Surveys vs Satellites: How to measure weather shocks in Sub-Saharan Africa

Details

Supervisors

Van den Broeck, Goedele

Faculty

Faculté des bioingénieurs

Degree label

Master [120] : bioingénieur en sciences et technologies de l'environnement, à finalité spécialisée

Abstract

enThis study examines the extent to which climate indicators derived from ERA5, the ECMWF atmospheric reanalysis product, can be used to predict the agricultural weather shocks reported by smallholder farmers in six sub-Saharan African countries (Ethiopia, Malawi, Mali, Niger, Nigeria and Tanzania). Household survey data are taken from the World Bank’s Harmonised Panel, which covers more than 59,000 households between 2008 and 2021. Fourteen indicators derived from ERA5, including temperature thresholds, extreme precipitation and the Standardised Precipitation-Evapotranspiration Index (SPEI), are assessed using bivariate logistic regressions and the area under the ROC curve (AUC). Combined models are then constructed through variable selection using Adaptive LASSO, supplemented by an analysis of false negatives and a comparison of subgroups according to four dimensions of social vulnerability. The ERA5 indicators offer acceptable predictive performance for drought shocks in Ethiopia, Malawi and Niger, and show particularly strong results for flood shocks in Malawi (AUC = 0.84). SPEI indicators consistently emerge as the most frequently selected predictors. However, no country reaches an acceptable threshold for rain shocks, and false negative rates remain high, exceeding 40% in some cases. Subgroup analysis reveals no systematic disadvantage for the most vulnerable households. Overall, ERA5 can serve as a valuable complement to household survey data for monitoring large-scale weather shocks but cannot replace information reported by farmers themselves. The difference between remote sensing measurements and farmers’ perceptions is not a matter of social exclusion but reflects agricultural and climate heterogeneity.