Evaluation of ensemble-based numerical weather predictions of solar irradiance over Germany

Forecasting solar irradiance is essential for grid stability and solar power integration. This study evaluates the performance of ensemble-based numerical weather prediction (NWP) models, which incorporate stochastic perturbations to represent forecast uncertainty. Five operational ensemble prediction systems (EPS) and a reference model are assessed using ground-based observations from 27 stations in Germany over summer of 2024. Forecasts with lead times under 24 hours are evaluated using metrics such as the Continuous Ranked Probability Score (CRPS), rank histograms, and spread-skill relationships. All NWP models outperform the reference in accuracy, with ICON-D2-EPS being the most accurate and MOGREPS-G and IFS-ENS the most reliable. A case study using WRF-Solar EPS explores the impact of different configurations and stochastic schemes, revealing mixed results. This work introduces a visual representation of the CRPS decomposition adapted for solar irradiance forecasts, and presents a comparison of nested WRF-Solar ensemble setups. The findings highlight the value of ensemble forecasts for solar irradiance and the importance of post-processing to improve reliability in operational applications.