Reliable estimation of surface solar radiation is essential for climate analysis and solar energy planning, particularly in data-limited regions such as coastal Africa. This study investigates the long-term variability of surface solar radiation and evaluates the performance of machine learning models for its prediction using a comprehensive reanalysis dataset spanning 1940–2024. Five radiation components—net surface solar radiation (SSR), clear-sky net radiation (SSRC), downward surface solar radiation (SSRD), clear-sky downward radiation (SSRDC), and total surface radiation (TSR)—were analyzed to quantify the influence of atmospheric attenuation caused by clouds, aerosols, and water vapor. Five machine learning algorithms—Linear Regression (LR), Gradient Boosting (GB), Random Forest (RF), k-Nearest Neighbours (KNN), and Artificial Neural Network (ANN)—were implemented and evaluated using train–test split, k-fold cross-validation, and leave-one-out validation. The results reveal strong interannual and multi-decadal variability in solar radiation, with clear-sky radiation consistently exceeding all-sky radiation, confirming the dominant role of atmospheric modulation in the region. Among the tested models, Linear Regression achieved near-perfect predictive performance (R²≈ 1.0) with the lowest error statistics, indicating that surface solar radiation over coastal Africa is largely governed by linear radiative processes. Gradient Boosting and Random Forest also demonstrated high accuracy (R² > 0.98), while the Artificial Neural Network showed poor generalization due to overfitting. The findings demonstrate that computationally efficient and physically interpretable machine learning models can reliably estimate long-term solar radiation in coastal Africa. This provides a robust scientific basis for solar resource assessment, photovoltaic system design, and climate-resilient renewable energy planning across the region.
Umoh et al. (Thu,) studied this question.