A GEOSPATIALLY-ENHANCED MACHINE LEARNING FRAMEWORK FOR SOLAR RADIATION FORECASTING IN NIGERIA

Abstract

Precise forecast of solar radiation is essential for renewable energy development, environmental sustainability, and climate-informed decision-making, particularly in tropical regions with high atmospheric variability. This study evaluates the predictive performance of two advanced machine learning models Artificial Neural Network (ANN) and Extreme Gradient Boosting (XG­-Boost) in estimating solar radiation (RADIA) using long-term meteorological data from Nigeria and Digital Elevation Model (DEM)-derived radiation values obtained via ArcGIS 10.6. The data span 29 years with major climatic predictors such as maximum temperature (TMAX), minimum temperature (TMIN), relative humidity (REHU), length of sunshine (SUNH), wind speed (WINS), (DEM), and evaporation pitch (EVPI). DEM-calculated values of ranged between 1200 to 2200 kWh per meter square per year and were divided into 5 solar potential classes. An extensive analytical model was used, which included correlation analysis, Variance Inflation Factor (VIF), residual diagnostics, predicted and observed comparisons and a variety of performance indicators. Results indicated significant relationships between solar radiation and atmospheric variables, with low VIF values confirming the absence of serious multi-collinearity. Comparative analysis showed that XG-Boost outperformed ANN, exhibiting superior residual behavior and closer alignment with the ideal prediction line. Its predictive accuracy was confirmed by strong performance metrics (MAE = 0.0282, MSE = 0.0010, RMSE = 0.0316, MAPE = 0.5412). Feature importance analysis identified maximum temperature and relative humidity as the most influential predictors. Thus, XG-Boost demonstrated strong predictive capability and reliability in forecasting solar radiation (1,200-2,200 kWh/m2/year), which can be used in the renewable energy planning, climate-wise agriculture, and sustainable policy development.