Evaluation of ensemble prediction systems in terms of global horizontal irradiance in Germany

In solar power forecasting, the consideration of uncertainties in the forecast is becoming increasingly important and is being taken into account more and more. Ensemble predictions systems can provide information about the uncertainty of cloud cover and solar irradiance. The goal of this study is to evaluate the quality of different operational ensemble predictions systems (EPS) in Germany in terms of global horizontal irradiance (GHI). Therefore, the models ICON-EU-EPS (horizontal grid spacing approx. 20 km, 40 members), IFS-ENS (horizontal grid spacing approx. 20 km, 50 members), GEFS (horizontal grid spacing approx. 28 km, 30 members), and WRF-Solar-EPS (horizontal grid spacing 3 km, 30 members) are evaluated. The 00UTC run is considered for all models. The evaluation period is summer 2022 with observational GHI data for 25 sites from Deutscher Wetterdienst (DWD). A quality check of the observational data with the python library libinsitu reveals that more than 95% of the observational data is of high quality. For reasons of computational time, WRF-Solar-EPS forecasts are evaluated for selected days only. The forecast quality is evaluated using rank histograms and the continuous ranked probability score (CRPS) among other metrics. The results show that the quality is similar for all models, while WRF-Solar-EPS does not necessarily stand out despite its higher spatial resolution. The rank histograms reveal that all models are highly over-confident, meaning that most observations lie outside of the range of all members. Post-processing methods and model calibration is not part of this study but has the potential to increase the forecast quality.