Extreme weather impacts do not improve conflict predictions in Africa

This paper investigates whether detailed information on extreme weather impacts improves conflict forecasts in mainland Africa (1994-2012). Using the Generalized Random Forest (GRF) algorithm, the study integrates data on droughts, floods, crop production shocks, and tropical cyclones from ISIMIP2a into predictive models. While extreme weather impacts alone modestly predict violent conflicts, socio-economic and conflict history indicators remain stronger predictors. Adding extreme weather impact information does not improve forecast models including conflict history, governance, and socio-economic variables, partly due to spatial correlations between extreme weather and other conditions. The study concludes that extreme weather impacts lack unique information for forecasting annual conflict incidence in Africa.

Sidney Michelini, Barbora Šedová, Jacob Schewe, Katja Frieler