Characteristics of Winter Wolf Kill Sites in the Southern Yellowstone Ecosystem

Habitat selection occurs on a hierarchical scale (Johnson 1980), and it is generally assumed that animals select habitat to maximize their fitness (Thomson et al. 2006; DeCesare et al. 2014). This includes successful foraging, and for a predator successful foraging means first encountering prey and then successfully pursuing and killing the prey. Maximizing encounters and detection of prey by selecting areas of high prey density increases the likelihood of a successful predation event (Hopcraft et al. 2005; Bergman et al. 2006). Wolves Canis lupus are opportunistic predators (Becker et al. 2008) and tend to select the most abundant prey (i.e., highest biomass; Kunkel and Pletscher 2001; Smith et al. 2004; but see research on coastal wolves by Darimont et al. 2008; Adams et al. 2010; Lafferty et al. 2014), although research also suggests wolf prey selection is further influenced by prey vulnerability (Bergman et al. 2006; Mattioli et al. 2011). Wild ungulates comprise the majority of the diet of wolves (Mattioli et al. 1995; Gazzola et al. 2007; Imbert et al. 2016), and unsurprisingly, wolves also favor areas of high ungulate density (Hebblewhite and Pletscher 2002; Garrott et al. 2005; Alexander et al. 2006; Bergman et al. 2006), particularly in winter (Fritts and Mech 1981; Forbes and Theberge 1996).

In the greater Yellowstone ecosystem, USA, elk Cervus elaphus are the most abundant ungulate (Houston 1982; Hand et al. 2014) and are the primary prey of wolves in this region (Stahler et al. 2006; U.S. Fish and Wildlife Service [USFWS] et al. 2006). In Yellowstone National Park, wolf winter landscape selection parallels elk landscape selection (Mao et al. 2005). Winter elk habitat preferences include low elevation (Boyce et al. 2003; Garrott et al. 2005), gentle slope (Bowyer et al. 1998), high solar radiation (e.g., south-facing slopes; D'Eon and Serrouya 2005; Henry 2009), and low snow water equivalent (Henry 2009). These areas have shallower snowpack (Keating et al. 2007) and facilitate movement for ungulates. In winter (December–March), ungulates typically congregate when accumulating snow decreases their ability to move freely and limits access to food (Demarais and Krausman 2000). Grouping benefits individuals in the herd through increased vigilance and decreased risk of predation through dilution (Roberts 1996; Creel and Winnie 2005).

Wolf habitat use is affected by prey availability and physical geography (Alexander et al. 2006; Oakleaf et al. 2006; Whittington et al. 2011; Uboni et al. 2015). Wolves select hunting areas based on prey abundance (Kauffman et al. 2007, but see Merkle et al. 2017) and increased prey catchability (i.e., less energy expenditure and increased chance of success; Hebblewhite et al. 2005; Hopcraft et al. 2005), or they select areas that facilitate the strategy of a coursing predator (Husseman et al. 2003; Milakovic et al. 2011). As group-hunting, coursing predators, wolves often rely on the pursuit of prey over open terrain (Estes and Goddard 1967; Mech 1970) where kills likely occur when the prey tires. In rougher terrain, wolves also use hiding cover for stalking prey or capture prey in “terrain traps” such as ravines and gullies (Kunkel and Pletscher 2000, 2001).

From approximately mid-December to mid-March, elk in northwestern Wyoming were fed hay on 22 elk feedgrounds managed by the Wyoming Game and Fish Department [WGFD] (Dean et al. 2003) and were fed pelleted alfalfa on the federally (USFWS) managed National Elk Refuge. Annual elk surveys indicated nearly 80% of elk in this area wintered on feedgrounds, whereas 20% wintered on native range (Dean et al. 2004). Elk feedground densities were high and reached 300–2,500 elk/km² on the National Elk Refuge in severe winters (Smith 2005). Native winter range surrounded these areas, although elk distribution decreased with distance from feedgrounds (Smith and Robbins 1994; Barnett and Stohlgren 2001).

Our objective was to investigate whether factors influencing winter wolf kill sites differed from random points and specifically how distance to elk feedgrounds (i.e., kills on or near feedgrounds vs. on native winter range) affected kill site characteristics. We examined patterns of wolf kill site characteristics in northwestern Wyoming from 2000 to 2008 in a recovering and expanding wolf population. We first documented wolves on feedgrounds in 1999 (USFWS 2000), and wolves visited feedgrounds in every year of our study. We presumed due to extremely high elk density on feedgrounds that habitat characteristics at feedground kill sites were a function of elk distribution vs. kills on native winter range where habitat (e.g., canopy cover or terrain roughness) may have played a more important role in kill site location. We expected to find wolf kills in areas that facilitate their pursuit style of hunting (i.e., open, flat areas). Although wolf kill site habitat analysis is not new (e.g., Kunkel and Pletscher 2001; Husseman et al. 2003; Bergman et al. 2006; McPhee et al. 2012), the presence of elk feedgrounds and corresponding seasonally high elk densities in our study area were unusual, and wolf kill site patterns have not previously been reported in this area.

Our study area encompassed ∼2,300 km² and included the Gros Ventre river drainage, portions of Grand Teton National Park (i.e., east of the Teton Range), and U.S. Forest Service lands as well as private lands (Figure 1). Rugged mountains, ridges, deep drainages, and open sage Artemisia tridentata flats characterized the area. Elevations ranged from ∼1,800 to 3,600 m. Low elevations were dominated by sagebrush Artemisia spp. with lesser amounts of riparian cottonwood Populus angustifolia forests and willow Salix spp.; mid-elevations were generally forested and consisted mainly of lodgepole pine Pinus contorta, Douglas fir Pseodotsuga menziesii, and aspen Populus tremuloides, whereas Engelmann spruce Picea engelmannii and subalpine fir Abies lasiocarpa dominated the higher elevations. Primary prey species for wolves were elk, moose Alces alces, mule deer Odocoileus hemionus, and bison Bison bison.

Within our study area, there were three state-managed elk feedgrounds on U.S. Forest Service land in the Gros Ventre drainage, located ∼8 km apart: Alkali (0.42 km²), Patrol Cabin (0.17 km²), and Fish Creek (0.29 km²). Feedgrounds were maintained in an effort to minimize conflict between livestock and elk (e.g., eating hay and concern for spread of brucellosis; Dean et al. 2004) and to increase opportunity for harvest (Smith 2001). Sporadic supplemental elk feeding began in the Gros Ventre drainage in 1929, with a daily feeding regimen existing since the mid-1960s (WGFD 2007). Feeding occurred via a team of horses and a sleigh, and feedground location was chosen to ensure ease of access by horse-drawn sleigh. The Gros Ventre feedgrounds were characterized by open grassy areas with little to no shrubs or trees. Although Patrol Cabin and Fish Creek feedgrounds were located on flat and open areas, Alkali was in a gently sloping bowl and was bordered by patchy aspen stands. Elk quotas (i.e., number of elk managed on a feedground as set by the Wyoming Game and Fish Commission in 1985) were 650, 800, and 1,000 for Alkali, Patrol Cabin, and Fish Creek, respectively (WGFD 2007). However, actual elk numbers typically exceeded these quotas (e.g., Gros Ventre feedgrounds winter 2005–2006: 3,221 elk; winter 2006–2007: 2,921 elk; WGFD 2007).

In summer (May–September), we set traps (no. 7 McBride; Livestock Protection Co., Alpine, TX) along trails and roads frequented by wolves and also at livestock depredation sites (Mech 1974; Ream et al. 1991). In winter (December–March), we darted wolves from a helicopter by following the details in Kreeger et al. (2002) and Ballard et al. (1991). We immobilized wolves with Telazol delivered via a dart rifle (Palmer Cap-Chur Equipment, Inc., Powder Springs, GA) at the rate of 10 mg/kg during trapping and 17 mg/kg during helicopter capture (Ballard et al. 1991; Kreeger et al. 2002). We collared wolves with very high frequency mortality-sensing MOD500 collars or global positioning system (GPS) Gen-III TGW-3590 collars (Telonics, Inc., Mesa, AZ). We located radio-collared individuals weekly from fixed-wing aircraft and three to seven times per week from the ground by using roads and accessible trails. We collared and monitored 48 wolves in total from 2000 to 2008, and the number of individuals simultaneously radio-collared ranged from 3 wolves in 2000 to 11 wolves in 2006. From 2000 to 2005, we primarily monitored a single pack (Teton pack) that had at least one radio-collared individual in every year of our study. From 2006 to 2008, we monitored four packs of wolves.

Although wolves kill prey other than large ungulates, elk make up the majority (>90%) of winter wolf kills in this area (Smith et al. 2004; USFWS et al. 2006). We located ungulate carcasses by using ground and aerial telemetry of radio-collared wolves and backtracking (i.e., following wolf tracks) on skis or snowshoes from December to March each year. We investigated potential kills as soon as wolves left the immediate area. Fresh kills (<12 h) containing large amounts of meat were not investigated until the following day to minimize disturbance. By examining the site around the carcass, the carcass, and the hide, we distinguished predator-killed carcasses from scavenged carcasses (Wade and Bowns 1982; Acorn and Dorrance 1990).

We included only carcasses classified as known and probable (n = 295) as determined by evidence of predation in our analysis. We defined a kill site as the location of the carcass unless drag marks indicated the carcass had been moved. In this case, we considered the location of rumen or copious blood in the snow as the kill site. We recorded the location of kill sites with a handheld GPS unit (Garmin e-Trex, Garmin, Olathe, KS).

We examined variables related to wolf kill site locations by comparing characteristics at kill sites with random locations within available habitat (see below for description of available habitat). Kill sites in the vicinity of feedgrounds were likely influenced by their proximity to the feedgrounds because elk were often found to move on and off feedgrounds (Jimenez et al. 2006). We selected habitat characteristics known to affect wolf (Oakleaf et al. 2006) and elk (Boyce et al. 2003; Hebblewhite et al. 2005; Mao et al. 2005) distributions and predator–prey interactions (Kunkel and Pletscher 2000; Mech and Peterson 2003; Hebblewhite et al. 2005) (Table S1, Supplemental Material). These included canopy cover, elevation, slope, aspect (radians), terrain roughness, elk distribution, distance to feedground, distance to forest edge (edge; tree canopy >10%), and distance to packed surfaces (Table S1, Supplemental Material). Distance to packed surfaces included plowed roads and groomed snowmobile trails because these surfaces both facilitate ease of travel for wolves and their prey. Using ArcGIS (ArcMap 10.0; ESRI, Redlands, CA), we developed a basemap of the study area for habitat analyses and an elk distribution map and overlaid kill site locations (Figures 1 and S1, Supplemental Material). We buffered each kill site and random location by a 50-m radius to account for variation in the immediate surrounding habitat. At each site, we used spatial and attribute queries in ArcGIS to get a mean value for elk distribution and each habitat characteristic within the buffer. To create the elevation layer, we merged 10-m digital elevation maps from Grand Teton National Park and U.S. Geological Survey for Wyoming, resampled to the 30-m resolution, and derived slope and aspect data from this layer (Spatial Analyst extension; ESRI). Because aspect results in circular data, we transformed aspect into the linear variables northness (cosine aspect) and eastness (sin aspect) (Kneib et al. 2007). We derived a terrain roughness measure that measures the variability of slope, aspect, and elevation (Geomorphometry and Gradient Metrics Toolbox; Evans et al. 2014). Values ranged from 0 to 8.85, with low values indicating flat areas and higher values indicating steep, rugged terrain. For canopy cover, we used a National Land Cover Database Zone 21 Tree Canopy Layer, 30-m resolution (Homer et al. 2015). Within R version 3.1.2 statistical programming language (R Development Core Team 2016), we created a histogram of canopy cover values from 10,000 random points within the elk distribution layer to obtain an estimate of the distribution of canopy cover values (Bartnick et al. 2013). The distribution indicated a break at 15% canopy cover, suggesting a natural break for forested (tree canopy ≥15%) vs. nonforested (tree canopy <15%), and we reclassified canopy cover as a categorical variable where nonforested canopy was the reference category. We projected all layers to Universal Transverse Mercator-12 North American Datum of 1983 with 30-m resolution.

Elk data were collected cooperatively by the USFWS, Grand Teton National Park, and WGFD. Radio-collared elk (∼60/y) included both cows and bulls that were aerially located one to two times per month. Elk in the Jackson, Wyoming, area have shown high fidelity to winter and summer ranges (Boyce 1991). In addition, elk in this region are highly conditioned to supplemental feeding and return to the same feedgrounds year after year (Cole et al. 2015). We aggregated 1,429 locations (2000–2006) of radio-collared elk collected from December to March and created an elk distribution layer (individuals per square kilometer) (Spatial Analyst extension; ESRI).

The outcome of habitat selection analysis depends, in part, on the boundaries defined as available habitat (Garshelis 2000; Huston 2002). We defined available habitat as areas where wolves and elk occurred (Minta 1992). Although wolves do kill prey other than elk, elk make up the majority (>90%) of wolf kills in this region (Smith et al. 2004; Jimenez et al. 2006), and thus we felt justified in using elk locations to define available habitat. Within ArcGIS (ArcToolbox Data Management), we generated 100,000 random points within the elk distribution layer (where elk distribution was >0/km²) bounded by the extent of wolf home ranges merged over all packs and all years (Figure S2, Supplemental Material). Home ranges changed little from year to year and in this application, we felt a rough estimate of home ranges was appropriate. Following Northrup et al. (2013), we randomly selected (sample function in R base package) an increasing number of random locations (i.e., 295, 590, 885, etc.) and performed logistic regression until coefficients and model selection results converged at 2,360 random points (8 times the number of kill sites).

We conducted all statistical analyses using R version 3.1.2 (R Development Core Team 2016). Our global model contained all candidate models and included all combinations of the 10 explanatory variables, as well as biologically meaningful interactions (distance to feedground by edge, by slope, and by terrain roughness). We coded site as a binary response variable (1 = wolf kill, 0 = random point) in an unmatched case–control logistic regression model where the probability of a selected site being a kill site is assumed to be a function of site characteristics. We considered all possible combinations of predictor variables and with and without the hypothesized interactions. We examined variables for partial collinearity using Pearson's correlation. Correlations were <0.36 for all pairs of variables except slope and roughness (0.93) and elevation and roughness (0.55). Correlation of these variables would be expected given that roughness is derived using these variables.

We ranked models using Akaike's Information Criterion corrected for small sample size (AIC_c), and models with delta AIC_c (ΔAIC_c) larger than 2.0 were considered to have inferior relative fit (Burnham and Anderson 2002). We assessed the relationship between site status (kill or random) and the explanatory variables by using odds ratios. The odds ratio (calculated as e^β) can be interpreted as the increase or decrease in the odds of a point being a kill site given a 1-unit change (or a change in categories) in the habitat variable (Hosmer et al. 2013). Last, as a measure of how well the model discriminated between kills and random nonkill sites, we computed the area under the receiver operating curve adjusted for model optimization via five-fold cross-validation. Areas under the curve closer to 1 indicated greater discrimination.

We investigated 295 wolf kill sites classified as known or probable from January 2000 to February 2008. Thirty-four percent (n = 100) were on feedgrounds and 66% (n = 195) were on native winter range (Figure 1). The majority of kills came from two packs (Buffalo and Teton; 79%, n = 233; Table S2, Supplemental Material) with largely the same home range in different years (see USFWS et al. 2001, 2002, 2003, 2004, 2005, 2006; Jimenez et al. 2007, 2008, 2009 for home ranges). We also documented kills from five additional packs in our analysis from 2000 (n = 15), 2005 (n = 6), 2006 (n = 12), 2007 (n = 28), and 2008 (n = 1) (Table S2, Supplemental Material). Prey composition was 95% (n = 280) elk, 3% moose (n = 8), 1% coyote (n = 3), 0.7% mule deer (n = 2), and 0.3% each bison (n = 1) and beaver (n = 1).

Four models had ΔAIC_c ≤ 2, suggesting any model could potentially be the best model. All four models included canopy cover; distance to feedground; edge (i.e., distance to forest edge); elevation; elk distribution; terrain roughness; slope; and the interactions of distance to feedground to edge, to roughness, and to slope (Table 1). The best-fit (lowest AIC_c) model contained these 10 variables plus eastness. This model had twice the model weight (0.44) of the second-best model (ΔAIC_c = 1.37; weight = 0.22), which also included northness, although neither northness nor eastness was significant. Distance to packed surfaces (i.e., roads or groomed snowmobile trails) was included in the third-best model, although again was not a significant predictor variable. Given the model similarity, we model-averaged the coefficients (Table 2; Burnham and Anderson 2002). Variables negatively correlated with kill sites included canopy cover, distance to feedground, elevation, and roughness, whereas elk distribution was positively related to kill sites (Tables 1 and 2).

The odds of a location being a kill site were higher (odds ratio = 1.16; 95% CI = 1.11–1.21) at sites with greater elk distribution. Mean elk distribution at kill sites was 7.21/km² (SE = 0.54) compared with 1.47/km² (SE = 0.05) at random sites. The odds of a location in a forested area (canopy ≥15%) being a kill site were lower (odds ratio = 0.36; 95% CI = 0.23–0.58) compared with nonforested areas (Table 2). Only 8% (n = 30) of kill sites were in canopy cover classified as forested (tree canopy ≥15%). A site was less likely to be a kill site (odds ratio = 0.34; 95% CI = 0.19–0.61) in areas with greater terrain roughness. Slope, eastness, northness, distance to packed surfaces or forest edge, and all specified interactions did not differ significantly between kill sites and random locations (i.e., odds ratio CIs overlapping 1; Table 2). The cross-validation estimated area under the receiver operating curve for the top model was 0.92, indicating a good model fit.

Our results support previous research (Kunkel and Pletscher 2000; Hebblewhite and Pletscher 2002; Alexander et al. 2006; Bergman et al. 2006; McPhee et al. 2012) indicating a positive relationship between wolf kill sites and prey abundance (elk distribution; Figure S1, Supplemental Material), and a negative relationship with factors inhibiting coursing behavior (i.e., canopy cover and terrain roughness; Kunkel et al. 1999; Husseman et al. 2003; McPhee et al. 2012; Mech et al. 2015). Availability and vulnerability influence prey selection (e.g., Kunkel and Pletscher 2001; Smith et al. 2004; Bergman et al. 2006; Mattioli et al. 2011), whereas kill site locations are primarily driven by prey density (e.g., Hebblewhite et al. 2005) and prey catchability (e.g., Hopcraft et al. 2005; Keim et al. 2011). By selecting for areas of high prey abundance and density, predators increase their chance of encountering prey. However, because making a kill is dependent first upon encountering prey (Hebblewhite et al. 2005), kill site habitat for wolves is perhaps influenced more strongly by prey availability and prey habitat selection than by other elements of the landscape. Indeed, Mao et al. (2005) showed evidence for wolf habitat selection paralleling that of elk in Yellowstone National Park. The presence of feedgrounds in our study area led to a nonnatural distribution of elk on the landscape, creating high densities of elk in generally flat, open areas and de facto “hunting grounds” (sensu Kauffman et al. 2007) for wolves. In this artificial environment, it is likely not the habitat that was being selected for by elk or wolves (when hunting on feedgrounds), but simply the presence of abundant food.

The odds ratio for canopy cover (0.36) indicated a substantial relationship with wolf kills and nonforested areas. Ninety percent of kills in our study were in nonforested areas (<15% canopy cover). However, we also recognize the relative ease of finding kills in open canopy compared with forested areas that may have biased our data. Open canopy areas are conducive to extended chases and the coursing, cooperative hunting strategy of wolves running prey to find and pursue vulnerable individuals. Wolves may simply have higher hunting success in areas where there were more elk due to higher detection probability, higher encounter rates, or increased probability of finding vulnerable individuals. Likewise, within areas of high prey density, wolf hunting success was perhaps higher in areas that facilitate their pursuit style of hunting (i.e., less rough) and where prey vulnerability and catchability are increased (Hopcraft et al. 2005).

Secondary analysis (Tables S3 and S4, Supplemental Material) indicated native winter range sites were positively associated with higher terrain roughness and canopy cover and negatively associated (albeit weakly) with distance to forest edge and elevation compared with feedground sites. We caution, however, against any strong conclusions from the association with canopy cover given the small number of sites with >15% canopy cover (n = 26 native winter range; n = 4 feedground). We speculate the rougher terrain associated with winter range kill sites (e.g., steep, narrow gulleys or basins) is a function of the difficulty of making a successful kill with a more difficult opponent (i.e., bulls compared with calves). Concurrent wolf prey composition research found 21% of kills on native winter range were bull elk compared with only 5% of feedground kills (S.P.W. and M.D.J., unpublished data), although the proportion of (live) bull elk to cows or calves was much lower on feedgrounds compared with winter range (A. Courtemanch, Wyoming Game and Fish Department, personal communication). Terrain roughness or ruggedness is important for prey concealment (Riley and Dood 1984; Frair et al. 2005), prey escape terrain (Sappington et al. 2007), and predators as hiding cover (Kruuk 1986). In highly rugged areas, wolves may also have been able to reduce detection by prey or prey may have been vulnerable to being “trapped by terrain.” However, the general lack of terrain on the feedgrounds potentially overstates the strength of this interpretation.

The behavioral response of elk to the presence of wolves is further evidence that “free food” is a major driver of elk using the feedgrounds. Preliminary results indicated elk either remained on the feedground after wolves killed elk there, left the feedground but returned within 1–2 d, or left the feedground where wolves killed elk and moved to an adjacent feedground without wolves (S.P.W. and M.D.J., unpublished data). Interestingly, behavioral response differed among feedgrounds, with elk leaving both Alkali and Fish Creek feedgrounds when wolves made kills there to gather on Patrol Cabin feedground. Selection of habitat is an important antipredator defense (Bergerud et al. 1984; Kie 1999). Although the analysis from this research is not yet completed, we speculate that elk behavioral response was a predator defense strategy and potentially related to snow depth or hiding cover (i.e., vegetation) for wolves. At Patrol Cabin, visibility to detect approaching predators was better (i.e., little topography or vegetation) than at Alkali or Fish Creek, and the large groups of elk compacting snow likely facilitated movement for elk (and wolves).

The two primary areas where we found kills (i.e., sites in the north in Grand Teton National Park [native winter range] and the south area along the Gros Ventre river [native winter range and feedground sites]; Figure 1) are more easily accessible (e.g., on skis [north] or snowmobiles [south]) than the largely roadless area in between. Although we may have occasionally missed kills in this “middle” area, we are confident this was a very infrequent occurrence for several reasons. First, during weekly winter monitoring flights, wolves were never located on a kill in this area. Second, the home range of two packs (Buffalo and Teton packs in separate years) encompassed both the north and south areas. As shown by GPS collar data, wolves in these packs frequently moved through the middle area, and it was rare for wolves from these packs to take longer than 24 h to travel from north to south or vice versa through this area (M.D.J., unpublished data). Second, it is not uncommon for wolves to spend multiple days on a kill (Peterson and Ciucci 2003), depending on pack size. During our study, we observed wolves spending 1–4 d on kills when pack sizes ranged from 2 to 13 wolves. Therefore, we feel confident that kills were rarely made outside these major north and south areas. Last, the middle area was where home ranges from different packs overlapped (see USFWS et al. 2001, 2002, 2003, 2004, 2005, 2006; Jimenez et al. 2007, 2008, 2009), and wolves tend to spend less time in overlap areas due to the risk of interspecific strife (Peters and Mech 1975; Mech 1994).

Concurrently, our carcass detection methods (i.e., radiotelemetry and backtracking in snow) may have biased our estimates of prey composition toward larger prey due to longer handling time (i.e., wolves spend more time with a larger prey item) and greater evidence of a kill with a large ungulate. However, winter diet analysis and predation studies indicate large ungulates are the predominant prey of wolves in winter (Kunkel et al. 1999; Smith et al. 2004; Marucco et al. 2008). Thus, any other prey items (e.g., small mammals) likely made up a small proportion of wolves' diets (Trejo 2012) and were not likely to be a major driving factor in space use.

Consistent with other studies, our research illustrates wolves exploit areas of high prey distribution. Throughout Wyoming, elk feedgrounds and wolf home ranges overlap (see USFWS et al. 2001, 2002, 2003, 2004, 2005, 2006; Jimenez et al. 2007, 2008, 2009). With ∼80% of elk in northwestern Wyoming using feedgrounds in winter (Dean et al. 2004), wolves are expected to continue to frequent feedgrounds. This pattern of high prey density and high wolf activity is likely to hold in other areas where wolves recolonize and has important implications for both wildlife managers and livestock producers. Many of these high winter elk density areas become areas of high livestock density during spring and summer (M.D.J., unpublished data). High concentrations of elk raise concerns for spread of disease (e.g., brucellosis [Brucella abortus] or chronic wasting disease), and conditions on elk feedgrounds have proven ideal for the spread of disease (Dean et al. 2004; Smith 2005; Scurlock and Edwards 2010; Smith 2012). The issue of brucellosis transmission from wild to domestic ungulates is an ongoing concern (Maichak et al. 2009; Scurlock and Edwards 2010), and as chronic wasting disease moves closer to Wyoming's elk feedgrounds, the potential effects are unknown (Smith 2013; Wyoming Game and Fish Department 2016) but could be devastating to elk populations, especially those clustered at high densities (Smith 2013; Monello et al. 2014). Predators, including wolves, have the potential to reduce incidence of ungulate disease through direct predation or indirect effects such as behavioral changes (Barber-Meyer et al. 2007; Jolles and Ezenwa 2015). On the contrary, wolves pushing elk off of feedgrounds or between feedgrounds could lead to an increase in disease transmission. Understanding the spatial use of wolves and how that might relate to prey species may help predict areas with increased likelihood of wolf–prey interactions, areas where wolves may have a higher impact on prey populations, or areas of wolf–livestock conflict. In the future, managers will continue to face the issue of having high concentrations of ungulates, either wild or domestic, and the obvious attraction this has for wolves.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Table S1. Median, minimum (Min), and maximum (Max) values of continuous habitat variables and elk Cervus elaphus density used to assess habitat covariates associated with wolf Canis lupus kill site locations on and off elk feedgrounds vs. random (available) points from 2000 to 2008.

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S1 (17 KB DOCX).

Table S2. Table shows year, pack, and prey composition for winter (December–March) wolf Canis lupus kills (n = 295) from 2000 to 2008 in the southern Yellowstone ecosystem, Wyoming, USA.

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S2 (15 KB DOCX).

Table S3. Summary of the candidate models (ΔAIC_c ≤ 2) quantifying wolf Canis lupus kill sites on elk Cervus elaphus feedgrounds (n = 100) vs. kill sites on native winter range (n = 195) in the southern Yellowstone ecosystem, Wyoming, USA, from 2000 to 2008. Column abbreviations are degrees of freedom (df), Akaike's Information Criterion corrected for small sample size (AIC_c), delta AIC_c (ΔAIC_c), and AIC_c weights (w_i).

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S3 (18 KB DOCX).

Table S4. Model-averaged coefficient (β) and odds ratio (OR) and corresponding 95% confidence intervals for the top three (≤2 ΔAIC_c) models comparing wolf Canis lupus kill sites on elk Cervus elaphus feedgrounds (n = 100) vs. kill sites on native winter range (n = 195) in the southern Yellowstone ecosystem, Wyoming, USA, from 2000 to 2008. ΔAIC_c = delta Akaike's Information Criterion corrected for small sample size. We used an indicator variable for forested (>15% canopy cover) sites. We coded sites as a binary response variable (1 = native winter range kill site, 0 = feedground kill site). Bold indicates confidence interval does not contain 0 (coefficient) or 1 (odds ratio). Asterisk (*) indicates unrounded confidence intervals did not overlap 1.

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S4 (18 KB DOCX).

Figure S1. Map showing an index of elk Cervus elaphus density with wolf Canis lupus kill sites in the southern Yellowstone ecosystem, Wyoming, USA, from 2000 to 2008.

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S5 (410 KB PDF).

Figure S2. Map depicting annual wolf Canis lupus home ranges and kill sites (black dots) in the southern Yellowstone ecosystem, Wyoming, USA, from 2000 to 2008. The polygon represents the approximate home ranges of three packs. Home range data were collected for monitoring purposes and came from relatively few locations (≤20) in some cases and changed minimally between years during this study (see all wolf home ranges in USFWS et al. 2001, 2002, 2003, 2004, 2005, 2006; Jimenez et al. 2007, 2008, 2009).

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S6 (442 KB PDF).

Reference S1. Mech LD. 1966. The wolves of Isle Royale. Washington, D.C.: U.S. Government Printing Office, U.S. National Park Service Fauna Series No. 7.

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S7 (65,508 KB PDF); also available at: http://npshistory.com/series/fauna/7.pdf (65,508 KB PDF).

Reference S2. Smith BL. 2005. Disease and winter feeding of elk and bison: a review and recommendations pertinent to the Jackson bison and elk management plan and environmental impact statement. Bozeman, Montana: Greater Yellowstone Coalition.

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S8 (442 KB PDF); also available at http://irmaservices.nps.gov/datastore/v4/rest/DownloadFile/151634?accessType (360 KB PDF).

Reference S3. [USFWS] U.S. Fish and Wildlife Service, Nez Perce Tribe, National Park Service, and USDA Wildlife Services, Meier T, editor. 2004. Rocky Mountain wolf recovery 2003 annual report. Helena, Montana: U.S. Fish and Wildlife Service

Found at DOI: http://dx.doi.org/10.3996//032016-JFWM-024.S9 (322 KB PDF).

We acknowledge the assistance of Sarah Dewey, John Stephenson, and Steve Cain of Grand Teton National Park. We also thank personnel of the Bridger-Teton National Forest and the WGFD. Dylan Taylor, Karen Colclough, Lydia Dixon, Rebecca Hansen, Ronnie Hegemann, Lindsay Reynolds, Leah Samberg, Miguel Licona, John Stephenson, and Pat Leslie assisted in data collection. Special thanks to pilots Gary Lust, Dave Stinson, and Bob Hawkins at Sky Aviation; Mark Duffy at Central Copters; and Lisa Robertson for hours of flight time. We also thank two anonymous reviewers and the Associate Editor for editing expertise. Funding was provided by USFWS and Craighead Beringia South. This study was in partial fulfillment of a Master's degree for S.P.W. at Prescott College.

Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.