Air pollution and meteorological factors significantly enhance solar radiation forecasting accuracy in the western mediterranean region: a machine learning approach

Introduction Solar energy plays a critical role in meeting global energy demands and combating climate change. While meteorological factors are widely used in solar radiation forecasting models, the impact of air pollution parameters remains underexplored, particularly in the Western Mediterranean Region of Türkiye. Methods Regression models were developed using four machine learning algorithms: Support Vector Machine (SVM), Random Forest (RF), Extra Trees (ET), and LightGBM; under two scenarios: M1 (meteorological data only) and M2 (meteorological data combined with air pollution variables: PM10, PM2.5, SO 2 , CO, NO 2 , and NO X ). Data from three provinces (Antalya, Burdur, and Isparta) were analyzed. Model performance was evaluated using R 2 , RMSE, and MAPE metrics, with statistical significance assessed via Wilcoxon signed-rank tests on 10-fold cross-validation scores. Results Incorporating air pollution variables (M2 scenario) substantially enhanced forecasting accuracy across all provinces and algorithms (p = 0.001). The coefficient of determination reached R 2 = 0.86 in the best-performing models, with relative improvements exceeding 14% over the M1 scenario in Antalya. Ensemble methods (ET and LightGBM) consistently outperformed other algorithms. Feature importance analysis identified temperature and relative humidity as dominant predictors, while CO showed the highest importance among air pollution variables. Discussion These findings confirm that air pollution data constitutes an indispensable component for high-accuracy solar radiation forecasting in the Western Mediterranean Region. The results directly inform regional solar energy planning and power generation strategies, and highlight the need to incorporate air quality data into future forecasting frameworks.