Generating Plausible Crop Distribution and Performance Maps for Sub-Saharan Africa Using a Spatially Disaggregated Data Fusion and Optimization Approach

Agricultural production statistics reported at country or sub-national geopolitical scales are used in a wide range of economic analyses, and spatially explicit (geo-referenced) production data are increasingly needed to support improved approaches to the planning and implementation of agricultural development. However, it is extremely challenging to compile and maintain collections of sub-national crop production data, particularly for poorer regions of the world. Large gaps exist in our knowledge of the current geographic distribution and spatial patterns of crop performance, and these gaps are unlikely to be filled in the near future. Regardless, the spatial scale of many sub-national statistical reporting units remains too coarse to capture the patterns of spatial heterogeneity in crop production and performance that are likely to be important from a policy and investment planning perspective. To fill these spatial data gaps, we have developed and applied a meso-scale model for the spatial disaggregation of crop production. Using a cross-entropy approach, our model makes plausible pixel-scale assessment of the spatial distribution of crop production within geopolitical units (e.g. countries or subnational provinces and districts). The pixel-scale allocations are performed through the compilation and judicious fusion of relevant spatially explicit data, including production statistics, land use data, satellite imagery, biophysical crop “suitability” assessments, population density, and distance to urban centers, as wells as any prior knowledge about the spatial distribution of individual crops. The development, application and validation of a prior version of the model using data from Brazil strongly suggested that our spatial allocation approach shows considerable promise. This paper describes efforts to generate crop distribution maps for Sub-Saharan Africa for the year 2000 using this approach. Apart from the empirical challenge of applying the approach across many countries, the application includes three significant model improvements, namely (1) the ability to cope with production data sources that provided different degrees of spatial disaggregation for different crops within a single country; (2) the inclusion of a digital map of irrigation intensity as a new input layer; and (3) increased disaggregation of rainfed production systems. Using the modified spatial allocation model, we generated 5-minute (approximately 10-km) resolution grid maps for 20 major crops across Sub- Saharan Africa, namely barley, dry beans, cassava, cocoa, coffee, cotton, cowpeas, groundnuts, maize, millet, oil palm, plantain, potato, rice, sorghum, soybeans, sugar cane, sweet potato, wheat, and yam. The approach provides plausible results but also highlights the need for much more reliable input data for the region, especially with regard to sub-national production statistics and satellite-based estimates of cropland extent and intensity