Propagation of rainfall dataset differences into crop model outputs and economic indicators

Gridded precipitation products (GPPs) are widely used in agricultural analyses in data-scarce regions, yet their suitability is often assessed using meteorological performance alone. This study evaluates how differences among six GPPs (CHIRPS, CPC, ERA5, MERRA2, MSWEP, and TAMSAT) propagate into simulated maize yield and economic indicators, using the Agricultural Production Systems sIMulator (APSIM) in two contrasting semi-arid systems, Kaffrine (Senegal) and Kongwa (Tanzania). By linking GPP evaluation to crop-model propagation and economic analysis within a single framework, this study offers a more comprehensive basis for assessing GPP suitability than meteorological validation alone. Rainfall performance varied across indicators and sites, with higher agreement for seasonal totals than onset, and higher detection skill for dry days than moderate or heavy rainfall. Several GPPs over-represented light rainfall and under-represented heavy rainfall, reflecting spatial averaging. These differences translated into site-specific yield responses driven by early-season rainfall: in Kaffrine, yield and nitrogen responses remained similar across datasets (KGE = 0.05–0.35), whereas in Kongwa, yield variability and optimal input levels were more sensitive to dataset choice (KGE = −0.24–0.60), particularly under wetter conditions. Across sites, yield response shapes to nitrogen and planting density remained similar, but economically optimal input levels varied across datasets, indicating that agreement in rainfall indicators alone is insufficient to infer stability of crop model outputs. Evaluating GPP suitability for agronomic applications requires crop model diagnostics alongside meteorological validation, with explicit environmental context, as demonstrated here within a single process-based model (APSIM) across two semi-arid maize systems.