Evaluation of WRF physics ensemble performance in forecasting extreme precipitation events over Germany

The effectiveness of numerical weather prediction models in forecasting precipitation and temperature extremes is highly dependent on the correct combination of the parameterization schemes as well as the grid resolutions used. For precipitation, the parameterization schemes of microphysics, cumulus and the planetary boundary layer are decisive for the correct forecast of the event. The WRF model is one of the most widely used numerical weather prediction models to estimate such extreme phenomena. In order to identify the optimal combination of these parameterizations for the European region, WRF is used here to simulate eight extreme precipitation events that occurred in the Schleswig–Holstein and Baden–Wurttemberg regions in Germany. The events were selected from the German Weather Service (DWD) catalog and exceeded DWD severe weather warning level 3 (i.e., precipitation > 40 mm/h – W3). A two-way nesting approach is used with 9 and 3 km spatial resolutions. The initial and boundary conditions are obtained from the ERA5 dataset at 0.25° × 0.25° resolution. To model each event, thirty different parameterization configurations were used, accounting for all possible combinations of five microphysics (MP), three cumulus (CU), and two planetary boundary layer (PBL) parameterization methods, yielding a total of 240 simulations. To determine the performance skill of each setup, the multi-criteria decision analysis Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) is employed. Six categorical and five statistical metrics are used as input in the TOPSIS algorithm to calculate each member's performance rank. The analysis conducted here revealed that an increase in the grid spatial resolution from 9 km to 3 km did not result in a substantial improvement in the accuracy of the forecast in time or in the estimation of precipitation intensity. When considering each event individually, the optimal combination, according to the TOPSIS ranking algorithm, is seasonally and geographically dependent, with specific members appearing more frequently in the top-ranking positions. When all events are treated as one to determine the best performing simulating members, the Morrison double-moment (MDM) scheme, along with the Grell-Freitas (GF) CU and the YSU PBL schemes, is found to be the most effective set up, followed by the WRF single-moment 5-class, along with the GF and the YSU schemes.