Impact of wind speed and soil frost on electricity distribution system reliability

Research article Impact of wind speed and soil frost on electricity distribution system reliability Juha Haakana a,* , Otto R¨ ais¨ anen a , Markku Karhunen b , Ilona L´ ang-Ritter c , Jukka Lassila a a LUT University, Finland b Finnish Environmental Institute, Finland c Finnish Meteorological Institute, Finland ARTICLE INFO Keywords: Distribution system Fault Interruption Power system Reliability Soil frost Wind gust Wind speed ABSTRACT Wind is a significant cause of power outages in overhead line networks. Present electricity distribution network management systems provide an extensive amount of data on network faults. These data combined with the openly available weather and soil information allow to analyze the impact of wind and soil frost on electricity system reliability more accurately than in the past. The objective is to determine the effect of soil frost on wind-induced faults in the electricity distribution system and identify variables that provide the best correlation between wind speed and the number of power supply interruptions. Five geographical areas in Finland are analyzed based on extensive electricity distribution interruption statistics collected in the country between 2008 and 2018. We introduce a methodology to analyze the effect of wind speed and soil frost and present numerical data analyzed for the areas under consideration. The methodology employs a multivariate linear regression model to show the dependence between wind, soil frost, and power supply interruptions. The results indicate that there is a strong positive correlation between the number of power supply interruptions and wind speed, and furthermore, in the multivariate model, adding the variable of soil frost improves correlation in all the areas under consideration. On average, the correlation with wind gusts and the number of power supply interruptions is approximately 0.67. However, with the soil frost variable, the correlation increases to 0.72. The results show that soil frost should be considered in studies on interruption modeling in Nordic conditions. 1. Introduction Strong winds and windstorms can cause significant damage to society. For instance, power outages are one of the most common and important consequences of windstorms [1]. Finnish statistics show that one-third of all customer-experienced supply interruptions are caused by wind and storms. Other significant sources of interruptions are unknown causes, snow and ice, and component failures, accounting for 18 %, 14 %, and 11 % of interruptions, respectively. If we consider the time of interruptions experienced by end users instead of the number of interruptions, the proportion of wind-induced supply interruptions is even higher, up to 50 % [2]. Considering that today’s society is highly dependent on a continuous supply of power, hardening of power lines and preparedness * Corresponding author. E-mail address: [email protected] (J. Haakana). Contents lists available at ScienceDirect Heliyon journal homepage: www.cell.com/heliyon https://doi.org/10.1016/j.heliyon.2024.e40846 Received 3 February 2024; Received in revised form 12 November 2024; Accepted 28 November 2024 Heliyon 10 (2024) e40846 Available online 4 December 2024 2405-8440/© 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). for windstorms are crucial in mitigating the socioeconomic impacts of power outages ([3]; [4]). Additionally, climate change has a significant impact on infrastructures of society [5]. Electricity distribution infrastructure is one of the sectors that is already suffering from climate change. Global warming itself is not directly critical to the electricity infrastructure, but its impacts can be seen, for instance, in increased costs caused by extreme weather events, such as storms and higher snow loads [6,7]. It is well known that a significant number of interruptions in the electricity distribution system are due to strong winds, windstorms [8,9], and snow loads affecting the overhead line network [2]. For instance, it is estimated that in the United States, approximately 80%–90 % of the power outages are due to extreme weather events [10]. In general, fault events causing interruptions of supply occur typically in network sections where falling trees may damage the power system. Laurila et al. [11] studied windstorms in Northern Europe (including Finland) and found that the wind gusts of cold-season windstorms are of the largest spatial scale, and the strongest wind gusts are recorded during the cold season (October–March). Laapas et al. [12] also found that the annual maximum wind speeds in Finland are typically observed in winter. Most of the wind-induced power outages are due to the damage caused by uprooting or breaking trees falling on electricity distribution or transmission lines [13]. Finland is among the most forested countries in Europe, forests covering approximately 70 % of Table 1 Key characteristics of the datasets used in studies on electricity network interruption analysis. A hyphen indicates information not disclosed or missing in the paper. The last row indicates the characteristics of this paper. OMS =outage management system, VMS =vegetation management system. Publication year, reference Analysis period Data types non-DSO Soil data types Data types,DSO Outage event types Area size [38] 1996.1998, 1999 Weather, soil,forest Soil drainage Infrastructure, OMS Hurricane 2 DSOs [39] 2000–2004 Population –Infrastrucutre, VMS –1 DSO, 764k customers, 28.6k km overhead line [40] 1998–2004 Landcover, soil, weather Soil drainage, soil depth Infrastructure, OMS Hurricane and ice storm 3 DSOs, 5.7M customers [41] 1997–2005 Landcover, soil, weather Soil moisture Infrastructure, OMS Hurricane 1 DSO, 125 000 km 2 [42] 1997–2005 Landcover, soil, weather Soil moisture Infrastructure, OMS Hurricane 59 700 km 2 [43] 1997–2005 Landcover, soil, weather Soil moisture Customer count, infrastructure, OMS Hurricane 59 700 km 2 [44] 1997, 1998, 2004–2005 Landcover, soil, weather Soil moisture Customer count, infrastructure, OMS Hurricane 78 500 km 2 [45] 2005–2014 Landcover, soil, weather Soil moisture OMS, infrastructure Blizzard, hurricane, nor’easter, thunderstorm 1 DSO, 1.2M customers, 29K km overhead line [46] 2011 Forest, landcover, weather –Infrastructure, OMS, VMS Hurricane 4600 km 2 [33] 2005–2014 Landcover, soil, weather Soil moisture Infrastructure, OMS Blizzard, hurricane, nor’easter, thunderstorm 1 DSO [34] 1995–2005 Forest, landcover, population, topography, weather Soil moisture OMS Hurricane 1 DSO [35] 2007–2019 Landcover, vegetation, weather Soil moisture Infrastructure, OMS Blizzard, hurricane, nor’easter, thunderstorm 1 DSO [35] 2005–2017 Landcover, soil, weather Soil moisture Infastructure, OMS, VMS Extratopical, thunderstorm  [47] 2005–2018 Landcover, topography, vegetation, weather Soil depth, soil moisture, soil type Infrastrucutre, OMS Nor’easter, snow and ice events 1 DSO [8] 2010–2018 Forest, weather –Customer count, infrastructure, OMS Extratopical 2 DSOs [48] 2010–2016 Forest, weather –Infrastructure, OMS Blizzard, extratropical, snow and ice events 1 DSO, 16 385 km 2 , 1.2M customers [37]Landcover, soil, weather, forest, vegetation Soil moisture Infrastructure, OMS Extratropical and tropical storms 40,000 km 2 , 3.5M customers [49] 2017–2019 Forest, topography, weather . Risk locations Snow events 1 DSO, 9.2K overhead line, 80k customers [9] 2005–2014 Weather –Customer count, infrastructure, OMS Windstorm, thunderstorm 3.3M customers, 375k km power line This study 2008–2018 Weather, soil Soil temperature OMS Extratropical and thunderstorms 3.3M customers, 375k km power line J. Haakana et al. Heliyon 10 (2024) e40846 2 the area of Finland [14]. Furthermore, a large proportion of power lines especially in Finnish rural areas are located above ground and in forests. Information on the topic is available from the Energy Authority of Finland (EA) and the National Land Survey of Finland (NLS). For instance, in 2020, the proportion of overhead power distribution lines below 70 kV was 52 % [15], and in 2019, the proportion of overhead lines in the proximity of forests was 78 % (NLS, 2019). The figure contains all the lines surrounded by forest and the lines located close to single trees. Seasonally freezing ground is a common phenomenon in areas with sufficiently long periods of below-freezing temperatures. The main factors affecting the duration and depth of soil frost are air temperature, soil type, vegetation, precipitation, and snow cover [16]. In the cold season, soil frost is typically present and anchors the trees to the ground thus making them more resistant against strong wind gusts [14,17,18]. 1.1. Effect of climate change Depending on location, climate change can have a varying effect on distribution network reliability. For instance, in northern locations, such as Finland, the climate is warming even faster than in the rest of the world. A recent study by Rantanen et al. [19] showed that in the Arctic region the warming has been at least four times as fast in the past 43 years (1979–2021) as the global average. In Finland in particular, the warming has been 2–3 times as fast as globally on average [19]. In a warmer climate, the depth of soil frost will decrease [20], and the period of soil frost is likely to become shorter in Finland [21]. Lehtonen et al. [21] state that the winter soil frost period will become roughly one month shorter by the 2050s, and the Finnish winters will have only short periods of soil frost. This could lead to an increase in wind-induced faults in the winter period. Regarding windstorms, there is no clear consensus about how climate change affects them, nor about how the windiness will change in Finland. Ruosteenoja et al. (2022) produced model-derived climate projections for Finland by using results of the CMIP6 global climate model (Phase 6 of the Coupled Model Intercomparison Project, [22]), which is the newest generation of global climate models (GCMs). According to Ruosteenoja et al. (2022), the projected changes in the mean wind speed are minor, and by the 2080s changes in monthly mean wind speeds will vary between −12 and +10 %. Kjellstr¨ om [23], for instance, also states that there is no clear trend in the future in the occurrence of strong winds in the Nordic countries. Even though the changes in the occurrence of strong winds are small, compound events, such as strong wind and heavy precipitation simultaneously can increase the risk of extensive damage to critical infrastructure [24,25]. Gregow et al. [26] state that in Finnish forests the wind-induced risks will increase because of the lack of soil frost in the future. Additionally, several studies agree that although extratropical cyclones do not seem to amplify on average, the most extreme windstorms may get stronger in the future (Sinclair et al., 2020; Priestley and Catto, 2022). Especially in the southern and middle regions of Finland, the risk of wind damage will increase due to the shorter and milder soil frost period [14,21]. 1.2. Effect of wind and soil frost on power system interruptions Numerous studies have been made about wind risks to the forests in Finland and other Nordic countries [26–31], and some studies have focused on the impacts of soil frost and winds on forestry [21,26,29,32]. The effect of wind on power supply interruptions has been investigated in a multitude of studies considering a wide variety of variables. However, there is a paucity of research related to the risks and impacts of soil frost or the combined effect of soil frost and wind-induced power supply interruptions on the electrical grid. The effect of soil has previously been considered on studies [33–35,36] that have identified the anchoring effect of the soil during extreme events as a contributing factor for wind-induced interruptions. However, these studies have focused on the phenomenon of increased soil moisture reducing the anchoring effect and have not considered soil frost. Notably in the study [37] a large set of 218 variables were considered however soil temperature or soil frost was not included in the study. Table 1 presents an overview of recent literature covering the research on the effect of wind and soil frost on electrical power system interruptions. The table shows that wind is the key element in several studies, but soil frost has not been addressed. Thus, this study, by combining the effects of both wind and soil frost, bridges a research gap. 1.3. Objective and contributions This study has two objectives. The first objective is to determine the effect of soil frost on wind-induced interruptions to assess the reliability of the electricity distribution system. The second objective is to define the best-correlating variables for soil frost and wind speed. The analysis is carried out with the programming language R and MS Excel Data Analysis Toolpack. The effects are analyzed for five case areas in Finland. The main contributions of this paper are as follows: 1. The effect of soil frost and wind speed on the distribution network interruption frequency is determined. 2. The correlation of multiple wind and soil frost variables is investigated, and the optimal power relation of wind speed and interruption frequency is defined. A further outcome of the present study is that the analyses are based on an extensive amount of statistical data on power supply interruptions from a long period. J. Haakana et al. Heliyon 10 (2024) e40846 3 1.4. Structure of the paper Section 2provides the background for electricity supply interruptions. Section 3describes the methodology used in the analyses. Section 4shows the results of the regression analyses. Section 5discusses the results of the paper, and finally, Section 6summarizes the findings of the paper. 2. Background The analysis is based on extensive interruption statistics from the years 2008–2018 gathered from Finnish electricity distribution system operators (DSOs). The data are divided into five areas shown in Fig. 1. The areas are numbered from 1 to 5, and the letters in brackets refer to the geographical location of the area (1(SW): southwestern Finland, 2(SE): southeastern Finland, 3(E): eastern Finland, 4(W): western Finland, and 5(N): northern Finland). Hereafter, we refer to these areas by their number and letter code. The regional division is based on the requirement of anonymization of the local DSOs and their customers. The condition for data sharing for research purposes requires a minimum of six DSOs operating in each area. The data contain all types of fault interruptions, and the causes of interruptions are divided into several groups, such as interruptions caused by wind, snow loads, or lightning. Nonmeteorological causes like animals are also listed. This study focuses solely on interruptions caused by wind. The data cover over 95 % of the interruption events in the Finnish electricity distribution networks in a typical year. The exception is the year 2015, when a change was made in the method of data collection, and thus, the data coverage was low. In this context, the term interruption refers to a situation where one or more customers experience an interruption of power supply that lasts at least 3 min. 2.1. Data The methodology developed in this paper is based on four datasets: an interruption dataset gathered from Finnish DSOs, the DSOs’ technical statistics dataset obtained from the Finnish Energy Authority (EA), wind speed data (average and wind gust speed measurements) from the Finnish Meteorological Institute (FMI), and soil temperature data (ERA5-Land reanalysis) from the European Centre for Medium-Range Weather Forecasts (ECMWF). All data, apart from the interruption dataset, are open data. The interruption dataset is not publicly available, and it is provided with the location data anonymized to five different areas containing a minimum of six DSOs to ensure anonymity. All the data were thus aggregated to the same level as the interruption dataset. The data cover the years 2008–2018, for a total of 4018 days. 2.1.1. Interruption and network data The interruption dataset includes information of the location (area), type, cause, time, number of customers affected, and length of the interruption for every interruption reported by the Finnish DSOs. The dataset was filtered to include only the wind-induced Fig. 1. Areas of Finland based on the interruption dataset. Red lines represent the medium-voltage overhead line network of Finland. J. Haakana et al. Heliyon 10 (2024) e40846 4 interruptions in the overhead line network, because they are in the focus of this study. The data were aggregated to a temporal resolution of one day. The technical statistics dataset from the Finnish Energy Authority was used to obtain the total medium-voltage overhead line lengths of the different areas for each year under study. The interruption rate (number of interruptions per day per kilometer) of the medium-voltage overhead line (OHL) network was then calculated based on these datasets and used as the target variable in the study. 2.1.2. Environmental data In the analysis, a total of nine variables were used: the number of wind-induced interruptions, the length of overhead lines, three soil temperature variables, and four wind speed variables. The soil temperature is determined at three different depths, whereas the different wind speed variables result from different ways to aggregate the data across measurement stations and time, as detailed in Section 3.2. Hourly wind speed (average wind speed and wind gust speed) measurements were obtained from the open database of the Finnish Meteorological Institute [50]. The wind speed dataset included measured wind speeds from 450 stations. The data were aggregated to daily values. Soil temperatures in three different depths (0–7 cm, 7–28 cm, and 28–100 cm) were obtained from the ERA5-Land meteorological reanalysis data produced by the ECMWF. Meteorological reanalysis datasets aim to represent past weather and climate conditions with complete data coverage [51]). Reanalysis datasets are produced by combining meteorological observations with forecasts of numerical weather prediction models. ERA5-Land has an hourly temporal resolution and a spatial resolution of 0.08 ×0.08◦, which is approximately 9 ×9 km at the equator. To our knowledge, only a very limited number of studies have discussed the accuracy of ERA5-Land soil temperatures, and in Finland, such a comparison has only been made to the predecessor of ERA5-Land, ERA Interim, which had a much lower spatial and temporal resolution than ERA5-Land. In the master’s thesis of Rimali [52], the resolution of the soil temperature data was higher compared with the very coarse resolution of ERA Interim. Previous studies have used the soil temperature of ERA5-Land Lembrechts et al. [53], and thus, the decision was to use ERA5-Land also in this study. The data are more easily accessible than the national, not openly available soil frost observation data of the Finnish Environment Institute. 2.2. Background analyses Interruptions are divided into five geographical areas defined in the study (Fig. 1). The characteristics of the areas are shown in Table 2. The areas roughly correspond to the network areas of the DSOs; however, in the case of large DSOs, the network areas are split into smaller sections that are located in different areas. The data of these DSOs are divided into areas in proportion to the overhead line length having an accurate coordinate-based location based on the open data provided by the National Land Survey of Finland (NLS, 2019). The network characteristics contain medium-voltage network lines, which are mainly 20 kV lines. The areas differ significantly from each other; for example, Area 5(N) is located in northern Finland, which is sparsely populated. The network structures are also different between areas, and the environmental locations of the lines vary. In the southern and western parts of the country, the remaining overhead lines are more often located in fields, whereas elsewhere in the country they are mainly built in forests. For instance, the largest proportion of weatherproof overhead lines is found in Area 1(SW) located in southwestern Finland. This makes a significant difference in the failure probability of a line. Furthermore, in the southern parts of the country, a large proportion of overhead lines have been replaced by underground cables over the past ten years. Thus, for a fair comparison, the annual interruption statistics are weighted by the annual overhead line lengths to ensure that the decreasing overhead line length because of the increasing underground cable length is correctly taken into consideration. The term weatherproof is used for overhead line sections for which the average tree height in the proximity is less than the height of the powerlines. The highest wind gusts are found in Area 1(SW), where the daily average peak gust is 11.8 m/s. The wind gusts are lowest in Area 3 (E), being 25 % lower than in Area 1(SW). Moreover, there is significant seasonal variation in interruptions of supply in the electricity distribution systems. This is illustrated in Figs. 2 and 3, which show the average numbers of interruptions caused by all sources of faults and interruptions caused by wind. The interruption frequencies are presented in proportion to the overhead line length. Table 2 Characteristics of five geographical areas in Finland. In the table, the term weatherproof is used for overhead line sections for which the average tree height in the proximity is less than the height of the powerlines. 1) (EA, 2019), 2) (LUKE 2019) ,3 (FMI, 2022). Area 1 (SW) Area 2 (SE) Area 3 (E) Area 4 (W) Area 5 (N) Total Network length, total (km) 1) 30 823 11 431 41 112 49 014 20 526 152 906 Network length, overhead line (OHL) (km) 1) 12 864 8368 33 839 28 838 14 358 98 267 Network length, underground cables (km) 1) 17 959 3062 7272 20 177 6168 48 470 Weatherproof OHL (km) 1,2) 5415 2345 3507 7770 2338 21 375 Proportion of weatherproof OHL (%) 1,2) 42 % 28 % 10 % 27 % 16 % 22 % Number of customers (millions) 1) 1.19 0.34 0.68 1.05 0.39 3.66 OHL length per customer (m/customer.) 1) 11 24 50 27 37 27 Average height of trees (m) 2) 15.7 13.9 14.4 15.6 9.6 13.4 Average temperature in a year (◦C) 3) 5.8 5.3 4.2 3.7 1.9 4.2 Days below 0 ◦C in a year (1/a) 3 80 95 108 108 147 538 Average peak wind gust on a day (m/s) 3 11.8 11.1 8.9 10.4 10.1 10.5 Number of meteorological stations 123 40 96 104 87 450 J. Haakana et al. Heliyon 10 (2024) e40846 5 It can be observed from Figs. 2 and 3 that the occurrences of interruption are close to each other in Areas 1–4. The same result is seen with the statistics of interruptions caused by all sources of faults (Fig. 2) and interruptions caused by wind (Fig. 3). The figures indicate that interruptions occur more often in autumn and mid-winter. In mid-summer, the number of interruptions is also higher than in spring and at the beginning of autumn. In the Finnish conditions, wind-induced interruptions are emphasized in autumn and midsummer. The hypothesis is that soil frost reduces the number of wind-induced interruptions in winter and spring; however, the number of snow-induced interruptions increases when snow loads increase the stress and number of interruptions in the overhead line network. Ten storms that have caused most electricity interruptions in Finland between 2008 and 2018 are listed in Table 3. It can be seen that these storms caused 28 % of all the wind-induced interruptions, whereas wind gusts of the other storms caused the rest 72 % of the wind-induced interruptions. 3. Methodology In this section, the methodology to analyze the effect of wind speed and soil frost on faults in the electricity distribution network is presented. There are numerous methods available for the analysis; however, the study focuses on the linear regression model. It was selected as the methodology to estimate the effect of wind speed and soil frost on power system interruptions. The reason behind selecting the linear regression model is that it is well known, and it provides a transparent analysis to evaluate results based on limited amount of input variables [54]. Thus, the results can be easily incorporated in power system planning and asset management. The other models, such as different decision trees or machine learning models, may yield a higher correlation and slightly better estimates; however, they may not necessarily enhance the usability of the results. 3.1. Statistical analyses The frequency of interruptions, i.e., the number of interruptions per day per km of overhead line, was the object of primary interest. To this end, the daily numbers of interruptions were divided by the length of the overhead line. The data were analyzed separately for each of the five areas. Pearson’s product-moment correlation coefficient and the linear model [55] were used to analyze the data. The marginal effects of different variables were calculated as follows. First, consider a linear univariate model for interruption frequency yi= α i+βxi+ ε i.(1) In such a model, one unit of x changes y by the amount coefficient β in area i, whereas α is constant, and ε is an error term. For the linear model, the term β is the effect size. The variable x can be wind speed, the power of wind speed, or soil frost. The confidence interval (CI) of the coefficient β determined in the study was calculated as CI =[βlow,βhigh],(2) where [βlow,βhigh]is the usual confidence interval of β [55]. In these analyses, the response variable is the frequency of interruptions. It represents the average number of interruptions in each area per day per power line kilometer. In this paper, the significance of the results is considered by using the P-value. In addition, to avoid the peril of P-value hacking, the P-values were multiple testing corrected by the false discovery rate (FDR) method presented in Ref. [56]. The FDR method provides another approach to confirm the validity of the results, while being less overly conservative than the Bonferroni correction. Further, the paper considers a multivariate model with an interaction term [57]. The model assesses the combined effect of wind speed and soil frost on the power network interruption frequency yi= α +β1x1i+β2x2ix1i+ ε i,(3) where one unit of x 1 affects y by the amount β1 in area i, multiplication of the binary variable x 2 , and the variable x 1 affects y by the amount β2 in area i. In the model, x 1 is the wind speed or its power, β1 is the effect of wind speed, x 2 is the binary variable of soil frost, Fig. 2. Average number of all monthly interruptions per overhead line length experienced by electricity end users. J. Haakana et al. Heliyon 10 (2024) e40846 6 and β2 is the effect of soil frost. The definitions previously considered for the confidence intervals of the univariate model related to the determination of the coefficients are valid for the multivariate linear models. 3.2. Effect of wind speed In the analyses, two primary wind speed variables were used to determine the best correlation between wind speed and occurrence of interruptions. The speed variables are the 10 min average wind speed (WS) and the maximum of the 3 s wind speed for a 10 min measurement period (WG) for each measurement station. From the 10 min values, the maximum values were determined for each hour. In the analyses, daily average and maximum values of the wind speed variables were used. From these values, regional values were determined for the wind speed variables. To determine the best correlation between wind speed and the number of wind-induced electricity interruptions, we considered four ways to aggregate the data. Both the WS and the WG were taken into consideration by calculating the daily average or maximum values for the five regions under study based on measurements obtained from several stations. In the analysis, also the use of cubes of the variables was considered because the power delivered by wind is proportional to the third power of wind speed. This can be derived from the kinetic energy of a moving mass: Ek=1 2mv2,(4) where v is the speed. To solve the mass m, consider a surface perpendicular to the flow (e.g., the canopy of a tree). The mass flux through this surface is ˙ m= ρ Av,(5) where ρ is the density of the air and A is the area of the surface. Thus, the mass equals m=Δt˙ m,(6) where Δt is the time interval. The power is given by P=Ek Δt= 1 2Δt˙ mv2 Δt=1 2 ρ Av3∝v3.(7) However, the literature indicates that the relation between wind speed and wind damage can be higher than the third power [32]. Fig. 3. Average number of monthly wind-induced interruptions per overhead line kilometer experienced by electricity end users. Table 3 Power supply interruptions in 2008–2018 caused by ten strongest windstorms. Number of interruptions caused by wind Proportion (%) Date Name of windstorm 5911 7.2 % 26 Dec. 2011 Tapani 3253 4.0 % 13 Dec. 2013 Seija 3167 3.9 % 17 Nov. 2013 Eino 2932 3.6 % 27 Dec. 2011 Hannu 1491 1.8 % 30 Nov. 2012 Antti 1449 1.8 % 27 Aug. 2016 Rauli 1403 1.7 % 23 Nov. 2008 not named 1269 1.6 % 2 Oct. 2015 Valio 926 1.1 % 1 Dec. 2013 Oskari 920 1.1 % 4 Jun. 2009 Toivo 58 884 72.2 % - Others J. Haakana et al. Heliyon 10 (2024) e40846 7 Thus, we examined the correlation of the wind speed and the number of power interruptions with powers between 1 and 20. The aggregations of wind speed variables used in the analysis are described in Table 4. The definitions and explanations of the wind speed under consideration were determined by the following formulations: - WS max,max , the daily maximum of the hourly maximum taken over the maximum 10 min wind speeds at the measurement stations in the area - WS ave,max , the daily maximum of the hourly average taken over the maximum 10 min wind speeds at the measurement stations in the area - WG max,max , the daily maximum taken over the maximum wind gusts at the measurement stations in the area - WG ave,max , the daily maximum of the hourly average taken over the maximum wind gusts at the measurement stations in the area WSmax maxa=(maxtmaxiWSi,t)a,i=1…N,t=1…T,a=1...20 (8) WSave maxa=(maxt∑N i=1WSi,t N)a ,t=1…T,a=1...20 (9) WGmax maxa=(maxtmaxiWGi,t)a,i=1…N,t=1…T,a=1...20 (10) WGave maxa=(maxt∑N i=1WGi,t N)a ,t=1…T,a=1...20 (11) where a is the power of wind speed, i is the group of measurement stations within the area under consideration, N is the number of measurement stations, t is the group of time periods within a day, and T is the number of time periods within a day. The methodology to determine the dependence of wind speed and the number of fault interruptions is shown in Fig. 4. 3.3. Method to analyze the effects of soil frost The effect of soil frost is assessed with the selected wind speed aggregation determined in the previous analysis concerning wind speed. In this paper, the analysis is carried out with soil temperature information from the ERA5-Land reanalysis, because the information is openly available and the actual soil frost is only sparsely measured. The soil temperature was compared from each depth (0–7 cm, 7–28 cm, and 28–100 cm) with three cut-off values of the soil temperature, 0 ◦C, −0.5 ◦C, and −1.0 ◦C, respectively. For simplicity, it was assumed that soil frost is present if the soil temperature is below the set temperature cut-off value. Pearson’s product moment correlation coefficients were calculated for the multivariate linear regression model where the wind is one explanatory factor for the wind-induced interruptions and soil frost-based variable is another factor. The effect of soil frost is determined as the multiplication of wind speed aggregation and binary frost variable. Soil frost correlation is assessed for each soil temperature cut-off value and depth. The methodology to select the best-correlating soil frost depth and temperature is presented in Fig. 5. 4. Results The results of the study are based on the methodology presented in the previous section. 4.1. Effects of wind on the number of faults in the electricity distribution network As a result of the analysis, correlation coefficients were determined for the wind speed variables used in the study. The correlation of the wind speed and fault interruptions was calculated with powers between 1 and 20. The results of powers 1, 3, and 10 are presented in Table 5 and the powers between 1 and 20 in Appendix Table A1. Based on the results (Table 5 and Appendix Table A1), WG ave,max 10 is the wind speed variable providing the best correlation for the majority of the areas considered in the study. The correlation coefficients of the tenth power are significantly higher compared with the smaller first and third powers. Moreover, in general, it can be stated that the hourly average wind speed (both wind gusts and 10 min wind speed) over the measurement stations in the area provides a higher correlation compared with the maximum wind speed over the measurement stations in the area. The reason for this is that the larger the area influenced by the windstorm is, the wider the effects of Table 4 Aggregation of different wind speed measures in the analysis. Option Definition Variable Aggregation (station level) Aggregation (time level) 1WS max,max 10-min wind speed Maximum Daily maximum 2WS ave,max 10-min wind speed Average Daily maximum 3WG max,max Wind gust Maximum Daily maximum 4WG ave,max Wind gust Average Daily maximum J. Haakana et al. Heliyon 10 (2024) e40846 8 the storm are. A closer consideration of the best correlations reveals that in one of the areas, Area 1(SW), the best correlation is found with WG ave,max 20 . The effect of wind gusts on the occurrence of interruptions is presented in Table 6, which shows the results of a univariate regression model fitted for the wind speed. It shows the coefficient of wind speed, its confidence intervals, P-value, FDR value, and the modeled average interruption frequency for the areas under study. The interruption frequency is determined with the model, and the wind gust distribution from a period of 11 years. The P-value and FDR determined in the study indicate that the results are significant even though the effect sizes vary between the areas. Based on the analysis, the coefficients of wind are between 2.8E-13 and 3.0E-12 on average, depending on the area under consideration. Considering the interruption frequency, the numbers of interruptions vary between 0.01 and 0.32 interruptions per day per 1000 km of overhead line. It can be observed that the farther south the area is, the more interruptions are caused by wind. This is reasonable because the trees are taller (Table 1) and soil frost is less common in the south, and thus, it has a lower effect in the south. It can be noticed that the modeled interruption frequency of Area 1(SW) is higher compared with the other areas. This can partly be explained by one major storm having the strongest effect in Area 1(SW). The statistics also show that one major storm in December 2011 (named Tapani in Finland and cyclone Dagmar in the other Nordic countries) caused 25 % of all wind-induced interruptions. Fig. 6 presents the distribution of wind gusts occurring in Area 4(W), the numbers of interruptions caused by different wind speeds, and the fitted number of interruptions based on the selected tenth power of WG ave,max . The data are from a period of 11 years between 2008 and 2018. The figure shows that the wind speed WG ave,max has almost a normal distribution, but the interruptions caused by wind occur with high wind speeds so that the probability of an interruption increases exponentially when the wind speed increases. Thus, a small number of highly windy days cause most of the wind-induced interruptions. In Area 4(W) this means, for instance, that 64 % of all wind-induced fault interruptions occur on days when the maximum wind speed is 15 m/s or higher, such days accounting for 8 % of all days. On average, for all the areas these percentages are 68 % and 10 % for the wind speed of 15 m/s or higher. The same percentage values, on average, with the wind speed of 10 m/s or higher are 92 % and 51 %, respectively. The figure also shows that the model fits Fig. 4. Methodology to determine the statistical dependence between wind speed and occurrence of interruptions in electricity distribution. J. Haakana et al. Heliyon 10 (2024) e40846 9 Table A.1 (continued) Area 1 (SW) Area 2 (SE) Area 3 (E) Area 4 (W) Area 5 (N) Average WS ave,max 10 0.72 0.39 0.69 0.75 0.4 0.59 WG max,max 10 0.39 0.6 0.36 0.63 0.07 0.41 WG ave,max 10 0.80 0.61 0.75 0.78 0.44 0.68 WS max,max 11 0.37 0.64 0.69 0.62 <0.01 0.58 WS ave,max 11 0.77 0.36 0.67 0.75 0.39 0.59 WG max,max 11 0.39 0.58 0.31 0.63 0.06 0.39 WG ave,max 11 0.84 0.61 0.74 0.78 0.42 0.68 WS max,max 12 0.37 0.64 0.68 0.62 <0.01 0.58 WS ave,max 12 0.81 0.33 0.65 0.74 0.38 0.58 WG max,max 12 0.39 0.57 0.27 0.62 0.05 0.38 WG ave,max 12 0.87 0.6 0.73 0.77 0.41 0.68 WS max,max 13 0.36 0.63 0.67 0.62 <0.01 0.57 WS ave,max 13 0.84 0.30 0.63 0.73 0.37 0.57 WG max,max 13 0.38 0.55 0.24 0.62 0.04 0.37 WG ave,max 13 0.89 0.59 0.72 0.77 0.4 0.67 WS max,max 14 0.36 0.63 0.66 0.61 <0.01 0.57 WS ave,max 14 0.86 0.27 0.6 0.73 0.36 0.56 WG max,max 14 0.37 0.53 0.21 0.62 0.03 0.35 WG ave,max 14 0.91 0.58 0.71 0.76 0.39 0.67 WS max,max 15 0.35 0.62 0.65 0.61 <0.01 0.56 WS ave,max 15 0.88 0.24 0.58 0.72 0.35 0.55 WG max,max 15 0.36 0.51 0.18 0.61 0.02 0.34 WG ave,max 15 0.92 0.57 0.69 0.75 0.39 0.66 WS max,max 20 0.32 0.59 0.61 0.58 <0.01 0.42 WS ave,max 20 0.93 0.15 0.49 0.68 0.31 0.51 WG max,max 20 0.3 0.43 0.1 0.59 0.01 0.29 WG ave,max 20 0.95 0.50 0.64 0.72 0.36 0.63 Fig. A.1. Interruptions caused by wind in Area 4(W). The bars show the mean number of interruptions, and the line diagrams illustrate two different fits for the daily mean number of interruptions with the wind speed WG ave,max , which is the daily maximum of average hourly measured maximum wind gusts. The fits are for the third and tenth powers of WG ave,max . The data are from the years 2008–2018, consisting of 4018 days and 22 881 interruptions. J. Haakana et al. Heliyon 10 (2024) e40846 16 Fig. A.2. Interruptions caused by wind in Area 1(SW). The bars show the mean number of interruptions, and the line diagram depicts the proportion of days with the wind speed WG ave,max , which is the daily maximum of average hourly measured maximum wind gusts. The data contain 3457 days without soil frost and 561 days with soil frost. The total number of interruptions in the electricity distribution network was 21 979 without soil frost and 1273 with soil frost. Fig. A.3. Interruptions caused by wind in Area 2(SE). The bars show the mean number of interruptions, and the line diagram depicts the proportion of days with the wind speed WG ave,max , which is the daily maximum of average hourly measured maximum wind gusts. The data contain 3106 days without soil frost and 912 days with soil frost. The total number of interruptions in the electricity distribution network was 8649 without soil frost and 413 with soil frost. J. Haakana et al. Heliyon 10 (2024) e40846 17 Fig. A.4. Interruptions caused by wind in Area 3(E). The bars show the mean number of interruptions, and the line diagram depicts the proportion of days with the wind speed WG ave,max , which is the daily maximum of average hourly measured maximum wind gusts. The data contain 2718 days without soil frost and 1300 days with soil frost. The total number of interruptions in the electricity distribution network was 21 784 without soil frost and 3628 with soil frost. Fig. A.5. Interruptions caused by wind in Area 4(W). The bars show the mean number of interruptions and the line diagram depicts the proportion of days with the wind speed WG ave,max , which is the daily maximum of average hourly measured maximum wind gusts. The data contain 2924 days without soil frost and 1094 days with soil frost. The total number of interruptions in the electricity distribution network was 20 984 without soil frost and 1897 with soil frost. J. Haakana et al. Heliyon 10 (2024) e40846 18 Fig. A.6. Interruptions caused by wind in Area 5(N). The bars show the mean number of interruptions, and the line diagram depicts the proportion of days with the wind speed WG ave,max , which is the daily maximum of average hourly measured maximum wind gusts. The data contain 2154 days without soil frost and 1864 days with soil frost. The total number of interruptions in the electricity distribution network was 685 without soil frost and 313 with soil frost. References [1] L. Ma, V. Christou, P. Bocchini, Framework for probabilistic simulation of power transmission network performance under hurricanes, Reliab. Eng. Syst. Saf. 217 (2022) (2022) 108072, https://doi.org/10.1016/j.ress.2021.108072. ISSN 0951-8320. [2] Finnish Energy, Energy distribution interruptions 2010–2019 – Finnish Energy, available at: https://energia.fi/uutishuone/materiaalipankki/sahkon_ keskeytystilastot_2010-2019.htmarginal#material-view. (Accessed 18 December 2023). [3] M. Tian, Z. Dong, L. Gong, X. Wang, Line hardening strategies for resilient power systems considering cyber-topology interdependence, Reliab. Eng. Syst. Saf. 241 (2024) (2024) 109644, https://doi.org/10.1016/j.ress.2023.109644. ISSN 0951-8320. [4] C. Winzer, Conceptualizing energy security, Energy Pol. 46 (2012) 36–48, https://doi.org/10.1016/j.enpol.2012.02.067. ISSN 0301-4215. [5] A. Schweikert, P. Chinowsky, X. Espinet, M. Tarbert, Climate change and infrastructure impacts: comparing the impact on roads in ten countries through 2100, Procedia Eng. 78 (2014) (2014) 306–316, https://doi.org/10.1016/j.proeng.2014.07.072. ISSN 1877–7058. [6] C. Fant, B. Boehlert, K. Strzepek, P. Larsen, A. White, S. Gulati, Y. Li, J. Martinich, Climate change impacts and costs to U.S. electricity transmission and distribution infrastructure, Energy 195 (2020) (2020) 116899, https://doi.org/10.1016/j.energy.2020.116899. ISSN 0360-5442. [7] G. Hou, K.K. Muraleetharan, V. Panchalogaranjan, P. Moses, A. Javid, H. Al-Dakheeli, R. Bulut, R. Campos, P.S. Harvey, G. Miller, K. Boldes, M. Narayanan, Resilience assessment and enhancement evaluation of power distribution systems subjected to ice storms, Reliab. Eng. Syst. Saf. 230 (2023), 2023, 108964, ISSN 0951-8320. [8] R. Tervo, I. L´ ang, A. Jung, A. M¨ akel¨ a, Predicting power outages caused by extratropical storms, Nat. Hazards Earth Syst. Sci. 21 (2021) 607–627, https://doi. org/10.5194/nhess-21-607-2021. [9] J. Jasi¯ unas, T. Heikkinen, P.D. Lund, I. L´ ang-Ritter, Resilience of electric grid to extreme wind: considering local details at national scale, Reliab. Eng. Syst. Saf. 232 (2023) 109070, https://doi.org/10.1016/j.ress.2022.109070. [10] M.M. Alam, Z. Zhu, B. Eren Tokgoz, J. Zhang, S. Hwang, Automatic assessment and prediction of the resilience of utility Poles using unmanned aerial vehicles and computer vision techniques, Int. J. Disaster Risk Sci. 11 (2020) 119–132, https://doi.org/10.1007/s13753-020-00254-1. [11] T.K. Laurila, H. Gregow, J. Corn´ er, V.A. Sinclair, Characteristics of extratropical cyclones and precursors to windstorms in northern Europe, Weather Clim. Dynam. 2 (2021) 1111–1130, https://doi.org/10.5194/wcd-2-1111-2021. [12] M. Laapas, I. Lehtonen, A. Ven¨ al¨ ainen, H. Peltola, 10-year return levels of maximum wind speeds under frozen and unfrozen soil forest conditions in Finland, Climate 7 (2019) 62, https://doi.org/10.3390/cli7050062. [13] R.J. Campbell, Weather-Related Power Outages and Electric System Resiliency, Congressional Research Service (CRS), Publication Date, 2012. Last accessed 18 December 2023, url: https://sgp.fas.org/crs/misc/R42696.pdf. (Accessed 1 August 2012). [14] A. Ven¨ al¨ ainen, I. Lehtonen, M. Laapas, K. Ruosteenoja, O.P. Tikkanen, H. Viiri, V.P. Ikonen, H. Peltola, Climate change induces multiple risks to boreal forests and forestry in Finland: a literature review, Global Change Biol. 26 (2020) 4178–4196, https://doi.org/10.1111/gcb.15183. [15] Energy Authority of Finland, Technical indicators of electricity grid operation 2020 (in Finnish: S¨ ahk¨ overkkotoiminnan tekniset tunnusluvut 2020), available at: https://energiavirasto.fi/verkkotoiminnan-julkaisut, 2021 (last access: 18 December 2023). [16] Y. Iijima, A.N. Fedorov, H. Park, K. Suzuki, H. Yabuki, T.C. Maximov, T. Ohata, Abrupt increases in soil temperatures following increased precipitation in a permafrost region, central Lena river basin, Russia, Permafr. Periglac. Process. 21 (1) (2010) 30–41. [17] H. Peltola, S. Kellom¨ aki, H. V¨ ais¨ anen, Model computations of the impact of climatic change on the windthrow risk of trees, Climatic Change 41 (1999) 17–36, https://doi.org/10.1023/A:10053998223191999. [18] C. Saad, Y. Boulanger, M. Beaudet, P. Gachon, J.-C. Ruel, S. Gauthier, Potential impact of climate change on the risk of windthrow in eastern Canada’s forests, Climatic Change 143 (2017) 487–501, https://doi.org/10.1007/s10584-017-1995-z. [19] M. Rantanen, A.Y. Karpechko, A. Lipponen, K. Nordling, O. Hyv¨ arinen, K. Ruosteenoja, T. Vihma, A. Laaksonen, The Arctic has warmed nearly four times faster than the globe since 1979, Commun. Earth Environ. 3 (2022) 168, https://doi.org/10.1038/s43247-022-00498-3. [20] S. Kellom¨ aki, M. Maaj¨ arvi, H. Strandman, A. Kilpel¨ ainen, H. Peltola, Model computations on the climate change effects on snow cover, soil moisture and soil frost in the boreal conditions over Finland, Silva Fenn. 44 (2) (2010) 213–233, https://doi.org/10.14214/sf.455. [21] I. Lehtonen, A. Ven¨ al¨ ainen, M. K¨ am¨ ar¨ ainen, A. Asikainen, J. Laitila, P. Anttila, H. Peltola, Projected decrease in wintertime bearing capacity on different forest and soil types in Finland under a warming climate, Hydrol. Earth Syst. Sci. 23 (2019) 1611–1631, https://doi.org/10.5194/hess-23-1611-2019. J. Haakana et al. Heliyon 10 (2024) e40846 19 [22] V. Eyring, S. Bony, G.A. Meehl, C.A. Senior, B. Stevens, R.J. Stouffer, K.E. Taylor, Overview of the coupled model Intercomparison project phase 6 (CMIP6) experimental design and organization, Geosci. Model Dev. (GMD) 9 (2016) 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016. [23] E. Kjellstr¨ om, G. Nikulin, G. Strandberg, O.B. Christensen, D. Jacob, K. Keuler, R. Vautard, European climate change at global mean temperature increases of 1.5 and 2◦C above pre-industrial conditions as simulated by the EURO-CORDEX regional climate models, Earth Syst. Dynam. 9 (2018) 459–478, https://doi.org/ 10.5194/esd-9-459-2018. [24] O. Martius, S. Pfahl, C. Chevalier, A global quantification of compound precipitation and wind extremes, Geophys. Res. Lett. 43 (2016) 7709–7717, https://doi. org/10.1002/2016GL070017. [25] J. Zscheischler, S. Westra, B.J.J.M. van den Hurk, S.I. Seneviratne, P.J. Ward, A. Pitman, A. AghaKouchak, D.N. Bresch, M. Leonard, T. Wahl, X. Zhang, Future climate risk from compound events, Nat. Clim. Change 8 (2018) 469–477, https://doi.org/10.1038/s41558-018-0156-3. [26] H. Gregow, H. Peltola, M. Laapas, S. Saku, A. Ven¨ al¨ ainen, Combined occurrence of wind, snow loading and soil frost with implications for risks to forestry in Finland under the current and changing climatic conditions, Silva Fenn. 45 (1) (2011) 35–54, https://doi.org/10.14214/sf.30. [27] A. Jalkanen, U. Mattila, Logistic regression models for wind and snow damage in northern Finland based on the National Forest Inventory data, For. Ecol. Manag. 135 (1–3) (2000) 315–330, https://doi.org/10.1016/S0378-1127(00)00289-9. [28] A. Zubizarreta-Gerendiain, P. Pellikka, J. Garcia-Gonzalo, V.-P. Ikonen, H. Peltola, Factors affecting wind and snow damage of individual trees in a small management unit in Finland: assessment based on inventoried damage and mechanistic modelling, Silva Fenn. 46 (2) (2012) 181–196, https://doi.org/ 10.14214/sf.441. [29] H. Gregow, Impact of Strong Winds, Heavy Snow Loads and Soil Frost Conditions on the Risks to Forests in Northern Europe. PhD Thesis, University of Eastern Finland, 2013. [30] S. Suvanto, M. Peltoniemi, S. Tuominen, M. Strandstr¨ om, A. Lehtonen, High-resolution mapping of forest vulnerability to wind for disturbance-aware forestry, For. Ecol. Manag. 453 (2019), https://doi.org/10.1016/j.foreco.2019.117619. [31] O.F. Eikeland, F. Maria Bianchi, I.S. Holmstrand, S. Bakkejord, S. Santos, M. Chiesa, Uncovering contributing factors to interruptions in the power grid: an Arctic case, Energies 15 (2022) 305, https://doi.org/10.3390/en15010305. [32] H. Valta, I. Lehtonen, T.K. Laurila, A. Ven¨ al¨ ainen, M. Laapas, H. Gregow, Communicating the amount of windstorm induced forest damage by the maximum wind gust speed in Finland, Adv. Sci. Res. 16 (2019) 31–37, https://doi.org/10.5194/asr-16-31-2019. [33] R. Eskandarpour, A. Khodaei, Machine learning based power grid outage prediction in response to extreme events, IEEE Trans. Power Syst. 32 (4) (2017) 3315–3316, https://doi.org/10.1109/TPWRS.2016.2631895. [34] D.B. McRoberts, S.M. Quiring, S.D. Guikema, Improving hurricane power outage prediction models through the inclusion of local environmental factors, Risk Anal. 38 (12) (2018) 2722–2737, https://doi.org/10.1111/risa.12728. [35] D. Cerrai, D.W. Wanik, M.A.E. Bhuiyan, X. Zhang, J. Yang, M.E.B. Frediani, E.N. Anagnostou, Predicting storm outages through new representations of weather and vegetation, IEEE Access 7 (2019) 29639–29654, https://doi.org/10.1109/ACCESS.2019.2902558. [36] D. Cerrai, P. Watson, E.N. Anagnostou, Assessing the effects of a vegetation management standard on distribution grid outage rates, Elec. Power Syst. Res. 175 (October 2018) (2019) 105909, https://doi.org/10.1016/j.epsr.2019.105909. [37] P.L. Watson, A. Spaulding, M. Koukoula, E. Anagnostou, Improved quantitative prediction of power outages caused by extreme weather events, Weather Clim. Extrem. 37 (2022), https://doi.org/10.1016/j.wace.2022.100487. [38] H. Liu, R.A. Davidson, D.V. Rosowsky, J.R. Stedinger, Negative binomial regression of electric power outages in Hurricanes, J. Infrastruct. Syst. 11 (4) (2005) 258–267, https://doi.org/10.1061/(asce)1076-0342(2005)11:4(258). [39] S.D. Guikema, R.A. Davidson, H. Liu, Statistical models of the effects of tree trimming on power system outages, IEEE Trans. Power Deliv. 21 (3) (2006) 1549–1557, https://doi.org/10.1109/TPWRD.2005.860238. [40] H. Liu, R.A. Davidson, T.V. Apanasovich, Spatial generalized linear mixed models of electric power outages due to hurricanes and ice storms, Reliab. Eng. Syst. Saf. 93 (6) (2008) 897–912, https://doi.org/10.1016/j.ress.2007.03.038. [41] S.R. Han, S.D. Guikema, S.M. Quiring, K.H. Lee, D. Rosowsky, R.A. Davidson, Estimating the spatial distribution of power outages during hurricanes in the Gulf coast region, Reliab. Eng. Syst. Saf. 94 (2) (2008) 199–210, https://doi.org/10.1016/j.ress.2008.02.018. [42] S.R. Han, S.D. Guikema, S.M. Quiring, Improving the predictive accuracy of hurricane power outage forecasts using generalized additive models, Risk Anal. 29 (10) (2009) 1443–1453, https://doi.org/10.1111/j.1539-6924.2009.01280.x. [43] S.D. Guikema, S.M. Quiring, Hybrid data mining-regression for infrastructure risk assessment based on zero-inflated data, Reliab. Eng. Syst. Saf. 99 (2011) 178–182, https://doi.org/10.1016/j.ress.2011.10.012. [44] R. Nateghi, S. Guikema, S.M. Quiring, Power outage estimation for tropical cyclones: improved accuracy with simpler models, Risk Anal. 34 (6) (2014) 1069–1078, https://doi.org/10.1111/risa.12131. [45] D.W. Wanik, E.N. Anagnostou, B.M. Hartman, M.E.B. Frediani, M. Astitha, Storm outage modeling for an electric distribution network in Northeastern USA, Nat. Hazards 79 (2) (2015) 1359–1384, https://doi.org/10.1007/s11069-015-1908-2. [46] D.W. Wanik, J.R. Parent, E.N. Anagnostou, B.M. Hartman, Using vegetation management and LiDAR-derived tree height data to improve outage predictions for electric utilities, Elec. Power Syst. Res. 146 (2017) 236–245, https://doi.org/10.1016/j.epsr.2017.01.039. [47] D. Cerrai, M. Koukoula, P. Watson, E.N. Anagnostou, Outage prediction models for snow and ice storms, Sustain. Energy, Grids Netw. 21 (2020) 100294, https://doi.org/10.1016/j.segan.2019.100294. [48] S. Hartling, V. Sagan, M. Maimaitijiang, W. Dannevik, R. Pasken, Estimating tree-related power outages for regional utility network using airborne LiDAR data and spatial statistics, Int. J. Appl. Earth Obs. Geoinf. 100 (2021) 102330, https://doi.org/10.1016/j.jag.2021.102330. [49] O. R¨ ais¨ anen, S. Suvanto, J. Haapaniemi, J. Lassila, Crown snow load outage risk model for overhead lines, Appl. Energy 343 (2023), https://doi.org/10.1016/j. apenergy.2023.121183. [50] Finnish Meteorological Institute: Wind gust observation data 2008-2018, available at: https://en.ilmatieteenlaitos.fi/download-observations,License:CCBY4.0, (last access: 18 December 2023). [51] J. Mu˜ noz Sabater, ERA5-Land hourly data from 1950 to present, Copernicus Climate Change Service (C3S) Climate Data Store (CDS) (2019), https://doi.org/ 10.24381/cds.e2161bac (last access: 18 December 2023). [52] A. Rimali, Analysis of Frozen Ground in Finland: Affecting Environmental Factors, Trends in Northern Finland and Applicability of Satellite Data, MSc, 2020. Thesis. [53] J.J. Lembrechts, J. van den Hoogen, J. Aalto, M.B. Ashcroft, P. De Frenne, J. Kemppinen, M. Kopecký, M. Luoto, I.M.D. Maclean, T.W. Crowther, J.J. Bailey, S. Haesen, D.H. Klinges, P. Niittynen, B.R. Scheffers, K. Van Meerbeek, P. Aartsma, O. Abdalaze, M. Abedi, J. Lenoir, Global maps of soil temperature, Global Change Biol. 28 (2022) 3110–3144, https://doi.org/10.1111/gcb.16060. [54] T.M.H. Hope, Linear regression, Machine Learning: Methods and Applications to Brain Disorders 67–81 (2020), https://doi.org/10.1016/B978-0-12-8157398.00004-3. [55] Bent Jørgensen, The Theory of Linear Models, Book 21, Chapman & Hall/CRC, Boca Raton, FL, 1993. Chapman & Hall/CRC Texts in Statistical Science. [56] Y. Benjamini, Y. Hochberg, Controlling the false discovery rate: a practical and powerful approach to multiple testing, J. Roy. Stat. Soc. B 57 (1995) 289–300. [57] N. Huntington-Klein, The Effect: an Introduction to Research Design and Causality, CRC Press, 2021. [58] A. Ven¨ al¨ ainen, H. Tuomenvirta, M. Heikinheimo, S. Kellom¨ aki, H. Peltola, H. Strandman, V¨ ais¨ anen, Impact of climate change on soil frost under snow cover in a forested landscape, Clim. Res. 17 (2001) 63–72, https://doi.org/10.3354/cr017063. J. Haakana et al. Heliyon 10 (2024) e40846 20