In this study, the Weather Research and Forecasting (WRF) model is used to produce short-term regional climate simulations with several configurations for the Carpathian Basin region. The goal is to evaluate the performance of the model and analyze its sensitivity to different physical and dynamical settings, and input data. Fifteen experiments were conducted with WRF at 10 km resolution for the year 2013. The simulations differ in terms of configuration options such as the parameterization schemes, the hydrostatic and non-hydrostatic dynamical cores, the initial and boundary conditions (ERA5 and ERA-Interim reanalyses), the number of vertical levels, and the length of the spin-up period. E-OBS dataset 2 m temperature, total precipitation, and global radiation are used for validation. Temperature underestimation reaches 4–7 °C for some experiments and can be reduced by certain physics scheme combinations. The cold bias in winter and spring is mainly caused by excessive snowfall and too persistent snow cover, as revealed by comparison with satellite-based observations and a test simulation without snow on the surface. Annual precipitation is overestimated by 0.6–3.8 mm day⁻¹, with biases mainly accumulating in the period driven by large-scale weather processes. Downward shortwave radiation is underestimated all year except in the months dominated by locally forced phenomena (May to August) when a positive bias prevails. The incorporation of downward shortwave radiation to the validation variables increased the understanding of underlying problems with the parameterization schemes and highlighted false model error compensations.

The mean climatological distribution of convective environmental parameters from the ERA5 reanalysis and WRF regional climate simulations is evaluated using radiosonde observations. The investigation area covers parts of Central and Eastern Europe. Severe weather proxies are calculated from daily 1200 UTC sounding measurements and collocated ERA5 and WRF pseudo-profiles in the 1985–2010 period. The pressure level and the native ERA5 reanalysis, and two WRF runs with grid spacings of 50 and 10 km are verified. ERA5 represents convective parameters remarkably well with correlation coefficients higher than 0.9 for multiple variables and mean errors close to zero for precipitable water and mid-tropospheric lapse rate. Monthly mean mixed-layer CAPE biases are reduced in the full hybrid-sigma ERA5 dataset by 20–30 J/kg compared to its pressure level version. The WRF model can reproduce the annual cycle of thunderstorm predictors but with considerably lower correlations and higher errors than ERA5. Surface elevation differences between the stations and the corresponding grid points in the 50-km WRF run lead to biases and false error compensations in the convective indices. The 10-km grid spacing is sufficient to avoid such discrepancies. The evaluation of convection-related parameters contributes to a better understanding of regional climate model behavior. For example, a strong suppression of convective activity might explain precipitation underestimation in summer. A decreasing correlation of WRF-derived wind shear away from the western domain boundaries indicates a deterioration of the large-scale circulation as the constraining effect of the driving reanalysis weakens.

This study evaluates snow depth (SD) from several data sources: a combined satellite-based and in situ snow water equivalent product from the Copernicus Global Land Service (CGLS), a dataset constructed from temperature, precipitation, and relative humidity using a snow model (CARPATCLIM), two state-of-the-art reanalyses by ECMWF (ERA5 and ERA5-Land), and Weather Research and Forecasting (WRF) regional climate simulations at grid spacings of 50 km and 10 km. SD observations from weather stations are used as a reference for the pointwise comparison. The study area covers the Pannonian Basin region (part of Central and Eastern Europe). Results are presented for the 2006–2010 and 1985–2010 periods. All datasets adequately reproduce the average day-to-day variation of SD but with different error magnitudes. The ERA5 reanalysis and the CGLS product represent SD remarkably well, with correlation coefficients above 0.9 and mean errors close to zero. On the other hand, ERA5-Land and CARPATCLIM overestimate daily mean SD by 2–3 cm for some stations and display lower correlations (0.7–0.9) during the 26-year time span. The WRF simulations significantly overestimate SD in the melting period (February–March). Reduction of the grid spacing from 50 to 10 km does not improve the results. The excessive snow cover might negatively impact land–atmosphere interactions in the model and lead to biases like temperature underestimation found in previous regional climate model evaluation studies. The results indicate that even in regions where snow is not a major climatic factor, SD errors can be substantial and should be considered in model evaluation and adaptation. Over the Carpathian Mountain ranges, SD from the different data sources diverges to the extent that the sign of the monthly mean model bias changes depending on the choice of the reference dataset.