Compound drought and heat exposure and lean-season food insecurity in Zimbabwe: agro-ecological heterogeneity in a 2009-2025 district panel

Handsome Bongani Nyoni , Lesly Rudo Chawira , Perez Livias Moyo , Zororo Kudzai Chinwadzimba , Rutendo Sibanda, Mashord Itayi Zenda, Happy Mathew Tirivangasi

Food security early warning in sub-Saharan Africa relies entirely on rainfall, although co-occurring hazards can produce impacts larger than the sum of their separate effects. Researchers have rarely applied compound-event framing to African food security outcomes, but harmonized sub-national data now make it testable.We test whether co-occurring growing-season moisture deficit and heat predict district-level food insecurity in Zimbabwe better than moisture deficit alone, using a panel of 84 district and urban-council units over 2009 to 2025 (1,342 district-seasons). Harmonized IPC-compatible phase classifications are combined with six-month SPEI and counts of days above each district’s own 90th percentile of daily maximum temperature. Nested two-way fixed effects models are compared on within-sample fit and leave-one-season-out error, with clustered, Conley, and permutation inference. Pooled nationally, the compound interaction is small (β = +0.030, p = 0.083), not robust to spatial dependence (Conley p ≥ 0.105; season-block permutation p = 0.35), and does not improve out-of-sample prediction. Within the driest third of districts, it is +0.116 (p < 0.001) and survives both checks (permutation p = 0.004). A pooled three-way model shows the driest third differs from the wettest (difference +0.051, p = 0.029), although the joint test across strata is weaker (p = 0.087). The compound mechanism is concentrated on the most marginal land.A single national compound-risk estimate would therefore understate risk where it is largest.Stratifying seasonal outlooks by agro-ecological zone costs little: ERA5-Land is operational and freely available, and the district-specific hot-day threshold is already a per-stratum calculation.