Investigation of Drought Propagation Dynamics and Cereal Crop Yield Prediction using Machine Learning Models under Global Warming in Ethiopia

Meteorological drought is characterized by both magnitude and duration, and its spatiotemporal dynamics drive its propagation to other drought type. Ethiopia’s agricultural sector faces significant challenges in achieving accurate crop yield prediction, a critical requirement for effective resource management and ensuring food security for its rapidly growing population. This study analyzes drought propagation dynamics across Ethiopia for 1981-2022 period and their impacts on cereal crop yields-maize, sorghum teff and wheat over 1993-2022. Multi-scale meteorological drought indices (SPI, SPEI at 1-12 months) from observed and ENACT rainfall and temperature data were integrated with root-zone soil moisture anomalies to map spatio-temporal propagation. Lagged correlation and maximum-correlation lag times quantified lead times between meteorological and soil moisture droughts. Random Forest and XGBoost models were applied for yield prediction, evaluated using RMSE and R2. Results indicate spatially variable drought regimes, with rapid onset in lowlands and slow cumulative patterns in highlands, alongside crop specific vulnerability periods. Maximum temperature and medium-term drought consistently dominated as predictors, highlighting combined heat and moisture stress effects. Recommended adaptation include breeding for heat-tolerant maize and wheat, implementing water-saving agronomy and supplemental irrigation, adjusting planting dates to avoid peak heat, and diversifying cropping systems to mitigate climate risks.