Abstract. Empirical high-resolution surface wind fields, automatically generated by a weather diagnostic application, the WegenerNet Wind Product Generator (WPG), were intercompared with wind field analysis data from the Integrated Nowcasting through Comprehensive Analysis (INCA) system and with dynamical climate model wind field data from the non-hydrostatic climate model COSMO-CLM. The INCA analysis fields are available at a horizontal grid spacing of 1 km x 1 km, whereas the COSMO model fields are from simulations at a 3 km x 3 km grid. The WPG, developed by Schlager et al. (2017, 2018), generates diagnostic fields at a high resolution grid of 100 m x 100 m, using observations from two dense meteorological station networks: The WegenerNet Feldbach Region (FBR) and its alpine sister network, the WegenerNet Johnsbachtal (JBT). The high-density WegenerNet FBR is located in southeastern Styria, Austria, a region predominated by a hilly terrain and small differences in altitude. The network consists of more than 150 meteorological stations. The WegenerNet JBT contains eleven meteorological stations at elevations ranging from about 600 m to 2200 m in a mountainous region in northern Styria. The wind fields of these different empirical/dynamical modeling approaches were intercompared for thermally induced and strong wind events, using hourly temporal resolutions as supplied by the WPG, with the focus on evaluating spatial differences and displacements between the different datasets. For this comparison, a novel neighborhood-based spatial wind verification methodology based on fractions skill socres (FSS) is used to estimate the modeling performances. All comparisons show an increasing FSS with increasing neighborhood size. In general, the spatial verification indicates a better statistical agreement for the hilly WegenerNet FBR than for the mountainous WegenerNet JBT. The results for the WegenerNet FBR show a better agreement between INCA and WegenerNet than between COSMO and WegenerNet wind fields, especially for large scales (neighborhoods). In particular, COSMO-CLM clearly underperforms in case of thermally induced wind events. For the JBT region, all spatial comparisons indicate little overlap at small neighborhood sizes and in general large biases of wind vectors occur between the dynamical (COSMO) and analysis (INCA) fields and the diagnostic (WegenerNet) reference dataset. Furthermore, gridpoint-based error measures were calculated for the same evaluation cases. The statistical agreement, estimated for the vector-mean wind speed and wind directions show again a better agreement for the WegenerNet FBR than for the WegenerNet JBT region. In general, the difference between modeled and observed wind directions is smaller for strong wind speed events than for thermally induced ones. A combined examination of all spatial and gridpoint-based error measures shows that COSMO-CLM with its limited horizontal resolution of 3 km x 3 km and hence, a too smoothed orography, is not able to represent small-scale wind patterns. The results for the JBT region indicate that the INCA analysis fields generally overestimate wind speeds in the summit regions. For strong wind speed events the wind speed in the valleys is underestimated by INCA, however. Regarding the WegenerNet diagnostic wind fields, the statistics show decent performance in the FBR and somewhat overestimated wind speeds for strong wind speed events in the Enns valley of the JBT region.
Statistical Downscaling of Gridded Wind Speed Data Using Local Topography
