Motion-triggered video cameras reveal spatial and temporal patterns of red fox foraging on carrion provided by mountain lions

Study area

Our study area covered approximately 2,300 km² in the southern GYE north of Jackson, Wyoming (Fig. 2). Our research was conducted on Bridger-Teton National Forest (United States Forest Service, USFS Authorization ID JAC760804), Grand Teton National Park (NPS Permit GRTE-2012-SCI-0067), and National Elk Refuge (USFW permit NER12). Elevations ranged from 1,800 m to >3,600 m contributing to long winters with high snow pack. The terrain was characterized by glacially scoured sagebrush valleys, densely forested north-facing slopes, and uplifted south facing cliffs with abundant evidence of landslides. Ungulate prey such as pronghorn (Antilocapra americana) and mule deer (Odocoileus hemionus) migrated out of the study area during winter, while elk (Cervus elaphus) remained and aggregated in lower elevations where they received supplemental feed (Jones et al., 2014). Red fox competitors for carrion resources included mountain lions, black bears (Ursus americanus), grizzly bears (Ursus arctos horribilis), and gray wolves, as well as coyotes (Canis latrans), bald eagles (Haliaeetus leucocephalus), golden eagles (Aquila chrysaetos), ravens (Corvus corax) and black-billed magpies (Pica pica). Potential prey species for foxes in our study area included voles (Microtus spp.), northern pocket gophers (Thomomys talpoides), deer mice (Peromyscus maniculatus), snowshoe hares (Lepus americanus), chipmunks (Tamias spp.), Uinta ground squirrels (Spermophilus armatus), ruffed grouse (Bonasa umbellus), and dusky grouse (Dendragapus obscurus). The availability of small mammals varied seasonally, as some species hibernated (chipmunks and ground squirrels) and others were influenced by accessibility determined by snow and ice pack. Red foxes also consume invertebrates (Macdonald, 1977) and seeds from whitebark pine (Pinus albicaulis; Cross, 2015), which were also present in our study area. Further details about the flora and fauna of the study area are found in Elbroch et al. (2013).

Mountain lion capture and collar programming

Our mountain lion capture protocols followed guidelines outlined by the American Society of Mammalogists (Sikes et al., 2011), as previously described in Elbroch et al. (2013) and were approved by two independent Institutional Animal Care and Use Committees (IACUC): the Jackson IACUC (Protocol 027-10EGDBS-060210) and National Park Service IACUC (IMR_GRTE_Elbroch_Cougar_2013-2015), with permission to handle cougars granted by the Wyoming Game and Fish Department (Permit 297). We captured mountain lions with baited cage traps and scent-trailing hounds from 2012–2015. Captures took place during winter months when snow facilitated the detection of mountain lions by researchers and cooler temperatures created safer animal handling conditions. Mountain lions were immobilized with ketamine (4.0 mg/kg) and medetomidine (0.07 mg/kg), and their vital signs were monitored at 5-min intervals. We fit them with a GPS collar (Lotek Globalstar S or Iridium M, Newmarket, Ontario; Vectronics Globalstar GPS Plus, Berlin, Germany), and the effects of the capture drugs were reversed with Atipamezole (0.375 mg/kg). We remained with the mountain lions until they were mobile and then we observed their departure from the capture site.

Data collection at mountain lion kills

CyberTracker-certified researchers (Evans et al., 2009; Elbroch et al., 2011) conducted site investigations of aggregated GPS locations, or “clusters” (Anderson Jr & Lindzey, 2003), in the field, as previously described by Elbroch et al. (2014). We defined clusters as two or more GPS locations within 150 m of each other, at least 4 hrs in duration and at most, 2 weeks apart; we included clusters made at any time of day or night. When we discovered prey during site investigations, we judged whether there was sufficient meat to draw in scavengers (Elbroch et al., 2017). If so, we placed paired motion-triggered cameras (Bushnell Outdoor Products, Overland Park, KS, USA) with 32 GB cards to record activity at the kill. Cameras were programmed to record 60 s videos with 30 s delay intervals, and they gathered data at the carcass for varying lengths of time, dependent upon the availability of carrion, whether the carcass was moved, and the battery life of the camera (Elbroch et al., 2017). Between 2012 and 2015 we placed motion triggered cameras at 232 mountain lion kill sites of prey >25 kg, 138 during the winter months and 94 during the summer.

Red fox behavior analyses

We cataloged each video collected at mountain lion kills in an Access database, recording species present, species interactions, vocalizations, and other behaviors (e.g., feeding, resting, moving). For each kill site, we determined whether foxes were detected, confirmed that they had indeed scavenged the carcass, and also estimated the number of foxes at the carcass. As we were unable to reliably differentiate between individuals, especially from coarse black and white footage obtained at night, we defined the minimum number of foxes at each kill site as the maximum number of individuals counted in any video from the kill in the same frame (i.e., at the same time). We estimated “total” fox feeding time by summing the 60 s videos in which one or more foxes were visibly feeding on the carcass. We classified behaviors such as looking, listening, and moving (gaining a vantage point to investigate a sound or sight) as vigilance behavior.

We compiled all data and conducted all our analyses across two seasons, summer (May–October) and winter (November–April), based on typical weather patterns of the region (SNOTEL weather station 506), and the influence that weather has on food availability for foxes and mountain lion foraging behaviors. First, we tested whether there was a difference in the proportion of mountain lion kills at which foxes were detected scavenging in summer versus winter with a two-proportions z-test, and then we compared the mean number of foxes detected at kills in the two seasons with a t-test.

We determined the mountain lion’s departure time and date from video data, and confirmed it with GPS data collected from collars. We separated the data into summer and winter, as detailed in ‘Spatial analyses to determine the distribution of red fox scavenging’. Then we compared red fox presence before and after the mountain lion’s departure (binary data) with a McNemar test, a variation of a contingency test that can compare paired categorical rather than continuous data, to determine whether foxes were waiting for mountain lions to leave before scavenging.

We classified a mountain lion as “in residence” from the time that the kill was made, as determined with GPS data, to the final feeding bout at the carcass, as determined by video footage and GPS data. In cases when red foxes fed while mountain lions were in residence, we also analyzed fox vigilance behavior. To estimate vigilance, we divided the number of 60 s videos in which red foxes were feeding by the total number of videos in which red foxes were recorded exhibiting any behavior (e.g., feeding + moving, watching, listening). Then we used a t-test to determine whether foxes exhibited different proportions of their time feeding at the carcass while the mountain lion was still feeding from the carcass versus after it had departed.

Spatial analyses to determine the distribution of red fox scavenging

Although it was well documented that prey aggregate in winter versus summer in our study system (Jones et al., 2014), we conducted simple nearest neighbor analyses in ArcGIS to determine whether the mountain lion kills we sampled with cameras were also more clustered in winter than summer. We did this for kills in the southern half of our study area (southern half of red rectangle in Fig. 2), where we had the most data. We compared these distances with a t-test to determine whether they were statistically different.

Then we identified five landscape variables as potentially important predictors of where foxes were likely to occur at mountain lion kill sites: elevation (m), aspect (transformed into categories of North, East, South, West), terrain ruggedness (vector ruggedness measure; VRM), habitat type, and distance to forest edge. We estimated elevation using a digital-elevation model (DEM) at a resolution of 30 m (http://datagateway.nrcs.usda.gov/). We then used ArcGIS 10.0 Spatial Analyst Tools to derive values of slope and aspect from the DEM. In addition, we also derived a vector ruggedness measure (VRM) from the DEM (following Sappington, Longshore & Thompson, 2007). A Gap Analysis Program (GAP) land cover (gapanalysis.usgs.gov/gaplandcover) was used, at a resolution of 30 m, which included 87 cover classes, which we reclassified into four cover classes based on similarity of land cover types: (1) open meadows, (2) shrub-steppe, (3) forested, (4) and riparian zones. We converted all forested lands from the GAP into a polygon layer in ArcGIS to and then created edge layers as the perimeter of each forested section. All distance variables were standardized by subtracting their means and dividing by their standard deviations, to aid in model convergence and for direct comparison of variables (Gelman & Hill, 2006).

We conducted our analyses for two seasons, as described above. We examined the relationship between mountain lion predation sites (following methods in Elbroch et al., 2013) and where foxes scavenged at predation sites by comparing known kill sites where foxes were documented to kill sites where they were absent. To do so, we used generalized linear mixed models that allowed for random effects to control for variation among samples. We included the number of days a camera recorded videos while a carcass was present as a random intercept to account for sampling bias that may have resulted from different sampling intervals (Elbroch et al., 2017).

Prior to modeling, we used a correlation matrix to evaluate collinearity (—r— >  0.6) among predictor variables. No predictor variables were correlated (—r— <  0.50) and therefore, all variables remained in the modeling process. We then modeled all possible combinations of the five predictor variables. For the categorical variables of land cover and aspect, we used forested habitats and southerly aspect as a reference categories, both of which are commonly used by prey species.

We calculated Akaike’s Information Criterion adjusted for small sample size (AICc), ΔAICc, and Akaike weights (w_i) for each model (Burnham & Anderson, 2002). We considered our top-ranked model and those models with ΔAICc values ≤2.0 to have provided the strongest empirical support for explaining variation in the locations where red foxes scavenged mountain lion kills; in cases where top models included nesting, we retained the simplest models to avoid the inclusion of uninformative parameters (Burnham & Anderson, 2002; Arnold, 2010). Lastly, we calculated evidence ratios for top models (Burnham, Anderson & Huyvaert, 2011) and assessed effect sizes of informative parameters by calculating Cohen’s d test statistics (difference between the means/pooled standard deviation). Statistics were conducted in R (R Core Team, 2017) with the lme4 (Bates et al., 2015) and MuMIn (Barton, 2016) packages.

Mountain lion kill sites monitored and red fox scavenging

On average a camera remained at a kill site for 6.4 (±SD = 3.7) days in the winter and 6.1 (±3.3) days in the summer. Because we omitted mountain lion kills <25 kg from this analysis, we detected red foxes at 149 kills rather than the 151 documented in Elbroch et al. (2017). Foxes scavenged a greater proportion of winter kills than summer kills (70.3% in winter vs. 48.9% in summer; z = 3.291, P = 0.001), and on average scavenged longer at winter kills than summer ($\bar{X}=102.7\pm 138.3$ min for winter vs. $\bar{X}=39.7\pm 74.0$ for summer; t₁₃₅ =  − 3.014, P = 0.003). When foxes were present, there were approximately 1.5 times more foxes per kill in winter than summer (1.83 vs. 1.16 foxes per kill, respectively; t₁₃₅ = 5.94, p < 0.001).

Mountain lion presence influenced red fox scavenging in summer, but not in winter. Red foxes preferentially scavenged after the departure of the mountain lion in summer (20 kills while the mountain lion was still feeding from the carcass vs. 36 kills after the mountain lion had departed; ${\bar{X}}_{135}=11.25$, P = 0.001), but scavenged regardless of the presence of the mountain lion in winter (61 kills while the mountain lion was still feeding from the carcass vs. 65 kills after the mountain lion had departed; χ²₁₃₅ = 0.25, P = 0.62). At kills at which foxes scavenged while mountain lions were present, foxes did not exhibit any changes in vigilance after the mountain lion departed (t ₆₃ = 5.94, P = 0.80).

Spatial patterns of red fox scavenging

Winter mountain lion kills were more aggregated than summer kills (t ₉₈ = 4.17, P < 0.001). In winter, kills were 456.6 (CIs [361.5–551.8]) m apart, and in summer, 1007.8 (CIs [762.4–1253.3]). Our analyses also identified different spatial attributes of red fox scavenging, dependent upon season, though these patterns were weak overall. Two top models provided insights into winter patterns (Table 1). The top-ranked model received 2.7 times more empirical support than our second model based on an evidence ratio, and emphasized variation in distance to forest edge (β = 0.473, SE = 0.263) in locations where red foxes were scavenging mountain lion kills versus where they were not. Nevertheless, this effect was small (Cohen’s d = 0.34). Our second-ranked model contributed insights absent in our top model, and showed that red foxes were more likely to be detected at higher elevations than lower (β = 0.279, SE = 0.194). Again, this effect was small (Cohen’s d = 0.27). Our null model, which only included the intercept and random effect (Table 1), was only 2.01 ΔAICc from the top-ranked model, emphasizing the fact that the covariates we tested in our models little influenced the spatial patterns of red fox scavenging in winter.

For summer, our analysis identified one top model that accounted for 26.8% of model weights (Table 1), and emphasized variation in the elevation at which foxes were scavenging. In contrast to winter, foxes were more likely to occur at mountain lion kills at lower elevations relative to where mountain lions kills occurred across the landscape (β =  − 0.472, SE = 0.245). This effect was small, approaching medium (d = 0.41). As during winter, the null model performed well, only 2.14 ΔAICc from the top-ranked model (Table 1), indicating that red foxes were not strongly selecting for any spatial attributes of mountain lion kills at which to scavenge.