*3.4. Comparison of the 10-Year Return levels of Maximum Wind Speed between OBS vs. ERA and WM*

Figure 7 show that even if there is still large variability and some systematic biases, the majority of the OBS and WM comparisons are closer to 1:1 line than the corresponding OBS and ERA comparisons. Also in Figure 8, all three statistics show improvement when comparing differences between OBS and WM to differences between OBS and ERA. Taken over all the comparisons, the D statistic of two-sample Kolmogorov–Smirnov test is decreasing from 0.543 to 0.195, R2 of the linear regression increasing from 0.228 (95% confidence interval 0.196–0.262) to 0.320 (95% confidence interval 0.286–0.355), and the mean difference decreasing from −1.96 to −0.77 (Mann–Whitney *U*-test *p* < 2.2e-16).

**Figure 7.** Scatterplot comparisons of 10-year return levels of wind speed between OBS and ERA (black) and between OBS and WM (red) at the station locations of 40 weather stations, including eight wind directions, three soil types, and two states of soil frost. OBS stands for return levels derived from observed weather station data, ERA for return levels derived straight from ERA-Interim reanalysis, and WM for ERA-Interim derived return levels downscaled with wind multiplier approach.

**Figure 8.** Boxplots of 40 station level comparisons of D statistic of two-sample Kolmogorov–Smirnov test (left), mean differences (middle), and coefficient of determination R<sup>2</sup> (right) between return levels derived from observations and ERA-Interim reanalysis (ERA) and between observations and wind multiplier downscaling (WM).

At some station locations, the application of wind multipliers led to more biased return levels compared to ones derived straight from the ERA-Interim. When considering all the 40 stations and each of three comparison statistic, 23% of the cases showed deterioration of results. However, for only three of the 40 stations, all three comparisons statistics were worsening (not shown).

A majority of the most pronounced overestimations of WM compared to OBS above return levels of 15 m/s (Figure 7) were from a single station at relatively high altitude combined with a large lake in the direction of the largest overestimations. There was also some similarity in the stations characterized by restricted openness, for which wind multipliers overestimated return levels around 10 m/s. For example, the station producing largest overestimations around return levels of 12–13 m/s was located in the residential area on the top of a hill.

#### **4. Discussion**
