processes out of which impact emerges The first section of this paper introduces the “impact challenge” facing complex programs such as the CPWF. The second explores the characteristics required of Participatory Impact Pathways Analysis (PIPA). The third describes PIPA in terms of its component parts and their relation to 5 This is the CPWF’s Development Objective 40 existing tools and approaches, and the literature. The fourth offers an account of how PIPA is used in practice with CPWF projects and their teams. The paper concludes with a discussion of the value added by PIPA to agricultural R4D and to the practice of evaluation in general. The “Impact Challenge” Facing R4D Projects and Programs The success of R4D projects and programs such as the CPWF depends upon achievement of intended results. This, in turn, depends on (i) sound project and program management geared to meeting the outcome expectations of funding agencies and (ii) maintaining and increasing resources as projects proceed beyond the pilot stages and the program gathers momentum. There is a close-knit relationship between these two issues particularly when funds come from diverse sources. Convincing arguments are required to persuade multiple funding agencies of the likely potential uptake of research products and services by networks of diverse partner organizations and of the eventual impacts of these on a wide range of beneficiaries. Without an initial wellfounded and persuasive ex-ante account of how project managers, basin coordinators and theme leaders predict their projects will have impact, and later ex-post evidence of impact, the executing organizations’ efficacy and their very right to existence is cast in doubt (Ryder-Smith, 2002; OECD, 2006; Anderson, Bos and Cohen, 2005). Both management and funds are vulnerable without critical and timely information for informed decision-making and effective ways of communicating anticipated and actual results to funding agencies. This information should come from monitoring and evaluation and, initially, from ex ante impact assessment. Plausible impact assessment must quantify impacts achieved or to be achieved and then make a convincing case that the project or program being assessed will contribute or has contributed to that impact (EIARD, 2003). To be able to do so requires understanding and communication of the R4D processes being employed, and the theory or theories supporting them. Monitoring and evaluation has the potential to provide this information but often does not, in part because evaluative inquiry as an organizational learning system is highly underdeveloped (Cousins et al., 2004). It is not uncommon to keep impact assessment and monitoring and evaluation separate. For example, in the CGIAR System, within which this work is being conducted, impact assessment, both ex-ante and ex-post, has been viewed as a legitimate research activity while M&E has been viewed as an accountability mechanism but not contributing to research (Horton, 1998). M&E in the CGIAR has largely been based on the use of logical frameworks to identify and report on milestones, which in a research sense is of limited interest. The logical framework was originally developed by the US Department of Defence in the late 1960s (Horton et al. 1993 p. 113) and since then has been modified and widely used by development agencies throughout the world (Rush and Ogborne, 1991; Cedric, Cedric, Saldanha and Whittle, 1998; Schmitz and Parsons, 1999; Kellogg Foundation, 2004) as well as in the private and public sectors (McLaughlin and Jordan, 1998; Cooksey et al., 2001). The logical framework builds a causal chain of how a project or program will achieve its development goal (Figure 1). The chain begins with identifying activities and shows how these will produce project outputs if a certain set of assumptions and necessary conditions are met. The next step in the hierarchy is to show how outputs will achieve the project purpose and then how that purpose achieves the goal, or final expected impact. 41 While the typical logical framework does show a causal chain, in practice it tends to be a very simple one, often with just one level of outcomes between production of project outputs and the eventual goal. In practice, whole chains of intermediate outcomes link project outputs with eventual impact. Also the opportunity for a detailed description of causality within the logical framework tends to be weak and provides only superficial explanations of causation. More seriously, logframes can lead to a false idea of the linearity and predictability of impact pathways which project and program managers find seductive. As a result, managers tend to stick with their original logframes developed at the outset and do not regularly revisit them to reassess the underlying assumptions. Longer-term outcomes resulting from the purpose GOAL THEN Medium-term outcomes resulting from use of outputs PURPOSE IF THEN What the project produces that others use OUTPUTS ACTIVITIES Assumptions and necessary conditions IF THEN What the project does with its resources Assumptions and necessary conditions Assumptions and necessary conditions IF Figure 1: The Logical Framework The potential of program theory In recent years a number of R4D scientists have increasingly begun to look beyond logical frameworks to program theory to help remedy this lacuna (Horton, 1998; Douthwaite et al. 2003). Logic modelling is largely limited to normative theory–what is expected to happen. Program theory is concerned with both normative and causative theory (Chen, 2005). Causative theory explains how use of project outputs lead to a chain of intermediate outcomes and eventual impact. It is an explanation of process based on either stakeholder theory or scientific theory. Examples of scientific theory is the published learning-selection model of early grassroots adoption and adaptation of technology (Douthwaite, 2002; Rogers, 2003) innovation decision process. Scientific theory is different to stakeholder theory as Chen (2005, p. 41) explains: “Stakeholder theory is implicit theory. It is not endowed with prestige and attention as is scientific theory; it is, however, very important from a practical standpoint because stakeholders draw on it when contemplating their program's organization, intervention procedures, and client-targeting strategies. Stakeholders' implicit theories are not likely to 42 be systematically and explicitly articulated, and so it is up to evaluators to help stakeholders elaborate their ideas.” The use of program theory in R4D projects offers a number of benefits. Evaluators would help project staff to articulate their implicit theories and where appropriate suggest appropriate scientific theory on which to base all or part the project or program’s causative theory. Subsequent M&E would then become tools in a legitimate research exercise that would contribute to knowledge by: (i) testing stakeholder-implicit theory and potentially establishing it as new scientific theory; and, (ii) validating scientific theory in different conditions. M&E of the validity of a project’s causative theory would support learning and change and adaptive project management, thus making project impact more likely. Information from M&E would also help refine the causative theory and contribute to process knowledge about how research outputs do, or do not, lead to developmental outcomes and impacts. Such process understanding can help improve the plausibility of qualitative ex-ante and ex-post impact assessment. Some donors have begun calling for changes in evaluation and impact assessment practice in R4D projects, changes that program theory could help deliver. The Task Force on Impact Assessment and Evaluation, European Initiative for Agricultural Research for Development (EIARD), a group of European donor agencies, wrote: “Impact assessments and evaluations should not be limited to directly measurable impacts; they should seek to capture the complexity and non-linear nature of agricultural innovation and sustainable development. Impact assessments and evaluations should also be integrated as far as possible into research programmes, to facilitate internal learning processes and changes that enhance the probability of impact.” (EIARD, 2003, p. 329) EIARD (2003) then went on to recommend that evaluators make explicit the model of how innovation occurs both for ex-ante and ex-post impact assessment. Program theory is starting to be used in R4D projects. Douthwaite et al. (in press) report the use of impact pathways evaluation to monitor and evaluate the development, adaptation and adoption of integrated weed management techniques in Northern Nigeria. Impact pathways evaluation develops and uses a causal model of how adoption and adaptation is expected to take place, and makes explicit mention of its roots in program theory (Douthwaite et al., 2003). The Association for Strengthening Agricultural Research in Eastern and Central Africa, (ASARECA, 1999), uses an impact chain to represent the several intermediate steps and actors along the way to impact. Projects and programs use their resources through planned activities to produce outputs. With the intervention of other actors these outputs are transformed into outcomes. The resulting impact chain is characterized by a time dimension and organizational level. Depending on the complexity of the impact chain, ASARECA acknowledges that it can become difficult to ascertain the proportion of credit due to which actor for what impacts -- the classical “attribution problem”. While the ASARECA approach goes beyond the simple logical framework by allowing the identification of chains of intermediate outcomes, and by introducing an organizational dimension, it falls short of program theory as it does not make causal theory explicit. 43 The International Development Research Centre (IDRC) has been working for a number of years to develop Outcome Mapping that focuses on making explicit the changes in behaviour that are expected as a result of project and program intervention (Earl et al. 2001). Outcome Mapping is similar to PIPA in a number of ways. Like PIPA, Outcome Mapping usually begins with a participatory workshop, it takes a learning-based and use-driven view of evaluation, and it identifies the stakeholders that the project needs to influence to achieve its expected outcomes. PIPA is different in two important aspects. Firstly, PIPA attempts to integrate both a results- and actor-orientated view, while Outcome Mapping focuses on the latter (Ambrose, 2007). The use of problem trees in PIPA makes it more accessible to project staff already used to working with logic models. Secondly, PIPA uses network mapping to explore how stakeholders are linked to, and influence each other, and how the project aims to change the existing network. Outcome Mapping does not consider this dimension, taking more of a project-centric view. Hartwich and Springer-Heinze (2004, p.5) argue for improving the impact orientation of agricultural research by means of impact pathways. However their conceptualization of an impact pathway is similar to the logical framework with just one level of outcome. The CGIAR Science Council also encourages progressing beyond the normative use of logical frameworks. The Science Council’s mission is to “enhance and promote the quality, relevance and impact of science in the CGIAR” and one of the functions it plays is to analyze CGIAR Centres’ medium term plans (www.sciencecouncil.cgiar.org). The Science Council recently requested that CGIAR Centres prepare for each CGIAR Centre project a “description of the plausible impact pathway from research outputs through outcomes to the ultimate impacts” (Science Council, 2006, p. 3). They acknowledge that the logical framework they ask to be prepared is by definition “only a simplified version of the impact pathway from outputs to outcomes to one level of intended impacts” (Science Council, p. 5). The Science Council requests that the plausible account of the full impact pathway be given in a written description called the “project narrative”. A plausible narrative would imply some discussion of theories of causality, and would be greatly helped by the use of program theory. We have, so far, argued that R4D projects and programs should increasingly use program theory because it has the potential to (i) raise the status of M&E to a research activity and thus be more likely to be taken seriously and attract resources, (ii) provide sound assessments of what changes will or might occur, (iii) provide descriptions of how project research outputs might or have achieved developmental outcomes and impact, and, (iv) provide process information to assist project and program management as well as to improve ex-ante and ex-post impact assessment. Program theory is already being used in a R4D context under the name of “impact pathways” and we choose to continue this tradition. The CPWF’s requirements for Impact Pathways (IP) Analysis In collaboration with other CPWF participants, the Impact Group agreed upon three general and two technical characteristics that IP Analysis must fulfil to meet the requirements of the CPWF. In general terms, it must be capable of providing (i) a better appreciation of the existing and potential impact of research on water use in agriculture to justify current and future funding, (ii) a deeper understanding of what impacts the CPWF expects to attain, and how and (iii) a 44 framework for an effective M&E approach that fosters and tracks progress towards achieving impact. In more technical terms, the model must also be capable of (i) making explicit each project’s causative theories and, (ii) generating quantifiable measures of the likely intermediate and final outcomes and impacts for which managers and funders hold the projects accountable. Design of Participatory Impact Pathways Analysis (PIPA) We chose to base PIPA on ideas from program theory (Chen, 2005), organizational learning (Argyris and Schön, 1974) and network theory (Cross and Parker, 2004). The characteristics of PIPA will be discussed in terms of the two technical requirements. Make project causative theory explicit Causative theory describes how project and program research outputs are adopted and promulgated. There has been an increasing recognition in agricultural R4D that two types of adoption are important: scaling-out and scaling-up. Scaling-out is the increasing adoption of project outputs from farmer to farmer, community to community, within the same stakeholder groups. It is a horizontal spread, as shown in Figure 2. Scaling-up is a vertical institutional expansion, based largely on a desire or need to change the rules of the game. It can be driven by the influence of first-hand experience, word-of-mouth and positive feed back, from adopters and their grassroots organizations on policy makers, donors, development institutions, and the other stakeholders who then have an interest in building a more enabling environment for the scalingout process. Sometimes the process is reversed and driven by political conviction. Interventions at a higher scale, for example policy research, can affect scaling-out processes at lower ones, as shown in Figure 2. 45 Figure 2: The concepts of scaling-out and scaling-up (Douthwaite et al. 2003) Combining logic models with network maps In PIPA project impact pathways are described in terms of an Impact Pathways (IP) logic model and network maps. The IP logic model is a flowchart that shows the chains of outcomes that link outputs to eventual developmental impacts. It is similar to Chen’s (2005) change model, except that where possible it incorporates one or more published (confirmed) causative theories as recommended by Renger and Titcomb (2002). The network maps give additional detail to the causative theory. PIPA builds on an innovation systems perspective that recognizes that scaling-out and –up are brought about by the formation and actions of networks of stakeholders in what is essentially a social process of communication and negotiation (Douthwaite, 2002; Hall, Mytelka and Oyeyinka, 2004). Network maps are drawn for the beginning of the project and for the future, usually two years after the project has finished. The “future” network is essential for the project to achieve eventual impact, because if no one is using or promulgating the project outputs after the end of the project, the project will not achieve its goal. Clarifying and making explicit how the project will build its “future” network helps project staff identify the key stakeholders that the project needs to engage with to achieve scaling-out and scaling-up of project outputs. The network maps are crucial to PIPA. The network maps include the ‘softer’ behavioral and relational dimensions of a project or program’s impact pathways, complementing the ‘harder’ mechanistic description given by the IP logic model. A number of writers have identified the 46 need to blend ‘hard’ and ‘soft’ to gain a fuller understanding of change and innovation processes (Checkland and Scholes, 1990; Douthwaite et al. 2001; Campbell et al. 2001). The network maps also help compensate for a weakness of logical frameworks and other types of logic models that do not give sufficient information about the actors involved in bringing about developmental change. For example, logical frameworks commonly contain narrative statements in them without people, “rice yields increased by 25% in pilot sites”. Network maps play a similar function to the concept of ‘reach’ (Montague, 1997) introduced to provide actor information in traditional logical frameworks; (McLaughlin and Jordan, 1998; Mayne, 2001). Reach refers to the sphere of influence -- i.e. the "with whom?" (partners and stakeholders),"for whom?" (direct and indirect beneficiaries) and "how many or how much?" (proportion of beneficiaries) -- over which an organization wishes to spread its resources. EIARD (2003) has noted that agricultural development comes about through complex and non linear processes. This reality is not represented in logic models, but it is implicit in network maps. Network maps show relationships between actors involved in an innovation process and can: “incorporate mutual and circular processes of influence as well as simple linear processes of change. This enables them to represent systems of relationships exhibiting varying degrees of complexity and chaos.” (Davies, 2003, p. 2). Integrated impact narrativ The IP logic model and the network maps are woven together by an impact narrative. We, and others, have found that textual descriptions can make up for or supplement the incompleteness that is an inevitable concomitant of flow-charts, diagrams, and matrices, useful as these undoubtedly are (Cooksy et al. 2001; Mayne, 2004). The impact narrative describes the relationships between the outcomes in the IP logic model with the network maps. By virtue of the demand that the narrative create an integrated unity, the IP group and project personnel find that the process of creating it subjects the assumptions on which the project is based to exacting scrutiny. This enhances the comprehensibility and reinforces the plausibility of both the logic model and the network maps, and hence the overall causative theory. This scrutiny helps project managers and staff to develop a better, more robust and complete impact pathways for their project or program. The impact narrative is more than the more traditional “narrative summary” that accompanies a logical framework. That is usually little no more than a statement of each of the project’s goals, outputs, and activities and inputs (Horton et al., 1993). It is also substantially richer than the stand-alone “impact narrative” used to provide an account of significant program efforts and milestones and the effects of the program on its target population (Taylor and Fugate, 1993; Hamilton, 2005). It is similar to Mayne’s (2004) “performance stories” although CPWF impact narratives, because of their ex ante orientation, explain what is expected to happen while performance stories recount what has happened. In terms of the relationship between program theory and theories of action (Figure 8) the whole process of developing the IP logic model, the network maps and then writing the impact narrative works to improve the project or program’s espoused theory about how they will 47 achieve impact by making explicit project members’ theories-in-use. The process used to construct project and program impact pathways (i.e., program theory) is described in the next section. Quantifiable measures of outcomes and impacts The Impact Group’s IP logic model goes further than identification of the likely intermediate and final outcomes and impacts. It quantifies these so that managers and funding agencies can be clear about the magnitude, in appropriate units of measurement, of what is expected from the project. Mayne (2004) has highlighted the importance of having clear, quantified statements of expectations. It is not practicable to measure everything, but without a concrete statement of expected results, – “….. all one has is results information.” (op. cit. p. 34). The two quantitative techniques are geographic extrapolation domain analysis and scenario analysis. The effective use of the latter depends upon the prior execution of the former and so geographic extrapolation domain analysis will be discussed first. Geographic extrapolation domain analysis Simply stated, geographic extrapolation domain (GED) analysis helps identify where one would expect a technology to be adopted. GED analysis uses Weight of Evidence (WoE) techniques using data from geographic databases to calculate where in the tropics one is likely to find areas with similar socio-economic and agro-ecological conditions as found in CPWF project pilot sites. The purpose is to determine, ex-ante, the sites most likely to offer the potential for successful adoption of research products and services generated by CPWF. With this information, the project and/or the CPWF can then plan to scale out into areas that offer the greatest likelihood of success so as to augment and maximize their impact and thereby optimize the use of the financial contributions of the agencies funding the research. GED analysis is so far unable to take into account similarities between the institutional environments of sites in the most probable replication areas making the technique less useful for the purposes of determining the success of scaling-up. Indeed, it is unlikely that GED or any other quantitative technique will ever be able to account for the any uncontrolled institutional factors that influence results in different social contexts (Dahler-Larsen, 2001). Scenario analysis Scenario analysis has gained in importance over more predictive approaches in a number of global environmental assessments over the last 20 years, because it allows for including surprises and unexpected developments outside of currently existing boundary conditions. Scenario analysis is used to quantify project impact pathways over a 25-year time scale. The analysis is carried out using an existing water and food supply and demand quantitative modeling framework called IMPACT-WATER. The framework allows economic policies, including trade policies, and climate outcomes of other basins and regions to be taken into account when building scenarios for the impact of different project research outcomes. 48 How impact pathways are developed for CPWF projects Project impact pathways are developed basin by basin. The process begins with an Impact Pathways Workshop at which two or more representatives from each project work to develop the inputs required to build their project’s IP logic models and network maps. The workshop is facilitated by members of the Impact Group. A “road map” of the entire process is shown in Figure 3. The purpose of the workshop is to clarify and surface the participants often implicit program theory. The first part of the workshop clarifies a linear “logic model” view of the project’s impact pathways, that focuses on outputs and outcomes. The second part clarifies an actor-orientated view focussing on the relationships needed to achieve impact. Developing an actororientate view of a project's IP Developing a results-orientated view of a project's IP PRODUCTS PRODUCED DURING WORKSHOP PRODUCTS PRODUCED AFTER WORKSHOP Causal Analysis / Problem Tree (Draft produced before workshop) Helps understand project rationale and what needs to change Extrapolation Domain Analysis Outputs Vision What the project will produce Where project is going- Goal "Now" network map Necessary relationships in place to produce the OUTPUTS "Future" network map Scenario Analysis Logic model (Results-orientated) Two descriptions of the project's impact pathways Iterative process Impact Narrative Necessary relationships to achieve the VISION Network maps Integration of both views (Actor-orientated) Project Timeline How project goes from outputs to vision Figure 3: The PIPA Process Impact Pathways Workshop: clarifying and making participants’ program theory explicit The nature of the workshops Workshops employ strategies for participation and the sharing of power that have already proven successful in earlier CGIAR projects involving evaluative inquiry and capacity development (Horton 2001). These strategies derive from principles of “negotiated rationality” (Lincoln and Guba, 1985; Guba and Lincoln, 1989) and “deliberative, democratic evaluation” (House, 2004). They include the inclusion of all participating stakeholder views, a willingness to share power, extensive dialogue to make value positions explicit, and deliberation to allow parties to change their positions if they encounter new and persuasive information. 49 A negotiated process for developing the impact pathways model for each project is time consuming and can be expensive. However, it is an effective process to ensure that stakeholder reality drives the IP models and not merely researcher assumptions. Value for money is exacted from the process by using the workshops as occasions for capacity building and for exchanging information from similar but widely dispersed projects. Unit of analysis The unit of analysis of PIPA is the project because this is what the CPWF funds. CPWF Projects last for 3 to 5 years, while it can take 20 years to go from basic research to developmental impact (Collinson and Tollens, 1994). A CPWF Project therefore cannot expect to achieve highly aggregated developmental impacts such as poverty reduction in the lifetime of the project. Nevertheless, workshop participants are stretched to think and plan beyond their current projects The diagram in Figure 4 is presented to workshop participants and the point made that while a project has little control over whether it achieves impact, that influence is not zero and can be maximized by identifying impact pathways and following them during the project cycle. Impact pathways may well involve looking for subsequent project funding after the end of the current one. Figure 4: Project influence on outputs, outcomes and impact Clarifying a linear view of a project’s impact pathways In preparation for an Impact Pathways workshop, the first two authors develop a draft problem tree for each project from the respective project proposals. This is considered necessary because CPWF project proposals are written in different styles, and generally do not use logical frameworks. It can be quite difficult for an outsider to grasp the project’s program theory. A problem tree is a visual problem-analysis tool used to identify problem situations and their key 50 causes starting with the root cause. We, and others (Renger and Titcomb, 2002) have found that it is an excellent tool for clarifying, building and communicating a project’s underlying logic. The managers and staff of each project are asked to reflect on the draft problem tree and to bring their own comments and modifications with them to the workshop. The first exercise in the workshop (see Figure 3) is for the project groups to modify and redraw their problem trees on cards and poster paper and then present them in plenary (see Figure 5). The next exercise is for the project groups to convert their problem trees into objective trees. This involves reframing the problem positively by describing the situation when the problem has been solved. For example, “food insecurity” becomes “food security”. The idea of reframing in the positive is shared with Appreciative Inquiry (Whitney and Trosten-Bloom, 2003) and other so-called “asset-based” approaches which have found that people are more motivated by positive outcomes than by problems. Figure 5: Constructing and presenting project problem trees helps clarify a linear view of a project’s impact pathways Photo: Boru Douthwaite, taken January 2006 in Volta Impact Pathways Workshop Constructing the problem tree helps clarify which problems the project is tackling and hence what its outputs should be. The next step in the workshop is for each project to construct a vision of project success two years after the end of the project. The visioning exercise is adapted from Appreciate Inquiry and is based on the question: “You wake up two years after the end of your project. Your project has been a success and is well on its way to achieving its goal. Describe what this success looks like: • What is happening differently now? • Who is doing what differently? • What have been the changes in the lives of the people using the project outputs, and who they interact with? • How are project outputs scaling-out and scaling-up?” 51 The visioning exercise has proved very useful because usually existing project espoused theory about goals are caged in very general terms, if described at all. The vision also provides the context for the “future” actor network map constructed in the second part of the workshop. An context for the “future” actor network map constructed in the second part of the workshop. An example of a project vision is shown in Box 1. Box 1: Example of a project vision – CENESTA What is happening differently now? • • • Extension and research are working together to support farmer-led research and are working with local community-based organizations as their interface in Honam and Merek Local communities are better organized; their organizations are based on traditional water and natural resource management organizations; have revived use of traditional knowledge and institutions; have local legitimacy and also recognized by the government Enhanced water productivity with positive impact on livelihoods Who is doing what differently? • Extension and research are working together and both working at the service of farmers and pastoralists • Local communities more independent: solving their own problems and conflicts • Government is starting to develop policies for the Karkheh River Basin as a whole What have been the changes in the lives of the people using the project outputs, and who they interact with? • Greater self-confidence among local communities • Better relationship between government and local communities • Farmers/pastoralists are using more productive and appropriate technologies and producing more food • Local communities have learned how to develop participatory technologies based on traditional knowledge and new technologies to improve their livelihoods and this is starting to spread to other communities • Greater farmer income How are project outputs disseminating? • By local community-based organizations with support from the Government where hended What political support is nurturing this spread? How did that happen? • Growing political support for cooperation between research and extension to serve farmers better in technology development and extension • Growing political support for the role of customary institutions • Political support gained by showing productivity gains with these new approaches which leads to food self-sufficiency (national policy) and more efficient use of government resources The final exercise in this first part of the workshop is for the project groups to develop a timeline of key events and activities that show how the project outputs are developed and then what needs to happen to those project outputs to achieve the vision. 52 Clarifying an actor-orientated view of a project’s impact pathways The second stage of the workshop involves asking participants to construct two network maps, one for the present and one corresponding to their vision, two years after the end of the project. The participants are also asked to transfer the map data into matrices. The “now” network map shows the existing relationships between the project partners and their links to other stakeholders and the ultimate beneficiaries of the project outputs. The relationships mapped include research, provision of funding, scaling-out and scaling-up. The “future” network shows the relationships that the participants think are necessary to achieve their respective visions. Before participants draw this network the facilitator reminds them of the concepts of scaling-out and scaling-up, and stresses that their respective projects will only achieve their vision and goal if a network of organizations actively works to scale-out and scale-up their project outputs after the end of the project. Once the two maps are drawn, the facilitator then asks them to compare and contrast them. They are also told that if the “future” map is very different from the “now” map, and usually it is, then this implies that the project must work to build these new relationships before the end of the project as the relationships are unlikely to spontaneously emerge afterwards. This need to forge new relationships suggests additional ways of working with existing partners and points at which new stakeholders should enter the project. Participants develop a relationship action plan as part of the workshop. After the Workshop Development of project IP logic models After the workshop the facilitators in their role as evaluators synthesize the objectives tree, the project outputs, vision and timeline into the project IP logic model. The IP logic model is a flow chart that shows both scaling-out and scaling-up processes (Figure 6) by which project outputs are increasingly used and promulgated such that they contribute to developmental outcomes. A published causative theory is integrated into the IP logic models of the projects carrying out participatory research in pilot sites. The theory describes how scaling-out and scaling-up occur as a result of iterative and interactive experiential learning (Douthwaite et al. 2003). The narrative for this change model is as follows: The project partners work in the pilot sites to develop, adapt and validate new technologies and their use strategies, in partnership with key stakeholders. The pilot site trials lead to the participants—farmers, scientists, extension workers, etc.—going through experiential learning cycles that lead to individual and collective changes in attitudes and perceptions, experimentation, adaptation and collective changes in attitudes and perceptions, experimentation, adaptation and adoption (Box 2, Figure 6). End-user adoption increases in the pilot sites based on positive feedback and promotion by the first adopters, and scaling-out begins as the technologies and strategies begin to spread to other villages. At the same time scaling-up begins as the project partners and stakeholders, who are taking part in the field work, gain ownership of the project outputs and begin to promote them in their own organizations. Early adopters begin to see real increases in income as a result of adoption and this helps fuel continuing positive feedback which drives an acceleration of adoption from farmer to farmer (scaling-out). 53 Positive feedback also drives an increase in institutional knowledge and support for the project outputs (scaling-up). 54 Figure 6: IP Logic Model for the Strategic-Innovations-in-Dryland-Farming Project Project Activities carried out in Pilot Sites with stakeholders and ultimate beneficiaries 1 Crop Related Outputs Crop Related Outcomes 3 Crop production guides or manuals for MoFA Scaling Out Best-bet soil and water conservation and management options manuals Changes in stakeholders attitudes and perceptions 2 Improved cropping systems in Northern Ghana Higher crop yields 8 5 Farmers plant to Farmers routinely avoid crop loss due generate organic to draught, majority matter , e.g. have intensified composting and cropping systems cover cropping Drought tolerant varieties developed Improved knowledge of stakeholders at pilot sites Farmers using appropriate tillage methods to conserve soil moisture Farmers using drought probability map and drought tolerant varieties Drought probability map Soil and water conservation improved in farmlands in N. Ghana Scaling Up Adoption of National variety release committee releases varieties Iterations of technologies learning and changes cycle in practice Wider adoption of project outputs beyond pilot sites 7 Adoption of project outputs by MoFA for extension after project finishes Stakeholders modify and innovate Communities trained on efficient fish production techniques Manuals on fish culture in dugouts and dugout maintenance Manuals on appropriate water harvesting systems 11 up Scaling Methods developed to institutionalize dialogue about water use among multiple users Water Related Outputs 4 Scaling Out Dugouts enhanced to retain water Communities have knowledge of low-cost domestic waterharvesting systems Improved soil fertility More time for income generating activities for women More water available for domestic needs Adequate water supply for dry season agriculture Project Goals Improved income for rural households High labour productivity Improved food security and rural livelihoods High land and water productivity 11 Reduction in water related diseases Community dugouts efficiently utilized for fish production Changes to housing structure to meet water harvesting needs Water Users Associations formed and strengthened Improved utility of community dugouts 10 Majority of communities in Northern Ghana have constructed and are using domestic water harvesting systems Effective management of community water resources 6 Water Related Outcomes 9 55 Drawing project network maps The Impact Group takes the network maps and matrices drawn in the workshop and redraws them using the Social Network Analysis (SNA) software package UCINET and NetDraw in order to make them easier to understand and use. The maps drawn in the workshops show all the relationships (e.g., research, provision of funding, scaling-out) and while useful for showing which are the most central (i.e., most linked) actors, they can be somewhat confusing. The software allows separate maps to be drawn for each relationship which has proven invaluable for clarifying theory-in-use about how relationships currently work and how they need to change in the future. This clarification comes through an iterative question and answer process involved in writing the Impact Narrative. Writing the Impact Narrative The first step in writing the Impact Narrative is that the Impact Group sends the draft project IP logic model and network maps back to the workshop participants, together with clarifying questions. If the project works in pilot sites, then we explain the Douthwaite et al. (2003) scaling-out and scaling-up theory-of-action and ask them if it applies to their project. Members of the Impact Group, again in their role as evaluators, then write the first drafts of the Impact Narratives based on the answers. This in turn throws up more questions and clarifications. In each round we press the workshop participants to quantify expected outcomes as much as possible for the reasons expressed earlier. The iterative process of writing the impact narrative changes both the IP logic model and network maps as the projects’ respective program theory improves and becomes clearer. For example, the Strategic-Innovations-in-Dryland-Farming Project’s scaling-out network maps changed radically (Figure 7). The process helped the project clarify that they expect seven different organizations, including their own, to be involved in extending project outputs to the ultimate beneficiaries. At present only three organizations are doing this, so this implies that before the end of the project they need to forge relationships with four new organizations. Not all these relationships are likely to work equally well in scaling-out project outputs, nor had most of the relationships yet been formed. Hence the network maps introduced the ideas that i) work had to be done to build relationships, ii) the relationships are likely to develop in unknown ways, producing both opportunities and threats to the project achieving eventual impact and (iii) these relationships should be monitored. None of this was in the original project description, nor in the IP logic model. Hence drawing the network maps helped improve the project’s causative theory by introducing ideas of relationship building and development, uncertainty, non linearity and opportunity. We integrate the IP logic model and network maps in the impact narratives by cross-referencing the network maps as much as possible with the outcomes and the scaling-out and scaling-up processes shown in the logic model. We then present the results of the extrapolation domain analysis and the scenario analysis to provide further quantification of likely impact. The finished output includes a four-page executive summary and the main text (see http://impactpathways.pbwiki.com for an example). The executive summary is designed to be 56 the basis for communication materials such as a press-release, web-page or glossy handout for donors. The main text contains within it sufficient description of the project’s impact pathways to be the basis of monitoring and evaluation to test and update the project. Figure 7: Scaling-out Network for the Strategic-Innovations-in-Dryland-Farming Project (i) Networks drawn based on information from the Impact Pathways workshop Now Future (ii) Networks redrawn after iterative question and answer process. (i) Now (ii) Future Understanding PIPA from an organizational learning perspective Research from the field of organizational learning helps understand how PIPA works. Argyris and Schön’s (1974) stated that people act on the basis of theories of action. Theories of action are the mental models that people use with regard to how to act in situations and which influence the ways they plan, implement and review their actions. Argyris and Schön’s (1974) distinguish between two types of theories of action – espoused theory and theory-in-use. A project or 57 program’s espoused theory is equivalent to its program theory written down in the form of a logic model or impact narrative. A project’s theories-in-use are found in the project staff and stakeholders’ usually tacit understandings of how change happens that affects how they implement the project. Argyris (1980) and later Patton (1997) state that developing congruence between the two can lead to greater effectiveness, thus suggesting that projects are more likely to achieve their development outcomes if there is closer agreement between program theory and practitioners’ theories-in-use. PIPA works to incorporate practioners’ theories-in-use into the project theory to achieve this congruence. It also works to include published theory where appropriate. Our initial results suggest that the network mapping in particular is a powerful tool in making explicit project staff’s implicit theories about how relationships need to develop to achieve scaling-out and scaling-up. This actor-orientated view of project’s impact pathways is usually missing in conventional logic models. = Impact Pathways (Douthwaite et al, 2003) Program Theory (Chen, 2005) Normative Theory (What is expected - projec t milestones, etc.) + Causative Theory = (Explanations of causation) Espoused Theory Theories of Action (Argyris and Schön, 1974) (Theories of action as explained to others) Theory-in-use Greater congruence increases project effectiveness (Personal theories of action, often implicit) Figure 8: Program Theory, Theories of Action and Impact Pathways PIPA and its contribution to R4D projects PIPA uses the outputs of a workshop to produce two descriptions of projects impact pathways: an IP logic model and actor network maps. The process of constructing and refining these two descriptions helps clarify and make explicit (i) assumed causal linkages between project outputs, outcomes and impacts and (ii) the relationships between organizations necessary for this to happen. Much of the clarification and surfacing of program theory come from refining the network maps, while writing the project’s impact narrative. The Impact Group, as evaluation specialists, give advice, question assumptions and suggest relevant theory to further improve the theory upon which a project has been conceived. 58 Once developed, the impact narrative helps a project better understand and communicate what it is doing, with whom it is doing it, and why. This makes the project more fundable because it presents a cogent, rational argument for support to funding agencies. It helps with project monitoring and evaluation because it permits managers to compare what they have predicted should be happening with what is actually happening. It also helps the project members develop a shared understanding of their project which can help with implementation, in part by identifying and giving focus to high priority activities and relationships. Moreover, constructing impact pathways for the projects in a basin helps project leaders, the basin coordinator and the CPWF Secretariat better identify complementarities and synergies between projects, thus contributing to the broader field of basin research program development. The workshops themselves have been found to foster better inter-project understanding and programmatic spirit. The added value of PIPA with respect to evaluation and impact assessment in the field of agricultural research-for-development is the explicit use of concepts from program theory (Chen, 2005) and organizational learning (Argyris and Schön, 1974) to clarify and describe projects’ impact pathways. These impact pathways are built of a number of hypotheses and assumptions about how research will lead to adoption, changes in peoples’ behaviour and developmental outcomes such as poverty reduction. The hypotheses and assumptions may be based on stakeholder-implicit theory or scientific theory. Hence, monitoring and evaluation of project and program impact pathways becomes a research activity with the potential to (i) test stakeholderimplicit theory and publish it as scientific theory and (ii) evaluate scientific theory in new contexts. This research process will yield new knowledge and insights into the processes by which research outputs do or do not achieve developmental impacts. This understanding is increasingly recognized as essential in the adaptive management of existing projects and conceptualizing of new interventions designed to improve living conditions of the rural poor. Such process understanding is also needed to give plausible ex-ante assessments of impact. A second contribution is that this is the first time that concepts from program theory have been integrated with extrapolation domain analysis and scenario analysis to produce a qualitative and quantitative ex-ante impact assessment approach that includes both quantitative and qualitative elements. A third contribution is the emphasis PIPA places on networks. One of the important long-term effects of projects is the networks they form, strengthen or undermine. Actor networks help projects identify linkages, and think about how they wish to alter and strengthen them so as to achieve their purpose and goal. Actor networks, kept up to date, can help projects monitor and evaluate their progress in this regard. Analyzing actor network maps can help projects prioritize their relationships and thus foster a strong network without incurring overly high transaction costs. The analysis can also clarify the essential future partnerships that need to exist after the end of the project. Network maps help projects achieve impact by showing the multiple linkages between partners and thus the multiple ways in which ideas and technologies can interact and be developed and diffused (see Figure 7). This helps people see that they are part of a network, and it is the network, not just their organization alone, that will achieve impact. It also helps people appreciate that the interactions between actors, indicated by the links in the map, make the innovation process inherently unpredictable in the medium and long-term, thus placing more 59 emphasis on the need for continual monitoring and evaluation to support adaptive project management. The novelty of PIPA to the field of evaluation is the use of network maps as a method to describe a project’s “reach”. PIPA follows Mayne’s (2004) counsel to make explicit the detailed expectations for each project. The activities involved, including the preparation of current and future network maps helps make explicit practitioners’ theories-in-use particularly about the relationships that will be required for their projects to accomplish the results they seek. PIPA supports the ex-post analysis of impact. By making explicit and then monitoring and evaluating progress along impact pathways, the project provides invaluable process documentation for impact evaluation after the project has finished. EIARD (2003) states that one of the requirements of good impact evaluation is that the impact pathways are described, hence if PIPA is carried out the evaluator’s job is to verify them. Finally, PIPA offers project managers and evaluators a practical set of tools that can provide (i) a better appreciation of the existing and potential impact of research to justify current and future funding, (ii) a deeper understanding of what impacts projects and programs might attain and how and (iii) the framework for an effective M&E approach that fosters and tracks progress towards achieving impact. Network maps help projects achieve impact by showing the multiple linkages between partners and thus the multiple ways in which ideas and technologies can interact and be developed and diffused (see Figure 7). This helps people see that they are part of a network, and it is the network, not just their organization alone, that will achieve impact. It also helps people appreciate that the interactions between actors, indicated by the links in the map, make the innovation process inherently unpredictable in the medium and long-term, thus placing more emphasis on the need for continual monitoring and evaluation to support adaptive project management. The novelty of PIPA to the field of evaluation is the use of network maps as a method to describe a project’s “reach”. PIPA follows Mayne’s (2004) counsel to make explicit the detailed expectations for each project. The activities involved, including the preparation of current and future network maps helps make explicit practitioners’ theories-in-use particularly about the relationships that will be required for their projects to accomplish the results they seek. PIPA supports the ex-post analysis of impact. By making explicit and then monitoring and evaluating progress along impact pathways, the project provides invaluable process documentation for impact evaluation after the project has finished. EIARD (2003) states that one of the requirements of good impact evaluation is that the impact pathways are described, hence if PIPA is carried out the evaluator’s job is to verify them. Finally, PIPA offers project managers and evaluators a practical set of tools that can provide (i) a better appreciation of the existing and potential impact of research to justify current and future funding, (ii) a deeper understanding of what impacts projects and programs might attain and how 60 and (iii) the framework for an effective M&E approach that fosters and tracks progress towards achieving impact. References Ambrose, K. 2007. Outcome Mapping in Ecuador: Enhancing Learning in the M&E Processes. Capacity.Org, Issue 30, March 2007 Anderson, J.; Bos, M.S. and Cohen, M.J. 2005 Impact Assessment of Food Policy Research: A stocktaking. Impact Assessment Discussion Paper 25. IFPRI Argyris, C. 1980. Inner contradictions of rigorous research, Academic Press, New York, USA Argyris, C. and Schön, D. 1974. Theory in practice: Increasing professional effectiveness, Jossey-Bass, San Francisco, USA. ASARECA, 1999. Workshop on Impact Assessment of Agricultural Research in Eastern and Central Africa. Entebbe, Uganda 16-19 November, 1999. Chen, H. T. 2005. Practical Program Evaluation: Assessing and Improving Planning, Implementation, and Effectiveness. Sage Publications, California, USA. Campbell, B., Sayer, J.A., Frost, P., Vermeulen, S., Ruiz-Perez, M., Cunningham, A. and Ravi, P. 2001 Assessing the performance of natural resource systems. Conservation Ecology 5(2), 22. [online] http://www.consecol.org/vol5/iss2/art22/index.html Saldanha, C.D. and Whittle, J.F. (1998) Using the Logical Framework for Sector Analysis and Project Design: A User's Guide. Asian Development Bank, Manila, the Philippines Checkland, P. and Scholes, J. 1990. Soft Systems Methodology in Action. John Wiley, Chichester, England Collinson M.P. and E. Tollens. 1994. The impact of the international research centers: measurement, quantification and interpretation. Issues in Agriculture: 6. CGIAR Secretariat, Washington, DC, USA. Cooksy, L. J., Gill, P., and Kelly, P. A. (2001). The program logic model as an integrative framework for a multimethod evaluation. Evaluation and Program Planning, 24(2), 119128. Cousins, J.P.; Goh, S.C.; Clark, S; & Lee; L.E. (2004) Integrating Evaluative Inquiry into the Organizational Culture: A Review and Synthgesis of the Knowledge Base. The Canadain Journal of Program Evaluation Vol. 18, No. 2 pp 99-141. Cross, R. and Parker, A. (2004) The Hidden Power of Social Networks. Harvard Business School Press, Boston, USA. 61 Dahler-Larsen, P. (2001) From Program Theory to Constructivism. Evaluation Vol. 7 No. 3 pp 331-349. Sage: London. Davies, R. 2003. Network Perspectives in the Evaluation of Development Interventions: More than a Metaphor. EDAIS Conference November 24-25, 2003 Douthwaite, B. 2002. Enabling Innovation: A practical guide to understanding and fostering technical change. ZED Books, London, England. Douthwaite, B., Ekboir J.M., Twomlow, S. and Keatinge, J.D.H. 2004. The Concept of Integrated Natural Resource Management (INRM) and Implications for Developing Evaluation Methods In: Shiferaw, B., Freeman, H. A. and Swinton, S. M. (eds.) Natural Resource Management in Agriculture: Methods for Assessing the Economic and Environmental Impacts of Management Practices. CAB International, Wallingford, England pp. 321-340 Douthwaite, B., Schulz, S., Olanrewaju, A., Ellis-Jones, J. In Press. Impact pathway evaluation of an integrated Striga hermonthica control project in Northern Nigeria. Agricultural Systems Douthwaite, B.; Kuby, T, van de Fliert, V. and Schulz, S. 2003. Impact Pathway Evaluation: an approach for achieving and attributing impact in complex systems. Agricultural Systems, 78 (2): 243 – 265 Earl, S., Carden, F. and Smutylo, T. 2001. Outcome Mapping: Building learning and reflection into development programs. IDRC, Canada, http://www.idrc.ca/booktique EIARD, 2003. Impact assessment in agricultural research for development. Agricultural Systems 78 pp 329-336 Guba, E.G. and Lincoln, Y.S. 1989. Fourth Generation Evaluation. Sage: Newbury Park, California, USA. Hamilton, W. V. (2005). Writing an Impact Report (Technology and Resource Spotlight Tip Sheet). cahe.nmsu.edu/tipsheets New Mexico State University: College of Agriculture and Home Economics. Hall, A; Mytelka, L.; and Oyeyinka, B (2004) Innovation systems: Implications for agricultural policy and practice. ILAC Brief 2. IPGRI, Rome, Italy Hartwitch, F., and Springer-Heinz, A. 2004. Enhancing the impact of agricultural research: An impact pathways perspective. ISNAR Briefing Paper, February 2004. Downloaded from http://www.isnar.cgiar.org/publications/pdf/bp-66.pdf on 7th September, 2006 Horton, D. 1998. Disciplinary roots and branches of evaluation: some lessons from agricultural research. Knowledge, Technology and Policy. 10(4) pp. 31-66 62 Horton, D.; Ballantyne, P.; Peterson, W.; Uribe, B.; Gapasin, D.; and Sheridan, K. 1993. Monitoring and Evaluating Agricultural Research: A Sourcebook. CAB International: Wallingford, UK. Horton, D. (Ed.) 2001. Learning About Capacity Development Through Evaluation. The Hague: International Service for National Agricultural Research. House, E.H. 2004. The Role of the Evaluator in a Political World. The Canadian Journal of Program Evaluation Vol. 19, No. 2 pp 1-16. Kellogg Foundation, 2004.Logic Model Development Guide. W.K. Kellogg Foundation, One East Micigan Avenue East, Battle Creek, Michigan 49017-4058, USA. Lincoln, Y.S. and Guba, E.G. 1985. Naturalistic Inquiry. Beverly Hills, California, USA: Sage Mayne, J. 2001. Addressing Attribution Through Contribution Analysis: Using Performance Measurements Sensibly. The Canadian Journal of Program Evaluation Vol. 16, No. 1, pp 124. Mayne, J. 2004. Reporting on outcomes: setting performance expectations and telling performance stories. The Canadian Journal of Program Evaluation Vol. 19 (1) pp. 31-60 Patton, M. Q. 1997. Utilization-Focused Evaluation. 3rd Edition. Sage Publications, Thousand Oaks, USA. McLaughlin, J.A. and Jordan, G.B. (1988). Logic Models: A Tool for Telling Your Program’s Performance Story. Evaluation and Program Planning, 22 pp 65-72 Montague, S. 1997. The Three Rs of Performance: Core concepts for planning, measurement, and management. Performance Management Network: Ottawa. OECD 2006. Promoting Pro-Poor Growth: Poverty Impact Assessment. Organization for Economic Co-operation and Development. 20 rue des Grands Augustins, 75006 Paris, France. Patton, M. Q. 1997. Utilization-Focused Evaluation. 3rd Edition. Sage Publications, Thousand Oaks, USA. Rogers, E.M. 2003. Diffusion of Innovations. 5th Edition. Free Press, New York, USA Renger, R. and Titcomb, A. 2002. A Three-Step Approach to Teaching Logic Models American Journal of Evaluation. 23: 493-503 Rush, B. and Ogborne, A (1991). “Program Logic Models: Expanding Their Role and Structure for Program Planning and Evaluation.” Canadian Journal of Program Evaluation, 6:2. 63 Ryder-Smith, D. 2002. Institutionalizing Impact Orientation. Natural Resources Institute, University of Greenwich, London, England Science Council, 2006. Guidelines for preparing 2007-2009 Medium Term Plans and 2007 Financial Plans. FAO, Rome. Taylor, C.L. and Fugate, Anne (1993). Writing the County Report of Accomplishment: Accomplishments/Impact Narrative. Circular PE-42, Agricultural Education and Communication Department, Florida Cooperative Extension Service, Institute of Food and Agricultural Sciences, University of Florida, Gainesville, Florida Whitney, D. and Trosten-Bloom, A. 2003. The Power of Appreciative Inquiry: A Practical Guide to Positive Change. Berrett-Koehler, San Francisco, USA 64 Selecting sites to prove the concept of integrated agricultural research for development Farrow, A.a, Opondo, C.b, Tenywa, M.c, Rao, K.P.C.d, Nkonya, E.e, Njeru, R. f, Lunze, L.g a Centri Internacional de Agricultura Tropical (CIAT), Kawanda, Uganda AHI/ICRAF c Makerere University,, Kampala, Uganda d ICRISAT e IFPRI f ISAR g INERA b Introduction The Sub-Saharan Africa Challenge Programme (SSACP) is a research effort that seeks to address the failings of the top-down dissemination from agricultural research through extension to smallholder producers, traditionally followed in sub-Saharan Africa. The SSACP seeks to implement and prove the effectiveness of an alternative approach namely Integrated Agricultural Research for Development (IAR4D) in three pilot learning sites (PLS) that represent three African contexts – East and Central Africa, West Africa and Southern Africa (FARA, 2008). In each PLS there are three teams (named taskforces) that will test the concepts of IAR4D. A major component of the IAR4D concept and that which differentiates it from conventional or other participatory approaches is the establishment and maintenance of innovation platforms (IPs). The research design for the proof of concept of IAR4D requires that each of the three task forces in each of the three PLS works on 4 independent innovation platforms (research design ref). Each IP is based in a particular territory, which for the purposes of the SSACP are named ‘sites’1. The pilot learning site for East and Central Africa is located at the borders of Rwanda, Uganda and the Democratic Republic of the Congo (DRC) and is named the Lake Kivu Pilot Learning Site (LKPLS). Due to the difficulties of conducting research in mountainous environments such as those found in the LKPLS it was agreed by all Task Forces that more then one innovation platform could be located in the same ‘site’, meaning that six action sites were sought. This report describes the methodology used to select sites for the LKPLS. The first section describes briefly the implications of the SSACP research design on site selection in LKPLS. This is followed by a characterisation of the LKPLS and the stratification of candidate sites. The third section is an account of site selection workshops which determined the levels of interventions by organisations promoting agricultural research and development and an appraisal of the critical issues in each candidate site. The report concludes with the final choices of sites in the LKPLS. 1 These sites are different from the Pilot Learning Sites 65 ‘Sites’ as part of the SSACP research design Across the three Pilot Learning Sites a consensus has emerged that each ‘site’ ought to be a local governmental unit. This offers the potential for dialogue with local policy makers and will help ensure that, while desired, positive spill-over effects are confined to the local governmental unit during the project implementation phase. The three task forces within LKPLS are working closely on the interactions between agricultural productivity, natural resource sustainability, markets and policy themes. The interactions between these themes imply that the three task forces work in common sites and potentially with common partners. At the same time the research design asserts that each of the three task forces in LKPLS will work with four innovation platforms giving a total of 12 IPs in each PLS. In order to reconcile a research design of 12 IPs with the need to collaborate2 it was decided that more than 1 Innovation platform would be formed in each site. Each IP is considered unique because the problem and entry-points are likely to be different for each task force even though some of the partners may be the same. At the same time it was decided that four IPs would be established in each country while each task force ought to establish and develop at least one IP in each of the three countries, and two IPs in one of the three countries; an example framework for task force site selection is shown in Table 1. Table 1. Example of organisation of common sites in the Lake Kivu PLS DRC Site 1 DRC Site 2 TF1 & TF2 TF2 & TF3 Rwanda Site 1 Rwanda Site 2 TF2 & TF1 TF1 & TF3 Uganda Site 1 Uganda Site 2 TF1 & TF2 & TF3 TF3 The research design of the SSACP requires that counter-factual sites are chosen for each action site where IPs will be established and developed. These counter-factual sites must be as similar as possible to the action sites with respect to the agro-ecology, farming system, market linkages, culture and demography. However the counter-factual sites must have experienced greater penetration and coverage by agricultural research for development organisations or projects. Given the limited number of districts (3rd level administrative units) within the LKPLS and hence the difficulty of finding a suitable counterfactual, the most appropriate size for a site is the 4th level administrative unit, which is a sub-county in Uganda, a secteur within Rwanda and a groupement within DRC. A plan for site selection was formulated during the period October-December 2007, between the launch meeting of the LKPLS in Kigali and a further meeting of the FARA coordinating team and the taskforce leaders in Kampala in December 2007. During this time the PLS was 2 and given the mountainous terrain 66 characterised, and the plan was refined and adjusted to respond to changes in the SSACP research design. The final plan for site selection consisted of 7 steps: 1) 2) 3) 4) 5) 6) 7) Census of the sub-counties, secteurs and groupements Definition of low and high market access Modelling of market access Identification of candidate sites Develop diagnostic tool for site selection Appraisal of candidate sites Final selection of sites Characterisation of the Lake Kivu Pilot Learning Site Much information regarding the characteristics of the LKPLS can be found in the original choice of Pilot Learning Sites (Thornton et al, 2006) and the report of the LKPLS validation team (Bekunda, et al, 2005). It was felt, however, that these ought to be revisited and the quantitative approach of the former combined with the qualitative assessment of the latter. In a partner workshop held in Kigali in October 2007 the members of the three task forces listed criteria that could affect productivity and environmental sustainability, and the success of agricultural enterprises (Table 2). The heterogeneity of the variables in Table 2 was also captured, but it was considered by the taskforce leaders that a mapping exercise was necessary to confirm the perceptions of the project partners. Table 2. Criteria determined by partners for site characterisation and an assessment of their variability within the Lake Kivu PLS (ref of Kigali meeting) Perceived Variability Variable Within PLS Within sites Partners, farmer organisations, large little networks Access to markets large moderate Rainfall moderate little Population density moderate little Infrastructure (roads, hospitals, moderate little schools) Production system moderate moderate Sources of incomes moderate moderate Terrain large large Soils large large Food security situation moderate large Settlement patterns ? Gender issues ? Conflict resolution ? Land tenure systems ? 67 The most important criteria to consider in the site selection phase are those variables that exhibit large variation within the LKPLS but which are relatively homogeneous within a sub-county, secteur or groupement. Variables which display large variability in the PLS but little at the site level should be controlled for in the choice of counterfactuals, while those variables that show little variability at the PLS but large variations within sites should be controlled for once sites have been selected. An analysis of the variance of different criteria shows that the standard deviation of values of annual precipitation3 for the whole PLS has a value of 256. This value is larger than the standard deviation of annual precipitation in every one of the 244 potential sites in the PLS (figure x). The difference between the standard deviation of the elevation4 values for the whole PLS as compared to the individual sites was not so large although the standard deviation is still large when compared to the individual sites (Figure 1). Terrain - measured here using slope - shows great variation within the PLS as well as within individual sites. 3 4 Annual precipitation from WorldClim (Hijmans et al, 2005) SRTM elevation and derived slope (Reuter et al, 2007) 68 Figure 1. Histograms of standard deviation of biophysical variables in the LKPLS Analysing the variation in population density depends on high quality spatial data. Unfortunately there are no spatial datasets that allow give population density at the fifth administrative level; comparing PLS variability with site variability of population density is therefore not possible. The only dataset that shows population density within sites – the LandScan suite of products – is based on a model, rather than observations (ORNL, 1998). An important research question that the task forces are trying to address is the degree to which the biophysical and socio-economic conditions at the site affect the engagement with markets and the enhancement of productivity and investment in NRM. Market access is a key hypothesis for many of the interventions being planned for the Lake Kivu PLS (FARA, 2008) as such it was decided5 that a key variable to be used in the stratification of sites would be the access to markets. The research design for the proof of concept of IAR4D (FARA, 2008) does not insist on 5 During a meeting of taskforce leaders, lead institute and FARA representatives in Kampala in December 2007 69 stratification of sites. By choosing sites in the three countries, however, we already introduce limited stratification according to the broad policy environments of each country. The PLS would be stratified to indicate sites accessible to a diverse set of markets (good market access), sites with access to a limited set of markets (poor market access), and sites with very poor access to all market types which would be excluded from the sample of potential sites. Sites would then be selected to ensure that of the 2 sites in each country one would have good market access and the other poor market access, with a counterfactual also selected for each site (Table 3). Table 3. Stratification of sites in the PLS according to market access DRC Good market access Poor market access Rwanda Uganda IAR4D Counterfactual IAR4D Counterfactual IAR4D Counterfactual Site 1 Site 3 Site 5 Site 7 Site 9 Site 11 Site 2 Site 4 Site 6 Site 8 Site 10 Site 12 Modelling access to markets in LKPLS A number of studies have developed or modified methods of determining access to markets (e.g. You and Chamberlin, 2004; Deichmann, 1997, Baltenweck & Staal, 2007; Farrow and Nelson, 2001). For this study we follow the methodology developed by ASARECA (2005) for a regional perspective of access to multiple markets. The spatial distribution of access to markets is based on models rather than observations but is augmented with expert opinion. The Model The modelling environment is a geographical information system (GIS) and the time is calculated using a costdistance algorithm and the model seeks the shortest path to all potential markets. Both raster (grid cells) and vector (points and lines) based modelling frameworks are possible and each offers advantages. Vector models are useful where movement is principally along paths and roads and where cross-country movement is disallowed. The vector framework is particularly appropriate in urban and developed country settings although it has been utilised (with certain modifications) in Africa (Deichmann, 1997; Baltenweck and Staal, 2007). For more general purposes and in developing countries where data on road quality and tracks are less reliable or up-to-date, a raster approach is often more suitable. In this case a ‘friction’ surface is created which describes the ease or difficulty of movement. For this application we have chosen to use a raster modelling framework6 in which the size of the grid cell is set at 100metres by 100metres. The model calculates for each market the time required from to arrive from all the cells in the grid and the path that would need to be taken. Cells are then allocated to their closest market. 6 For a description of this accessibility model see: Farrow and Nelson, 2001 70 The algorithm itself is conceptually easy to understand but the credibility of the model results depends on the construction of a friction surface that reflects the prevailing modes of transport and the barriers that constrain movement. Constructing a friction surface Common variables used in what we call the 'friction surface' include roads, land cover, barriers (such as customs posts at national borders, or rivers), and navigable rivers or boat routes (such as on Lake Kivu), and urban areas. Each of these variables has to be given an appropriate friction value depending on the modes of transport most appropriate for a particular context or problem. For the Lake Kivu PLS it is assumed that producers or traders have access to some form of motorised transport and the speeds for the roads (and thus the time required to traverse a grid cell) are set according to the quality of the road where that information is available. Boat services are an important means of transport across Lake Kivu For the background friction, i.e. those areas between the roads, we have used land cover data from the Africover dataset (FAO, 1994), this same dataset was also used to define urban areas. Barriers are limited to lakes and the national borders. There is also another factor that modifies the friction surface which is the slope of the surface. Slope increases the time needed to cross a cell irrespective of the fact that one is climbing or descending, this is less true of a bicycle than of a fully laden truck, but makes the computation easier. The values used for the friction surface can be seen in Table 4. Table 4.Time values used in creation of the friction surface Surface type Time to cross 100m cell Roads: Tarmac road 7 seconds (approx speed 50kmhr) Murram road 10 seconds (approx speed 35kmhr) Other road 14 seconds (approx speed 25kmhr) Tracks 24 seconds (approx speed 15kmhr) Landcover: Urban areas 10 seconds (approx speed 35kmhr) Herbaceous Cropland 150 seconds Tree-based Cropland 200 seconds Grassland 200 seconds Forests 400 seconds Barriers: Lakes 500 seconds National borders 2 hours Slope: 0-12° has no effect 12-30° increases friction by *2 slopes > 30° increases friction by *3 71 Choice of Markets For this study we follow the methodology adopted by ASARECA (2005) and distinguish between four types of markets: • Regional markets • Cross-border markets or transit points • National markets • Local markets All partner institutions were requested to identify markets for each of these classes; those that were located and used in the model are listed in Table 57 Table 5. Markets used in the accessibility modelling Regional markets: DRC Goma, Bukavu Rwanda Kigali Uganda Kampala National markets: DRC Goma, Bukavu, Butembo Rwanda Kigali, Ruhengeri, Byumba, Gitarama, Kibuye, Gisenyi Uganda Mbarara, Kampala Cross-border locations/markets: DRC-Uganda Bunagana, Ishasha (minor) DRC-Rwanda Goma/Gisenyi, Cyangugu/Bukavu, Kibuye RwandaGatuna/Katuna, Rugarama/Kyanika (minor) Uganda Local markets DRC: Rutshuru, Bunagana Rutshuru Beni, Kasindi Beni Butembo Butembo Sake, Kichanga, Masisi Masisi Kibumba, Goma Nyiragongo Minova, Nyabibwe, Kalehe Kalehe Local markets Rwanda: Mahoko, Gisenyi Rubavu Vunga, Kora, Gasiza Nyabihu Base Rulindo Gakenke Gakenke Kabaya Ngororero Byumba, Gatuna Gicumbi 7 Some market locations, mainly in DRC and in Kisoro (Uganda), could not be located on maps or were thought too distant from the PLS (See Annex 1) 72 Byangabo, Ruhengeri Musanze Kigali, Kibuye, Gitarama, Cyangugu, Ruhango Elsewhere Local markets Uganda: Kabale, Rubanda, Muko, Bufundi, Rubaya, Maziba, Kamwezi, Bukinda, Mparo Kabale Kisoro, Nyakabande, Cyanika, Bunagana Kisoro Kanungu, Kayonza, Burema, Rwanga, Ishasha, Kirima, Kambuga Kanungu Rukungiri, Nyarushanje, Kebisoni, Katobo, Kagunga, Ruhinda, Bugangari, Rukungiri Kikarara, Rwenshama Rwahi, Ngoma, Rubaare Ntungamo Kampala Elsewhere Local markets Burundi: Kirundo The accessibility model does not take into account the attractiveness of the markets and the basic algorithm is unable to discriminate between targets. Nevertheless an element of attractiveness can be introduced by considering different thresholds for the time needed to reach each of the market types (Figure 6). A location is considered to have good access to a regional market if it is within 3 hours, while the threshold for a national market would be 2 hours and a local market 1 hour. For cross-border markets the thresholds would be 1 ½ hours for a minor cross-border market, and 3 hours for a major cross-border market. Figure 6. Access time threshold for each market type The results of the model can be seen in Figure 7 and it is apparent that the density of the road network in Rwanda facilitates good market access – this is in clear contrast to DRC, where the roads are poor, and in Uganda where the major markets are distant. The quality of the spatial data is again an issue and Rwanda has excellent road data compared to the other two countries. Nevertheless the accessibility model offered information that would have been difficult and timeconsuming to collect otherwise and the project partners were comfortable with the results. 73 Figure 7. Time thresholds for market types Choosing sites based on market access Accessibility to different market types is combined (Figure 8) to indicate which areas are accessible to a diverse set of markets - these will be considered ‘good’ market areas, while those that can only access a limited set of markets (e.g. just local) would be our ‘poor’ market areas. Locations with universally poor areas are excluded from our sample 74 Figure 8. Diversity of market access and potential sites within the LKPLS The results of this process were shared with project partners at a meeting in Gisenyi in February 2008. Partners were invited to share their thoughts on the process as well as the results and were asked to make modifications to the sets of potential sites (Table 6) and decide on candidate sites which would be further characterised by a field visit and appraisal. In Uganda all sub-counties in the districts of Rukungiri and Kanungu were considered too remote, while sub-counties in Ntungamo and Bushenyi districts were considered to be in agro-ecosystems that were not representative of the LKPLS. As such only sub-counties in the districts of Kabale and Kisoro were included in the stratification. In Rwanda all areas were considered but the group decided to concentrate on the districts of Musanze, Nyabihu, and Rubavu which have similar agro-ecosystems and are located in the corridor between the towns of Ruhengeri and Gisenyi. However other sites along the RuhengeriKigali axis were also chosen for further characterisation. 75 In DRC areas at the northern tip of the LKPLS boundary, and west of Masisi were not considered due to the remoteness of these areas and the insecurity due to various armed groups operating there. Table 6. Good and poor market access administrative units and candidate sites (in blue) Country Good Market Access Poor market access Uganda Kisoro Kisoro • Chahi • Nyarusiza • Nyakabande • Busanza DRC Kabale • Bubare • Hamurwa • Muko Kabale • Buhara • Ikumba • Kaharo • Kitumba • Kyanamira • Rwamucucu • Bukinda • Kamuganguzi • Rubaya • Bufundi Kalehe • Mbinga-sud • Buzi Masisi • Muvunyi-Shanga Rutshuru • Bwenza • Jomba • Kisigari • Rugari Nyiragongo • Monigi • Kibati • Kibumba Nyiragongo • Buvira Rutshuru • Busanza Masisi • Muvunyi-Matanda • Kamuronja Kalehe • Mbinga-nord Rwanda Musanze • Cyuve • Muhoza • Remera • Shingiro • Kinigi • Nyange Gicumbi • Kaniga • Cyumba • Mukarange • Shangasha • Manyagiro • Byumba 76 • • Rwaza Gataraga Gakenke • Cyabingo • Kivuruga • Kageyo Rubavu • Bugeshi • Busasamana • Mudende • Cyanzarwe • Kanzenze • Rubavu • Nyakiliba • Rugerero • Gisenyi • Nyundo Rulindo • Kisaro Burera • Rwerere Nyabihu • Karago • Jenda • Bigogwe • Kabatwa Characterisation and selection of candidate sites The objective of the characterisation of the candidate sites was to be able to choose sites that would allow the investigation of the efficacy of the IAR4D principles and compare the results of IAR4D with conventional agricultural research for development approaches. To enable this investigation it was necessary to ensure that action sites have had as little as possible outside AR4D interventions as possible, while also finding counter-factual sites that have a similar context to the action sites but which have experienced more AR4D interventions. In each country four sites will be chosen, 2 action sites and 2 counterfactual sites. The action and counterfactual sites are stratified according to market access with 1 action site having good market access and another with poor market access, this is repeated for the counterfactual sites. Action sites will be chosen from the list of candidate sites according to the level of agricultural research for development between 2003 and 2008. All the villages in each site will be assessed and will be classified into 2 types: (a) clean villages that have neither had IAR4D nor conventional projects in the last 2-5 years; and (b) conventional approach villages that have had projects identifying, promoting and disseminating technologies in the past 2-5 years. Sites with 77 most clean villages will be chosen as action sites while sites with a mixture of clean and nonclean villages will be chosen as counterfactuals. To this end a tool was developed to ascertain the previous research and development activities in the previous 5 years in both the agricultural and other sectors, as well as to identify critical issues in the sites. There 5 major outputs of the characterisation of the candidate site were: 1. Census of villages in each sub-county, secteur or groupement taken 2. For each village the current agricultural research for development activities were determined 3. For each village the agricultural research for development activities in the past five years were determined 4. Inventory of potential stakeholders completed 5. Assessment of critical issues in the sub-county made Diagnostic tool The diagnosis of sites for final selection will rely on key informants from the candidate sites and at the next higher administrative area (district and territoire). A semi-structured questionnaire will be used (Annex 2) and the results compared using methods of triangulation. The questionnaire instrument was filled in during a group discussion. Informants 1. Sub-county or "Joint action forum" chairpersons. 2. Sub-county chiefs or executive secretary of the sector. 3. Development extension workers, cellule coordinator, line agriculturists and community development workers in the sub-county/secteur/groupement. 4. Sub-county/secteur/groupement based NGOs and farmer leaders. The tool was developed by project partners but did not undergo pre-testing before being tested in the field. The pilot testing was carried out in Uganda with all taskforce leaders present, most Ugandan partners and some key task force members from Rwanda. Modifications to the instrument were made in situ and were updated for use in Rwanda and DRC ensuring that the instrument was as efficient as possible and was used consistently in all three countries. The location and dates of the group discussions can be seen in boxes 1-3. 78 Box 1. Summary of site appraisal in Uganda The teams visited the Sub-Counties in Kabale and Kisoro districts between 17th and 19th March 2008 to test the tool for site selection. The itinerary was: 17th March 2008: Rubaya and Bubare sub-counties 18th March 2008: Hamurwa8, Muko and Bufundi sub-counties 19th March 2008: Busanza, Nyakabande, Chahi and Nyarusiza sub-counties 1. Select sites based on criteria A wrap-up meeting with all taskforce members was convened to discuss the selection of sites. Villages with few AR4D activities were found from almost all the sites that were visited. It was noted that the agro-ecosystems of the two districts (Kabale and Kisoro) are different and that it would be difficult to mix action and counterfactual sites between districts. Summaries were made of the sites and decisions were taken on the preferred research sites: Poor market access Bufundi: (Little intervention - no service providers, steep slopes, access to some water) – Chosen as action site Busanza: (Mixed - involvement of stakeholders, access to water) – Potential action site but no obvious counterfactual site Nyarusiza: (Little intervention- Few actors, little access to water) – Potential action site but no obvious counterfactual site Rubaya: (Mixed - many service providers, steep slopes, access to water) – Chosen as counterfactual site Good market access Chahi: (Little intervention -no service providers, volcanic soils, gentle slopes) – Chosen as action site Bubare: (Mixed- livestock activities in valley bottoms) Muko: (Mixed- service providers many; prices determined by traders; close to the main road) Nyakabande: (Mixed, volcanic soils, gentle slopes) – Chosen as counterfactual site 2. Data processing of info collected It was noted in the de-briefing meeting that the decisions on choosing sites should be more systematic and it was proposed to create indices based on the thematic sections of the survey instrument. This would allow the criteria in the 8 All candidate sites were visited but despite an appointment the meeting in Hamurwa sub-county did not take place (the reason given was that this day was a market day). The leadership in this particular candidate site did not seem too keen on participating in the planned meetings 79 tool to be quantified. From the team’s experiences in Uganda it was recommended that an additional criterion, be considered, i.e. leadership and commitment at the site level. 3. Lessons learnt and to be shared The purpose of sharing the lessons learnt was to increase the efficiency in Rwanda and DRC. One key lesson was that before the visit, the teams should procure a list for the villages and assign codes. This ought to also include some background information (e.g. development plans) which would support the information for the site selection process. 4. General reflection There was discussion on the consistency of the site selection and the critical issues encountered by the lake Kivu Validation team, i.e. poor markets; significant land degradation; polices that inadequately support development; and low adoption of technologies. It was felt by the site selection team that the criteria used to select the action and counterfactual sites were consistent with the validation team’s perceptions of the major issues in the PLS. Box 2. Summary of site appraisal in Rwanda The teams visited the secteurs in Musanze, Nyabihu, Rubavu, Gakenke and Burera districts between 2nd and 3rd April 2008. The itinerary was: 2nd April 2008: Nyange, Kivuruga, Gataraga, and Bigogwe secteurs 3rd April 2008: Rwerere and Mudende secteurs 1. Select sites based on criteria As with Uganda a debriefing meeting was convened at the conclusion of the group discussions in the candidate sites. Summaries were made of the sites and decisions were taken on the preferred research sites: Poor market access Mudende: (Little intervention - there are very few organizations dealing with agriculture, although several NGOs have an education, peace and reconciliation or HIV agenda, there are volcanic soils with high production potential, with gentle slopes) – Chosen as action site for market and productivity entry points Rwerere: (Little intervention - there are very few organizations dealing with agriculture, although several NGOs have an education, peace and reconciliation or HIV agenda, low potential soils, mainly Oxisols and Ultisols, intensively cropped for long time, with steep slopes) – Chosen as action site for NRM Bigogwe: (Some intervention- non volcanic Oxisols, except in a small portion of the sector. It is a new open land, where soils are still fertile, but fragile with high risk of rapid fertility decline, generally flat and gently sloped with a portion with steep slopes) – flat and gently sloped parts chosen as counterfactual for 80 Mudende, while the counterfactual for Rwerere will be the hilly part of Bigogwe Good market access Gataraga: (Little intervention – very few service providers, volcanic soils with high production potential, generally flat and gently sloped with only a small portion with steep slope) – Chosen as action site. Nyange: (Some development activities - low potential soils, mainly Oxisols and Ultisols, intensively cropped for long time) - Chosen as counterfactual for Gataraga sector and has at least 5 villages with little intervention Kivuruga: (Some development activities - non volcanic soils, steep slope) not selected because it presents completely different conditions from other sectors: different soil types and landscape 2. Lessons learnt and to be shared Secondary data collection on the 5 sectors selected for the study, especially quantitative data to support market, productivity and NRM critical issues and their magnitude should be collected. These could include long term weather data, population census and number of households, crop productivity and production, income generating activities, etc. It was recommended that DRC team prepare the list of villages per groupement (soft copy) and secondary data compiled prior to conduct site selection survey. Also the translation of the questionnaire to French was discussed. This was thought essential as all DRC partners are francophone. Even if questionnaire is in French, the final data collected will be translated back into English for entry. Box 3. Summary of site appraisal in DRC The teams visited the groupements in Kalehe (Sud-Kivu), Masisi, Nyiragongo, and Rutshuru territoires between 16th and 17th April 2008 and on 21st May. 16th April 2008: Buzi, Muvunyi-Shanga, Kamuronja and Muvunyi-Matanda groupements 17th April 2008: Kibati, Kibumba, Jomba and Busanza groupements 21st May 2008: Rigari and Kisigari groupements 1. Select sites based on criteria No decisions were taken on the preferred research sites after the meetings, instead the data were sent to the taskforce leaders to assess the different criteria and choose sites. During the planning one of the sites (Kisigari) was replaced with an alternative site (Kibati). 81 Good market access Muvunyi-Shanga: (Little intervention – very few service providers, low prices, soil erosion severe and frequent flood of the lowlands, the landscape is flat along Lake Kivu, that is where banana is produced, annual crops are produced in the hills where soil erosion is severe) – Chosen as action site. Buzi: (Many development activities – same conditions as Muvunyi-Shanga) Chosen as counterfactual for Muvunyi-Shanga groupement. Kibumba: (Many development activities, but not in agriculture - the main market for foodstuffs for Goma, such as vegetables, but also bean and potatoes. Although the soil fertility is good, the production is low because of poor service to the producers, lack of productive crop varieties. Vegetable production was once very competitive in that area but no longer today. NRM problems are severe soil erosion and fertility) - potential action site but no obvious counterfactual. Kibati: (Many development activities - productive soils are very limited because there are fresh lavas. The main activity is therefore wood production) Poor market access Busanza: (Little intervention - very distant from Goma and security is not guaranteed) Jomba: (Some intervention - Jomba is too distant from Goma, therefore difficulty of accessibility on muddy road) Kamuronja: (Little intervention- productivity is good; soils are fertile without any specific problems. Soil erosion is not a problem. The main crops are sweet potatoes, bean and cassava. Productivity constraints are lack of seeds and poor crop management, as a consequence of poor service to producers) – Chosen as action site but no practical counter-factual Muvunyi-Matanda: (Some intervention but in humanitarian activities productivity is medium. It is the only groupement with the high density of livestock, mainly dairy cattle) – would have been a good control site for Kamuronja particularly that part with more agricultural production. However, there has been some looting on the road in the past month. Security in DRC has been a major concern and as a result of the assessments of the sites it was decided to assess two more groupements with poor market access: Rugari and Kisigari (which was originally chosen as a candidate site). Kisigari: (Little intervention – hilly, land scarcity a problem) – Chosen as action site. Rugari: (Some intervention - hilly, land scarcity a problem) – Chosen as counterfactual for Kisigari. 82 Figure 9 shows the location of the action and counterfactual sites chosen as a result of the characterisation, as well as the candidate sites that were not chosen. Figure 9. Final choice of action and counter-factual sites in LKPLS The values of the indices for the sites can be seen in Table 7. The maximum possible score was 3 and the minimum possible score was 1. It can be seen that there are greater differences in the agricultural research for development activities between sites than between the three countries. It is also evident that the values in DRC suggest that the counterfactual sites had fewer interventions than the action sites. However, the team in DRC felt that there were enough villages with no AR4D interventions in each site to be able to test IAR4D. Similarly in the counterfactual sites villages that had experienced agricultural interventions in the preceding five years were identified. 83 Differences between poor market areas and good market areas were also not particularly large, but there is a tendency for fewer interventions and stakeholders in the poor market access areas of all three countries. Table 7. Index of agricultural research for development activities in selected sites Uganda AR4D Index Market Index Productivity Index NRM Index Rwanda AR4D Index Market Index Productivity Index NRM Index DRC AR4D Index Market Index Productivity Index NRM Index Good market Access Chahi (A) Nyakabande (C) 2.5 1 2.4 1.73 2.33 2 1.77 1.43 Good market Access Gataraga (A) Nyange (C) 2.5 2.25 2.67 2.33 2.15 1.93 2.13 1.77 Good market Access Muvunyi-Shanga (A) Buzi (C) Poor Market Access Bufundi (A) Rubaya (C) 2.5 1.5 2.73 1.91 2.8 2.08 2.42 2.07 Rwerere (A) 2.5 2.73 2.27 2.33 Poor Market Access Mudende (A) Bigogwe (C) 2.75 1.75 2.11 1.89 2.21 2.21 2 2.33 Poor Market Access Kisigari (A) Rugari (C) 1.5 2.33 1.86 1.75 1.89 1.71 2.6 2.5 2.31 2.2 2.5 2 2 1.67 1.81 1.69 Conclusions and discusion The site selection in the Lake Kivu Pilot Learning Site evolved with the framing of the overall research design for the Sub Saharan Africa Challenge Programme and was accomplished using a mixture of methods, tools and data. While the practicalities of field project implementation were considered they were never the principal reason for choosing sites. The rigour of the SSACP research design ensured consistency in the choice of sites between the three countries and offered an objective measure by which to assess the sites. Apart from the scientific rationale behind the research design there are practical advantages to this approach such as the transparent explanation of the choice of candidate sites to the local participants and policy-makers. Nevertheless the process of site selection allows for the articulation of local needs and the expression of critical issues within the candidate sites, which resulted in a more nuanced set of information on which to base the choice of action sites and ensure that counter-factual sites were as similar as possible. 84 References ASARECA, 2005. Fighting poverty, reducing hunger and enhancing resources through regional collective action in agricultural research for development. ASARECA (Association for Strengthening Agricultural Research in Eastern and Central Africa) Strategic Plan 20052015, August 2005, Entebbe, Uganda. 94 pp. Baltenweck, I., and Staal, S., 2007. Beyond One-Size-Fits-All: Differentiating Market Access Measures for Commodity Systems in the Kenyan Highlands. Journal of Agricultural Economics, Vol. 58, No. 3 Bekunda, M., Mudwanga, E.B., Lundall-Magnuson, E., Makinde, K., Okoth, P. Sanginga, P., Twinamasiko, E., and Woomer, P.L., 2005. Findings of the Lake Kivu Pilot Learning Site Validation Team: A Mission Undertaken to Identify Key Entry Points for Agricultural Research and Rural Enterprise Development in East and Central Africa. FARA, Accra, Ghana, 5th to 30th October 2005. Deichmann, U., 1997. Accessibility Indicators in GIS. United Nations Statistics Division, Department for Economic and Policy Analysis, New York. FAO, 1994. Africover – Eastern Africa Module: Land cover mapping based on satellite remote sensing. Available at: http://www.africover.org/documents.htm FARA, 2008. Sub Saharan Africa Challenge Program (SSA CP) Medium-Term Plan 2009-2010. Accra, Ghana, June 2008. Available at: http://cgmap.cgiar.org/documents/SUBFiles/SSA_2009-2011_MTP.DOC Farrow, A. and Nelson, A., 2001. Accessibility Modelling in ArcView 3: An extension for computing travel time and market catchment information. Software manual, CIAT, Cali, Colombia. Available at: www.ciat.cgiar.org/access/pdf/ciat_access.pdf Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. and Jarvis, A., 2005. Very High Resolution Interpolated Climate Surfaces for Global Land Areas. International Journal of Climatology, 25, 1965–1978. Oak Ridge National Laboratory, 1998. LandScan Global Population 1998 Database. Online documentation available at: http://www.ornl.gov/sci/landscan/landscanCommon/landscan_doc-main.html Reuter, H. I., Nelson, A. and Jarvis, A., 2007. An evaluation of void-filling interpolation methods for SRTM data. International Journal of Geographical Information Science, 21 (9), 983 1008 Thornton, P.K., Stroud, A., Hatibu, N., Legg, C., Ly, S., Twomlow, S., Molapong, K., Notenbaert, A., Kruska, R., and von Kaufmann, R., 2006. Site selection to test an integrated approach to agricultural research for development: combining expert 85 knowledge and participatory Geographic Information System methods. International Journal of Agricultural Sustainability, (2006) 4, 39–60 You, L., and Chamberlin, J., 2004. Spatial Analysis of Sustainable Livelihood Enterprises of Uganda Cotton Production. Eptd Discussion Paper121 . Washington, D.C., International Food Policy Research Institute. Available at: http://www.ifpri.org/divs/EPTD/DP/eptdp121.htm 86 Presión de la Sigatoka Negra y Distribución Espacial de Genotipos de Banano y Plátano: Resultados de 19 Años de Pruebas con Musáceas Ramírez, J.a,b, Jarvis, A.a,b y Van den Bergh, I.c a Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia. Bioversity International, Regional Office for the Americas, Cali, Colombia c Bioversity International, Parc Scienfique Agropolis II, Montpellier, France. b Resumen La Sigatoka negra (SN) es la enfermedad de la hoja de mayor importancia económica del cultivo de banano y plátano. En plantaciones para exportación, la enfermedad es controlada mediante el uso de funguicidas cuyo uso desmesurado y continuo no sólo es extremadamente dañino para el ambiente y la salud de los trabajadores, sino que también está fuera del alcance de los pequeños agricultores –pobres- quienes no pueden pagar los altos costos de producción. El desarrollo y utilización de cultivares resistentes a la enfermedad ofrece una alternativa mucho más sostenible. Basados en los resultados del IMTP (Programa Internacional de Pruebas de Musáceas), presentamos análisis de la interacción genotipo x medio ambiente x patógeno con el objetivo de entender los controles del ambiente sobre la Sigatoka negra, y la respuesta relativa de diferentes genotipos hacia la enfermedad. Se discuten las implicaciones de estos nuevos resultados, dando las bases de un modelo de soporte de decisiones para definir el potencial de diferentes genotipos en sitios específicos, y evaluar el riesgo del ataque de la Sigatoka negra. Palabras clave: banano, Sigatoka negra, IMTP, modelos, medio ambiente, interacción GxExD Abstract Black leaf streak is the most economically important leaf disease of banana and plantain. In export plantations, the disease is controlled by the heavy use of pesticides, which uncontrolled and continuous use not only is extremely harmful to the environment and the health of the workers, but also beyond the reach of poor farmers who cannot afford high production costs. The development and deployment of cultivars resistant to the disease offers a more sustainable alternative. Here we present genotype x environment x pathogen analyses based on IMTP results focused on understanding the environmental controls on Black leaf streak disease, and the relative response of different genotypes to the disease. The implications of these new insights are discussed, providing the basis for a decision support model for defining the potential of different genotypes in specific sites, and evaluating risk of Black leaf streak disease attacks. Keywords: bananas, black leaf streak, IMTP, models, environment, GxExD interaction Introducción La Sigatoka negra (Mycosphaerella fijiensis Morelet) es la enfermedad más destructiva que afecta los cultivos de banano y plátano en el mundo (Fullerton, 1994; Fullerton & Stover, 1990; Stover, 1984); descubierta en 1963 como “mancha negra de la hoja”, en Fiji, en el distrito 87 Sigatoka (Rhodes, 1964). El patógeno empezó su dispersión desde el sudeste asiático, llegando a África (Zambia) en 1973 ( Gauhl, 1994). Diez años después de su descubrimiento fue observada en Honduras (Stover & Dickson, 1976) y rápidamente se extendió por los países vecinos (Stover & Simmonds, 1987). La severidad del patógeno se magnifica en sistemas como el del banano y el plátano, en los cuales la propagación vegetativa (reproducción asexual) y la ocupación de grandes extensiones de tierras con un clon relativamente uniforme facilita ataques epidémicos de la enfermedad (Orozco-Santos, 1998). La aparición y el desarrollo de una enfermedad es el resultado de la interacción de tres factores principales: planta susceptible, agente patógeno y condiciones ambientalmente favorables (Valadares et al., 2007). El medio ambiente es un componente relevante, previniendo en muchos casos la presencia de una enfermedad incluso cuando tanto el patógeno como el hospedero están presentes. La actividad parasítica de M. fijiensis, por ejemplo, disminuye progresivamente con el incremento de la altitud (Mourichon y Fullerton, 1990; Mouliom Pefoura y Mourichon, 1990; Fouré y Lescot, 1988), y es afectada directamente por la temperatura (particularmente cuando está debajo de 20ºC) llegando a modificar significativamente el número de ciclos del patógeno por cada ciclo de producción del cultivo (Mourichon, 1995; Cordeiro et al., 2005; Valadares et al., 2007; Gauhl, 1994). La adaptabilidad de los genotipos depende tanto de las condiciones bióticas como abióticas de la región de estudio. Esta adaptabilidad puede ser medida en términos de la probabilidad de éxito del genotipo en cuestión o también en términos de la respuesta agronómica de dicho genotipo ante un medio ambiente determinado. En este documento, se usan datos del Programa Internacional de Pruebas de Musáceas (IMTP, por su nombre en inglés) para entender claramente la interacción entre el medio ambiente, la severidad de la Sigatoka negra y el rendimiento de los genotipos (interacción GxAxE). Materiales y Métodos El desarrollo presentado aquí combina dos diferentes ejercicios de modelación: el primero atiende a desarrollar un modelo (o modelos) de presión de enfermedad para mapear la presión de la Sigatoka negra a partir de parámetros climáticos y el segundo pretende analizar el rendimiento y la respuesta de diferentes genotipos ante la enfermedad y los diversos factores medioambientales. Variables independientes: clima Para los parámetros climáticos, usamos diecinueve variables bioclimáticas (Busby, 1991) con 30-arco segundos (~1km) de resolución espacial derivadas de WorldClim (Hijmans et al., 2005), y nueve variables de balance hídrico derivadas de observaciones del satélite de lluvias tropicales TRMM con 15-arco minutos de resolución más la elevación de cada sitio. Modelación de la presión de la Sigatoka negra La modelación de la presión de la enfermedad comprende el desarrollo de modelos matemáticos a través de regresiones multivariadas entre variables de respuesta a la enfermedad y los parámetros climáticos. La construcción del modelo usa dos pasos básicos: el proceso de desarrollo del modelo y el proceso de validación del modelo. Se seleccionaron los genotipos susceptibles Pisang Mas, Pisang Berlin, Niyarma Yik y SF215/NBA14 y se creó un set de datos para cada variable. 88 Pre-selección de variables Se están usando un número considerable de variables independientes para explicar las variables dependientes (HMJM y PDE), por tanto, para evitar problemas con nuestra modelación y de esta manera producir los mejores ajustes para los diferentes sets de datos se usó (1) una regresión de mejor modelo (PROC REG / selection=Rsquare; SAS, 2002) para encontrar el número ideal de variables como el número de variables produciendo el último cambio significativo en el valor de R2; (2) se modeló con todas las variables dependientes (PROC REG; SAS, 2002) para seleccionar las variables descriptoras completamente independientes entre ellas para desarrollar modelos; y (3) regresiones stepwise y backward (PROC REC / selection=stepwise, selection=backward; SAS, 2002) para seleccionar variables por significancia estadística Implementación de los modelos de regresión Se seleccionaron y espacializaron los mejores modelos, cuya cobertura espacial se ajustó a altitudes entre 0 y 1500 m.s.n.m y precipitaciones anuales entre 1000 y 3000 milímetros. Se transformaron a una escala de 1 a 10 usando 1 como el mínimo valor de severidad y 10 como el máximo. Generación del ensamble espacial de modelos Los modelos mapeados fueron validados en un total de 95 sitios usando la Raíz de la Diferencia Media Cuadrada (RMSD). Los modelos con RMSD menor a 2.7 se agruparon y el valor final de la presión de la enfermedad se calculó como el promedio de los modelos que pasaron el proceso de validación. Se calculó la confianza usando el coeficiente de variación (C.V.) de los modelos seleccionados. Modelación de la productividad genotípica Para el proceso de modelación de la productividad de genotipos se utilizaron tres diferentes pasos: (1) organización de los datos de entrada y agrupamiento de genotipos y (3) generación de modelos de productividad para cada uno de los diferentes grupos de genotipos. Datos agronómicos y de enfermedad y agrupamiento de genotipos Se usaron datos de enfermedad y rendimiento de 36 diferentes genotipos de banano y plátano, que incluyen especies silvestres, landraces, híbridos y somaclones. Se usó el peso del racimo como variable independiente. Se usaron dos diferentes clasificaciones de genotipos en este documento: (1) una clasificación de acuerdo al genoma y (2) una clasificación de acuerdo a la respuesta a la enfermedad (clasificación LP, con cinco grupos: altamente susceptible, susceptible, parcialmente resistente, resistente, altamente resistente) y se analizó la estabilidad estadística de la clasificación LP usando la función discriminante (PROC DISCRIM; SAS, 2002) Modelos de productividad por cluster Se desarrollaron regresiones lineales multivariadas (PROC REG; SAS, 2002) para cada uno de los grupos de genotipos para ambas clasificaciones usando las variables climáticas y de enfermedad como predictores y el peso del racimo como la única variable dependiente. La respuesta de cada uno de los modelos se transformó a una escala de 1 a 10 usando una regresión lineal simple. Finalmente se calculan los valores de presión de la enfermedad para cada una de las zonas de productividad de los diferentes clusters así como el tamaño mismo de la zona para 89 evaluar el impacto potencial del ataque de M. fijiensis en el desempeño de los diferentes genotipos de banano y plátano. Resultados y Discusión Variables independientes: clima Un amplio rango de parámetros medioambientales puede usarse para explicar las variaciones en el ataque de la enfermedad, cada variable es identificada con un prefijo y un número excepto la altitud (Alt), la lista de variables bioclimáticas de Busby (1991) y nueve variables derivadas del satélite TRMM: precipitación total anual TRMM 1, precipitación del mes más húmedo (TRMM 4), precipitación del mes más seco (TRMM 3), número de días con lluvia (TRMM 2), número promedio de días secos acumulados (TRMM 5), número de días de días de estación seca (TRMM 6), número máximo acumulado de días de estación seca (TRMM 7), relación Ea/Ep (TRMM 8) y el número de días de crecimiento (TRMM 9). Modelación de la presión de la Sigatoka negra Pre-selección de variables El método de selección de variables usando el coeficiente R2 mostró que para el PDE el número ideal de variables es 4 y 6 para el PDE usando dos modelos, mientras que para la HMJM el número ideal es 6 y 8 usando también dos modelos. La variable PDE se mostró como la de mejor respuesta, con mejores valores de R². Todos los modelos desarrollados con los procedimientos stepwise y backward mostraron tanto diferentes variables como diferentes valores del coeficiente de determinación R². Implementación de los modelos de regresión La extrapolación de la severidad de la enfermedad (figuras 1) cubre las principales áreas de producción bananera en Suramérica, Centro América y el Caribe, África, Asia y las zonas bananeras del norte de Australia. La mayoría de zonas en estos modelos tiene media a muy alta presión (zonas oscuras), lo que puede ser observado en el suroeste de Brasil, sureste de México, Belice, la zona Sahel, el norte de Madagascar, el norte de Zambia, el norte de Angola, Camerún y Costa de Marfil. Las zonas con baja presión se pueden observar en Zaire, el este de Perú y el norte de Venezuela. 90 (a) (c) (b) (d) Figura 1 Modelos de presión de SN: (a) PDE = -88.45071 + 0.04733 * altitud + 0.84255 * bio_6 0.01773 * trmm_1, (b) PDE = -96.57598 - 2.42454 * bio_2 + 3.27692 * bio_3 - 0.06041 * bio_4 + 3.05658 * bio_5 - 3.03858 * bio_11 - 0.02881 * trmm_1; (c) PDE = 104.26284 - 0.01875 * bio_4 + 0.01355 * bio_12 - 0.02725 * trmm_1; (d) PDE = -270.88287 + 0.13675 * altitude - 2.82893 * bio_3 0.0629 * bio_4 - 1.79853 * bio_5 + 4.4071 * bio_10 - 0.10015 * bio_13 El modelo 3 muestra alta presión al sur de Brasil y sus límites con Argentina. La mayoría de zonas en Centroamérica y el Caribe, Asia así como el norte de Australia muestran alta y muy alta presión de la enfermedad (zonas gris oscuro y negras); el modelo 4, por otro lado, muestra zonas de alta presión en el sudeste de Brasil y dentro de sus límites con Paraguay y Bolivia mientras que Centroamérica y el Caribe están de nuevo mostrando alta y muy alta presión de la enfermedad. Generación del ensamble espacial de modelos La validación individual de los modelos mostró que los mejores modelos fueron 1 y 2 con RMSD menores a 2.7. El ensamble total de modelos presentó una RMSD de 2.25. El modelo final (figura 2a) muestra patrones de severidad y dispersión de la enfermedad alrededor del mundo que indican que la más alta severidad de la enfermedad está definitivamente en Centroamérica, y especialmente en Belice, Costa Rica, Honduras y Panamá, en donde toda el área se considera de alta presión, el coeficiente de variación (figura 2b) está particularmente mostrando alta confianza, y en donde desde el descubrimiento de la SN es el lugar en donde se han encontrado los más fuertes ataques (Stover y Dickson, 1976). De acuerdo a los modelos, las zonas húmedas con temperaturas entre 20 y 25ºC en África y Latinoamérica presentan alta presión de la enfermedad tal como lo describen muchos autores; estas zonas también tienen baja evaporación a de esa manera un ambiente acorde para el desarrollo y dispersión de la SN. 91 (a) (b) Figura 2. (a) integración final de modelos; (b) mapa de confianza de la integración final de modelos Modelación de la adaptabilidad genotípica Datos agronómicos y de enfermedad y agrupamiento de genotipos La función discriminante mostró que la clasificación LP es estable excepto para los genotipos FHIA-23 pasando de susceptible a parcialmente resistente, PA 03-22 pasando de susceptible a altamente resistente, FHIA-18 pasando de parcialmente resistente a altamente resistente, Burro Cemsa de resistente a altamente resistente, FHIA-03 de resistente a parcialmente resistente, FHIA-21 de resistente a parcialmente resistente, Pisang Ceylan de resistente a altamente resistente y Pisang Mas de resistente a susceptible. Modelos de adaptabilidad por cluster No se lograron encontrar respuestas matemáticas para todos los clusters; sin embargo, las respuestas encontradas para la mayoría de los clusters fueron satisfactorias pese a no estar aún validadas. Los clusters genómicos AAB (R=0.70, 99.9%, 18 datos) (figura 3a), AAAA (R=0.732, 99.9%, 33 datos) (figura 3b) y AAAB (R=0.934, 99.9%, 35 datos) (figura 3c) presentaron respuestas interesantes en términos del ambiente y la enfermedad. Los cambios en productividad de un clima a otro se deben al efecto combinado de las respuestas de los diferentes genotipos dentro del cluster, la influencia del ataque de la enfermedad y el gradiente medioambiental presente entre las diferentes zonas agroclimáticas. (a) (c) (b) (d) (e) Figura 3 Modelos de productividad por cluster (a) AAB: BW = -1.27985 + 2.79491 * presión - 0.86148 * trmm_5; (b) AAAB: BW = -325.34530 + 0.61711 * presión + 4.09831 * bio_3 + 0.04432 * bio_4 - 0.17050 * bio_5 + 0.73405 * bio_8 - 0.85050 * bio_9 + 0.01592 * trmm_1 + 0.69859 * trmm_6 - 0.16475 * trmm_7; (c) BW = 325.34530 + 0.61711 * presión + 4.09831 * bio_3 + 0.04432 * bio_4 - 0.17050 * bio_5 + 0.73405 * bio_8 0.85050 * bio_9 + 0.01592 * trmm_1 + 0.69859 * trmm_6 - 0.16475 * trmm_7; (d) PR: BW = -10.07114 + 3.18476 * presión + 0.01250 * bio_14 - 0.14693 * bio_9 + 0.25249 * trmm_2 - 0.24029 * trmm_ 3; (e) HR: BW = 0.38835 + 1.53465 * presión + 0.14602 * bio_14 - 0.23812 * trmm_3 92 El mapa de productividad del cluster AAAA (figura 3a) muestra zonas de alta productividad en el sureste de África y el este de India; sin embargo, hay algunas áreas en Centroamérica (Belice, Costa Rica, Honduras, Panamá), Suramérica (Colombia, Venezuela y Brasil), y el Caribe en el que muchos genotipos muestran bajo desempeño debido a la alta presión; estos sitios requieren posiblemente una estrategia sostenible para controlar la enfermedad. Los genotipos incluidos en el cluster AAAB (figura 3b) muestran en general menos productividad que los genotipos en el cluster AAAA; toda la zona central de Brasil no es adaptable para estos genotipos, aunque por otro lado hay zonas en donde presentan muy alta productividad; las zonas en donde estos genotipos son productivos presentan de 2 a 5 meses consecutivos secos en el año, en dichos meses, el hongo M. fijiensis probablemente sufre un quiebre en su ciclo de desarrollo (Porras y Pérez, 1997). La clasificación LP, por su parte, produjo dos clusters con respuestas definidas, uno con 35 datos y otro con 40 datos, con una correlación de 0.81 (99.9%) (figura 3d) para el cluster parcialmente resistente (PR) y 0.61 (99.9%) (figura 3e) para el cluster altamente resistente (HR). El cluster PR muestra alta productividad considerable en el este de India, norte de Vietnam y el este de Myanmar, mientras las zonas de media productividad sólo son observables en el centro de Zambia, centro de Madagascar, este de Colombia, centro de Honduras, este de Cuba y el noreste de México. Hay zonas marginales y muy marginales (zonas en las que el rendimiento se ve significativamente afectado por la presión de la enfermedad y el clima) que cubren muchos de los países de Latinoamérica (Cuba, República Dominicana, sur de México, Belice, Honduras, Nicaragua, Costa Rica, Panamá, Colombia, Perú, Brasil y Venezuela). El clima más apropiado para el desarrollo del banano es un clima caliente y húmedo a través del año con vientos fuertes; los factores que determinan la distribución de la SN son precipitación en exceso (más de 100 mm por mes) (Simmonds, 1966) y un rango de temperaturas entre 10-40ºC con un óptimo entre 2530ºC y una media mínima de 15.5ºC. Aunque los genotipos incluidos en el cluster HR son altamente resistentes, no responden con altos rendimientos relativos (máximo 75%) puesto que su constitución genómica no permite tales respuestas, un análisis independiente sobre cada genotipo permitiría determinar cuáles genotipos en especial responden con mayores pesos de racimo que otros. Análisis complementarios Los tamaños de las zonas en donde los diferentes grupos de genotipos son adaptables varían de un cluster a otro, llegando a variar incluso el número de zonas cubiertas y los valores medios, máximos y mínimos de presión y presencia de la enfermedad (tabla 1). En general, se encontró que la presión de la enfermedad disminuye a medida que la productividad aumenta para los clusters AAAA, AAAB y PR y aumenta para los clusters HR y AAB; esto significa que para AAAA, AAAB y PR la productividad aumenta debido en mayor medida a una disminución en la presión de la enfermedad y en menor medida a la mejora en las condiciones medioambientales específicas de la zona; mientras que para AAB y HR el aumento en la productividad se debe a una combinación de condiciones medioambientales favorables para el crecimiento de estos genotipos en lugar de a una disminución significativa en la presión de la presión de SN. 93 Tabla 1 Distribución de la presión de la enfermedad (PE) en áreas de adaptabilidad por cluster Área Rango de potencial PE PE PE Zona Productividad Cluster de cobertura (media) (mínima) (máxima) Comparativa (km2·10^3) AAAA 1,768.4 6.48 4.4 9.4 AAAB 1,829.8 7.09 4.6 10.0 1 0-2 AAB 4,086.4 7.90 4.8 9.7 PR 4,725.2 7.37 4.1 10.0 HR 6,440.6 8.59 5.8 10.0 AAAA 11,961.1 7.39 4.4 10.0 AAAB 3,749.1 7.22 4.4 10.0 2 2-4 AAB 10,986.0 6.91 4.1 8.5 PR 12,396.2 7.50 4.0 10.0 HR 15,081.8 7.16 4.1 10.0 AAAA 6,766.4 8.01 4.9 10.0 AAAB 3,003.5 7.40 4.5 10.0 3 4-6 AAB 482.0 5.65 4.0 6.0 PR 3,241.0 7.87 4.4 10.0 HR 671.5 6.27 4.0 9.6 AAAA 1,543.2 8.23 5.2 10.0 AAAB 2,030.0 7.66 4.3 10.0 4 6-8 PR 621.3 8.29 4.9 10.0 HR 9.3 5.66 4.4 7.5 AAAA 414.6 8.18 6.2 10.0 5 8-10 AAAB 1,490.7 8.07 4.1 10.0 PR 252.8 7.81 6.2 10.0 Es importante notar que la mezcla de genotipos en los diferentes clusters genómicos puede llevar a tener respuestas tanto resistentes como susceptibles en un solo cluster y por tanto para un único valor de severidad de la enfermedad podría haber muchos diferentes valores de rendimiento; esto puede llevar a problemas en predicciones y posiblemente a falta de respuestas en algunos clusters. Respecto a la clasificación LP, debe notarse que en este caso el agrupamiento está produciendo una mezcla de genotipos con la misma respuesta pero con definitivamente diferentes características genéticas y por lo tanto diferentes rendimientos. Conclusiones 1. La variabilidad espacial de la SN puede ser explicada mediante una serie de variables climáticas, incluyendo la altitud, la precipitación anual, la precipitación del mes más seco y el más húmedo y las temperaturas mínimas y máximas. Este análisis muestra que una combinación de conocimiento experto, datos de campo y datos ambientales espaciales pueden usarse para desarrollar modelos matemáticos que expliquen la variabilidad en la presión de la enfermedad y dilucidar las posibles interacciones ambientales con los patógenos. 2. Hay algunas diferencias observables al nivel de clusters cuando se consideran las diferencias entre los genotipos incluidos en cada uno de ellos; el genotipo FHIA-25, por ejemplo, está incluido en el grupo genómico AAB, que de hecho no muestra zonas de alta ni muy alta productividad, pero en la clasificación LP está incluido en el grupo altamente resistente que tiene 94 un área total de alta productividad de 0.67 millones km2; hay ciertas diferencias que deben considerarse antes de tomar decisiones acerca de la posible liberación de los genotipos. 4. Dentro de las principales limitantes en la aplicación de estos modelos multivariados están el incremento de la dispersión del patógeno a través del tiempo y el efecto del cambio y la variabilidad climática no sólo sobre el rendimiento y la respuesta a la enfermedad sino también sobre la presión de la enfermedad y la dinámica del patógeno. Bajo condiciones de clima diferentes, estos modelos podrían prestarse para también evaluar posibles cambios temporales en presión de la enfermedad en series de tiempo anuales y por décadas. Investigaciones futuras deben combinar estos datos con otras variables tales como prácticas de manejo y datos de suelos así como también aplicaciones de pesticidas y fungicidas. Referencias Busby, J.R., 1991. BIOCLIM – a bioclimatic analysis and prediction system. Plant Protection Quarterly. 6, 8-9 Cordeiro, Z. J. M; Matos, A. P; Ferreira, D. M. V; Abreu, K. C. L. M. Manual para identificaçao e controle da Sigatoka-negra da bananeira. EMBRAPA Mandioca e Fruticultura Tropical: Cruz das Almas, 2005. 36p. Fouré, E; Lescot, T. 1988. Variabilité génétique des Mycosphaerella finéodés au genre Musa. Mise en évidence de la présence aun Cameroun sur bananier et plantain d’une cercosporiose (Mycosphaerella musicola) au comportement pathogéne atypique. Fruits 45: 407-415. Fullerton, R. A. 1994. Sigatoka leaf diseases. In Compendium of Tropical Fruit Diseases, eds. R. C. Ploetz, G. A. Zentmyer, W. T. Nishijima, K. G. Rohrbach and H. D. Ohr, pp. 12-14. APS Press, St. Paul, MN, USA. Fullerton, R. A. and Stover, R. H (eds.). 1990. Sigatoka leaf spot diseases of bananas. Proceeding of an International workshop held at San José, Cost Rica, 29.3-1.4.1989, INIBAP, Montpellier. Gauhl, F. Epidemiology and Ecology of Black Sigatoka (Mycosphaerella fijiensis Morelet) in Plantain and Banana (Musa spp.) in Costa Rica, Central America. 1994. PhD thesis originally presented in German. INIBAP, Montpellier, France. 120pp. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G., and Jarvis, A. 2005. Very high resolution interpolated climate surfaces for global land areas. International Journal Of Climatology, 25, 1965-1978. Mouliom Pefoura, A; Mourichon, X. 1990. Développement de Mycosphaerella musicola (maladie de Sigatoka) et M. fijiensis (maladie des raies noires) sur bananiers et plantains. Etude du cas particulier des productions d’altitude. Fruits 45: 17-24. 95 Mourichon, X; Fullerton, R. A. 1990. Geographical distribution of the two species Mycosphaerella musicola Leach (Cercospora musae) and M. fijiensis Morelet (Cercospora fijiensis), respectively agents of Sigatoka and Black leaf streak diseases in Bananas and plantains. Fruits 45: 213-218. Mourichon, X; Carlier, J; Fouré, E. 1997. Enfermedades de Sigatoka: Raya negra de la hoja (Sigatoka negra), Enfermedad de Sigatoka (Sigatoka Amarilla). Enfermedades de Musa: hoja divulgativa Nº 8. Mourichon, X. 1995. Les Cercosporioses des Bananiers: Éléments sur la Biologie des Interactions et les Stratégies de Lutte. Orozco-Santos, M. 1998. Manejo integrado de la Sigatoka negra del plátano. SAGAR, INIFAP, CIPAC. Campo Experimental Tecomán. Tecomán, Colima. Folleto técnico No. 1 División Agrícola 95pp Porras, A; Pérez, L. 1997. The role of temperature in the growth of the germ tubes of ascospores of Mycosphaerella spp., responsible for leaf spot diseases of banana. Infomusa – vol 6, Nº2. Rhodes, P. 1964. A new banana disease in Fiji. Commonwealth Phytopathological News 10:3841. Simmonds, N. W. 1966. Bananas. In: Sastry, P. S. N. 1988. Agrometeorology of the banana crop. Agricultural meteorology. World Meteorological Organization. CagM Report No. 29. Statistical Analysis System. SAS (r) 9.1. SAS Institute Inc., Cary, NC, USA. 2002-2003. SAS (r) 9.1 (TS1M3) Stover, R; Dickson, J. 1976. Banana leaf spot diseases caused by M. musicola and M. fijiensis var. difformis: a comparison of the first Central American epidemics. FAO Plant Protection Bulletin 24(2): 36-42. Stover, R. H; Simmonds, N. W. 1987. Bananas (3rd ed.). Longman Scientific & Technical, Harlow U.K. Stover, R. 1984. Las manchas producidas por las Sigatokas en hojas de bananos y plátanos. In: Curso internacional de reconocimiento, diagnóstico y control de Sigatoka negra del plátano y banano. 14-18 Mayo, Tulenapa, Colombia 15pp. Valadares, R; Cintra de Jesus, W; Avelino, R. Influencia das mudanças climáticas na distribuiçao espacial da Mycosphaerella fijiensis no mundo. In: Anais XIII Simpósio Brasileiro de Sensoramiento Remoto, Forianópolis, Brasil, 21-26 abril 2007, INPE, p 443-447. 96 Identifying candidate sites for crop biofortification in Latin America Zapata-Caldas, E.a, Hyman, G.b, Pachón. H.b, Monserrate, F.A.b, Vesga-Varela, L.c a Department of Geography, Universidad del Valle, Cali, Colombia Centro internacional de Agricultura Tropical (CIAT), Cali, Colombia c Universidad Industrial de Santander, Bucaramanga, Colombia b Abstract Agricultural science can address a population’s malnutrition through biofortification – plant breeding and biotechnology to develop crop varieties with high nutrient contents. These improved varieties should be grown in areas with populations at risk of nutrient deficiency and in areas where the same crop is already grown and consumed. Information on the population at risk of nutrient deficiency is rarely available for sub-national administrative units, such as provinces, districts, and municipalities. Nor is this type of information commonly analyzed with data on agricultural production. This project developed a method to identify populations at risk of nutrient deficiency in zones with high crop production, places where biofortification interventions could be targeted. Results Nutrient deficiency risk data were combined with crop production and socioeconomic data to assess the suitability of establishing an intervention. Our analysis developed maps of candidate sites for biofortification interventions for nine countries in Latin America and the Caribbean. Results for Colombia, Nicaragua, and Bolivia are presented in this paper. Interventions in northern Colombia appear promising for all crops, while sites for bean biofortification are widely scattered throughout the country. The most promising sites in Nicaragua are found in the centernorth region. Candidate sites for biofortification in Bolivia are found in the central part of the country, in the Andes Mountains. The availability and resolution of data limits the analysis. Some areas show opportunities for biofortification of several crops, taking advantage of their spatial coincidence. Results from this analysis should be confirmed by experts or through field visits. Conclusions This study demonstrates a method for identifying candidate sites for biofortification interventions. The method evaluates populations at risk of nutrient deficiencies for sub-national administrative regions, and provides a reasonable alternative to more costly, informationintensive approaches. Background Biofortification is the improvement of agronomic characteristics and the nutritional content of crops through plant breeding or modern biotechnology [1]. Taking advantage of the natural genetic diversity of crops, different varieties of a crop are crossed to develop new cultivars with 97 higher levels of desired nutrients. These new varieties can be disseminated to farmers in areas where nutrient-dense crops could address problems of nutrient deficiency and malnutrition. Several studies have shown that biofortification can improve nutritional status and that it is economically viable [2, 3, 4, 5]. Major international programs have been initiated to breed crops with higher levels of iron, zinc, Vitamin A, and amino acids [6,7] Biofortified crop varieties should be disseminated and used in places where nutrient deficiency is a problem and where the crops of interest are being produced and consumed in sufficient quantity to achieve impact. If these conditions are not met, then investments in biofortified crops will fail to reach the intended beneficiaries. A growing body of research has demonstrated the benefits of geographic targeting for poverty reduction and improving nutrition [8,9,10]. Thus, the targeting of interventions is an important problem that any nutritional initiative must address. Our analysis combines agricultural, nutritional, and socioeconomic information to assess candidate sites for crop biofortification in nine countries of Latin America and the Caribbean. Results for three of the countries are presented here. Candidate sites for biofortification interventions are found in areas where high prevalence of nutritional risks, high production and consumption of target staple crops, and high risk of poverty converge. This analysis is a preliminary step, before more detailed research on the candidate sites can determine their potential for impact. Methods The analysis first employed a procedure to prioritize indicators of nutrient deficiency risk. Next, weighted overlay was used to generate scores indicating the degree of confluence of factors important for biofortification. Data were collected to reflect the demand for nutrition interventions and the presence of the staple crops that are the current focus of biofortification research to improve nutrient content. The method assigned scores to the collected variables that are relevant to targeting biofortification interventions. The variables were weighted according to their importance to the result. The scores were then summed at the pixel level to provide the final result map. The following sections describe the data collected and the weighted overlay procedure. Data on risk of nutrient deficiency Assessing the demand for nutrition interventions over a large region calls for the development of information characterizing the magnitude and geographic distribution of nutrient deficiencies. A literature review of indicators of nutrient deficiency was carried out to determine the most appropriate indicators and how they could be used. Our assessment of the literature suggested a hierarchical organization of nutrient risk indicators based on how well they depict the problem. Indicators were grouped into three categories – biochemical measures, anthropometric measurements of children, and socioeconomic status (Figure 1). The measures were then classified according to the literature review into risk levels of nutrient deficiency. The class breaks and assignments of scores for the weighted overlay method are consistent with the scientific literature regarding risk levels of the three categories of nutritional indicators. The 98 indicators of nutritional risk were linked to administrative division maps and analyzed in a geographic information system. Use biochemical data to define risk of specific nutrients at department level Zinc: Vitamin A: Protein: Iron: seric zinc seric retinol none hemoglobin If biochemical data does not exist, use anthropometric data to define deficiency risk of any of the four nutrients at departmental level. Iron, Zinc, Vitamin A and/or protein: low height for age Use socioeconomic data to identify administrative districts at risk of deficiency of any of the four nutrients. Iron, Zinc, Vitamin A and/or protein: people with unmet Basic needs or those under extreme poverty line Figure 1 - Schema for selecting nutrient deficiency risk indicators The type of nutrient risk indicator used in the analysis was determined according to usefulness first, and then according to data availability. Biochemical measures of nutrient deficiency, such as hemoglobin levels in the blood, are direct measures of a person’s nutrient status, and as such are the preferred indicator. Unfortunately, health surveys often lack such data, especially for measures of Vitamin A and levels of amino acids. When biochemical indicators were unavailable, anthropometric indicators were the preferred next option. Since national health surveys often include anthropometric measurements of children less than 5 years of age, this indicator is often available [11]. Finally, if neither biochemical nor anthropometric data are available, poverty measures and maps can serve as indicators of risk of nutrient deficiency. Data on population and poverty Biofortification interventions are more likely to be successful where there are substantial rural populations living in poverty. Rural population data were developed from the Gridded Population of the World data set [12]. The 1-km global data set was resampled to 10-km resolution to conform to the framework of the analysis. Poverty index maps were derived from vector maps at the 2nd administrative level for Latin America based on the basic needs method [13]. These maps were converted to raster format. Data on crop production and consumption Biofortification interventions necessarily must be implemented where farmers grow the crop and consumers provide a local market. Several measures of the presence of the target crop for biofortification were collected and mapped. Biofortification is more likely to have a nutritional impact where there is a high level of production and consumption of the crop. Crop data sets for this analysis were derived from 10-km resolution crop production maps available for the world 99 [14, 15]. Food consumption data were acquired at department level from the Living Standards Measurement Study [16]. FAO production statistics and food balance sheet data provided contextual information for the analysis [17]. Weighted overlay analysis The first step in carrying out a weighted overlay analysis was to convert input data to the same spatial format and framework. A raster format was developed with 10-km spatial resolution to match the crop production data. All vector maps were converted to raster formats with corresponding 10-km pixel resolution. The literature review mentioned above had classified risk of nutrient deficiency into low, moderate, and high, and in some cases added an additional category of very high. Values of 3 (low), 6 (moderate), and 9 (high) were assigned when the classification comprised three categories. Values of 3 (low), 5 (moderate), 7 (high), and 9 (very high) were assigned when the classification included four categories. All other data were divided into terciles and assigned three values depending on whether they fell into the lowest (3), middle (6), or highest (9) tercile. The next step was to assign influence weights to each variable according to the importance of that variable to biofortification interventions. The risk of nutrient deficiency and the presence of crop production were considered to be the most important factors, and each was assigned an influence weight of 30%. Poverty intensity and rural population density were both assigned influence weights of 20%, since the weights must add up to 100%. The weighting scheme can be altered in the future, after dialogue with country experts on the preliminary results presented thus far. Figure 2 illustrates the method for the lower left pixel in a hypothetical map [18]. For each pixel, the assigned influence weights were multiplied by the corresponding variable value and then summed to derive the final score: Score = (a * .3) + (b * .3) + (c * .2) + (d * .2) Where a is the indicator of nutrient deficiency risk, b is the level of crop production, c is the poverty intensity, and d is the rural population density. The example in Figure 2 shows a high value of 7 for nutrient deficiency risk and a moderate value of 5 for crop production. With a value of 2, poverty intensity is low for the pixel. A rural population density value of 9 is high. Applying these values to the equation above yields a final score for the pixel of 6 (scores are rounded to the nearest integer). 100 Figure 2 - Use of the Weighted overlay method for lower left pixel in a hypothetical map Adapted from ESRI, 2006 [18] The map resulting from this weighted overlay procedure shows high, moderate, or low scores depending on the confluence of factors relevant to biofortification interventions. The highest scores indicate areas where the combinations of factors suggest a candidate site for implementing a biofortification program. The maps were further improved by eliminating isolated pixels surrounded by non-similar values through application of a spatial filter to the data. Finally, the highest two or three scores were chosen for the final map. Results Colombia Biofortification interventions in Colombia could potentially be implemented in any of the four physiographic regions – the coast, mountains, savanna (Llanos), and Amazon (Figure 3). 101 Population density is highest in the inter-Andean valleys of the mountain regions, areas such as the Bogotá plain (Cundinamarca) and the Cauca Valley. The savanna, Amazon, and coastal regions have far fewer people, but higher proportions of their population living in poverty (Figure 3c and d). Figure 3 - The geographic distribution of rural population, poverty intensity, and risks of nutrient deficiency in Colombia a) hemoglobin levels and b) stunting (height for age) in children less than 5 years of age, c) poverty intensity, and d) rural population density. 102 Sources: ICBF, 2005 [19] (iron deficiency); MACRO International, 2007 [11] (nutrient deficiency); Schnuschny and Gallopin, 2004 [13] (poverty intensity); CIESIN et al., 2004 [12] (rural population density) All departments in Colombia have either moderate or high risks of iron deficiency as indicated by hemoglobin levels surveyed in the Demographic and Health Survey [19] (Figure 3a). A group of departments in the north has high risks of iron deficiency. The map of stunted children shows a group of four departments with moderate levels of nutrient deficiency risk (Figure 3b). The federal district has high risk of nutrient deficiency as indicated by stunted children. Colombian crops that are the focus of biofortification efforts are found mainly in the hills and valleys of the mountain region (Figure 4). Nariño, Santander, and Antioquia are important regions for beans. Cassava production is most dense in the northern part of the country. Key areas of rice production include the Llanos (Meta department), the Amazon regions bordering the Andes Mountains, and many coastal regions in the northern part of the country. Maize has a fairly wide distribution throughout the country, with high production in Antioquia and Córdoba. Figure 4 - Crop production in Colombia a) bean, b) rice, c) maize, d) sweet potato, and e) cassava Source: You and Wood, 2006 [14] 103 High anemia levels in northern Colombia suggest this area as a best bet for candidate sites to implement crop biofortification aimed at reducing iron deficiency (Figure 5). In particular, the Córdoba department could be a focus for improved cassava, sweet potato, maize, and rice. High scores also were found in the southern parts of both Magdalena and Sucre departments. The result map indicates potential sites for bean biofortification in the northern part of the country and some smaller areas scattered throughout the country. Figure 5 - Candidate sites for iron biofortification in Colombia a) bean, b) rice, c) maize, d) sweet potato, and e) cassava, as indicated by hemoglobin levels Source: AgroSalud, 2007 [6] Candidate sites for biofortification with zinc, amino acids, and/or vitamin A are similar to those for iron (Figure 6). The Córdoba department in northern Colombia could be a focus of intervention for all crops. One exception to the focus on the northern part of the country is the pattern for bean biofortification, where pockets of bean production throughout the Andes coincide with moderate levels of stunting or high poverty intensity. 104 Figure 6 - Candidate sites for zinc, amino acids, and vitamin A biofortification in Colombia a) bean, b) rice, c) maize, d) sweet potato, and e) cassava, as indicated by height-for-age Source: AgroSalud, 2007 [6] Nicaragua Only general deficiency risk, based on anthropometry, could be evaluated for Nicaragua because of the lack of biochemical data on specific nutrients (Figure 7). High and very high risk levels are found in the northern departments. Moderate risks are found in the southeast part of the country, with low risks in the east. 105 Figure 7 - Nutritional risk of zinc, vitamin A, and amino acids deficiency in Nicaragua, as indicated by height-for-age Source: INEC, 2002 [30] Crop production is mostly focused in the western part of Nicaragua (Figure 8). Much of the humid east lacks large-scale production. Maize cultivation is concentrated in the departments along the Pacific Ocean. Bean production is most dense to the west of Lake Nicaragua and a group of departments in the center-north region. 106 Figure 8 - Crop production in Nicaragua a) bean, b) rice, c) maize, and d) cassava Source: You and Wood, 2006 [14] Consumption of beans, rice, and maize generally follows production patterns (Figure 9). The exceptions are Río San Juan and Atlántico Sur departments where per capita consumption is high. However, these departments have relatively low population density. 107 Figure 9 – Food crop consumption, poverty, and population in Nicaragua a) bean, b) rice, c) maize, and d) cassava consumption by department, e) poverty intensity, and f) rural population density Sources: World Bank, 2008 [16] (consumption); Schuschny and Gallopin, 2004 [13] (poverty intensity); and CIESIN et al., 2004 [12] (rural population density). Nicaraguans consume large quantities of maize and beans, moderate quantities of rice, and modest amounts of cassava or sweet potato. Nicaragua neither imports nor exports large volumes of maize, beans, rice, and cassava [18]. Thus, consumption of biofortified varieties of these crops – mostly grown within the country – would be likely to reach the intended beneficiaries. The center-north region stands out as a likely candidate for biofortification interventions (Figure 10). Matagalpa department shows candidate sites for bean, rice, maize, and cassava. Jinotega department shows candidate sites for rice, maize, and bean. Bean and cassava candidate sites are concentrated in relatively small areas in the center-north of the country. Maize and rice candidate sites are distributed widely, following production zones of these crops. 108 Figure 10 - Candidate sites for zinc, amino acids, and vitamin A biofortification in Nicaragua a) bean, b) rice, c) maize. and d) cassava as indicated by height-for-age Source: AgroSalud, 2007 [6] Bolivia Maize is the most important crop of those that are the target of biofortification initiatives in Bolivia (Figure 11). Rice and cassava production are important in Santa Cruz department. Bean 109 production is overwhelmingly concentrated in Santa Cruz, with much of it for export [20] . Santa Cruz department is Bolivia’s most important in the context of agricultural production. Figure 11 - Crop production in Bolivia a) bean, b) rice, c) maize, d) sweet potato, and e) cassava production Source: You and Wood, 2006 [14] Indicators of risk of nutrient deficiency are moderate, high, or very high throughout Bolivia (Figure 12). Both anemia and stunting indicators suggest the poorest conditions in the western, Andean part of the country. Poverty intensity is higher in the west as well. Crop production and risk of deficiencies do not neatly coincide. While Santa Cruz has comparatively lower risk factors for nutrient deficiencies, its high crop production could make it a focus of biofortification to address nutrient deficiencies, even though they are less severe compared to other countries. The Santa Cruz department could also be the source of biofortified foods for the rest of the country, to the extent that it serves as a breadbasket region. 110 Figure 12 - Variables considered in identifying candidate sites for biofortification in Bolivia a) iron deficiency risk and b) height-for-age in children less than 5 years old, c) poverty intensity, and d) rural population density Sources: MACRO International, 2007 [11] (nutritional deficiency); Schuschny and Gallopin, 2004 [13] (poverty intensity); and CIESIN et al., 2004 [12] (rural population density) 111 Four departments could be strong foci for biofortification in Bolivia – La Paz, Cochabamba, Chuquisaca, and Santa Cruz (Figures 13 and 14). The focus area could extend from the central part of La Paz department towards the southeast near the border with Paraguay. The result maps showed Santa Cruz to be of less interest, mostly due to the relatively lower levels of nutrient deficiency risk. However, Santa Cruz is the most important agricultural region of Bolivia, with good potential for the adoption of biofortified crops. Figure 13 - Candidate sites for iron biofortification in Bolivia a) bean, b) rice, c) maize, d) sweet potato, and e) cassava as indicated by hemoglobin levels Source: AgroSalud, 2007 [6] 112 Figure 14 - Candidate sites for zinc, vitamin A, and amino acids biofortification in Bolivia a) bean, b) rice, c) maize d) sweet potato, and e) cassava as indicated by height-for-age Source: AgroSalud, 2007 [6] Discussion This study revealed candidate sites for biofortification interventions in Latin America. Data availability, scale problems, and issues specifically related to biofortification need to be addressed to improve the capacity to identify the best sites for disseminating nutrient-dense crop varieties in the region. The following discussion addresses some of these issues. This research demonstrated that data limitations for geographic targeting of nutrition interventions can be overcome. However, our data collection effort has shown that simply better information could improve geographic targeting of interventions. Very few surveys provide biochemical data on nutrient status. In the three examples described above, two of the countries had hemoglobin data, and none of them had biochemical information indicating risk of zinc, amino acids, or vitamin A deficiency. Anthropometric measures of childhood nutrition are more widely available, but even these can be outdated, depending on the frequency of surveys carried out. 113 The varying resolution of input data for this geographic targeting exercise reduces its usefulness to some degree. Users of the analysis should be aware of these scale-related problems. The “ecological fallacy” especially limits the analysis when department level data is used [21, 22]. The level of nutritional deficiency risk reported for a department may be very high in some parts of the administrative unit and very low in others. For this reason, the results reported here should only be used after consultation with experts who know the situation in a country, or after on-theground verification of conditions. Geographic targeting based on identification of the most severe problem areas is sometimes inappropriate because nutrient deficiency risks may be uniformly severe throughout a country. For example, the level of stunting among children less than 5 years of age is very high in every department of Guatemala [23]. Several other countries only have two categories of deficiency risk. Where nutrition problems are severe everywhere, agricultural considerations such as potential for adoption and level of production should take precedence. In other cases, areas with severe nutritional problems could be served by other interventions aimed at reducing nutrient deficiency, such as supplementation or diet diversification programs. Again, Bolivia provides an example. The department with comparatively less deficiencies – Santa Cruz – may have the greatest potential for biofortification interventions. Even though this department is relatively less poor, moderate nutrient deficiencies are present An additional benefit of geographic targeting can be realized by looking for opportunities where more than one crop can be biofortified in a particular region. Programs promoting biofortified crops can realize marginal returns from setting up initiatives for several crops in the same region. These benefits can improve the efficiency of testing biofortified varieties and disseminating them. Ideally, the population of a given place would consume more than one biofortified food. For example, Córdoba department in Colombia could benefit from improved rice, beans, maize, cassava, and sweet potato for supplying diets with higher levels of iron, zinc, protein, and vitamin A. Expert opinion should be used to guide any targeting exercise, thus addressing the data and analysis limitations discussed above. We solicited comments on the results of the weighted overlays from our network of collaborators. There was general agreement about the location of candidate sites for biofortification. Comments tended to focus on contextual conditions for which the analysis could not account. For example, some regions produce crops for export or for animal feed. Others produce crops for urban markets where nutrient deficiency problems may be insubstantial. In other cases, cultural conditions may hinder implementation of biofortification interventions. For example, the people of a region may be accustomed to consuming whitefleshed sweet potatoes – not the high Vitamin A orange-fleshed varieties. Reviews and comments from experts are essential for targeting biofortification interventions. The data sets and methods described in this paper are oriented towards the current status of information available to conduct a multi-country assessment. Recently, new methods have been applied to create high-resolution nutrition deficiency maps [24]. One such method – called small area estimation – relies on both national censuses and representative household surveys. Using sophisticated statistical analysis, a nutrition risk variable in a household survey, such as height for age, is mapped onto the census geography to create maps at the 2nd administrative level. We 114 are aware of five implementations of this method for mapping malnutrition – in Panama, Dominican Republic, Ecuador, Cambodia, and Bangladesh [25, 26, 27, 28, 29]. Until this method is validated and more widely applied, the approach described in this paper provides a low cost alternative for assessing populations at risk of nutrient deficiency. Conclusions This study demonstrates a method for identifying candidate sites for biofortification interventions. The method uses available secondary data at the finest available spatial resolution. The study and accompanying data can be used for identifying populations at risk of nutrient deficiencies. It allows designers of large regional nutrition interventions to recognize localities that merit further consideration for inclusion in programs to reduce nutrient deficiency. The research combines agricultural production and health information to support decision-making and program implementation, addressing the need to efficiently target interventions to the populations that need them most. Competing interests EZ, GH, HP, FM, and LV have no competing interests in this study. Authors' contributions EZ developed and pre-processed much of the data, designed and carried out the computer modeling, and wrote the initial manuscript in Spanish. GH led development of the database and contributed to the study design and methodology. HP led the development of nutrition data and their classification as indicators of nutrient deficiency risk. FM developed much of the database and linked statistical information to maps. LV conducted a literature search and developed a table of indicators of the risk of nutrient deficiency. All authors participated in the research design and interpretation of results. They all read and approved the final manuscript. Acknowledgements This research was supported by grants from the Canadian International Development Agency (CIDA) and the Bill and Melinda Gates Foundation. Elizabeth Barona and Claudia Perea at the International Center for Tropical Agriculture (CIAT) assisted in the development of data and metadata used in this project. The research was carried out under the AgroSalud and HarvestPlus biofortification initiatives. References AgroSalud [http://www.agrosalud.org]. Accessed on 09 February 2009. AgroSalud: Maps Gallery [http://www.agrosalud.org/descargas/Galeria_Final/Guatemala/indicadores_nutricionales/a ntropometirco_riesgo_deficiencia_nutricional.gif]. Accessed on 9 February 2009. 115 Baker JL, and Grosh, ME: Poverty reduction through geographic targeting: how well does it work? World Dev 1994, 22(7):983-995. Bangladesh Bureau of Statistics: Local Estimation of Poverty and Malnutrition in Bangladesh. In collaboration with UN World Food Programme; 2004:43 Bigman D, and Fofack H (Eds): Geographical Targeting for Poverty Alleviation: Methodology and Applications. Washington, D.C: World Bank Regional and Sectoral Studies; 2000. Bouis HE: Economics of enhanced micronutrient density in food staples. Field Crops Res 1999, 60(1-2):165-173. CIESIN, IFPRI, and CIAT (Center for International Earth Science Information Network, International Food Policy Research Institute, and Centro Internacional de Agricultura Tropical): Global Rural-Urban Mapping Project (GRUMP). Palisades, NY: CIESIN, Columbia University; 2004. ESRI: ArcGis 9.2® Desktop 2006 [http://webhelp.esri.com/arcgisdesktop/9.2/index.cfm?id=5188&pid=5186&topicname=We ighted_Overlay] FAO: FAOSTATS [http://www.fao.org/index_ES.htm]. Accessed on 09 February 2009. Freedman D: Ecological inference and the ecological fallacy. Prepared for the International Encyclopaedia of the Social and Behavioral Sciences. Technical Report No. 549. 1999:7. Haas JD, Beard JL, Murray-Kolb LE, Del Mundo AM, Felix A, and Gregorio GB: Ironbiofortified rice improves the iron stores of nonanemic Filipino women. J Nutr 2005, 135:2823-2830. ICBF (Instituto Colombiano de Bienestar Familiar): Encuesta Nacional de la Situación Nutricional en Colombia (ENSIN). Bogotá, ICBF; 2005. INEC (Instituto Nacional de Estadisticas y Censos): Encuesta Nicaraguense de Demografia y Salud (ENDESA) 2001. Informe Basico. Managua: INEC; 2002. MACRO International: Demographic and Health Surveys [http://www.measuredhs.com/]. Accessed on 09 February 2009. Morris SS, Flores R, and Zuñiga M: Geographic targeting of nutrition programs can substantially affect the severity of stunting in Honduras. J Nutr 2000, 34:2514-2519. Nestel P, Bouis H, Meenakshi JV, and Pfeiffer W: Biofortification of staple food crops. J Nutr 2006, 136:1064-1067. 116 Pfeiffer WH, and McClafferty, B: HarvestPlus: breeding crops for better nutrition. Crop Sci 2007, 47:88-105. Robinson WS: Ecological correlations and the behavior of individuals. Am Sociol Rev 1950, 15(3):351-357. Rogers BL, Macias KE, and Wilde P: Atlas of hunger and malnutrition in the Dominican Republic. In Working Paper in Food Policy and Applied Nutrition. Medford MA: Friedman School of Nutrition Science and Policy, Tufts University; 2007, 36:32. Rogers BL, Wirth J, Macias K, and Wilde P: Mapping hunger in Ecuador: a report on mapping malnutrition prevalence. In Working Paper in Food Policy and Applied Nutrition. Medford MA: Friedman School of Nutrition Science and Policy, Tufts University; 2007, 37:37. Rogers BL, Wirth J, Macias KE, Wilde P, and Friedman DR: Mapping hunger: a report on mapping malnutrition prevalence in the Dominican Republic, Ecuador, and Panama. In Working Paper in Food Policy and Applied Nutrition. Medford MA: Friedman School of Nutrition Science and Policy, Tufts University; 2007, 34:32. Rogers BL, Wirth, J, Macias KE., and Wilde P: Mapping hunger in Panama: a report on mapping malnutrition prevalence. In Working Paper in Food Policy and Applied Nutrition. Medford MA: Friedman School of Nutrition Science and Policy, Tufts University; 2007, 35:27. Schuschny A, and Gallopin G: La distribución espacial de la pobreza en relación a los sistemas ambientales en América Latina. In Medio Ambiente y Desarrollo. Chile: Comisión Económica para América Latina y el Caribe (CEPAL); 2004 (87):42. Stein AJ, Nestel P, Meenakshi JV, Qaim M, Sachdev HPS, and Bhutta ZA: Plant breeding to control zinc deficiency in India: how cost-effective is biofortification? Public Health Nutr 2007, 10(5):34. Tomoki F: Micro-level estimation of prevalence of stunting and underweight in Cambodia, draft report for World Food Programme and ORC Macro. 2002. Van Jaarsveld PJ, Faber M, Tanumihardjo SA, Nestel P, Lombard CJ, and Spinnler Benade AJ: β-carotene-rich orange-fleshed sweetpotato improves the vitamin A status of primary school children assessed with modified -relativedose- response test. Am J Clin Nutr 2005, 81:1080-1087. Voysest VO: Bean believers: former Bolivian miners turn a staple grain legume into gold. Growing Affinities. Colombia: CIAT (Centro Internacional de Agricultura Tropical): 1999:1, 6, 7. World Bank: Living Standards Measurement Study (LSMS). 2008. [http:econ.worldbank.org/] Accessed on 09 February 2009. 117 You L, and Wood S: An entropy approach to spatial disaggregation of agricultural production. ScienceDirect 2006, 90:329-347. You L, Wood S, and Wood-Sichra U: Generating plausible crop distribution maps for SubSaharan Africa using a spatially disaggregated data fusion and optimization approach. ScienceDirect 2009, 99:126-140. 118 9.2 Sustainable and Equitable use of Ecosystem Services Challenges to Managing Ecosystems Sustainably for Poverty Alleviation: Securing WellBeing in the Andes/Amazon. Situation Analysis prepared for the ESPA Program. Amazon Initiative Consortium, Belém, Brazil Porro, R.a,b, Borner, J.a,b, Jarvis, A.a, Fujisaka, S.c a Centro Internacional de Agricultura Tropical Amazon Initiative, Embrapa, Tr. E. Pinheiro s/n, Belém-PA, Brazil c Consultant. Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia. b Executive Summary This report focuses on the Amazo The Ecosystem Services and Poverty Alleviation Program (ESPA) was initiated in 2007 by the Natural Environment Research Council (NERC), the Department for International Development (DfID), and the Economic and Social Research Council (ESRC) of the UK. ESPA is a global program in its initial stages that will promote research and capacity-building to achieve sustainable ecosystem management and increased well-being in developing countries. n basin and the eastern Andean slopes (herein referred to as the Andes/Amazon ecosystem or region). The Amazon is the largest fresh water system and tropical forest in the world. Large portions of the region are still covered by relatively intact primary forests that provide substantial locally and globally valuable ecosystem services (ES). Rural population densities in the region are among the lowest in the world. As such, the Andes/Amazon is a contrast to other ESPA target areas that are characterized by scarce and degraded resources used by often overwhelming numbers of the poor. Hence, in the Andes/Amazon, ESPA should focus on promoting resource conservation before valuable ES are irreversibly lost due to actions by resource users ranging from poor slash-and-burn farmers to large timber and commodity farming interests. A rationale for this approach is that rebuilding ecosystem services in ecologically degraded areas is generally much more costly than preventing their loss in the first place. As an agricultural colonization frontier, the Amazon has lost some 84 million ha of native forests over the last few decades – a loss accompanied by losses of locally and globally valuable ES. A “situation analysis” of ES and poverty in the Andes/Amazon was conducted September 2007 March 2008. Findings are intended to help guide ESPA in terms of research and capacitybuilding priorities. A macro-scale approach was taken to examine ES, well-being, and management needs. The work was accompanied by an extensive consultation with local, national and regional stakeholders. The introductory chapter sets out the objectives of the situation analysis, and the approach of the study. It also briefly discusses the relationships among ES and poverty in the context of this situation analyses. The discussion settles on key findings of another recent study that has reviewed the literature on this relationship on a global scale. The situation analysis adopts existing definitions of ES, which are understood to be the “processes and conditions through 119 which ecosystems support human life” or, more generally, the “benefits that people obtain from ecosystems”. No single poverty definition is adopted throughout the report. Depending on data availability and analytical approaches it employs different poverty concepts and explores implications if necessary. Stakeholder consultations reinforced the need to adjust standard poverty measures to better capture the ES dimensions of wellbeing in the Andes/Amazon. Moreover, the concept of poverty itself was challenged in favour of a wellbeing oriented approach. The report focuses on key issues: Paramount ES provided by the Andes/Amazon ecosystem to local populations and to the global society, and the main threats and challenges to the provision of these services are identified (Chapter 2). The benefits that local populations derive from using ES are characterized (Chapters 2 and 5). Promising options to manage ES provision in ways that also prevent or help to alleviate poverty are identified and characterized (Chapters 3 and 4). Key results of stakeholder consultations and related priorities for research and capacity building are summarized in Chapter 5. Chapter 6 summarizes the key messages of all chapters and proposes three core areas to be addresses by research and capacity-building in the ESPA program. Prototype research projects and promising impact pathways are proposed. By chapter, Chapter 2 provides a spatial assessment of ES and poverty in the Andes/Amazon. The literature review and the stakeholder consultation allowed for the identification of the most important ES. However, not all ES could be quantified and assessed spatially due to data limitations. Attempts to quantify services included direct measures or measures of the natural resource base for any particular service provision. Services examined were water quantity and quality, local climate regulation, carbon as an indicator for global climate regulation services, soil related services, and a set of services associated with terrestrial and aquatic biodiversity. The spatial assessment confirms that rural inhabitants are most vulnerable to changes in ES provision. Particularly traditional and indigenous populations have developed strong dependencies on locally abundant ES and goods. Hence, relative resource abundance does not mean low vulnerability. Especially, ES that are subject to natural variability and human pressures (e.g. water flow and quality, local climate, forest products) introduce an important source of uncertainty even into relatively well adapted livelihood strategies. A key contribution of Chapter 2 is to illustrate some of the spatial and long-term temporal dimensions of ES provision, which may help to better target future ESPA program activities. Chapter 3 reviews the diverse options available to manage ES and their potential effects on the poor. Management options (MO) are classed as enabling (e.g., technologies, property rights, environmental education, public-private partnerships, credit, and insurance), incentives (e.g., payments for environmental services, subsidies, inputs, and certification or eco-labeling), and disincentives (e.g., taxes, regulations, fines, and imprisonment). It becomes clear that the MO of choice in the past have been disincentive-based. In large and sparsely populated areas, where few actors can have large impacts, the need to constantly enforce disincentive MO may make them less cost-effective than incentive-based MO. Research is needed to support the current trend in favour of such MO to determine where and under what conditions they represent true alternatives. Options to manage ES should not be understood as substitutes for social policies and basic public services. The lack of the latter is often the root cause of poverty in the 120 Andes/Amazon. What’s needed is a better understanding of how to combine enabling and incentive MO for ES management in order to allow for the poor to capture benefits. Chapter 4 reviews factors underlying successful programmes and projects that have implemented management options in the Andes/Amazon. Lessons learned are discussed. Reviewed projects dealt with conservation and recuperation of ES and ecosystems; impacts on well-being; and innovative approaches. Project impacts are discussed in terms of economic benefits, reversal of environmental degradation or ES conservation, local added value, redistribution of benefits, empowerment of communities, and potential of resources transfer from wealthier to poorer sectors. Again, incentive-based MO, such as certification and incentives from ecotourism, seem to have more potential to benefit the poor. Pilot experiences need to be replicated and scaled out. Chapter 5 summarizes the main outcomes of the stakeholder consultation and discusses environmental policy approaches in the Andes/Amazon. Recommendations include: better definition, assessment, and valuation of ES; assessment of contributions of ES to wellbeing; development of management options that contribute to wellbeing; development and support of pilot studies; and improving capacities of institutions dealing with ES and poverty alleviation. Chapter 6 recommends three core areas to be included in the ESPA agenda for the Andes/Amazon. The first area involves primarily biophysical, the second interdisciplinary, and the third primarily socio-economic and policy research: 1. Understanding and predicting spatial and temporal dynamics of key locally and globally valued ES (especially, forest products, local climate regulation, water quality/quantity and fish resources) with a special focus on a. Moving away from the traditional spatial scales of study (individual sites) to policy relevant regional scales such as that addressed in this situation analysis. Also taking into account the important implications of geographic and environmental differences throughout the region on the development of locally adaptive and effective regional policy. Recognizing the impact of trans-frontier and trans-continental linkages especially for climate and water b. Identifying critical thresholds of change in the provision of ES under human impacts, such as deforestation, and climate change and devising monitoring, prevention, adaptation, and mitigation measures to ensure that significant thresholds that would lead to increased poverty are not crossed through ecosystem mismanagement c. Developing and disseminating practical methods to monitor and document local changes in ES provision and spatial-temporal policy support systems to indicate the driving forces of such changes and test in silico preventative policy measures 2. Understanding, measuring and valuing the contribution of each of locally valued ES to generate wellbeing among heterogeneous local stakeholder groups with a special focus on a. Developing comparative frameworks and integrate ES-related welfare into indexbased poverty measures b. Identify location and stakeholder specific vulnerability indicators 121 3. c. Developing and disseminating methods and tools to forecast natural and policyinduced changes in ES provision and their likely impacts for local wellbeing d. Establish and institutionalize regional knowledge management practices to feed research results from this into the previous and next area for prioritization of actions. Promote incipient initiatives to implement incentive based approaches (e.g. certification, payments for environmental services, ecotourism) to ES management and related comparative research to extract lessons learned with a special focus on: a. Globally and locally valued ES which are affected by externalities of local income generating activities b. How, where and for whom incentive-based management options need to be combined with enabling management options in order to maximize benefits for the poor c. Developing and disseminating decision-frame works and related tools for policy makers to decide where and under what conditions incentive-based management options will work and what can be done if minimum conditions are not in place Chapter 6 ends with a series of prototype projects to address key research questions in each of these areas, suggests promising impact pathways and capacity-building components. The full report is available from: http://www.ecosystemsandpoverty.org/index.php/2008/andesamazon-ecosystems-services-andpoverty-alleviation/ 122 Paying for avoided deforestation in the Brazilian Amazon: From cost assessment to scheme design Börnera, J., Wunder, S. b a Amazon Initiative & International Centre for Tropical Agriculture, CIAT, c/o Embrapa, Tr. E. Pinheiro s/n, Belém-PA, Brazil b Center for International Forestry Research, CIFOR, c/o Embrapa, Tr. E. Pinheiro s/n, BelémPA, Brazil Abstract Reducing emissions from deforestation and degradation (REDD) is considered a significant climate change mitigation opportunity. The Brazilian Amazon has traditionally had the highest forest loss in the world and, thus, represents a likely target for future REDD initiatives. This paper presents an ex-ante assessment of the potential REDD costs in two of the three largest states in the Brazilian Amazon using official land use and cover change statistics. The two states, Mato Grosso and Amazonas, have historically experienced rather different land use dynamics. The findings focus on the opportunity costs of REDD and suggest that at least 1 million ha of projected deforestation in Mato Grosso and Amazonas could be compensated for at current carbon prices until 2016. Total costs may differ between US$ 330 million and over US$ 1 billion depending on how payment mechanisms are designed. Implications of payment scheme design for the political economy of REDD are discussed. Key words: Opportunity costs, REDD, payments for environmental services, carbon supply, land use Does REDD make sense in the Amazon region? Both the International Panel on Climate Change (IPCC) and the Stern Review on the Economics of Climate Change reckon that avoiding deforestation accounts for a significant share of the global potential for climate change mitigation through forest related activities (IPCC 2007, Stern 2007). For many years, Brazil has been the country with the highest areas of tropical forest clearing by far. Its dynamic agribusiness sector has led an aggressive expansion of the agricultural frontier in the Amazon region. Chomitz and Thomas (2001) found that, until 1996, more than three quarters of deforested land has ended up under pasture. In fact, extensive cattle production continues to strongly dominate land use in the Brazilian Amazon, even if more recent evidence indicates that cropland now expands faster than pastures in relative terms (Morton et al. 2006). Model based simulations suggest that between now and 2050 primary forest clearing in the Amazon region may release up to 32 Pg of carbon into the atmosphere – an amount roughly twice the global annual anthropogenic greenhouse gas emissions (GHGs) (Soares-Filho et al. 2006). While farmers, the local and probably also the national economy have benefited from clearing forests for agriculture (Andersen et al. 2002), continuous deforestation not only accelerates 123 climate change but also threatens the provision of other important globally and regionally important ecosystem services, such as biodiversity protection, hydrological, and local as well as regional climate regulation. Thus, it seems wise to intensify the search for flexible policy mechanisms that translate the demand for such global public services into local economic incentives for conservation. Traditional command-and-control policies have been rather ineffective in curbing deforestation in the Brazilian Amazon. The Código Florestal has been the prime legal instrument for forest conservation on private lands since 1965. But due to lax enforcement, illegal deforestation contributes the lion’s share to forest loss in the Brazilian Amazon. During 2005-06, deforestation rates had dropped sharply. At the international Conference of the Parties on climate change in December 2007 in Bali (COP13), many hoped this was a lasting reduction, to be attributed to better rural licensing systems, increased fines for illegal clearings, and other policy actions by the Brazilian government under its ambitious Plan to Combat Deforestation.9. However, in early 2008 the Brazilian Space Research Centre (INPE) reported that deforestation has accelerated again sharply during the second half of 2007, probably in response to the recovery of international soy and meat prices. Infrastructure expansion and other development policies combined with high food-commodity prices and rising demand for biofuels will add to Brazilian agricultural land demand and to forest-conversion pressures in the foreseeable future. Enforcing command-and-control policies at the scale of the Amazon region is thus unlikely to work as a stand-alone strategy. It is against this backdrop that the debate on Reduced Emissions from Deforestation and Forest Degradation (REDD) has gained momentum, both internationally and inside Brazil. The COP13 decided to include REDD in future negotiations on mitigation mechanisms for countries that have not adopted any emission reduction targets. Several proposals to implement REDD in the Brazilian Amazon were also presented. Drawing on its experiences with Bolsa Floresta, a pilot compensation scheme for avoided deforestation on smallholdings, the government of the Brazilian State of Amazonas proposed a REDD scheme at the state level (Government of Amazonas 2007). An NGO consortium sketched the outlines of a proposed payment for environmental services (PES) scheme for avoided Amazon deforestation10. A sub-group of these NGOs presented a report that provides the scientific underpinning for a national-level REDD scheme to boost Amazon conservation (Nepstad et al. 2007). The evidence presented in the following builds on calculations made by the authors for the first two proposals. The challenge of quantifying potential REDD supply has both a temporal and a spatial dimension. First, credible temporal baselines are needed to project forest-cover change relatively far into the future. Second, the total cost of implementing a payment scheme has to be estimated for different locations with variable environmental and economic conditions. Spatial disaggregation generally contributes to better targeting of direct payments, which will result in 9 “Cutting down deforestation in the Brazilian Amazon”. Report published by the Brazilian Ministry of th Environment at the COP13, December 12 2007, Bali, Indonesia. 10 Pacto pela Valorização da Floresta e pelo o fim do Desmatamento na Amazônia (Forest Valuation Pact). http://www.icv.org.br/publique/media/PactopelaValorizacaodaFlorestaepeloFimdoDesmatamento_sumario.pdf 124 more efficient PES scheme (Wünscher et al. 2008). Yet, scientific assessments of the supply side of Amazon REDD have so far been scarce. In a multiple-country background study for the Stern Review, Grieg-Gran (2006) estimated avoided deforestation in Brazil to cost US$1.2-1.7 billion, depending on whether timber rents are included. Nepstad et al. (2007) expected avoiding 6.3 Pg of carbon emissions in the Amazon over 30 years to cost considerably more (US$ 8.2 billion)11. In spite of the diverging total cost estimates, both studies suggest that REDD at current carbon prices might be competitive vis-à-vis the conservation opportunity costs12 of private development of Amazon land for crops and pastures. Current Brazilian deforestation can be said to occur at four different levels of (il)legality. First, landowners can legally clear up to 20% of their land area (private landowners in the Amazon are required to keep 80% of their farm area as a Legal Forest Reserve.). Secondly, they could pass that legal clearing threshold and develop a so-called ‘environmental deficit’ on their land – a phenomenon that is widespread (and tolerated) in many old frontier areas. Third, private individuals could invade and clear forest on weakly enforced state land (terra devoluta), in the realistic hope of establishing land tenure over time. Finally, land invasion could happen in declared national parks, indigenous and extractive reserves, etc. To counteract the third and fourth types of deforestation, international REDD payments could be used for financing improved command-and-control systems. However, at least inside existing parks and reserves, compensation payments would appear pointless, because the Brazilian federal or state governments have legally delimited them to ensure protection. Thus, this study will focuses on direct compensations to private landowners. This refers to the first and, possibly, to the second legality scenario – given strong political pressures to lower the 80% legal reserve threshold or allow landowners to pay their way out of ‘environmental deficits’. PES-type compensations will likely become an important element in Amazon REDD schemes. To make forest conservation attractive to landowners, such transfers have to exceed their land opportunity cost – at least as long as command-and-control policies are not duly enforced. Hence, this article aims to contribute to the REDD debate in two ways. First, it evaluates the economic feasibility of REDD using municipal-level production data for the private lands of two of the largest Brazilian states, with a combined area equal to 47% of the Legal Amazon). Secondly, it uses these results to provide guidance for REDD design that combines cost effectiveness with equity concerns. Section 2 provides an overview of the two case study areas and the context for REDD in the Brazilian Amazon. Section 3 describes the methods and data used to arrive at the results presented in section 4. Section 5 interprets the results from a political economy perspective and section 6 presents the main implications of this study. Finally, section 7 discusses some key assumptions and compares the findings with those from other REDD opportunity cost studies. Future perspectives of REDD in the Amazon are discussed as well. 11 Per ton of carbon values are less diverging. See Section 7 for explanation. The economic returns to converting forest to other uses minus the current economic benefits derived from the standing forest 12 125 Study area: Brazilian Amazon, Mato Grosso, and amazons Only roughly 25% of land in the Brazilian Amazon is private. About 35% is indigenous territory or protected by federal- or state-level protected areas. The remaining land is considered public with weakly enforced tenure (terra devoluta) (Toni 2006). State indigenous territories or protected areas cover over 30% of total area in Amazonas state and 20% in Mato Grosso. Land concentration is comparatively high in the Amazon (see table 1), with regional Gini indices remaining around 0.85 between 1950 and 1996. During the same period, the Gini index reduced from 0.9 to 0.8 in Amazonas and remained almost constant at 0.85 in Mato Grosso (ADA 2002). Both the small share of private lands and the high concentration of land ownership have important implications for REDD, which will be addressed in Section 5. Figure 1 shows the location of Mato Grosso and Amazonas and the main roads and riverways, while table 1 gives comparative statistical figures. Amazonas is the largest and second-least densely populated state in Brazil. Per-capita income is among the lowest in Brazil -- especially outside the capital Manaus with its free-trade zone. Amazonas is remotely located from the main Brazilian markets in the South and its cities are mainly accessible only through fluvial transport. Despite some large-scale cattle operations, smallholders with less than 100 ha own more that a third of private land. Crops (annual and permanent) and pasture each account for about 40 % of total land use. In recent years the state implemented many environmentally friendly policies, increasing protected areas and creating positive incentives for conservation. As a combined result of economics and policies, deforestation in Amazonas has been low, both in absolute and relative terms. 126 Figure 1: Location and main transport ways of the states of Amazonas and Mato Grosso Table 1: Key features of Amazonas (AM) and Mato Grosso (MT) states Units AM MT [million 1.57 0.90 Area km2] Forest cover (2006) [%] 90 36 Forest carbon (2006) [Mt C] 16 000 3 600 [km2 910 (0.1) 6 650 (2.5) Average annual forest loss (2000-6) (%)] [people Population density (2000) per 1.79 2.77 2 km ] [US$ Income per capita (2000) per 1 148 1 901 year] Share of farms smaller than 100 ha (1995/6) [%] 94 60 Total area of farms smaller than 100 ha (1995/6) [%] 34 3 Sources: UNDP, IBGE, FAO, Houghton et al. (2001) *Calculated from FAO data (2000-5) Brazil 8.51 56 n.a. 31 030 (0.6)* 19.92 1 962 89 20 127 In contrast, Mato Grosso lies in the heart of the so-called ‘Arc of Deforestation’ at the southern edge of the Amazon basin. It has a relatively dense road network and is well connected to the main population centers in Brazil’s Center and South. Mato Grosso has a strong commercial agricultural sector, dominated by extensive cattle and soy production (IBGE 1995/6). Soy and cattle expansion are also responsible for Mato Grosso being the Brazilian state with highest deforestation -- in the last decade more than one third of total forest loss in the Brazilian Amazon. The state has historically adopted policies that favour land-extensive economic development. In 1999, the government of Mato Grosso introduced a Licensing System for Rural Properties (SLAPR) (Fearnside 2003), which many hoped would help curbing deforestation rates. Today, however, enrolment in the SLAPR is still below 30%, and much of the recent pickup in deforestation has been registered in Mato Grosso13. Figure 2: Municipal deforestation rates in Amazonas and Mato Grosso during 2000-06 13 Brazilian Space Research Institute (INPE): Online Communication 24.01.2008 (http://www.inpe.br/noticias/noticia.php?Cod_Noticia=1318) 128 Figure 2 shows the distribution of average 2000-6 deforestation rates in Amazonas and Mato Grosso, which serve as baselines for future deforestation in the REDD opportunity-costs calculations below. Deforestation is far higher in Mato Grosso than in Amazonas both relatively and absolutely. Although growth in total land under agricultural crops (in Mato Grosso, especially soy) has been faster than expansion of pastures, pasture remains the predominant converted land cover in both Amazonas and Mato Grosso. As Figure 3 shows, soybeans have started to dominate the land-use mix in a few municipalities in the centre and southeast of Mato Grosso, some of which lie in the transition zone between Amazonas and Cerrado biomes. In Amazonas, crops generally have a higher share in the municipal crop mix than in Mato Grosso, due to the more diverse and subsistence-oriented smallholder sector. That said, in 2006, municipalities in Amazonas had on average 2% of their total area deforested, as opposed to Mato Grosso, where 21% had been denuded from natural forests. In the westernmost remote municipalities in Amazonas, the little land that was converted during 2000-06 is exclusively covered by crops, which be explained partially by their large indigenous territories. Deforestation rates are high in Mato Grosso in both soybean- and pasture-dominated areas, suggesting both activities contribute considerably to forest loss. Figure 3: Dominance of crops vs. pastures on deforested land in Amazonas and Mato Grosso states between 2000-06. 129 Data and Methods One can estimate the opportunity costs of forest conservation using various approaches, ranging from economic optimization or general equilibrium models (Cattaneo 2002, Börner et al. 2007) to using land prices as surrogates for the discounted stream of future deforestation returns - see Grieg-Gran (2006) for a discussion. Nepstad et al. (2007) calculate REDD opportunity costs based on simulated returns to soy and cattle production on land their model predicts will be cleared in the future. In their approach, land opportunity costs depend heavily on distance to roads and on suitable soil and climate conditions. The Nepstad et al. study considers only returns to timber, cattle, and soy bean production. Slashand-burn agriculture, an important element in Amazonian agricultural landscapes was not considered. Moreover, the profit rates are based on simulations and not on actual data. Hence, we believe that the complementary approach presented in this paper (i.e. based on INPE annual deforestation rates and municipal agricultural production data from the Brazilian Institute for Geography and Statistics (IBGE)) can help to complete the picture. The IBGE Municipal Agricultural, Animal, and Extractive Production data bases (PAM/PPM/PEV)14 holds annual information about total cultivated area, yields, and total production value for all Brazilian municipalities. These data are not field measurements, but expert estimates collected in annual consultations of local extension agents, government officials and IBGE staff. Comparisons with the latest agricultural census (1995/96) show that the expert estimates put forward in the PAM/PPM/PEV data bases largely correspond to measured census data as far as municipal averages of yields and prices are concerned. Satellite-based annual deforestation measurements from INPE are frequently higher than the PAM/PPM/PEV estimates of growth in cattle herds and cultivated area, which leads us to be less confident in the latter. In the Amazon region, technical coefficients15 and cost information are not available at the municipal level. The estimates thus heavily rely on national-level estimates for main agricultural crops from the Brazilian Agriculture Yearbook (FNP 2007) and Amazon-specific estimates by Margulis (2004) for cattle ranching and Pokorny and Steinbrenner (2005) and Barreto et al. (1998) for timber harvesting. All monetary figures have been converted to 2006 US dollars using the Brazilian consumer price index IPCA and the average 2006 exchange rate. The following opportunity-cost estimation is limited to private landholdings, since direct payments to farmers invading public lands could easily create perverse incentives for additional forest clearing. For Amazonas State, calculations rely on the rural land register published by the National Institute for Colonization and Agricultural Reform (INCRA). INCRA data are often inconsistent with agricultural census information, which reflects the considerable uncertainty with regard to land-tenure data in Brazil. Especially in Mato Grosso, where aggressive land grabbing has taken place for many years, INCRA data are also inconsistent with municipal boundaries. Hence, INCRA data are used only for Amazonas, whereas estimates for Mato Grosso are restricted to farms registered in the SLAPR (i.e. roughly 25% of farms in the rainforest areas of the state). 14 Portuguese ancronyms used by the IBGE Parameters of agricultural production, e.g. amount of labor and other inputs needed to produce a given level of output (yield). 15 130 Figure 4 depicts the main analytical steps to calculate opportunity cost of REDD. Past municipallevel deforestation rates are calculated from INPE PRODES16 data and projected linearly into the future for the period 2007-16. INCRA and SLAPR data serve as the basis for calculating the share of private land in each municipality. While the SLAPR database for Mato Grosso directly records remaining forests on private land, forestland on private properties in Amazonas state needs to be estimated. It is assumed the amount of forest left in Amazonas corresponds to total private land less land currently under pastures and crops. This may overestimate remaining forests in 2006, as one would expect a minor share of private land to be in fallow (3% in the agricultural census of 1995/6). Baseline Private Land Private Forest Land use INPE INPE 2000-2006 2000-2006 INCRA/ INCRA/ SLAPR SLAPR IBGE IBGE 2000-2006 2000-2006 IBGE IBGE 2000-2006 2000-2006 Land use expansion Gross Return per Land use Cost/Benefit ratios per Land Use IBGE IBGE 2000-2006 2000-2006 IBGE IBGE 2000-2006 2000-2006 FNP FNP + others + others Opportunity cost (PES scenario) Figure 4: Data sources and calculation steps for REDD opportunity costs. As mentioned, land-use mixes for each municipality are calculated on the basis of PAM and PPM data. PPM data on cattle-herd size per municipality is used to impute pasture cover, assuming 1995/96 stocking rates to remain constant in both states. State-level growth rates of land under pastures and crops (permanent and annual) are then applied to estimate the growth of land in particular land use categories, such as annual subsistence crops produced in slash-andburn systems, traditional cash crops, fibres, and fruits. Each land-use category is represented by the single crop with the highest share in 2000-6 total land use expansion, e.g. soy beans for the category cash crops in Mato Grosso. Gross per-hectare returns of crops were also calculated from PAM and PEV data. No such information is available for timber extraction, so yields and per-ton extraction costs reported by Pokorny and Steinbrenner (2005) and Barreto et al. (1998) were used in calculations for Amazonas. Timber yields for Mato Grosso were adjusted according to estimates provided by the Forest Management Unit of the Environmental Secretariat of Mato Grosso17. Gross returns from each selected land-use category were converted to net profits as follows: 16 INPE’s Program for the Calculation of Deforestation in the Amazon (PRODES) publishes annual deforestation estimates for the Amazon. 17 Personal Communication: Secretariat of the Environment (SEMA), Forest Management Unit 13.05.2007 131 ck (1) ) bk where Пik is net per-ha profit per ha of crop k in municipality i , GR are annual gross per-ha returns in municipalities calculated from the PAM/PPM/PEV data base, whereas b and c are per-ha gross returns and total costs, respectively, derived from other sources. Π ik = GRik * (1 − Profitability of extensive cattle operations was taken from Margulis (2004), assuming his highend estimates to apply for Mato Grosso and low-end estimates for Amazonas -- cattle ranching being less capitalized in the latter than in the former. Vosti et al. (2002) and others show that land use after deforestation often follows similar patterns, which we call land-use trajectories. For example, forests are often cleared first for annual subsistence crops, after which land is put under pasture or repeated cycles of fallow-based slash-and-burn agriculture. To calculate REDD opportunity costs, hypothetical land use trajectories were set up that represent a sequence of individual land use categories. Figure 5 shows examples of such trajectories in a stylized form. timber extraction food crops (1) extensive cattle (2) fallow fallow (3) cash crops NPV extensive cattle (1) (2) (3) annual deforestation fallow fallow cash crops years Figure 5: Stylised sequences of land uses applied in the opportunity-cost estimations Note: Percentages represent hypothetical shares in the municipal land-use mix Figure 5 depicts how municipal opportunity costs were calculated from individual land-use sequences at the plot level. All-land use trajectories start with timber extraction, followed by subsistence-crop production in the second year, while then some land goes into pasture (1), some into crop-fallow cycles (2), and some is used for cash crops (3). Net present values (NPV) of land-use trajectories in a given municipality were calculated (see equation 2) using a 10% discount rate over a ten-year planning horizon, and are reported in Table 2 below. 132 The same amount of new land was assumed to be opened each year, such that the ten-year period 2007-16 covers the accumulated NPV of the benefits derived from the corresponding land-use trajectories (see Equation 3). This step is necessary, because the NPV of a given 10-year land use trajectory beginning, say, in year 2010 is worth less to the farmer than beginning the same trajectory one year earlier. The municipal land-use mix was adjusted annually according to the state-level growth rates of agricultural land versus pastures during 2000-06. The shares of subcategories within these two broad categories of land use were held constant over time for each municipality. NPV j = ∑ t Π k =1,t =1 (1 + r ) t =1 + Π k = 2,t = 2 (1 + r ) t =2 + ... + Π k = K ,t =T (1 + r )t =T (2) where NPVj is the net present value per ha of land use trajectory j in a given municipality and k depicts the different crops/land uses that follow each other during a ten year planning horizon in j. NPVi = ∑ t ∑ s NPV j j (1 + r ) t ijt (3) where NPVi is the net present value per hectare (average opportunity cost of avoided deforestation) in municipality i, s is the share of land use trajectory j in the total municipality’s annual land use expansion, and NPVijt is the net present value of the ten-year land use trajectory j in year t of the REDD scenario, while r is the discount rate. Transport costs are accounted for by creating a cost index, which reduces the municipality’s net agricultural returns proportionally to how far it is located from the state capital. Transport costs are assumed to be zero in the municipality of the state capital, and then increase linearly with distance up to a maximum of 20%, i.e. profits in the remotest municipality are only 80% of gains prior to calculating transport costs. We thus ignored for the sake of simplicity that difficult access conditions in the remotest areas could lead to higher reductions in net profits for bulky produce, due to their more pronounced sensitivity to transport costs. Finally, we assumed that carbon dioxide emissions resulting from deforestation correspond to the total carbon content in above-ground vegetation. Hence, opportunity costs per ton of avoided carbon dioxide emissions are equal to per-hectare opportunity costs divided by average carbon content (see next section). Presenting opportunity costs per ton of carbon dioxide (the commonly traded unit on existing carbon markets) allows evaluating the competitiveness of REDD carbon both in terms of municipal averages (see Figure 6 below) and in terms of land-use trajectories (see Figures 7 and 8). Analysis and Results How large gains would landowners forego? Table 2 presents average profits calculated for the main expanding land-use categories in Amazonas and Mato Grosso. It shows that soybean plantations are clearly the most profitable 133 land-use option among those that contribute to forest loss in the two states. For the sake of simplicity, it is assumed that no returns are derived from standing forests, so the profits from converted uses are identical to the opportunity cost of conserving the forest. Table 2: Net returns and importance of crops and land use categories in the opportunity cost estimation (10% discount rate, 10-year period) Total net Average Average Share in total return annual net NPV of Land 2000-06 return Use Trajectory expansion* [US$/ha] [US$/ha] [US$/ha] [%] Amazonas Timber extraction 24-791 Extensive cattle 694 ranching 39 86 Food crops (corn) 39 475 6 Cash crops (coffee) 93 650 3 Fruits (water melons) 41 393 1 Fibres (malva) 24 307 4 Mato Grosso Timber extraction 109-734 Extensive cattle ranching 59 Cash crops (soybeans) 171 * Shares in total expansion refer to land use categories. 719 1 080 84 16 Note that NPV values for land-use sequences are strongly influenced by the returns to timber extraction in the municipalities that report timber extraction in past years. Due to fallow periods, during which returns to land are zero, the NPV for staple crops is considerably lower than for cattle production, even though average annual returns are equal. Values in the last column of Table 2 show the share of each land-use category in total 2000-6 expansion of agricultural land. In the case of crop categories, these values correspond to the crops shown in brackets in the first column. Opportunity costs per ton of carbon dioxide depend heavily on the amount of biomass and, hence, carbon content per hectare of primary forest, which varies widely across the Amazon region (Saatchi et al. 2007). Houghton et al. (2001) present data from seven independent studies analysing carbon content of forest biomass in the Amazon. To provide a conservative estimate of opportunity costs, this study adopts the lowest estimate presented in the Houghton et al. study (110 tons C per ha) for forests in the state of Amazonas, and assumes that 20% of this would be kept as an insurance reserve. For Mato Grosso, the same procedure was applied to more detailed carbon content data provided to the authors by the Instituto Centro de Vida (ICV)18. 18 Instituto Centro de Vida (www.icv.org.br) is a subscriber to the Forest Valuation Pact, and was intensively involved in the research underlying the Pact. 134 Spatial distribution and abatement cost curves Figure 6 shows average REDD opportunity costs per ton of carbon dioxide at the municipal level. Average values are highest in Mato Grosso, although many municipalities with high opportunity costs lie in savanna (cerrado) regions19 with lower natural biomass density. In Amazonas, many high opportunity cost municipalities lie alongside road and fluvial transport ways (see Figure 1). Opportunity cost differences in pasture-dominated parts of Mato Grosso are mainly caused by high returns to timber extraction prior to forest conversion. In general, opportunity costs differ remarkably across space -- not only between but also within the two states. Figure 6: Municipal opportunity costs per ton of carbon dioxide in Amazonas and Mato Grosso 19 Municipalities were defined as being “savanna-dominated” if savanna areas were larger than forest areas. However, only areas classified as forest in the INPE data base were considered in this study’s calculations. 135 NPV US$ per ton of CO2 4 +95% Perc, -5% Perc CCX permanent CCX temporary 3 2 1 0 0 100000 200000 300000 400000 500000 Avoided deforestation (ha) Figure 7: Opportunity cost per avoided ton of carbon dioxide in the State of Amazonas. Notes: - CCX permanent – full average price of per ton of CO2 at Chicago Climate Exchange - CCX temporary -- includes a 39% rebate on permanent carbon prices. - Grey areas represent values that lie in a 5-95% sensitivity range. 15 NPV US$ per ton of CO2 12 +95% Perc, -5% Perc CCX permanent CCX temporary 200000 800000 1000000 1200000 9 6 3 0 0 400000 600000 Avoided deforestation (ha) Figure 8: Opportunity cost per avoided ton of carbon dioxide in the State of Mato Grosso. Notes: - CCX permanent – full average price of per ton of CO2 at Chicago Climate Exchange - CCX temporary -- includes a 39% rebate on permanent carbon prices. - Grey areas represent values that lie in a 5-95% sensitivity range. 136 For the moment all calculations assume zero transaction costs (to be relaxed in next section). Figures 7 and 8 present carbon-dioxide emission abatement costs (REDD supply curves) for Amazonas and Mato Grosso, respectively. As a benchmark, both figures include 2006 average prices for permanent carbon credits traded at the Chicago Climate Exchange (CCX) carbon market. However, since the authors expect that REDD payments are likely to be introduced in the form of temporary carbon credits, the figure shows a hypothetical price line with a 39% rebate on current CCX prices. The rebate was calculated following Dutschke and Schlamadinger (2003), given that carbon credit buyers will have to reinvest in new credits by the time their temporary credits expire (i.e. here assumed after ten years). The CCX carbon market is voluntary, which means that prices per ton of carbon dioxide are at the lower end if compared, for example, to carbon prices in the European Union Emissions Trading Scheme or project-based transactions under the Kyoto Protocol. Whether REDD carbon will be traded in the form of permanent or temporary certified emission reductions has not yet defined, which is why we consider both options in Figures 7 and 8. The grey ‘bands’ in Figures 7 and 8 show the result of sensitivity analyses varying key parameters such as product prices and per-ha carbon content by ±30%, to account for expected market fluctuations and perceived uncertainties. The supply curve for Amazonas shows that more than one third of deforestation is worth less than US$1/tCO2, and thus profitable to buy out under almost any carbon-market scenario. Going towards the right the curve starts sloping, but there is in Amazonas no deforestation worth more than US$3/tCO2 -- at least at the aggregated municipal-average level. The situation is slightly different in Mato Grosso. While around half of deforestation is worth less than US$3/tCO2, with a relatively flat curve, the other half is more heterogeneous and rises to values around US$12/tCO2. How much REDD is economically feasible? What does this mean for the competitiveness of REDD as a land-use option? Table 3 compares the opportunity-cost results in Mato Grosso’s SLAPR areas and in Amazonas State to three carbon-price situations (rows 1-3): (1) maximum price (i.e. the hypothetical price needed to buy out all deforestation) (2) permanent CCX price (value in 2006) (3) temporary CCX price (same as (2), but with a 39% discount – see above). On the payment side, two generic scenarios (two last columns) are shown. First, “opportunitycost payment” (Scenario I) implies that each farm receives differentiated compensation payments corresponding to their pure opportunity cost values. Graphically, this corresponds to the area under the emission abatement-cost curves in Figure 7 and 8. The (extreme) assumption here is that payments can be perfectly differentiated, so that provider economic rents are fully eliminated. Secondly, under “marginal pricing” (Scenario II) all providers receive the same uniform payment, determined by the farm with the highest opportunity cost. Graphically, payment value thus not only corresponds to the area under the supply curve, but to the entire price-times-quantity rectangle: cheap REDD suppliers (on the left-hand side of the curve) 137 capture a “provider surplus”, i.e. the difference between the market price and their individually lower costs of supplying REDD. The maximum carbon price (row 1) needed to compensate all deforestation costs would be almost US$13/tCO2 – most of all due to a few municipalities in Mato Grosso’s SLAPR areas with very high opportunity costs for conserving forests. Focusing first on Scenario I (pure opportunity-cost compensation), this would lead to payments of US$680 million to achieve zero deforestation in all SLAPR areas of Mato Grosso by fully covering all producers’ economic returns from deforestation. In Amazonas, the total would be only US$143 million, both because there is less deforestation and because the average per-hectare opportunity cost is lower. The permanent CCX price of US$3.88/tCO2 in 2006 (row 2), would compensate farmers to reduce Mato Grosso’s SLAPR deforestation by two-thirds, at a total cost of US$381 million; the permanent CCX price would also compensate for all projected forest loss in Amazonas. At temporary CCX prices of US$2.32/tCO2 (row 3) – a conservative estimate – 40% of SLAPR areas enter REDD at costs of US$212 million, while US$123 million can compensate for 93% of Amazonas deforestation. Hence, at current carbon price ranges, the bulk of deforestation can potentially be compensated, especially on the low-opportunity cost lands that predominate in Amazonas. Table 3: Opportunity costs and area coverage in Mato Grosso (SLAPR) and Amazonas under different payment scenarios and carbon prices (10% discount rate, 10-year period) Scenario I Scenario II Opportunity cost payment Marginal pricing payment Units Mato Grosso Amazonas Mato Grosso Amazonas (1) Maximum price (MT US$/tCO2 12.36) and (AM US$/tCO2 3.24)* Total opportunity cost mill US$ 680 143 2 745 363 Reduced forest loss % 100 100 100 100 Reduced forest loss ha 1 375 385 564 849 1 375 385 564 849 (2) CCX permanent price (US$/tCO2 3.88) Opportunity cost mill US$ 381 143 677 363 Reduced forest loss % 62 100 62 100 Reduced forest loss ha 850 122 564 849 850 122 564 849 (3) CCX temporary price (US$/tCO2 2.32) Opportunity cost mill US$ 212 123 274 239 Reduced forest loss % 40 93 40 93 Reduced forest loss ha 554 842 525 094 554 842 525 094 What if one has to compensate farmers at a fixed marginally determined price, rather than ‘just’ their pure individual opportunity costs (Scenario II, last column)? Obviously, this does not change the amount of forest area protected, but distribution-wise a ‘provider’s surplus’ is created, thus increasing costs. Potentially, this economic rent can be sizeable, especially at high carbon prices and heterogeneous producer costs. For the maximum price situation (line 1), costs 138 in Mato Grosso’s SLAPR areas would quadruple to US$2.7 billion, three fourths of which would accrue to low-cost suppliers as windfall gains (i.e. compensations paid in excess of opportunity costs). At temporary carbon prices (3), these gains are less astronomic. For instance, for Mato Grosso’s SLAPR areas the costs rise only from US$212 to US274, since this corresponds to the low-sloping section of the supply curve. But for Amazonas, costs still more than double, from US$123 to US$239 million, because a large part of Amazonas’ potential REDD credits are very low-cost and would fetch economic rents even under moderate prices. These findings for Scenario II have important implications for REDD design. Rising carbon prices would multiply economic rents accruing to low-cost providers. There would thus be large efficiency gains for REDD buyers in introducing some sort of differentiated payment system (according to location, producer types, land values, etc.) that caters to highly variable provider opportunity costs. The flip side is that price differentiation would also eat into the ‘provider’s surplus’, which represents the potential welfare gain on behalf of farmers, including for poverty alleviation. In practice, probably neither a uniform nor a fully differentiated price is very likely, but for analytical purposes they represent extreme scenarios that help us understand the competitive and distributional consequences of different payment modalities. The results prove to be particularly sensitive to the returns from timber extraction. One-off timber rents can in some cases be sizeable, and since they accrue at the beginning of each landuse cycle, they are not being time-discounted. They can thus potentially gain high influence on the overall NPV results. However, timber rents are also often at least partially captured by actors other than the landowner proper, and their harvesting may happen well in advance (and causally divorced) from the deforestation process proper. Setting timber extraction profits to zero, for analytical purposes, would allow REDD transfers at temporary CCX prices to compensate more than 80% of forest loss in Mato Grosso and 100% of forest loss in Amazonas at current (temporary) carbon prices. This reconfirms that the timber economy, and the second “D” in REDD, merit further analysis. Apart from timber rents, total opportunity costs are most sensitive to beef prices, e.g. a 30% price reduction decreases total opportunity costs by 9% in Mato Grosso and 10% in Amazonas, followed by soybean prices (Mato Grosso) and food crop prices (Amazonas). That is due to the dominance of the related land uses in overall crop mix. Prices per ton of carbon dioxide are particularly (and proportionally) sensitive to changes in the amount of tradable emission reductions assumed per hectare of avoided deforestation. Finally, discount rate changes also affect total opportunity costs to a considerable extent. For example, reducing the assumed 10% discount rate to 5% would increase total costs in Mato Grosso by roughly one third. How large could transaction costs be? Of course, opportunity costs are only one part of the story: transaction costs also need to be paid for through the REDD resources. Relatively little is known about the transaction costs of payments for environmental services (PES) schemes in general, less so for still to-be-developed direct REDD compensations to landowners. Transaction costs are defined as all costs of an environmental services payment scheme that are not transfers proper. Transaction costs occur both on behalf of the carbon buyer (e.g. having to monitor compliance) and the seller (e.g. having to comply with payment modalities). 139 Ex-ante transaction-cost estimates have to be interpreted with caution. May et al. (2003) note that many incipient carbon-based PES schemes have incurred extremely high transaction costs, mainly because of the difficulties involved in developing forest carbon projects in an uncertain market environment. As a consequence, pioneering carbon investors have required projects to repeatedly revise strategies throughout project implementation. In general, PES schemes seem to require relatively large start-up costs, while running costs tend to be more manageable, as shown for a series of carbon projects in Indonesia (Cacho et al. 2005). Turning to South America, in two Ecuadorian PES cases of Pimampiro (watershed protection) and PROFAFOR (carbon sequestration), start-up costs were US$76/ha and US$184/ha, respectively, while recurrent annual per-hectare transaction costs in the operational phase were only US$7 and US$3 (Wunder and Albán 2008). In the Amazon, the authors expect transaction costs to arise mainly in the categories presented in Table 4. Especially if smallholders are to be involved in Amazon-REDD schemes the need to handle a large number of small volume transactions will represent a major challenge for the desing of direct payment mechanisms. Table 4: REDD transaction costs and implications for REDD in the Amazon Transaction cost category Comments 1. Information and Currently, carbon markets are not prepared for large-scale REDD procurement in the Amazon and carbon buyers have traditionally been reluctant to invest in carbon projects in the forestry sector. Procurement costs can therefore be expected to be significant. 2. Scheme design and Large-scale REDD schemes may incur significant negotiation negotiation costs, especially if they contemplate payments from national government budgets that need to be negotiated with the civil society. 3. Implementation Existing organisations and institutions needed to be strengthened and systems like SLAPR implemented in all areas covered by REDD. Establishing and running payment mechanisms (especially in the case of direct payments to landowners) are likely to contribute the lion’s share to this cost item. 4. Monitoring In some states, rural licensing systems are in place that would allow annual deforestation monitoring at farm-level scales. The technology for satellite-based deforestation monitoring is relatively well developed and much more cost-effective than ground-based monitoring. 5. Enforcement and Enforcement costs might be considerably reduced by delivering protection payments only after verification of effectively avoided deforestation. Given weakly enforced property rights in large parts of the Amazon, enforcing theses rights (e.g. in and around protected areas) might prove crucial to assuring additionally of REDD and, hence, represent a relevant source of transaction costs. 7. Verification and These cost items have shown to be an important barrier for smallcertification (Approval) scale carbon forestry projects (Cacho et al., 2005), but are expected to decrease with project size. Source: Adapted from Milne (1999) 140 Based on information from Environmental Secretariat of Mato Grosso, a hypothetical state-level REDD scenario was set up. The scenario involves the creation of a carbon payment fund that cooperates with existing government and civil society organizations in implementing direct REDD payments to land owners in Mato Grosso. This allows assessing likely transaction costs in the categories 3, 4, and 7 of Table 4. Start-up costs are estimated at US$7.5/ha and annual implementation costs at US$4.5/ha of avoided forest loss. Recurrent costs are thus slightly higher than what Grieg-Gran (2006) calculated for the Costa Rican national PES scheme (US$3/ha/yr). Depending on biomass density, transaction costs in Mato Grosso with these absolute values would range from US$0.07 to 0.24 per ton of carbon dioxide during a 10 year period, or a total of US$49 million. Given temporary CCX prices, thus would marginally shift up the emission abatement cost curve in Figure 8, so that cost-effectiveness in terms of deforestation avoided would be reduced by roughly 3%. This addresses the transaction costs of buyers or intermediaries, but what about service providers? Poor transport infrastructure (e.g. in Amazonas’ remote areas) can potentially drive up their transaction costs in negotiating contracts and cash in payments. REDD initiatives might learn important lessons from other experiences with decentralized conditional cash transfers, such as the Brazilian Family Assistance Program (Bolsa Familia) and the Amazon State’s avoided deforestation program Bolsa Floresta (Hall 2006). The political economy of redd The Amazon framework conditions for REDD described in Section 1 also have implications in terms of: 1. Who may be the winners and the losers? 2. Which areas become eligible for REDD? 3. What share of the REDD potential can be considered truly additional? First, REDD will only attract large-scale investments if additional emission reductions can be credibly demonstrated. For a region with highly unequal land and power distribution like the Brazilian Amazon, smallholders and forest-dwelling communities may not be the prime beneficiaries if additionality is put at the forefront. Chomitz (2006) shows that less than 20% of one-time forest clearings in the Amazon are small-scale, i.e. smaller than 20 ha. Larger clearings are generally out of the range of smallholders. To the extent it is necessary to compensate those who would benefit from (legal) deforestation, and thus suffer the opportunity costs of conserving the forest, a rather high share would need to go to medium-sized and large commercial farmers. On the other hand, for a REDD programme to be politically acceptable in Brazil, and to avoid significant leakage to the smallholder sector, it may turn out to be beneficial to invest a share of REDD money that is more than proportional to the related threats into rewarding good forest stewards and local communities for assistance in monitoring protected areas. A general sense of fairness will be crucial for the political acceptance of REDD, both in ES buyer and seller countries. 141 An example may underscore this point. The Forest Valuation Pact, a recently proposed scheme to compensate farmers for not deforesting, to be funded primarily with Treasury resources, received mixed public reactions. It was criticised that Brazilian taxpayers should pay for services that benefit society globally especially when the beneficiaries would be large commercial landowners with a history of aggressive land clearing (such as in Mato Grosso) – i.e. rewarding “bad” rather than the “good guys”. However, public acceptance of such compensations, at national level, would likely be higher if REDD was funded by international carbon markets, rather than Brazilian taxpayers. Second, only some of the highly threatened forests in the Brazilian Amazon can potentially be protected through direct REDD payments, because much of the land cleared is public or has insecure tenure. Direct payments to farmers on land with deficient land tenure rights will be inefficient – and paying land grabbers to desist from invasions would likely create perverse incentives for others to simulate similar clearing threats to claim compensation. As for the large protected areas and indigenous territories, many lie in remote and relatively undisturbed areas where de facto threats are low, and payments here could easily become “hot air”. Deforestation within protected areas has been relatively low, compared to outside (see Ferreira et al. 2005 for a comparison of deforested areas in and outside protected areas), though part of this may be explained by remoteness rather than protection status. Studies of less remote protected areas in the state of Pará show illegal deforestation there can be almost as high as the regional averages (Velásquez et al. 2006). Yet, from a legal point of view, paying REDD in these areas based on opportunity costs is highly questionable. At best, one could imagine the use of REDD to cofinance the creation of new protected areas, or subsidize recurrent costs in ways that clearly diminish threats to standing forests as carbon stocks. Third, in the opportunity cost estimation it was assumed that all privately owned forests are potentially available for REDD. Yet as mentioned, Brazilian forest retention standards require 50-80% of private property in the Amazon region to remain under forest. Although few farmers de facto comply with this requirement, REDD in these areas would legally not be additional. Conversely, restricting payments exclusively to legally convertible forests on private properties would dramatically reduce the scope for REDD. Some combination of improved command-andcontrol tools and incentives is probably necessary. Finally, a similar efficiency versus fairness trade-off can apply at the level of distinct states within Brazil. The previous discussion showed that the currently competitive REDD options for the environmentally pro-active Amazonas state would allow the state to reduce deforestation in private lands by 92% for a sum of US$123 million, while for Mato Grosso, which has a history of aggressive agricultural expansion, it would cost nine times as much (~US$1.1 billion) to reduce deforestation by less than half (47%). In other words, if funds were allocated exclusively according to the criteria of additionality, Mato Grosso could receive the bulk of REDD payments and still continue clearing forest with high opportunity costs for its economic development, while Amazonas would have receive less transfers and be almost barred from further land clearing. This disparity results from agricultural market dynamics and the basic economics of deforestation, but also in part because Amazonas state previously declared many more protected areas than Mato Grosso. If the federal government operates the REDD system, the distribution of resources between states should surely be guided largely by additionality concerns, but must also 142 reward ‘good past stewardship’ (e.g. through co-financing for national parks, reserves, etc.). Otherwise, a backlash against these environmentally progressive policies could occur, which would also negatively affect the protection of carbon stocks. Alternatively, states could be allowed to negotiate REDD contracts independently from the federal government. However, such a scenario will be more vulnerable to leakage from states that are successful in capturing carbon rents to states that are not. Discussion How do the presented results compare to other REDD opportunity cost studies? Nepstad et al. (2007) estimated potential productivity of beef and soybean production based on suitability of climate and soil conditions and at spatially more disaggregated scales than ours. Hence their emission abatement cost curve does include very high-cost abatement options at its upper end. Including all, not only private, land plus the use of a 5%, instead of 10%, discount rate and a 30, instead of 10, year time period for cost accounting boosts their estimate of total opportunity costs to over US$200 billion for the whole Brazilian Amazon. Because they include not directly threatened, but potentially suitable, forests, the carbon unit-cost estimates in the Nepstad et al. study are not directly comparable with the values presented here. Nevertheless, the authors share the conclusion that REDD in the Amazon is a highly competitive mitigation option given the prices at which carbon is traded on both voluntary and non-voluntary markets. Swallow et al. (2007) estimated emission abatement cost for sites in the Peruvian Amazon. Their approach is based on cost-benefit analyses of existing land-use systems and observed land-use changes. The study presents values that correspond to this study’s findings for the state of Amazonas, where more than 90% of emission reductions are competitive at current carbon prices. At a 10% discount rate Swallow et al. estimate that the majority of carbon emitting land use changes between 1998 and 2007 could be compensated for at less than US$5/tCO2. This study’s approach to estimating opportunity costs of REDD in Mato Grosso and Amazonas required the following key assumptions: 1. Deforestation on private land is equal to the municipal level deforestation rate. This potentially underestimates true total opportunity costs, because private deforestation rates are expected to be higher than those in protected areas or public land. Preliminary results from the Brazilian Agricultural Census 2006, for instance, suggest that forest on private lands in Mato Grosso between 1995 and 2006 has been reduced at an average annual rate of 5%, i.e. about twice the 2000-2006 rate at the state level. 2. REDD-compatible benefits from the standing forest, e.g. extraction of non-timber forest products, are zero. This assumption leads to a potential overestimation of per ha opportunity costs. For the type of farmers that most contribute to deforestation in the Amazon (i.e. commercial cattle and agricultural producers), it is expected that non-timber forest products play a minor role in resource use decisions. 3. Current municipal land-use distribution and profits are fully replicated on deforested land. The direction of bias introduced by this rigid assumption is ambiguous, and depends on the relative weight of new opportunities (e.g. technological progress, price changes, 143 new crops such as biofuels) versus incremental limitations (e.g. running into soil fertility or producer capital constraints). 4. Forest clearing is primarily motivated by the expected returns to the land uses that follow deforestation. Hence, this method do calculate the opportunity costs of forest conservation fails if deforestation is motivated by speculation (e.g. on obtaining land tenure security). Yet, since the focus of this paper is on private land (implying relatively secure tenure), this assumption generally holds. Deforestation rates on private land, the actual net returns to individual land uses, and the carbon content of forests can all be expected to vary considerably across the Amazon. The upcoming Brazilian agricultural census will provide more solid data for illuminating the first two factors. Other changes in assumptions could also influence the results. Differentiation of returns for cattle-based activities, i.e. ranching versus dairy farming and land-intensive/ modernized versus land-extensive/ rudimentary operations could reveal more land units at the high-cost end. A more detailed assessment of transport costs would likely reduce the opportunity costs for remote land units (of which there are many in the state of Amazonas) and bulky commodities. Conclusions and Policy Perspectives The empirical assessment of likely REDD opportunity costs in the Brazilian states of Amazonas and Mato Grosso, based on Brazil’s official agricultural statistics, clearly supports previous claims that REDD can be a cost-effective way of reducing deforestation in the Brazilian Amazon. This conclusion is valid in the market-remote Amazonas state with its conservationist policies and low deforestation rates, but equally in the agribusiness-oriented Mato Grosso state with its vibrant soy and beef industries and a history of aggressive forest clearing. A partial assessment of approximate transaction costs does not seem to alter this fundamental conclusion: at current carbon prices, paying for protecting forests is a good deal with wide options. Nonetheless, the comparison of the two very different states in the Amazon also shows that (at current carbon prices and demand) zero deforestation is an unrealistic goal to be achieved through REDD: some high-value uses of converted land cannot be “bought out” through REDD. In addition, only a minor share of deforestation happens on lands with private secure tenure, or at the least with effective control over third-party access rights. Direct REDD payments can therefore not fully substitute for improved command-and-control policies in the Amazon region. In fact, REDD could also co-finance this improvement and, at the same time, reduce the costs of enforcement, especially in areas where conservation opportunity costs are low. Hence, direct REDD payments can be a meaningful complementary strategy, providing positive economic incentives, i.e. “carrots” that will help increasing the political acceptability of “stick” policies to effectively reduce deforestation. At current carbon prices, how much deforestation would REDD really reduce, and at what costs? The answer from above was “almost all deforestation in Amazonas (525 094 ha), and half to two thirds in Mato Grosso’s SLAPR areas (554 842 ha), at somewhere between US$330 million and US$1 billion of total costs” – depending on the payment modality (uniform rates vs. differentiated cost-aligned compensations) and whether permanent or transitory CCX carbon 144 prices (the latter implying a 39% price discount) apply. Taking the two states together this corresponds to roughly 360 million tons of reduced carbon emissions over a ten year period. Nevertheless, it has to be kept in mind that only about a quarter of private land in Mato Grosso is licensed under SLAPR. Under the heroic assumption that SLAPR-registered farms are fully costrepresentative of all farms in Mato Grosso, state-wide costs would range somewhere between US$1.2 and US$4 billion – again depending on the assumptions about payment modes and carbon prices. This large variance of estimates points to the importance of designing the payment mechanism in a way that combines cost-effectiveness with equity considerations. What else should decision-makers bear in mind when planning REDD initiatives in the Brazilian Amazon? First, given favourable opportunity costs for REDD, it might be beneficial to separate the carbonsupply for the “deforestation” and “forest degradation” elements. One pathway is to offer payments for reduced-impact logging that minimizes carbon losses. Another option is to adopt a “log-and-protect” strategy of extracting only the most valuable timbers and then setting aside the resulting secondary forests for strict conservation. A full assessment of the cost-effectiveness of REDD, however, needed to account for losses incurred throughout the entire value chain of agricultural production in the Amazon. As a result, governments might decide to tax income from private REDD agreements to make up for losses in productive activity, which would further increase total costs. Second, the above observed difficulty of precisely estimating highly variable opportunity costs in space might be alleviated through the use of more sophisticated economic techniques. This study’s results suggest price differentiation between REDD suppliers can make REDD considerably cheaper (see Senario I and II in table 3). Experiments with inverse auction systems where producers ‘self-reveal’ their costs and preferences have progressed sufficiently to also pilot these techniques in the Amazon, thus validating ex-ante cost estimates and avoiding overor underpaying individual farmers due to aggregation errors. Third, who would pay for REDD on a massive scale, and at what price? Only some markets currently accept REDD carbon. With roughly 47 Mt CO2/yr (available at current CCX prices) from private lands in Amazonas and Mato Grosso being thrown into the world market, the above assumed constant prices on existing voluntary markets might in fact drop significantly, unless there is a simultaneous hike in demand. Fourth, REDD will have local economy impacts that depend on the degree of diversification of local economies and the potential to maintain output and labour demand at reduced rates of expansion into forest land. Socio-economic impact assessment, therefore, needs to be part of feasibility studies. That said, the lion’s share of forest in the Brazilian Amazon is replaced by extensive cattle production, which has shown considerable potential for intensification. Finally, the REDD scenario on which the presented calculations are based would only pay for those private land areas that will be deforested. However, it is illusionary to predict exactly where deforestation is bound to happen. Furthermore, even if this was possible, paying only for 145 threatened areas will relocate part of conversion pressures to areas not covered (leakage). To counteract the inevitable imprecision of spatial predictions and leakage, payment schemes may need to have a broader spatial coverage of all private areas potentially at risk, and/or raise the carbon stocks set aside as ‘insurance reserve’. This will make REDD schemes more expensive than suggested above. Acknowledgements Technical and financial support from the following institutions is gratefully acknowledged: Embrapa Amazonia Oriental (GIS Laboratory), Sustainable Development Secretariat (SDS) of the State of Amazonas, Instituto Centro de Vida (ICV in Mato Grosso), Zentrum für Internationale Migration und Entwicklung (CIM), and the European Union. References Andersen, L. E., Granger C. W. J., E. J. Reis, D. Weinhold, and S. Wunder. 2002. The dynamics of deforestation in the Brazilian Amazon. Cambridge University Press, Cambridge, UK. ADA 2002. Diagnóstico e cenarização macrossocial da Amazônia Legal: Estrutura fundiária na Amazônia Legal – 1950/1995, Belém. Barreto, P., Amaral, P., Vidal, E., and Uhl, C. 1998. Costs and benefits of forest management for timber production in eastern Amazonia. Forest Ecology and Management, 108: 9-26. Börner, J., Mendoza, A., and Vosti, S.A. 2007. Ecosystem services, agriculture, and rural poverty in the Eastern Brazilian Amazon: Interrelationships and policy prescriptions. Ecological Economics, 64: 356-373. Cacho, O., Marshall, G.R., and Milne, M. 2005. Transaction and abatement costs of carbon-sink projects in developing countries. Environment and Development Economics, 10, 597-614. Cattaneo, A. 2002. Balancing agricultural development and deforestation in the Brazilian Amazon. International Food and Policy Research Institute (IFPRI), Washington D.C. Chomitz, K.M. 2006. At Loggerheads? Agricultural expansion, poverty reduction, and environment in the topical forests. World Bank, Washington D.C. Chomitz, K.M. and THOMAS, T.S. 2001. Geographic patterns of land lse and land intensity in the Brazilian Amazon. Development Research Group, The World Bank, Washington D.C. Dutschke, M. and Schlamadinger, B. .2003. Practical issues concerning temporary carbon credits in the CDM. In HWWA - Diskussionspapier Nr. 227. Fearnside, P.M. 2003. Deforestation control in Mato Grosso: A new model for slowing the loss of Brazil's Amazon forest. Ambio, 32. 146 Ferreira, L.V., Venticinque, E., and Almeida, S. 2005. O desmatamento na Amazônia e a importância das áreas protegidas. Estudos Avançados, 19. FNP 2007. AGRIANUAL. FNP Consultoria e Comercio, São Paulo, Brazil. GOVERNMENT OF AMAZONAS 2007. Amazonas Initiative on Climate Change, Forest Conservation, and Sustainable Development. Government of Amazonas, Manaus, Brazil. Grieg-Gran, M. 2006. The cost of avoiding deforestation. Report prepared for the Stern Review of the Economics of Climate Change. International Institute for Environment and Development (IIED), London. Hall, A. 2006. From Fome Zero to Bolsa Família: Social policies and poverty alleviation under Lula. Journal of Latin American Studies, 38, 689-709. Houghton, R.A., Lawrence, K.T., Hackler, J.L., and Brown, S. 2001. The spatial distribution of forest biomass in the Brazilian Amazon: a comparison of estimates. Global Change Biology, 7, 731-746. IBGE (1995/6) Censo Agropecuário Instituto Brasileiro de Geografia e Estatística (IBGE). IPCC (2007). Forth Assessment Report: Summary for policy makers. Intergovernmental Panel on Climate Change. Margulis, S. 2004. Causes of Deforestation of the Brazilian Amazon. World Bank, Washington D.C. May, P.H., Boyd, E., Veiga, F., and Chang, M. 2003. Local sustainable development effects of forest carbon projects in Brazil and Bolivia. IIED Catalogue. Milne, M. 1999. Transaction costs of forest carbon projects. Centre for International Forestry Research (CIFOR) Bogor, Indonesia. Morton, D.C., Defries, R.S., Shimabukuro, Y.E., Anderson, L.O., Arai, E., Del Bon EspiritoSanto, F., Freitas, R., and Morisette, J. 2006. Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. Proceedings of the National Academy of Sciences of the United States of America, 103, 14637-14641. Nepstad, D., Soares-Filho, B.S., Merry, F., Moutinho, P., Rodrigues, H.O., Bowman, M., Schwartzman, S., Almeida, O., and Rivero, S. 2007. The costs and benefits of reducing carbon emissions from deforestation and forest degradation in the Brazilian Amazon. WHRC, IPAM, UFMG, Report presented at the COP13, Bali, Indonesia. Pokorny, B. and Steinbrenner, M. 2005. Collaborative monitoring of production and costs of timber harvest operations in the Brazilian Amazon. Ecology and Society, 10. 147 Saatchi, S.S., Houghton, R.A., Dos Santos Alvala, R.C., Soares, J.V., and Yu, Y. 2007. Distribution of aboveground live biomass in the Amazon basin. Global Change Biology, 13: 816-837. Soares-Filho, B.S., Nepstad, D.C., Curran, L.M., Cerqueira, G.C., Garcia, R.A., Ramos, C.A., Voll, E., Mcdonald, A., Lefebvre, P., and Schlesinger, P. 2006. Modelling conservation in the Amazon basin. Nature, 440, 520-523. Stern, N. 2007. The Economics of Climate Change. The Stern Review Cambridge University Press, New York. Swallow, B., Noordwijk, M.V., Dewi, S., Murdiyarso, D., White, D., Gockowski, J., Hyman, G., Budidarsono, S., Robiglio, V., Meadu, V., Ekadinata, A., Agus, F., Hairiah, K., Mbile, P., Sonwa, D., And Weise, D. 2007. Opportunities for avoided deforestation with sustainable benefits. An Interim Report by the ASB Partnership for the Tropical Forest Margins, Nairobi, Kenya. Toni, F. 2006. Gestão Florestal na Amazônia Brasileira: Avanços e obstáculos em um sistema federalista. CIFOR/CIID/IDRC. VELÁSQUEZ, C., VILLAS BOAS, A., and SCHWARTZMAN, S. 2006. A challenge for integrated environmental management in agricultural frontier territory in western Pará, Brazil. Revista de Administração Pública, 40. Vosti, S.A., Witcover, J., and Carpentier, C.L. 2002. Agricultural intensification by smallholders in the western Amazon: From Deforestation to sustainable land use, Washington D.C. Wünscher, T., Engel, S., and Wunder, S. 2008. Spatial targeting of payments for environmental services: A tool for boosting conservation benefits. Ecological Economics, 65, 822-833. Wunder, S., Albán, M. 2008. Decentralized payments for environmental services: the cases of Pimampiro and PROFAFOR in Ecuador. Ecological Economics, 65, 685-698. 148 Is soil carbon sequestration part of the bundle of ecosystem services provided by conservation agriculture in the Andes? Quintero, M.a , Comerford, N.b, Estrada, R.D.c,,Marin, F.d a Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia. University of Florida, USA c Consultant. Center for Tropical Agriculture (CIAT), Cali, Colombia. d Universidad Nacional de Colombia, Palmira, Colombia b Abstract The positive role that conservation agriculture has on ecosystem services in the Fuquene watershed (in the Andes of Colombia) derives from better soil water retention, sediment retention and water infiltration located in the Andes. However, soil carbon retention could be an additional soil ecosystem service inside the “bundle” of services provided by conservation agriculture (CA) in this watershed. The objective of this research was to compare soil organic carbon (SOC) content and its stability in different soil aggregate sizes under traditional tillage and CA (reduced tillage, green manuring, permanent cover) in a potato-based production system. Soil organic matter (SOM) and SOC contents were not significantly increased with CA, and the stability of carbon was lower in production systems with these practices. This was explained by lower soil moisture with traditional practice, and by the naturally high OM content and deep A horizon (~0.2g/g, 72 cm A horizon) of these soils. Although the stability of carbon was not increased, the CA practices are ensuring the non-disruption of aggregates which has a positive effect on water-related soil characteristics such as porosity, hydraulic conductivity and soil moisture. Media grab In the Andes conservation agriculture practices ensure soil aggregate protection against mechanical breakdown reducing the probability of emitting C to the atmosphere and improving water movement throughout the soil which favors the volume of stream flows Introduction Much of the world’s carbon is held in soils (more than 41%) and another significant part is in the atmosphere, as carbon dioxide (20%). However, soil disturbance for crop production is reducing soil carbon and augmenting the atmospheric carbon pool. Golchin et al (1995) classifies light fraction soil organic carbon (SOC) into free particulate organic carbon and occluded organic carbon. The SOC light fraction has been found to be a sensitive indicator of managementinduced effects on SOC (Bremer et al. 1994). SOC accumulation to some degree depends on the amount of soil disturbance. Disrupting macroaggregates exposes the microaggregate carbon pool to decomposition (Bajracharya et al., 1997). Management systems involving high C inputs and reduced tillage should favor C storage 149 directly by reducing aggregate breakdown and by enhancing SOM-mediated aggregation (Angers, 1992; Carter, 1992 and Beare et al. 1994). Conservation agriculture is one of these management systems practiced in the watershed of the Fúquene Lake which is located in the valleys of Ubaté and Chiquinquirá, north of Bogotá, the capital of Colombia. These practices were introduced as a measure to control the sediments that are released from potato farms on very steep slopes and that are causing the eutrophication of the lake. This lake provides potable water to more than half a million people downstream. Although the benefits to reduce sediments and to increase net income of farmers are recognized (Rubiano et al. 2006) there are not studies about the impact of these practices in soil carbon protection. In consequence, the objective of this research was to determine carbon content and its stability in stable soil aggregates in two different systems (traditional tillage vs. conservation agriculture CA). To achieve this, the protected carbon in soil micro and macro aggregates was measured using sonication techniques. The hypotheses were that with CA: 1) soil organic matter content is increased, 2) The stability of carbon contained in aggregates is greater, and 3) the SOM (and the SOC) is more stable in smaller size fractions of aggregates. Methods Two potato production systems (traditional tillage, CA with minimum tillage with incorporation of green manures and permanent plant cover for 7 years) were compared at six sites (3 sites per system) within the Fuquene watershed. The soils are Andisols and are classified as Lithic Hapludands (IGAC, 2000). The sites were selected with the same characteristics in terms of: 1) landscape position; 2) land cover; 3) slope; and 4) rainfall intensity. At each site, two pits were dug, soil horizons were identified, and three soil samples were taken per soil horizon. Fresh samples were segregated and classified by size using dry-sieving with a nest of sieves with 5, 2, 1, 0.5 and <0.5 mm screen size. Additional samples were taken to measure saturated hydraulic conductivity, soil moisture, porosity and bulk density. In general, three horizons were found it in the profiles with average thicknesses of 72 cm (horizon I, top), 21 cm (horizon II) and 56 cm (horizon III, bottom). Soil organic matter content of each aggregate size class was extracted using a sonication procedure (North, 1976; Six et al., 2001). Through this procedure, some of the SOM in the aggregates was extracted while the rest of the SOC remained in the aggregates even after sonication. The organic matter extracted from the aggregate by sonication was called AOM (aggregate OM) as it contained fine organic matter from inside aggregates. Organic matter remaining in the same particle size class after sonication was termed particulate organic matter (POM). Different levels of energy were applied to see how AOM and POM is affected by the degree of disruption. The AOM and POM were measured through the loss on ignition procedure, and %AOM was calculated as percentage of total SOM (AOM+POM). All SOM measurements were converted to estimates of SOC concentration by multiplying by the Van Bemmelen factor of 0.58 (Lal et al, 1998). It was expressed as percent of total organic matter in each sample. 150 Data analyses The effects of production system on %AOM and SOC were analyzed using analysis of variance (ANOVA) with soil horizon, type of management practice and aggregate size fraction as fixed effects. Differences were considered significant at p<0.05. There were significant effects of horizon and size fraction on %AOM and SOC. Therefore further statistical analysis was done separately for each size fraction and horizon. A post hoc comparison procedure with the TukeyCramer adjustment was used. The SOM and AOM values were correlated to physical soil characteristics using a linear model. Results Soil organic carbon concentration and tillage systems The average concentration of SOC in the soil profile was not significantly affected by the management system, and averaged 0.12 g/g with CA and 0.09 g/g with traditional tillage. SOC was significantly higher in the top horizon (0.13 g/g) than in the deepest horizon (0.05 g/g). Aggregate stability and SOM content As the ultrasonic energy applied to the soil increased, more aggregates were destroyed, increasing the amount of AOM removed (Table 1). The effect of size fractions on % AOM was highly significant (p<0.01). The 5 mm fraction released significantly less organic matter than the smallest size fraction (0.5 mm). Table 1. Average values of AOM for different aggregate size fractions. Aggregated Organic Matter (AOM) (%) Size fraction Average Tukey group (mm) 0.5 1.0 2.0 5.0 99.354 95.301 87.067 83.248 a ab ab b Note: Inside the same column, averages with the same letter are not statistically different There was no significant size fraction x tillage system interaction for the AOM. However, %AOM was higher in the 2 and 1 mm size fractions under CA than with traditional tillage (figure 1). Also, %AOM was higher in the horizon II and III with CA than with traditional tillage, but with similar values in horizon I (figure 2) 151 Conservation Agriculture Traditional Agriculture Conservation Agriculture Traditional Agriculture 100 120 98 96 Aggregated Organic Matter (%) Aggregated Organic Matter (%) 100 80 60 40 20 94 92 90 88 86 84 82 80 I 0 0 1 2 3 4 5 6 Size fraction (mm) II III Horizon Figure 1. Effect of size fraction on %AOM Figure 2. Effect of horizon on %AOM variable. AOM, SOM and physical soil characteristics Simple correlation analyses using a linear model showed that, for CA, aggregate organic matter (g/g) was negatively correlated with the aggregate size (5 mm) (p < 0.01; figure 1) and positively correlated with soil moisture (p < 0.05; data not shown). Also, there was a positive correlation of total organic mater (g/g) with hydraulic conductivity, total porosity and macro-porosity in both conservation and traditional agriculture treatments (data not shown). The soil moisture and hydraulic conductivity were higher in CA. The average soil moisture for the soil profile in CA was 52% while in traditional agriculture was 39%. The average hydraulic conductivity was 12 cmh-1in CA soil profiles and 5 cm h-1in traditional agriculture. The total organic matter was negatively correlated with bulk density in conservation and traditional agriculture systems. In general, bulk density had a negative correlation with saturated hydraulic conductivity, total porosity, macro porosity and soil moisture. Discussion SOM and SOC contents The soils of the study sites are Andisols and according to IGAC (2000) are classified as Lithic Hapludands. The high concentration of organic carbon is in line with the high organic matter concentration characteristic of Andosols. The accumulation of organic matter in these soils is determined by the environmental conditions of the paramo ecosystem, which is characterized by low temperatures and high plant biomass (pastures) inputs and low decomposition rates. The lack of significant differences between SOC concentration in CA vs. traditional agriculture sites may reflect the difficulty of detecting small changes in SOM against such a high background level, even after 7 years of CA in these soils. The high accumulation of OM in these soils is reflected in the large depth (mean 72 cm) of horizon I. Also, the capacity of these high organic matter soils to store further C may be near maximum. Finally, the lack of significant differences between the production systems may be due to the fact that in the traditional tillage system, potato is rotated every 2-3 yrs with 2-3 yrs of pastures (average of 2.7t DM/ha/yr). Therefore the benefits of CA could be more related to nutritional benefits, reduction of runoff, and improvement of water movement in the soil profile rather than change in SOC per se (see below). 152 Aggregation and SOM stability The results showed no effect of production system on aggregate stability of organic carbon stability in the aggregates, rejecting the second hypothesis. In fact, the aggregates from CA in horizons II and III, and from the 2 and 1 mm size fractions, released more organic matter than the equivalents from the traditional system. The higher stability of traditional agriculture soil aggregates may result from greater drying of the soil than under CA. Soil moisture was higher in CA sites, especially in horizons II (63 vs. 44%) and III (48 vs. 32%). In Andisols, the presence of minerals such as ferrihydrate and allophone results in irreversible hardening when the soil is dried beyond a certain level (Maeda et al.‚ 1977).; the drier the soil, the stronger the aggregates. Other authors have also found that the strength of the bonds between organic materials and mineral particles decreases with increasing water content, resulting in lower stability (Reid and Goss, 1982; Perfect et al., 1990, Gollany et al. 1991; Caron and Kay, 1992 cited in Lal et al. 1992). However in the CA sites the risk of releasing that SOC contained in aggregates is low because of the use of minimum tillage. While it is recognized that micro-aggregates protect SOC (Six et al, 2000), we found that in these soils the trend was counter to the third hypothesis, as OM in aggregates of 5mm size fraction was more stable than in the smaller fractions. Similar results were obtained in Spodosols of Florida where the highest strength was obtained in macroaggregates (Sarkhot et al, 2005). This trend is less apparent in horizon III, which could be related to the fact that the content of clay is higher in deeper horizons, making the OM more strongly attached to the microaggregates (the average percentage of clay content in horizon III was 39%, in horizon II 32% and 19% in horizon I). The percentage of OM released after applying higher amounts of energy (11.7-15.4 kJ) was high on most soil samples (>80% of total organic matter), and only 17.4% of soil samples released <80% of the total organic matter of the samples. This means that most of the organic carbon is in the aggregate pool and the remaining is POM. This result is in line with other studies that have found that 90% of SOM was located within soil aggregates (Jastrow et al., 1996). This highlights the importance of conserving the aggregation of these soils and reducing its mechanical breakdown by tillage or soil erosion. It is worth noting that this sonication procedure measures the stability of aggregates to mechanical breakdown and does not indicate the susceptibility to microbial breakdown. Conclusions and recommendations In Andisols of this Colombian watershed, CA practices had a negligible effect on SOM and SOC concentration. This may be due to the already high SOM content in these soils. In these soils, the benefits of CA (minimum tillage and permanent cover) are related to improving soil characteristics important for increasing infiltration and storage and reducing runoff and erosion, such as hydraulic conductivity, porosity and bulk density. Probably, it also contributes to increase nutrients availability and to reduce soil runoff. In addition, CA practices did not increase SOM stability in aggregates, which may be related to the higher soil moisture in the CA system. It is important to note that CA ensures that the accumulated OM is not released from aggregates as soil disturbance in minimal in this system. 153 Since more than 80% of the total organic matter is contained in the aggregates it is important to avoid the disruption of aggregates by mechanical forces in these paramo soils. Also the importance of macro aggregates for SOC stability is important in these soils, even more than micro aggregate stability, contrary to our initial expectations. Finally, there is a need to apply the same methodology to explore the effects of CA on SOM and SOC content and its stability in other type of soils with lower C OM and with different clay contents. Acknowledgements This paper presents findings from PN22 ‘Payment for Environmental Services (PES) as a mechanism for promoting rural development in the upper watersheds of the tropics’, a project of the CGIAR Challenge Program on Water and Food. Also this research is part of a Master thesis of the University of Florida. Special thanks to Arnulfo Rodriguez from the soil physics laboratory of CIAT who helped during the field work and lab analyses. Literature cited Angers, D.A. 1992. Changes in soil aggregation and organic carbon under corn and alfalfa. Soil Sci. Soc. Am. J. 56:1244-1249. Bajracharya, R.M., Lal, R. and Kimble, J.M. 1997. Soil Organic Carbon Distribution in Aggregates and Primary Particle Fractions as Influenced by Erosion Phases and Landscape Position. In Soil processes and the carbon cycle, ed. Lal, R., Kimble, J.M., Follet, R.F. and Stewart, B.A. CRC Press. Boca Raton, USA. 353-367 Beare, M.H., Hendrix, P.F. and Coleman, D.C. 1994. Water-stable aggregates and organic matter fractions in conventional and no-tillage soils. Soil Sci. Soc. Am. J. 58:777-786. Bremer, E., H.H. Janzen and A.M. Johnston. 1994. Sensitivity of total, light fraction and mineralizable organic matter to management practices in a Lethbridge soil. Can. J. Soil Sci. 74:131-138 Carter, M.R. 1992. Influence of reduced tillage systems on organic matter, microbial biomass, macroaggregate distribution and structural stability of the surface soil in a humid climate. Soil Tillage Res. 23:361-372 Golchin, A., P. Clarke, J.M. Oades, and J.O. Skjemstad. 1995. The effects of cultivation on the composition of organic matter and structural stability of soils. Aust. J. Soil Res. 33: 975993. Instituto Geografico Agustin Codazzi (IGAC). 2000. Estudio general de suelos y zonificación de tierras del departamento de Cundinamarca. Bogotá, D.C. Colombia. 154 Jastrow, J.D., T.W. Boutton, and R.M. Miller. 1996. Carbon Dynamics of aggregate-associated organic matter estimated by 13C-natural abundance. Soil Sci. Soc. of Am. J. 60:801-807. Lal, R., J.M. Kimble, R.F. Follet, and B.A. Stewart. 1998. Soil processes and the carbon cycle. CRC Press LLC. Boca Raton, Florida. USA. Maeda, T., Takenada, H. y Warkentin, B.P., 1977. Physical properties of allophane soils. Advances in Agronomy. 29:229-261. North, P.F. 1976. Towards an absolute measurement of soil structural stability using ultrasound. J. Soil Sci. 27:451-459. Rubiano, J., Quintero, M., Estrada, R.D., Moreno, A., 2006. Multiscale analysis for promoting integrated watershed management. Water International 31(3), (in press) Sarkhot, D.V., N. B. Comerford, E.J. Jokela, W.G. Harris, and J. Reeves. 2005. Soil Carbon in a Forested Sandy Coastal Plain Spodosol: Methods considerations and initial results. PhD. Diss. University of Florida. Six, J., Paustian, K., Elliott, E.T., Combrink, C., 2000. Soil structure and organic matter: I. Distribution of aggregate-size classes and aggregate-associated carbon. Soil Sci. Soc. of Am. J. 64, 681-689. Six, J., G. Guggenberger, K. Paustian, L. Haumaier, E.T. Elliott, and W. Zech. 2001. Sources and composition of soil organic matter fractions between and within soil aggregates. European J. Soil Sci. 52:607-618. 155 Ex ante Analysis of Legumes: The Dilemma of Using Legumes as Forage for Animal Nutrition during the Dry Season or as Green Manure for Soil Improvement Quintero, M.a, Holmann, F.a, Estrada, R. D.b a b Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia. Consultant. Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia. Introduction Soil nutrient depletion is a common problem faced by both subsistence farming and commercial crop production in developing countries and can be attributed to the nutrient uptake by agricultural crops, which is higher than the amount of nutrients available in the soil. This is also a major cause of soil degradation (Frossard et al., 2006). Research carried out over the past decades has clearly evidenced a direct relationship between soil degradation, food insecurity, and poverty (Lipper, 2001). The most important animal production system in developing countries is the mixed livestock production system (von Kaufmann, 1999) and can be found in Nicaragua as well as other Central American countries, where most farms are small, located in hillside areas undergoing different stages of degradation, and combine livestock production with the planting of subsistence crops such as maize and beans (INTA, 2002). Natural pastures are the most important source of feed for livestock but their quality and quantity are seriously limited during the dry season, which lasts from 4 to 6 months, causing shortage of forage and animal undernutrition (PASOLAC, 2002). Furthermore, because of the problem of grass shortage, producers allow cattle to freely graze the dry vegetation, which makes the problem of overgrazing—another major source of soil degradation—even worse (FAO, 2000). On the other hand, milk production significantly decreases during the dry season and, as a result, milk prices increase by 40%-50% as compared with its prices during the rainy season. Improved animal nutrition during the dry season would therefore significantly improve family incomes in these mixed production systems. In the past, several alternatives have been used to correct forage shortage or deficiencies during the dry season. These have included the use of net energy sources, ranging from forage cane to legumes, the latter contributing protein and complementing energy sources and available grasses. However, the competitiveness of using legumes for animal nutrition versus their use to improve soil quality and, as a result, crop productivity has seldom been analyzed. This study therefore assesses the economic benefits of (1) a short-term alternative, which consists of establishing legumes for use as supplement, mixed with crop residues, to increase milk production and farmer incomes during the dry season when milk prices are higher; and (2) a medium-term alternative, which consists of establishing legumes as green manure at the same sites where maize and beans are planted and then incorporate these legumes into the soil to improve its fertility and, accordingly, improve agricultural productivity in subsequent years. 156 Objective This study aims to (1) perform an ex ante analysis of the expected economic and environmental benefits of using the legume Canavalia brasiliensis either as green manure to improve agricultural productivity or as forage to improve milk production during the dry season; and (2) compare these benefits with the subjective perception of producers living in hillside areas of Nicaragua that have mixed maize-beans-livestock production systems regarding these new alternatives. Current Status of Research on Canavalia Use as green manure The effect of nitrogen fertilization on subsequent crops is greatest when legumes are used as green manure. However, the N available due to decomposition of crop residues may be released before the roots of the new crop are established and can properly tap this source. The N can therefore be lost due to volatilization, denitrification, or leaching (Millar et al., 2004). When Canavalia was established at the end of the rainy season for subsequent growth during the dry season and then incorporated into the soil, the increase in marginal productivity of the following maize harvest corresponded to an application of 50 kg N/ha (Burle et al., 1999). Although this suggests that Canavalia residues supply an important amount of N, the amount of N symbiotically fixed has not yet been determined. Use as supplement for animal nutrition The biomass of maize and bean stubble is the most important forage reserve for animal nutrition during the dry season. Although the available dry matter (DM) of these stubbles is relatively high, its low protein content (~ 4%) and digestibility (~ 40%) reduce animal productivity significantly, leading to both lower milk productivity and animal weight loss as compared with the rainy season. The nutritional value of maize and bean stubble can be improved significantly by introducing legumes such as Canavalia (Said y Tolera, 1993). The advantage of Canavalia is that it is very tolerant to drought. Preliminary experiments show that Canavalia is well accepted by goats and sheep in Nicaragua. Recent results show a raw protein content from 20% to 25% and a digestibility of 80% (CIAT, 2006). Materials and Methods Collection of primary data Data came from a survey of 10 producers of the Pire river watershed, located in the Department of Estelí in northern Nicaragua. The survey, conducted in September 2007, aimed to collect information on land use, animal inventory, use of inputs, and use of family and contracted labor to estimate animal and crop production costs (i.e., maize and beans), productivity, and income from the sale of milk, meat, maize, and beans. 157 The survey also gathered information on how producers perceive the use of the legume C. brasiliensis and what their expectations are to justify the adoption of Canavalia, based on the following: a) the minimum amount of milk that should be produced in excess of the average dry-season production for producers to adopt Canavalia as animal supplement; or b) the amount of fertilizer (i.e., urea) that producers considered that could be saved, while maintaining the same maize and bean production, to adopt Canavalia as green manure. Ex ante economic evaluation Based on average survey results, an ex ante economic evaluation was made of the economic benefits that would be produced if this legume was cultivated as green manure or used as animal supplement. The ECOSAUT model was used (Quintero et al. 2006). This optimization model uses linear programming, to evaluate land uses under multiple criteria—social, economic, and environmental. These decision-making criteria or variables are defined according to the production system (land use) evaluated and the evaluation objective. The agroecosystem is accordingly simulated to better understand the effects that the incorporation of C. brasiliensis will have on producers’ income and if the expectations producers expressed during the field visit are fulfilled. To conduct this evaluation, the following scenarios were analyzed over a 5-year period: Scenario 1. Baseline This is the current land use scenario of the farms included in the survey. For this study, the baseline is defined as a farm type showing the average values of production costs, income, and productivity obtained in the survey. The land use system is mixed—maize and beans are grown and both milk and meat are produced. The farm area is 12 ha, of which 10 ha are sown to Jaragua grass (Hypharrenia rufa) and 2 ha are planted to maize and beans. The Jaragua grass is not fertilized and its biomass production decreases during the dry season, from 1.6 to 0.6 t DM/ha. Milk production also decreases during these months. Maize is planted first, at the onset of the rains (June). Once the maize has formed ears, the plants are folded for drying and beans are grown in half of the area (1.0 ha), using these dry stalks as support. Beans are planted at the end of the rainy season, around September-October, and are harvested at the beginning of the dry season (December-January). Scenario 2. Canavalia for animal nutrition This scenario also corresponds to a combined crop/livestock production system, but C. brasiliensis is also grown, intercropped with maize in the area where beans are not planted (1.0 ha). In this case, the legume is used for livestock nutrition during the dry season to increase onfarm milk production. This evaluation assumed an annual production of C. brasiliensis of 2 t DM/ha. The same distribution of land in pastures and grasses as found in the baseline is maintained. 158 Scenario 3. Canavalia for soil improvement This scenario corresponds to the same scheme described in Scenario 2 above, with the difference that the legume is incorporated into the soil to improve fertility and, as a result, improve the productivity of subsequent plantings of maize and beans. This is why the legume is incorporated into the soil as green manure. It is assumed that the incorporation of Canavalia contributes 64 kg N/ha and replaces the traditional application of N in the form of urea (52 kg/ha) in maize and bean crops. It is only necessary to continue applying the complete fertilizer (12-30-12 NPK) at 82 kg/ha. Scenario 4. Canavalia for animal feeding with sorghum This scenario was developed because many producers (especially those with more livestock) plant sorghum at the end of the rainy season in order to have sufficient biomass to feed livestock during the dry season, in addition to maize stubble. The main objective is to produce biomass as source of forage for livestock. As a result, producers use a high planting density to maximize forage production and not grain production. Scenario 5. Canavalia in rotation with maize to improve soils throughout the farm This scenario explores the maximum potential of the farm in terms of generating income by gradually substituting the area (2 ha/yr) currently under Jaragua grass with a rotation of maize and Canavalia over a 5-year period. The purpose of this scenario is to explore the contribution of C. brasiliensis as mechanism to improve soil fertility and make the system more sustainable by subsequently introducing improved pastures, such as Brachiaria brizantha cv. Toledo, as well as an energy source, for example sugarcane. Ex ante environmental evaluation The environmental ex ante evaluation was focused on the effects that the incorporation of Canavalia brasiliensis into the crop rotation might have on environmental externalities such us sediment and water yields. This analysis was conducted applying SWAT (Soil and Water Assessment Tool) for an area with biophysical conditions similar to those found in the visited farms. These conditions refer to soil, climatic and topographic characteristics and that were collected for the study area. The value of soil characteristics considered in this analysis were obtained from the analysis of local soil samples conducted by the soil research component of this project (personal communication with S. Douxchamps). It includes information of texture and total C for the superficial soil horizon. In addition some information about soil type units was extracted from the Land Use Plan of Estelí (Plan de Ordenamiento Territorial in Spanish) (MARENA, 2001) and use to complement the information on texture and organic matter for subsurface soil horizons. Using the soil texture information, the hydraulic conductivity, available water content and bulk density values were derived using the Soil Characteristic Tool (Saxton and Rawls, 1985; Saxton et al. 1986) that is applicable to mineral soils. In table 1, the values used in the SWAT modeling are shown. 159 Table 1. Soil characteristics used for the SWAT modeling Horizon Depth (cmm) Bulk Density (g/cm3) Saturated Available Hydraulic Water Content Conductivity (mm/hr) (cm/cm) % C % Clay % Silt % Sand A 0-20 1.13 0.15 22.44 23.4 28 32 40 B 20-70 1.32 0.1 1.2 6 54 18 28 The climatic data used consisted on daily values of precipitation, maximum and minimum temperature; and mean monthly temperature, radiation and wind velocity. The data sets for the period of January 1987 - December 2006 were obtained at INETER (Instituto Nicaraguense de Estudios Territoriales) The topographic data was directly obtained from the Digital Elevation Model of the River Pire watershed at a resolution of 90 m. To do this an area of 154 ha was selected near the farms where experiments on Canavalia brasiliensis are being held, and which GPS points where captured during the field visit on 2007. The climatic, soil and topographic data was integrated in SWAT to derive the values of sediment and water yields, surface runoff, lateral flow, percolation, evapotranspiration, and soil water for the following land use scenarios: 1) current maize-pasture system, 2) maize rotated with Canavalia brasiliensis which residues are left on the soil surface as green manure, 3) maize rotated with Canavalia brasiliensis that is grazed after 90 days of growth. In figure 1, the schedule of planting for each scenario is shown. It is worth to note that these scenarios were assessed for the portion of land that is only planted with maize and not followed by other crop such as beans (see above description of scenarios 1-3). Land use scenario Tradditional maize rotation Maize rotated with C.brasiliensis as green manure Maize rotated with C.brasiliensis as forage Jan Feb Mar Apr May Jun Jul Ago Sep Oct Nov Dec Fallow Maize Fallow Fallow with residues of C.brasiliensis Maize C.brasiliensis Fallow Maize C.brasiliensis-grazing Results Tables 2-5 present the average production costs of maize, beans, milk, and meat as well as average values of productivity, farm area distribution in different land uses, use of family and contracted labor, and herd composition. Table 6 presents the producers’ expectations regarding the reduced requirement of fertilizers or the increase in milk production expected with the inclusion of C. brasilensis as green manure (in the former case) or animal supplement (in the latter). Table 7 presents the production costs and expected productivity of this legume. ECOSAUT Model results Table 8 shows the values for each scenario used for the ex ante evaluation of potential economic benefits derived from the incorporation of C. brasiliensis into the land use system of producers of the Pire river watershed as well as from the incorporation of other potential energy and protein 160 sources that could help overcome the shortage of feed for livestock during the dry season and recover soil fertility. Benefits of C. brasiliensis under the current land distribution scheme (Scenario 1 versus Scenarios 2 and 3) Based on the results obtained, the incorporation of C. brasiliensis as green manure (Scenario 3) slightly decreased the net income as compared with the baseline (5%). The opposite occurred when this legume was used as animal feed (Scenario 2) because the net income of producers was increased by 5% (Table 8). The urea applied in the baseline scenario is replaced in Scenario 3 with the incorporation of the legume into the soil. The reduction in net income obtained by using Canavalia as green manure can be attributed to the fact that, although the incorporation of the legume reduces the cost incurred for purchasing fertilizers, the requirement of contracted labor to plant the legume increases and the purchase of legume seed implies an additional cost. As a result, the benefit represented in reduced fertilizer costs does not compensate for the additional cost of planting the legume. On the other hand, the increased income due to the incorporation of Canavalia for animal nutrition can be attributed to the increase in milk production, specifically during the dry season. Milk production during the dry season increased from 2 to 3 lt/day, representing a 26% increase in annual production as compared with the baseline. In addition, the increase in income is not only due to a greater volume of milk produced during the dry season, but also the higher price of milk during this time of scarcity (US$ 0.27/lt during rainy season compared with 0.32/lt during the dry season). Therefore the benefits of using Canavalia as animal feed are related to the increases in milk production and not to increases in stocking rate or meat production, which are maintained. Benefits of using C. brasiliensis as animal feed when complemented with sorghum as energy source (Scenario 2 versus Scenario 4) The positive effect of complementing Canavalia with an additional energy source such as sorghum is reflected in the 80% increase in producers’ net income as compared with Scenario 2 where only Canavalia is incorporated as additional source of feed for livestock. This increase can be attributed to the fact that milk production increased substantially by 137% due to the merging of three factors: (1) increased production potential from 3.7 to 4.4 lt/day; (2) doubling of animal stocking rate from 7 to 14 cows/farm; and (3) increased sale of milk during the dry season, which took advantage of the better prices that are characteristic of that season. This increase in milk production can be explained by the incorporation of an additional energy source into the system, which allows the additional protein resulting from the incorporation of Canavalia during the dry season to be used more efficiently. In other words, sorghum helps balance the additional protein provided by the legume. Benefits of incorporating Canavalia in the rest of the farm (Scenario 5) This scenario aims to estimate the benefits of gradually replacing those areas sown with Jaragua grass with an improved pasture, for example, B. brizantha. A maize/Canavalia rotation is used to 161 gradually improve the soil in all areas sown to native pastures undergoing degradation. In the case of improved pastures, energy usually becomes a limiting factor so it is necessary to plant sugarcane to improve the nutritional balance for the better-quality pasture. These two factors make it possible to increase the number of cows (although not their production potential) in comparison with all other evaluated scenarios. With the renewal of the pasture and the incorporation of sugarcane, a stocking rate of 30 cows/farm can be used, with a milk production potential of 3.5 lt/day per cow. Compared with the baseline, this increases milk production by 17% and stocking rate 4.3 times. Net income increases 2.8 times as compared with the baseline. The increase in income can also be attributed to the fact that the income derived from maize production increases because the areas that are replaced with B. brizantha were previously planted to maize/Canavalia. In other words, the maize grown in the area sown to pastures is additional to the area normally planted to this crop on the farm. Results from the SWAT modeling The results from SWAT modeling show that the incorporation of Canavalia brasiliensis reduces both, the sediment and the water yield by 32 and 10%, respectively. This is related with an important reduction on the surface runoff by 35%. The reduction of the surface runoff is related to improvements on water percolation and water lateral flow and the increment of the evapotranspiration (Table 2 and 3). However there were not obtained differences between using Canavalia brasiliensis as green manure or as forage. These two options have the same effects in terms of water and sediment yields as well as on the other water balance variables (runoff, lateral flow, soil water, percolation and evapotranspiration) (Table 2). Table 2. Results of SWAT modeling for different maize-based rotations Land use scenario Tradditional maize rotation Maize rotated with C.brasiliensis as green manure Maize rotated with C.brasiliensis as forage Evapotranspiratio n (mm) Surface runoff (mm) Lateral Flow (mm) Percolation (mm) Sediment yield (t/ha) Water yield (mm) Soil water (mm) --- --- --- --- --- --- --- 4.62% -35.21% 5.25% 3.63% -31.91% 10.64% 3.11 % 4.61% -35.20% 5.25% 3.65% -32.08% 10.62% 3.11 % 162 Table 3. Water balance for a 20-yr period: Traditional maize-pasture rotation vs. C.brasiliensisbased rotations Tradditional maize rotation Maize rotated with C.brasiliensis as green manure or forage Difference Surface runoff (mm) 1925.897 1247.801 -678.096 Lateral Flow (mm) 433.047 455.79 22.743 Groundwater (mm) 2316.36 2474.024 157.664 Water yield (mm) 4675.304 4177.615 -497.689 Evapotranspiration (mm) 12398 12970.85 572.85 Although the total annual water yield is reduced with the C.brasiliensis-based scenarios, there are increments on it during the dry months (figure 2). In figure 3, the difference on monthly water yields between traditional maize rotation and C.brasiliensis rotation is shown. 250 200 mm 150 100 50 0 Jan Feb Mar Apr May Jun Montanuela station Jul Aug Sep Oct Nov Dec Condega station Figure 2. Average monthly precipitation (mm) for two stations near the study area 163 5 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec mm -5 -10 -15 -20 -25 Figure 3. Annual average difference on water yield (mm) from changing traditional maizepasture rotation to maize-legume rotation It is worth noting that the effect of Canavalia brasiliensis varies throughout the years as the rainfall varies yearly. The precipitation datasets showed that there is a great variation on annual rainfall (figure 4). The lowest rainfall was registered on 1992 with 493 mm/yr and the highest on 1998 with 1384 mm/yr. During the wettest year the sediment yield for the traditional maize rotation is 70 /ha/yr and for the C.brasiliensis-based scenarios is 57t/ha/yr. In the driest year it was 9.6 t/ha/yr and 4.5 t/ha/yr for the traditional and legume-based scenarios, respectively. 1600 1400 Precipitation (mm) 1200 1000 800 600 400 200 19 87 19 88 19 89 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 0 year Figure 4. Annual rainfall 1983-2005 164 Discussion Is it feasible to achieve this proposed redistribution of land uses in these scenarios? One prerequisite is stability in the prices of maize, beans, and milk, which helps producers perceive greater economic security to incur in the initial investment that this type of change requires. The aversion of producers to assume the risk implicit in the increase of area planted to crops or the introduction of new crops and pastures was confirmed during the field visit, when producers said that price instability was the principal limitation to increasing the area planted to crops. This was corroborated by the fact that the area on farms dedicated to crops is very similar, regardless of the variations in total farm size. For example, a 17-ha farm and an 8-ha farm will always have 2 ha planted to crops. In farms under 5 ha, the area planted to crops is only slightly less. Another factor that could limit the feasibility of incorporating the proposed changes in these scenarios is the local availability of labor. Contracted labor would necessarily increase from 90 to 384 man-days as compared with the baseline, or the family labor dedicated to agriculture activities would increase by more than 100%. Another factor that could currently be hindering the expansion of the agricultural area and the purchase of livestock are the high interest rates reported by producers. These rates range between 10% and 26% in real terms. As a result, the system never generates sufficient surplus to support higher investments in the future. The effective term is substantially reduced, which especially affects long-term investments in livestock. By expanding the area planted to crops, the stocking rate could be increased on several farms of the Pire river watershed and, as a result, the inventory of cattle available in the area could dwindle. This, in turn, would have an impact on the prices of livestock and would not only curtail the feasibility of Scenarios 2, 4, and 5 but also affect foreseen economic benefits. In this sense, it would be better to improve the birth rate of livestock using better-quality forage, for example C. brasiliensis, and improved pastures, such as B. brizantha. Environmental Impacts The results from SWAT modeling permit to quantify the environmental effects that the incorporation of Canavalia brasiliensis would have on environmental externalities that are important to society such as sediment and water yields. It is clear that the main benefit of incorporating this legume to the current land use system is that the sediment yields could be reduced. This effect is related with the fact that the legume provides some cover to the soil during the wettest months (September and October) (Figure 2). However this is not the case for water yield20. According to the results, the water yield is reduced when C.brasiliensis is either planted as forage or as green manure. This reduction is directly related with a reduction on the surface runoff that is too high and is not compensated by the improvements on lateral flow and groundwater. The reduction on runoff might be explained by the improvement on land cover provided by the legume and by an important increment on the 20 Water yield (mm H2O). Total amount of water leaving the land and entering main drainage. Water yield= Surface runoff + Lateral flow + Groundwater + Transmission losses. Transmission losses are minimal in this case. 165 evapotranspiration (table 3) with respect to the traditional maize rotation were the only crop is maize. Although this total water yield reduction, the C.brasilliensis scenarios increase it during the drier months (November to March) where water yield is most important as an externality. The benefits of modifying this externality should be valuated to determine if that increment at the watershed level could be significant if the legume is introduced in several farms. In another hand, the lack of differences between using the Canavalia brasiliensis as a green manure or as forage has to be done with the rainfall behavior. It was simulated that both, the cut of green manure and its posterior deposition on soil surface or the grazing of the legume, occurs after December when the crop biomass is high enough for these purposes. Due to on December the rainfall is minimal; the marginal impact of having a cover crop is insignificant because the soil is not exposed to the impact of rain drops. It was expected that the soil water content might change with the incorporation of the legume as green manure. However there were not changes with respect to the scenario where the legume is grazed. Probably the contribution as a green manure is not big enough to counteract the effect of high temperatures and water deficit. The ongoing field measurements of soil moisture and organic matter in the experimental plots will permit to confirm or reject this ex-ante results. An eventual rejection will provide insights for calibrating the model for future ex-ante analysis. In consequence, from the farmer perspective, the environmental benefits of incorporating the legume to its current land use system could be only the reduction of soil loss during the rainy season since the effects on soil water appear to be not greater (only an increment of 3%). However these predicted impacts need to be verified during the implementation of the C.brasiliensis-based scenarios in the selected farms. Besides the farm-level environmental impacts the effects of the legume on subsequent maize harvests need to be measure in the field since the incorporation of OM, N and the possible increment on soil water could increment the maize yields. Apart from the farm-level effects, the aggregated effect of having several farms under the C.brasiliensis-based scenarios in the watershed could be greater and significant in terms of soil loss reduction and water yields. For this purpose it is still indispensable to obtain soil data for all existing soil types in the watershed and river flow measurements in order to run and calibrate SWAT at this scale. This step will be crucial to establish the trade off between reducing sediment yields vs. water yields. In case of confirming the potential reduction on total water yields after the incorporation of C.brasiliensis to the production systems, it will be necessary to compare the total benefits of introducing the legume to the system (economic farmer benefits derived from improvements in dairy or maize productivity + society benefits derived from sediment retention) with the cost for the society derived from total water yield reduction. Also it would be necessary to analyze the cost of reducing total water yield and compare it with the value of benefits derived from the increments of water yields during the dry season. 166 Conclusions Producers’ expectations regarding the benefits of C. brasiliensis and its adoption potential In the survey producers expressed that they would be willing to adopt C. brasiliensis as green manure if the use of fertilizers was reduced in 112 kg urea/year (i.e., 51 kg N/ha) and 112 kg NPK/year (i.e., 12-30-12). Taking into account that legume productivity in this ex ante evaluation was considered to be 2 t DM/ha per vegetative cycle and that this legume presents 20% protein, producers’ expectations would be satisfied because this represents 64 kg N/ha (without counting the N fixed through Rhizobium. Regarding the adoption potential of C. brasiliensis as animal feed, producers said that they would be willing to incorporate this forage into their systems if the daily milk production increased by 1.95 kg/cow during the dry season. If Scenario 1 (baseline) is compared with the other scenarios, the incorporation of Canavalia alone increases daily milk production, but does not succeed in meeting producers’ expectations. Production barely increased by 0.7 lt/day in Scenario 2, 1.4 lt/day in Scenario 4, and only 0.5 lt/day in Scenario 5. However, on-farm milk production can be increased beyond the expectations of producers by increasing the carrying capacity of farms as result of incorporating other technologies such as sugarcane and improved pastures. Possible disparities between ex ante and ex post analyses This study is framed within a broader experimental study that tries to measure changes in maize and milk productivity when C. brasiliensis is incorporated into the production systems of selected farms in the Pire river watershed in Nicaragua. The data derived from these experiments will allow these same scenarios to be evaluated ex post. Taking into account the interdependences between income and the different characteristics of production systems, the possible variations in the value of several of these regarding those used in the ex ante evaluation could imply changes in the net income of producers—the objective of this evaluation. These characteristics are listed below and should be taken into account when collecting data during the experimental phase: (a) Increase in maize productivity per ha, due to increases in biomass caused by the incorporation of C. brasilensis as green manure. This study did not assume any increment on maize productivity after incorporating the legume into the soil. (b) Contribution of N made by C. brasilensis. This study assumed a legume production equivalent to 2 t DM/ha, with 20% protein. (c) Frequency of planting of C. brasiliensis necessary to maintain the increases in maize and bean productivity over time. This study assumed that it was necessary to rotate the legume with maize every year. (d) The amount of N supplied by C. brasiliensis that is really tapped by the crop. The study assumes did not discount from the total N contributed the part that may be lost either by leaching or volatilization. 167 Maximum milk production potential of cows on selected farms. This study estimated that these cows would reach a maximum production of 4.4 lt/day with better-quality feed using C. brasiliensis and sorghum. In relation to the ex-ante environmental analysis the results may vary as some input data could change after experimentation and ex-post measurements. These data is related with the following variables: (a) Soil characteristics for the different scenarios: Saturated hydraulic conductivity, organic matter content, organic carbon, available water content and bulk density. Any variation on these parameters will affect the water balance. (b) The real estimation of % of residues remaining on the soil surface after cutting the C.brasiliensis. This will affect specially the surface runoff and sediment yields. (c) Any improvement of maize biomass after the incorporation of the legume as green manure. If the biomass is increased it could reduce the runoff. The improvement on data and a hydrological modeling at the watershed scale will permit to determine accurately the impacts on water and sediment yield in order to establish the trade off between these two environmental externalities derived from different land use scenarios. Table 2. Land use and labor of small livestock producers in the Pire river watershed, Estelí, Nicaragua. Average (n = 10) Variable Land use (ha) Natural grasses Improved grasses Crop area (maize + beans) Forest Total 8.90 0.65 2.10 0.95 12.6 Use of labor (no. of permanent annual workers) Family members (no.)2 1.0 Dedicated to agricultural activities (%) 65 Dedicated to livestock production (%) 35 Contracted (no.) 3 0.19 Dedicated to agricultural activities (%) 100 Dedicated to livestock production (%) 0 1 Same area used (i.e., the same lot) to plant maize and beans. 2 Generally the head of the household. 3 For agricultural activities related to the planting, cleaning, or harvesting of maize and beans. 168 Table 3. Animal inventory and milk production on small cattle farms of the Pire river watershed in Estelí, Nicaragua. Variable Livestock inventory (no.) Milking cows Dry cows 2-year-old heifers 1- to 2-year-old heifers 0- to 1-year-old calves Bulls Total animal units (AU)1 Stocking rate (AU/ha) Milk production (kg/cow per day) Dry season Rainy season Milk prices ($/kg) Dry season Rainy season 1 Average (n = 10) 2.4 2.2 1.8 1.1 2.5 0.2 7.7 0.81 2.1 4.1 0.32 0.27 Animal units (AU) per hectare of grass. 1 cow = 1.0 AU; one 2-year-old heifer = 0.8 AU; one 1- to 2year-old heifer = 0.6 AU; one 1- to 2-year-old calf = 0.3 AU; 1 bull = 1.3 AU. 169 Table 4. Milk and meat production, income derived from livestock production, and production costs of milk and meat on small cattle farms of the Pire river watershed, Estelí, Nicaragua. Average (n = 10) Variable 1 2 3 Milk production Total (kg/farm per year) Per day (kg/farm) Per cow (kg/day) 3,624 9.93 4.14 Meat production (kg/farm per year)2 452 Value of livestock production Milk Meat Total 1,029 360 1,389 Production costs (a) Supplementation costs ($/farm per year) Hay Concentrate Molasses Subtotal 111 18 4 132 (b) Lease of pastures during the dry season (US$/farm)1 195 (c) Family labor Allotted to livestock-related activities (no. full-time workers) Opportunity cost of family labor (US$/farm per year)3 0.35 345 Total livestock production costs Farm (US$/year) Per kg milk Per kg meat 673 0.18 0.31 Pay for day’s work ($/day) 8.3 80% of the producers interviewed regularly lease pastures during the dry season. The average was 9.1 heads of cattle during 2-3 months, at an average cost of US$3.85/head per month. Sale of weaned calves. 128 days of work, estimated on the basis of the commercial value of a day’s work in the area of $2.70/day. 170 Table 5. Production costs, productivity, and income due to the sale of maize and beans on small cattle farms in the Pire river watershed, Estelí, Nicaragua. Variable Maize Beans (n=10) (n=10) Production costs ($/ha) (a) Inputs Land preparation (animal power)1 Herbicides2 Fertilizers3 Subtotal Area planted (ha) (b) Labor Contracted4 Family member5 Subtotal Total production costs ($/ha) Unit production cost ($/kg) Production (kg/ha) Sale price ($/kg) Total production value ($/farm per year) Self-consumption Destined amount (kg/family per year) Value self-consumption (US$) Income due to sale of agricultural surplus ($/farm per year)6 Payment of day’s work ($/day)7 1 2 3 4 5 6 7 14.8 13.2 71.3 99.3 0 15.1 45.2 60.3 2.1 1.9 102.6 335.8 438.4 81.0 335.8 416.8 308 0.19 2,387 0.27 1,340 280 0.43 1,308 0.66 1,448 1,079 581 222 283 759 8.3 1,165 10.1 Land is not previously prepared for bean cultivation because beans are planted immediately after maize harvest. Glyphosate. 82 kg complete fertilizer and 82 kg urea/ha are used in the case of maize; 52 kg complete fertilizer and 52 kg urea/ha are used in the case of beans. Price of complete fertilizer = $0.45/kg; price of urea = $0.42/kg. 38 day’s work contracted for 2.1 ha of maize and 30 day’s work contracted for 1.9 ha of bean, at a cost of $2.70/day’s work. Equivalent to 124 day’s work for 2.1 ha of maize and 124 day’s work for 1.9 ha of beans, assuming an opportunity cost of $2.70/day’s work. Does not include the cost of family labor. Value of production minus the cost of contracted labor and purchased inputs. 171 Table 6. Reduction in the use of fertilizer in maize and bean crops or increase in the amount of milk produced during the dry season that producers perceive as necessary to adopt the legume Canavalia brasiliensis. Item Amount Milk production 1.95 • Increased productivity/cow to justify the adoption of Canavalia as forage (kg/cow per day) • Value of additional milk production $ 112.30 (per cow) Green manure • Reduction in use of fertilizer/ha in maize and beans, maintaining the same productivity to justify the adoption of Canavalia • Value of reduced use of fertilizers 112 kg NPK 112 kg urea $ 104.2 (per ha) Preference • Producers who preferred to adopt Canavalia (%) (jjjjj) Only for producing maize and beans (kkkkk) Only for milk production For both alternatives 30 20 60 172 Table 7. Estimated production costs, expected productivity, and unit production cost of Canavalia brasiliensis in Nicaragua. 1 2 3 4 Category Amount (US$) Production cost (US$/ha) Labor1 Fertilization Herbicides Seed3 Total 121.5 0 28.5 35 185 Production (kg DM/ha) 2,000 Unitary production cost (US$/kg) 0.0925 45 day’s work distributed as follows: 17 for planting and 28 for cleaning and herbicide application, at a cost of $2.70/day’s work. 3 lt/ha at $9.50/lt. Based on 35 kg/ha, at a cost of $1/kg. Around 1750 kg of leaves and 250 kg of grain. 173 Table 8. Characteristics of production systems in each scenario evaluated (annual value). 1 2 3 4 5 6 7 Scenario 3 Canavalia as green manure 2,849 1,147 807 1,261 631 266 135 Scenario 4 Canavalia + sorghum 2,994 1,098 798 1,277 631 266 90 Scenario 2 Canavalia for animal nutrition 3,169 1,098 798 1,692 641.48 266 141.05 5,700 1,098 798 4,068 1,218 266 365 Scenario 5 Rotation of paddocks with maize/Canavalia 8,383 1,098 798 6,770 2,561 266 384 1 2 ---- 1 2 --- 1 2 1 1 2 --- 1 2 --- ---- 1 --- 1 1 ------10 --4,470 ------10 --5,740 ------10 --4,470 ----4 6 --13,640 1.5 ------8.5 23,102 3 3.7 3 4.4 3.5 7 7 7 14 30 Characteristic Scenario 1 Baseline Net income Income due to maize1 Income due to beans2 Income due to milk3 Income due to meat4 Family labor5 Contracted labor6 Crop/grass distribution (ha/year) Beans Maize C. brasiliensis as green manure C. brasiliensis for animal nutrition Sugarcane Cratylia argentea Sorghum Jaragua grass Brachiaria brizantha Milk production (lt/year) Milk production (lt/day per cow) Meat production (kg/year) No. cows/year7 Calculated with a sale price to producer of US$270/t and a productivity of 2.4 t/ha per year intercropped in the same plot with beans. The same productivity is expected if maize is grown with C. brasiliensis as green manure (2 t DM/ha). If used as green manure, C. brasiliensis replaces 100% of the urea traditionally used. The estimated contribution of N of C. brasiliensis is equivalent to 38 kg N/ha, which surpasses current levels of application of urea (128 kg/ha per year). If C. brasiliensis is used as forage, it does not have any impact on maize productivity and is assumed to have 20% protein content, 50% protein digestibility, and 2.0 Mcal of metabolizable energy/kg. Sale price to producer is US$660/ton and productivity is 1.3 t/ha per year. A similar productivity is expected if beans are grown after C. brasiliensis is incorporated into the soil as green manure. Sale price to producer is US$0.27/lt during the rainy season and US$0.32/lt during the dry season. Sale price to producer is US$1200/t. Family labor is the total of annual day’s work required for all farm activities minus the number of day’s work contracted per year indicated by producers during the field visit. Price of contracted day’s work is US$2.70. Calculated taking into account that a cow requires 0.034 t digestible protein/semester and 2400 Mcal metabolizable energy/semester. 174 References Burle, M.L., D.J. Lathwell, A.R. Suhet, D.R. Bouldin, W.T. Bowen, and D.V.S. Resck. 1999. Legume survival during the dry season and its effect on the succeeding maize yield in acid savannah tropical soils. Trop. Agric. (Trinidad) 76:217-221. CIAT, 2006. Annual Report 2006. Cali, Colombia. FAO. 2000. World soil resources report: land resource potential and constraints at regional and country level 90, Rome. Frossard E., Bünemann E.K., Carsky R., Compaoré E., Diby L.N., Kouamé V.H., Oberson A. and Taonda S.J.-B. 2006. Integrated nutrient management as a tool to combat soil degradation in Sub Saharan Africa. In: T. Bearth, Becker B., Kappel R., Krüger G. and Pfister R. (eds.) Afrika im Wandel. vdf Hochschulverlag Zurich (in press). INTA. 2002. Programa de reconversion competitiva de la ganaderia bovina. Instituto Nicaraguense de Tecnologia Agropecuaria, Managua, Nicaragua. Lipper, L. 2001. Dirt poor: Poverty, farmers and soil resources investment. FAO Economic and Social Development Paper 149 (http://www.fao.org/DOCREP/). Millar, N., J.K. Ndufa, G. Cadisch, and E.M. Baggs. 2004. Nitrous oxide emissions following incorporation of improved-fallow residues in the humid tropics. Global Biogeochemical Cycles 18. Ministerio del Ambiente y los Recursos Naturales – MARENA. 2001. Plan de ordenamiento de la microcuenca Estelí-Estanzuela. BID, Helsinki Consulting Group, POSAF, NDF. Managua, Nicaragua. 65 p. PASOLAC. 2002. La alimentación de ganado vacuno durante la estación seca. Memoria de la gira y taller regional de ganadería, Nicaragua, Honduras y El Salvador, Mayo 27-30, 2002. Documento No. 346, Serie Técnica 13/2002. Programa para la Agricultura Sostenible en Laderas de América Central (PASOLAC). Agencia Suiza para el Desarrollo y la Cooperación (SDC). Managua, Nicaragua. Quintero, M. Estrada, R., & García, J. (2006). Model of optimization for ex-ante evaluation of land use alternatives and measurement of environmental externalities- ECOSAUT. CIATCIP-GTZ-CONDESAN-WFCP. Potato International Center. Lima, Perú. 76 p. http://www.infoandina.org/apc-aafiles/237543fdce333f3a56026e59e60adf7b/Sistematizaci_n_Per__RV_lez.pdf. Said, A.N., and A. Tolera. 1993. The supplementary value of forage legume hays in sheep feeding-feed intake, nitrogen-retention and body-weight change. Livestock Production Science 33:229-237. 175 Saxton, K. E., Rawls, W. J., Romberger, J. S. and Papendick, R. I. 1986. Estimating generalized soil water characteristics from texture. Soil Sci. Soc. Amer. J. 50(4):1031-1036. Saxton, K. E., Rawls, W. J. 1985. . Soil Water Characteristics. Hydraulic Properties Calculator. USDA- Washington State University. http://hydrolab.arsusda.gov/soilwater/Index.htm von Kaufmann, R.R. 1999. Livestock development and research in the new millennium, 24 pp. ILRI. Nairobi. 176 9.3 Climate change and risk The Impact of Climate Change in coffee-growing regions Laderach, P.a, Jarvis, A.a,b, Ramirez, J.a a International Centre for Tropical Agriculture (CIAT), AA6713, Cali, Colombia Bioversity International, Regional Office for the Americas, c/o CIAT, AA6713, Cali, Colombia b Introduction There is now little doubt that the climate is changing and will continue to change. Global Circulation Models (GCMs) all point in the direction of higher mean temperatures and changes in precipitation regimes, indicating that traditional coffee growing regions may disappear and new regions may appear. For sustainable coffee sourcing, participants of the global coffee supply chain need to have an estimate of where coffee will grow in the future and how the suitability of these areas will be. The objective of this paper is to show how coffee production areas change under progressive climate change. The impact of climate change on coffee (Coffea arabica) is assessed. The results of the present study are part of a private-public partnership project called AdapCC (www.adapCC.org) of the “Deutsche Gesellschaft fuer Technische Zusammenarbeit” (GTZ) and Cafedirect, a UK-based fair-trade hot-drink company. The ppp project focused on four coresourcing areas in Latin America (Piura in Peru, Nicaragua, Chiapas and Veracruz in Mexico). In continuation the results of the Nicaragua study are presented. Materials and Methods The methodology applied was based on the combination of current climate data with future climate change predictions from 4 global circulation models for 2020 and 18 models for 2050. The data of the current climate and the climate change was used as input to Maxent, a crop prediction model. The evidence data used for Maxent were compiled from existing databases, scientific publications, expert knowledge, and Google Earth. The analysis focused on the specific municipalities that were of interest to the regional project partners and provide predictions of the future climate and predictions of the suitability of current coffee-growing areas to continue growing coffee by 2020 and by 2050. Results In Nicaragua the yearly and monthly rainfall will progressively decrease and the yearly and monthly minimum and maximum temperatures will progressively increase by 2020 and by 2050. 177 35 300 30 250 20 150 15 Temperature (ºC) Precipitation (mm) 25 200 Current precipitation Precipitation 2050 Precipitation 2020 Mean temperature 2020 Mean temperature 2050 Current mean temperature Maximum temperature 2020 Maximum temperature 2050 Current maximum temperature Minimum temperature 2020 Minimum temperature 2050 Current minimum temperature 100 10 50 5 0 0 1 2 3 4 5 6 7 Month 8 9 10 11 12 Figure 1. Climate trend summary for sample sites 2020 and 2050 for 10 coffee-growing municipalities of Nicaragua The overall climate will become more seasonal in terms of variation through the year in temperature with temperature in specific municipalities increasing by about 1.0°C by 2020 and by about 2.3°C by 2050. In contrast, seasonality of the climate will not change in precipitation with the maximum number of cumulative dry month staying constant at 6 months. Precipitation for specific municipalities will decrease 70 to 100 mm by 2020, and 100 to 130 mm by 2050. Figure 2. Current suitability for coffee production for 10 coffee-growing municipalities of Nicaragua. 178 Figure 3. Suitability for coffee production in 2020 (above) and mean coefficient of variance of bioclimatic variables 2020 (below) for 10 coffee-growing municipalities of Nicaragua. Figure 4. Suitability for coffee production in 2050 (above) and mean coefficient of variance of bioclimatic variables 2050 (below) for 10 coffee-growing municipalities of Nicaragua The current coffee-growing areas that are today highly and moderately suitable will equally become significantly less suitable by 2020, although there will be some areas that become more suitable by 2020. By 2050 there will be further decreases in suitability to as low as 30 – 50%. With progressive climate change, areas at higher altitudes will become more suitable for producing coffee. The optimum coffee-producing zone is currently an altitude of 1200 masl, by 2020 the optimum altitude increases to 1400 masl, and by 2050 increases further to 1600 masl. Increasing altitude compensates for the increase in temperature. Between now and 2020 areas at altitudes around 800 masl will suffer the highest decrease in suitability and areas around 1600 masl the highest increase in suitability. By 2050 the corresponding altitudes will be 1000 masl and 1700 masl. 179 Figure 5. The relation between the suitability of areas for coffee production and altitude for current climates, and predicted for 2020 and 2050 for 10 coffee-growing municipalities of Nicaragua (left). The relation between the change in suitability of areas for coffee production and altitude for 2020 and 2050 compared with current suitability for 10 coffee-growing municipalities of Nicaragua (left). Conclusions The results show that the change in suitability as climate change occurs is site-specific. There will be areas that become unsuitable for coffee, where farmers will need to identify alternative crops. There will be areas that remain suitable for coffee, but only when the farmers adapt their agronomic management to the new conditions the area will experience. Finally, there will be areas where today no coffee is grown but which in the future will become suitable. These areas will require strategic investments to enable them to develop for production of coffee. Climate change brings not only bad news but also a lot of potential. The winners will be those who are prepared for change and know how to adapt. 180 Global Impacts and Implications of Climate Change on Banana Production Systems Jarvis, A., Ramirez, J., Guevara, E., Zapata, E. Centro Internacional de Agricultura Tropical, Cali, Colombia Abstract There is now little doubt that climate change is a reality to which the world must adapt to. Predictions show that climate change will bring both opportunities and challenges for the agricultural sector, although most crops will suffer reductions in productivity with temperature changes > 2oC. We present analyses on the impacts of climate change on banana production systems. First, we query the likely impacts on banana production sites using 18 global climate models (GCMs) for the year 2050 derived from the 4th IPCC assessment report. All regions with banana production suffer increase in mean annual temperature within the range of 1.5 – 3.2oC, with West African countries suffering the highest temperature increases. Precipitation changes are highly variables, with Carribean countries suffering significant reductions (>100mm less rainfall per year), whilst East Africa, South Asia and Ecuador having significant increases (>100mm). We also present decadal changes in growing environments for the world’s major producing countries, demonstrating different fortunes for countries such as Cuba (significant drying trend) versus Colombia (steady increase in precipitation). Second, we apply broad adaptation models (EcoCrop) for banana under current and future conditions. We show significant losses in climatic suitability for banana occurring in many lowland areas of Latin America (e.g. Amazon, Venezuela) and Africa (coastal West Africa), whilst suitability increases (but with high levels of uncertainty) for many sub-tropical zones (South Brazil, Australia, China) and coastal zones of Ecuador, Peru and Colombia. On average, global suitability for banana increases by 6%, but many of these gains occur in regions with low density of banana production. Third, we analyse the impacts of climate change on potential climate-induced disease pressure for black leaf streak (black Sigatoka), and present some good news for banana producers currently losing productivity to this harmful disease. Almost all major banana producing regions become less impacted by black leaf streak as maximum temperatures in the hottest months exceed the heat thresholds of the fungus. The only areas negatively affected are in Southern Brazil, south-east Paraguay, northern Vietnam and central Myanmar. Finally, we show some adaptation pathways that the research and production communities might take to adapt banana production to changing conditions. We develop a matrix of significant constraints that must be overcome in the future if banana production is to be sustained, or indeed enhanced if the opportunities that climate change presents are exploited fully. Keywords: climate change, banana, adaptation, suitability, black leaf streak. 181 Resumen Hay ahora muy pocas dudas de que el cambio climático es una realidad a la que el mundo debe adaptarse. Las predicciones de la comunidad cientifica muestran que el cambio climático traerá tanto oportunidades como retos para el sector agropecuario, especialmente considerando que muchos cultivos sufrirán reducciones en productividad si los aumentos de temperatura llegan a más de 2ºC. Presentamos un análisis sobre los impactos del cambio climático en los sistemas de producción de banano. En primer lugar, investigamos los impactos más probables en los sitios de producción bananera usando 18 modelos de clima global para el año 2050 del cuarto informe del IPCC (Panel Intergubernamental de Cambio Climático). Todas las regiones del mundo con producción bananera sufrirán un incremento en la temperatura media anual dentro de un rango de 1.5 a 3.2ºC. Los países de África occidental sufrirán los incrementos más notables en temperatura. Los cambios en precipitación son altamente variables; los países del Caribe sufrirán reducciones significativas (más de 100mm de disminución en la precipitación anual), entretanto el este de África, el sur de Asia y Ecuador tendrán aumentos de más de 100mm. Presentamos, además, cambios decadales en los ambientes de crecimiento de banano para los países de mayor producción mundial, lo que muestra fundamentalmente diferentes suertes para países como Cuba (tendencia de sequía significativa) en relación a Colombia y Ecuador (incremento gradual en precipitación). En segundo lugar, dado que la capacidad de un cultivo para crecer en un medio ambiente determinado (adaptabilidad) es una función –en gran medida- del clima, aplicamos modelos de adaptación para banano bajo las condiciones actuales y futuras (año 2050): habrá pérdidas significativas en adaptabilidad climática para el cultivo en muchas tierras bajas de Latinoamérica (e.g. Amazonas, Venezuela) y África (costas de África occidental), mientras que en muchas zonas subtropicales (sur de Brasil, Australia, China) y costeras de Ecuador, Perú y Colombia la adaptabilidad se incrementa (aunque con bajos niveles de certeza). En promedio, la adaptabilidad del banano se incrementa en un 6%, pero muchos de los aumentos ocurren en regiones con baja densidad de cultivo. En tercer lugar, analizamos los impactos del cambio climático en la presión climáticamente inducida de la Sigatoka Negra, cuyo hongo causante es altamente sensible a los cambios en la humedad, precipitación y temperatura. Hay algunas buenas noticias para los productores de banano que actualmente tienen pérdidas de productividad (o altos gastos en agroquímicos) debido a esta enfermedad: casi todas las grandes áreas de producción bananera sufrirán menos impacto de Sigatoka Negra. Esto será posible debido a que el aumento en las temperaturas máximas de los meses más calientes excederá los umbrales de calor dentro de los cuales el hongo tiene buen desarrollo. Las únicas áreas afectadas negativamente (aumento de presión de Sigatoka Negra) están en el sur de Brasil, el sureste de Paraguay, el norte de Vietnam y el centro de Myanmar. Finalmente, mostramos algunos caminos de adaptación que la investigación y las comunidades productivas podrían tomar para adaptar la producción bananera a las condiciones cambiantes. 182 Desarrollamos una matriz de principales limitaciones que debe ser satisfecha en el futuro si se pretende hacer sostenible la producción bananera a nivel mundial. Palabras clave: cambio climático, banano, adaptación, adaptabilidad, Sigatoka negra. Introduction The 4th Assessment of the IPCC (IPCC, 2007) concluded that there is now no doubt that humans are affecting the climate. The report outlines how the climate has already changed over the past 100 years, in some cases quite significantly, and provides the latest results of modelling of the global climate system to predict the likely expected changes over the coming century. Depending on how rapidly the world reacts to the climate change crisis, temperatures are predicted to increase by as much as 6oC on average across the globe to 2100. The implications of such changes are widespread, affecting almost all sectors of the economy. Regardless of whether the change be 2oC or 6oC, the various sectors of a country must address the problem by adapting the means by which they operate to maintain or enhance productivity in the face of change. Agriculture is among the most vulnerable sectors of the economy to climate change due to its very direct reliance on the climate for productivity. The IPCC report suggests minor increases in productivity for a handful of well-studied major crops so long as temperatures do not rise more than 2oC. Given the current evidence, and the latest state-of-the-art global climate models (GCMs), it is highly likely that by 2050 temperatures will have increased by an amount greater than 2oC. The picture is therefore fairly bleak for many major crops, however we still know very little about the expected changes in productivity, especially for crops that come after the big 4: maize, wheat, barley and rice. In order for society to take measures in adapting to climate change, it’s important to have some understanding of what you are adapting to. For that, we rely on GCM predictions of future climate, which come with inevitable uncertainty. Here we provide an analysis of the best-bet impacts and implications of climate change on the banana sector globally, focussing on two specific aspects: productivity and black leaf streak disease prevalence. Materials and Methods The analysis is comprised of four stages: 1. Compilation of expected changes in climate for current banana production zones 2. Niche-based mapping of suitability change for banana 3. Analysis of impacts of climate change on prevalence of black leaf streak 4. Analysis of some possible adaptation pathways to dealing with climate change Current and future climate data The IPCC 4th Assessment report was based on the results of 21 global climate models (GCMs), data for which are available through an IPCC interface (www.ipcc-data.org), or directly from the institutions developing each individual model. The spatial resolution of the GCM results is however inappropriate for analyzing the impacts on agriculture as in almost all cases grid cells of 183 over 100km are used. This is especially the case in heterogeneous landscapes such as those found across the Andes, with just one cell covering the entire width of the range in some places. Downscaling is therefore needed to provide higher-resolution surfaces of expected future climates if the true impacts of climate change on agriculture are to be understood. Two approaches are available for downscaling; 1) re-modeling of impacts using regional climate models (RCMs) based on boundary conditions provided by GCMs to generate climate surfaces with over 20km of spatial resolution for specific regions, or 2) statistical downscaling whereby resolution is reduced using interpolation and explicit knowledge of fine-scale climate distribution and correlations between major climatic variables. Whilst the use of RCMs is more robust from a climate science perspective, it requires significant re-processing, and RCMs are only available for a reduced number of GCM models. It is only realistic to include 1-2 RCMs in any analysis (due to the high processing requirements), and so in the context of this project the use of an RCM for only one GCM would result in the inability to quantify uncertainty in the analysis, which we feel is inappropriate. We therefore have used statistically downscaled data derived from a larger set of GCMs. We downloaded and re-processed IPCC 4th Assessment climate change results from 18 of the most reputable GCMs (Table 1) and applied a statistical downscaling method to produce 10km, 5km and 1km resolution surfaces of future monthly climate (maximum, minimum, mean temperature and precipitation) for the time period representing 2020 (for 4 models) and 2050 (for 18 models). In both cases the emissions scenario A2a (business as usual) has been used. Table 1. Global climate models used for the analysis Originating Group(s) Bjerknes Centre for Climate Research Canadian Centre for Climate Modelling & Analysis Canadian Centre for Climate Modelling & Analysis Canadian Centre for Climate Modelling & Analysis Météo-France Centre National de Recherches Météorologiques CSIRO Atmospheric Research CSIRO Atmospheric Research Max Planck Institute for Meteorology Meteorological Institute of the University of Bonn Meteorological Research Institute of KMA LASG / Institute of Atmospheric Physics US Dept. of Commerce NOAA Geophysical Fluid Dynamics Laboratory US Dept. of Commerce NOAA Geophysical Fluid Dynamics Laboratory NASA / Goddard Institute for Space Studies Institut Pierre Simon Laplace Center for Climate System Research National Institute for Environmental Studies Frontier Research Center for Global Change (JAMSTEC) Center for Climate System Research National Institute for Environmental Studies Frontier Research Center for Global Change (JAMSTEC) Meteorological Research Institute National Center for Atmospheric Research Hadley Centre for Climate Prediction and Research Met Office Center for Climate System Research (CCSR) National Institute for Environmental Studies (NIES) Country Norway Canada Canada Canada MODEL ID BCCR-BCM2.0 CGCM2.0 CGCM3.1(T47) CGCM3.1(T63) OUR ID BCCR_BCM2 CCCMA_CGCM2 CCCMA_CGCM3_1 CCCMA_CGCM3_1_T63 GRID 128x64 96x48 96x48 128x64 France CNRM-CM3 CNRM_CM3 128x64 Australia Australia Germany Germany Korea China CSIRO-MK2.0 CSIRO_MK2 CSIRO-Mk3.0 CSIRO_MK3 ECHAM5/MPI-OM MPI_ECHAM5 64x32 192x96 N/A ECHO-G MIUB_ECHO_G 96x48 FGOALS-g1.0 IAP_FGOALS_1_0_G 128x60 USA GFDL-CM2.0 GFDL_CM2_0 144x90 Year 2050 2020-2050 2050 2050 2050 2020 2050 2050 2050 2050 2050 USA GFDL-CM2.0 GFDL_CM2_1 144x90 USA France GISS-AOM IPSL-CM4 GISS_AOM IPSL_CM4 90x60 96x72 Japan MIROC3.2(hires) MIROC3_2_HIRES 320x160 2050 2050 2050 2050 Japan MIROC3.2(medres) MIROC3_2_MEDRES 128x64 Japan USA MRI-CGCM2.3.2 PCM MRI_CGCM2_3_2a NCAR_PCM1 N/A 128x64 UK UKMO-HadCM3 HCCPR_HADCM3 96x73 Japan NIES-99 NIES-99 64x32 2050 2050 2050 2020-2050 2020 184 The statistical downscaling has been performed using the WorldClim dataset for current climate (Hijmans et al., 2005, available at http://www.worldclim.org) and a spline interpolation technique. Specifically, the centroid of each GCM grid cell is calculated, and the anomaly in climate assigned to that point. A spline- algorithm is then used to interpolate between the points to the desired resolution. The higher-resolution anomaly is then summed to the current distribution of climate (derived from WorldClim) to produce a surface of future climate and 19 bioclimatic variables are finally derived from monthly downscaled variables according to Busby (1991). The method assumes that the current meso- distribution of climate remains the same, but that regionally there is a change in the baseline. Whilst in some specific cases this assumption may not hold true, for the great majority of sites it is unlikely that there will be a fundamental change in meso-scale climate variability. In addition to working on changes in the climate baseline we also have available data on year-toyear variability in climate; an important component of climate change which can have significant impacts on agricultural production and food security. We downscaled monthly data from 8 GCMs for each year from the 20th century through to 2100. Throughout the analysis, the use of 18 GCM models allow the framing of results in terms of the associated uncertainty in future climate. Crop adaptability We follow the methodology of Lane and Jarvis (2008) for examining the impacts of climate change on productivity. Physiologically-based mechanistic crop models are available for only a small percentage of the world’s crops, and althoguh for banana such models already exists they are not currently applicable worldwide. In the absence of mechanistic crop models, the Ecorop model (http://ecocrop.fao.org/) provides a simple method to evaluate climate change impacts on a wide range of crops, including banana. The Ecocrop model contains information on the edafoclimatic requirements for 1,300 cultivated species considering optimal conditions and limits to adaptation. Ecocrop is implemented in DIVA-GIS (Hijmans et al., 2005) and has been interfaced with monthly precipitation and temperature data to permit mapping of suitability on a global scale based solely on climate data. We apply the banana ecocrop model for current conditions, and for the 18 future 2050 climatic conditions from the different models. We then calculate an average future suitability, and by looking at the difference between future and current adaptability we report a range of basic descriptive statistics on changes in adaptability in different countries and regions of the world. We also report uncertainty in our projected changes by calculating the coefficient of variation (proportion of the standard deviation in the predictions’ average) in adaptability change between the 18 different climate models. Changes in black sigatoka prevalence We follow the methodology of Ramirez et al. (this volume) for analysing the changes in climatically-induced disease pressure from black leaf streak disease (BLS, or black sigatoka). Ramirez et al. (this volume) generated statistical models for predicting BLS disease pressure through analysis of field experimental data. We apply the same model to future climate conditions, and calculate the future level of BLS disease pressure for each of the 18 GCM future climates for 2050. Like for Ecocrop, we take an average and report the change in disease pressure through descriptive statistics for countries and regions, and also classify uncertainty using the coefficient of variation between GCM model predictions. 185 Adaptation pathways A range of adaptation pathways exist for dealing with (and profitting from in some cases) the changes in climate. Here we only give examples of possible adaptation options, and is by far not an exhaustive list. Results and Discussion Predicted climate changes in banana production zones The GCM data for both 2020 and 2050 in banana production zones was queried to describe the likely changes for each banana producing country in the world (Table 2). On average, banana producing zones increase in precipitation by just 6mm of annual rainfall, although some countries suffer significant increase in precipitation (e.g. Indonesia, Pakistan, Kenya, Ecuador) and other quite significant decreases (e.g. Honduras, Nicaragua, Grenada). On the whole, Europe (sub-tropical banana zones) and the Caribbean suffer the greatest drying (reduction of 124mm and 110mm per annum), whilst Asia and Sub-Saharan Africa get the wettest (increase of 59mm and 42mm per annum respectively). Temperatures increase by 2.3oC on average globally, though the increases are highest in North Africa (2.9oC), Europe (2.6oC) and sub-Saharan Africa (2.5oC). The Caribbean suffers the least temperature increase of 1.7oC. 186 Table 2. Descriptive changes in average climate for the banana production zones of each banana producing country in the world for 2050 Country Region Bangladesh Bhutan Brunei Cambodia China Cyprus India Indonesia Japan Laos Malaysia Myanmar Burma Nepal Asia Pakistan Papua New Guinea Philippines Solomon Islands Sri Lanka St Vincent Grenadines Taiwan Tanzania Thailand Turkey Vanuatu Vietnam Australia Australia Antigua and Barbuda Barbados Cuba Dominican Republic Dominicana Caribbean Grenada Guadeloupe Haiti Jamaica St_Lucia Trinidad and Tobago Albania Bulgaria France Europe Greece Macedonia Portugal Spain Latin Argentina America Belize Bolivia Brazil Colombia Costa Rica Ecuador El Salvador Guatemala Guyana Total Annual Precipitation Temperature Precipitation Consecutive precipitation mean coefficient of coefficient of seasonality dry months change temperature variation variation change change (mm) change (ºC) (%) (%) 113.822 2.230 -4.297 0.0 3.972 4.385 135.103 2.651 -9.262 0.0 9.223 10.908 14.258 1.783 1.070 0.0 2.127 2.064 20.366 2.070 0.576 0.0 4.527 2.661 52.539 2.394 -1.313 -1.0 5.668 6.326 -42.590 2.211 -7.392 1.0 11.703 4.390 185.577 2.413 -15.326 -1.0 8.921 4.231 185.577 2.413 -15.326 -1.0 8.921 4.231 42.806 2.209 1.720 1.0 2.683 4.359 33.169 2.213 -0.886 0.0 4.353 4.245 -0.434 1.848 1.322 0.0 2.975 2.120 107.909 2.252 -6.038 0.0 4.867 4.342 178.806 257.852 2.594 3.013 -7.156 -19.018 0.0 -1.0 9.329 51.713 5.390 5.809 68.346 1.858 1.308 0.0 3.827 2.462 37.752 1.697 0.503 0.0 3.208 1.309 145.589 1.586 0.848 0.0 7.693 0.941 41.959 1.747 1.824 0.0 4.445 1.596 -138.300 1.559 0.206 1.0 9.424 1.169 11.933 101.217 41.497 -97.396 -9.637 3.105 81.401 1.764 2.277 2.138 2.856 1.498 2.007 2.384 2.180 -4.038 -1.925 1.487 1.986 1.896 -0.496 0.0 0.0 0.0 1.0 0.0 0.0 0.0 4.633 7.796 4.833 7.247 4.183 3.276 12.514 1.644 3.715 3.282 5.615 1.033 3.030 4.187 -86.311 1.535 5.433 1.0 13.653 1.145 -105.667 -72.869 1.562 1.768 2.679 2.862 1.0 0.0 11.495 8.344 1.066 1.933 -95.659 1.835 3.283 1.0 10.271 1.985 -129.420 -139.639 -111.697 -81.332 -73.158 -127.657 1.546 1.589 1.539 1.821 1.726 1.572 1.300 0.194 2.281 1.711 2.910 0.952 0.0 0.0 1.0 0.0 1.0 0.0 8.718 10.385 9.121 8.650 5.506 9.884 1.294 1.075 1.271 2.230 2.023 1.157 -186.898 1.998 1.704 1.0 10.460 2.487 -161.794 -125.967 -106.611 -114.509 -138.139 -109.854 -116.618 47.917 -145.478 -13.887 1.029 51.070 -67.955 116.421 -54.633 -97.506 27.845 2.627 2.744 2.594 2.512 2.792 2.332 2.673 2.590 2.205 2.857 2.576 2.315 2.080 2.108 2.299 2.373 2.366 17.746 20.148 10.389 11.619 24.306 10.154 14.362 -4.084 2.800 3.465 1.844 -2.015 -0.459 -3.294 -3.097 0.940 -2.802 1.0 1.0 1.0 2.0 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 8.263 12.214 4.893 7.266 13.552 8.029 11.510 10.026 7.501 7.134 5.562 4.778 7.333 6.058 7.238 5.315 8.092 9.339 7.563 6.618 5.395 7.174 5.276 6.308 4.181 3.255 4.315 3.861 3.370 2.291 2.763 3.367 3.607 3.356 187 Examining the yearly time-series of change, the climate change trajectories of some banana producing regions for specific countries can be seen (Figure 1). On the whole the trends are fairly gradual and linear, although for some countries there is significant yearly and decadal variability in climate (e.g. Vanuatu). Other countries have non linear changes, for example in Japan where it gets slightly dryer before the trend changes towards increase in precipitation around 2050. 4.0 3.5 2099 Modeling time-limit Temperature anomaly (ºC) 3.0 2.5 2.0 2050 Modeling time-limit 1.5 2020 Modeling time-limit 1.0 0.5 1870 Baseline 0.0 -600 -400 -200 0 200 400 600 Precipitation anomaly (mm) Haiti Central African Republic Burundi China Ecuador Cuba Venezuela Japan Colombia Mexico Myanmar Burma Vanuatu Costa Rica Figure 1. Yearly-time series of changes in climate for selected countries using the NCAR-CCM3 GCM model Impacts on crop adaptability The results of the Ecocrop model demonstrate an average increase of 6% in climatic suitability for bananas globally. The average increase in suitability for the ten highest producing countries is also 6%, though it is sub-tropical banana growing regions in China, Brazil and India that contribute most to this increase. Thailand and Colombia both suffer slight decreases in suitability (-5% and -2% respectively). The biggest winners are in Sub-Saharan Africa, especially Kenya (+23%), Rwanda (+23%), Uganda (+24%) and Ethiopia (20%), and sub-tropical Latin America (e.g Paraguay with an increase of 26%). The biggest losers are in the Caribbean (e.g. Barbados with -9% change), SE Asia (e.g. Cambodia with -6% change) and West Africa (e.g. Togo with 8% change). The changes in banana suitability are shown as a map in Figure 2. 188 Figure 2. Changes in climatic suitability for banana across the globe for 2050 There is considerable variability in changes within countries (Figure 3). All regions of China, Burundi, Taiwan and Vanuatu experience increases in suitability, whereas in Venezuela almost all regions suffer decreases. The remainder of countries tend to have a range of impacts, with some regions increasing in suitability and others decreasing. On the whole, the modelling of crop adaptability show that the 2050 climate has the potential to harbor more banana production, but significant shifts in the current distribution of banana are required to achieve such increases. These changes would require the expansion of banana production in some countries, decrease in others, and within countries there is evidence of significant shifting importance of different regions. 189 40 Banana adaptability change 30 20 10 0 -10 -20 -30 Laos Taiwan Brazil South Africa Bangladesh Ecuador Costa Rica Colombia China Vanuatu Japan Burundi Myanmar Venezuela Cen. Africa Mexico Cuba Haiti -40 Figure 3. Projected changes in banana climatic suitability for 2050 for selected countries, bars representing within country range and extremes. Impacts on black sigatoka Following the model of Ramirez et al. (this volume) the impacts of climate change on black leaf streak disease pressure indicate an average global decrease of 1.3%, and 3% in the top-ten producing nations. The picture of change is almost the opposite as that for climatic suitability for production – sub-tropical regions which currently do not suffer from black leaf streak become much more suitable as low temperature thresholds are exceeded and the disease becomes more prevalent. Almost all tropical regions are predicted to experience less pressure form sigatoka as increases in temperature push the maximum temperatures above the threshold for the fungal disease. The biggest losers are China (23% increase in disease pressure), Taiwan (22% increase), Swaziland (51% increase), South Africa (31% increase), and Paraguay (22% increase). Almost all tropical countries experience decreases of 3-8% in disease pressure. The results are shown in a map (Figure 4), and for selected countries the mean and ranges of changes are shown (Figure 5). 190 Figure 4. Changes in climatic suitability for black leaf streak disease across the globe for 2050 10 BLS disease severity change 8 6 4 2 0 -2 -4 -6 Laos Taiwan Brazil South Africa Bangladesh Ecuador Costa Rica Colombia China Vanuatu Japan Burundi Myanmar Venezuela Cen. Africa Mexico Cuba Haiti -8 Figure 5. Projected changes in black leaf streak disease suitability for 2050 for selected countries, bars representing within country range and extremes. 191 Adaptation pathways A range of adaptation pathways become evident from the above results. The pathways are largely site-specific, and depend on the current banana cropping systems and the predicted changes. In simple terms, the following three options, in order of severity, would be required: 1. Change crop management 2. Varietal change 3. Migrate to different zone or change crop With increasing magnitudes of expected changes, different severity of adaptation measures are needed. As a worst case resort banana producers may need to relocate their crop, or change to a different crop altogether. However, in some cases these pathways may be designed to maximise the potential rather than mitigate negative changes. Cross-cutting across these three types of adaptation is the potential of research and development to provide novel responses to climatic changes. These may lie in the development of new management regimes, promotion of precision and site-specific agricultural approaches that match the management to site-specific edafo-climatic variability, or in technology development, including breeding of new varieties with novel biotic and abiotic traits. Conclusions We present here the results of a modelling exercise to understand the impacts and implications of climate change on banana productivity and black leaf streak disease pressure. We show that climate change is not all bad news for the banana sector, with average increases in crop suitability, and significant decreases in black sigatoka disease pressure for many tropical countries. However, there are some hotspots where significant problems are likely to be experienced, requiring quite significant adaptations in order to overcome the challenges of climate change. The analysis shown here looks only at two components of the banana production system: adaptability and black leaf streak. Many other facets of the production system have not been analysed and may have even greater significance to the banana sector. No matter what the predictions say, change is inevitable. The banana sector must continually adapt to changing biophysical conditions as well as social and economic changes which have plagued the banana sector in the past decades and will continue to do, most likely at a faster rate than climate change. Nevertheless, fundamental changes in the climate baseline to 2050 will require fundamental changes in production systems, and will likely have profound impacts on the global balance of production, especially with the increasing role of sub-tropical banana production. References Busby, J.R., 1991. BIOCLIM – a bioclimatic analysis and prediction system. Plant Protection Quarterly. 6, 8-9. Hijmans, Robert J., Cameron, Susan E., Parra, Juan L., Jones, Peter G., and Jarvis, Andy. 2005. Very high resolution interpolated climate surfaces for global land areas. International Journal of Climatology, 25, 1965-1978 IPCC, 2007. Climate change 2007: Synthesis report. Summary for policymakers. Technical report. URL http://www.ipcc.ch/ipccreports/ar4-syr.htm 192 Lane, A; Jarvis, A. 2007. Changes in Climate will Modify the Geography of Crop Suitability: Agricultural Biodiversity can Help with Adaptation.. SAT eJournal. Volume 4: Issue 1. Hijmans, R. J.; Guarino, L.; Jarvis, A.; O'Brien, R.; Mathur, P. 2005. DIVA-GIS. Available from http://www.diva-gis.org. Ramírez, J.; Jarvis, A.; Van den Bergh, I. 2008. Presión de la Sigatoka Negra y Distribución Espacial de Genotipos de Banano y Plátano: Resultados de 19 Años de Pruebas con Musáceas. In: Proceedings XVIII Reunión Internacional ACORBAT 2008. Guayaquil, Ecuador Nov. 10 – 14 de 2008. 193 Helping small-holder farmers to manage drought risk through insurance: A case study of dry bean production in Honduras Díaz-Nieto, J.a, b, Cook, S.E.a, Jones, P.a, Laderach, P.a a b Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia The University of Sheffield, Sheffield, United Kingdom. Abstract The struggle to find sustainable formal insurance for droughts in developing countries captures the attention of many in the development community for good reason. Droughts disrupt the development process via poor access to credit and an unwillingness of the poor to invest. This paper first examines the problems with traditional approaches to formal drought insurance and then examines the potential promise of index-insurance that is event driven. A combined weather-generation and crop simulation modelling approach is used to estimate site specific risks. The method is demonstrated in a case study for dry bean production in Honduras. Keywords: Risk, poverty, drought insurance, dry beans, Central America, Honduras, Introduction This paper begins by examining the role of crop insurance in assisting agricultural investment, indicating the potentials, opportunities and pitfalls. While traditional approaches to crop insurance have been lacking, new approaches involving index-insurance weather events linked to yield shortfalls are being tested around the world. To date, these new approaches have largely neglected the potential of site-specific crop growth models that can be linked to specific weatherbased insurance. That is the major focus of this paper. Crop growth simulation models are linked to a weather generation process to estimate risks of dry bean production for different locations in Honduras. The method is demonstrated by linking site-and soil specific plant growth with drought insurance that is risk-rated to cover drought risks for dry beans. Climatic risk is a major problem for poor farmers in the tropics. The fear of climatic risk hinders investment that drives development and so presents a significant obstacle to changes that might otherwise enable people to climb out of poverty. Equally important, just as individual households are beginning to escape the gripes of poverty, weather shocks can and do stop that progress. The literature that describes these poverty traps that are linked to risk is growing (e.g. Stephen Dercon’s edited book Insurance Against Poverty). Farmers adopt a range of strategies to cope with risk, including avoidance, management or risk-sharing. Possibly the most widely used method of risk sharing in the developed world is formal insurance. Yet formal insurance is used by very few poor farmers in the developing world, who are obliged to rely almost entirely on less effective mechanisms of risk avoidance, or risk sharing through informal arrangements and selfinsurance. 194 Can crop insurance alleviate poverty? Risk increases poverty Farmers face crop losses due to drought, floods, frosts, fire, pests, theft and other hazards. Of these, the most prevalent are weather risks that affect hundreds of millions of poor farmers each year. Nearly 80% of farmers interviewed in Ethiopia cited harvest failure caused by drought, flooding or frost as the event that caused them most concern (Dercon, 2002). Pandey et al. (2001) revealed a huge drop in income for rice farmers in Orissa as a result of drought. A review of chronic rural poverty, (Bird et al., 2002) identified exposure to risk as a major modifiable reason for chronic poverty, noting the widespread evidence that correlates risk with poverty. Dercon (2005) has numerous chapters that demonstrate a strong link between shocks and poverty. Increasingly studies are finding that the poor in developing countries are a transitory group that moves in and out of poverty on a regular basis. Shocks from a wide range of risk related events stop progress and send households who are making progress back to the poverty ranks. These poverty traps justify some type of public intervention using both equity and efficiency criterion. As Dercon concludes “social protection may well be good for growth.” [page 2, Dercon (2005)]. Self-insurance is a common method of coping with risk Farmers are well aware of the detrimental impacts of risks, and instinctively adopt a range of informal self-insurance methods (Table 1). While these methods have been documented to reduce the adverse impact of weather risks on poverty (Webb and Reardon 1992; Morduch, 1999), they do so by spreading investment internally without reference to the actual risks involved, and consequently dampen capital formation through inefficient use of resources (Hazell et al., 2000). Excessive fear of weather hazards also prevents poor farmers from taking reasonable risks that might otherwise lead them out of poverty. Many case studies show how the consequent chronic under-investment holds back farming systems from development (Webb and Reardon, 1992 and Rosenzweig and Binswanger, 1993). Table 1. Self-insurance measures and their impact (Skees, 2003 and Dercon, 2002) Self insurance measure How it acts as a barrier to development Diversification is often recommended however it Diversification is not beneficial if it involves diversifying away from the most productive practices. Accumulation of financial reserves and Financial reserves are not re-invested but are stocks on farm stored as a preventative measure. This is effective for independent risks, but less so Reliance on off-farm income generation for correlated risks when there is high competition and low wages. Selling of assets when everyone is trying to sell Selling assets (e.g. cattle) lowers prices and it may involve a net loss. The fear of losing an entire crop due to unfavourable weather holds back farmers from Avoidance of investment (e.g. fertilizing) costly but more productive investments (such as fertilizing). 195 Formal insurance is a more effective coping mechanism Insurance enables farmers to take reasonable investment risks, such as buying fertilizer in areas afflicted by occasional drought. By means of indemnity payments farmers can survive periods of financial stress that might otherwise ensue when a hazard strikes. Formal insurance has been used for centuries to manage catastrophic risks by means of a transparently determined estimate of the risky event. This transparent and mutual sharing enables risk to be spread more widely, far beyond the immediate community where the event is sustained, resulting in more sustainable and affordable insurance. The potential impact of insurance relates to two factors: the degree of improvement that insurance is likely to bring to individuals and the number of individuals who are likely to benefit. A well-designed insurance scheme has several major advantages over selfinsurance: • Protection of general capital more rapidly enhancing the opportunity to move to enterprises with higher mean incomes, • After establishment the scheme is self-financing: mutual benefit occurs to both insurer and insured, • Insurance is a tried-and-tested means of encouraging reasonable risk taking while discouraging excessive risk, • Insurance is progressive; insurers can increase the range of hazards they cover as knowledge accumulates about likelihood of events, • Insurance is an effective method for communicating knowledge about risk through prices that reflect the best available scientific knowledge, and improved knowledge about risk can lead to better management practices and • Insurance can help smooth incomes and reduce the frequency of falling into a poverty trap for those who are generally forced to sell off productive assets when there is a shock. Why isn’t insurance used more widely in poor countries? Paradoxically people in the developing world who are most seriously affected by risk, are poorly served by insurance (Wenner and Arias, 2003). Insurance nonetheless has been used to manage risk in developed countries for centuries, apparently to mutual benefit. One assumption is that weather insurance does not exist in developing countries because the cost reduces the demand from poor farmers with little surplus. However an evaluation of a micro-insurance scheme by Ahuja and Jutting (2004) concluded that it is not a problem of affordability but organization that could be overcome partially by incorporating insurance into established micro-financial services. Sakurai and Reardon (1997) also found that the demand for formal insurance is likely to be high where alternative self-insurance mechanisms are not adequate for reducing vulnerability. Furthermore, the literature on use of informal credit and the accompanying high cost to the poor suggests that the poor may be willing to pay for effective insurance. Several governments offer crop insurance schemes to farmers to overcome apparent market failure (Table 2). To date most remain either fully government owned or heavily subsidized, mainly due to the fact that no private insurance company considers it prudent to cover such widely correlated risks. Miranda and Glauber (1997) and Skees et al. (1999) add that one of the main reasons for the failure of publicly owned insurance schemes is that they offer either multiple peril or all risk programs, for which indemnity payments are unsustainable on the basis of premiums alone. Administrative problems have therefore contributed to a generally abysmal history, in response to which many governments and private companies decline to invest. 196 Table 2. Examples of crop insurance schemes Location Comments Crop hail insurance has been offered for over 100 years. Private sector insurance provides single peril insurance profitably. The government provides multiple peril insurance, USA however the scheme suffers from high loss ratios and increased premium subsidies have been used to mask poor actuarial performance. A heavily subsidised government insurance scheme existed for most crops. All uncontrollable risks were covered and Brazil premium rates were equal across zones and crops, which led to manipulation of the scheme. The program has had a high loss ratio and only high-risk farmers have benefited. State owned companies offer area index insurance but it has been unsuccessful and has a claims ratio that averages 500 India percent. Failure is attributed to the fact that premiums and claims were not equitably distributed across crops and states. A heavily subsidized insurance scheme existed, however participation rates were low and indemnity payments slow to Morocco reach policyholders. In 1999/2000 indemnity payments outstripped premiums and Reinsurance resources were consumed. Agricultural insurance in Uruguay was available at a limited Uruguay scale and was under state monopoly. The limited uptake is mainly due to the unofficial policy of automatic disaster relief. In 2001 Mexico was the first developing country to experiment with weather indexes. Weather markets were used to reinsure part of the multiple crop government insurance programs. Weather index-insurance for individual farmers is a voluntary Mexico program. Although more cost effective, coverage is lower than with the former insurance. Groups of farmers working through the FONDOs offer some promise for using weather index insurance in Mexico to basically start a mutual insurance system. Source Wenner and Arias (2003) Rezende Lopes and Leite da Silva Dias (1986) Hess (2003) and Manojkumar et al. (2003) Skees et al. (2001) and Wenner and Arias (2003) Wenner and Arias (2003) Hess (2003), Stoppa and Hess (2003) and Wenner and Arias (2003) Problems with formal insurance In almost all cases, insurance has failed not because of the principle but because of technical and administrative problems. The most common problems are described briefly below: Technical problems • Information asymmetry between insurer and insured: since farmers know more about the likelihood of crop failure on their farm than the insurer, this asymmetry in knowledge will deter development of effective crop insurance. Insurers cannot write policies unless they can estimate, with reasonable accuracy, the likelihood of the insured event. This requires reliable and accurate historical data that are lacking in many poor countries. In most cases 197 • accurate estimation of insurable events is not possible and remains a major reason for failure. Adverse selection and moral hazard: insurance schemes should not preferentially encourage farmers in high-risk situations to buy policies (adverse selection). Nor should policies insure unwise behavior (moral hazard). Historical data are required to identify high-risk farmers and independent estimates are needed to reduce moral hazard. Once again the lack of data is a major obstacle. Administrative problems • Corruption and political bias: several agricultural insurance schemes were initiated by governments and are either damaged by political bias or outright corruption. An example of this has been the corruption of local insurance officials in a Mexican scheme (Wenner and Arias, 2003). • High administration costs: early schemes for pro-poor agricultural insurance were seriously hampered by high administrative costs to oversee contract preparation and payment. Individual contracts are unworkable for small farmers and validation of individual claims is expensive and impracticable. • Lack of Reinsurance: Reinsurance is essential to the long-term viability of nearly all multiple peril crop insurance schemes since risks such as drought risk are highly correlated creating large losses for the insurer. Without Reinsurance multiple peril and geographically concentrated schemes are vulnerable to collapse (Miranda and Glauber, 1997). The Reinsurance market is extremely thin in this respect and offers little support. Alternative financial instruments have been proposed (Goes and Skees 2003) but their use to date has been limited due to many of the fundamental problems associated with organizing abusefree crop insurance schemes. • Lack of organizational infrastructure: a major shortcoming of existing insurance has been its inability to deliver support to poor farmers, who are, by definition, also deprived of access to finance and information technologies. Given high fixed administrative costs, larger policy holders (i.e., farmers of larger operations) are more likely to benefit the most from traditional crop insurance schemes. Weather insurance to cover crop loss Whilst it is crop loss that farmers wish to cover, it is more secure for the insurer to offer protection against weather events. Standard and independent procedures for weather data collection exist and hence it is easier to obtain the likelihood of insured events that are likely to create large crop failures. Weather markets took off in the 1990s in the North American energy sector (Turvey, 2001). Given the obvious relationship between weather and crop yield, agricultural economists began to explore the potential for weather insurance to manage agricultural risks. The principles of weather insurance are summarized in box 1 and explained in detail in Hazell et al., (2000); Skees (2000); Varangis, (2001); Skees et al., (2001) and Bryla et al., (2003). Although there remain many challenges to be resolved, weather insurance has the potential to address many of the problems faced by formal insurance as discussed below. We summarize below the potential advantages and disadvantages to a weather based system, together with possible solutions. 198 Advantages • Reduction of the information asymmetry problem: quantitative independent estimates of probabilities of weather events at specific sites reduce the risks to insurers, who can consequently lower the cost of premiums to policy-buyers. • Reduction in adverse selection: premiums are based on the most accurate estimates of sitespecific probabilities. While some risk of adverse selection remains where farmers have better estimates of long-term probabilities than the insurer – the risk is greatly reduced. • Moral hazard and corruption: since the trigger for indemnity is unrelated to individual farmer decisions, moral hazard is avoided and corruptions greatly reduced. • Administration costs: the trigger for indemnity payments is an easily measurable weather event, rather than yield loss. This removes the cost of inspection and loss adjustment. Cost can be further reduced by using standard unit contracts (Skees et al., 2005). • Reinsurance: while weather index-insurance does not remove the correlated risk, such contracts offer new possibilities for sharing these types of risks. Jaffee and Russell (1997) argue that on a long-term scale, catastrophic risk sharing is viable. Spatially distributed historical records allow risks to be spread across uncorrelated areas. The geographical specificity of events allows alternative reinsurance schemes to be exploited. • Ability to use existing organizations to access the rural poor: relaxing the need for direct inspection opens up new options for distribution. One of the most widely advocated options is to offer such index-insurance through established micro finance institutions (MFI). In fact there are mutual benefits between the insured and MFI, since high climatic risk is a significant obstacle to credit being offered to farmers (Hess, 2003, Skees 2003). Other options include distributing the insurance via farming co-operatives (Black et al., 1999) or disaster relief organizations (Goes and Skees 2003) Challenges [and solutions] • Lack of historical data: historical data, which is largely unavailable in poor countries, is essential to obtain the frequency of weather events. [Where such data does not exist it can be simulated using grid based weather simulation models.] • Weather must explain variation in yield: while it is an accepted fact that weather explains much of the variability of crop yield, the quantitative nature of this relationship must be established for it to provide a basis for insurance (Skees et al., 2005). [This relationship can be quantified by means of crop growth simulation models.] • Basis risk: temporal basis risk occurs when in total there has been a sufficient amount of precipitation, but the timing has been unfavourable and therefore crops have suffered. Spatial basis risk is whereby the rainfall station being used to administer payments registers adequate precipitation, but in fact a farmer didn’t experience the registered amount of rainfall. [Spatial basis risk is greatly reduced by offering site specific insurance.] • Secure measurement: the insurer requires independent and reliable data regarding indemnifiable events. [This requires tamper proof weather stations; an option is for these to be put in place by the insurance provider. This also has the dual purpose of improving the network of climatic data in the region.] Need to develop an insurance product The discussion above should make it clear that one needs site-specific weather insurance to offer the most effective risk-sharing scheme. Firstly the evidence is compelling that insurance against 199 weather risks is needed. Secondly a major obstacle to development of sound index-insurance products is the basis risk represented by non-specific schemes. Finally this paper demonstrates a site-specific and soil-specific method of estimating drought risk for bean farmers in Honduras. Method: developing a drought insurance scheme for Honduras For a given site and soil the method estimates the frequency of drought events that are associated with crop yield loss and the premium that would be required to indemnify, given the frequency of the event. The basis of this relationship is explained in Box 1. The method breaks down into four steps: 1. Establishing a transparent insurance process that relates indemnifiable events, indemnity payments and premiums 2. Generation of weather data for specific sites, from which to determine frequencies of events 3. Relating weather events to their likely impact on crop yield by means of the crop growth simulation model: DSSAT 4. Relating likely yield loss to readily-determined weather indices Box 1: Principles of weather insurance The insurer offers protection against a defined weather event, normally a hazard such as drought, frost or excess rainfall. Since much more is known about the weather than its consequences, these events are of more certain frequency, and provide the basis for policies against which farmers or their advisors can take out policies. The premium relates to the probability of the event and the size of indemnity according to the general formula (Brown and Churchill, 1999): Premium = f (Indemnity. Probability of occurrence) While the relationship between premium and indemnity must be determined solely on the basis of probabilities of events (viz. their expected frequency), certain parts of a scheme are adjustable to suit both parties. The trigger for indemnity payment and its size can be adjusted - through the strike event - to suit the preferences of individual customers. For example, a farmer who believes he can manage all but the most serious events may choose a contract with low premium that pays indemnities only against the most exacting trigger. Conversely, a farmer who is in a more vulnerable situation may prefer a contract that pays smaller indemnities more frequently; or against larger premiums. Insurers may reduce payments to be proportional. Background: site selection Honduras was chosen as an exploratory study site. This country is often hit by droughts that have a serious impact on dry bean crops that are important source of food and income for poor farmers. It is estimated that 16 million kilograms of dry bean yield - more than one third of expected yields - were lost as a result of the drought in 2001 (CEPAL, 2003). This study therefore pursues a bean specific (activity specific) and site-specific rainfall insurance scheme. Insurance premiums were estimated for six locations distributed throughout the bean-growing region of Honduras, as shown in Figure 1. 200 SAN VIL SIG SIGC SIGB PAR Figure 1. Location of trial sites in Honduras shown in relation to the grid used for climate interpolation by MarkSim Step 1: Establishing an insurance process In the proposed insurance process the insurer agrees to indemnify policy-holders in the event of drought. A drought insurance premium is established for a particular location on the basis of the average indemnity payment that is expected at the location. This is estimated on the basis of the frequency of the drought events. In any given year, payment is triggered by a drought event, or ‘strike’. Drought is deemed to occur when the rainfall falls below a pre-determined ‘strike’ level. Rainfall is expressed as an index that is weighted to account for the temporally variable effects on crop yield. The size of indemnity payment is scaled according to the severity of drought, up to a maximum limit determined by the insurer. The strike and maximum indemnity may be adjusted by the insurer to improve the viability or attractiveness of the insurance scheme. Step 2: Generating site-specific weather data An insurable event is normally defined on the basis of historical data. Such data does not exist for Honduras (or many other areas in the developing world) at the spatial resolution required, so we start by generating pseudo-historical data using the MarkSim weather-generating model, designed specifically for tropical weather systems. (Jones et al., 2002). MarkSim is capable of simulating daily rainfall and temperature data for any point in the tropics at a resolution of 10 arc minutes (approximately 17km² in the Central American region). Technical details can be found in Jones and Thornton (1993), Jones and Thornton (1999), Jones and Thornton (2000) and Jones et al. (2002). In this study MarkSim is used to simulate long series of climate in selected grid cells over Honduras. The simulated data is then used as input for the crop simulation model and the long data series’ are used to calculate recurrence probabilities. 201 Step 3: Establishing an activity specific relationship between rainfall and yield While payment is triggered solely by the weather event, this event has to be defined in a way that reasonably represents the likely degree of yield loss. Accordingly, the weather data is transformed by use of a crop simulation model into weighted indices that incorporate the following factors: • Effect of crop phenology: crop phenology influences the sensitivity of final yield to rainfall shortages at specific times. This is taken into account by weighting rainfall during the growing season according to the degree of sensitivity. The heaviest weights are assigned to periods when the crop appears most sensitive to rainfall deficits. • Effect of soil variation: the crop-weather system also exhibits significant interactions with soil water, through the varying ability of soils to store and release rainwater. This is managed through soil-specific rainfall weighting schemes and contracts (hence premiums) to reflect the different levels of risk in different soils. Weather data was transformed into likely crop yield variation using the DSSAT model. Technical details of the model can be found in Tsuji et al. (1994) and Boote et al. (1998). This crop simulation model operates on a daily time step and represents crop growth features such as sensitivity to water, soil, climate and crop management. The soil is represented as a onedimensional profile, horizontally homogenous but consisting of a number of vertical soil layers (Jones et al., 2003). Researchers around the world have used DSSAT for over 15 years (e.g. Alexandrov and Hoogenboom, 2000, O’Neal et al., 2002 and Jones and Thornton, 2003). In this study DSSAT is used to model the growth of dry bean crops in relation to climatic and soil variations. Cultivar and soil input data for the model were obtained from generic databases that accompany DSSAT. For the purpose of this study, we based simulation on the widely adopted cultivar, Rabia de gato+. Step 4: Relating weather indices to yield loss The sensitivity of dry bean yield to rainfall deficits at different times during the growing season was analyzed by modifying the rainfall data input to DSSAT and observing the effect on simulated yield. Ninety-nine years of daily climate data were generated using MarkSim. For each year a likely planting date was estimated based on the temperature and rainfall data. Sensitivity of the crop to rainfall deficits was assessed by comparing the simulated yields with or without ‘droughts’ in 10-day windows throughout the crop cycle. The analysis used ninety-nine years of simulated climate files. Weights were assigned to each of the 10-day windows according to their relative influence on simulated yield. Table 3 indicates the influence of crop stage and soil type on sensitivity. Crops were most sensitive to drought during flowering or early grain fill (days 30 to 50). Sandy soils were more sensitive to short-term drought because of low water holding capacity. 202 Table 3. Influence of soil type and timing of rainfall on yield. Days after planting Crop stage Sensitivity to rainfall expressed as a weight Clay soils Sandy soils Day 1 to 10 Planting / Seedling 0.1 0 Day 11 to 20 Seedling / Flowering 0.2 0 Day 21 to 30 Flowering 0.2 0.2 Day 31 to 40 Flowering / Grain fill 0.2 0.4 Day 41 to 50 Grain fill 0.2 0.3 Day 51 to 60 Grain fill / Maturity 0.1 0.1 Day 61 to 70 Maturity 0 0 Day 71 to 80 Maturity 0 0 Day 81 to 90 Maturity 0 0 Results Basic insurance scheme Figure 2 shows the results of simulation from site SIG. The lower than average yields (bar down) correspond partially with lower than average weighted precipitation (dots). When the weighted precipitation index drops below the rainfall strike, payment is made at a level proportion to the discrepancy, up to a maximum payout. (in this case the maximum payment is $100 for a maximum deviation value of zero weighted rainfall). The premium is related to the average payout. For the strike of a 60% negative deviation from the average rainfall, it is estimated at $1.44/ha. This example used a shallow sand soil for simulation and consequently the strike is probably conservative, since weighted rainfall deficit triggers payment only two times, or 1:50 years. After consultation with users the strike would probably be adjusted to allow more frequent claims, against which premiums would increase. 203 Yield Proportional payout Maximum payout Strike Weighted rainfall Payout amount 100 Premium = 1.44 50 % deviation from the long term mean for this site 0 -50 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 05 01 -100 Years Figure 2. Design of insurance for a shallow sand soil at site SIG (note: only lower than average yields and weighted rainfall are shown). 204 100 Yield Proportional payout Maximum payout Years Strike Years Weighted rainfall Figure 3. Payments and premiums for six sites within the bean growing area of Honduras 205 95 90 85 80 75 70 65 60 55 50 Years 95 90 85 80 75 70 65 60 55 50 45 40 35 Years 45 40 35 SAN Premium = 16.23 30 25 SIGC Premium = 3.77 30 -100 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 % deviation from the long term mean for this site % deviation from the long term mean for this site Payout amount Payout amount SIG Premium = 1.44 25 0 20 -100 15 0 20 50 01 05 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 01 05 -100 10 50 Payout amount 100 01 05 -50 % deviation from the long term mean for this site Payout amount -50 15 100 Payout amount 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 01 05 % deviation from the long term mean for this site 0 10 -50 % deviation from the long term mean for this site Payout amount 50 01 05 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 01 05 % deviation from the long term mean for this site 100 100 SIGB Premium = 5.72 50 0 -50 -100 Years 100 VIL Premium = 8.71 50 0 -50 -100 Years 100 PAR Premium = 8.97 50 0 -50 -100 Site-specific variation of drought risk For the same soil type, the premiums for six sites within Honduras indicate a more than 10-fold variation in risk (Figure 3). For a strike of less than 65% of the average rainfall, premiums varied between a low of $1.44/ha at SIG to a high of $16.12/ha at SAN. An area average scheme would charge a pure premium of approximately $7.50/ha plus whatever was deemed necessary to cover the uncertainty created by spatial variation. Such a scheme would be unnecessarily expensive and it would also encounter the risk of adverse selection. Farmers at location SAN would be much more willing to pay a premium of $7.50 /ha, since they would be more likely to claim more in indemnities than what they would pay in premiums. Conversely farmers at location SIG may consider this premium unreasonable. Soil-specific effects The impact of rainfall deficits is strongly influenced by the water holding capacity of the various soil types. In all cases weighting the rainfall improved the correlation between rainfall variation and simulated yield, suggesting that the modelling process improves the representation of variation of drought effects likely to be experienced on various soil types. In the case of sandy soils, correlations remained at only 35%, even after weighting, illustrating the risk of basing insurance premiums on rainfall alone. The indexing method effectively ‘normalizes’ rainfall on the basis of simulated influence. While this improves the relationship between rainfall and likely yield variation, it also re-scales variation in ways which may call for subsequent adjustment of triggers. Accordingly, strikes may be varied to modify the insurance scheme for soils expected to be of low or high risk. Rainfall indices for clay soils -which have a high available water capacity (AWC)- tend to be conservative. While the correlation between weighted rainfall and simulated yield is good (~60%) - many yield-reducing events miss the trigger because weighting overdampens the influence of drought. In such cases, it would be appropriate to modify the strike to increase the frequency of payment. Silty loams appeared insensitive to short-term rainfall deficits. Indices for growing season rainfall correlated poorly with simulated yield, suggesting that precedent soil moisture is most important for soils with very high AWC. Verification of MarkSim results MarkSim has been validated using standard statistical tests, including the ability to simulate monthly averages, variance and wet and dry spell persistence. (Jones and Thornton, 1993). In this case MarkSim was verified by comparing simulated crop specific weighted rainfall data with long-term observations for two sites where data is available. 74 years of historical precipitation data was obtained for Palmira (Valle, Colombia) and 40 years for Guatemala City (Guatemala). The same number of years were simulated for the corresponding locations. Data was compared for the bean growing season only, based on the likely planting date for the primera planting season. This was estimated for both observed and simulated data sets by assuming that planting took place on the first day following a five-day cumulative rainfall of 50mm between the 1st of March and the 31st of June. Once the planting date was identified the precipitation-weighting scheme was applied to both modelled and observed data and the cumulative frequency distribution curve was obtained as shown in figure 4. MarkSim appears to simulate well the frequency of low weighted rainfall events for Palmira. The Kolmogorov Smirnov statistic does not indicated a significant difference between sample populations (DeGroot, 1975). Closer inspection however reveals that some discrepancies occur at lower rainfall values that are likely to be of importance to insurers. Taking the weighted precipitation value of 15mm or less for 206 Palmira, MarkSim data suggests that this has a probability of approximately 0.1 (i.e. once every ten years) however the observed data reveals a probability of approximately 0.15 (i.e. once every six or seven years). Similarly the data for Guatemala City illustrate that there are marked discrepancies in the cumulative frequencies of the low values. The simulated frequencies are generally higher than the observed. The MarkSim data suggest that a weighted rainfall value of 40mm or less has a probability of approximately 0.12 (once every eight years) whereas the observations illustrate that it is more likely to be a probability of 0.25 (once every four years). This is bad news for an insurer relying on MarkSim to set premiums. It will almost certainly mean that the uptake of the scheme by farmers will be well accepted but that premiums will eventually rise. More work is required to attach error values to MarkSim generated weather data for each location; this would then allow insurers to reflect the level of accuracy in the site specific premium prices. Furthermore verification of the low yield years identified by the weighted rainfall index is necessary, for this corresponding observed yield data is required. 207 1 0.9 Palmira observed Palmira simulated Cumulative frequency 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50 60 70 80 90 100 Weighted rainfall value (mm) 1 0.9 Guatemala City Observed Cumulative frequency 0.8 Guatemala City simulated 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 20 40 60 80 100 120 140 160 Weighted rainfall value (mm) Figure 4. Comparison of simulated with observed rainfall frequencies for Pamira (top) and Guatemala City (bottom) Comparison of simulated with observed rainfall frequencies for Guatemala City. 208 Conclusions and suggestions for further work This study demonstrates the potential application of weather generation and crop simulation models to design site-specific and soil-specific drought insurance for less developed countries. Such areas are unlikely to possess sufficient historical data on which to base conventional methods of insurance. The results can be included in weather index-insurance schemes that relate trigger events, indemnities and premiums according to best estimates of drought frequencies and their effects. This method could be repeated for any site in the tropics and for any crop for which DSSAT has been validated. High levels of basis risk are a major source of uncertainty in weather-based insurance. This research demonstrates that spatial variation can introduce substantial basis risk even within the relatively short distances across the bean growing areas regions of Honduras. Sample premiums varied at least ten-fold between sample sites within the bean growing area of Honduras. Insurance schemes that do not include this variation in their estimates expose both insurer and insured to unnecessary serious basis risk. Additional basis risk was identified due to the interaction between climate and soil, in particular due to the different water holding capacities of soils. Although this exploratory study has illustrated how use of simulation models may be able to address some of the problems outlined in the introductory review, many challenges remain. These include: • Further verification of model output: for an insurance product based on simulated data the issue of verification if of great importance, and further verification is needed of the accuracy of MarkSim to accurately simulate frequency probabilities. • Assigning confidence levels to simulated data: future work is required to assign confidence values to account for uncertainties in model estimates. • Mechanism to update premiums during mid-season: it has been suggested that weather index insurance should be sold at the latest up to two weeks before the crop cycle begins. This may prove to be a major limitation for farmers, who may not have funds available to purchase insurance before the crop cycle. Mid-season premiums may be preferred for which an appropriate statistical method of updating premiums based on prior events would need to be identified. However, ENSO signals can be strong in Central America and this challenge can create serious intertemporal adverse selection problems. Weather risk may need to be conditioned based upon the ENSO signal. This could significantly increase the premium rates. • Assessing farmer preferences for insurance contract design: farmer preferences regarding contract design, trigger selection, premium cost, indemnity payments, and distribution need to be ascertained and incorporated into the final product. • Distribution methods: the distribution of such index-insurance directly influences the impact the scheme has on poverty alleviation. There is a need to carefully investigate and design a method for offering and distributing the insurance so that it has the greatest impact on poverty alleviation and that if possible to organize the insurance so that it can be used to promote rural development and adoption of progressive management techniques. 209 References Ahuja, R. and Jutting, J. (2004) ‘Are the poor too poor to demand health insurance?’, Working paper no. 118. Indian Council on International Economic Relations, New Delhi, India [Available online at: http://www.icrier.res.in/wp118.pdf]. Alexandrov, V.A. and Hoogenboom, G. (2000) ‘The impact of climate variability and change on crop yield in Bulgaria’, Agricultural and Forest meteorology, Vol.104, pp.315-327. Bird, K., Hulme, D., Moore, K. and A. Shepherd. (2002) Chronic Poverty and Remote Rural Areas, ODI, London. Black, J. R., Barnett, B. J. and Hu, Y. (1999). ‘Cooperatives and Capital Markets: The Case of Minnesota-Dakota Sugar Cooperatives’, American Journal of Agricultural Economics Vol.81, pp 1240-46. Boote, K.J., Jones, J.W. Hoogenboom, G. and Pickering, N.B. (1998) ‘The CROPGRO model for grain legumes’, in Tsuji, G.Y., Hoogenboon, G. and Thornton, P.K. Understanding options for agricultural production, Kluwer, London. pp.99-128. Brown, W. and Churchill, C.F. (1999) ‘Micro-Insurance: Providing Insurance to Low-Income households Part 1 Working draft’. [Available online at:https://amis.ai.mvirtconf.nsf 0/0E16A7927944086E8525695E0073DB3D/$File/Micro-InsurancePart1.pdf?OpenElement, accessed March 2004]. Bryla, E., Dana, J., Hess, U. and Varangis, P. (2003) ‘The use of price and weather risk management instruments’, International conference report: Paving the way forward for rural finance. [Available online at: http://www.basis.wisc.edu/ live/rfc/cs_03a.pdf, accessed March 2004]. CEPAL (2003) ‘El impacto socioeconómico y ambiental de la sequía de 2001 en Centroamérica’. [Available online at: http://www.ccad.ws/documentos/sequia /indiceyportada.pdf, accessed April 2004] DeGroot, M.H. (1975) Probability and statistics, Addison-Wesley, Massachusetts. Dercon, S. (2002) ‘Income risk, coping strategies and safety nets’, Center for the Study of African economics, Oxford University, Oxford. [Available online at: http://www. wiwiss.fuberlin.de/w3/w3collie/SocPolWS03/DerconSurvey.pdf, accessed April 2004]. Dercon, S. (2005) Insurance against poverty, Oxford University Press, Oxford. Goes, A. and Skees, J.R. (2003) ‘Financing natural disaster risk using charity contributions and exante index insurance’. [Available online at: http://www.globalagrisk.com /pubs/Charity%20Cat%20Bonds.pdf, accessed March 2004]. 210 Hazell, P., Larson, D., Mac Isaac, D. and Zupi, M. (2000) ‘Weather insurance project study: Ethiopia’, The World Bank/ Fondazione/Salernitana Sichelgaita/ IFPRI, Washington. Hess, U. (2003) ‘Innovative financial services for rural india: Monsoon-Indexed Lending and insurance for small holders’, Agriculture and Rural development working paper 9, The World bank, USA. [Available online at: http://www.itfcommrisk.org/documents/weather/India.pdf, accessed March 2004]. Jaffee, E.M. and Russell, T. (1997) ‘Catastrophe Insurance, capital markets and uninsurable risks’, The Journal of Risk and Insurance, Vol.64, pp.205-30. Jones, J.W., Hoogenboom, G., Porter, C.H., Boote, K.J., Batchelor, W.D., Hunt, L.A., Wilkens, P.W., Singh, U., Gijsman, A.J. and Ritchie, J.T. (2003) ‘The DSSAT cropping system model’, European Journal of Agronomy, Vol.18, pp.235-65. Jones, P.G. and Thornton, P.K. (1993) ‘A rainfall generator for agricultural applications in the tropics’, Agriculture and forest meteorology, Vol.63, pp.1-19. Jones, P.G. and Thornton, P.K. (1999) ‘Fitting a third order Markov rainfall model to interpolated climate surfaces’, Agricultural and forest meteorology, Vol.97, pp.213-231. Jones, P.G. and Thornton, P.K. (2000) MarkSim: software to generate daily weather data for Latin America and Africa, Agronomy Journal, Vol.93(3), pp.445-53. Jones, P.G. and Thornton, P.K. (2003) ‘The potential impacts of climate change on maize production in Africa and Latin America in 2055’, Global environmental change, Vol. 13, pp.51-9. Jones, P.G., Thornton, P.K., Diaz. W. and Wilkens, P,W. (2002) MarkSim: a computer tool that generates simulated weather data for crop modelling and risk assessment, CIAT, Cali. Manojkumar, K., Sreekumar, B. and Ajitkumar, G.S. (2003) ‘Crop Insurance Scheme: A case study of banana farmers in Wayanad district’ Discussion Paper No.34, [available at: http://krpcds.org/ publication/downloads/54.pdf, accessed November 2004] Miranda, M.J. and Glauber, J.W. (1997) ‘Systematic risk, reinsurance and the failure of crop insurance markets’, American Journal of Agricultural Economics, Vol.79, pp.206-15. Morduch, J. (1999) ‘Between the state and the market: Can informal insurance patch the safety net?’, The World bank research observer , Vol.14(2), pp.187-207. O’Neal, M.R., Frankenberger, J.R. and Ess, D.R (2002) ‘Use of CERES-Maize to study effect of spatial precipitation variability on yield’, Agricultural systems, Vol.73, pp.205-25. Pandey, S., Behura, D., Villano, R. and Naik, D. (2001) ‘Drought risk, farmers’ coping mechanisms, and poverty: a study of the rainfed rice system in eastern India’ in Peng, S. 211 and Hardy, B. (eds.), Rice research for food security and poverty alleviation, International Rice Research Institute, Philippines, pp.267-274. Rezende Lopes, M. and Leite da Silva Dias, G. (1986) ‘The Brazilian Experience with Crop Insurance Programs’, in Hazell, P., Pomareda, C. and Valdez, A., Crop insurance for agricultural development: issues and experience, John Hopkins University Press, Baltimore, pp 240-62. Rosenzweig, M.R. and Binswanger, H.P. (1993) ‘Wealth, Weather Risk and the Composition and Profitability of Agricultural Investments’, Economic Journal, Vol.103(416), pp.56-78. Sakurai, T. and Reardon, T. (1997) ‘Potential demand for drought insurance in Burkina Faso and its determinants’, Americ Journal of Agricultural Economics, Vol.79, pp.1193-1207. Skees, J. (2000) ‘A Role for Capital Markets in Natural Disasters: A Piece of the Food Security Puzzle’. Food Policy, Vol. 25, pp. 365-378. Skees, J. R. (2003) ‘Risk management challenges in rural financial markets: blending risk management innovations with rural finance’, Thematic Paper Presented at Paving the Way Forward for Rural Finance: International Conference on Best Practices., U.S. Agency for International Development, USA., [available at: http://www.basis.wisc.edu/live/rfc/theme_risk.pdf, accessed March 2004]. Skees. J., Gober. S., Varangis. P., Lester. R. and Kalavakonda, V. (2001) ‘Developing a rainfallbased index insurance in Morocco’, World Bank policy research working paper 2577. Skees. J., Hazell. P. and Miranda. M. (1999) ‘New approaches to public / private yield insurance’, EPTD Discussion Paper No. 55, International Food Policy Research Institute. Skees. J., Varangis. D., Larson. D. and Siegel. P. (2005) ‘Can financial markets be tapped to help poor people cope with weather risks?’, in Dercon, S (ed.), Insurance against poverty, Oxford University Press,WIDER Studies in Development Economics. Stoppa, A. and Hess, U. (2003) ‘Design and use of weather derivatives in agricultural policies: the case of rainfall insurance in Morocco’, International conference Agricultural policy reform and the WTO: where are we heading?, Italy. Tsuji, G.Y., Uehara, G. and Balas, S. (1994) DSSAT version 3 volume 2, University of Hawaii, Hawaii. Turvey. C.G. (2001) ‘Weather derivatives for specific event risks in agriculture’, Review of Agricultural Economics, Vol.23(2), pp.333-351. Varangis, P. (2001) ‘Innovative approaches to cope with weather risk in developing countries’, The climate report, Vol.2(4). [available at: http://www.guaranteedweather .com/page.php?content_id=74, accessed March 2004]. 212 Webb, P. and Reardon, T.A. (1992) ‘Drought impact and household response in East and West Africa’, Quarterly Journal of International Agriculture, Vol.31(3), pp.230-246. Wenner, M. and Arias, D. (2003) ‘Agricultural insurance in Latin America: Where are we?, International conference report: Paving the way forward for rural finance’, Inter-American Development Bank, Washington, [available at: http://www.iadb.org sds/doc/RURAgroInsuranceWennerArias.pdf, accessed March 2004]. 213 Risk sharing through insurance: Weather indices for designing micro-insurance products for poor small-holder farmers in the tropics Díaz Nieto, J. a, b, Cook, S.E.a, Lundy, M. a, Fisher, M.c, Laderach, P.a a Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia b The University of Sheffield, Sheffield, United Kingdom. c Independant Consultant, Cali, Colombia Abstract Agriculture is inherently risky. Drought is a particularly troublesome hazard that has a documented adverse impact on agricultural development. A long history of decision-support tools have been developed to try and help farmers or policy makers manage risk. We offer sitespecific drought insurance as a significant addition to this process. Drought insurance works by encapsulating the best available scientific estimate of drought probability at a site within a single number- the insurance premium, which is offered by insurers to insurable parties in a transparent risk-sharing agreement. The proposed method is demonstrated in a case study for dry beans in Nicaragua. Keywords: Poverty, drought, dry beans, MarkSim, DSSAT, micro-insurance, indexed insurance. Introduction Agriculture is inherently risky. A review of rural poverty identified exposure to risk as a major modifiable reason for chronic poverty, noting the widespread evidence that correlates risk with poverty (Bird et al., 2002). Production risks include, but are not limited to climatic hazard, which of all the hazards agriculture faces is perhaps the most difficult one for agriculturalists to manage. Drought is the most serious of the natural hazards globally in terms of loss of life, accounting for 44% of reported deaths in the period 1974-2003 (EM-DAT, 2004). The mere expectation of drought is sufficient in some cases to reduce agricultural production. Nearly 80% of farmers interviewed in Ethiopia cited harvest failure caused by drought and other natural hazards as the event that caused them most concern (Dercon, 2001). Pandey et al. (2001) revealed a huge drop in income for rice farmers in Orissa state in India as a result of drought. This work is substantiated further by experience from more recent droughts in the region. The impacts of drought extend beyond the loss of production. Sakurai and Reardon (1997) include increases in local interest rates due to a rise in households seeking credit, a decline in farm labor demand, a reduction in local wages due to greater numbers seeking off-farm employment, drops in livestock prices due to distress sales of livestock and increases in food prices coinciding with low financial resources. In this paper we attempt to set the stage for providing crop insurance as a mechanism for poor smallholder farmers to cope with drought. We firstly review the relation between drought and poverty and then explore some of the issues involved in spreading risk by means of crop insurance and the issues of making it available to poor smallholders. We go on to summarize 214 recent developments in spatial information and insurance products. We follow with a discussion of how pseudo-historical weather may be generated to substitute for inadequate, unreliable or insufficient data as input to crop simulation models to derive relationships between specific weather events and crop yield. We briefly illustrate this with a case study or smallholder producers of drybeans in north-central Nicaragua. We conclude by discussing the hazards in designing crop insurance instruments with emphasis on poor smallholder farmers. Drought, risk and poor smallholder farmers Drought is a widespread and common natural hazard Although drought is the major cause of crop loss throughout the semi-arid tropics, in this article we shall focus on Central America, with reference to our work to design a drought index on which to base an insurance product for poor bean farmers (Díaz Nieto, 2006 ). Drought is an especially serious problem for small-scale producers, most of whom do not have access to irrigation, for example, in Nicaragua only 8% of the land is irrigated (World Bank, 2001), and almost none of this is in the central-north region where most poor bean growers are located. Droughts cause food and income insecurity through both acute effects and chronic secondary effects. Acute effects are immediate crop failure, which in extreme cases leads to hunger and even starvation. Secondary consequences of drought include increases in local rates of interest due to an increase in the number of households seeking credit and a decline in the demand for farm labor leading to a reduction in local wages due to greater numbers seeking off-farm employment. Livestock also suffer hunger and starvation leading to falling prices due to distress sales. Food prices increase coincidental with falling financial resources available to rural households as sources of income dry up (Sakurai and Reardon, 1997). The rural poor are often, indeed usually, found on lands that are marginal for one reason or another, such as low fertility soils, steep slopes and remoteness. They are especially vulnerable to drought. Large numbers of people are affected. Numerous studies have shown a strong link between risk, vulnerability and poverty (Rosenzweig and Binswanger, 1993; Mosely and Krishnamurthy, 1995; World Bank, 2000; Dercon, 2001). Poor households lack resources with which to absorb the shocks of natural hazards. Even small disruptions in the flow of income can have serious implications for them, so poor farmers commonly use informal and self-insurance measures to avoid risk. As discussed in more detail below, while these measures can help survival (e.g. Webb and Reardon, 1992), most studies conclude that they are not the most effective tools for risk management, since they reduce the impact of a hazard at the expense of more profitable activities (Morduch, 1995; 1999; Barrett et al, 2001). The poorest regions will be most affected by global change The Third Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) concluded, inter alia, that climate change due to increasing global temperatures would disproportionately affect the poor, who have scant resources to adapt to its effects (Watson et al., 2001). In particular, the El Niño-La Niña phenomenon in the western Pacific is expected to become increasingly frequent and severe. During El Niño events, rains in Central America are 215 much below average so that droughts will become more common and intense in the face of climate change. Many marginal lands in the tropics are used in some form of slash-and-burn management. As populations increase, the slash-and-burn cycle becomes more frequent and typically harvests decrease markedly caused by soil erosion, nutrient depletion, and weed invasion. These problems will become more acute in the face of climate change. Risk and Insurance Strategies for coping with risk and their effects on livelihoods Most of the modern risk-avoidance measures are not readily available in developing countries, hence farmers in these regions are obliged to adopt traditional informal risk coping mechanisms (Wenner and Arias, 2003) (Table 1). Table 1. Risk management tools Self insurance measures Crop diversification Maintaining financial reserves Reliance on off-farm employment Other off-farm income generation Selling family assets (e.g. cattle) Avoidance of investments in expensive processes such as fertilizing (especially in high-risk years) Accumulation of stocks in good years Removal of children from education to work on farm Modern risk avoidance measures Production contracting Marketing contracting Forward pricing Futures options contracts Leasing inputs Invest in fertilizer, use long-term forecasts Acquiring crop and revenue insurance Custom hiring (Source: Wenner and Arias, 2003; Skees et al., 2001; Hess, 2003) Many argue that informal self-insurance measures are a barrier to poverty alleviation and indeed reinforce poverty (Rosenzweig and Binswanger, 1993; Brown and Churchill, 1999; Barrett et al., 2001). The general effect is that traditional risk-coping mechanisms not only sustain poverty but actually hinder development. They do this because risk-averse strategies firstly generally use resources inefficiently and secondly fail to exploit more productive investments and technologies that in the long term would result in more productive systems (Hazell et al., 2000; World Bank, 2001). For example, when faced with the possibility of losing an entire crop due to drought, farmers may lessen risk by minimizing investment in the crop by not applying fertilizer. They do this because making the additional investment increases their loss should the crop fail. Likewise, selling family assets such as cattle at a time when everyone else is also trying to sell their assets will lower prices. As a result, such assets are of little use in smoothing the effect of the drought shock. Worse, if the asset was bought at a time when prices were buoyant, as in a time of plenty, selling will incur a net loss (Skees, 2003). Furthermore if an animal dies of starvation, all investment in it is lost. 216 Another common risk-coping mechanism is seeking off-farm income. This may be effective for idiosyncratic risks, but the tactic is less effective when a geographically-extensive risk event, such as drought, which is typically wide-spread, occurs. This is because the amount of labor on offer increases so that conditions become more competitive and wages fall. Moreover, as economic conditions worsen, the amount of work available typically lessens as employers seek to cut costs. Informal insurance is therefore a relative ineffective strategy to cope with covariant risk events, such as drought. Repeated shocks further undermine it as a coping strategy (Dercon, 2003). A survey in India found that 30% of respondents cited loss of wages, income or work as the major impact of a risk event (Hess, 2003). Forty-five percent said that they would borrow money to tide them over the crisis, leading to increased indebtedness. In reality, the option to smooth consumption by borrowing is generally not available to small-holder farmers with low incomes. Financial institutions are unwilling to lend to these borrowers precisely because of their vulnerability to drought and the consequent likelihood of default on loan repayments (Hess, 2003). Indian banks, who lend to farmers in irrigated areas, are constrained by the risk of drought from extending credit to farmers in non-irrigated areas (Mishra, 1994). Goes and Skees (2003) argue that ex post disaster-relief plans can have unintended negative impacts on economic development. In the worst-case scenario ex post relief can increase risk exposure in the long-term by promoting dependence on charitable relief. In addition, government assistance has to be very careful not to encourage new economic activity in areas that are unreasonably vulnerable to natural disasters (Skees et al., 2001). Risk sharing through insurance is an option but has traditionally not been available to the poor. The purpose of formal risk-management strategies is to enable investment in more profitable activities through transparent sharing of risk. For the reasons explained above, the informal riskaversion mechanisms that poor households use mean that they are unlikely to invest in new technologies that could lead to increased wealth. For this reason, poor people exposed to risk find it difficult to break out of the poverty cycle. Formal insurance has provided benefits to individual consumers for centuries and in the last few years has also been suggested as a pro-poor tool for managing risk (van Oppen, 2001). A growing number of micro-insurance products (products offered to insure items in the range of a few hundreds of dollars) are now being offered in poor countries in the areas of life, health and property insurance and in some cases, schemes for crop insurance. This growing interest in micro-insurance products as development tools is associated with the expansion of micro-credit schemes (Morduch, 1999). There is also a growing recognition of the mutual benefits of risk management as a tool for poverty alleviation. Micro-insurance is not only justified on the basis of humanitarian need. Properly designed, it also makes economic sense for the organization offering it (Dercon, 2003). Micro-insurance is one of a number of products that can be sold under the collective title of micro-finance and an initial question to consider is whether insurance is the most appropriate of 217 these tools to address weather risks (Brown et al., 2000). Insurance needs to be evaluated against other tools such as savings, mutual plans or credit. Formal strategies such as insurance are most effective where there is a high degree of uncertainty and when there is a lot to lose (Brown and Churchill, 1999; Zupi, 2001) (Figure 1). Weather risks naturally fall into this category. The uncertainty is large because long-term weather forecasting is as yet imprecise. Moreover, the level of loss can be severe because a severe drought may lead to the entire failure of the crop. Figure 1. Illustration of the potential for the insurance in managing situations where there is high uncertainty and a lot to lose. Informal strategies become less effective in these situations. (Source: Brown and Churchill, 1999) Insurance can be thought of as exchanging the irregular uncertainty of large losses for regular small premium payments. A general rule of thumb seems to be that the larger the proportional loss in assets and income to the household, the fewer alternatives there are to recover from the loss (Brown and Churchill, 1999). Insurance is one of the few viable options for poor people to manage uncertain events that can cause large losses. There are few examples of micro-insurance on which to assess its impact. Documented microinsurance schemes operating in poor countries have in general reported encouraging results. In an empirical study of a crop insurance scheme, Mishra (1994) found that although the scheme was not financially viable, it provided many socio-economic benefits for both farmer and the insurer. Farmers benefited from insured production, which led to increased investment and wealth, while the insurers benefited from a broader base of creditworthy customers. If the financial weaknesses of crop insurance in developing countries can be overcome (Bryla et al., 2003), these socio-economic benefits could flow more generally amongst poor small-holder producers. 218 Previous experience with insurance has not been good Although we have made the case for crop insurance above, crop-insurance schemes in general in the tropics have a sorry record (Skees et al., 2001). Several governments have developed crop insurance schemes. To date, most agricultural insurance has been either fully publicly owned or has involved large government subsidies to schemes operated by private companies. Unfortunately most of them have failed. The main reason for failure of publicly-owned insurance schemes is because they were either multiple-peril or all-risk programs (Skees et al., 2001). This means that virtually any cause of crop failure has been insured, resulting in excessive indemnity payments. It also results in moral hazard, which is when there is no incentive for the insured to use the best possible practices to avoid yield loss. A second problem is that risks are widely correlated or systemic, that is a weather risk event affects many crops at the same time over an extensive geographic area (Miranda and Glauber, 1997). Related to moral hazard is asymmetrical information, which is when the insured knows more about the risk of crop failure than the insurer. Skees (2003) believes that the problem of hidden and asymmetrical information is the underlying cause of failure of many schemes. A corollary is that a successful scheme requires symmetrical information, where both the insurer and the insured have equal understanding of the risk. A further key factor for a successful scheme is to overcome systemic risk, which requires some form of long-range risk-sharing mechanism such as re-insurance There are several further problems that have led to the failure of insurance schemes in the past, and which a successful scheme must avoid: • Adverse selection, where farmers facing lower-than-average risks opt out of the scheme leaving only farmers with higher-than-average risk. • Public insurers are often mandated to extend their insurance cover to small farms and this can add enormously to administration costs. • When insurers know that the government will automatically cover most losses, they have had little incentive to pursue sound insurance practices when assessing risks, a version of moral hazard. • Corruption has been a problem. Examples include inspectors receiving bribes averaging 30% of the value of the indemnity payment to the farmer. Governments have also undermined public insurers for political reasons. Recent developments in provision of spatial information and insurance products Weather micro-insurance has been proposed as a viable tool to help poor farmers manage weather risk, which translates into crop production risk. The principles behind weather insurance have been widely discussed (Skees et al., 2001; Bryla et al., 2003; Hess, 2003; Stoppa and Hess, 2003; Varangis et al., 2003). A review of the principles and experience of the insurance processes follows. 219 Principles of weather insurance Two broad principles govern the viability of insurance. The first is that risk-sharing can only occur when both parties (the insurer and insured) have accurate information about a hazard and its likelihood. This has been the basis of insurance for over three centuries and Skees (2003) maintains that a sound weather insurance product is transparent and symmetrical, so eliminating both moral hazard and adverse selection. The second requirement is that the risk sharing must be broad enough to overcome co-variate risk (the risk that all crops insured in a scheme are affected), given that major weather events typically have broad geographic coverage. The probabilities of occurrence of adverse weather event that reduce crop yield can usually be estimated from historical weather data. However, some areas are riskier than others. In an insurance scheme the probability of occurrence must be identified for specific areas and be agreed by both parties (symmetry of information.) Crop yield indices are a relatively new method for insurance products that have been applied as area average indices (Skees et al., 2001). Indemnity payments are made to policy-holders when the area-average yield for a particular season falls below a predetermined long-term area average. The index in this case is some percentage of the long-term area average yield. The scheme is in operation in USA, India, Sweden, Mongolia and Quebec in Canada. Although area average indices avoid the traditional problems of adverse selection and moral hazard, they may not be appropriate for developing countries where long and reliable yield data are not available (Skees et al., 2001). Moreover, yield data for developing countries are normally for research stations. Not only may the research station location not be representative of the area as a whole, but research station yields are known to overestimate farmers' yields by 30% or more (Davidson, 1965). Insurance based on weather indices is another relatively recent development, in which weather events, not yield, are the basis for determining indemnity payment. In the 1990s weather markets started developing in North America, mainly as a result of the privatization of the energy sector in the USA. Producers sought to manage revenue fluctuations associated with weather variation (Turvey, 2001) by means of both the derivatives and insurance markets. Agricultural applications appeared as a spin-off from these markets, since many of the weather risks of concern to the energy sector also affect the agricultural sector through crop losses. Compared to area-average indices, weather-based indices have the advantage that weather data are generally more accessible and reliable than yield data. This is especially the case in developing countries (Skees, 2003). Weather-related crop insurance products succeed or fail on their ability to present accurate information about weather-related risks that are specifically associated with yield loss. The critical step is to identify the relationship between an insured weather event and consequent crop loss. A key attribute of weather-based index insurance is its simplicity and transparency, which not only increase the products' profitability, but also makes them more attractive to global insurance markets (Miranda and Vendenov, 2001). Weather-index insurance also provides a hedge against the cause of the yield loss, rather than its cost, which is the underlying concept of insurance 220 against yield reduction. This removes the need to estimate prices (Turvey, 2001; Skees et al., 2001), a critical component of many of the traditional yield-triggered insurance schemes. Methods of drought insurance The main challenge in developing weather insurance: Basis risk A summary of the main challenges involved in developing good weather insurance schemes are summarized in Table 2 and discussed further below. Table 2. Summary of main challenges that need to be addressed and possible areas of action Challenge Basis risk Precise actuarial modeling Reinsurance1 – without reinsurance correlated weather insurance is likely to fail. Security and dissemination of measurements, the insurance will ultimately depend on the objectivity and accuracy of the measurement. Education – customers may not understand this new generation of products Marketing – for the product to be successful it is critical to think carefully about how, when and where the insurance product is sold. Including marketing on a higher level i.e. reinsurance markets. Payment of the premium – Expecting the poor to pay a premium could be quite difficult Possible solution Careful design of index insurance parameters Selling via micro finance institutions that understand the risks. Requires historical data and actuarial models If no data is available then this is where weather generators come in useful Reinsurance. CAT-bonds2 Install tamper proof rainfall stations Need some education to help customer assess whether it will benefit them or not Insurance as a component of MFI loans. Charities may have a role to play (Goes and Skees, 2003). For example charities are always ready to provide support after a disaster – what about providing support before the disaster? Or the insurance cover could be purchased by the charity and the indemnity also be administered by them. 1 Reinsurance is where the primary insurer covers its exposure to a given risk with a third party. Reinsurers typically count on enormous financial resources and spread risk from narrowly-based insurers to a broader, often global, basis. Well-known reinsurers include Munich Re in Germany, Swiss Re in Switzerland and Lloyds in the UK. 2 Catastrophe bonds, commonly called CAT-bonds, are financial derivatives used by insurance companies to hedge their exposure to catastrophic risk. They usually pay a higher interest rate than the premiums charged by reinsurers to cover the same risk. In the event that the trigger is met, often that the issuing company’s payout for a specified catastrophe is exceeded, the capital is “forgiven”, that is the bonds become worthless. CAT-bonds are most often issued to cover exposure to risks of earthquakes and hurricanes. (Source: Skees, 2003) 221 The greatest challenge facing weather-based insurance products is basis risk (Miranda and Vedenov, 2001;Skees et al., 2001; Turvey, 2001; World Bank, 2001; Skees, 2003). Basis risk occurs when the insurance index does not accurately represent loss: a weather index may not trigger a payment when there has indeed been a loss; or payment may occur without serious loss. The insurance product will not be attractive to potential customers if they think that the basis risk is too high (Skees et al., 2001). A feasibility study of rainfall indices for Nicaragua concluded that even within departments a single index did not adequately represent the spatially variability of risk (World Bank, 2001). In each department there was at least one weather station where the data were markedly different from the others. A short study by Diaz-Nieto et al (2006a) using simulated data for Honduras also revealed that a single weather index was not appropriate for a country the size of Honduras. Basis risk is caused by the need to model complex heterogeneous systems within a single index. There are three sources of basis risk (Table 3). Table 3. Summary of main challenges that need to be addressed and possible areas of action. Basis risk Details Solutions Temporal risk The level of impact of a weather phenomenon will vary according to the time at which it occurs during the crop cycle. E.g. a shortage of rainfall at just before maturity may kill a crop, whereas just after seedling may have little effect. A rainfall deficiency may occur at one location causing crop losses, but this rainfall deficiency did not occur at the recording location and so no payment is triggered. A rainfall deficiency may kill a drought sensitive crop, whereas a drought resistant crop will survive through longer periods of drought. Indices that represent the temporal variability in sensitivity to rainfall deficit. Spatial risk Crop specific risk Offset the risk by offering sitespecific contracts that account for spatial variability. Offset the risk by tailoring the insurance to specific crops. (Source: World Bank, 2001) Specialized contracts can be designed to offset much of temporal, spatial and crop-specific basis risk (Miranda and Vedenov, 2001). However, doing so may increase administrative costs and, more importantly, increase the complexity involved in marketing and distribution. An alternative to overcome basis risk is a larger number of standard contracts that cover all possibilities and priced accordingly, and allow the insureds to select the contract they consider most appropriate (Turvey, 2001). Establishing the correlation between crop yield and rainfall index The fundamental requirement of a rainfall index is that rainfall must explain a large proportion of the variability in yield (Skees et al., 2001; Turvey, 2001; Skees, 2003; Stoppa and Hess, 2003). As a first step, it is essential to establish the cause and effect relationship (Turvey, 2001), so that the index represents critical rainfall deficits that account for crop yield losses. It is not sufficient, for example, to posit that a rainfall deficit of 30% of the long-term average will trigger payment because this provides no information about the timing of rainfall in relation to crop demands at different growth stages. 222 Defining the weather events that cause the most serious yield losses and that cover as many of the loss-causing events as possible requires a considerable investment in research (Skees et al., 2001). Furthermore it is critically important that both parties agree that the weather index adequately explains the variability in crop yields (Stoppa and Hess, 2003). Few customers would be inclined to purchase insurance that they did believe protected them against risk. Limited availability of yield and climate data on which to base indices Stoppa and Hess (2003) suggested that to develop effective weather-index insurance the weather variable must not only be measurable but adequate historical weather records must be available from which to estimate probabilities of a risk event occurring. In spite of this, many of the feasibility studies into the use of weather-based indices in developing countries provide indices based on relatively few data. Reliable long-term datasets of weather in developing countries are very limited and this presents a major potential challenge. It is noteworthy that countries with poor infrastructure are precisely those places where an effective insurance product could have most impact. The danger is that poor regions, which have greatest demand for insurance, are those which are excluded, precisely for reasons of poor infrastructure associated with poverty. An alternative approach, which we describe below, is to use statistical models and process-based simulation models, based on decades of scientific analysis, to generate ‘pseudo-historical’ data of climate and yield. Where possible these pseudo-historical data can be complemented with such weather data as are available. Methodologies to obtain weather indices There are five methods by which a relation might be established between climate and the yield of a particular crop taxon at a given site. In this context, "site" means the area to which the available weather data may reasonably be applied. The size of the site will obviously depend heavily on the topography. Long-term crop yield and weather data Where sufficient data are available this is clearly the best option. Nevertheless, some cautions are necessary. Firstly, crop yields have been steadily increasing in industrialized agriculture for at least the last 60 years with improved germplasm, mainly by changing harvest index (the proportion of total plant yield that is partitioned to the harvested product, Gifford et al., 1984). This had been made possible by improved agronomy such as the precision placement of fertilizer so that the crop plants were not required to scavenge for nutrients to the extent that they did formerly. Secondly, there is incontrovertible data that show that the global climate has changed and is continuing to change (Watson et al., 2001). The last decade has been the hottest since formal temperature records were started in the 1850s, with five of the last six years the hottest ever recorded. Any serious analysis will take account of both factors to the extent possible. One possibility is to use the historical weather data to generate pseudo-historical yields for the current germplasm using an appropriate crop simulation model. Weather data generated from historic data and used as input to crop simulation models Where reliable data are available for a site of temperature (monthly maxima and minima, or mean monthly temperature and mean daily range) and mean monthly rainfall, the weather generator MarkSim (Jones et al., 2002) may be used to generate as many years' data as are 223 necessary (up to 4999 sets each of 99 years) of maximum and minimum temperatures, solar radiation and rainfall. Briefly, MarkSim uses a Markov method (sometimes called a Markov chain) to generate weather data. The Markov method is a statistical technique used to describe a time-series of discrete states. For rainfall, the states are either a day in which no rain falls, or a day in which it rains. The Markov method determines the state on any particular day based on the states of the either the previous day or a sequence of previous days. For tropical systems, where the weather is controlled by convective circulation rather than the west to east movement of frontal systems as in temperate climates, a third order model, that is three consecutive days, are required to represent weather satisfactorily (Jones and Thornton, 1997). The generated data can then be used as input to the appropriate simulation model for the crop of interest. Weather data generated according to location and used as input to crop simulation models When no reliable weather data are available, it is possible to use MarkSim to interpolate on a multi-dimensional surface for weather derived from more than 20 000 sites throughout the tropics. A limitation of this approach is that the pixel size is 10 arc minutes a side, about 18 km at the equator. This creates some limitations for very mountainous regions in that the generated weather is a mean distribution for the pixel. It may not be representative of the weather for the extremes of altitude that lie within it. We used this methodology to generate weather indices for bean farmers in north-central Nicaragua. We report on this in more detail below. Crop sensitivity determined by expert opinion and probabilities generated by some method of interpolation of those weather data that are available. This method is considerably less reliable than the methods discussed above, but in some circumstances it may be all that is available. Be that as it may, any scheme based on such an unscientific approach must be viewed with considerable suspicion and due caution used in its application to generate an insurance product. Crop sensitivity determined by expert opinion from similar sites and probabilities generated by some method of interpolation of those weather data that are available. Sites that are homoclimatic can be selected with the Homologue procedure (Jones et al., 2005). This method suffers from the same drawbacks as above, but with the added uncertainty of applying expert knowledge from another place. Payout index highly correlated with yield loss In a weather insurance scheme it is not the actual crop loss that is insured but the loss-causing event, which in this case is a specified adverse weather event. Therefore the way in which the relationship between weather and crop losses is expressed in an insurance index needs to be carefully thought out and appropriately designed. A producer will be interested in a weatherinsurance scheme that is highly likely to pay out when (s)he does indeed suffer a crop loss. Ideally the relationship between weather and crop yield can be extracted from long historical records of both. In practice, as in the case of drybean yields in Nicaragua, data are typically very scarce. It was therefore necessary to design a methodology that allowed weather insurance to be developed in these circumstances. 224 Case study for dry beans in Nicaragua We used MarkSim to generate 99 years of weather data for each 10-arc minute pixel in northcentral Nicaragua. They used these data as input to the Decision Support System for Agrotechnology Transfer (DSSAT, Jones et al., 2003) drybean model, using a four typical soils with textures ranging from sand to clay and either deep or shallow profile. They used the genetic coefficients for the variety Rabia de Gato, whose physiological characteristics are similar to the traditional varieties grown in the region. In summary, we simulated yields for 99 years for eight soils for each of 151 10-arc minute pixels, that is, almost 120 000 separate crops of drybeans. It is worth adding that the DSSAT drybean model provides detailed physiological and agronomic data and soil-water balance for each of the 71 to 75 days of the crop growth cycle, a total of almost 9 million crop-days of data. For each soil within each pixel, we used the 99 years' data to determine the threshold rainfall necessary in successive 10-day periods during crop growth. They did this by minimizing the residuals using a simplex routine to fit multiple linear regressions. Rainfall less than this threshold was termed a deficit for that ten days. For each of the 99 years where there were deficits, they summed them to give the total deficit for the growing season. Using the relation of total rainfall deficit against yield we set levels of deficit that would trigger an indemnity payout in a hypothetical insurance instrument. The probabilities of reaching a given level of deficit were then calculated for each of the eight soils for each pixel. The probabilities of reaching deficits of 50 and 70 mm, averaged over all eight soils for simplicity, are presented in Figure 2. 225 Figure 2 Probability of accumulated rainfall deficits of 50 and 70 mm during the growth of dry beans during the first growing season in north central Nicaragua. Based on these data, it was then straightforward to design an insurance instrument for each soil within each pixel. The details of a hypothetical contract are shown in Table 4. Tables 5 and 6 show hypothetical growing seasons that do not reach, and do reach, respectively, the trigger level. 226 Table 4. Sample insurance contract. RAINFALL INSURANCE CONTRACT Reference weather station (e.g.) San Dionisio INETER weather station Crop (e.g.) Dry beans – drought tolerant type Reference soil type (e.g.) Deep sand Sowing window (e.g.) 15 May to 15 June (e.g.) First day after 5 consecutive rainy days over Sowing date rule 5mm each Trigger value (e.g.) -70mm Premium price (e.g.) US$3 Indemnity (e.g.) US$5 for every mm of rainfall deficit after the trigger value Minimum rainfall requirements (given crop and soil stated above) Day Day Day Day Day Day Day Day Day 1 to 11 to 21 to 31 to 41 to 51 to 61 to 71 to 80 to 10 20 30 40 50 60 70 80 90 MIN 0 10 10 25 40 40 40 30 0 RAIN DEF a. TOTAL Rainfall deficit Calculation of indemnity payments: 1. MIN is the minimum rainfall that is required for your crop in each of the 10 day windows. 2. RAIN is the rainfall observed at the reference weather stations (you may enter this into the RAIN box, however it is the official rainfall recorded at the weather station that determines whether you are entitled to an indemnity payment). 3. DEF is the rainfall deficit. This is calculated by subtracting MIN from RAIN (only negative values are taken into account). 4. Indemnity payments occur when the TOTAL rainfall deficit is equal to or less than the trigger value. 5. The rainfall deficit is the sum of the 10 day rainfall deficits. Table 5. Example of a season not entitled to an indemnity payment (total rainfall deficit does not reach the trigger value of -70mm) Day Day Day Day Day Day Day Day Day 1 to 11 to 21 to 31 to 41 to 51 to 61 to 71 to 80 to 10 20 30 40 50 60 70 80 90 MIN 0 10 10 25 40 40 40 30 0 RAIN 34.9 22.4 0.6 33.8 0 57.6 73.4 161.8 112.9 DEF -9.4 -40 a. TOTAL Rainfall deficit -49.4 227 Table 6. Example of season resulting in an indemnity payment (total rainfall deficit exceeds the trigger value of –70mm) Day Day Day Day Day Day Day Day Day 1 to 11 to 21 to 31 to 41 to 51 to 61 to 71 to 80 to 10 20 30 40 50 60 70 80 90 MIN 0 10 10 25 40 40 40 30 0 RAIN 5.8 3.6 0 9.5 4.1 23.5 12.6 2 96.1 DEF -6.4 -10 -15.5 -35.9 -16.5 -27.4 -28 a. TOTAL Rainfall deficit -139.7 This exercise shows that it is feasible for any given location to simulate the yield of any particular crop for which there is a simulation model in the DSSAT series. Discussion Sound insurance requires best estimates of hazard probability. It also requires agreement about the likelihood of the hazard occurring. Errors in estimation of the hazard can be due to three sources: • An incomplete model in which the weather event cannot be related to the loss, • Spatial and (b) temporal variation in which the model is complete, but data are incomplete, and • Basis risk. Incomplete model. Exclusion of major factors such as soil and crop cultivar Soil specificity The effectiveness of rainfall is strongly influenced by soil characteristics. In soils that have low water-storage capacity, the impact of rainfall shortages will be felt much sooner than in the case of soils with high water-storage capacity. Conversely, when soils are dry, small falls of rain can be more effective on sandy soils compared with clay soils, which require more water to "wet up". Soil texture, soil depth and water-holding capacity are key factors to take into account in designing an effective insurance scheme. Farmers growing crops on very risky soils will need indemnity payments more often than farmers on less risky soils, which must be reflected in both a soil-specific payout structure and in the cost of the insurance coverage. We base our comments here on our experience in the Nicaragua case study described above. We used a range of generic soils with both deep and shallow profiles. As expected, sandy soils were much droughtier than heavier-textured soils and especially if they were shallow. In designing an insurance instrument based on modeling as described above, it is relatively simple matter to obtain the information necessary for the actual soils in question and adjust the index criteria accordingly. Cultivar specificity Rainfall requirements will also vary greatly from crop to crop and within the same crop depending on the cultivar. Drought-tolerant varieties will naturally withstand rainfall deficits more successfully that drought-sensitive varieties. Therefore in order to improve the relationship 228 between the rainfall weather index and crop losses, the rainfall indices need to be tailored specifically to the crop variety. The implications of this for modeling are that the genetic coefficients must be known for the cultivar or cultivars in question. Ideally these should be the outcome of carefully-designed experiments. Nevertheless, it is possible to make some informed guesses as to what the coefficients should be, based on phenological data from different latitudes for the cultivar in question. But the guessing should only be undertaken by experts with a clear understanding of how the particular model represents physiological factors such as photoperiod response and the thermoregulation of plant development. Planting date In rain-fed agriculture, which is implicit in designing a drought index, sowing date varies from season to season depending on the onset of rain at the start of the growing season. Since weather insurance schemes will be sold in advance when there is no information about what the weather will be, a transparent system is needed that incorporates variable planting dates into the insurance products. Both insurer and insured will need to know the exact start and end dates within which the observed rainfall will be taken into account for determining indemnity payments. To maximize the effectiveness of the insurance product, the method used to establish the sowing date used in the product must reflect the actual planting date as closely as possible. Spatial error Crop yields from research stations are typically 30%, or more, higher than those of farmers' fields (Davidson, 1965), so that using them as the basis for estimating the effect of a given weather event on farmers' yields is particularly dangerous. This is apart from the problem of whether or not a particular research station is representative of a given geographic area of interest. Moreover, weather risk varies spatially. To reflect this spatial variation of risk in the premium, methods to estimate it in risk evaluation are needed so that the insured pays the price of the risk they actually confront. There are limitations in generating weather by interpolation on the MarkSim weather surface. For the African and Latin American tropics the resolution is 10 arc minutes, about 18 km near the equator. In mountainous areas, this resolution is simply not fine enough to represent the weather for a particular farmer's field. For Asia the resolution is 2.5 arc minutes, about 4 km near the equator, better, but still a problem in mountainous country. The resolution of MarkSim was at least in some degree constrained by the large size of the data files necessary in relation to the limited capacity of computer hard drives in the relatively recent past. This is no longer much of an issue, so MarkSim could be refined by improving its resolution using the recently-available digital terrain models (DTMs) derived from the Shuttle Radar Topographic Mission (SRTM). The SRTM DTMs have a resolution of 90 m, which is more than adequate for modeling rainfall. This problem can also be addressed if one has actual rainfall data for a given site (for a sufficient span of years, of course). One can then use these data as input to MarkSim. As for temperature data input for MarkSim, one can correct the interpolated MarkSim temperature data for the pixel by correcting the pixel's mean altitude to the actual altitude of the site in question using the adiabatic lapse-rate of 0.6 °C per 100 m. 229 Temporal error, estimating extreme events from short-run data It is common to think that 50 years' (or so) weather data is sufficient to estimate yield variation in crops. We caution that this is a dangerous assumption. Engineers design structures and other works to withstand a given frequency of extreme weather, for example, a river levy to withstand a one in 100 year flood, termed more simply a 100-year flood. Clearly, a short run of historical data (50 years or even less) is only a limited sample of a very large population. Using such limited data alone to generate probabilities of climate risk will lead to seriously underestimated risk since by definition, only the extremes encompassed by the actual data are represented. To estimate the frequency of an event not represented in the data, engineers apply a log Pearson function to the yearly extremes within an historical data set. They then use the fitted Pearson function to predict the probabilities of extreme events that lie outside the range of the observed data. Obviously some variation of this approach must also be applied to short runs of historical data used to generate probabilities of weather-based risk events. A different component of temporal factors is some method of incorporating the El NiñoSouthern Oscillation (ENSO) phenomenon. Recent studies have shown that the ENSO has a profound effect on weather, not only in the eastern Pacific but more generally globally. Although this may make long-term forecasts more reliable, it is not yet clear how this can be applied in practical terms. We flag the topic here as one that needs to be followed closely. Consequences of basis risk As the chequered history of insurance shows, commercial viability is essential to ensure a selfsustaining insurance process. Viability of insurance is determined by the design of the insurance process, which encourages risk-sharing on the basis of transparent agreements between the insurer and the insured about drought probabilities. A key part of this agreement is the provision of accurate estimates, and in this respect we have concerns about potentially imprudent application of insurance. Insurance with excessive basis risk will be expensive or, worse, may invoke moral hazard since farmers will believe themselves to be protected whereas in fact they are not. Index-based schemes seem particularly vulnerable to basis risk, since their prime attraction is cost reduction through insuring weather events rather than actual inspectable loss. This requires trust and agreement that the weather event on which the premium is based is associated with actual loss. Practical implications: Technical considerations in the design of an effective weather insurance scheme A weather-index insurance scheme should ideally take into account the following scientific and technical details: Payable index Several models, typified by the DSSAT series, are available to simulate crop yield. The minimum climatic variables required as key drivers are daily maximum and minimum temperatures, solar radiation and rainfall. In principle, such models could be used to determine whether farmers receive an indemnity or not, by inputting the current weather data into the 230 model as they becomes available. Although this approach is scientifically sound, it is unlikely to be thought transparent by either the insured or the insurer. The requirement of a weather index simply means that a complex relationship between one climatic variable, such as rainfall in the case of drought, and crop yield must be converted into a simple index. Moreover, the index must be easily understood by all parties so that the trigger event for an indemnity payment is clearly defined. Accurate estimation of payment probabilities Insurance companies will need to know how often they will be paying out indemnities based on each of the weather stations they are using as a reference for payments. In some cases these weather stations will not have the necessary historical data to determine this probability. A method therefore needs to be established that will enable accurate estimation of the probability at points where the historical data are inadequate or lacking. Conclusions We present methods of providing low-cost, site-specific drought insurance products for any crop in any location in the tropics. We explain the benefit of insurance to risk takers, and especially those with minimal resources, from which it should become apparent that the major contribution this innovation offers is that it streams best available science about natural hazards directly to decision makers, through the medium of commercially-viable insurance products. Insurance provides decision-support to manage drought risk. The basis of the method, the insurance premium, transmits the best-available estimate of drought probabilities. Estimates are only as accurate as the predictive model that produces them and we reflect here on three sources of basis risk that are likely to occur when modeling crop drought risk: structural uncertainty of the model; spatial error and temporal error. Structural uncertainty increases when the model fails to represent processes that significantly influence drought risk. In this respect, a model that depends solely on correlation between rainfall and yield will not represent systematic and significant yield variations that are caused by temperature, soil, crop variety or a number of other factors. Spatial error introduces a second major source of basis risk, since it is rare that weather data, and even more so, yield data, are available with sufficient density to enable simple interpolation over large areas. Even where dense networks of weather stations exists, the degree of bias towards non-marginal sites is unknown, hence its ability to represent higher risk, marginal areas. Thirdly, error can occur due to unexplained temporal error caused by inadequate data runs. A purely empirical estimation of low probability events requires long-runs of data. As a final comment, there is a need to involve the key stakeholders of the insurance system, the insured, the insurers and the re-insurers, in the definition of models that accurately reflect the risks faced in the farmer's field. Their buy-in is needed not only to ensure ready acceptance of the actual index or indices that are produced, but also for the underlying assumptions on which they are based. This is especially true for the re-insurance firms who will either “reward” or “punish” the scheme via premium costs depending on what they perceive to be an adequate representation of risk. 231 References Barrett, C. B., Reardon, T. and Webb, P., 2001. Non farm income diversification and household livelihood strategies in rural Africa: Concepts, dynamics and policy implications. Working paper. URL: http://www.inequality.com/publications/working_papers/BarrettReardon-Webb_IntroFinal.pdf. Accessed 16 March, 2006. Bird, K., Hulme, D., Moore, K. and Shepherd, A., 2002. Chronic poverty and remote rural areas. ODI, London. URL: http://www.odi.org.uk/PPPG/publications%5Cpapers_reports%5Ccprc%5Cwp13.pdf. Accessed 16 March, 2006. Brown, W. and Churchill, C., 1999. Providing insurance to low-income households Part I: A primer on insurance principles and products. USAID - Microenterprise Best Practices (MBP) Project, Bethesda. Brown, W., Green, C. and Lindquist, G. (2000). Cautionary note for MFIs and donors considering developing micro insurance products. Development Alternatives Inc., USAID’s Microenterprise Best Practices Project, Bethesda. URL: http://www.usaidmicro.org/pdfs/mbp/a_cautionary_note_for_microfinance_institutions.p df. 42pp. Accessed 27 March, 2006. Bryla, E., Dana, J.; Hess, U. and Varangis, P., 2003. The use of price and weather risk management instruments. In Risk Management: Pricing, Insurance, Guarantees. Working Paper 2002-45. World Bank, Washington. Davidson, B.R., 1965. The Northern Myth: A Study of the Physical and Economic Limits to Agricultural and Pastoral Development in Tropical Australia. Melbourne University Press, Carlton. Dercon, S., 2001. Income risk, coping strategies and safety nets. Center for the Study of African Economics, Oxford University, Oxford. URL: http://www. wiwiss.fuberlin.de/w3/w3collie/SocPolWS03 /DerconSurvey.pdf, accessed 13 March, 2006. Dercon, S. 2003. Risk and poverty: a selective review (Or: Can social protection reduce poverty?). Mimeo, Department of Economics, Oxford University. Díaz Nieto, J., Cook, S., Lundy, M., Fisher, M., Sánchez, D. and Guevera, E. 2006. Final report: A system of drought insurance for poverty alleviation in rural areas. Centro Internacional de Agricultura Tropical CIAT, Project PE-4 Internal Report. EM-DAT 2004. The OFDA/CRED (Office of U.S. Foreign Disaster Assistance /WHO Collaborating Centre for Research on the Epidemiology of Disasters) International Disaster Database. http://www.em-dat.net, Université Catholique de Louvain, Brussels, Belgium. 232 Gifford, R.M., Thorne, J.H., Hitz, W.D. and Giaquints, R.T., 1984. Crop productivity and photoassimilate partitioning. Science 225, 801-808. Goes, A. and Skees, J.R., 2003. Financing natural disaster risk using charity contributions and ex ante index insurance. Presented at the American Agricultural Economics Association Annual Meeting, Montreal, Canada, July 27-30, 2003. URL: http://www.globalagrisk.com/pubs/Financing%20Natural%20Disaster%20Risk%20Using %20Charity%20Contributions%20and%20Index%20Insurance.pdf. Accessed 13 March, 2006. Hazell, P., Larson, D., Mac-Isaac, D. and Zupi, M., 2000. Weather Insurance Project Study: Ethiopia. The World Bank/ Fondazione Salernitana Sichelgaita/IFPRI, Washington. Hess, U., 2003. Innovative financial services for rural India: Monsoon-indexed lending and insurance for small holders. Agriculture and Rural Development Working Paper 9. World Bank, Washington, D.C. Jones, J. W., Hoogenboom, G., Porter, C. H., Boote, K. J., Batchelor, W. D., Hunt, L. A., Wilkens, P. W., Singh, U., Gijsman, A. J. and Ritchie, J. T., 2003. The DSSAT cropping system model. European. Journal of. Agronomie 18: 235-265. Jones, P.G. and Thornton, P.K., 1997. Spatial and temporal variability of rainfall related to a third-order Markov model. Agricultural and Forest Meteorology. 86: 127-138. Jones, P.G., Thornton, P.K., Díaz, W. and Wilkens, P.W. 2002. MarkSim: A computer based tool that generates simulated weather data for crop modelling and risk assessment. CD-ROM Series, CIAT, Cali, Colombia. Jones, P.G., Díaz, W. and Cock, J.H. 2005. Homologue: A computer system for identifying similar environments throughout the tropical world. CIAT, Cali, Colombia. Miranda, M. J. and Glauber, J. W. 1997. Systematic risk, reinsurance, and the failure of crop insurance markets. American Journal of Agricultural Economics 79: 206-215. Miranda, M. J. and Vedenov, D. M., 2001. Innovations in agricultural and natural disaster insurance. American Journal of . Agricultural Economics 83: 650-655. Mishra, P. K., 1994. Crop insurance and crop credit: Impact of the comprehensive crop insurance scheme on cooperative credit in Gujarat. Journal of International Development 6: 529568. Morduch, J., 1995. Income smoothing and consumption smoothing. Journal of Economic Perspectives 9: 103-114. Morduch, J. 1999. Between the state and the market: Can informal insurance patch the safety net? World Bank Res. Obs. 14, 187-207. 233 Mosely, P. and Krishnamurthy, R., 1995. Can crop insurance work? The case of India. Journal of Development Studies 31: 428-450. Pandey, S., Behura, D., Villano, R. and Naik, D., 2001. Drought risk, farmers’ coping mechanisms, and poverty: A study of the rainfed rice system in eastern India. In, Rice Research for Food Security and Poverty Alleviation. Peng, S. and Hardy, B. (Eds). International Rice Research Institute, Philippines. 67-274. Rosenzweig,M.R. and Binswanger,H.P., 1993. Wealth, weather risk and the composition and profitability of agricultural investments. Econonmic Journal 103: 56-78. Sakurai, T. and Reardon, T., 1997. Potential demand for drought insurance in Burkina Faso and its determinants. American Journal of Agricultural Economics 79: 1193-1207. Skees, J.R. 2003. Risk management challenges in rural financial markets: Blending risk management innovations with rural finance. Paving the Way Forward for Rural Finance: An International Conference on Best Practices, June 2 – 4, 2003, Washington, DC. URL: http://www.globalagrisk.com/pubs/Risk%20Mgmt%20Challenges%20in%20Rural%20Fi nancial%20Markets.pdf. Acessed 12 March, 2006. Skees, J.R., Gober, S., Varangis, P.,Lester, R. and Kalavakonda, V., 2001. Developing rainfall based index insurance in Morocco. Policy, Research working paper ; no. WPS 2577, World Bank, Washington. URL: http://web.worldbank.org/WBSITE/EXTERNAL/PUBLICATION/INFOSHOP1/0,,conte ntMDK:20124289~menuPK:278886~pagePK:162350~piPK:165575~theSitePK:225714, 00.html Accessed 16 March, 2006. Stoppa, A and Hess, U., 2003. Design and use of weather derivatives in agricultural policies: The case of rainfall index insurance in Morocco. International Conference: Agricultural Policy Reform and the WTO: Where Are We Heading?, Capri, Italy, June 23-26, 2003. Turvey,C.G., 2001. Weather derivatives for specific event risks in agriculture. Review of Agricultural Economics 23: 333-351. van Oppen, C., 2001. Insurance: A tool for sustainable development. J. Insur. Res. Prac. URL: http://microfinancegateway.org/files/2709_file_02709.doc. Accessed 16 March, 2006. Varangis,P., Skees, J., and Barnett, B., 2003. Weather indexes for developing countries (CHP). World Bank, University of Kentucky and University of Georgia, Washington. Watson, R.T., et al,. 2001. Climate Change 2001: Synthesis Report. Intergovernmental Panel on Climate Change, Geneva. Webb, P. and Reardon, R., 1992. Drought impact and household response in East and West Africa. Quarterly Journal of International Agriculture 31: 230-246. 234 Wenner, M. and Arias, D., 2003. Agricultural insurance in Latin America: Where are we? Paving the Way Forward for Rural Finance: An International Conference on Best Practices, Case Study. GlobalAgRisk, Inc., Lexington. World Bank. 2000. Managing catastrophic risks using alternative risk financing and insurance pooling mechanisms. The World Bank, Washington, DC. World Bank. 2001. Innovations in managing catastrophic risks: How can they help the poor? URL: http://www.proventionconsortium.org/files/wharton_010801/questions_for_sessions.pdf, Accessed 16 March, 2006. Zupi, M., 2001. The connection between micro-finance and micro-insurance in LDC: A review on literature and experiences. Internal Report, IFPRI, Washington. 235 Identifying the Role of Crop Production in Land Cover Change in Brazil, 1990-2006 Elizabeth Barona A.a a Intenational Center for Tropical Agricultural, Cali, Colombia Abstract Crop production in Brazil has changed significantly over the last decade. New crops are being cultivated to satisfy the world’s growing demand for Brazilian export products —a demand that has caused substantial changes in land use and cover, mainly characterized by the increase in large-scale mechanization of agriculture, deforestation, and intensification of agricultural land use. Brazil currently provides crop production information at the municipality level. This information was analyzed using Geographic Information Systems (GIS) to examine changes in the spatial distribution of the production of various crops and livestock in Brazil for 1990-2006. In addition, to better understand the relationship between agricultural expansion and deforestation, spatial data on agricultural expansion and deforestation over the Legal Amazon were statistically analyzed for 2000-2006. The results indicate that changes in the spatial patterns of crops have indeed taken place in central and northeastern Brazil as well as in the southern Amazon region. The areas to crops such as soybean and sugarcane expanded, surpassing the total area planted to domestic food crops, which, in turn, recorded a significant decrease in area. This crop expansion has exerted pressure on other crops and livestock, pushing them further into the Amazon forest region during 19902006. In the same period, pasture was the predominant land use in the Legal Amazon; however, results indicate that the area planted to soybean increased whereas the area under pasture decreased. Statistical analyses revealed that, in those areas with over 50% forest, deforestation was strongly related to agricultural expansion. Deforestation was related to pasture expansion in the states of Mato Grosso and Rondônia, but not to soybean expansion. On the other hand, soybean expansion in Mato Grosso seems to be correlated to a decrease in pasture. An increase in pasture was also observed in the states of Para, Acre, and Rondônia, leading to the hypothesis that soybean expansion in Mato Grosso displaced pasture to other states, thereby indirectly causing deforestation elsewhere. Table 1, summarizes the percentage of change in area harvested in 31 seasonal crops in Brazil from 1990 to 2006. Soybean, maize, sugarcane, rice, beans, cassava, wheat, and cotton were the eight most important crops in terms of area harvested in Brazil in 1990. However, only soybean, maize, and sugarcane showed an increase in area harvested in 2006. 236 Table 1. Change in area of 31 seasonal crops in Brazil during 1990-2006 CROPS Soybean (Soja) Maize (Milho) Sugar Cane (Cana de açúcar) Beans (Feijão) Rice (Arroz) Cassava (Mandioca) Wheat (Trigo) Cotton (Algodão) Sorghum (Sorgo) Tabacco (Fumo) Oats (Aveia) Castor beans (Mamona) Potato E. (Batata inglesa) Peanut (Amendoim) Triticale (Triticale) Watermelon (Melancia) Barley (Cevada) Sunflower (Girassol) Pineapple (Abacaxi) Onion (Cebola) Tomato (Tomate) Sweet Potato (Batata doce) Broad Bean (Fava) Melon (Melão) Linen (Linho) Mallow (Malva) Garlic (Alho) Jute (Juta) Rye (Centeio) Pea (Ervilha) Rami (Rami) TOTAL 1990 11,487,303 11,394,307 4,272,602 4,680,094 Total harvested area (ha) 1995 2000 11,675,005 13,656,771 13,946,320 11,890,376 4,559,062 4,804,511 5,006,403 4,332,545 2006 22,047,349 12,613,094 6,144,286 4,034,383 % Harvested area 2006 40.24 23.02 11.21 7.36 Harvested area % Harvested change area change 1990 to 2006 1990 to 2006 10,560,046 93.46 1,218,787 10.79 1,871,684 16.57 -645,711 -5.71 3,946,691 1,937,567 2,680,989 4,373,538 1,946,163 994,734 3,664,804 1,708,875 1,138,687 2,970,918 1,896,509 1,560,175 5.42 3.46 2.85 -975,773 -41,058 -1,120,814 -8.64 -0.36 -9.92 1,391,884 137,758 274,098 193,200 286,703 158,326 1,103,536 153,961 293,425 165,179 76,427 176,767 801,618 528,061 310,462 182,010 208,538 151,731 898,008 722,200 495,706 323,998 151,060 140,826 1.64 1.32 0.90 0.59 0.28 0.26 -493,876 584,442 221,608 130,798 -135,643 -17,500 -4.37 5.17 1.96 1.16 -1.20 -0.15 83,583 0 67,986 105,067 94,723 0 79,347 69,458 104,948 0 80,509 145,507 110,777 101,088 92,996 82,177 0.20 0.18 0.17 0.15 27,194 101,088 25,010 -22,890 0.24 0.89 0.22 -0.20 0 33,167 74,646 60,869 0 44,384 74,676 62,054 0 60,406 66,505 56,720 67,829 66,845 63,314 58,893 0.12 0.12 0.12 0.11 67,829 33,678 -11,332 -1,976 0.60 0.30 -0.10 -0.02 62,629 92,137 55,946 74,261 43,900 41,179 44,357 36,857 0.08 0.07 -18,272 -55,280 -0.16 -0.49 7,842 4,061 21,192 17,149 13,294 2,855 6,073 12,758 11,399 5,321 3,759 13,269 21,350 18,679 12,682 10,486 0.04 0.03 0.02 0.02 13,508 14,618 -8,510 -6,663 0.12 0.13 -0.08 -0.06 3,016 4,395 10,798 1,651 2,647 654 1,114 6,755 1,467 4,179 2,932 1,677 0.01 0.01 0.00 1,163 -1,463 -9,121 0.01 -0.01 -0.08 7,139 2,868 465 447 0.00 -6,692 -0.06 43,497,198 45,068,169 44,022,212 54,796,077 11,298,879 Source: IBGE - 2006 237 A Framework for Assessing the Impact of Agricultural Drought in Developing Countries Hyman, G.a, Jones, P.a, Fujisaka, Sb a International Center for Tropical Agriculture,Cali, Colombia Consultant. Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia. b Abstract Concerns about climate change, demographic changes and rising food prices are motivating increased interest in drought among the international agricultural research and development community. Yet very few systematic studies assess the size and extent of drought impacts on developing-country farmers. Where does drought hit hardest? What populations are affected? What are the crop yield losses due to drought? This research builds on a farming systems framework to assess the worldwide impact of drought in developing countries. The framework brings together a unique combination of socioeconomic and biophysical data, including information on climate variability, population, poverty and agricultural production. In order to put drought in its larger agricultural context, it also includes information on other constraints to crop production. The framework supports the formation of recommendations to lessen drought impacts in developing countries by targeting interventions to the local context. Possible location-specific solutions include agricultural intensification, crop diversification, livelihood diversification and exit from agriculture. The research project formulates data and tools to support investment decisions on research and development related to agricultural drought. Keywords: drought, impact assessment, priority-setting, developing countries, rural development 238 Are Crop Wild Relatives a Useful Source for Genetic Traits Related to Abiotic Resistance in the Context of Climate Change? Ramirez, J.a, Jarvis, A.a,b, Gamboa, D.E.a, Guevara, E.a a Centro Internacional de Agricultura Tropical (CIAT), Cali, Colombia. Bioversity International, Regional Office for the Americas, Cali, Colombia b Abstract Since early domestication, crops have evolved from their wild relatives, migrating into new areas and being selected for favorable traits to the point in some cases where today’s crops bear little resemblance to their ancestors. Yet wild relatives of modern and traditional crops still prove to be a useful source of genes for crop breeding for developing resistance to both biotic and abiotic stresses. In the 21st century we expect to experience climate change at a rate not before experienced in recent history, and hence the agricultural sector is faced with the challenge of adapting crops for the conditions to come. Here we examine the potential use of crop wild relatives in breeding for increased abiotic stress resistance based on a simple analysis of climate conditions where three major crops and their wild relatives are found. The analysis mined global climate datasets for current and future climatic conditions and datasets on crop distribution to compare the abiotic adaptation of crops versus their wild relatives. Principal components analyses determined whether wild relatives presented wider adaptability for current climate conditions and for four different climate change scenarios or vice versa. We found that the wild relatives for millets are currently the most distinct in climatic adaptation compared to the cultivated crops. In the case of potato, we show that wild relatives have an increasing role to play in breeding, with more traits available to adapt to future conditions than for the current climate. Keywords: wild relatives, climate change, abiotic stresses, breeding, domestication. Introduction Since early domestication, crops have evolved from their wild relatives, migrating into new areas and being selected by farmers and breeders alike based on favorable traits to the point in some cases where today’s crops bear little resemblance to their ancestors. Yet wild relatives of modern and traditional crops still prove to be a useful source of genes for crop breeding for developing resistance to both biotic and abiotic stresses (Hajjar and Hodgkin, 2007). Conventional crop breeders have used crop wild relatives especially for the introduction of biotic resistance into crops (Lane and Jarvis, 2007; Hajjar and Hodgkin, 2007), but the advent of novel molecular tools are overcoming some of the constraints to use and it is widely expected that crop wild relatives will play a more important role in crop improvement in the coming years (Hajjar and Hodgkin, 2007). The challenges associated with climate change are likely to increase the demand for abiotic resistance in crops. In the 21st century we expect to undergo a global change in climate at a rate 239 not experienced in recent history. Temperatures are expected to increase by 1.1-6.4oC to 2100, with shifts in total annual precipitation and its distribution through the year (IPCC, 2007). The agricultural sector is faced with the challenge of adapting crops for the conditions to come. Pertinent questions exist as to what the priorities for breeding are, what the biological limits to adaptation are for each crop, and where important traits may be found as input to breeding programs. Here we ask questions pertinent to the latter. Are crop wild relatives likely to provide traits useful in breeding to adapt to expected changes in climate? We use datasets on climate, crop distribution and crop wild relative distribution to evaluate this, focusing on three crops as representative of a broader spectrum of crops. Materials and Methods Our approach was to compare the climatic conditions of sites where crop wild relatives occur with the climatic conditions of areas currently under cultivation of their respective crop under both current and future conditions. We assume that a wild crop relative population collected in a specific site is genetically well adapted to the climatic conditions at that site. We are interested in demonstrating if crop wild relatives have unique climatic adaptations that might be of use for crop improvement to address current and future abiotic stresses. Statistical analyses are therefore employed to quantify the extent to which the crop wild relatives overlap with the climatic adaptation of the crop under current conditions, and using the results of global climate models we examine the extent to which these patterns shift into the future. The crops analyzed in this approach were millet (genus Pennisetum and Eleusine), potato (genus Solanum), and wheat (genus Triticum and Aegilops). Crop wild relative data The Global Biodiversity Information Facility (www.GBIF.org) was used to generate a spatially explicit database of occurrences of crop wild relatives. The crop wild relatives selected for each crop were based on a thorough literature review of taxonomy. A total of 15,005 entries were found including all wild relatives of all genera under analysis. Crop distribution data Total world harvested area grids (derived from FAOSTAT) for each of the three crops were used to define the climatic conditions where each crop is found. To reduce the size of the datasets for statistical analysis, representative random points were selected (3 to 5 percent of the total amount of pixels in the crop distribution) to provide 10,000 unique sites were the crop is known to exist. Climate data The climate data used was derived from WorldClim (http://www.worldclim.org, Hijmans et al, 2005), representing long term averages of monthly maximum, minimum, and mean temperature and monthly precipitation. The monthly variables were reduced to 19 bioclimatic variables (Busby, 1991) and altitude. Four future calibrated and statistically downscaled global climate grids (available from http://www.worldclim.org/futdown.htm) based on the HADCM3 model and the CCCMA model for the A2a scenario and the B2a scenario and specifically for the year 2050 were used for analyses of future changes in climatic conditions. 240 These climatic parameters (current climate, HADCM3 scenario A2a, HADCM3 scenario B2a, CCCMA scenario A2a, CCCMA scenario B2a) were extracted for each one of the points of random locations where the cultivated crops are reported, and for the current climate extracted for the points where wild relatives are reported. It was assumed that the cropped area is not changing in the next 50 years to allow a general analysis on the potential availability of CWR genetic traits for future climate conditions. Statistical Analyses All of the bioclimatic variables and the altitude were standardized (PROC STANDARD; SAS 2002) to produce a global mean of zero and a total variance of 1. To assess a reliable determination of the specificity of each crop and wild relatives in terms of abiotic adaptation and/or requirements, principal components analysis (PCA) was performed on the wild relatives and crop distribution on a crop by crop basis (PROC PRINCOMP; SAS 2002). A comparison between the climate conditions where each crop’s wild relatives are currently found and the cultivated crop was then performed for each climatic condition (current and the four future scenarios). All these procedures were performed using the Statistical Analysis System 9.1.3 under UNIX (Solaris 9). According to the proportion of the variance explained by each component, the number of components required was determined and biplots for each crop were generated to assess the overlap of climatic adaptation between the crop and its crop wild relatives. In order to quantify the extent to which the climatic adaptation of crop wild relatives overlaps with that of the crop, we employed two methods. First, the average of both principal components (PC1 and PC2) was calculated for wild relatives and cropped genotypes for each crop, giving the coordinate of the center of each point-distribution cloud of the biplot. The distance between these two points (distance between means-DBM) was then calculated and the overlap was evaluated in terms of this distance: the higher the distance the lower the overlap and hence the greater potential for novel climatic adaptations that might come from the crop wild relatives. In addition to this, frequency surfaces were computed each 0.5 units for each principal component: a Cartesian plane divided into a set of cells with 0.5 units of size was first obtained and then for each cell the number of occurrences for both wild and cropped genotypes were counted (frequencies). This produced a colored surface of frequency values. The percent of wild relatives outside the crop distribution surface was counted for each climate scenario and crop. This approximation quantifies the number of crop wild relatives with climatic adaptations outside of the current crop distribution, and hence likely to provide genes not currently available in the crop genepool. Coupling this with future climate scenario it is also possible to gauge the extent to which the crop wild relatives will provide novel traits for adapting to predicted future conditions. Results and Discussion Wild relatives versus crop climatic adaptation under current climate conditions All crops show a wider climatic adaptability than for the wild relatives species from which they have evolved (Figures 1). This indicates that domestication and breeding have made cropped varieties suitable for a wider range of environmental conditions. Variables with highest weights in all biplots, and that therefore generated general trends of data distribution for all crops under analysis were annual precipitation and annual mean temperature. Figure 1 shows the frequency 241 surface produced after the principal component analysis. Only the first and the second principal component were used for the biplots as these two components explained the great majority of variance for the three crops under analysis. For millet (Figure 1a) there is a great concentration of both wild and common genotypes for negative values of PC2 and positive values of PC1; this concentration can be seen also in potato (Figure 1b) but it is located in the first quadrant of the plane. In the case of wheat (Figure 1c) the wild relative distribution seems to be distributed uniformly in both axes, which indicates that domestication has lead to a broadening of climatic conditions in all directions (greater and lesser temperature and precipitation). a b c Figure 1. Biplots for current climate adaptations of crops versus their wild relatives for a) millet (left); b) potato (center) and c) wheat (right) Wild relatives of millet show very good suitability for environments that produce positive values in PC1; indicating sites with significant variations in monthly temperatures. There are two differentiated groups of genotypes: one of genotypes in environments that produce positive values in PC2 and the other contains wild genotypes in environments that produce negative values in PC2. There is a group of especially important wild relatives (E. coracana ssp. africana and P. purpureum) that are adapted to climates with annual mean temperatures under 15ºC and extreme precipitation. Cropped genotypes are suitable for environments produce extreme negative values in PC1 (wide range of monthly variations in temperature and/or precipitation). Distinctive attributes of the cultivated millets are their adaptability to adverse agroecological conditions, requirement of minimal inputs, and good nutritional properties (Garí, 2001). Wild millets, in the other hand, comprise a diverse range of wild grasses that are related to the cultivated millets, including wild millet relatives and wild millet-like grasses. Wild millets play important roles in local food security, especially during drought crisis or in arid ecosystems. Wild millets, therefore, represent fundamental genetic sources that deserve consideration and integration in biodiversity conservation and rural development programs, because of their role and potential in both food security and agricultural development (Garí, 2001). Potato wild relatives seem be distributed in a relatively uniform way over both components axes. Almost all potato wild relatives are present in environments with low monthly variations in temperature (most limiting factor). There are some wild species (S. colombianum and S. longiconicum) that seem to be adapted to conditions where no crop is grown with altitudes between 2500 and 3000 meters, annual mean temperatures near to 10ºC and annual precipitation 242 between 1200 and 4000 millimeters. This indicates some potential for breeding, as Smillie et al (1983) suggested that wild potato temperature stress tolerance varies according to altitude, and that wild relatives will only be useful as resources for breeding for abiotic traits for certain altitude bands. There are two groups of wheat wild relatives, with the first located in environments with negative values in PC1 and positive values in PC2, and the second located in environments with negative values in PC2 and both positive and negative values in PC1. Most of the wild relatives are present in sites with low precipitations and temperatures near to the crop’s global average, while the crop has expanded into zones of high annual mean temperatures where no wild relatives are reported. For wheat there is one wild species (T. cereale) outside of the crop climatic range that is suited to environments with annual mean temperatures between -2 and 10ºC but there are only 13 occurrences of this species. According to Bedõ et al (2005), wild relatives for wheat are especially useful if they are located in cold environments, as they would be useful for winter wheat breeding, and this species shows some promise. Usage of wild relatives in the face of climate change Under future climate scenarios the same analyses were performed in order to see if crop wild relatives are more or less important for adaptation to future conditions. The shift in future climate conditions where the crops are currently grown means that the central points of the distribution change relative to the central point for the crop wild relatives, and the shape of the distribution cloud also changes. Table 1 shows the distances between means (DBM) and the percent of wild relatives (WR) outside the crop distribution for each crop and each climate change scenario. In general, it was observed that some wild relatives are increasing their potential usages while other ones are decreasing. Significant variability is also evident between climate change scenarios and models. 243 Table 1. Overlap between climate adaptation of wild crop relatives and the crop for current and future climatic conditions Crop Genus Year Model Scenario DBM PC1 PC2 Millet Eleusine, Pennisetum 2000 None None Millet Eleusine, Pennisetum 2050 CCCMA A2a Millet Eleusine, Pennisetum 2050 CCCMA B2a Millet Eleusine, Pennisetum 2050 HADCM3 A2a Millet Eleusine, Pennisetum 2050 HADCM3 B2a Potato Solanum 2000 None None Potato Solanum 2050 CCCMA A2a Potato Solanum 2050 CCCMA B2a Potato Solanum 2050 HADCM3 A2a Potato Solanum 2050 HADCM3 B2a Wheat Aegilops, Triticum 2000 None None Wheat Aegilops, Triticum 2050 CCCMA A2a Wheat Aegilops, Triticum 2050 CCCMA B2a Wheat Aegilops, Triticum 2050 HADCM3 A2a Wheat Aegilops, Triticum 2050 HADCM3 B2a DBM: difference between means; WR: Wild relatives %WR Outside 2.47 0.1430.028 20.73 2.06 0.113 0.044 17.74 2.09 0.116 0.041 19.12 2.01 0.108 0.049 16.24 2.09 0.115 0.046 18.70 1.22 0.049 0.111 7.05 2.98 0.162 0.339 15.60 2.78 0.143 0.320 12.82 2.93 0.147 0.332 17.70 2.83 0.129 0.325 15.04 0.74 0.0040.033 0.37 0.95 0.0410.007 2.84 0.85 0.0350.014 2.03 0.90 0.0380.010 0.54 0.77 0.0330.009 0.37 First it should be noted that some wild genepools have better potential for use in crop breeding than others. The least useful was in the case of wheat, where low percentages of the wild species occur outside the current and future crop distribution surface (only 0.37% for the current conditions and a maximum of 2.84% in the future). The genepool with most promise is in millet, with more than 20% of wild genotypes outside the crop distribution surface under current conditions, and a minimum of 16.24% under future conditions. 244 For potato the potential of crop wild relatives for providing useful traits increases considerably when the future climate conditions of current potato growing regions is examined. A similar increase in potential is evident for wheat, but the potential remains low (under 3% of crop wild relatives have climatic adaptations different to the cultivated crop). Millet wild relatives tend to lose some potential for providing traits for adaptation to future conditions, but still have 18-20% of crop wild relatives providing potentially novel adaptation traits. Conclusions In all crops studied the wild relatives have a more restricted climatic adaptation to their cultivated crops. This indicates that during domestication, crops have diversified outside their natural environments. However, for some crops there are wild relatives available as sources of traits for abiotic resistances not available in the crop genepool. Millet wild relatives are most distinct, followed by potato and finally wheat. Specifically, wild species such as E. coracana ssp. africana and P. purpureum for millet, S. colombianum and S. longiconicum for potato, and T. cereale for wheat are found in climates distinct from the current distribution of the cultivated crop. In the context of climate change, the wild relatives are shown to become more important for providing useful traits to adapting crops to future abiotic stresses in the case of potato, and continue to be important for millets. References ARC 9.2. Arc/Info. Environmental Systems Research Institute, Inc. 1982 – 2006. Bedõ, Z; Láng, L; Veisz, O; Vida, Gy. 2005. Breeding of Winter Wheat (Triticum aestivum L.) for Different Adaptation Types in Multifunctional Agricultural Production. Turk J Agric For. 29 (2005) 151-156. Busby, J.R., 1991. BIOCLIM – a bioclimatic analysis and prediction system. Plant Protection. Q. 6, 8-9. Garí, J. A. 2002. Review of the African Millet Diversity. International workshop on fonio, food security and livelihood among the rural poor in West Africa. IPGRI/IFAD, Bamako, Mali, 19-22. November 2001. Gitay, H; Suárez, A; Jon Dokken, D; Watson, R. T. 2002. Climate Change and Biodiversity. Intergovernmental panel on Climate Change. IPCC Technical Paper V. 16-19p. Hajjar, R; Hodgkin, T. 2007. The use of wild relatives in crop improvement: a survey of developments over the last 20-years. Euphytica, 1-13. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G., and Jarvis, A. 2005. Very high resolution interpolated climate surfaces for global land areas. International Journal Of Climatology, 25, 1965-1978. 245 IPCC, 2007. Climate change 2007: Synthesis report. Summary for policymakers. Technical report. URL http://www.ipcc.ch/ipccreports/ar4-syr.htm Lane, A; Jarvis, A. 2007. Changes in Climate will Modify the Geography of Crop Suitability: Agricultural Biodiversity can Help with Adaptation.. SAT eJournal. Volume 4: Issue 1. Smillie, R; Hetherington, S; Ochoa, C; Malagamba, P. 1983. Tolerances of wild potato species from different altitudes to cold and heat. Biomedicine and biologic sciences. 159:2. Statistical Analysis System. SAS (r) 9.1. SAS Institute Inc., Cary, NC, USA. 2002-2003. SAS (r) 9.1 (TS1M3) 246