lnland Valley Characterization Repo $ Regional Characterization of lnland Valley Agroecosystems in Sikasso, Mali and Bobo-Dioulasso, Burkina Faso through integration of Remote Sensing tiloDal positioning System, and Ground-Truth Data in a Geographic information Systems Framework Prasad S. Thenkabail and Christian Nolte lnland valley watershed system highlighting the valley bottoms Uplands Intensely cultivated valley bottoms Moderatelycultivared valley bonoms Uncultivated valley bonoms International Institute of Tropical Agriculture with lnland Valley Consortium Inland Valley Characterization Report 3 Regional Characterization of Inland Valley Agroecosystems in Sikasso, Mali and Bobo-Dioulasso, ~urkinaF aso through integt0;ltiono f Remote Sensing, Global Positioning Systems and Ground-Truth Data in a Geographic Information Systems Framework Prasad S. Tllenkabail and Christian Nolte International Institute of Tropical Agriculture with Inland Valley Consortium Preface The inland valley characterization report series of the Resource and Crop Management Division (RCMD) is intended for the wide dissemination of results of research about the inland valley agroecosystems of sub-Saharan Africa. These research reports will address issues relating to characterization and diagnosis concerning inland valley agroecosystems. The range of subject matter is expected to contribute to existing knowledge on improved agricultural principles. practices, and policies that affect the sustainable development of these potentially rich and productive agroecosystems of sub-Saharan Africa. These reports summarize results of studies by IITA researchers and their collaborators; they are generally more substantial in content than journal articles. The research report series is aimed at scientists and researchers within the national agricultural research systems of Africa, the international research community, policy makers, donors, and international development agencies. Individuals and institutions in Africa may receive single copies free of charge by writing to: The Director Resource and Crop Management Division International Institute of Tropical Agriculture PMB 5320 Ibadan Nigeria The Authors Dr. Prasad S. Thenkabail is a Visiting Scientist, Remote Sensing Specialist, at Agroecological Studies Unit of RCMD, IITA. Dr. Christian Nolte is an Associate Scientist, Wetland Agroecosystem Agronomist, at Agroecological Studies Unit of RCMD, IITA. 'E 1995 Internztional Institute of Tropical Agriculture zz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ZS:MOJ ' ~ 6 :1qi ed NJ, IRSpUE7 104 zz sassep asn-puel 91 aqi u! sadLl JaAo3-pueljo uo!inq!Jis!p aXerua3iad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . [z Lpnis s!qi u! pagguap! sadh JaAo3-pueT . . . . . . . . . . . . . . eaie Lpnis aql jo spueldn uo pue 'sa8u!~j Lal le~ie 'smo~loqL a l p U! uo!id!i3sap pue uo!mq!~ls!p uo!lela8a~q ayl '~a!J08al~a3s n-puel £1 " " " " ' z1 . . . . . . . . . . ~. .L. .L .0~.3.SI .s N. .n./ O. V. .d. 0.1 .8.~.! ~. ~. 0. 3. 3.EE.aJ).l? . L.p.m s aql U! S[!OS $0 UO!Illq!llS!a Vale Lpnls ayl Lq paJaAO3 sauoz almsoJ3ew aql ~qs axpu! uo!iela5a~ (NL) adde em ~ g e w a q ~ zI . . ' . . " . ' . . " . . ' . . " . ' . . " ZS:MOJ '~61:qledE lEp NJ, lESpUE7 Lq pJaAO3 dew d x n a~q j o suo!%a~p uel pue sauoz l!os pue [e3!8o[o~ao.18eI laha? oI . . . . . . . . . . . . . . “ . . eaie Lpms ayl u! y~omlaup eoi pue 'sluamalllas 'uo!lelndod . . " . " . ' . . " . (Z ain%ga as) liodai d x n M ayl jo d ~ msuo !8al-pue[ aql ol pua8a7 . . . . . . . . . . . . . . . . sauoz pas pue [1?3!80[03a0~811? (aha[ ayl Su!q!~osap sJalaureJed zP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . s.uo !snl3uo3 pue L ~ e w w n'I~A elep q l ~ l - p u n o i 8w o y swaisLs La l l e~p uelu! jo uialled 8u!ddoin aqi Bu!u!w~alaa 8.p t E " ' elep ~ ~ N I - ~ U I I OwJo~i j par\!iap sLalleh puelu! $0 s~!ls!ialn~ieqn[e n!8oloqdiom ~ ' p EE " " ' . . . . . . e.a.ie. L.p.nl.s .ai.!l.ua. .aq.l. u!. a.n.u.an.b.as.i>.d.oi.a q. l. s.so.ia.e. u.~.al.le.d.u o.!ie.A !llnn jo L p n i ~9 .p E E yiomlau peoi aqi pue s,uawalilas woij a3ueis!p ol o a d s a ~q l ! ~sa! i!sualu! uo!ie~!i[n3 S.P EE . . . . . . . . . . . . . . . . . . . - uo!ieA!ilnn a3!i jo uo!lnq!~ls!p pue Li!sualu~ OE . , . . 1.v.p a3uanbasodol ssoiae sa!i!suam! uo!ieA!i[nn pue sLalle~p uelu! jo L ~ o r n a ~~u ~. p PZ " " " ' tz spueldn pue 'saZu!y Lal le~'s uro~~oLqa l le~jo uo!lez!ia13e~eq3 asn-puel 6.p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .e.ie.p .N.L. B. u.!s.n .qiomlau peoJ aql pue siuawa~ilas8 u!ddeN 2.b EZ . . . . . . . . . . . . . . . . . . . . . . . . . .sa 8u!ij Lal le~p ue surolio EZ . . . . . . . . . . . . . . .quo Lal le~8 u!ddem 1.b !ssnns!a pue s l l n s a ~. AI 81 ". '. . .s.m.a.nL. s. L. .a l.p. p.u.el.u!. .aq.l .u!. 3.!u.r.eu.L.p .ia.le.m. .o.i 1.3.a.ds.a.~q. l ! L~8 oloipL~ T . E . ~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L80lolpr(q pUE IIejU!ex E'z I [ surajsLs Bu!uried t . z .2 I T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .u o!ieia8a~ E'Z'Z I T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I T " " ' . . ' " ' " ' " " " " ' " . . . . " ' " . . . Sl!"S 2.2'2 le!ialew waled 1 '2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . eaie Lpws aql jo s3!1s!iai3eieq3 les!sLqd 2.2 . . . . . . . . ' . . ' . . . . " . . " . . . earn ipms aql bq paiaho:, sauoz apssoi3em 1. z " " " ' " ' " ' " ' " " " " " " s~!~s!~ameieeqine p avllales pue eaJe Lpws '11 [ . . . . . " . . punoi%q3ea pue uo!innpoiiuI .I 10. TM-derived vegetation indices for the final land-use classes in Landsat TM path:197, row:52, covering the regions of Bobo-Dioulasso, Burkina Faso and Sikasso, Mali . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 11. Land-use distribution in the different land regions of the WURP map determined using Landsat TM data of path: 197, row52 in the regions of Bobo-Dialousso, Burkina Faso and Sikasso, Mali . . . . . . . . . . . . . . . . . . . . . . . 28 12. Land-use distribution in the different agroecological and soil zones determined using Landsat TM data of path: 197, row52 in the regions of Bobo-Dialousso, Burkina Faso and Sikasso, Mali . . . . . . . . . . . . . . . . . . . . . . . 29 13. Distribution of valley bottoms, valley fringes, and uplands and their cultivation status in the study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 14. Inland valley morphometric characteristics determined using Landsat TM data of path: 197, row:52 (Bobo-Dioulasso, Burkina Faso, Sikasso, Mali) . . . . . . . . . . . . . 32 15. Morphological characteristics of inland valleys in the regions of Bobo-Dioulasso, Burkina Faso and Sikasso, Mali . . . . . . . . . . . . . . . . . . . . . . . . 35 16. Cultivation pattern of valley hottoms, valley fringes and uplands with respect to distance from settlements and the road network for different level I zones of Landsat path:197, row:52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 17. Relative distribution of land-use patterns within a 90 m by 90 m plot at different toposequential components in the Boho-Dioulasso, ~ u r k i n Faa so and Sikasso, Mali study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 18. Relative distrihutinn of land-use types within a cropping pattern at different toposequential components in the Bobo-Dioulasso, Burkina Faso and Sikasso, Mali study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 19. Farmland cropping pattern in hotto~nsa nd at hydromorphic and nonhydromorphic fringes of inland valleys and on uplands in the regions of Boho-Dioulasso, Burkina Faso and Sikasso, Mali . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Figures 1. Landsat TM and SPOT HRV acquisitions for level I1 characterization of inland valley agroecosystems in the level 1 agrnecological and soil zones of IITA (inserted between pp.213) 2. Landsat TM and SPOT HRV acquisitions for level I1 characterization of inland valley agroecosystems in the land regions of West Africa . . . . . . . . . . . . . . . . . . . . . . . . 3 3. Crosssection showing a model inland valley as defined in this study . . . . . . . . . . . . . 4 4. Population density in the study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 5. Location of the study area ant1 IITA level I zones . . . . . . . . . . . . . . . . . . . . . . . . . 7 6. Spatial distribution of the WURP land regions in in the study area . . . . . . . . . . . . . . 8 7. Geological formations in the study area . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . 14 8. Soils distrihution in the study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 9. Characterization of agricultural systelns in the study area . . . . . . . . . . . . . . . . . . . . 16 10. Characterization of cropping systems in the study area . . . . . . . . . . . . . . . . . . . . . 17 11. Monthly rainfall in 6 stations in the study area expressed as percentage of annual total . 18 12. Ground-truth site location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 13. Bispectral plots of the mean digital values for the final 16 land-use classes of Landsat TM bands (a) TM4 versus TM3 and (b) TM4 versus TM5 . . . . . . . . . . . . . 25 Plates - see centerfold - 1. False color composite of TM 4 (red), TM3 (green), TM5 (blue) highlighting the broad inland valley bottoms near Sikasso, Mali. For example, the bottom widths of valley bottoms were: 250 to 500 meters in Peniasso, and 500-800 meters in Banankoni and Lotio 2. Distinguishing the inland valley bottoms from the valley fringes and uplands through ratio RGB image of TM4lTM7 (red), TM4lTM3 (green), and TM4lTM2 (blue) 3. Inland valley bottoms delineated and mapped for the entire study area of Landsat Path:197, Row52 (Sikasso, Mali; and Boho-Dioulasso, Burkina Faso) (see the legend for color key) 4. Land-use classes mapped for the different components of the toposequence (valley bottoms, valley fringes, and uplands) for the entire study area of 3.12 million ha covered by Landsat path:197, row:52 in Sikasso, Mali and Boho-Dioulasso, Burkina Faso (see the legend for color key) 5. Land-use classes in a sub-area near Boho-Dioulasso, Burkina Faso; note the relatively high density of cultivation across the toposequence in areas adjoining roads and settlements (see legend for the culor key) 6. Land-use classes in a suh-area near Sikasso, Mali; note the dramatic differences in land use in Farako reserve forest and the area in the immediate neighbhourhood of Sikasso (see legend for color key) 7. Rice cultivation in the inland valley bottoms surrounding Sikasso, Mali. Observe the contrast between the valley bottoms of the Farako forest reserve from that of the valley bottoms near and north Sikasso 8. Location of the potential benchmark research sites for technology development research activities in the study area covered by Landsat TM path:197, row:52 (Sikasso, Mali; Bobo-Dioulasso, Burkina Faso) 9. Land-use classes uf valley bottoms in a sub-area (near Toussiana, Burkina Faso) of Landsat Path:197, Row:52. The valley system of Kamao (immediately west of Toussiana) is one potential valley for technology development research activity 10. Normalized difference vegetation index (NDVI) map of sugarcane fields in a sub-area ( ) of Landsat path: 197, row:52; the higher the NDVI, the greater is the biomass and vigour leading to higher productivity 11. Spatial distrihution of areas of significant upland cultivation (see white color) in the entire study area of Landsat TM path: 197, row:52; areas inside the polygon indicate insignificant cultivation 12. Spatial distrihution of areas of significant cultivation (see white cn!or) in inland valleys (valley bottoms plus valley fringes) in the entire study area of Landsat TM path:197, row:52; the areas encircled hy the polygons 1, 2, 3, and 4 indicate insignificant cultivation Introduction and Background The International Institute of Tropical Agriculture (IITA) is currently conducting a characterization of inland valley agroecosystems in West and Central Africa (Izac et al. 1991). A macro- (subcontinental) scale stratification (level I) of West and Central Africa led to a map of 18 agroecological and soil zones (figure 1). Each of these zones represents an area of more than 10 million ha (table 1). Overall, about 36 million ha spread across 11 countries of West and Central Africa are targeted for level I1 characterization. This will be mesoscaled (regional, semidetailed). For this purpose, 1 I Landsat Thematic Mapper (TM), and 5 Lz systkmepour I'obscrvation de la terre (SPOT) high r-solution visible (HRV) data bases were acquired in sample areas of level I. Each satellite scene over a particular area will be referred to as a study area. Each study ar,-a covers one or more of IITA's level I agroecological and soil zones (figure 1 and table 1). The ohjectives of level I1 (regional) characterization include inventorying and mapping inland va!ley bottoms; determining land use and land cover of inland valleys and their uplands; determining the cultivation intensities relative to settlements and the road network; mapping the spatial distribution of inland valley systems: testing hypotheses, such as relationships between uplandlinland valley cultivation patterns and cultivation patterns within and across agroecological zones; and determining the location of potential benchmark research sites for technology development activities. The level II characteristics are reported with respect to the subcontinental (level I) agroecological and soil zones (figure 1) and also with respect to geological and geomorphological land regions of the Wetland UtilizationResearch Project (WURP) in West Africa (Windmeijer and Andriesse 1993). The location of satellite images on the WURP map is shown in figure 2. Readers of this inlanil valley characterization report are referred to the resource and crop management research monograph of IITA hy Thenkahail and Nolte (1995a) for the background, definitions (see illustration in tigure 3), rationale, ohjectives, approach, and methodology of level 11 characterization which remain consistent across the study areas. A brief overview of this ~nethodologyi s described below. 1 . The valley bottoms were delineated using image enhancement, display. and digitizing techniques. 2. The valley fringes were mapped by delineating the areas immediately adjoining valley bat-toms, by using a search radius on either side of valley bottoms equivalent to the mean fringe width measured during the ground-truthing. 3. Land-uselland-cover studies were carried out separately for valiey bottoms, valley fringes, and uplands using unsupervised classification of multihand data. 4. Several other characteristics such as the percentage area of inland valleys cultivated at varying distances from roads and settlements, have heen extracted through spatial data manipulation (e.g., boolean logic interpolation and contiguity analysis) of Geographic Information Systems (GIS) datalayers. 5. A methodology for key sitelkey watershed selection was developed based on expert knowledge and GIS modeling of various spatial datalayers. 6 . Ground-truth data, Global Positioning Systems (GPS) data, and other data sources were incorpo- rated into digital image analysis. Analyses were conducted using Earth Resources Digital Analysis System (ERDAS 1991). Table 1 Parameters describing the level I agroecological and soil zones ~ ~ ~ ~ ~~ ~ ~ ~-~~ p ~ ~ p ~ Level I Agroecological zone L G P ~ Major FA0 soil ~ r e a ~ AEZa according to IITA's (days) groupingC (million ha) d e f ~ t i o n 1 northern Guinea savanna 15 1-180 Luvisols 25.2 2 southern Guinea savanna 181-210 Luvisols 18.4 3 southern Guinea savanna 181-210 Acrisols 12.4 4 southern Guinea savanna 18 1-2 10 Ferralsols 11.9 5 southern Guinea savanna 181-210 Lithosols 10.7 6 Derived savanna 21 1-270 Ferralsols 47.2 7 Derived savanna 21 1-270 Luvisols 24.9 8 Derived savanna 21 1-270 Nitosols 14.2 9 Derived savanna 21 1-270 Arenosols 14.0 10 Derived savanna 21 1-270 Acrisols 11.7 11 Derived savanna 21 1-270 Lithosols 10.8 12 Humid forest > 270 Ferralsols 150.1 13 Humid forest > 270 Nitosols 27.2 14 Humid forest > 270 Gleysols 19.2 15 Humid forest > 270 Arenosols 18.9 16 Humid forest > 270 Acrisols 18.0 17 Midaltitude savanna Ferralsols 45.4 18 Midaltitude savanna Nitosols 12.3 Notes: a. AEZ: level I agroecological and soil zones b. LGP: length of growing period c. Names refer to the soil classification scheme of FAO/UNESCO (1974) d. The area figures are for West and Central Africa and were determined using the AREA procedure of IDRlSI (1992) Plate 1F alse color composite of TM4 (red), TM3 (green), TM5 (blue) highlighting the broad inland valley bottoms near Sikasso, Mali. For example, the bottom widths of valley bottoms were: 250-500 m in Peniassa, and 500-800 m in Banankoni and Lotio Plate 2 Distinguishing the inland valley bottoms from the valley fringes and uplands through ratio RGB image of TM4fI'M7 (red), TM4FIM3 (green), and TM4fl'M2 (blue) Plate 3 Inland valley bottoms delineated and mapped for the entire study area of Landsat Path: 197, Row: 52 (Sikasso, Mali; and Bobo-Dioulasso, Bnrkina Faso) along with land-use classes. Valley bottoms were 8.6 percent (0.27 million ha) of the total study area (3.13 million ha) nsignificant f t i l 'mlilnd~.f ringes i i ~ n i r i r a n t , f a l r n l a n d sh, ottoms I 1 i- Yes 0.035 study area Dioulassr,. Sikasso, Orrrditra, Koloko, K<,udrlugrx~K, ookx. Fammana, Klela, Kouliala. Kouri Notes: 1 . Level I agroecological and soil zones (see figure I) h. Geological and geomorphological land regions in West Africa according to Hekstra d al. (1983) and Windmeijer and Andriesse (1993) (see figure 2) c. Wetland utilization research project (WURP), see Hekstra et al. (1983) The results and discussions will he presented ant1 discussetl for the following zones (see tahle 4 and figures 5 and 6): 1. Level I zones: AEZ 1 (45% of the study area) and AEZ 2 (12%); and 2. WURP land regions: land region 3.3 (68%) and land region 2.8 (24%); other land regions occu- pied very low percentages of the overall stutly area: land region 3.1 (2%) and land region 3.2 ( 6 % ) ,a nd hence characteristics relative to these two zones were not reported. The study area comprises two zones with different lengths of growing period: the northern Guinea savanna with 151-180 days and the southern Guinea savanna with 181-210 days (see figure 5 and tahle 1). Land region 3.3 falls in this study area predominantly in the northern Guinea savanna (figures 5 and 6). The mean vegetation densities were nearly uniform across the different macrozones in the study area (tahle 5). The Normalized Difference Vegetation Index (NDVI) values of 0.34 ttr 0.39 (tahle 5) are indicative of fairly vigorous vegetation. This was mainly due to the date of data acquisition (27 Septe~nher1 991) which is the peak rainy season in the region with crops in vegetative or critical growth phases. Other vegetation. such as grasses and shruhs, is vigorous to very vigorous in growth. 2.2 Physical characteristics of the study area 2.2.1 Parent material. The study area covers four different land regions (LR) of the WURP map (see figure 2 and table 2, and figure 6). However, two land regions occupy only 2% and 6% of the study area, whereas the major part falls into land region 3.3 (68%), the Plateaux region with sedimentary deposits (paleozoic sandstones) as geological formation. In the northwestern part of figure 6 land region 2.8 occurs (24% of the overall study area). It is part of the Interior Plains region of West Africa with sedimentary deposits (paleozoic sandstones, tillites) as parent material. The gwlogical formations constituting the region are given in tigure 7. Much of the study area is covered by sandstones and tillites of Cambrian age. Ahout 90% of the ground-truth sites were located here. According to Si~npara( 1995) a northwest-southeast axis divides the geologic formations of that region. The sandstone south of Sikasso is of a hard nature whereas north of Sikasso a soft sandstone occurs. Blanchet (1992) sees these changing sandstone characteristics as a major reason for independently circulating subsurface groundwater levels in the region which affects the hydrologic dynamic in inland valleys. The study area is predominantly sandstone with recent alluvial deposits along certain valleys (figure 7). The recent alluvial deposits in the valleys are as a result of periodic tlooding in the area. The tlooding deposits fertile soil and makes cultivation of rice and other crops extremely attractive. 2.2.2. Soils. According to the classification of the level I map, Luvisols are to be found as the major soil grouping in the study area (see figure 1 and table I). In reality, the study area is characterized by very heterogenous soil conditions on uplands and the nonhydromorphic part of inland valley fringes. Figure 8 tlisplays the areal spread of 24 map units as given on the soil map of the world (FAOIUNESCO 1977) for that area. Their composition is listed in table 6. Based on the composition rules of F A 0 (1978) the following tlistribution of soil units was calculated: - 56% Luvisols. with 61% gleyic and 37% ferric Luvisols - 17% Nitosols, with 95% distric and 5% eutric Nitosols - 11 % Regosols, with 87% eutric Regosols - 9% Lithosols - 7% Others (Gleysols. Cambisuls, Vertisols, Fluvisols, Acrisols) 2.2.3 Vegetation. Vegetation in the study region is mostly part of the shrub savanna in the Sudan- Guinea savanna transition zone. Alhergel et al. (1993) report Buryospr~~mumpa rkii (Karitd), Rrminalia macroprc2ru, Ptc.rocurpus erinucrrus, Cordia afn'canu, Parkia biglobosa, and Khaya sc,nc,galensEs as dominating species to he fountl. At the lower part of inland valley fringes, a tree- savanna-type canopy (gallery hrest) occurs. 2.2.4 Farming systems. The agricultural systems are as diverse as the soils in the study area. Manyong et al. (in preparation) characterized farming systems in the middle belt of that area as market-driven and being in the intensification phase (figure 9). The influence of market access and good infrastructure on farming systems is also obvious in the area around Sikasso and between Bobo- Dioulasso and Toussiana. Farming systems here are in the market-driven expansion and early intensification phases, respectively. That means infrastructure, such as road networks, has reached a level sufticient enough to enable farmers to grow at least one cash crop as a major objective of farm households. Between these two regions, around Koloko un the Burkina Faso site, lies a region where population density is still the major driving force for farming systems hut land availability is increasingly scarce. Therefore, Mauyong et al. (in preparation) classified that region as population driven expansion phase (figure 10). Market factors mainly drive the farming (figure 9) with cotton and sorghum being the main upland crops (see their spatial distribution in figure 10) and rice being the main lowland crop. Table 4 Level 1 agroecological and soil zones and land regions of the WURP map (Hekstra et al. 1983; Windmeijer and Andriesse 1993) covered hy Landsat T M path: 197, r o w 5 2 (Boho-Dioulasso, Burkina Faso and Sikasso, Mali) Macroscale zones Area Percentage of entire study areaa (ha) A. agroecological and soil zone (AEZ) AEZ 1 1,418,302 45 AEZ 2 389,234 12 outside level I zoneb 1,328,110 43 entire study area 3.135,856 100 B. WURP land regionsC land region 2.8 761,281 24 land region 3.1 44.270 2 land region 3.2 land region 3.3 2,138,169 68 Notes: I percentages are "rounded-off" to nearest integer b. results for the area "outside ledel I" are not reported c. geological and gmmorphological land regions in West Africa according to Hekstra et al. (1983) and Wind~neiizra nd Andriesse (1993) Table 5 Thematic Mapper (TM) vegetation indices for the ~nacroscalez ones in the study area Vegetation indices Macroscale zones Per~.entage Ratio Normalized Midinfrared Midinfrared of the entirz vld = VI = VII = V12 = study area TM4 TM4-TM3 TM4 TM4 TM3 TM4+TM3 TM5 TM7 AEZ 1" 45 2.27 0.39 1.10 3.31 AEZ 2 12 2.19 0.37 1.01 2.76 entire study area" 100 2.27 0.39 1.07 3.05 land region 3.3L' 68 2.21 0.38 1 .OO 2.89 land region 2.8 24 2.01 0.34 1.02 2.86 Notes: n. Level I agrowological and soil zone (see figure I and table I ) b. Study area covered by i i ~ l ls cene of Landsat-5 path: 197, row52 c. Geological and geomorphological land regions in West Africa according to Hakstra at al. (1983) and Windmeijer and Andriesse (1993) (see figure 2) d. VI = vegetation index Table 6 Distribution of soil units in the study area (according to FAOIUNESCO 1 9 7 7 ) ~ , ~ May FAO- Extension in Dominant %d Associated cdd Inclusions %d unit svmhol studv area(ha) soil soils 1 Lf 41-2a 606183 Lf 60 Fg 2 0 2 0 23 Rd 7 12695 Rd 50 k§ 20 Be 10 20 24 Lf 39 31737 Lf 60 k§ 20 20 Notes: a. This tahle is a legend for figure 8 b. Names of soil units according to FA0 (1974): ahhreviations: A = Acrisols J = Fluvisols V = Vertisols g = gleyic B = Camhisols L = L~~visols d = distric h = humic G = Gleysols N = Nitosols e = eutric p = plinthic I = Lithosols R = Regosols f = ferric v = vertic c. Soils in map unit nu~nherI , 11, 14, 21, and 24 have a petric phase; soils in imp unit number 2, 16, 20, and 23 have a petroferric phase d. Percentages according to the algorithm of F A 0 (1978); areas derived by planimeter measuring 13 N LEGEND I. . . . .[ Sonbrlone + l i i i i t o icombnoni t Rocen1 o u m i o l deporirr 1--Ssh,sr + q ~ o r l l i l e/p recombr ionJ VXA s c nrsts, quor,zi1s,, sllnbSlon91, grcn,fic gnstrs. lnoNr RFRffRRl Groni le . rsh,rf /bossmen1 c a m p ~ e r i '?io,,/e, g#DD,o Figure 7 Geological formations in the study area N LEGEND 1-1 country border 1-( ~ o j o ro od rysfem S!O t.; Il0.n. 7 1 ino or rood e system ~ ~ drive~n ~ l t .h~ , i t , ~o r i optnn om v]MO.IH~d~r*i~v Ie n rntsn,it~sotionp nore /oor/y stopel 1-1 ~opu io t iond riven mr.nsificotion pnore (eor~ys togo) Mor#ot dr ivm #zpons!on pnore 1- Piiii:)p.p./otiond riven Pxpons,on phoJ# 1-1 NO do10 Figure 9 Characterization of agricultural systems in the study area 2.3 Rainfall and hydrology Rainfall in the study area shows a monomodal pattern with 80% of the annual amount falling in four months, between June and September (figure I I). The annual total varies at stations in the study area among 832 mm (KoundougoulKouka in the northeast), 987 mm (Koutiala in the northwest), 1072 mm (Sidkradougou in the southeast, at the lower right comer of the study area), and 1310 mm (Sikasso in the southwest). Evapotranspiration around Sikasso is around 1800 mm (Blanchet 1992). The long- term spatial distribution of the rainfall in the study area varies between 1000 mm to 1250 mm, with rainfall decreasing towards the northern portion of the study area. Most of this rainfall occurs between the months of May ant1 September. 5 I P M A M I J A S O N D Figure 1.l Monthly rainfall in 6 stations in the study area expressed as percentage of annual total [Sivakumar et al. (1984). Sivaku~nara nd Gnoumou (1987)l 2.3.1 Hydrology with respect to the water dynamic in inland valley systems. The hydroldgical dynamics is best understood with a knowledge of the topographical or terrain features in the study area. The inland valley systems are. typically, characterized by large bottom and fringe widths with gentle transversal slopes of 0-2 degrees. Mild to very mild longitudinal slopes along with tlat to near-tlat valley bottoms in the first- to fourth-order streams do not facilitate quick drainage of water downstream. This results in shallow water being spread across inland valley bottoms of third- and fourth-order streams (which often exceed 500 In in width) as the water drains off from the first- and second-order streams. The inundation will last for several months during the peak rainy season (July through October). These characteristics are generally found in the whole of the study area and are best exemplified in plate 4: (i) the area between Nialninasso and Nougoussouala north of Sikasso, Mali, (ii) the area between Bama and Desso. nor& of Sikasso, Mali, and (iii) the area in the immediate vicinity of Sikasso. Mali. These areas are very well suited for inundated rice cultivation. Blanchet (1992) determined in the Peniasso watershed near by Sikasso in 1991 runoff coefficients (surface runoff in percent of rainfall) between 3.2% and 13.0% for eight events. The runoff coefficients were calculated to be 8-13% at the heginning and in the middle of the rainy season whereas at the end of the rainy season (September) they come down to 6% (Blanchet 1992). Runoff as a percentage of total discharge (defined as surface runoff [ruissc 60% at the start of the rainy season and between 30 and 50% at the end. Ground-Truth Data Ground-truth data were collected from 25 valley bottoms, 17 valley fringes, and 36 upland plots. In valley fringes, 17 sample plots were located at the upper, nonhydromorphic part and 8 sample plots at the lower hydromorphic part. Ground-truthing took place between 25 September and 1 October 1993 to correspond seasonally with the satellite overpass date of 27 September 1991. The locations of ground-truth sites are shown in figure 12. At each location (where a GPS reading was taken) there were 1 - 2 plots with the GPS location reading being taken on the center of the road and the plots falling on either side of the road. The time frame of the project necessitated a heavy reliance on archived (or historical) satellite data. Real-time or near-real-time ground-truthing is not a feasible proposition due to numerous difticulties involved, such as a high uncertainty in obtaining a good quality real-time satellite image as a result of cloud and harmattan problems, and difficulties in planning ground- truthing activities across several study areas of West and Central Africa at short notice. A consideration of such data collection prccedures has been discussed in the monograph by Thenkabail and Nolte (1995a). The data collection strategy, parameters measured or observed, and the methods and procedures used to collect and analyze parameters remain the same across study areas. These are described in the same monograph. A comprehensive inland valley database has been developed by Ofodile et al.(in press). Only a shortcut of the parameters measured is presented in the following paragraph. The location of each ground-truth site was determined using a global positioning system (GPS) Garmin 100-SRVYR. Locations noted were geographic co-ordinates (latitudellongitude) in degree, minutes, md seconds and universal transverse mercator (UTM) co-ordinates (x,y) in meters. The accuracy of these GPS readings was usually within + 30 m. GPS was also used to collect ground- control points to georeference the satellite image. A total of 21 ground-control points in prominent locations, such as a road crossing a river (over the center of the bridge) and a road crossing a railway line were recorded. These ground-control points were well spread across different portions of the image. Land-use measurements were made along a transect in a 90 m by 90 m plot in valley bottoms, at valley fringes (hydromorphic and nonhydromorphic), and on uplands. GPS-location readings were taken at the center of the valley bottoms. Leaf area index (LAI) of the canopy was measured in the same 90 m by 90 m plot of valley bottoms and uplaqds. Land-cover types recorded at each site were trees, shrubs, grasses, cultivated farms, barren farms, barren lands, built-up areas or settlements, roads, and others. Different combinations of these land-cover types led to specific land-use categories (see tables 7, 8, and 9). Several other characteristics recorded at each inland valley site included: valley bonom width (m), valley fringe width (m), transversal slope (degree), stream order (number), and qualitative observations, such as occurrence of a central stream in the bonom, nature of the water discharge, status of water management systems, and soil moisture conditions (see Thenkabail and Nolte 1995a). LEGEND Y0/00 rood .rrr.m [ I L - 1 Mhor rood #y,r,nr -1 Gl S ~ I J E [ U I J~~L~I U ~ ~ ! U % ! S 1 spulrldn 'OU SSElO sy,ouo!inquls!p pun uo!idu3sap uo!iw%a1\ LioJainn asn-puel 1ulaads Table 8 Land-cover types identit'ied in this study Code Land-cover type description Code Land-cover type description 1 water 6 harren farms 2 trees 7 barren lands 3 shrubs 8 built-up arealsettlernent 4 grasses 9 roads 5 cultivated farms 10 others Table 9 Percentage distribution of land-cover types in the 16 land-use classes for Landsat TM path:197, row:52 covering the regions of Bobo-Dioulasso, Burkina Faso and Sikasso, Mali - - Code of Code of landcover types land-use classes 1 2 3 4 5 6 7 8 9 10 1 8 6 19 63 3 1 Note: Table 13 provides the exact cultivated areas for different components of the toposequence Results and Discussion The first step in establishing the characteristics of inland valleys in the study area involved georeferen- cing the satellite image to Universal Transverse Mercator (UTM) coordinates. This was done using 19 of the 21 GPS-location data points gathered at different prominent locations of the image during the ground-truthing. The image was georeferenced with an accuracy of about 2 pixels (ahout 60 m). This precise georeferencing made possihle an accurate study of inland valley characteristics, such as their land use and cultivation intensities in different components of the toposequence (valley hottoms, valley fringes, and uplands). 4.1 Mapping valley botton~sa nd valley fringes The study region is characterized hy valleys with large botttrm widths as exemplified in a suharea around Sikasso. Mali (plate I). The characteristics of this image include: 1 . a false color composite (FCC) image of hand TM4 (red). TM3 (green), and TM5 (blue) surrounding Sikasso, Mali: 2. an image displayetl with an magnifying factor of 1; and 3. an image that highlights the flat or ncar-flat hottoms which are seasonally inundated. Inland valley hottoms were distinguishwl from neighhouring fringes and uplands through image ehancenlent and display techniques such as the ratio RGB image of TM41TM7 (red), TM4lTM3 (green), and TM4lTM2 (hlue) (plate 2). These highlighted valley hottoms were delineated through digitizing. The illustration in plate 2 tlernonstrates: 1. an enhanced image. ohtained by using a ratio red-green-hlue (RGB) image of Landsat T M hands TM4lTM7, TM4lTM3. TM4lTM2; as a result, valley hottoms showed up in a white or cream Coltrred network oistreams. very easily distinguished from their fringes and surrounding uplands; and 2. an image, displayetl with a magnification factor of 2, to highlight inland valley hottoms, clearly distinguished from other features: thereby. valley btrtto~nb oundaries could he easily and exactly digitized directly off the screen. The resulting valley bottoms in the entire stutly area of 3.12 million ha are shown in plate 3 along with their land-use classes (to be discussed later in this report). Plate 3 provides the spatial distribution of inland valley bottoms, their densities. and land-use characteristics. Following the definition in Thenkahail ant1 Nolte (1995a), inland valleys comprise valley bottoms and valley fringes (hydr~~morphaicnd nonhydromorphic parts) (see figure 3). Valley fringes adjoin valley bottoms and were mapped by a combination of image processing and GIS techniques as explained in the monograph of Thenkahail and Nolte (1995a). The mean widths of the valley fringes, measured during the ground-truthing, were used to "spread" on either side of valley bottoms and "mask" the image area other than that within this "spread" width. This results in "masking" the valley bottoms and the uplands in order to highlight only the valley fringes. The outcome is illustrated for sample areas in earlier reports of Thenkabail and Nolte (1995b and 199%). The same technique was adopted to map the valley fringes of the entire study area of Boho-Dioulasso, Burkina Faso and Sikasso. Mali. 4.2 Mapping settlements and the road network using TM data The following parameters were determined using the TM data for each of macnrscale zones covered by the study area (tahle 3): 1. presence or absence of major settle~nents; 2. number of major settlements; 3. presence or absence of major road network; and 4. density of road network (kmlkm'). All the zones had one major settlement or more than one (tahle 3). The biggest settlements were Boho-Dioulasso (about 3050 ha) in Burkina Faso and Sikasso (about 835 ha) in Mali. All other settlements were much smaller (between 15 ha and 490 ha). A total of 70 major settlements were mapped. The location of these settlements with respect to the macrcrscale zones are given in tahle 3 (even though these settlements exist in plate 4, the scale of the map makes it impossible to notice smaller settlements). All macroscale zones in the study area have a major road system. The density of the road network was lowest in AEZ 2 with 0.022 km/km2 compared with the other zones which have road-network densities hetween 0.027 and 0.040 kmlkm? (tahle 3). 4.3 Land-use characterization of valley bottoms, valley fringes, and uplands Land-use characteristics were mapped separately t'or valley hottoms. valley fringes and uplands. This involved using the CLUSTR unsupervised classification algorithm of Earth Resources Digital Analysis System (ERDAS.) and incorporating ground-truth information to arrive at the desired land-use themes and classes (see Thenkahail antl Nolte 1995a for an extensive discussicln of the methodology). Six nonthermal hands of TM were used in the classification process. The GPS data and the land-use and land-cover data were used along with the spectral vegetation indices to identify the spectral classes of unsupervised classification. An initial 50 spectral classes of unsupervised classification were reduced to the final 16 land-use information classes (tahle 7, and figures 13a and 13h) which were then mapped unifc~rmlya cross the stutly area. Each of these classes has a varying percentage of 10 land-cover types (tahle 8). The 16 land-use classes (tahle 7) were displayed against the 10 land-cover types in a matrix fcirmat in tahle 9 . For example, the class "significant farmlands of valley bottoms" (land-use class 10) contains 65% farmlands (59% harren farms plus 6 % cultivated farms), 8% trees, 9 % shrubs. 14% grasses. and 4% harren lands (table 9). This proportion of land-cover types will vary depending on the season. For example, in the dry season many of the farmlands are expected to be harren. However. the land-use class will remain the same. The varying proportion of land- cover types for a land-use class is a result of the heterogeneity crf the information classes even within a single pixel (28.5 ~n hy 28.5 ~ n )an d due to aggregating different spectral classes to a few predecided land-use classes. As mentioned earlier, in this study the original 50 spectral classes from unsupervised classification were aggregated to 16 land-use classes (table 6). Pure land-use classes (having a single land-cover type that occupies 100% of its area) are water, settlements, and roads. The 50 original spectral classes were reduced to 16 land-use classes by integrating the ground-truth, GPS, and ancillary data. The final mean spectral characteristics of the 16 land-use classes are provided in table 10 antl figures 13a and 13b. The vegetation during the satellite overpass date (27 Septemher 1991) was lush green as it was still the rainy season in the region. This results in relatively high vegetation indices for each class. The distinct clusters of each class are obvious from figures 13a and 13h. Settle~nents(c lass 14) and wads (class 15) showed high reflectance in thematic mapper hand 3 (TM3) resulting in their clustering at the extreme right side of the plot. Absorption in the TM3 and TM4 wavebands resulted in a single very distinct cluster for water (class 13). Upland forest classes - class 6 (very dense forest) and class 5 (dense forest) - predominantly consisted of trees (see tahle 7). Ell - O P - 0s r 09 ENL snnaA VNL (el pug9 NJ.1 espue-1~0s assel3 asn-puel 91 pug a y l l o j s a n p lel!%!pu ealu aq] JO slold [e~]3ads!a€ 1 a ~ n 3 ~ These classes also comprise gallery forests along the fourth- or higher-order streams. All these classes have high reflectance in TM4 (near infrared waveband) and low reflectance in TM3 (the red waveband). As a result, the vegetation indices for these classes are amongst the highest (see table 10). The uncultivated valley fringes (class 9) were significantly different in spectral characteristics compared to the other two classes for uncultivated areas (class 3 for uplands and class 12 for bottoms). This was mainly due to the higher vigor and greater density of valley-bottom vegetation relative to that of fringes and uplands. Gallery forest (lush green trees and shrubs) is mainly concentrated along the valley bottoms. This was also due to a higher percentage of sparse and short shrubs at the fringes (class 12) compared with the bottoms. The clusters of classes 2, 8, and 11 are very close in figure 12 because they both have a remarkably similar land-cover distribution (table 9). The classes showing significant farmlands at the fringes (class 7) and on the uplands (class 1) have similar spectral characteristics, but their clusters are notably different in position, illustrating significant farmlands in valley bottoms (class 10). This was mainly due to the cultivation of inundated rice (swamp rice) in the valley bottoms versus cropping of sorghum and cotton at the fringes and on the uplands. The presence of water in these valley bottoms caused a high absorption in the water absorption band TM5, resulting in low values of TM5 (figure 13b). Therefore, the midinfrared simple vegetation index one (MSVII = data of TM4 divided through data of TMS) discriminates better between class 10 and class 1 or 7, respectively, than the ratio vegetation index (RVI = data of TM4 divided through data of TM3). Class 16 (barren land) shows up as a wetland in these hispectral plots mainly due to a high soil moisture status because of rains during the date of overpass. The 16 land-use classes obtained for this study area (Landsat TM path:197, row:52) are shown in plate 4. These land-use classes were grouped as follows (table 7): 1. 7 classes of uplands (classes I , 2, 3, 4, 5, 6, and 16); 2. 3 classes of valley fringes (classes 7, 8, and 9); 3. 3 classes of valley bottoms (class 10, 11, and 12); and 4. 3 classes of others (class 13. 14. and 15). Three land-use classes were synonymous for each component of the toposequence (table 7): a. significant farmlands (classes 1 , 7, and 10): farmlands (cultivated farms + barren farms) constitute > 30 % 11f the total land area of this class: b. scattered farmlands (classes 2. 8, and 11): farmlands constitute 1 10 % hut 5 30 % of the total land area of this class: and c. insignificant farmlands (classes 3.9, and 12): farmlands constitute 5 10 % of the total land area of this class. The areas occupied by each of these 16 land-use classes have heen presented with respect to the WURP land regions in table I I and to the agroecological and soil zones of IITA in table 12. The land-use classes in plate 4 compress an area of 3.13 million ha. This plate is useful in getting a spatial view of land-use distribution in a region. Land-use maps for "windows" within this region of study provide excellent details (see, for example, plate 5). Land-use class 3 represents areas with savanna vegetation. It comprises 5 10% (insignificant) farmlands. The percentage area relative to the total geographic area occupied by this class varies between 6.5% (for land region 2.8) to 28.7 % (for AEZ 1) (see tables l l and 12). The entire study area had 23.2%. area of class 3. Land-use class 2 (for uplands). 8 (for valley fringes), and 11 (for valley bottoms) are also predciminantly savanna vegetation with 10% to 30% of the area being farmlands. The percentage areas covered by these predominantly savanna land-use classes, in each agroecological zone, were generally high. For example: in AEZ 1 the percentage areas covered by these classes were 25.9% for class 2 (uplands). 6.7% for class 8 (valley fringes). and 4.2% for class 1 I (valley bottoms). The overall percentage areas far savannas should include the percentages of the "pure" savanna class (class 3) and predominantly savanna classes (classes 2, 3, 8, and 11). Thereby, the overall savanna percentage areas were 65.5% (class 2, 3, 8, and 11) for AEZ I , 70.4% for AEZ 2, 65.4% for the entire study area, 67.1 % for land region 3.3, and 61.6% for land region 2.8. Dominance of these classes in land regions 3.3 and 2.8 (table 11); and AEZ 1, 2, and entire study area (table 12) are characteristics of this study area in the transition of northern Guinea savanna and Sudan savanna. On average, land-use class 3 consists predominantly of grasses (48%) followed by shrubs (24%), trees (9%), farmlands (9%), barren land (6%) and others, (4%) (see table 6). The forest classes 5 and 6 represent about 4% in different macroagroecological zones (table 11 and 12) as may be expected in this transition zone of the northern Guinea savanna and Sudan savanna study area. It has to be noted that, even within this area, a significant portion is gallery forests (trees along river banks) for higher-order streams (fifih-order or higher) (see these characteristics depicted in plate 4). The gallery forests along the fourth- or lower-order streams fall into the class "uncultivated valley bottoms" (land-use class 12) or part into the class "uncultivated valley fringes" (land-use class 9). Tnhle 10 TM-derived vegetation indices for the final land-use classes in Landsat TM path:197, row:52, covering the regions of Boho-Dioulasso. Burkina Faso and Sikasso. Malia ~~~ Land-use MEAN VALUES classes TM3 TM4 TM5 R V I ~ MSVII' 1 39.75 82.75 87.50 2.10 0.96 2 38.26 78.36 74.89 2.12 1.10 3 33.00 74.82 67.18 2.27 1.11 4 38.00 69.00 68.00 1.82 1.01 5 33.50 89.50 67.50 2.67 1.33 6 33.50 101.40 77.00 3.03 1.32 7 40.25 81.33 89.33 2.05 0.94 8 36.67 76.42 79.00 2.09 0.97 9 38.75 79.13 84.63 2.06 0.95 10 38.90 77.30 70.20 2.01 1.15 11 35.50 75.50 73.38 2.13 1.04 12 32.78 80.67 67.56 2.48 1.19 13 29.70 37.50 i5.00 I .26 1.50 14 50.20 74.10 108.40 I .48 0.68 15 53.10 7 1.20 106.30 1.45 0.67 16 39.00 64.00 78.00 1.64 0.82 Notes: I . Data were obtained from 10 sample subareas (n = lo), each of 200 pixel by 200 pixel arca, tin each of the 16 land-use classes; this was done using MASK and BSTATS options of ERDAS h. RVI = ratio vegetation index; data of TM band 4 are divided through data of TM band 3 r. MSVII = midinfrared simple vegetation index one; data of TM band 4 are divided through data of TM band 5 27 It is important to note that the areas of each land-use class in tables 11 and 12 contain a varying degree of land-cover types as defined in table 9. Land use is an identity name for varying combinations of land-cover types. Therehy, exact cultivated areas, for example, should be derived from tables 11 or 12 based on the distrihution pattern of land cover provided in table 9. The results of exact cultivated areas are presented in table 13. For example, the exact cultivated areas of the valley bottoms for AEZ 1 are calculated as follows. a. The areas of valley bottoms for AEZ 1 were (table 12): class 10 (valley hottoms with significant farmlands) with 7,717 ha; class 11 (valley bottoms with scattered farmlands) with 59,062 ha; and class 12 (valley bottoms with insignificant farmlands) with 62,045 ha. Adding up these three classes gives a total valley bottom area of 128,824 ha. Table 11 Land-use distribution in the different land regions of the WURP map (Hekstra et al. 1983, Windmeijer and Andriesse 1993) determined using Landsat TM of path:197, row52 in the regions of Bobo-Dialousso, Burkina Faso and Sikasso, Mali No. Land-use category Land region 3.3 Land region 2.8 area % of total area % of total land (ha) land region (ha) region Uplands I significant farmlands 144,642 6.4 124,549 17.8 2 scattered far~nlands 588,800 26.0 275,227 39.3 3 savanna vegetation 648,320 28.5 45,760 6.5 5 dense vegetation 72,326 3 . 1 4,969 0.7 6 very dense vegetation 10,950 0.5 6,905 1.0 Valley fringes 7 significant farmlands 58,423 2.6 38,874 5.6 8 scattered farmlands 196,267 8.7 63,776 9.1 9 insignificant farmlands 206,119 9.1 23,639 3.4 Valley bottoms 10 significant farmlands 13,898 0.6 6,788 1.0 1 1 scattered farmlands 87,115 3.9 46,755 6.7 12 insignificant farmlands 84,384 3.7 9,530 1.4 Others 14 built-up arealsettle~nents 4,787 0.2 1,056 0 .2 15 roads 4,941 0.2 1,577 0 .2 16 harrenldesert area 57,237 2.5 44,766 6.4 Note: For the co~npositiono f land-cover types and their distribution in each land-use class, see tables 8 and 9. 28 Table 12 Land-use distribution in the different agroecological and soil zones (AEZ) determined using Landsat TM of path: 197, row:52 in the regions of Bobo-Dialousso, Burkina Faso and Sikasso, Mali No. Land-use category AEZ 1 AEZ 2 Entire study area area % of total AEZ area 5% of total AEZ area % of total AEZ (ha) (ha) (ha) Uplands significant farmlands scattered farmlands savanna vegetation wetlands/marshland dense vegetation very dense vegetation Valley fringes significant farmlands scattered farmlands insignificant farmlands Valley bottoms significant farmlands scattered farmlands insignificant farmlands Others water built-up arealsettlements roads barrenldesert area - - - -- - - - - - - Note: For the composition of land-cover types and their distribution in each land-use class. see tables 8 and 9 b. the cultivated areas of land cover for the corresponding land-use classes (class 10, 11, and 12) were defined in table 8 to be 65% (59% cultivated farmlands plus 6 % barren farmlands) for class 10, 23 % for class 11, and 3 % for class 12; c. the resulting cultivated area is 20,462 ha (7717*0.65 + 59062*0.23 + 62045*0.03); and d. the valley bottom area cultivated (20,462 ha) as a percentage of total valley bottom area (128,824 ha) is 15.9% (table 13). The land-use characteristics mapped for the entire study area of 3.13 million ha is depicted in plate 4. This plate provides an excellent spatial depiction of land use in a regional context. For example, the areas with little or no cultivation are shown in violet (upland savannas), rose, and red-orange (upland forest), red (predominantly valley-fringe savannas andlor forest), and magenta (predominantly valley-bottom forest and/or savannas), These colors are dra~naticallys een in plate 4 and show up to occupy over 50% of the area. These colors contrast with those of significant cultivation (shown in gray for uplands, white for fringes, and cyan for bottoms) and scattered cultivation (seafoam for uplands; pine-green for fringes; and yellow for hottoms) mainly along roads and settlements (see roads and settlements in plate 8 and compare the distrihution of significant and scattered cultivation in plate 4). The detailed land-use characteristics are depicted forsub areas near two major settlements: Bobo- Dioulasso, Burkina Faso (plate S), and ~ ikassoM, ali (plate 6). Plate 5 illustrates the high cultivation intensities across the toposequence in areas nearer to settlement and road networks. In the immediate vicinity of Sikasso there is intense cultivation in valley botto~ns( mostly inundated rice) and valley fringes and uplands (mostly sorghum) (plate 6). In areas to the east of Sikasso, in the Farako forest resexwe, the land use dramatically changes to savannas (see dramatic differences in land use depicted for forest versus nonforest areas in plate 6). 4.4 Inventory of inland valleys and cultivation intensities across the toposequence The area of inland valleys (valley bottoms plus valley fringes) is a function of the density of valleys and their characteristics, such as their bottom width and fringe width. An inventory of inland valleys was made possible by this process of highlighting and mapping (see section 4.1). Using the same technique as enumerated in section 4.1 and illustrated in plates 1 through 3, inland-valley bottoms were mapped for the entire study area of 3.13 million ha (plate 3). The sparse network of inland- valley systems in the entire study area (plate 3) is obvious from their spatial distrihution. A quantitative assessment indicated low drainage densities (ratio of the length of the streams to the area encompassed by them in kmlkm') and coarse stream frequencies (ratio of length of the streams to the area encompassed by them in number/km2) (tahle 14). The drainage densities varied between 0.35 kmlkm2 and 0.48 km/km2 and stream frequencies varied between 0.48 number/km2 and 0.69 number/km2. Although the spatial coverage of inland valleys is sparse, this study area is characterized hy large valley-bottom ant1 valley-fringe widths (table 15). The large bottom widths are evident in plate 1. These large valley hottom and valley fringe widths account mainly for the considerable percentage of area covered by inland valleys (tahle 13) in spite of the coarse stream frequency (tahle 14) in all the different macroscale zones studird. The mapping strategy conceptualized for use with remotely sensed data, as outlined in detail in Thenkahail and Nolte (1995a). is to map consistently all valleys as inland valleys along fourth- or fifth-order streams. The decision where to draw the line between inland valleys and floodplains (usually at fourth- or fifth-order streams) is hased on ground-truth data. However, not all valleys below, say, fourth-order when mapped as inland valleys are actually likely to be inland valleys. Table 13 Distribution of valley bottoms, valley fringes, and uplands and their cultivation status in the study area Study area Percentage VALLEY BOTTOM AREA VALLEY FRINGE AREA UPLAND AREA of entire study area as a 5% of culti\,atzd as a %E of cultivated as a % (of cultivated total as a X of total total as a % of total total as a R of total geographic valley-bottom geographic \,alley-fringe geographic upland area area area area area area AEZ l h 45 9.1 15.9 17.8 16.0 72.2 20.5 AEZ 2 12 7.7 24.0 25.5 2 1.8 66.5 24.1 land region 2.8' 24 9.1 21.0 18.1 27.0 72.6 29.6 land region 3.3 68 8.2 17.0 20.4 16.7 70.5 19.2 Entire study area 100 8 .6 18.4 20.4 19.2 70.2 21.9 Nntr;: a. When valley bottoms + valley fringes + uplands are not equal to loo%, the rest of the area falls in water body, roads and settlements or "round-off errors b. Level I agroecological and soil zones (see figure 1) c. Geological and geomorphological land regions in West Africa according to Hekstra et al. (1983) and Windmeijer and Andriesse (1993) (see figure 2) Table 14 Inland valley morphometric characteristics determined using Landsat TM data of path: 197, row:52 (Bobo-Dioulasso, Burkina Faso, Sikasso. ~ a i i ) ~ , ~ - Characteristics of inland valley watersheds AEZ I AEZ 2 entire land land land land study region region region region area 3.3 3.1 2.8 3.2 Mean drainage density 0.42 0.48 0.40 0.43 - 0.35 - (kmlkm2) - - Mean stream 0.51 0.69 0.61 0.65 0.48 frequency (no.lkm2) Notes: a. Hekstra et al. (1983) classified: I. drainage densities (km/km2)a s: very low (0-03); low (0.3-0.6); medium (0.6-1.2); high (1.2-2.4); and very high > 2.4; and 2. stream frequencies (numherlkm2)a s: very coarse (0-0.5); coarse (0.5-1.0); medium (1-2); fine (2-3); and very fine (> 3) b. When the suhara was too small such as for land region 3.1 (2% of total study area), stream densities and frequencies were not calculated and hence were marked "-" Some of them are tloodplains. This is due to the high variation encountered in characteristics such as bottom widths and flooding regime. Since floodplains have a different hydrological regime, soil conditions (Raunet 1985), and cropping patterns, they are to he distinguished from inland valleys. However, a strict distinction is not possible due to practical reasons and hence all valleys of fifth- order and below have been mapped as inland valleys. The inland valley frequencies and densities were higher in (1) AEZ 2 compared to AEZ 1; and (2) Land region 3.3 compared to land region 2.8 (see table 14). However, as a result of the presence of valleys with larger bottom widths in AEZ 1, the percentage area of valley bottoms in AEZ 1 (9.1%) exceeded that of AEZ 2 (7.7%) (as area is also a significant function of bottom width). For the same reason, the valley bottom area of land region 2.8 (9.1 %) exceeded that of land region 3.3 (8.2%). As a result of the methodology used in this study (see Thenkabail and Nolte 1995a for details) valley fringe area is a direct function of valley frequencies and densities. As a result, the zones with higher frequencies and densities (AEZ 2 in comparison to AEZ 1; land region 3.3 in comparison t o land region 2.8; see table 14) had a higher percentage of valley fringe area (25.5% for AEZ 2 in comparison to 17.8% for AEZ I; and 20.4%.for land region 3 .3 in comparison to 18.1% for land region 2.8; see table 13). Both land regions have similar geology-sedimentary deposits (Cambrian sandstone, figure 7). Cultivation intensities (table 13) in the valley bottoms were highest for AEZ 2 (24%) as a result of nearness of this area to the major settlement of Sikasso, and conditions market- driven expansion phase (see tlgure 9). Rice is the major crops in the bottoms. The highest intensities of upland (29.6%) and valley-fringe (27%) cultivation was in land region 2.8 as a result of market- driven conditions and with well connected road network and with cotton as the major crop. Due to significant differences in the geographical areas studied (45% of the entire study area for AEZ 1, 12% for AEZ 2, 24% for lmd region 2.8, and 68% for land region 3.3, see table 12) a direct and realistic comparison of results across zones was not feasible. In a more regional context, in the entire study area, valley bottoms were 8.6% (see plate 3) valley fringes 20.4%, and uplands 70.2% (table 13). The cultivation intensities in the entire study area were nearly constant across the toposequence with around 20% (18.4% for valley bottoms, 19.2% for valley fringes, and 21.9% for uplands). The significant cultivation across the toposequence was as a result of: 1. cotton + sorghum-based (figure 10) market-driven intensification or expansion phase (figure 9); 2. market-driven cultivation in lowlands (mainly rice) (see plate 7, for example). 4.4.1 Intensity and distribution of rice cultivation. Rice cultivation forms an important component of inland valley cultivation in the rainy season in the entire study area, especially in the valleys surrounding Sikasso, Mali. The broad and flat or near-tlat valley bottoms offer an excellent opportunity for paddy rice cultivation during the rainy season as demonstrated in several valleys around Sikasso, Mali (see plate 7 for the spatial distribution of rice cultivation in valley bottoms near Sikasso and its surroundings). A total of 269,006 ha constitute valley bottoms in the entire study area, of which 18.4% (49,497 ha) are cultivated (tables 12 and 13). Of the cultivated inland valleys, 42% of the area (20,789 ha) had rice crop. Inland valleys with rice are primarily to be found in a large area near Bama, northwest of Bobo-Dioulasso (see plate 4) and in the vicinity of Sikasso (see plate 7). Potential inland valleys for paddy rice cultivation exist, especially valleys that have wide bottom width (typically second- and higher-order), and significant water submergence as shown near Niaminasso, Nougoussouala, and Sikasso (see plate 7). These valleys, however, would require appropriate technologies, such as low-cost water control measures (e.g., channels, levies, and bunding) and rice varieties adapted to inundated conilitions. 4.5 Cultivation intensities with respect t o distance f r o m sett lements a n d the r o a d ne twork Cultivation intensities of valley bottoms, valley fringes, and uplands were calculated relative to their distance from major settlements and major road networks through manipulation of relevant GIs spatial data layers, using such techniques as hoolean logic interpolation and contiguity analysis. Cultivation intensities of valley bottoms, valley fringes, and uplands at various distance limits (0-2 km, 2-4 km, 4-5 km, and > 5 km) from major settlements and major road networks in the different level I zones are presented in table 15. Five k ~ nw as consitlereil the greatest distance for farmers to commute on foot to their farms on any given tl;~y;a nt1 hence the maximum tlistance limit was set at 5 km. Generally, the cultivation intensities decreased with increasing distance from settlements and the road network for each component of the toposequence (table 15). However, in several cases such a fall in cultivation intensity between two distance limits was only marginal (within 1 or 2%). This is obvious from uplands in land region 2.8 where the cultivation intensity remained virtually constant. According to Manyong et al. (in preparation) this area is characterized by market-driven agricultural systems with cotton as the major cash crop which is likely to account for that effect. In most cases, however, the cultivation intensities were about 3% higher for distance limits within 0-5 km as compared to those heyond 5 k ~ n . 4.6 Study of t h e cultivation pat teru across t h e toposequence i n t h e en t i r e s tudy area Significantly cultivated areas at each component of the toposequence are spatially illustrated for the entire study area of 3.13 million ha for uplands (plate 1 I), valley fringes, and valley bottoms @late 12). The polygons I and 2 were drawn for regions with insignificant cultivation. The regions with insignificant upland cultivation (areas within polygon 1 in plate 11) also had insignificant inland valley (valley bottom plus valley fringe) cultivation (areas within polygon 1 in plate 12). Similarly, regions with significant upland cultivation (several areas outside polygon 1 in plate 11) also have significant inland valley cultivation (several arras outside polygon I in plate 12). It is obvious from these figures that in dominant portions of the study area a high correlation exists between cultivation patterns on the uplands and in inland valleys. This proves one of the hypotheses of Izac et al. (1991) that the degree of upland cultivation has a strong influence on the degree of inland valley cultivation. These results further contlrmed the findings of Thenkabail and Nolte (1995b) in the Save study area. The cultivation intensities were strongly intluenced by the presence of settlements and the road network (see plate 4 along with roads and settlements shown in plate 8). Data of cultivation intensities across the toposequence are summarized in Tahle 16. 4.7 Morphological characteristics of inland valleys derived from ground-truth data Data of measurements of some morphological characteristics of inland valleys gathered duringground- truthing are highlighted in table 16. Only the data for areas with (Cambrian) sandstone as parent material (see tigure 7) are illustrated since descriptive statistical analyses were only possible in these areas with enough ground-truth sit& per respective stream orders. Due to the small number of observations per stratum, no statistical test was performed. However, data in table 16 show a clear trend that the bottom and fringe widths increase considerably with increasing stream order. At the same time, the data illustrate a high variation in measurements of bottom and fringe widths at each stream order. About 50% of the area (shape ratio of 0.49, table 16) of the valleys along first- and second-order streams constitute the hottom. The lower-order valleys also have 0.5 or 0.8 degree transversal slopes with almost flat or near-tlat fringes. Fourth- and fifth-order valleys had mean bottom widths of 495 m and 978 m, respectively. As mentioned in section 4.4 some of the higher- order valleys are to be considered as tloodplains. All the valleys (100%) investigated were U-shaped. In this wet season investigati~~(nla st week of September) 69% of the valley bottoms had wet soil conditions. 21% were moist. and 10% had dried-out soils. 4.8 Deternlining the cropping pattern of inland valley systems from ground-truth data The ground-truth land cover data provided the following important inferences (table 17): 1. The nonhydromorphic valley fringes had significantly more shrubs (27.8%) when compared with valley bottoms (14.5%) and uplands (17%); and 2. Uplands had significantly more grasses (38.3%) when compared with valley fringes (27.5% for hydromorphic and 27.9% for nonhydromorphic) and valley bottoms (26.1 %). The valley bottoms were distinguishable through: 1. Significantly higher cultivation intensities (37.5%) in comparison to valley fringes (17% for nonhydrtrmorphic and 26.2% for hydromorphic fringes); and 2. Significantly fewer trees + shrubs + grasses (50.2%) compared to nonhydromorphic fringes (69%), and uplands (63.9%). However, the cultivation intensities based on ground-truth data were highly overestimated (table 17) when compared with the same tjgures from satellite data (table 13). For example, the cultivated areas for valley bottoms were 37.5% using ground-truth data compared to only 18.4% estimated by satellite data. This is due to factors such as: 1. dependency along road networks for ground-truthing; 2. possible hias in stopping for readings at more cultivated valleys rather than relying on selecting valleys on a purely random basis; and 3. the fact that the valleys along road networks are more likely to he cultivated than valleys away from them. T a b l e 15 Cultivation pattern o f valley hottoms. valley fringes and uplands with respzct to distance from settlements and the road network for different level I zones of Landsat path: 197. r o w : j 2 Valle) Bottom Valle) Fnnge Upland Dlrtancc liniit Cult~vateda im as prrcentage of topal \.alley Cultivated a e a as percentage of total valley Cultivated area as percentage o f total holrorn area wilhin the corresponding distance fringe area within the corrzsponding distance upland area within the corresponding (km) limit limit distance limit AEZ I AEZ 2 land land AEZ I AEZ 2 land land AEZ 1 A E Z 2 land land region region rezion rsglon region region 3 .3 2 . 8 3.3 2.8 3 . 3 2.8 Distance f rom sat lements Distance from roatl nztwork Table 16 Morphological characteristics of inland valleys in the regions of Bobo-Dioulasso, Burkina Faso and Sikasso, Mali I TM data - Subarea Stream Transversal slope Bottom widtha Nonhydromorph~c Shape ratiob Bottom width order fringe width (m)" nC avgC n avg s.e. n avg s.e. n avg avg n s.e.= (m) (m) (degree) Notes: Database did not provide enough sample points per subarea and stream order for statistical analyses a. r' value for bottom widths measured on the ground and with TM data is ... (n = ..., a = 0.01) h. shape ratio = bottom width + fringe width c. n = number of observations; avg = average; s.e. = standard error d. see figure 8; 90% of the ground-truth sites fall into land region 3.3 (figures 4 and 7) Table 17 Relative distribution of land-use patterns within a 90 by 90 m plot at different toposequential components in the Bobo-Dioulasso, Burkina Faso and Sikasso, Mali study area Toposequence Observation Uncultivated and fallow land Farmland Barren land Total and others trees shrubs grasses sub total cultivated barren sub total (%) ( W ) ("/.I (%) (%I (%) A . Inland-valley 28 9.6 14.5 26.1 50.2 37.5 3.2 40.7 9.0 100 hottom B. Hydromorphic 8 10.6 20.0 27.5 58.1 26.2 10.1 36.3 5.6 100 fringe C. Nonhydromorphic 16 13.3 27.8 27.9 69.0 17.0 3.0 20.0 11.0 100 fringe D. Upland 37 8.6 17.0 38.3 63.9 27.7 4.5 32.0 4.1 100 (1) (2) (3) (4) (5) (6) (7) (8) I . C and D (0.10) 2. A and C (0.01), C and D (0.05) 3. A and D (0.05) 4. A and C (0. lo), A and D (0.10) 5. A and B (0.05). A and C (0.05), A and D (0.05) 6 . A and B (0.10) 7. A and C (0.05) 8. A and D (0.05), C and D (0.01) Significant differences between any 2 groups (e.g., valley bottom versus uplands) were reported for tach Varameter (t.g.., trees, shrubs) at 0.01, 0.05, and 0.10 levels. For example, grasses were significantly different at 0.05 level between valley bottom (26.1 %) and uplands (38.3%); thls has been reported as A and D (0.05) The bias becomes more pmrninent when one considers the low density of road network in the region (see plate 8, for example). In addition to the above points, it is important to note that the ground-truth data depends on plot measurement in each location. The diversity and variability even within a given valley are typically overwhelming. This is so because a timely representative plot exists only in theory. In practice (in the field) one hardly gets a clear view of variability due to accessibility and time factors. The above-mentioned difticultie with ground-truth data can be overcome through the spectral capability of remotely sensed data. This capability will enable a proper characterization of spatial variabilities that occur within and between valleys or uplands. Similar large differences were found hetween ground-truth and remotely sensed estimates of cultivation in another study area in Gagnoa, C6te d'lvoire (Thenkahail and Nolte 1995~) .D ue to the season (last week of September) of ground-truthing (main cropping season with most crops in vegetative to critical growth phases) 72-92% of farms were cultivated in different components of the toposequence (table 18). Grasses were the most dominant characteristic land-cover feature of uncultivated and fallow lands, irrespective of the toposequence (table 18). Valley bottoms were dominated by rice and closely followed by sorghum or maize (tahle 19). All other components of toposequence (nonhydromorphic ant1 hydromorphic fringes and uplands) were dominated by sorghum and maize fields. One surprising aspect of these results were that the numbers of cotton fields were very low or nonexistent an observation which contradicts data from Manyong et al. (see figure 10). Table 18 Relative distribution of land-use types within a cropping pattern at different toposequential components in the Boho-Dioulasso, Burkina Faso and Sikasso. Mali study area - - - - Toposquence Uncultlvatzd and tallow land Farlnland posltlon trees shn~hs grasses sub- cultivated barren sub- total total " (76) (W) (%) ( % I (%) (%) A. Inland valley 28 19 29 52 100 92 8 100 bottom B. Hydromorphic 8 18 34 48 100 72 28 100 fringe C. Nonhydro- 16 19 40 4 1 100 85 15 100 morphic fringe D. Upland Note: n = number of observations Table 19 Farmland cropping pattern in hottoms and at hydromorphic and nonhydromorphii: fringes of inland valleys and on uplands in the regions of Boho- Dioulasso, Burkina Faso and Sikasso. Mali Toposequence position n Farms R~ce Cassava Sorshum Cotton Vegetables Plantation Barren or yam or ma~ze Observations in absolute figures A. Bottom 28 22 1 1 I 10 1 3 7 7 B. Hydromorphic fringe 8 8 0 0 6 0 2 2 6 C. Nonhydromorphic fringe 16 9 0 1 7 0 2 2 6 Observations in relative figures (%) A. Bottom 79 50 5 45 5 14 32 32 B. Hydromorphic fringe 100 0 0 75 0 25 25 75 C. Nonhydromorphic fringe 56 0 11 78 0 22 22 67 D. U ~ l a n d 51 0 0 63 11 11 26 26 Note: n = number of observations Benchmark Area or Watershed Selection for Technology Development The output spatial data layers of this study obtained from remotely sensed data (land use of valley bottoms, land use of valley fringes, land use of uplands, road networks, and settlements) were used for GIS modeling to select likely benchmark sites for technology development research. The data obtained from ground-truthing were incorporated into the above datalayers. The position data of each ground-truth site and the ground-control point data for georeferencing form integral components of the above datalayers. Expert opinion was sought to rate each of the above spatial datalayers on a scale of 1 to 5 (5 being the best). This was done by weighing each factor of each datalayer according to their impact on inland valley cultivation as conceived by the expert. This procedure has been discussed in detail by Thenkabail and Nolte (199%). These data were sought in standard forms from members of the Inland Valley Consortium. Four international research centers (IITA, WARDA, CIRAD, Winand Staring Centre and Wageningen Agricultural University) and seven national research systems from Republic of Benin, Burkina Faso. CCte d'Ivoire, Ghana, Mali, and Nigeria constitute the Inland Valley Consortium. Thirty scientists with considerable experience ant1 knowledge in inland valley agroecosy- stems and diverse background were requestetl to respond to the questionnaire. These scientists represented five international agricultural research centers (IITAINigeria, WARDAlCBte d'lvoire, Winand Staring Centre and Wageningen Agricultural UniversityIThe Netherlands, CIRADIFrance, and IMMIINiger) and seven national research institutions (Sierra Leone, Cote d'Ivoire, Mali, Burkina Faso, Ghana, Benin, and Nigeria). Twelve persons responded. The modal value of each variable pertaining to a spatial datalayer was taken and incorporated into GIS modeling using the GISMO routine of ERDAS. The expert opinion indicated the following: 1. Significantly cultivated valley bottoms (2 30% of the total area is cultivated), valley fringes (hydromorphic and nonhydromorphic part), and uplands at present will drive further exploitation of the lands still left in the same valley system andlor in neighboring valley systems. 2. Inland valleys that are near already cultivated uplands have the best chance of being exploited for cultjvation; the greater the degree of cultivation in the uplands, the greater are the chances of inland valley cultivation. 3. The nearer the settlements and road networks are to inland valleys, the greater are the chances of those valleys being exploited; proximity to settlements has a greater influence than proximity to road systems. 4. Inland valley utilization for agriculture is likely to peak when population density rises above 30 persons/km2; 5. The zones with a shorter length of growing period (e.g., northern Guinea savanna) are relatively more likely to have inland valleys utilized for agriculture compared to zones with a greater length of growing periotl (e.g.. equatorial forest). 'Isaialu! jo spaqsialem io s e a s p!iualod aql u! sioleioqelloa SXVN 30 sisaialu! aql se [lam se sanss! syuouoaa pue '~eiuaumoi!~ua'a !uqla '@!30s se qans siolaej [euo!l!ppe $0 les!eidde aAloAu! plnom s ! 'uo~!lem ioju! alqepene Ile ql!m seaie pai:,a[as 01 sl!s!~q znoiql [es!eidde p!dei e uo paseq aq [[!M sa!l!i\!lae qaieasai luawdo[ahap K%olouqaal,1 03 paqualem l o eaie qxeasai e 30 uo!~aa[as[ tug a u ' (sdem 'suadxa '.%.a) sa:,Jnos i a q o Lue moy uo!aemioju! pue f(01 aleld u! dem IAaN aql aas) i o % pp ue Li!suap uo! l~ la%a~ se q3ns sal!s a saq JOJ %U!SU~S alomai moij UO!I~UI~OJU! leuo!l!ppe pue :(6 a l e ~ d' aldmexa 103 'aas) sassel:, asn-puel aql pue 'q~omlaup eoi aql 'malsKs Kalle~a ql se qsns uo!lemioju! 8u!moqs l!eiap u! paddew a n g aleld u! palsa%%ns eaie uo!ieaol le!iuaiod aql JO @iaAaS '~uamdola~1a0p3 @!lualod lsaleai% aql ItIasa~dalq aym asoql ale sma1sLs Kalle~p uelu! lsaq aqL '8 aleld u! salsi!s pallg ~ I ! M paiq%!~y%!qa n s paqsialefi qaieasai yiemqsuaq lo uaLele1ep le!leds asaqljo seaie qsieasai ynmq:,uaq a s a u ,(8 aleld) sa!i!~!pe qaieasai ,uamdola~apL %olouqsa~JO J suo!~eaolJ saq aqi Klle!lualod ale leq) m a n aql palq%!lq%!q %u!lapom s ~ fa )q 'siaKe[eiep legeds sno!ieA 01 mdu! uadxa aAoqe aql uo pasea Summary and Conclusions This inland valley agroecosystem research report presents and discusses the results of a level I1 (regional or semidetailed or meso-) characterization study of inland valley agroecosystems in the Sikasso and Koutiala regions of Mali; and Bobo-Dioulasso, and Kouka regions of Burkina Faso. The total study area is about 3.13 million ha covered by Landsat-5 Thematic Mapper (TM) path:197, row:52. (See the exact co-ordinates in figure 4.) The study adopted the methodology recommended by Thenkabail and Nolte (1995a) which involved digital image analysis and integrationof the remotely sensed data with GPS and ground-truth data in an GIS framework. One hundred percent of the inland valleys that were studied were U shaped, 74% were fndamas (that is, inland valleys with potential for dry-season cropping). At the time of ground-truthing 69% of the inland valleys were wet, 21% were moist, and 10% were dry. The mean transversal slopes were generally mild with about 1.5 degrees for the first-, to third-order inland valley streams, and about 0.5 degrees for the fourth-order inland valley streams. The total study area (3.13 nill lion ha) comprised 8.6% valley bottoms (see plate 3), 20.4% valley fringes, and 70.2% uplands. Water bodies, roads, and settlements comprised the other 0.8% area. The valley-bottom distribution was sparse (see plates 2 and 3). The drainage density of 0.4 km/km2, and stream frequency of 0.61 number/km2 obtained in the study area were classified as low (0.3-0.6 km/km2), and coarse (0.5-1.0 number/km2), respectively, by WURP (1983). In spite of the low and coarse drainage densities, and stream frequencies in the study area, the percentage area of inland valleys (valley bottoms plus valley fringes) was significant mainly as a result of the large valley- bottom and fringe widths of the inland valley streams (first- to fourth-order streams). The mean bottom widths for the first- to third-order strea~nsw ere about 90 In, and increased dramatically for the fourth-order to about 400 m. Valleys with typically large bottom widths are illustrated in plate 1. The mean valley fringe (hydromorphic plus nonhydromorphic) widths were about 200 m for the first three inland valley streams and for the fourth-order stream about 920 m. Hence, even though the stream frequencies and stream densities were coarse and low respectively, the large sizes of the valley bottoms and valley fringes led to their significant percentages. Due to significant differences in the geographical areas studied (45% of entire study area for AEZ 1, 12% for AEZ 2, 24% for land region 2.8 and 68% for land region 3.3 see tables 11 and 12) a direct and realistic comparison of results across zones was not feasible. However, it may be noted that the valleys in AEZ 1 had greater bottom widths than valleys in AEZ 2, resulting in a higher percentage area of valley bottoms in AEZ 1 (9.1%) compared to AEZ 2 (7.7%). For the same reason, the percentage valley bottom area in land region 2.8 (9.1 %) exceeded that of land region 3.3 (8.2%). The study mapped 16 land-use classes (table I I ) which were derived from the various combinations of the land-cover types (table 8). The spatial distribution of the 16 land-use classes in the entire study area of 3.13 million ha (plate 4) dramatically highlights regions with insignificant or no cultivation (violet, red, magenta, rose, and red-orange) in cornparison to regions with significant cultivation (gray, white, cyan). The seafoam color is the region with scattered farming. The characteristics of land use are best depicted when mapped in a smaller scale for subareas such as in plates 5 and 6. The spatial distribution of land use and cultivation patterns relative to road networks and settlements is available from plates 5 and plate 6. Forest and nonforest boundaries are dramatically highlighted in plate 6. Several characteristics of the inland valley agroecosystems can also be inferred from the land- use maps of the subscenes. For example, significant cultivation, mainly with rice crop, is seen mostly in the third- and fourth-order valley bottoms in the vicinity of Sikasso (plate 6). These valley bottoms are, typically, broad and flat and are often flooded and have recent deposits of fertile alluvium. The grassland dominant savannas are most extensive in the study area. The overall savanna percentage areas were 65.5% for AEZ 1, 70.4% for AEZ 2, 67.1% for land region 3.3, 61.6% for land region 2.8, and 65.4% for the entire study area. The forest classes are predominantly trees along the river banks and were about 4% for all level I zones within the study area. This very low percentage of forest cover was only to be expected in the study area as it falls in the northern Guinea savanna and Sudan savanna. Compared to other level I zones studied, barren areas of the Sudan savanna were most extensive with 6 % area of the respective level 1 zone. The cultivation intensities were nearly the same across the toposequence with 18.4% for valley bottoms, 19.2% for valley fringes, and 21.9% for uplands. The significant cultivation across the toposequence was mainly attributed to the market-driven conditions. In most cases, cultivation intensities were about 3% higher for distance limits within 0-5 km from the road network and settlements compared to those areas beyond 5 km. The valley bottoms in the study area were characterized by flat or near-flat surfaces (see plate 1 for example) that have shallow flooding all through the rainy season. Rice cultivation forms an important component of lowland rainy-season cultivation in the entire study area, especially in the valleys surrounding Sikasso, Mali (plate 6). These broad and flat or near-flat valley bottoms of the study area offer an excellent oppurtunity for inundated rice cultivation during the rainy season as successfully demonstrated in several valleys around Sikasso, Mali. (See plate 7 for spatial distribution of rice cultivation in the valley bottoms near Sikasso and surroundings.) However, the area of valley bottoms available for cultivation far exceeds their current utilization. A total of 269,006 ha constitute valley bottoms in the entire study area, of which only 18.4% (49,497 ha) was cultivated. Of the cultivated inland valleys, 42% (20,789 ha) had rice cultivation. Except for a few inland valleys such as the large farm near Bama, northwest of Bobo-Dioulasso (see plate 4), and in the area surrounding Sikasso (plate 7), there is very little rice cultivation andior only partial exploitation of inland valley systems for rice cultivation in the study area. However, an extensive potential for rice cultivation exists, especially in the inundated valleys of third- and fourth- order streams such as those near Desso and Laranfiara in Burkina Faso (see plate 9, for example); Banankoni and Lotio watersheds surrounding Sikasso, Mali (see plate 3 and 7); and Niaminasso and Nougoussouala, north of Sikasso, Mali (plates 3 and 7). These valleys, however, require appropriate technology such as low-cost water control nleasures (e.g., channels, levies, ant1 bunding), and appropriate rice varieties for inundated conditions. The study showed a strong relationship between upland cultivation (plate 11) and inland valley cultivation (plate 12) proving one of the hypotheses of this study. The study highlighted the strengths of remotely'sensed data in the proper inventorying of parameters such as percentage nf cultivated areas. Such estimates based purely on ground-truth data provided significant overestimates of cultivated areas. For example, the cultivated valley-bottom areas estimated purely based on ground-truth data were 37.5% compared to 18.4% provided hy Landsat TM data. Similar differences were observed in another study area in Gagnoa, Cote d'lv8ire using SPOT HRV data Fenkaha i l and Nolte 199.5~). Information from the different georeferenced spatial datalayers generated mainly from Landsat TM data (e.g., land use of different components of the toposequence, cultivation intensities of inland valleys with respect to road network and settlements) was used in conjunction with expert knowledge, through GIs modeling, to determine the potential sites for technology development research activities. This led to a map of potential benchmark research sites in the study area @late 8). These potential benchmark research sites were characterized by: (a) near-flat valley bottoms; (b) large valley bottom widths (about 100 m for first- to third-order valleys; and about 400 m for the fourth-order valleys); (c) well connected road networks (typically, within 6 km); (d) proximity to settlements (typically, within 6 km); (e) rainy-season shallow inundation of flood water in valley bottoms; ( f ) mild to very mild transversal slopes (mean of about 1.5 degrees for first- to third-order valleys; and a mean of about 0.5 degrees for the fourth-order valleys); (g) large fringe widths (about 200 m for first to third- order valleys; and about 920 m for fourth-order valleys). A final selection of the research sites should involve a visit to the several potential sites shown in plate 8 by a team of scientists of diverse expertise for a quick reconnaissance, interviews with farmers, and interaction with NARS. Detailed maps of each of the interested potential locations that are shown in plate 8 should be drawn to be taken to the field (see one such example in plate 9). The study resulted in digital georeferenced data bases for the land use of uplands, valley bottoms, and valley fringes; inland valley bottom areas; inland valley fringe areas; upland areas; rice cultivation areas: and benchmark research area locations. Acknowledgements The authors are grateful to the Directorate General for International Cooperation (DGIS), The Netherlands, for financial support of satellite imagery acquisition. We thank the Winand Staring Centre and the International Institute for Land Reclamation and improvement, Wageningen, The Netherlands, for allowing us to adapt and print figure 2 (the WURP map). We would also like to thank Ms. R. Umelo for providing excellent editorial help. Mr. John Babalola provided excellent secretarial assistance throughout the preparation of this research report. Mr. O.L.1. Ajuka and Mr. Mustapha Wahah provided invaluable assistance during ground-truthing and other related activities. We acknowledge all their cheerful help with thanks. Mr. Sam Ofodile provided excellent support in statistical analysis and database management of all ground-truth data pertaining to inland valley characterization reports including this one. Ms. Doreen Oyetayo efticiently entered all the ground- truth data. The acknowledgements to the last two persons (Sam and Doreen) were inadvertently missing in the earlier two IV characterization reports and we apologize for it. References Albergel, J . , J.M. Lamachkre, B. Lidon, A.I. Mokadem, and W. van Driel (eds.). 1993. Mise en valeur des bas-fonds au Sahel. Typologie, Fonctionnnement hydrologique, Potentialites agricoles. Rapport final d'un projet CORAF-R3S, Ouagadougou, Burkina Faso, CIEH, 335 PP. Anderson, J.R., E.E. Hardy, J.T. Rhach, and R.E. Witmer. 1976. A land use and land cover classification system for use with remote sensor data. Geological Survey Professional Paper 964, United States of American Government, Printing Office, Washington, USA. Blanchet, F . 1992. Etude hydrologique et agronomique d'un bas-fond de la region de Sikasso (Mali Sud). These Diplome, CNEARC-EITARC-CIRAD-IRAT, 51 pp. IDRISI. 1992. A raster-based geographic analysis and image processing system developed by the Graduate School of Geography (Prof. Eastman and his team) at Clark University, USA. ERDAS. 1991. Field Guide. ERDAS Inc., 2801 Buford Highway, Suite 300, Atlanta, Georgia, USA, 392 pp. F A 0 (Food and Agricultural Organization of the United Nations), 1978. Report on the agro- ecological zones project. Vol. 1. Methodology and Results for Africa. World soil resources Rep. 4811. Rome, 158 pp. FAOIUNESCO 1974. Soil map of the world 1:5,000,000. The Legend. UNESCO, Paris. FAOIUNESCO. 1977. Soil map of the world, 1:5,000,000 Vol. VI. Africa. UNESCO, Paris. Hekstra, P., W. Andriesse, G. Bus, and C.A. de Vries. 1983. Wetland utilization research project, West Africa. Phase I. The Inventory: Vo1.l : Main Report Vol.11 : The Physical Aspects Vol.111 : The Agronomic, Economic and Sociological Aspects Vol.IV : The Maps WURP-Report, Iland regionIISTIBOKA, Wageningen, The Netherlands Izac, A.M.N., M.J. Swift, and W. Andriesse. 1991. A strategy for inland valley agroecosystem research in West and Central Africa, RCMP Research Monograph No. 5, Resource and Crop Management Program, International Institute of Tropical Agriculture, Ibadan, Nigeria. Izac, A.M.N., S.S. Jagtap, A.I. Mokadem, and P.S. Thenkabail (in preparation). Agroecosystems characterization in international agricultural research: The case of inland valley systems in West and Central Africa. Manyong, M., J. Smith, G.K. Weber, S.S. Jagtap, and B. Oyewole (in preparation). Macrochara. cterization of Agricultural Systems in West Africa: A Broad Overview. Resource and Crop Management Monograph, RCMD, IITA, Ibadan, Nigeria. Ofodile, S., Nolte, C., and Thenkabail, P.S. (in press). A comprehensive georeferenced ground- truth database of inland valley agroecosystems of West and Central Africa. Database. Resource and Crop Management Division, International Institute of Tropical Agriculture, Ibadan, Nigeria. pp.480. Raunet, M. 1985a. Bas-Fonds et Riziculture en Afrique. Approche Stmcturale Comparative. L' Agronomie Tropical 40 (3): 18 1-201. Raunet, M. 1985b. Les Bas-Fonds en Afrique et h Madagascar. G&morphologie-Gbchimie- Pedologie-Hydrologic. Zeitschriti fiir Gmmorphologie, Supplementband 52, Berlin-Stuttgart, 25-62. Simpara, M.B. 1995. Presentation de la zone Mali-Sud. Paper presented at the 6th Annual Workshop 23-24 March 1995, lnland Valley Consortium, WARDA, Bouakk, CBte d'Ivoire. Sivakumar, M.V.K., M. Konate, and S.M. Virmani. 1984. Agroclimatology of West Africa: Mali. Information bulletin No. 19, ICRISAT, Patancheru, India, 294 pp. Sivakumar, M.V.K., and Faustin Gnoumou. 1987. Agroclimatology of West Africa: Burkina Faso. Information bulletin No. 23, ICRISAT, Patanchem, India, 192 pp. Thenkabail, Prasad S., and C. Nolte. 1995a. Mapping and Characterizing Inland Valley Agroecosystems of West and Central Africa: A Methodology Integrating Remote Sensing, Global Positioning System, and Ground-Truth Data in a Geographic Information Systems Framework. RCMD Monograph No. 16, IITA, Ibadan, Nigeria. 62 pp. Thenkabail, Prasad S., and C. Nolte. 1995b. Regional characterization of inland valley agroeco- systems in Save, Bante, Bassila, and Parakou regions in southcentral Republic of Benin through integration of remote sensing, global positioning system, and ground-truth data in a geographic information systems framework. Inland Valley Characterization Report No. 1. Resource and Crop Management Division, IITA, Ibadan, Nigeria. 60 pp. Thenkabail, Prasad S., and C. Nolte 1995~ . Regional characterization of inland valley agroecosystems in Gagnoa, CBte d'Ivoire through integration of remote sensing, global positioning systems, and ground-truth data in a geographic information systems framework. lnland Valley Characterization Report No.2. Resource and Crop Management Division, IITA, Ibadan, Nigeria. 52 pp. Thenkabail, Prasad S., and C. Nolte. (in preparation). Level I1 characterization of inland valley agroecosystems in Kaduna, Minna, and Birnin Gwari regions of Nigeria through integration of remote sensing, geographic information system, and global positioning system. Inland Valley Characterization Report No.4. Resource and Crop Management Division, IITA, Ihadan, Nigeria (Draft report published in March 1993. To he modified) 112 pp. UNEPIGRID. 1993. United Nations Environmental Program, Global Resources Information Database. UNEP, Nairobi, Kenya. Windmeijer, P.N., and W. Andriesse. (eds.). 1993. Inland valleys in West Africa: An agro- ecological characterization of rice-growing environments. WURP-Report, Iland regionIISTI- BOKA, Wageningen, The Netherlands, Publication No.52. LIST OF INLAND VALLEY CHARACTERIZATION REPORTS 1. Regional Characterization of Inland Valley Agroecosystems in Save, Bante, Bassila, and Parakou Regions in South-Central Republic of Benin through Integration of Remote Sensing, Global Positioning System, and Ground-Truth Data in a Geographic Information Systems Framework. Prasad S. Thenkabail and Christian Nolte. 1995. 2.Regional Characterization of Inland Valley Agroecosyste~ns in Gagnoa, C6te d'Ivoire through Integration of Remote Sensing, Global Positioning System, and Ground-Truth Data in a Geographic Information Systems Framework. Prasad S. Thenkabail and Christian Nolte. 1995. 3. Regional Characterization of Inland Valley Agroecosystems in Sikasso, Mali and Bobo-Dioulasso, Burkina Faso through integration of Remote Sensing, Global Positioning Systems and Ground-Truth Data in a Geographic Information Systems Framework. Prasad S. Thenkabail and Christian Nolte, 1995.