*4.1. Main Findings*

In this study, we produced high-resolution results (dataset) of 10-year return levels of maximum wind speed, separately for seasons of frozen and unfrozen soil. This was done by utilizing wind speed data from ERA-Interim reanalysis, modelled soil frost data, and surface roughness and topography-based wind multiplier downscaling. Differences in wind speed return levels between seasons of frozen and unfrozen soil are of interest for practical forestry as frozen soil reduces wind damage risk in terms of uprooting of trees due to stronger tree anchorage, in opposite to unfrozen soil during strong winds. Mapping of the most exposed areas to wind damage risk could also provide support for risk management in forest planning and forestry.

Relatively small differences found in this study between the 10-year return levels of maximum wind speeds during the frozen and unfrozen soil conditions can mainly be explained by the large year-to-year variability in the part of year when the annual maximum wind speeds are occurring. In Finland, maximum annual wind speed is only rarely observed during the warmest season from June to September, and those cases are typically connected to more isolated convective weather events. The period from October to December is characterized by frequent large-scale wind storms, however, in most of Finland, the soil can still be unfrozen. During the coldest season from January onwards, the possibility to have soil frost is largest and the occurrence of high wind speeds is almost as frequent as in October to December period. As a result, datasets from where wind speed return levels were

derived for both frozen and unfrozen soil seasons were rather similar in the end, especially for spruce stands on clay or silt soil (CSS) and pine stands on sandy soil (SP). Peat is an efficient insulator and it takes longer cold periods before soil frost penetrates deeper into the soil, explaining the higher return levels of maximum wind speed for the unfrozen period in the case of pine stands on peatlands (PP).

The high resolution of wind multiplier downscaling produces also some fine local scale spatial variation in differences between seasons of soil frost. These differences can be quite large and have different sign in the opposing sides of topographic features like hills, ridgelines etc. Our results revealed that taking into account the impact of terrain variability and wind direction, the occurrence of maximum wind speed can change from frozen to unfrozen soil season.

### *4.2. Strengths and Limitations of Study Approaches*

There are multiple sources of uncertainty in our work due to the combination of results from several modelled datasets, use of statistical estimation approaches and aim to produce high-resolution information for the whole of Finland apart from the archipelago. Uncertainties related to return level estimation are somewhat increased in coastal areas. This applies especially for PP as some of the years have no frost at all, and therefore smaller dataset for return level estimation of wind speed in soil frost season. Generally, uncertainty related to 10-year wind speed return level estimates can be considered to be small (+/−1 m/s).

Also, the use of a single threshold of 20 cm for needed soil frost depth to provide sufficient support for anchorage of trees is a rather large simplification in our approach. In reality, needed soil frost depth is affected by various factors, like e.g., soil type, soil wetness, and tree species and other tree and stand characteristics. Furthermore, in the used soil frost model soil water content is assumed to be constant, therefore not representing the extreme cases of very wet or dry conditions. One rather unnecessary source of uncertainty is our use of modelled snow depth, especially as we restricted our work to the observed climate, with the possibility to use gridded observational datasets presumably having smaller uncertainties. However, our approach is justified as we are planning to take projected climate change into account in future work. In the end, [31] concluded that despite many uncertainties in soil frost modelling; their results were, in general, reasonable.

Unfortunately, point observations from operational weather stations do not provide the best possible basis for the validation of wind multiplier downscaled return levels of maximum wind speeds. Operational weather stations are usually founded on locations representing regional climatic conditions or for practicality reasons on specific locations. E.g., 17 out of 40 stations used here are located within flat and open airports and airfields for purposes of aviation. This, compared to the objective of wind multiplier downscaling, i.e., to take into account the influence of small-scale topographic and surface roughness derived variations in wind speed, does not provide an ideal starting point for validation of our results. This applies especially to our application in forestry, where the interest is focused more on areas where wind damage risk is increased. In fact, considering all 40 stations and each of eight wind directions, wind multiplier values are over 1.0, i.e., increasing the wind speed return level estimate derived from ERA-Interim reanalysis, in only 15% of the cases. More detailed and accurate validation would need a more specified measurement campaign in a more complex terrain, which would also make possible the fine tuning of the wind multiplier method. Still, our validation results indicated that wind multipliers improve the wind speed return levels derived straight from ERA-Interim reanalysis by producing return levels less biased as a whole.

The high-resolution wind speed return levels, taking into account the upwind surface roughness and topography, produced in this study may provide a valuable support for wind damage risk assessment. In wind damage risk assessment it is first calculated the critical wind speed (CWS) needed for wind damage of trees to occur based on different tree and stand properties and forest configurations [27,50–53], and further on estimated the probability of wind damage and the amount of damage, respectively, based on the probability of CWS see e.g., [52]. Our results could be used e.g., when probabilities of exceeding CWSs are assessed and how these probabilities are affected by the soil frost season.

Despite differences in wind speed return levels between soil frost seasons are small, even statistically non-significant for some forest-soil type combinations and parts of Finland when station level estimates are considered, it should be kept in mind that strong winds occurring during soil frost season are excluded from unfrozen soil season. As there was such a large year-to-year variability in the timing of annual maximum wind speeds, divided surprisingly evenly between soil frost seasons found in this study, return level estimates for both seasons are therefore smaller than ones estimated from annual maximum wind speeds. In this context, differences are expected to increase in the future climate of Finland, as according to [30], the ongoing climate change is expected to reduce frozen soil conditions by several weeks until the end of this century in Finland. This might also lead to more pronounced differences of wind damage risk between different parts of Finland.
