Protected areas are essential tools for reducing global biodiversity loss. To fulfill their ecological functions, protected areas must be connected to their surroundings, a requirement that is difficult to meet in landscapes intensively disturbed by anthropogenic activities. Therefore, protecting movement corridors at the edges of protected areas is crucial, especially for species with broad habitat needs, such as the American marten (Martes americana). However, habitat selection and space use patterns are dynamic processes, so we could expect that functional connectivity would vary temporally in response to changing environmental conditions and levels of human activities. In this study, we aimed to predict the location of movement corridors for the American marten in Forillon National Park and its periphery during two contrasted periods (snow-free: May–November; snow-covered: December–April). We used “seasonal” resource selection functions to identify core areas (interpreted as circuit “nodes”) and CircuitScape to delineate movement corridors between them based on the electrical circuit theory. Habitat selection patterns differed between periods, with martens avoiding open areas, high elevations, and road proximity during the snow-free period, while avoiding areas closer to secondary roads but selecting areas closer to primary roads and housing structures during the snow-covered period. Consequently, the location of movement corridors differed partially between periods; functional connectivity was favored by the presence of forest patches for both periods, while being constrained by open environments, especially during the snow-free period. Our study highlights the importance of modeling functional connectivity at fine temporal scales to provide movement corridors that fulfill the requirements of a species at each period of its annual cycle.

We captured American martens from September to December 2020 and from October to December 2021 using certified live traps (Tomahawk model 202, Wisconsin, USA, and Havahart model 1078, Pennsylvania, USA). Captured martens were physically restrained using a handling cone and anesthetized with an intramuscular injection (rear thigh) of BAM (combination of butorphanol, azaperone, and medetomidine; Wildlife Pharmaceuticals, Inc., Windsor, Colorado, USA). Martens weighing ≥ 650 g were fitted with a Litetrack 20 (Lotek Engineering Inc, Newmarket, Ontario, Canada, 21 g) GPS collar programmed to acquire location fixes every 6 h for nearly 10 months. From January 2021 to May 2022, we collected GPS locations from collared martens through remote data download, recaptures, or trapped individuals.

We used digital maps to obtain habitat covariates that could potentially influence the habitat selection and space use patterns of martens.  We also used 1: 20,000 digital terrain models, and digital maps of all roads, trails, and recreational facilities published by the MRNF. We conducted our habitat selection analyses using RSF models (Manly et al., 2002) at the 3^rd order of selection (sensu Johnson, 1980), thus contrasting habitat covariates found at the sites used by martens (i.e., GPS locations) with those found at sites available in individual home ranges (i.e., random points). To reflect seasonal variation in resource selection, we partitioned GPS locations into two distinct biological periods: snow-free (from April 12^th to November 14^th) and snow-covered (from November 15^th to April 11^th). For each GPS location and random point, we extracted a set of habitat covariates, including habitat type, elevation, and Euclidean distance to roads and housing structures. We parameterized seasonal RSFs using mixed logistic regression models and set individual martens as a random factor (i.e., random intercept) to prevent pseudoreplication and account for differences in sample sizes (Gillies et al., 2006). For both periods, we used the coefficients of the highest ranked RSF model to calculate, following Boyce et al. (2002; see equation 2), and map the relative occurrence probability over the entire study area.

We used CircuitScape (v. 4.0.5; McRae et al., 2008) to conduct our two functional connectivity analyses (one per period). CircuitScape needs two rasters to operate: the first indicating the core areas to be connected and the second representing landscape resistance to animal movement. We used our previously created maps of relative occurrence probability to develop our rasters (pixels 50 m × 50 m) of core areas and landscape resistance. For both periods, core areas were determined by selecting the 20% highest ranked habitats (i.e., pixels with a relative occurrence probability ≥ 80%) that had a minimum area of 1 km², to maintain suitable habitat patches as well as stepping stones. We thus assumed that pixels with a higher relative occurrence probability offered lower costs to travel than those with a lower relative occurrence probability (Beier et al., 2008). Following Finch et al. (2020), we defined functional connectivity corridors as contiguous pixels with a high current density (i.e., upper quartile, ≥75% of the maximum value) connecting at least two core areas of suitable habitat quality (with a relative occurrence probability ≥ 80%).