Decoding Cropland Mask Effects on the Explanatory Power of Remote Sensing and Reanalyzed Climate Data on Yield Anomalies in Africa

Abstract

Ensuring crop yield stability is crucial for food security in Africa, where agriculture faces increasing food demand amid considerable vulnerabilities. Remote sensing and reanalyzed data products offer the potential for capturing crop growth dynamics and understanding their drivers. However, the impacts of cropland masks on relative yield anomalies (RYA) and the contributions of variables across Africa and crops remain unclear. This study explores the explanatory power of air and land surface temperatures (AT and LST), precipitation, evapotranspiration, and soil moisture on maize, millet, and sorghum RYA in Africa for 2001–2020 under seven cropland masks with distinct configurations for temporal, crop type, and water supply systems. Results indicate that (a) North Africa was particularly affected by soil moisture variation and evapotranspiration, West Africa was strongly impacted by precipitation, Central and East Africa were highly influenced by mean AT and total precipitation, and South Africa was mainly affected by high LST, mean evapotranspiration, and precipitation variation. (b) Interactions between precipitation and LST improved the explanatory power of the multiple stepwise regression model from 67% to 73%, while that of the random forest model considering complex variable interactions reached 83%. (c) Variables with high contributions were less impacted by the choice of masks. Mask configurations with broader crop coverage compensated for the limitations of temporally static masks, while crop type identification enhanced explanatory power when using year-specific and crop-specific maps. Future research should integrate process-based crop models to better understand the mechanisms behind the diverse drivers of yield at the regional scale in Africa.