Introduction

The magnitude and rate of change in land-use cover presents a challenge in endeavouring to fully understand the responses of wildlife to ever-encroaching human environments (Torres et al. 2016; Watson et al. 2016). While concern is rightly given on the conservation of roadless areas across the globe (Selva et al. 2011; Ibisch et al. 2016; Ascensão et al. 2018), there is increasing interest on how already existing and growing road networks may impact wildlife persistence (Bennett 2017). The expansion of roads in human-dominated landscapes can exacerbate habitat loss and fragmentation, lead to direct mortality by vehicle collisions and increase disturbance and human pressures on wildlife (e.g., Mech et al. 1988; Fahrig et al. 1995; Saunders et al. 2002; van Langevelde and Jaarsma 2004; Shephard et al. 2008; Ceia-Hasse et al. 2017; Laurance et al. 2017). By studying wildlife responses to growing road networks, conservationists are better able to delineate and prioritize effective mitigation measures to ensure the persistence of wildlife and functional ecosystems, in an increasingly human-dominated world (Forman 2000; Boutin and Hebert 2002; Dellinger et al. 2013; Wadey et al. 2018).

Evidence shows that large carnivores have an ability to persist in human-dominated landscapes, therefore indicating that human-large carnivore separation is not a necessary condition for their conservation (López-Bao et al. 2017). The impact of linear infrastructures on large carnivores, and their behavioural responses, has been the focus of attention in recent times (e.g., Whittington 2005; Llaneza et al. 2012; Basille et al. 2013; Dellinger et al. 2013; Boulanger and Stenhouse 2014; Ordiz et al. 2014; Riley et al. 2014; Ceia-Hasse et al. 2017; Find’o 2019; López-Bao et al. 2019; Zeller et al. 2021; Proctor et al. 2019). For example, roads decrease Amur tiger (Panthera tigris altaica) survivorship and reproductive success (Kerley et al. 2002), isolate lynx (Lynx lynx) habitat patches (Kramer-Schadt et al. 2004), and lead to grizzly bear (Ursus arctos) avoidance of landscapes with high road densities (Mace et al. 1996). Furthermore, studies have shown how carnivores can perceive risks of crossing roads, with black bears (Ursus americanus) having elevated heart rates when crossing high-traffic volume roads (Ditmer et al. 2018) or higher movement speeds when crossing roads than during other situations (Zeller et al. 2021), and grizzly bears (Ursus arctos horribilis) increasing their selection of roads closed to traffic (Whittington et al. 2019).

For species such as wolves (Canis lupus), accumulating evidence is leading an interesting debate on the levels of wolf resilience towards landscape transformations and human disturbance (e.g. Llaneza et al. 2012, 2016; Lesmerises et al. 2013; Ahmadi et al. 2014; Chapron et al. 2014; Bojarska et al. 2020); which may be highly context-dependent (e.g., Eurasia vs. North America; Chapron et al. 2014; Sazatornil et al. 2016; Muhly et al. 2019). With an estimated wolf density around 2.55 wolves/100 km2 (López-Bao et al. 2018), and a mean paved road density at 2.7 km/km2 (Llaneza et al. 2012), wolves in Galicia are expected to interact with roads frequently, as seen in other studies (e.g. Mladenoff et al. 1995; Whittington et al. 2004; Kaartinen et al. 2005; Lesmerises et al. 2013; Zimmerman et al. 2014; Ronnenberg et al. 2017). Forest roads and other trails can be advantageous to wolves, assisting movement through territories and consequently hunting and patrolling efficiency (Thurber et al. 1994; Musiani et al. 1998; James and Stuart-Smith 2000; Ciucci et al. 2003; Gurarie et al. 2011; Zimmerman et al. 2014; Dickie et al. 2017; Newton et al. 2017). On the other hand, roads and trails increase human accessibility (Thurber et al. 1994) and can lead to wolf mortality, either directly (i.e. vehicle collisions, Boyd and Pletscher 1999) or indirectly (i.e. hunting and poaching, Mech et al. 1988; Sazatornil et al. 2016; Suutarinen and Kojola 2018).

Several studies have suggested an ability of wolves to perceive mortality risks associated with humans, and therefore adjust, for instance, their use of space and time accordingly (Thurber et al. 1994; Merrill and Mech 2000; Habib and Khumar 2007; Ahmadi et al. 2014; Zimmerman et al. 2014). In regions where human presence is low, temporal use of roads can span both day and night (e.g., Thurber et al. 1994). However, in human-dominated landscapes, wolves become more cryptic and nocturnal, decreasing the probability of encountering humans (Vilà et al. 1995; Ciucci et al. 1997; Theuerkauf et al. 2003a; Kusak et al. 2005; Zimmerman et al. 2014). This evidence supports the idea that wolf responses to roads may be largely influenced by their use and the levels of traffic intensity, rather than the presence of the roads themselves (Hebblewhite and Merrill 2008; Kohn et al. 2009; Lesmerises et al. 2013). Since traffic intensity varies between different road types, presumably so too would wolf activity (Kohn et al. 2009).

Roads can be categorised into several classes depending on traffic volume, speed limit as well as physical attributes, such as number of lanes and fencing (van Langevelde and Jaarsma 2004; Colino-Rabanal et al. 2011). Consequently, distinguishing paved roads from one another would be a necessary step when studying wildlife responses to various road classes (Zeller et al. 2021). The Regional Government of Galicia has jurisdiction over most roads (although some of them are national jurisdiction), and they are classified into the following: (1) motorway (national and regional), (2) basic and complementary primary network (national and regional), (3) secondary network (regional) and (4) county roads network (regional). A number of studies investigating wolf, and large carnivore, movement patterns in relation to paved and forest roads grouped paved roads into one single class, namely primary roads, when analysing road network data (Theuerkauf et al. 2009; Gurarie et al. 2011; Zimmerman et al. 2014; Ordiz et al. 2015; López-Bao et al. 2019). Classification of paved road networks into more than one class has been described (e.g. primary and secondary roads; e.g., Theuerkauf et al. 2007; Ahmadi et al. 2014); but still the lack of differentiation between paved road classes can make wolf interactions to paved roads more difficult to interpret (see for example Find’o et al. 2019; Zeller et al. 2021).

Presumably, wolves persisting in close proximity to roads may be able to distinguish between different traffic intensities (Kaartinen et al. 2005; Whittington et al. 2005). With that in mind, we aimed to evaluate contrasting responses of Iberian wolves to different paved roads and traffic volume levels in a high road density region by employing data of 25 GPS collared wolves. Iberian wolves in NW Iberia have traditionally persisted in areas with high levels of human activity, and thus provide a good opportunity to investigate wolf responses to paved roads. We considered paved roads as separate road classes, according to their physical characteristics and traffic volume. Based on estimated crossing rates (which accounted for different lengths of road classes and wolf monitoring periods) made by GPS-collared wolves, we hypothesised that (1) wolves cross all paved road classes, (2) but crossing rates will be higher in low speed and low traffic volume roads compared with high speed and high traffic volume roads, (3) adults and sub-adults will cross roads more frequently than juveniles, due to adults and sub-adults moving further distances to hunt, patrol or disperse, and (4) wolves should cross all road classes more frequently during night hours, thereby reducing encounters with humans. We also hypothesised that the level of human activity, i.e. traffic volume, being different across road classes, influenced road use. We also expected mortality by vehicle collision to be highest on high-use road classes than on low-use classes, mainly because of the higher speed limit and traffic volume experienced on these road classes and therefore increasing the chances of a wolf being road-killed.

Methods

Study area

Fieldwork was carried out in Galicia (NW Spain, ca. 42.5°N 8.1°W; Fig. 1, highlighted in black box). The high human population in Galicia, mean population density of ca. 93 inhabitants/km2 (INE 2010), scattered in many settlements across the countryside, has resulted in a vast paved road network (mean paved road density 2.7 km/km2, Llaneza et al. 2012; Fig. 1b). There also exists a large network of dirt gravel roads (passable by 4 × 4 wheel drive, Fig. 1c), created to serve as access mainly for forestry and agricultural purposes. Consequently, the landscape of Galicia is heterogeneous and intensively managed by human activities comprised mainly of cropland (32%), scrubland (11%), and forestry plantations comprising of Eucalyptus spp. and Pinus spp. (43%). The last estimate of the status of wolves in Galicia was carried out between 2013 and 2015, resulting in the estimate of a continuous range distribution of wolves in the area, and 90 reproductive packs (Llaneza et al. 2015).

Fig. 1
figure 1

Galicia, north-western region of Iberian Peninsula, highlighted in black box. A Individual wolf home ranges used in this study (estimated as Minimum Convex Polygons using 100% of GPS locations; n wolves = 25. B Paved road network in Galicia, bold black lines representing motorways, national and regional roads, light roads representing secondary and county maintained roads. C Humanised landscapes of Galicia

Full size image

The studied wolves

We investigated wolf response to paved roads by studying the spatial behaviour of 25 individuals equipped with GPS-GSM collars (Followit, Sweden). Between 2006 and 2013, wolves were captured with Belisle© leg-hold snares (Edouard Belisle, Saint Veronique, PQ, Canada) and chemically immobilised by intramuscular injection of medetomidine (Domitor®, Merial, Lyon, France; 0.10 mg/kg) using a blowpipe. All wolves were evaluated as clinically healthy at the moment of capture, and they only had minor lesions associated with trapping (i.e., skin abrasions). In Spain, wolves north of the Duero River are listed in Annex V of the European Habitats Directive (92/43/EEC). Fieldwork procedures were specifically approved by the Regional Government of Galicia. The wolves included in this study were captured under permits 19/2006, 71/2009, 86/2011, and 095/2013 from the Regional Government of Galicia (Spain). All fieldwork procedures adhered to the Spanish animal welfare regulations (Spanish Decree 53/2013).

Sex and age were determined in situ. Age was estimated by dental pattern and tooth wear (Gipson et al. 2000) and wolves were classified into three categories: juveniles (< 1 yr, n = 6), sub-adults (1–2 yrs, n = 10) and adults (> 2 yrs, n = 9). All wolves, except 4 pairs, belonged to different packs. We used the locations of GPS collars which were obtained every two hours for this study. Wolves were monitored an average of 156 days (range 23–390 days). For the purposes of this study, we estimated the home range of every individual as the Minimum Convex Polygon (MCP) using 100% of locations during the entire monitoring period for each wolf (Fig. 1a). Accordingly, the mean (± SD) wolf MCP of the 25 wolves studied was 262.05 ± 253.46 km2 (Appendix B).

The paved road network within wolf home ranges: Paved road classes

Firstly, we categorised paved roads into four classes depending on their purpose, speed limit and traffic volume (Table 1). We used official information on the Galician paved road network (information provided by the Regional Government of Galicia and Fomento Ministry of Government of Spain) to quantify the extent of the paved road network within every wolf home range (MCP). In addition, we also reviewed this paved road network within each wolf MCP by overlapping the official spatial information provided with high-resolution digital orthoimages, since some roads may not be included in the official data (Llaneza et al. 2018). All paved roads that were missing were added to the network manually using ArcGIS (ESRI, California, USA). Paved roads were classified in accordance to their purpose, speed limit and traffic volume, with number of lanes and fencing details given to describe wolf accessibility (Table 1). The length of each paved road class within each MCP was measured in km to calculate road class densities (km/km2).

Table 1 Main characteristics of the four paved road classes within Galicia and considered in this study
Full size table

Traffic flow

We obtained data of traffic flow from the official transport records of the Regional Government of Galicia (Autonomous Community maintained roads) and from the Spanish Ministry of Development (motorways and national roads, references in Appendix A). Traffic flow for motorways, national and regional roads, as well as a few secondary roads were obtained from those sources. No data on traffic flow was available for county roads and therefore this class was removed for this specific analysis. For the other three road types within wolf MCPs, traffic flow stations on roads crossed by wolves were utilised, and the mean traffic flow on a daily and every 2 h time interval were used to test the influence of traffic volume on wolf response to paved roads.

Wolf road-kill dataset

We compiled official records on wolf mortality caused by vehicle collisions between 1991 and 2012 from the Regional Government of Galicia (Appendix F). The sex was known for all 54 dead wolves: 25 (46.3%) female and 29 (53.7%) male. Estimation of age was possible in 48 cases: 27 (50%) adults, 12 (22.2%) sub-adults and 9 (16.7%) juveniles. No age classification was given for 6 individuals (11.1%). Monthly distribution of vehicle collision mortalities was graphed (Appendix G).

Data analysis

For each individual, consecutive GPS locations were connected via straight-line segments, and these line segments were used to determine crossing rates, by overlapping them with the paved road network. Only line segments derived from effective locations every 2 h were considered, removing the small number of cases of missing locations. GPS success rate was very high in this area. For example, for 5 wolves in our dataset and 36,104 locations (range 3654–11,861 per wolf), GPS success rate was, on average, 98% per individual (Planella et al. 2016). When a line segment from the initial GPS location to the successive location was bisected by a road an intersect point was created using ArcGIS. Since we considered a 2 h time interval, our approach allowed us to derive a minimum number of crossings.

To test for differences in the frequency of wolf crossings in relation to the road class, we analysed the number of crossings on each road class by calculating an “Index Cross” for each wolf, accounting for sampling effort and road length. Thus, we took into account that the number of crosses depended on the number of days that each wolf was sampled (there was a significant and positive correlation between the number of crosses of paved roads and the number of sampling days, Pearson correlation analysis, rp = 0.856, P < 0.0001, n = 25). We also considered the length (km) of every road class within wolf MCPs (there was a significant and positive correlation between the number of crosses and the length of each paved road class, Pearson correlation, rp = 0.474, P = 0.017, n = 25).

The Index Cross for each paved road class within a wolf MCP was calculated as follows (Eq. 1):

$${\text{Index cross = }}{{{\text{Number of crosses}}} \mathord{\left/ {\vphantom {{{\text{Number of crosses}}} {\left( {{{{\text{Number of monitoring days}}} \mathord{\left/ {\vphantom {{{\text{Number of monitoring days}}} {{\text{Length of road}}\left( {{\text{km}}} \right)}}} \right. \kern-\nulldelimiterspace} {{\text{Length of road}}\left( {{\text{km}}} \right)}}} \right)}}} \right. \kern-\nulldelimiterspace} {\left( {{{{\text{Number of monitoring days}}} \mathord{\left/ {\vphantom {{{\text{Number of monitoring days}}} {{\text{Length of road}}\left( {{\text{km}}} \right)}}} \right. \kern-\nulldelimiterspace} {{\text{Length of road}}\left( {{\text{km}}} \right)}}} \right)}}$$
(1)

where the number of crosses were standardised for all wolves by dividing this figure by the ratio between the number of wolf monitoring days and road length (km). We log-transformed the Index Cross for subsequent analysis. For each wolf, we also separated every Index Cross into three time periods within the circadian cycle. We split the 24-h period into: (i) night (only star and lunar light, i.e., difficult to make out horizon), (ii) twilight (the sun is 6 degrees below the horizon) and (iii) day, the remaining time. In order to account for wolf crepuscular activity (Merrill and Mech 2003; Theuerkauf 2009), we buffered twilight 2 h before sunrise and 2 h after sunset. Daylight Savings Time was also accounted for.

We built general linear mixed models (GLMMs) with Gaussian error distribution and identity link, to compare a set of seven competing models explaining wolf response to paved roads, using as a proxy the Index Cross (response variable), according to age, road class, time period and their interactions. Firstly, we considered (i) the null model, (ii) a model containing the variable wolf age (3 levels), (iii) a model considering the road class (4 levels, Table 1), and (iv) a model considering time period (3 levels). Three interaction models were also employed, including the interaction between road class and time period, in order to explore different temporal use of paved roads according to the road class, the interaction between age and road class and the interaction between age and time period. The interactions with age were considered to test for individual differences in crossing rates according to individual attributes. As repeated measurements of the wolf responses to different paved roads are unlikely to be independent for the same individual, the identity of the individual was treated as a random factor in these models. Sex was not considered in the set of competing models since we did not detect differences in crossing rates between the sexes (GLMM using Index Cross as response variable, sex as predictor, and wolf identity as random factor, P = 0.984).

In addition, for the small number of roads within wolf MCPs where data on traffic flow was available (n roads = 54), we built another GLMM with Negative Binomial error distribution and log link to evaluate how traffic flow (mean daily number of vehicles) influenced the number of crossings. We used as a response variable the mean daily number of crossings over a given paved road made by a wolf (i.e., count data, not distributed following a Poisson distribution). The identity of the individual was also treated as a random factor in this model. For this dataset, we compared this model against a model considering road class (i.e., the road class according to Table 1), and a model considering both predictors (traffic flow and road class), in order to get insights into the idea of whether traffic flow is more important than the type of road considered.

Akaike Information Criterion with a second order correction for small sample size (AICc) was used for model selection (Burnham and Anderson 2010). We also used the Akaike weights (wi) to determine the relative strength of support for each competing model (Burnham and Anderson 2010). We used the “glmmADMB” package for R software (Skaug et al. 2014) to run GLMMs, the ‘‘car” package (Fox et al. 2015) to evaluate the significance levels for model parameters, and the “bbmle” package to calculate Akaike weights (Bolker 2017). Additionally, we also correlated the mean hourly traffic flow against the number of wolf crossings, and the number of mortality events by type of paved road with the Index Cross using a Spearman’s Rank Correlation analysis. Finally, Kruskal–Wallis test was used to test for significant differences in the wolf response (i.e., Index Cross) to traffic flow across time periods. All statistical analyses were performed in R 3.0.2 (R Core Team 2015).

Results

On average, the studied wolves persisted in areas with densities of paved roads at the home range level (MCP) ranging from 0.75 to 2.95 km/km2, with a mean paved road density of 1.92 km/km2 (Appendix B). The majority of wolf MCPs (98%) contained primary and secondary roads (Appendix B), and eight out of twenty-five wolves (32%) had motorways within their MCPs, with 62% of these wolves having crossed them. Paved road length was positively and significantly correlated with MCP home range size (Spearman’s rank correlation, rs = 0.975, n = 25, P < 0.0001).

Overall, in 3915 sampling days, we recorded 29,859 wolf crossings over paved roads (mean number of crosses by wolf = 1194, range 43–4172) (Appendix C). Wolf crosses of all paved road classes were recorded at a mean rate of 0.022 crossings/day/km (95% CI 0.016–0.027). Considering this mean rate, and the minimum length of paved roads observed within a wolf home range (39 km; Appendix B), this means that the studied wolves were expected to cross a paved road at least once every day. Road class IV (county roads) and class II roads (primary roads) were crossed more frequently, followed by road class III (secondary roads) and road class I (motorways) (Fig. 2). The median index cross over class I roads was 0.001 crossings/day/km, 0.018 in class II roads, 0.010 in class III roads and 0.019 in class IV roads (Fig. 2). Crossings were most frequent at night (median, 0.014 crossings/day/km), then during dawn/dusk (0.003 crossings/day/km) and daylight (0.001 crossings/day/km) (Fig. 2). Adult (0.018 crossings/day/km) and sub-adult (0.028 crossings/day/km) wolves crossed roads more frequently than juveniles (0.009 crossings/day/km; Fig. 2), and all wolves crossed more frequently during night hours (Appendix D). The most parsimonious model explaining the variation in wolf crossings over paved roads was the model considering road class, time period and their interaction (wi = 1; Table 2). Wolf crossings over paved roads were significantly different across road types (Wald χ2 = 110.62, d.f. = 3, P < 0.0001; Appendix E); whereas time period and the interaction road class and time period did not show significant influence on wolf Index Cross (both P > 0.234; Appendix E).

Fig. 2
figure 2

Bar charts of median index cross (crossings/day/km) by road class, solar time, and wolf age with inter-quartile range displayed as error bars

Full size image
Table 2 Comparison of seven competing models built to explore the behavioural response of wolves to different paved roads in human-dominated landscapes of Galicia, NW Spain
Full size table

Traffic flow influenced the number of crossings registered by wolves. We detected a negative and marginal significant effect of traffic flow on the mean daily number of wolf crossings (Wald χ2 = 3.38, d.f. = 1, P = 0.066). The model containing traffic flow showed the highest support (wi = 0.50), compared to the model including road class (wi = 0.26) and the model with both predictors (wi = 0.25) (road class was not significant in both models (P > 0.382). The median response of mean traffic flow differed across time periods (Kruskal–Wallis, H = 367.50, d.f. = 11, P = 0.0001). Interestingly, when traffic flow and number of crossings were examined on a two hour basis, we observed a significant and negative correlation between mean traffic flow and the mean number of crossings (Spearman’s rank correlation, rs:− 0.385, n = 708, P = 0.0001; Fig. 3).

Fig. 3
figure 3

Mean number of crossings versus mean traffic across time (2-h interval)

Full size image

Out of the 54 records of wolves being road-killed in the area between 1991 and 2012, the majority of cases occurred on primary roads (class II) (61%), followed by secondary roads (class III) (28%), county roads (class IV) (9%) and motorways (class I) (2%) (Appendix F). We did not detect a significant correlation between the mean Index Cross by road class and the number of road kills (Spearman’s rank correlation, rs = 0.634, n = 4, P = 0.365; Fig. 4). But, the mean Index Cross for primary roads (class II) (0.026 crossings/day/km), and the number of road-killed wolves was the highest (Fig. 4). By looking at the monthly distribution of vehicle collision mortalities, it was observed that most collisions occurred in the early spring and late autumn months, with peaks in February and October (Appendix G).

Fig. 4
figure 4

Bar chart of mortality numbers against the Index Cross for each road class

Full size image

Discussion

Iberian wolves in Galicia live in a highly human-dominated landscape, with road densities (mean paved road density in wolf home ranges -MCPs-: 1.92 km/km2) well above road thresholds reported in most previous studies (between 0.23 km/km2 and 1.12 km/km2; e.g., Mech et al. 1988, 1989; Mladenoff et al. 1995, 1997, 1999; Theuerkauf et al. 2003a; Jȩdrzejewski et al. 2005; Kaartinen et al. 2005; Potvin et al. 2005; Mladenoff et al. 2009; Mattisson et al. 2013; Ronneberg et al. 2017). As expected, wolves crossed all four classes of paved roads in our study area; although further research is needed to explore in depth avoidance behaviours towards roads or particular landscape characteristics associated with crossing sites (see for example Baigas et al. 2017; Find’o et al. 2019; Zeller et al. 2021). Canada lynx (Lynx canadiensis) displayed no avoidance behaviour and were found to cross 2-lane highways every other day (Baigas et al. 2017). Elsewhere, wolves have been recorded crossing many paved roads, especially during dispersal events (e.g., Mech et al. 1995; Blanco and Cortés 2007). Since in this study we considered a 2 h time interval to calculate the number of wolf crossings, our approach allowed us to derive a minimum number of crossings.

Among road classes, wolves crossed more frequently low speed and low traffic volume roads, when compared with high speed and high traffic volume roads. County roads (class IV) had higher crossing rates in comparison to motorways (class I), presumably because pavement width is shorter, speed limit is slower and traffic volume is lower (see Skuban et al. 2017 for similar responses by brown bears). The higher density of county roads (class IV) within wolf MCPs (Appendix B) would also account for the higher crossing rates, as the probability of a wolf encountering this road class is greater. With the exception of two wolves, primary roads (class II) and secondary roads (class III) were present in all wolf home ranges (Appendix B). Interestingly, primary roads (class II) were the second most frequently crossed road class followed by secondary roads (class III). This road class has a wider pavement width and experiences higher speed limits and traffic volume compared with secondary roads (class III), which may lead to higher mortality rates, as discussed further (Fig. 4). The lower crossing rates shown by juveniles, may be related to the fact that wolves select breeding sites away from paved roads (Theuerkauf et al. 2003b; Iliopoulos et al. 2014; Sazatornil et al. 2016; Llaneza et al. 2018).

Of the wolves that had a section of motorway (class I) within their home range (n = 8), five crossed this road class and more than once for some individuals. However, we were unable to determine whether the five wolves that crossed a section of motorway used passages or crossed the motorway itself frequently. Crossings made by wolves on motorways have been described elsewhere (Licht and Fritts 1994; Kohn et al. 2000; Merrill and Mech 2000; Blanco et al. 2005; Ciucci et al. 2009; Lesmerises et al. 2013). Wolves were reported to cross a motorway after a period of local disturbance (i.e. roadwork completion, Lesmerises et al. 2013). In contrast, an increasing human presence on motorways has resulted in fewer crossings made by wolves in other regions (Alexander and Waters 2000; Musiani et al. 2010). As well as traffic, fencing can also affect wolf response to motorways. Fencing along a motorway can act as a barrier and lead to fewer crossings (Paquet and Callaghan 1996), while an unfenced motorway in Wisconsin had no”barrier effect” on wolf movements (Kohn et al. 2000). Remarkably, fenced motorways have not hindered the expansion of wolves in Europe (Blanco et al. 2005; Ciucci et al. 2009; Reinhardt et al. 2019). Overall, wolves crossed roads mainly during the night given that traffic volume increases during daylight hours. Similar patterns of nocturnal use of trails and paved roads were reported before (Vilà et al. 1995; Ciucci et al. 1997; Theuerkauf et al. 2003a; Blanco et al. 2005; Kusak et al. 2005; Hebblewhite and Merrill 2008; Houle et al. 2010; Gurarie et al. 2011; Zimmerman et al. 2014). In contrast, wolves living in areas with low traffic volume may not display such preferences for nocturnal activity (Thurber et al. 1994).

One of the major causes of death for wildlife in human-dominated landscapes is traffic mortality (Jaarsma et al. 2006). Over twenty-one years, the majority of wolves found road-killed were adults with mortalities occurring in the late winter/early spring (mating and breeding season) and early autumn months (dispersal). In comparison, high road mortality was observed for younger wolves in Italy but similarly to this study, adult mortality on roads occurred during the winter months (Lovari et al. 2007). We found wolf mortality to be highest on primary roads (class II), with the least number of mortalities occurring on motorways (class I). The high speed limit and traffic volume experienced on a primary road (class II) would certainly result in a higher probability of a wolf being road-killed. However, by contrast, wolf mortality was highest on motorways elsewhere in Spain, with fencing being reported as a key predictor in roadkill (Colino-Rabanal et al. 2011). Whilst traffic volume is an obvious explanation of mortality, the effects of fencing may not be immediately apparent, as they are usually thought of as an effective mitigation tool in reducing roadkill (Spanowicz et al. 2020). However, if a wolf enters a motorway through gaps in fencing, they may be forced to spend more time on the road searching for an exit, thus heightening their chances of being killed.

Road density may not greatly influence wolf distribution at a landscape level, rather it may be on a finer, home range level, that the presence of public-paved roads can have a negative influence on wolf movements in some regions (Ciucci et al. 2003; Zlatanova and Popova 2013). The variations seen in response to roads, and other human structures, can depend on many factors, including the rate of disturbance and persecution experienced (Ciucci et al. 2003; Sazatornil et al. 2016). Wolf persistence in human-dominated landscapes is supported by a number of behavioural adaptations to deal with risk and allow persistence (Theuerkauf et al. 2003a; Kusak et al. 2005; Ahmadi et al. 2014; Zimmerman et al. 2014; Llaneza et al. 2016). In Iberian Peninsula, wolves do not have a lot of options in inhabiting remote regions when available land is located at short distances from roads (Torres et al. 2016). Thus roads will continue to pose risks to wolves (Merrill and Mech 2000) but by selecting to cross roads with shorter pavement widths, slower speeds and lower traffic volume, predominately during night hours, wolves decrease the probability of being injured or killed by a vehicle.

Factors predicting wolf mortality on Galician roads should be considered in future landscape planning strategies in order to guarantee the conditions for long-term wolf persistence. An increasing focus should be allocated to identify drivers of wolf mortality on primary roads (class II) in our study area (Fig. 4), in order to identify segments of roads where mitigation efforts should be prioritised (e.g., crossing structures such as bridges, underpasses, or fencing; Grilo et al. 2010; Van der Ree et al. 2015; Spanowicz et al. 2020). Research has indicated that wildlife-vehicle collisions are not random events, but clustered (Clevenger et al. 2003; Ramp et al. 2006; Morelle et al. 2013; Keken et al. 2019). Models will be useful in determining these factors (Litvaitis and Tash 2008), and road-related predictors could include traffic volume, road alignment and road-side topography (Clevenger et al. 2003; van Langevelde and Jaarsma 2004; Barrientos and Bolonio 2009), as well as incorporating wolf movement/crossing patterns (this study). A predictive model in wolf-vehicle collision risk may be able to inform decision-makers on the exact location of where fencing, guard-rails or a crossing structure may be required (e.g., Malo et al. 2004). If fencing of particular segments along primary roads (class II) is to be implemented as a tool for reducing collision risk, funnelling animals towards suitable crossing structures should be prioritized (Huijser et al. 2016). Our study also highlights the importance of considering different paved road classes when studying the impact of roads on wildlife.