Performance Evaluation of Solar Irradiance Models at 11 Sites in South America

Abstract

Abstract. This study provides the first continental-scale, multimodel validation of solar irradiance products in South America, comparing satellite-based datasets (CAMS, Land Surface Analysis Satellite Applications Facility (LSA-SAF)) and reanalysis products (ERA-5, MERRA-2) against high-quality ground measurements from eleven stations spanning tropical, subtropical, and high-altitude environments. By using 15- and 60-min data, we quantify accuracy through relative mean bias error (MBE), mean absolute error (MAE), and root mean square error (RMSE). To ensure spatial consistency between gridded products and point measurements, model outputs were harmonized by selecting the grid cell containing the coordinates of each ground station. Results show that CAMS and LSA-SAF consistently outperform reanalysis products, with CAMS achieving up to 18% lower relative root mean square error (rRMSE) than ERA-5 in lowland tropical regions and LSA-SAF reducing rRMSE by 30–40% relative to MERRA-2 at high-altitude Andean sites. ERA-5 exhibits moderate errors (20–30%) at mid-altitude locations, which is consistent with commonly reported uncertainty ranges for satellite-based and reanalysis irradiance products, but larger discrepancies in complex terrain, while MERRA-2 systematically overestimates global horizontal irradiance (GHI) with relative mean bias error (rMBE) exceeding +40% at several stations. All models show pronounced degradation at large solar zenith angles (>60deg). These results fill a major gap in regional literature by establishing a unified benchmark for model performance across diverse climatic regimes in South America, providing guidance for solar resource assessment and photovoltaic project planning.