Long-term wind resource assessment in Ethiopia: a hybrid machine learning framework and grid complementarity analysis

Abstract

This study provides a comprehensive spatiotemporal assessment and predictive modeling of Ethiopia's wind energy potential to address the climate vulnerability of its hydropower-dominated grid. Using NASA POWER and ERA5 datasets (2000–2024), a high-resolution hybrid framework combining physical aerodynamic modeling with machine learning (ML) XGBoost was developed. The hybrid model significantly outperformed standalone approaches, achieving strong predictive accuracy (coefficient of determination: R 2 = 0.962; mean average error: MAE = 42.8 W/m 2 ) and reducing errors by 64.4% and 14.9% relative to physical and ML models, respectively, while overcoming spatial resolution limitations. The analysis indicates a national mean annual Wind Power Density (WPD) of 348.03 ± 20.65 W/m 2 , with the Southern Lowlands and Afar Triangle identified as high-potential Class 7 wind energy zones. A strong wind–hydropower complementarity is observed, with negative correlation ( r < 0) across 92.71% of the country. Annual Seasonal Complementarity Index (SCI) values are predominantly within the optimal 0.0–0.25 range (57.13% of region), indicating substantial potential for grid stabilization. This complementarity is most pronounced during the dry season (Winter > Spring), where peak wind availability (∼385 W/m 2 ) coincides with hydrological minima. Overall, the findings demonstrate that integrating wind resources can enhance grid reliability and support Ethiopia's Climate Resilience Green Economy (CRGE) objectives through improved energy balancing and regional power system stability.