**1. Introduction**

In the last few decades, wind storms have caused the most damage and economic losses in European forests, compared to all abiotic and biotic damage agents [1–4]. So far, winter storms have caused the most destructive damage in Western and Central Europe [3,5,6], e.g., storms like Vivian in 1990 (over 100 million m<sup>3</sup> of timber), Lothar and Martin in 1999 (over 175 million m3), Kyrill in 2007 (54 million m3) and Klaus in 2009 (50 million m3), respectively. Damages have increased in recent years also in northern Europe [4,6,7], where in 2015 Gudrun damaged 70 million m<sup>3</sup> and in 2007 Per damaged 12 million m3 of timber, mainly in Sweden. In Finland, over 25 million m<sup>3</sup> of timber has been damaged during storms since 2000, the most in autumn storms in 2001 (Pyry and Janika, 7.3 million m3) and in summer storm in 2010 (Asta, Veera, Lahja and Sylvi, 8 million m3), respectively. The increasing amount of damages in European forests may at least partially be explained by increasing volume of growing stock and changes in forest structure (e.g., age, tree species) related to changes in forest management practices [1,5,8,9]. Forest disturbances may also amplify or even cancel out the expected increase in productivity of forests under changing climate [4,10].

Some recent studies indicate increased storminess for some regions in Europe (see e.g., review by [11]). However, the majority of studies point towards decadal variation in storminess without any clear trend for a direction or another [12–15]. In Finland, slight weakening of annual mean (−0.09 ms−<sup>1</sup> decade<sup>−</sup>1) and maximum (−0.32 ms−<sup>1</sup> decade<sup>−</sup>1) wind speeds across 33 weather stations have been observed in the period of 1959–2015 [16], which is in accordance with widespread weakening of terrestrial near-surface wind speeds [17,18]. For future projections, the change in the extreme wind speed during the coming decades is still a somewhat unsolved issue and the outcome is largely dependent on the climate model used for the simulation [11,19,20].

However, the risk of wind damage to forests may still increase in Northern Europe under climate change even if the frequency and severity of wind storms do not increase. This is due to the shortening of the frozen soil period, which improves tree anchorage during the windiest season of the year from late autumn to early spring [21–24]. Moreover, storms may be accompanied by heavier rainfall, leading to more saturated soils and increased risk of wind damage [5]. When estimating the forest wind damage risk it is thus essential to know whether the extreme wind speeds occur during the frozen or unfrozen soil conditions. Typically, the windiest season in Finland is from October to March [25] and soil frost season starts in October–November and ends in April–May [26]. However, there is large year-to-year and regional variation in soil frost duration.

Even a 20 cm thick frozen soil increases the anchorage of trees and reduces substantially the risk of uprooting [27,28]. According to tree-pulling experiments in Finland, under frozen soil the type of failure was stem breakage, whereas under unfrozen soil conditions, about 80% of trees uprooted, respectively [28]. From the three economically and ecologically most important boreal tree species in Finland, Norway spruce (*Picea abies*) with the shallow rooting is the most vulnerable to uprooting, followed by Silver and Downy birches (*Betula pendula* and *Betula pubescens*), and Scots pine (*Pinus sylvestris*), respectively [27,28]. However, from late autumn to early spring, birches (without leaves) are not vulnerable to wind damage and therefore excluded from this study.

For snow free surfaces, the soil frost modelling can be done using cold season frost sum and soil characteristics alone [29]. The presence of snow and vegetation complicates the modelling, and requires a more sophisticated approach including the modelling of heat and water transfer [23]. An example of a relatively simple approach accounting for the main controlling factors was published by [30]. It was further developed and tested in the Finnish conditions by [31] for the calculation of soil temperatures in three common combinations of soil and forest types in Finland, i.e., dense Norway spruce stands on clay or silt soil, Scots pine stands on sandy soil, and Scots pine stands on drained peatlands. Soil frost conditions can vary a lot, even up to few months in mean duration, depending on soil type. Peat is effective insulator compared to mineral soils, therefore having shorter soil frost periods in similar climatic conditions [31].

The estimation of the return levels of maximum wind speed values (extreme winds) can be done using observational data representing conditions at the observing station location or using reanalyzed data like ERA-Interim [32], representing a larger area's averaged value, respectively. When studying the high-resolution spatial variation of extreme winds, the data has to be either downscaled from the reanalyzed coarse grid to a local value or upscaled from station point observations to areas located between the stations. Downscaling can be done by applying various spatial statistical tools, e.g., [33,34], or complex airflow models like e.g., WAsP [35], which are typically applied for wind power potential predictions. GIS-based methods for mapping the areas having highest wind damage risk have also been introduced, e.g., [36–38]. One computationally feasible approach for the estimation of the return levels of extreme wind speeds for large geographical areas with very high spatial resolution is the wind multiplier approach [39–41]. In this method, return levels obtained, e.g., from the reanalysed data, are downscaled to local wind speeds with help of land cover (roughness) and topography data. By applying GIS-tools such as ArcGIS, QGIS or R, it is rather straightforward to produce the required multipliers.

The reliable high-resolution information on the spatial variation of extreme wind speeds can enhance wind damage risk management in forest planning and forestry. In the above context, the objective of this study was to produce spatially detailed estimates (maps) of the 10-year return level maximum wind speed under current climate for unfrozen and frozen soil conditions in some of the most common combinations of forest and soil types in Finland. By utilizing soil frost calculations of [31] to determine the duration of soil frost seasons, the wind speed return level calculations were done for dense Norway spruce stands on clay or silt soil, Scots pine stands on sandy soil, and Scots pine stands on drained peatland. The coarse resolution estimates of the 10-year return level of maximum wind speed were based on 1979–2014 ERA-Interim dataset [32]. Downscaling to a 20 m grid was done by applying the wind multiplier approach [41].
