VEGA-RAD: Hybrid physical-statistical model for the daily prediction of solar radiation in the Amazon

Agreda-Vega J.F.; Sare-Lara E.; Rosales-Huamani J.A. · Revista Cientifica de Sistemas e Informatica (2026) · DOI: 10.51252/rcsi.v6i1.1454

Resumen

Daily solar radiation forecasting in the Peruvian Amazon represents a relevant challenge due to the high atmospheric variability that characterizes the region. In this study, VEGA-RAD (Vega Radiative Adaptive Dynamics) is formulated and evaluated as a hybrid physical–statistical model for daily solar radiation prediction in tropical environments. The model integrates an interpretable physical–astronomical proxy, stochastic temporal memory, and an adaptive statistical correction based on machine learning to capture residual nonlinearities. The analysis is conducted using daily ERA5 reanalysis data for the period 2017–2025, obtained through the Open-Meteo API. The results show a reduction in mean absolute error (MAE) from 1.699 to 0.477 kWh/m2/d and an increas e in the coefficient of determination (R2) from 0.635 to 0.854. These improvements are supported by paired inferential analysis (Wilcoxon) and non-parametric bootstrap resampling. In addition, conformal prediction intervals achieve coverage levels consistent with the nominal 90 % and 95 % levels, with a temporally stable average width, indicating a conservative and reliable quantification of predictive uncertainty. The proposed VEGA-RAD model is presented as a reproducible, interpretable, and robust tool for energy applications in Amazonian contexts. © Los autores.

Instituciones: Universidad Nacional de Ingeniería; Universidad Nacional de San Martín