Estimation of Global Solar Radiation from Extreme Temperatures Using Machine Learning Techniques

Abstract

This study aimed to model and predict the daily maximum solar radiation in the city of Puno (Peru) using meteorological variables associated with extreme temperatures through machine learning techniques. Records obtained from a DAVIS Vantage Pro 2.0 meteorological station were used, covering the period 2017–2024 and comprising 2,060 daytime observations. Three tree-based models were developed: Random Forest, XGBoost, and LightGBM, whose performance was evaluated using error and goodness-of-fit metrics, including root mean square error, mean absolute error, and the coefficient of determination. Exploratory analysis revealed a strong positive correlation between observed solar radiation and extraterrestrial radiation, confirming its role as the main physical forcing of the system. Among the evaluated models, Random Forest achieved the best predictive performance, with a root mean square error of 124.24 W/m2, a mean absolute error of 96.50 W/m2, and a coefficient of determination of 0.5135, outperforming XGBoost and LightGBM, which obtained coefficients of determination of 0.4482 and 0.3679, respectively. Variable importance analysis consistently identified extraterrestrial radiation and daily thermal amplitude as the most influential predictors. Overall, the results demonstrate that artificial intelligence approaches constitute effective and reliable tools for local solar radiation estimation in high-altitude regions, with potential applications in energy planning, environmental management, and climate studies.