Assessment of Global Numerical Weather Prediction Models for Short-Term Solar Irradiation Forecasting in Southeastern South America
Abstract
The accuracy of solar forecasts improves power system management, including energy dispatch and reserve handling. This work evaluates the performance of five numerical weather prediction models over southeastern South America over a three-year period. The first 24 hours of forecasts are analyzed at daily and hourly scales, taking into account seasonal and cloudiness discrimination. The models’ performance shows little variation across geographical locations or seasons, and forecast accuracy deteriorates with increasing cloudiness. ECMWF shows the best performance with a daily rRMSD of 16.7%, despite underestimating solar radiation under clear skies. ICON follows with 18.0%, showing better behaviour under clear sky conditions. NEMS30 and GFS reach 21.0% and 21.9% respectively, with GFS underestimating cloudiness more and NEMS30 performing better under cloudy conditions. These results provide the first detailed evaluation of model performance in the region and highlight opportunities for improvement through post-processing techniques.
Keywords
Solar forecast, GHI, NWP, Global Models