Features of Linear Convective Windstorms That Determine Their Impact on Northern Eurasian Forests

Abstract

Severe linear convective windstorms (SLCWs) account for 66% of the total windthrow area in Northern Eurasian forests. However, in many cases, these events do not result in forest damage. The aim of this study is to reveal the features of storms that determine whether or not they cause forest damage. The study examines the relationship between windthrow occurrence and the characteristics of SLCW (seasonality, wind gusts and accompanying rainfall), as well as their formation environments. The sample includes 351 SLCW events that occurred in Northern Eurasian forests between 1986 and 2024. These are subdivided into two subsamples: 181 SLCW events with wind gusts of ≥25 m s⁻¹, but without substantial damage to forests (SR_ND), and 170 SLCW events associated with windthrow (SR_WD). Since the subsamples are similar in terms of forest stand characteristics, their differences are likely linked to differences in the characteristics of the SLCWs themselves. In general, SR_WD events are accompanied by stronger wind gusts and rainfall than SR_ND events. The difference in rainfall amounts is more pronounced in the Integrated Multi-satellitE Retrievals for Global Precipitation Monitoring (GPM IMERG) satellite-derived data than in the data from weather stations. Springtime events contribute substantially more to SR_ND (26%) than to SR_WD (12%). According to the ERA5 reanalysis, SR_WD events form in conditions of greater thermodynamic instability and stronger wind shear than SR_ND events, i.e., under conditions that are generally more favorable for more severe windstorms. Obtained results can be further used to assess probable changes in forest damage caused by SLCW events based on projections of rainfall intensity and convective variables in a changing climate.

1. Introduction

Severe winds are among the most destructive abiotic disturbance agents affecting boreal and temperate forests worldwide [1,2,3]. Large-scale stand-replacing windthrow events have long-lasting consequences for forest ecosystems and society [4]. Windthrow events are typically induced by cold-season extratropical cyclones, severe convective storms, tropical cyclones, orographic winds [3,5] or the combined impact of windstorms and wet snow accumulation [6].

In Russia, convective winds, especially long-lived straight-line windstorms, are the primary cause of forest damage. Windthrow events caused by non-tornadic convective winds account for 66% of the total area of windthrow associated with all types of weather events [7,8,9]. Other convective events, such as microbursts [10] and tornadoes [11,12], also cause windthrow, but they are less damaging to forests due to their more localized nature.

An important research question is why some severe linear convective windstorms (SLCWs) cause widespread damage to forests while others with similar wind speeds only result in very local windthrow or no damage at all. Xi et al. [13] showed that wind speed and precipitation explain forest damage variation at the regional scale, while topographic, site and stand factors have key importance at the landscape scale. Numerous studies have considered the factors that predict forest vulnerability to the impact of wind, including the properties of forest stands themselves, as well as site characteristics and soil properties, by employing statistical approaches [14,15,16,17,18,19] or individual tree models [20,21]. In particular, it has been found that the age of stands and the amount of growing stock are the most important predictors of forest damage [19]. The probability of damage increases with a higher percentage of conifers and greater stand height [14]. Stand fragmentation can also affect the extent of the damage caused [22], e.g., it substantially increases the damage to trees growing near newly cleared forest areas [18]. Terrain properties, like slope exposure, windward/leeward index, convexity and valley depth, modify the wind flow by either increasing or decreasing the wind speed [5,19,23]. Soil properties like depth, texture, bulk density, moisture content, and organic matter content also influence the ability of forests to withstand wind disturbances [15,24].

Several studies have examined the contribution of atmospheric characteristics to the occurrence and degree of forest damage. A strong correlation was found between the degree of forest damage and the maximum observed wind gusts for non-convective storms [7,25]. In general, the total destruction of forest stands is associated with wind gusts exceeding 35 m s⁻¹ [26] or 40 m s⁻¹ [27]. According to the International Fujita Scale [28], total destruction of a forest canopy indicates the impact of hurricane-force winds (>33 m s⁻¹) or IF1+ damage in the case of convective storms.

The characteristics of precipitation that accompanies a windstorm, or the soil moisture that forms in the preceding days, may also influence windthrow occurrence. Indeed, heavy rainfall associated with storms can rapidly increase soil moisture, making trees more prone to windthrow [7,29,30]. Gardiner et al. [31] found that trees are most susceptible to wind damage immediately after heavy rainfall. A correlation between extreme rainfall and windthrow occurrence has been established in tropical forests [32,33,34] but is less studied in temperate and boreal forests.

In Northern Eurasia, the relationship between storm event characteristics events (wind gusts and precipitation) and windthrow occurrence has only been analyzed in a few areas, including the Central Forest Reserve [35], the Middle Ural [36], and the Perm Region [37]. These studies were performed over substantially different time intervals and were based on non-standardized data collection methodologies. Overall, the meteorological characteristics of SLCW events that determine whether they cause forest damage in Northern Eurasia remain largely unexplored. However, the availability of information about windthrow-causing weather events in the Russian forest zone [8,9] opens up opportunities for identifying relationships between storm event characteristics (and their formation environments) and related forest damage.

This study examines the characteristics of SLCW events that determine whether they cause forest damage. Our focus is primarily on meteorological variables, which have not been analyzed in previous studies in terms of their contribution to windthrow occurrence. For this reason, the storm reports considered were selected to minimize differences in the characteristics of the impacted forest stands.

In Section 2, we present information on the data collection and categorization of SLCW events into two groups: those with and without windthrow (SR_WD and SR_ND respectively). We then describe the characteristics of these groups, which were obtained using various data sources, including ground-based and satellite observations, as well as the ERA5 reanalysis data. Section 3 presents the main results of the study, including an analysis of the differences between the two groups of SLCW events and their interpretation. Section 4 discusses and summarizes the obtained results.

2. Materials and Methods

2.1. Compilation of SLCW Events Sample

We compiled a dataset of SLCWs in the Northern Eurasian forest zone based on reports from Russian weather stations and a previously compiled database of windthrow events [8,9] for the period from 1986 to 2024 (Figure 1).

In the first step, we selected the weather stations located within the forest zone. In particular, the proportion of forest-covered area within a 25 km and 100 km radius around each weather station was calculated according to the Vegetation Map of Russia [38], and a 25% threshold value (for 25 km radius) was used. A total of 844 weather stations in Russia’s forest zone (out of more than 1600 weather stations) meet this criterion and were considered in the subsequent analysis (Figure 1).

Next, we selected storm reports from these 844 weather stations based on routine 3-hourly observations provided by the All-Russian Research Institute of Hydrometeorological Information—World Data Center (RIHMI-WDC) [39]. In particular, we considered the variables ‘maximum wind gust over a three-hour period between observations’. The cases with wind gusts ≥ 25 m s⁻¹ were selected as storm reports in line with criteria of the Russian weather service (Roshydromet) for a hazardous wind event. To exclude non-convective winds, only warm-season reports (April–September) accompanied by thunderstorms were selected (Figure 1). It should be noted that routine 3-hourly observations from only about 200 of 844 selected weather stations were available for the entire period from 1986 to 2024. Other stations provided a complete data archive for the period 2011–2024 only. For these stations, we also compiled SLCW events from the monthly hazardous weather event reviews published in the Russian Meteorology and Hydrology journal (http://mig-journal.ru/en/archive-eng, last access: 15 December 2025). These reviews mention numerous convective windstorms reported at weather stations, along with additional information about whether the storm was convective or non-convective.

In this paper, we interpret individual observations at stations as separate SLCWs, even if these observations are associated with one large, long-lived windstorm, such as a derecho. Combining reports into a single event would contradict our objectives, since a single long-lived event may be associated with SLCWs that may or may not be accompanied by windthrow. For instance, such a situation was observed in southern Siberia in May 2020 [40]. In this paper, therefore, we will treat a report from one weather station as a single SLCW event. If a weather station reported wind gusts in two consecutive 3 h periods, only the report with the highest value (and its corresponding time) is treated as an SLCW event. Therefore, the terms ‘SLCW reports’ and ‘SLCW events’ are used interchangeably throughout this paper.

In total, we compiled a dataset of 249 SLCW events with observed wind gusts ≥ 25 m s⁻¹ in the forest zone. The timing of these events is known to within three hours, except in cases where the exact time of occurrence is known from the aforementioned monthly reviews.

Then, we considered the windthrow event database [8,9] to select SLCWs that were both reported by weather stations and caused forest damage. The database currently includes 2777 stand-replacing windthrow events. Of these, we considered 668 windthrow events that were caused by SLCWs with a known storm event date. Next, we retained only the 540 windthrows for which there was at least one weather station within a 100 km radius. We then analyzed wind gust observations at these weather stations on the dates of windthrow occurrence. We only retained observations with wind gusts ≥ 15 m s⁻¹, since the Russian Weather Service defines this as the threshold for an adverse meteorological phenomenon. Images from meteorological satellites (MVIRI and SEVIRI/Meteosat, AVHRR/NOAA and MODIS/Terra and Aqua) and weather radar data were used to clarify the timing of windthrow events and verify whether the wind reports from weather stations were related to the same convective storm (mesoscale convective system) that caused forest damage [8,9]. If several stations located near windthrow reported wind gusts between 15 m s⁻¹ and 25 m s⁻¹, we only kept the report of the strongest wind gust as one SLCW event. Conversely, if two or more stations reported wind gusts of ≥25 m s⁻¹, all these reports were included in the analysis as separate SLCW events (Table 1).

In total, we selected 170 reports with wind gusts ≥ 15 m s⁻¹ associated with windthrow events (SR_WD) for further analysis. Of these, 68 reports with wind gusts ≥ 25 m s⁻¹ were selected at the previous stage (Figure 1).

The resulting compiled dataset consisted of 181 SR_ND events and 170 SR_WD events. These 351 SLCW events occurred between 3 June 1986 and 5 July 2024 at 227 different weather stations in the Russian forest zone. A total of 142 weather stations reported one event, 57 stations reported two events, 21 stations reported three events, 5 stations reported four events, and 2 stations reported five and seven events, respectively. Weather stations that reported multiple events may report strong winds more frequently because they are located on a hill or in an open area (away from buildings or trees).

Figure 1. The workflow showing the compilation of the SLCW events sample. The data sources are [38] for the Vegetation Map of Russia; refs. [39,41] for the coordinates of weather stations’ routine 3-hourly observations; refs. [8,9] for the database of windthrow events.

Figure 1. The workflow showing the compilation of the SLCW events sample. The data sources are [38] for the Vegetation Map of Russia; refs. [39,41] for the coordinates of weather stations’ routine 3-hourly observations; refs. [8,9] for the database of windthrow events.

2.2. Determination of Forest Stand Characteristics for SLCW Events

The vulnerability of forests to the impact of wind is determined by several forest stand characteristics, primarily forest species composition, stands age and volume [2,14,15,16,17,18,19]. In this study, we estimated the percentage of forest-covered area, percentage of main forest types and the average growing stock averaged within a 25 km and 100 km radius around each weather station that reported storm events, based on satellite-derived products (Table 1). We compared their values for SR_WD and SR_ND to find the differences between these subsamples and assess their statistical significance.

Assessing the characteristics of forest stands has several limitations. In particular, dominating forest species and growing stock were defined for a fixed year (the growing stock was determined for 2010, while the dominant species were determined for 2023), whereas the storm events occurred in different years. However, high-resolution data for each analyzed year is currently unavailable. In addition, data on the age of forest stands, which determines vulnerability alongside species composition and growing stock, is also unavailable for the entire study region.

Since many of the considered SLCWs events occurred in spring, we also estimated the contribution of the seasonal state of vegetation (particularly deciduous trees) to windthrow. We estimated the average value of the Leaf Area Index (LAI) for high vegetation based on the ERA5-Land reanalysis data (0.1° cell size) [42]. LAI values were obtained for each date of SLCW occurrence, averaged over a 25 km radius around each weather station that reported severe wind, and compared for the SR_WD and SR_ND subsamples and for events occurring in different months of the year.

2.3. Determination of SLCW Event Features

2.3.1. Wind Gusts

Characteristics of storm events (Table 1) were determined with ground-based and satellite observations, as well as the ERA5 reanalysis data, to identify the differences between SR_ND and SR_WD in terms of wind gusts, precipitation and convective variables.

Both for SR_ND and SR_WD, observed maximum wind gusts were determined. For SR_WD, the location of the weather station in relation to the windthrow area determined the probability that the storm was reported or missed at the weather station. There were only 15 events where the distance between the weather station and the windthrow area was <2 km. However, observations can still be representative at a greater distance if a weather station was located in the path of the storm (i.e., behind or in front of a forest damage track; see Figure 2).

The SR_WD reports were subdivided into two subsamples, given the location of weather stations related to the windthrow and general storm propagation direction. The storm propagation direction was determined using a time series of satellite images (mainly MVIRI and SEVIRI/Meteosat) if the direction of the windthrow itself was not evident, in line with [43]. Storm reports from weather stations located near a windthrow area are referred to as SR_WD1 if the angle between the direction of storm propagation and the direction from the nearest points in the windthrow area to the weather station did not exceed 30° (red triangles at Figure 2). On the contrary, the reports from weather stations located on the side of a storm path—that is, where the angle between the direction of the storm path and the direction from the nearest points of the windthrow area to the weather station exceeded 30°—are referred to as SR_WD2 (yellow triangles at Figure 2). SR_WD2 events are generally less representative than SR_WD1 events, especially with regard to local events. The 30° angle criterion used to distinguish between SR_WD1 and SR_WD2 was primarily chosen to balance the number of cases in the two subsamples. In fact, most cases are characterized by an angle of up to 20° for SR_WD1 or close to 60° for SR_WD2 (Figure 2). Therefore, adjusting this threshold by 10–15° would not significantly impact the ratio of cases in the subsamples.

Note that we initially considered the reports from weather stations located at a distance ≤ 100 km from the windthrow area (Figure 1). This threshold value was determined empirically, based on the available data and scientific literature. For long-lived SLCWs (derechos), the length of the damage path can exceed 500 km, the maximum width can reach 50–70 km, and the length of the parent storm path can exceed 1000 km [43,44,45,46]. This highlights the importance of considering wind and precipitation reports from weather stations located a substantial distance from areas affected by windthrow for SR_WD1 reports. The maximum gap between the station and the windthrow area was 80 km for SR_WD1 and 37 km for SR_WD1. The mean values were 22 km and 14.6 km respectively for SR_WD1 and SR_WD2, and in only 7% of cases, the distance between the station and the windthrow area exceeded 50 km.

Table 1. Attributes determined for each storm report.

Name	Data Source
Storm event characteristics
Date and time of the storm report (UTC)	Routine observations at weather stations with 3 h and 12 h time steps, provided by RIHMI-WDC [39], and monthly reviews of hazardous weather events
Maximum observed wind gust, m s−1
Observed 12 h precipitation amount, mm
Satellite-derived daily precipitation amount (mm) in the grid cell corresponding to the location of the weather station (1998–2024 only)	GPM IMERG v.07 dataset [47] with 0.1° pixel size
Values of 16 convective instability indices (dynamical and thermodynamic), associated with SLCWs that were calculated 1 h before the event and extracted from the nearest point and found as the maximum within a 100 km radius of the weather station	ERA5-based dataset of convective variables for Northern Eurasia [43,48] with 0.25° pixel size
Forest cover characteristics
Percentage of forest-covered area and percentage of main forest types (dark coniferous, pine, deciduous and mixed) within a 25 km and 100 km radius of the weather station	Vegetation Map of Russia [38] with 230 m pixel size
Average growing stock (m3 ha−1) within a 25 km and 100 km radius of the weather station	GlobBiomass dataset [48] with 100 m pixel size
Daily mean Leaf Area Index (LAI) for high vegetation in 25 km radius around weather station	ERA5-Land reanalysis [42] with 0.1° pixel size
Forest damage characteristics
Distance between weather station and windthrow area (for SEWD only)	GIS database of windthrow events [8,9]
Location relative to windthrow area (SEWD1 or SEWD2)
Windthrow track length and area (can be the same for several SEWD reports related to a single windthrow event)

Table 1. Attributes determined for each storm report.

Name	Data Source
Storm event characteristics
Date and time of the storm report (UTC)	Routine observations at weather stations with 3 h and 12 h time steps, provided by RIHMI-WDC [39], and monthly reviews of hazardous weather events
Maximum observed wind gust, m s−1
Observed 12 h precipitation amount, mm
Satellite-derived daily precipitation amount (mm) in the grid cell corresponding to the location of the weather station (1998–2024 only)	GPM IMERG v.07 dataset [47] with 0.1° pixel size
Values of 16 convective instability indices (dynamical and thermodynamic), associated with SLCWs that were calculated 1 h before the event and extracted from the nearest point and found as the maximum within a 100 km radius of the weather station	ERA5-based dataset of convective variables for Northern Eurasia [43,48] with 0.25° pixel size
Forest cover characteristics
Percentage of forest-covered area and percentage of main forest types (dark coniferous, pine, deciduous and mixed) within a 25 km and 100 km radius of the weather station	Vegetation Map of Russia [38] with 230 m pixel size
Average growing stock (m3 ha−1) within a 25 km and 100 km radius of the weather station	GlobBiomass dataset [48] with 100 m pixel size
Daily mean Leaf Area Index (LAI) for high vegetation in 25 km radius around weather station	ERA5-Land reanalysis [42] with 0.1° pixel size
Forest damage characteristics
Distance between weather station and windthrow area (for SEWD only)	GIS database of windthrow events [8,9]
Location relative to windthrow area (SEWD1 or SEWD2)
Windthrow track length and area (can be the same for several SEWD reports related to a single windthrow event)

2.3.2. Rainfall Amount

We paid special attention to precipitation, since, alongside wind speed, it determines wind-related damage to forests [29,30,31,32]. Initially, we compiled the 12 h rainfall amounts from routine observations for each of the 351 considered SLCW events (for the corresponding days), recorded twice daily during the main observation periods closest to 08:00 and 20:00 local time (according to the Guide for Weather Observations at Stations). However, in dozens of cases, the rainfall amount associated with a storm event was divided between two consecutive 12 h observations. In such cases, we considered 24 h precipitation amounts.

Because of the sparsity of ground-based observation networks, weather reports may erroneously reflect the spatial distribution of convective wind gusts and precipitation [49,50,51]. Convective rainfall estimates at high spatial and temporal resolutions can be provided by weather radars [51] and satellite-based precipitation products. Although weather radar data has a higher spatial resolution of about 1 km, it does not cover most of Russia’s forest zone and has only been available for the last few years. On the contrary, satellite-based precipitation products provide near-global data coverage [52]. Among these, GPM IMERG (Integrated Multi-satellitE Retrievals for Global Precipitation Monitoring) provides precipitation estimates with relatively high spatial (0.1° × 0.1°) and temporal (30 min) resolution [53]. Furthermore, GPM IMERG outperforms other satellite-based products in reproducing spatio-temporal patterns and variability of extreme precipitation [54].

In this study, we used daily precipitation data from the GPM IMERG v.07 [47] alongside ground-based observations to estimate the amount of precipitation associated with SLCWs. We opted for daily data rather than half-hourly data, as previous studies [7,29,31] have demonstrated that heavy precipitation in the hours preceding a storm can increase the vulnerability of tree stands to wind impact by wetting the soil and crowns. In particular, we used the GPM_3IMERGDF (Final Precipitation L3, 1-day, 0.1-degree x 0.1-degree, V07) product for all the days in our sample that fell within the 1998–2024 period (201 days in total). Values were extracted from the grid cell nearest to the location of the weather stations and assigned to each storm report for further analysis (Figure 3).

For 333 out of 351 considered storm reports (occurring between 1998 and 2024), both observed precipitation and estimates according to the GPM IMERG L3 data were obtained. There is a moderate correlation between the two (Figure 4). The average and median observed precipitation (14.2 mm and 10.0 mm, respectively) is slightly higher than that estimated using GPM IMERG data (12.6 mm and 9.5 mm, respectively).

There are five cases in which the GPM IMERG underestimated the amount of precipitation several times (the observed 12 h precipitation exceeded 40 mm, while the GPM IMERG estimate was less than 10 mm). According to satellite images (MVIRI and SEVIRI/Meteosat-8), these cases were produced by local convective storms (meso-β scale). The underestimation of precipitation according to GPM IMERG data can be attributed to its low spatial resolution. Conversely, cases in which GPM IMERG substantially overestimated precipitation were associated with mesoscale convective systems (MCSs) of meso-α scale, particularly when two or three MCSs passed over the weather station in succession.

We calculated the average precipitation amount in forest-damaged areas according to the GPM-IMERG data and compared it with the observed values from nearby weather stations (see Figure 4b). As expected, the average precipitation in the damaged forest areas was higher than that observed at the nearest weather station, by an average of 3.3 mm. Windthrow events typically occur when storms intensify, producing heavy precipitation and severe downdrafts [55,56]. However, as weather stations are often located on the periphery of storm-affected areas, they may not always provide an accurate representation of storms. The correlation between the observed precipitation and the GPM IMERG data for these 151 cases is also moderate and statistically significant.

2.3.3. ERA5-Based Convective Variables

SR_WD and SR_ND events may form in differing convective environments. To estimate this difference, we analyzed so-called convective instability indices from a previously developed dataset for Northern Eurasia, covering the period 1979–2023 [43,57]. This dataset consists of 50 variables, calculated based on ERA5 reanalysis surface data and 20 standard vertical levels from 1000 to 300 hPa. The fields of convective variables have a spatial resolution of 0.25° and time steps of 1 h, as in the initial ERA5 data [58]. Indices for the dates and times of the formation of SR_WD and SR_ND were extracted from this dataset for further comparison.

In this study, we analyzed 16 variables that characterize the main ingredients of the development of a deep moist convection, which can produce SLCWs [59,60]. These include convective available potential energy (CAPE), convective inhibition (CIN) and lifted condensation level (LCL), which were calculated for the mixed layer (ML) from 0 to 1000 m above ground level (AGL); total precipitable water content (PW); wind shear for the 0–1 km (LLS), 0–3 km (MLS) and 0–6 km (DLS) layers; storm-relative helicity for the 0–1 km and 0–3 km layers (SRH); and five composite parameters: supercell composite parameter (SCP); significant hail parameter (SHIP); significant tornado parameter (STP); severe weather threat (SWEAT); energy–helicity index (EHI), which was calculated for the 0–1 km and 0–3 km layers. Additionally, ML WMAXSHEAR was calculated as the product of the DLS and the square root of doubled ML CAPE.

The values of the convective variables were calculated for the nearest ERA5 grid cell to the locations of the weather stations that reported severe winds (for both SR_WD and SR_ND). Additionally, the maximum values (excluding ML LCL and ML CIN) were extracted within a 100 km radius of the weather stations, in line with Chernokulsky et al. [43]. To estimate pre-convective environments, we used the values of the indices obtained one hour prior to the time of the storm report. In several cases, images from meteorological satellites (MVIRI and SEVIRI/Meteosat) were used to clarify the timing of the passage of a storm-producing MCS over a weather station.

2.4. Analysis of Differences Between Samples

We examined the differences in the above meteorological and forest stand characteristics (Table 2) between two groups of storm reports (SR_ND and SR_WD), and between three groups (SR_ND, SR_WD1 and SR_WD2). In particular, intergroup differences in wind gusts, precipitation amounts (based on ground-based observations and GPM IMERG data) and storm occurrence dates were estimated to establish whether these factors could account for the presence or absence of forest damage. We used the Kolmogorov–Smirnov (K–S) test to estimate the significance of differences, considering a p-value of 0.05 to be the significant level. In addition, we examined the relationships between various attributes (Table 2) using both Spearman’s rank correlation and Pearson’s linear correlation.

Table 2. The mean (numerator) and median (denominator) values of forest cover characteristics for the SLCW events (SR_ND, SR_WD1 and SR_WD2), calculated within a 25 km and 100 km radius around each weather station. Statistically significant differences are highlighted in bold.

Report Type (Number of Reports)	Forest Characteristics Around Each Weather Station
	Forest- Covered Area, % of Total Area	Proportion of the Main Forest Types (% of Total Area)				Growing Stock (m3 ha−1)	LAI for High Vegetation
		Dark- Coniferous	Pine	Deciduous	Mixed
Within 25 km radius
SRND (181)	64.8/ 65.2	3.8/ 1.2	12.6/ 10.0	22.0/ 22.4	23.5/ 21.2	132.5/ 128.2	3.40/ 3.37
SEWD (170)	61.5/ 65.0	3.7/ 0.6	10.3/ 7.4	20.8/ 21.4	24.6/ 22.0	133.8/ 132.6	3.50/ 3.58
Within 100 km radius
SRND (181)	66.2/ 65.6	5.5/ 2.1	11.3/ 10.0	23.0/ 24.4	24.2/ 22.0	139.1/ 139.9
SEWD (170)	64.5/ 63.8	4.6/ 0.7	8.9/ 7.2	24.4/ 26.3	25.5/ 24.4	142.3/ 142.4

Table 2. The mean (numerator) and median (denominator) values of forest cover characteristics for the SLCW events (SR_ND, SR_WD1 and SR_WD2), calculated within a 25 km and 100 km radius around each weather station. Statistically significant differences are highlighted in bold.

Report Type (Number of Reports)	Forest Characteristics Around Each Weather Station
	Forest- Covered Area, % of Total Area	Proportion of the Main Forest Types (% of Total Area)				Growing Stock (m3 ha−1)	LAI for High Vegetation
		Dark- Coniferous	Pine	Deciduous	Mixed
Within 25 km radius
SRND (181)	64.8/ 65.2	3.8/ 1.2	12.6/ 10.0	22.0/ 22.4	23.5/ 21.2	132.5/ 128.2	3.40/ 3.37
SEWD (170)	61.5/ 65.0	3.7/ 0.6	10.3/ 7.4	20.8/ 21.4	24.6/ 22.0	133.8/ 132.6	3.50/ 3.58
Within 100 km radius
SRND (181)	66.2/ 65.6	5.5/ 2.1	11.3/ 10.0	23.0/ 24.4	24.2/ 22.0	139.1/ 139.9
SEWD (170)	64.5/ 63.8	4.6/ 0.7	8.9/ 7.2	24.4/ 26.3	25.5/ 24.4	142.3/ 142.4

Table 3. The mean values of the storm report characteristics (mean dates, wind gusts and precipitation amounts) for the SR_ND, SR_WD1 and SR_WD2 reports. Statistically significant differences are highlighted in bold.

Report Type (Number of Reports)	Storm Event Date	Wind Gust, m s−i	Observed 12 h Rainfall Amount	Daily Rainfall Amount (GPM-IMERG)
SRND/SRWD	25.06/01.07	27.0/23.6	13.7/14.4	8.9/16.8
SRWD1/SRWD2	30.06/03.07	24.1/22.7	16.2/11.0	18.1/14.5

Table 3. The mean values of the storm report characteristics (mean dates, wind gusts and precipitation amounts) for the SR_ND, SR_WD1 and SR_WD2 reports. Statistically significant differences are highlighted in bold.

Report Type (Number of Reports)	Storm Event Date	Wind Gust, m s−i	Observed 12 h Rainfall Amount	Daily Rainfall Amount (GPM-IMERG)
SRND/SRWD	25.06/01.07	27.0/23.6	13.7/14.4	8.9/16.8
SRWD1/SRWD2	30.06/03.07	24.1/22.7	16.2/11.0	18.1/14.5

3. Results

3.1. Wind and Precipitation Characteristics of the SLCW

The compiled dataset includes a total of 351 SLCW events from 227 different weather stations in the Russian forest zone between 3 June 1986 and 5 July 2024. Of these, there are 181 SR_ND and 170 SR_WD reports, the latter of which include 112 SR_WD1 reports and 58 SR_WD2 reports.

Of the 351 analyzed SLCW events that occurred in the Russian forest zone between 3 June 1986 and 5 July 2024, wind gusts ranged from 15 m s⁻¹ to 36 m s⁻¹. For 249 reports, wind gusts were ≥25 m s⁻¹. It should be noted that over 30% of weather stations in the forest zone did not record any wind gusts of ≥25 m s⁻¹ during the study period. This indicates a relatively low frequency of these events in Russia’s forest zone, which can be explained by large distances between weather stations, and their locations in wind-sheltered areas. Most of the considered storm events occurred in the ER (Figure 5), mainly due to the higher density of weather stations compared to Siberia. This is also partly because the windthrow data series for the forest zone of the ER is longer than that for Siberia [8,9].

Of the 351 SLCW events, 181 were SR_ND events and 170 were SR_WD events. The latter included 112 SR_WD1 events and 58 SR_WD2 events. SR_ND events predominantly occurred on the eastern slope of the Ural Mountains, while SR_WD events were prevalent in the central part of the forest zone of the ER. These regions differ significantly in terms of storm characteristics, forest stands and climate. For example, the proportion of forest area on the eastern slope of the Ural Mountains is lower than in the ER forest zone, as is the proportion of weather stations located in wind-sheltered areas. While deciduous and pine forests dominate the eastern slope of the Urals, mixed and dark coniferous forests (which are more vulnerable to wind impact) are widely distributed in the ER forest zone [38]. Additionally, a significant proportion of SLCW events on the eastern slope of the Urals were accompanied by low precipitation, particularly in spring, whereas in the ER forest zone, most SLCW events occurred in summer and were accompanied by moderate or heavy rainfall. Finally, the annual precipitation amount in the central part of the ER forest zone was 40% greater than on the eastern slope of the Urals, determining differences in soil moisture.

3.2. The Main Characteristics of SR_ND and SR_WD and Their Difference

3.2.1. Forest Stands Characteristics for SR_ND and SR_WD

Table 2 and Figure 6 show the mean and median values of forest stand characteristics calculated within the 25 km and 100 km radii around each weather station reporting SLCW events. Based on the K-S test, no statistically significant differences were found between SR_ND and SR_WD for most characteristics that determine forest vulnerability to wind impact (forest-covered area, growing stock, percentage of mixed forests). Moreover, the proportion of dark coniferous and pine forests within a 100 km radius around weather stations for SR_ND is higher than for SR_WD. Therefore, we can hypothesize that differences in storm characteristics, rather than differences in forest stand vulnerability, determine whether or not wind-related forest damage occurs.

3.2.2. Wind Gusts

SR_WD gusts were significantly stronger than SR_ND gusts (on average, 28.5 m s⁻¹ vs. 27.0 m s⁻¹) when considering reports of ≥25 m s⁻¹ (181 SR_ND and 68 SR_WD; Figure 7). The full sample of SLCW reports contains 196 cases of hurricane-force winds (≥33 m s⁻¹), 12 of which caused forest damage. There were no significant differences in wind gusts between SR_WD1 and SR_WD2 (Table 3).

There is no significant correlation between windthrow characteristics (length and damaged area) and wind gusts reported by the nearest weather station. The weather stations reported hurricane-force winds, which caused both local (<100 ha) and large-scale (>10,000 ha) forest damage. The different scales of windthrow can be explained by the nature of the events. Although local downbursts and long-lived derecho-like storms may have similar peak intensities, their spatial footprints differ by tens of times. The number of stations reporting severe winds is a more informative predictor of the area affected by windthrow than peak wind gust. However, limitations in the observation network, such as low density and weather stations located in wind-sheltered environments, significantly undermine this relationship. In this study, we did not examine the relationship between the number of SLCW reports and windthrow characteristics since we considered each SLCW report separately, without grouping them by storm track.

3.2.3. Precipitation Amount

The difference between the SR_ND and SR_WD reports in terms of observed precipitation amounts is statistically significant, although the average value differs by only 0.7 mm (Table 2). But there is a much stronger difference between SR_WD1 and SR_WD2, as well as between SR_ND and SR_WD1. The median observed rainfall for SR_WD1 is 5 mm higher than for SR_ND. SR_WD1 reports characterize more intense rainfall associated with the central part of the storm’s path. In contrast, SR_WD2 events were located on the storm’s periphery or outside the precipitation zone.

The standard deviation of observed precipitation differs substantially between SR_ND and SR_WD (14.5 mm vs. 11.5 mm), reflecting differences in their distributions. The proportion of cases with light precipitation (≤3 mm) was 25% for SR_ND and 15% for SR_WD. Conversely, the proportion of cases with heavy precipitation (≥30 mm) was 13.8% for SR_ND and 10% for SR_WD (Figure 7). According to the weather station data, there was a total of 42 SLCW events with heavy rainfall (≥30 mm 12 h⁻¹), of which only 17 were SR_WD.

According to the GPM IMERG data (Figure 7, right panel), SR_WD events were accompanied by significantly heavier rainfall than SR_ND events. Unlike the observational data, the GPM IMERG data show that SLCW events with heavy rainfall were also mainly SR_WD. Of the 28 SLCW events with heavy rainfall (≥30 mm 12 h⁻¹) reported by the GPM IMERG data, 75% were SR_WD.

The differences in precipitation amounts between SR_ND and SR_WD, according to observations and especially GPM IMERG data, generally indicate that heavy rainfall substantially increases the probability of windthrow. This is particularly relevant for the SR_WD1 subsample, which more accurately characterizes the precipitation that accompanied windthrow events. These findings are consistent with previous studies indicating that heavy rainfall, when occurring before or during a windstorm, increases forest stand vulnerability to wind impact [29,30,31].

More interestingly, most of the windstorms accompanied by particularly heavy rainfall (8 out of 10 cases with an observed rainfall amount ≥ 50 mm/12 h, and 14 out of 18 cases with an observed rainfall amount ≥ 40 mm/12 h) did not cause damage to forests. Furthermore, none of the eight compound wind and rainfall extremes in our sample (events with wind gusts ≥ 25 m s⁻¹ and precipitation ≥ 50 mm/12 h) caused damage to forests. An explanation of this feature requires additional information about the characteristics of these storm events and their formation environments. In particular, the lack of windthrow can be associated with the localized nature of these storms and the limited area in which they impact forests. A similar feature was previously found in the Perm Region (eastern European Russia), where numerous local downbursts generally did not cause large-scale forest damage [37].

3.2.4. Convective Variables According to the ERA5 Data

Convective environments are significantly different between SR_ND and SR_WD, both at the weather station locations and within a 100 km radius of each location. In general, SR_WD events formed in environments that are more favorable for the development of SLCWs than SR_ND events in terms of convective instability, wind shear, storm-relative helicity, and composite variables (Figure 8). The average differences in ML CAPE and PW between SR_WD and SR_ND are approximately 257 J kg⁻¹ and 4.9 mm, respectively (Figure 8). Both SE_WD and SR_ND predominantly formed in environments with high ML CAPE (>500 J kg⁻¹). Low-CAPE events were quite rare, accounting for only 17.9% of events, 73% of which were SR_ND. More than half of the low-CAPE events occurred in April and May, whereas about 80% of the events in the entire sample occurred in the summer. The ML CAPE and PW values for the SR_WD reports are slightly higher than those for a sample of SLCW events that damaged forests from 2006 to 2021 [43]. However, regional studies based on small samples may show insignificant differences in ML CAPE and PW between SR_ND and SR_WD (e.g., [37]).

Most SLCW events (both SR_ND and SR_WD) were associated with DLS > 20 m s⁻¹ (59%), but stronger wind shear was generally more characteristic of SR_WD than for SR_ND, which aligns with previous results [37]. The proportion of weak-shear events (DLS < 15 m s⁻¹) 24.3% for SR_ND and just 6.0% for SR_WD. All of these events were associated with high CAPE, with an average ML CAPE of 1048 J kg⁻¹. The mean observed 12 h precipitation amount associated with such events was 24.9 mm, which is 1.75 times higher than the average for the entire sample. Four of the eight compound wind-precipitation extremes occurred under conditions of high CAPE and weak DLS. Since weak shear contributes to the formation of only slow-moving and often short-lived storms [56,60], the associated SLCWs were very localized and most likely can be classified as microbursts.

In the two-dimensional CAPE-DLS space, SR_ND produced two clusters (Figure 9a). One cluster occurred in the spring and had high shear and low CAPE. The other cluster occurred only in the summer and had weak shear and high CAPE. SR_WD formed a cluster with moderate to high ML CAPE (≥500 J kg⁻¹) and moderate to high DLS (≥15 m/s). It is noteworthy that the events with the highest ML CAPE (≥1500 J kg⁻¹) were mainly SR_WD.

In the precipitation-DLS space (Figure 9b), SR_ND also formed two clusters: events with weak precipitation (<10 mm) and very strong DLS (≥30 m s⁻¹), and events with weak or moderate DLS (<20 m s⁻¹) and very heavy rainfall (≥40 mm). For the first group, the lack of forest damage can be explained by weak precipitation and/or the seasonal state of vegetation, specifically the absence of leaves on trees in the spring. For the second group, the lack of forest damage is associated with their local impact on the forest canopy. Thus, the formation environments of SLCWs substantially affect forest damage.

Composite convective variables are generally higher for SR_WD than for SR_ND. On average, the EHI and SCP show the strongest differences, which are more than 1.6 times larger for SR_WD than for SR_ND. The average values of the composite variables for SR_WD are similar to those for the SLCW sample in the European Russia and the Urals [43], as well as for strong (≥32 m s⁻¹) warm-season SLCW events in Europe [60]. However, SR_ND also formed in areas with high composite variable values. In particular, ML WMAXSHEAR was >1000 m² s⁻² in 14 cases. In three of these cases, forest damage was caused by tornadoes, which were not considered in this study. In another six cases, windthrow events occurred more than 100 km away from weather stations that reported severe wind gusts.

3.2.5. Temporal Distribution of Storm Events

The vast majority of SLCWs have been reported in the afternoon or evening, according to local time. This is consistent with previous studies of convective windstorms in northern Eurasian forests [9,37,43]. No statistically significant differences in timing were found between SRND and SRWD events.

The seasonal distribution differs significantly between SR_ND and SR_WD (Figure 10), reflecting their distinct formation environments. The largest number of SR_ND events occurred in late May, with a secondary peak between mid-June and mid-July, which corresponds to the annual maximum in the development of a deep convection [11]. In contrast, the seasonal distribution of SR_WD only peaks in the summer, consistent with the general distribution of windthrow events in Russia [8,9].

The conditions in May differ from those in midsummer in terms of both forest vulnerability to wind impact and atmospheric conditions favorable to storm formation. In particular, vulnerability increases in summer compared to spring due to the growth of foliage on deciduous trees [2,23]. But on the other hand, soil moisture can decrease in summer compared to spring, which can lead to a decrease in vulnerability [30].

In this study, we evaluated seasonal changes in woody vegetation using LAI, calculated according to the ERA5-Land reanalysis, for each storm event date. There are no significant differences in LAI between SR_ND and SR_WD (Table 2), though the seasonal difference is significant (mean LAI values are 2.82 and 3.59 for spring and summer events, respectively). Therefore, differences in LAI may partially explain the seasonal distribution patterns of SR_WD and SR_ND.

Seasonal differences in the formation environments of SLCWs in Russian boreal forests are similar with to those in Europe [60]. Summertime events form predominantly in warm, moist air masses and are accompanied by stronger rainfall. Springtime events form in colder air masses with lower moisture content and are accompanied by weaker rainfall. In our sample, springtime events are characterized by slightly stronger wind gusts (an average of 26.5 m s⁻¹) and light precipitation (an average of 8.0 mm). Summertime events have slightly weaker gusts (an average of 25.1 m s⁻¹) but almost twice the precipitation (an average of 15.6 mm). Thus, seasonal differences in air mass characteristics determine the difference in rainfall amount associated with SLCWs, thereby explaining the increasing proportion of SR_WD in summer.

4. Discussion

In this study, we examined for the first time the characteristics of severe linear convective windstorm (SLCW) events to determine whether they lead to damage to temperate and boreal forests in Russia. We selected 351 SLCW events that occurred in the forest zone of Russia using routine meteorological observations and the previously compiled database of windthrow events [8,9]. We divided the sample into two subsamples: one with windthrow (SR_WD, 170 events) and one without (SR_ND, 181 events). As shown in previous studies [2,4,14,15,16,17,18,19], the occurrence of windthrow is determined by the characteristics of the windstorm itself and the characteristics of the forest stand and sites (in particular, terrain and soil).

As our goal was to determine the differences in the characteristics of windstorms between SR_WD and SR_ND subsamples, events were selected to minimize differences in other predictors of windthrow occurrence, primarily the characteristics of forest stands. When calculated within a 25 km radius around weather stations, there are no significant differences between the SR_WD and SR_ND subsamples in terms of the proportion of forest-covered area, the contribution of different tree species, growing stock volume and LAI.

The study’s approach and the data used have several limitations that should be taken into account. Firstly, the available land cover and land use maps do not provide data on forest age, one of the main predictors of windthrow occurrence [13,14,15,16,17,18,19]. Secondly, the vegetation map of Russia used in this study shows the dominant tree species in 2018, whereas the windthrow event data were obtained from 1986 to 2014. Thirdly, soil characteristics (thickness, moisture and texture), which are also important predictors of windthrow occurrence, were not considered due to a lack of data with the required spatial resolution for Russia. However, the impact of soil characteristics on windthrow occurrence is weaker than that of forest stand characteristics [18,19]. Growing stock data [48] was used instead of data on forest stand age.

The next limitation relates to the data used from weather stations. Firstly, the distances between weather stations, particularly in Siberia, exceed the typical spatial footprint of an SLCW event (except for long-lived ones). In the forest zone of European Russia, for example, the average distance between stations is about 47 km, whereas in Siberia, it is around 59 km. Observation density also affects the spatial distribution of identified cases (there are more cases in European Russia than in Siberia). Another issue that could affect the results is that some weather stations in the forest zone are located in sheltered areas and are therefore not well suited to observing SLCWs. This leads to a significant underestimation of wind gusts for SR_WD compared to SR_ND. In turn, SR_ND events were reported four to five times at the same stations, which are located in areas where strong winds are more likely to be reported.

The actual amount of precipitation in windthrow areas cannot be accurately determined from weather station data, as the spatial distribution of convective precipitation is highly uneven [61]. High-resolution weather radar data is better suited to assessing differences in precipitation intensity within and outside windthrow areas, but it is still very limited in Russia [62]. As weather radar data accumulates and becomes more accessible, this could be a focus for future research.

The GPM IMERG rainfall data, which were used alongside ground-based observations, have several limitations that have been highlighted in previous studies [63,64,65]. In particular, the resolution may not be accurate enough to adequately consider spatial variations in the intensity of convective precipitation. A time step of 30 min is insufficient to observe short-term peaks in precipitation intensity, which reduces the accuracy of total precipitation calculations. This results in only a moderate correlation between ground-based precipitation data and GPM IMERG satellite data, as was also highlighted for other regions [63,64,65]. However, such limitations are equally typical of both SR_ND and SR_WD events, meaning that the significant differences identified between the two types of event can be considered meaningful.

The approach used to select and classify SR_WD events into SR_WD1 and SR_WD2 also has limitations, primarily due to the large distance between the areas of windthrow and the weather stations. This distance may exceed the characteristic scale of the spatial variability of convective gusts and precipitation [66,67]. However, using a lower maximum distance threshold for weather stations and windthrow events (rather than using the value of 100 km) could have substantially reduced the sample size and led to SR_WD reports being misclassified as SR_ND. We consider the used thresholds to be the most optimal for the analyzed data. An analysis of the sensitivity of such threshold values could be performed in the future, once more detailed data (e.g., radar data) is available. The lack of open access to weather radar data generally limits many studies of severe convective events in Russia.

As a result, we do not consider the aforementioned limitations to be critical when drawing conclusions about the differences between the storms in the SR_WD and SR_ND subsamples. However, addressing these limitations leaves room for future research.

5. Conclusions

The main findings resulting from this study are listed below:

• SR_WD events are characterized by stronger wind gusts than SR_ND events (28.5 m s⁻¹ and 27.0 m s⁻¹, respectively), when comparing SR_WD and SR_ND events for SLCW reports obtained using the observed wind gust threshold (≥25 m s⁻¹).
• There is no correlation between observed wind gusts and forest damage characteristics, e.g., track length and area. This differs from previous studies [5,7], which focused on windthrow events caused by non-convective winds.
• The SR_WD and SR_ND samples differ significantly for precipitation amounts obtained from both station observations and satellite data. Heavy rainfall accompanying SLCWs generally contributes substantially to forest damage. According to the GPM IMERG satellite-derived precipitation estimate, the daily rainfall associated with SR_WD is 1.5 times greater than that associated with SR_ND. Based on station data, this difference is smaller, at around 5%, but still significant. This is consistent with the findings of previous studies on both temperate [7,31] and tropical [32,33] forests. However, this is the first time that the relationship between the amount of rainfall during a storm and the occurrence of windthrow has been considered for convective windstorms in boreal and temperate forests.
• The amount of rainfall observed depends significantly on the location of the weather station in relation to the storm’s path. The amount of rainfall associated with cases where weather stations are located along the path of the storm is significantly higher (by 47%) than for cases where weather stations are located to the side of the storm’s path. The sensitivity of windthrow occurrence to precipitation explains the significant difference in the seasonality of SR_ND and SR_WD events. Specifically, SR_ND occurred at two peak frequencies: one in the second half of May and one in summer, while SR_WD events mainly occurred in summer. The most significant differences between SR_ND and SR_WD were found in the convective environments, estimated based on the ERA5 reanalysis. On average, SR_WD cases formed in conditions of stronger wind shear, as well as in more unstable air masses with a higher moisture content. Therefore, SR_WD cases tend to occur in environments that are more favorable for the development of severe convective storms.
• In line with this general pattern and in confirmation of previous estimates [37], there are several notable features in the distribution of convective variables for SR_ND alone. There are two types of SR_ND events: those that occur under low CAPE and high shear, and those that occur under weak shear and high CAPE. The low-CAPE SR_ND events often formed in spring, in air masses with a relatively low moisture content. These events were accompanied by low precipitation (<10 mm), which may explain the lack of forest damage. The high-CAPE and low shear SR_ND events only occurred in summer and were associated with heavy rainfall. The lack of forest damage can be explained by the localized nature of these events and the small area of forest impacted.
• In turn, 64% of SR_WD formed under conditions of high DLS (>20 m s⁻¹) and high CAPE (>500 J kg⁻¹), which are especially favorable for the development of severe and long-lived SLCW events, including derecho-like storms that cause large-scale damage [43,44,45,46].

Our analysis reveals that the occurrence or non-occurrence of windthrow events caused by SLCWs, alongside forest stand vulnerability, is determined by the combined effects of wind and rainfall, storm longevity and spatial footprint. These characteristics are closely related to the environments in which storms form. In the Russian forest zone, where SLCWs account for 66% of the total windthrow area, the highest probability of windthrow events occurs in environments with high CAPE and strong wind shear. This can inform assessments of probable changes in forest damage caused by SLCW events based on projections of rainfall intensity and convective variables in a changing climate [68,69].

The features that were identified as determining the impact of linear convective windstorms on boreal and temperate forests in Northern Eurasia—including wind gusts, rainfall amounts, and convective environments—can be used alongside previously obtained climatologies of tornadoes [11] and windthrow events [8] to assess the climate risks to forests under climate change. The results may particularly help to refine previous assessments of climate risks in Russia [70], as well as being used to refine regional plans and strategies for adapting to climate change, particularly in forestry.