Validating a cassava production spatial disaggregation model in sub-Saharan Africa

by Kirsty L. Hassall, Vasthi Alonso Chávez, Hadewij Sint, Joseph Christopher Helps, Phillip Abidrabo, Geoffrey Okao-Okuja, Roland G. Eboulem, William J-L. Amoakon, Daniel H. Otron, Anna M. Szyniszewska

Cassava is a staple in the diet of millions of people in sub-Saharan Africa, as it can grow in poor soils with limited inputs and can withstand a wide range of environmental conditions, including drought. Previous studies have shown that the distribution of rural populations is an important predictor of cassava density in sub-Saharan Africa’s landscape. Our aim is to explore relationships between the distribution of cassava from the cassava production disaggregation models (CassavaMap and MapSPAM) and rural population density, looking at potential differences between countries and regions. We analysed various properties of cassava cultivations collected from surveys at 69 locations in Côte d’Ivoire and 87 locations in Uganda conducted between February and March 2018. The relationships between the proportion of surveyed land under cassava cultivation and rural population and settlement data were examined using a set of generalized additive models within each country. Information on rural settlements was aggregated around the survey locations at 2, 5 and 10 km circular buffers. The analysis of the original survey data showed no significant correlation between rural population and cassava production in both MapSPAM and CassavaMap. However, as we aggregate settlement buffers around the survey locations using CassavaMap, we find that at a large scale this model does capture large-scale variations in cassava production. Moreover, through our analyses, we discovered country-specific spatial trends linked to areas of higher cassava production. These analyses are useful for validating disaggregation models of cassava production. As the certainty that existing cassava production maps increases, analyses that rely on the disaggregation maps, such as models of disease spread, nutrient availability from cassava with respect to population in a region, etc. can be performed with increased confidence. These benefit social and natural scientists, policymakers and the population in general by ensuring that cassava production estimates are increasingly reliable.