*2.4. Comparison of Return Levels Derived from Point Observations, Reanalysed Data, and Wind Multiplier Downscaled Data*

The 10-year return levels of maximum wind speeds calculated from the observations (OBS) of 40 weather stations (Figure 1a) were compared to corresponding return levels calculated from the original ERA-Interim reanalysed data (ERA) and return levels downscaled with wind multiplier approach (WM). Weather station location coordinates were used to derive data from ERA and WM gridded datasets.

Besides general visual scatterplot comparison of ERA and WM values to OBS values, also the coefficient of determination R2, D statistic of two-sample Kolmogorov–Smirnov test, and mean differences were used to analyse the performance of ERA and WM to produce return level values similar to OBS. These statistics were also analysed at the station level. Comparisons were considered more at a qualitative than quantitative level, i.e., are return level values produced with WM approach improvement to original ERA values when considering similarity to values derived from weather station observations.

The D statistic of the two-sample Kolmogorov–Smirnov test was used as a measure of similarity of ERA and WM to OBS. Smaller values of D are considered as a good result, i.e., EDF (empirical distribution function) of WM is more similar than EDF of ERA compared to EDF of OBS. Mean difference statistic used was simply the mean of differences between ERA and WM to OBS at station level, over all the combinations of eight wind directions, three soil types, and distinction between frozen and unfrozen soil. Again, smaller values were considered as a good result as a difference between WM and OBS is smaller than a difference between ERA and OBS. R<sup>2</sup> was used as a goodness of fit of a simple linear regression between OBS and ERA or WM. Here, increasing R2 was considered as an improvement when comparing a regression of OBS and WM to a regression of OBS and ERA.

#### *2.5. Structure and Restrictions of Data Analyses*

For deeper understanding of results for calculated return levels of maximum wind speeds and their differences between frozen and unfrozen soil, we first considered independently underlying soil frost conditions (e.g., number of soil frost days and duration of soil frost) and wind conditions (e.g., timing of maximum wind speeds throughout year and between frozen and unfrozen seasons), respectively.

We also restricted our analysis to mainland Finland (see Figure 1a). The reasoning for this is the lack of years with soil frost in the archipelago, leading to increased uncertainty in the calculation of wind speed return levels. Also, the insufficient performance of wind multiplier method for the small Baltic Sea islands found by [41] supports our decision.

The territory of Finland was moreover divided into three sub-regions in the analysis of results. The three sub-regions were based roughly on the mean annual growing degree day sum (GDD) calculated using the threshold of 5 ◦C. The limits are GDD > 1200 ◦C days for southern, 1000 ◦C days < GDD ≤ 1200 ◦C days for central and GDD ≤ 1000 ◦C days for northern sub-regions, following also roughly the borders of boreal subzones.

Also, one smaller area (30 × 30 km) from northern Finland (Figure 1a) with a more complex topography (Figure 1b) was used to examine and present the more local scale behavior and influences of wind multiplier downscaling to 10-year return levels of wind speeds and differences between frozen and unfrozen soil seasons.

**Figure 1.** (**a**) Locations of 40 weather stations (black dots), division of the Finland into three parts, and location of detailed study area (square with black borders). (**b**) Topography and elevation (meters above sea level) of detailed study area.

#### **3. Results**
