Evaluating the performance of a WRF initial and physics ensemble over Eastern Black Sea and Mediterranean regions in Turkey

The physics- and initial-based ensemble approach for Weather Research and Forecasting (WRF) model is applied for predicting four different extreme precipitation events that occurred in summer and autumn over the two most flood-prone regions of Turkey, namely; the Eastern Black Sea (EBS) and Mediterranean (MED). A total of 48 runs, each of which includes two nested domains, is designed considering four microphysics (MP), three cumulus (CU), two planetary boundary layer (PBL) schemes, and two initial forcing datasets (ERA5 and Global Forecast System (GFS)) for each event. The forecast skill of the runs is measured through the hierarchic application of the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) method using five categorical and four statistical metrics. The best ten members for each event are validated on the independent events. The MP scheme is the most influential on precipitation estimates between the schemes, especially for the autumn events. However, the choice of a specific MP scheme varies with season and region. Similarly, PBL and CU schemes show a high dependency on event timing and location. Forecasts based on ERA5 yield the highest accuracy over EBS, while GFS-based forecasts are better in the MED region. Most of the top ten members are represented by runs from the 9-km domain. The model parameterization generally shows a more significant impact than model initiation on precipitation variability through spatiotemporal runs. Also, the dynamic downscaling of ERA5 through the WRF model indeed produces better precipitation estimates than ERA5 products over topographically complex regions. •The sensitivity of WRF-precipitation to the initial data and model physics is tested.•Model parameterization is more influential on the WRF outputs than the initial data.•The best model is variable with respect to the spatiotemporal features of the events.•The locally-run WRF improves the precipitation of the initial data source, ERA5.