Study area
We conducted our study in the western part (ca. 250 km2) of the Lower Silesia Forest, western Poland (N 51° 14.2′–51° 27.1′, E 14° 57.5′–15° 10.9′, Fig. 1). Average annual temperature during the study was 10 °C (daily mean temperature varying between −16 and 29 °C), and snow cover (up to 10 cm deep) persisted for 7–71 days each winter (data provided by the Institute of Meteorology and Water Management). The area is flat (altitude 120–160 m a.s.l.) and is predominantly covered by forests managed by the Polish State Forests, a governmental organisation specialised in timber production. Forests are primarily composed of even-aged Scots pine Pinus sylvestris plantations, which are managed throughout the year by regular clearcutting, planting, and fencing of young stands (Bena 2012). To improve the efficiency of forest management and to enable access to logging sites, the forest is divided into rectangular compartments of 750 m × 375 m, most of which are surrounded by straight forest roads of 3–5 m width. Many such compartments are intersected by an additional road, which further divides tree plantations into squares of 375 × 375 m. As a result, the landscape consists of rectangular patches divided by a network of straight forest roads (Fig. 1). The density of forest roads is almost uniformly 4 km/km2 throughout the study area. Forest roads are used throughout the year, primarily by motor vehicles of foresters and heavy-duty machinery for wood extraction and transportation. Additionally, traffic on forest roads increases seasonally due to hunting, antler collecting, as well as berry and mushroom picking. The public is not allowed to use motor vehicles on forest roads. Due to shallow snow cover, these roads are rarely ploughed. Only few public paved roads run through the study area, their density is 0.2 km/km2.
Wolves began recolonising the study area in the 1980s and the process accelerated in the 2000s (Wolsan et al. 1992; Nowak and Mysłajek 2016). During the first two winters of our study (2012/13 and 2013/14), the area was occupied by one wolf pack, which later split its territory into two. The numbers of wolves in the original and the two resulting packs were dynamic throughout the study, counting from four to eight individuals per pack in winter. Wolves prey predominantly on three species of ungulates: red deer Cervus elaphus, roe deer Capreolus capreolus and wild boar Sus scrofa (Nowak et al. 2011). Wolves are legally protected in Poland; therefore, there has been no legal wolf hunting in the area since 1995. However, wolf poaching is regularly reported in the whole country (Gula 2008; Gula et al. 2009; Nowak and Mysłajek 2016).
Classification of forest roads
We classified roads by visual inspection at crossroads. Based on this inspection and through 100 photo-trapping (Bushnell Trophy Cam HD Aggressor) sessions (50 in winter and 50 in summer) lasting 6–10 days, we functionally classified forest roads into three categories: (1) high-traffic road (density 0.4 km/km2): maintained hard-surfaced gravel road used frequently for wood transportation, forestry activities and hunting (traffic 5–20 motor vehicles per day), (2) medium-traffic road (0.8 km/km2): road with visible signs of frequent use and little vegetation covering the middle of road surface, used regularly (up to five vehicles per day, mean four vehicles per week), (3) low-traffic road (2.8 km/km2): mostly passable by four-wheel drive cars but with most of the surface overgrown by low vegetation, used only sporadically when logging or hunting activities occurred in its proximity. Overall traffic on forest roads peaked during the working hours of forestry personnel (0700–1600 h) and people rarely drove on forest roads during the night. In the study area, sunrise was between 0444 and 0803 h and sunset between 1553 and 2118 h of local time. We did not include logging trails in the analysis, as they appear only temporarily during wood harvesting.
Data collection
To investigate road selection using telemetry data, we captured three wolves using Belisle foot snares (Belisle Enterprises, Canada) and fitted them with GPS collars that also included VHF transmitters (LOTEK Wireless Inc., Canada). The three collared individuals belonged to two packs: a young female (monitored December 2012–January 2013) belonged to the original pack that later split its territory (see above), whereas a breeding female (January 2015–September 2016) and a young male (January–August 2016) belonged to the northern of the two packs created from the original pack (Fig. 1). There was no collared individual in the southern pack created by the split. The GPS collars obtained locations at various intervals: every 2 h for the young female, and at 11-h intervals for the other two individuals with additional 1-h intervals for 1-week periods every month from December to April. The Polish authorities (General Directorate for Environmental Protection: DOP-OZGIZ.6401.08.12.2011.ls, and Regional Directorate for Environmental Protection in Gorzów Wielkopolski: WPN-I.6401.347.2015.JK) approved capture and handling protocols.
To investigate road selection using snow-tracking data, between December and March of 2012–2017, we located fresh wolf tracks when snow cover permitted. We used recent (obtained during the last 24 h) GPS locations and VHF signals of collared individuals to locate tracks. During winters when no individual was collared (2011/12, 2013/14 and 2016/17) or when tracking the non-collared pack, we searched for wolf tracks on forest roads, driving the road network of 1.5 × 1.5 km mesh size intersecting the whole study area. In the latter case, snow tracking always started at places where wolves crossed roads. When we found fresh wolf trails, we followed them on foot while recording the tracks on a hand-held GPS (Garmin GPSmap 62 s) set to record locations at 20-m intervals. We followed the wolf trails in both directions as long as the tracks were visible, or only backwards in case of very fresh tracks.
Finally, to investigate road use for scent marking, during snow-tracking sessions, we recorded all wolf scent marks on the trail (urination, scratching, defaecation). We attributed wolf trails and scent marks to a road if they were located on the road surface or within 3 m from the road edge. To avoid overestimation of road use by wolves, we excluded from the analyses trails that were shorter than 1 km or entirely on roads.
Data analysis
To analyse road selection using GPS-telemetry data, we applied a generalised linear model (GLM) with binomial distribution and logit-link function in R 3.6.1 (R Core Team 2019) using a binary response variable (1: wolf telemetry location, 0: random point). We generated random points within the seasonal (winter: December–March, spring–autumn: April–November) home ranges (100% minimum convex polygons) of each GPS-collared individual. For each wolf and each season, we used the same number of random points as there were telemetry locations (2,963 random points in total). We included as fixed variables the habitat type as a categorical factor with four levels (0—off-road, 1—low-traffic road, 2—medium-traffic road, 3—high-traffic road) and period of day as a nominal factor (working hours of forestry personnel: 0700–1600 h, off-work: the remaining time). We considered a telemetry location or a random location as on a road (road types 1–3) if it was within a 15-m buffer of the midline of the road, which corresponds to the location accuracy of the GPS collars model that we used (Di Orio et al. 2003).
To analyse road selection using snow-tracking data, we applied a GLM the same way as described above, but without the variable “period of day”. We generated wolf trail points every 218 m (to achieve the same number as of telemetry locations) along the wolf trails recorded during snow tracking (2963 wolf trail points) and the same number of random points inside the minimum convex polygon encompassing all the wolf trails.
To assess the daily distances travelled by wolves, on roads and off-road, we used telemetry data obtained during 128 days of intensive sampling of GPS locations (1-h or 2-h intervals) in winter. We corrected distance travelled by dividing the cumulative straight-line distances between consecutive wolf locations by correction coefficients provided by Mills et al. (2006): 0.68 and 0.60 for 1-h and 2-h GPS fix intervals, respectively. To assess wolf travel speed on roads and off-road, we combined telemetry data (proportion of time spent on roads) with snow-tracking data (proportion of distance on roads).
To analyse wolf use of roads for scent marking, we used GLM with binomial distribution and logit-link function in R 3.6.1 to test whether the probability of scent marking was related to its location (on roads or off-road, road category) and distance to junctions. We included a binary response variable (scent mark location/random point along wolf trail) and three fixed variables: habitat type (0—off-road, 1—low-traffic road, 2—medium-traffic road, 3—high-traffic road), distance to the closest junction (m) and their interaction. We performed the analyses separately for each type of scent marking (scats, urination, scratching). We generated the random points along wolf trails in numbers equal to numbers of scent marks of each type.
All spatial analyses were performed in QGIS version 3.2.3 (Quantum GIS Development Team 2018). We used model selection based on the Akaike Information Criterion (AICc) to rank models in the MuMIn package (Bartoń 2019) in R 3.6.1.