Relative probability of predator-scat occurrence
We developed RSFs for predators (Manly et al., 2002), where ‘used’ samples were the locations of predator-specific scats along transect lines and ‘available’ samples were random locations within a 50-m buffer on each side of the line in a ratio of 1 scat:10 random points. We used an exponential RSF design fitted using logistic regression. We did not enter correlated variables (r > |0.6|) into the same model. We used model selection and Akaike’s Information Criterion corrected for small sample sizes (AICc) to identify the best-supported RSF for each species using a criterion of four ∆AIC points to distinguish between competing models.
We used terrain, landcover, and anthropogenic features expected to influence predator occurrence (Nielsen et al., 2002; Whittington et al., 2005; Knopff et al., 2014).We measured landcover as cover type derived from TM Landsat Imagery, vegetation “greenness” from the Normalized Difference Vegetation Index (NDVI), forest fires ≤ 14 years old (Hebblewhite, 2006), and cutblocks ≤ 20 years since harvest (Visscher & Merrill, 2009) as mean proportion of 30 x 30-m pixels within a 1.3-km radius (5.3-km2) buffer around a scat or random location. This buffer size reflected the average daily movement of black bears the shortest distance of all the predator species (see Appendix S1, Supporting Information). Forest edge was based on a 30-m buffer of conifer or mixed-deciduous forest with any other landcover type. We derived terrain features (slope, elevation, and terrain ruggedness) from a 30-m Digital Elevation Model. Locations were designated as within 30-m or farther than 30-m of a trail. Forest edge, proximity to roads, off-highway vehicle (motorized) trails, and waterways were measured as the shortest distance (km) to the nearest feature or as density (km km-2) within the buffer. We compared the predictions of scat-based RSF of wolves and bears to a telemetry-based RSF at 1,000 random points. Scat-based RSF values increased as telemetry-based values increased, but was not strictly linear (Appendix S2, Supporting Information).
To include a numeric component, we weighted the wolf RSF by a probability density function (PDF) based on kill rates derived from annual, wolf pack size of collared packs (Hebblewhite & Merrill, 2007; Berg, 2019), and weighted the RSF by the annual PDF in proportion of scats collected in each year. We weighted the grizzly bear RSF by estimates of grizzly bear densities, which were 2.4 times higher within BNP compared to outside of BNP ( Whittington & Sawaya, 2015; Government of Alberta, 2016). We smoothed the PDF values along the BNP border using a 12.9-km moving window, the size corresponding to the average home range for local grizzly bears (Nielsen et al., 2002). RSFs of cougars and coyotes were not numerically weighted because no numeric indices were available.