Received: 31 October 2023 Revised: 11 March 2024 Accepted: 19 March 2024 DOI: 10.1002/ecs2.4929 ARTICLE Methods, Tools, and Technologies How to find a wolverine: Factors affecting detection at wolverine (Gulo gulo) bait stations in western Canada Andrea Kortello 1 | Doris Hausleitner 2,3 | Garth Mowat 4,5 | Mirjam Barrueto 6 | Nicole Heim 7 | Lisa Larson 8 | Michael Lucid 9 | Anthony Clevenger 10 1 Grylloblatta Ecological Consulting, Nelson, British Columbia, Canada 2 Seepanee Ecological Consulting, Nelson, British Columbia, Canada 3 Selkirk College, Castlegar, British Columbia, Canada 4 Ministry of Forests, Nelson, British Columbia, Canada 5 Department of Earth, Environmental and Geographic Sciences, University of British Columbia, Kelowna, British Columbia, Canada 6 Department of Biological Sciences, University of Calgary, Calgary, Alberta, Canada 7 Nikki Heim Consulting, Canmore, Alberta, Canada 8 Mount Revelstoke and Glacier National Parks, Parks Canada, Revelstoke, British Columbia, Canada 9 Selkirk Wildlife Science LLC, Sandpoint, Idaho, USA 10 Western Transportation Institute, Montana State University, Bozeman, Montana, USA Correspondence Andrea Kortello Email: kortello@yahoo.com Funding information U.S. Fish and Wildlife Service; British Columbia Ministry of Forests; Columbia Basin Trust; Fish and Wildlife Compensation Program; Parks Canada; Western Transportation InstituteMontana State University; Woodcock Foundation; Wilburforce Foundation; Liz Claiborne Art Ortenberg Foundation; Volgenau Foundation; Cross Foundation; Great Northern Landscape Conservation Cooperative; Alberta Conservation Association; Mountain Equipment Co-op; McLean Foundation; Patagonia; Cameron Plewes; Alberta Sport Parks Recreation and Wildlife Foundation; National Geographic Society; Disney Conservation Fund; Norcross Foundation; Bow Valley Naturalists; Yellowstone to Yukon Conservation Initiative; Lake O’Hara Lodge; Alpine Club of Canada; The Wolverine Foundation; University of Abstract High individual detection success enables precise estimates of density and the ability to monitor trends in abundance for wolverine and other low-density species, information that is critical for the implementation and assessment of conservation measures. We evaluated a dataset that included six different wolverine capture–recapture studies over a large gradient in wolverine (Gulo gulo) density to provide recommendations for increasing detection. We examined factors related to bait station components, habitat, and seasonal timing. Accounting for variation in wolverine density and trap duration, our results suggest that bait stations setups having a run pole, frame, and camera to photograph unique ventral color patterns, in addition to hair snag devices, identify more individual wolverine than those without. The presence of snow is a habitat feature that also increases individual detection. Female detection rates were lower than male detection rates at the onset of the reproductive denning season in late February and early March compared with January and early February. We found latency to detection was independent of wolverine density, but greater in areas with human influence. Relatively high rates of genotyping success (55%) were predicted by even a single guard hair left at bait stations, while underfur required ~15 hairs for similar success. Longer This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2024 The Author(s). Ecosphere published by Wiley Periodicals LLC on behalf of The Ecological Society of America. Ecosphere. 2024;15:e4929. https://doi.org/10.1002/ecs2.4929 https://onlinelibrary.wiley.com/r/ecs2 1 of 22 Calgary; Natural Sciences and Engineering Research Council of Canada (NSERC); British Columbia Habitat Conservation Trust Foundation; Forest Enhancement Society of British Columbia; Helicat Canada; K3 Cat Ski Handling Editor: Jesse S. Lewis KORTELLO ET AL. sampling intervals reduced genotyping success in spring, more so for underfur than guard hair. Hair samples acquired from barbwire were of higher quality than those from either alligator clips or gun brushes. To improve individual detection for wolverine inventory and monitoring, we recommend deploying run pole setups in areas with low human disturbance that will retain snow into late spring. Extending the winter trapping effort into April and May could increase the chances of detecting denning females. Latency to detection suggests that traps should be active for more than a month, especially in human-influenced areas, but genotyping success suggests that traps should also be cleared of hair samples at smaller intervals of a month or less, during late winter/spring. KEYWORDS bait station, capture–recapture, detection, genotyping success, Gulo gulo, inventory, monitoring, noninvasive sampling, study design INTRODUCTION For species at risk, knowledge of population size and trend is a priority for successful management and conservation. Imprecise population estimates can hinder the ability to detect trends over time and preclude effective conservation actions. However, species at risk are distinguished by low or declining density, which can lead to difficulties with estimating remaining populations (Ellison & Agrawal, 2005; Green et al., 2020; Kéry et al., 2011; Schaub et al., 2007). Wolverine (Gulo gulo) are one such species. They are characterized by low population density and inhabit remote and often rugged terrain that makes assessment of population status challenging (Fisher et al., 2022). Population dynamics in North America are largely inferred from harvest records, and reliable knowledge of population trends is almost completely lacking (Fisher et al., 2022; Slough, 2007; see Barrueto et al., 2022 for the exception). The advent of noninvasive genetic sampling has greatly facilitated the ability to inventory rare species, using DNA from hair, scat, or other material (Schwartz et al., 2007; Waits & Paetkau, 2005). Early attempts to collect wolverine hair for noninvasive genetic sampling with box traps, barbwire corrals, or rub pads detected few wolverine (Mowat et al., 2003). Later bait station designs developed by Magoun et al. (2011) and Mulders et al. (2007) have been deployed with considerable success for estimating wolverine population size and density (Barrueto et al., 2020; Mowat et al., 2020; Royle et al., 2011), as well as distribution and connectivity (Fisher et al., 2013; Heim et al., 2017; Kortello et al., 2019; Sawaya et al., 2019). Accurate population estimates can be generated using spatially explicit capture–recapture methods (Conner et al., 2016; Pozzanghera, 2015; Royle et al., 2011) but this necessitates individual identification of animals. While modern statistical techniques can account for imperfect detection at bait stations (Efford, 2004; MacKenzie et al., 2017), the power to assess significant population change is facilitated by high detection probabilities (Ellis et al., 2014; Green et al., 2020; Keiter et al., 2017; Morin et al., 2022). Detectability is often related to density (Kéry, 2002; Royle & Nichols, 2003; Walsh et al., 2018) and for low-density species like wolverine, low detection rates add uncertainty to population estimates (Keiter et al., 2017). Consequently, understanding the factors affecting detection is crucial for developing effective sampling designs, but existing information for wolverine has been piecemeal. Robinson et al. (2017) examined factors affecting multispecies bait station detection in northern Idaho and southeastern British Columbia (BC). This included latency to detection (LTD), seasonal timing, and the influence of habitat on detection. However, low wolverine numbers in their study area and sparse detections limited their conclusions for this species. Bait stations for wolverine individual genetic identification can be set up in various ways, components include a hair snag device, a bait or lure attractant, and possibly a camera. As wolverine may visit a bait station but not leave a hair sample, Fisher and Bradbury (2014) determined that wildlife cameras increased the effectiveness of bait stations for species identification over genetic sampling alone but did not assess individual identification. Mowat et al. (2020) reported a weak effect of bait type on detection. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2 of 22 Wolverine habitat selection likely affects individual detection success in a hierarchical fashion. While density reflects the location of home ranges within an animal’s distributional range, individual detection may also be affected by behavior and habitat use at finer scales, such as foraging or denning habitats. For systematic sampling in remote terrain, safety and logistics often limit bait station placement choices, and habitat features that constrain access (i.e., avalanche terrain, roads) may affect detection (Fisher & Bradbury, 2014). For example, Stewart et al. (2016) showed that wolverine were more inclined to interact with bait stations in protected areas with low anthropogenic disturbance. As sparsely distributed scavengers that locate carrion, and hence bait stations, using olfactory cues (Green et al., 2012), landscape attributes that aid scent dispersal and facilitate travel or encounter rates may affect detection. Scent dispersal is a diffusive process affected by climate, particularly temperature and wind, and topography (Conti et al., 2020). Cool temperatures retain scent while surface features such as vegetation that provide wind resistance impact odor plumes, affecting scavengers’ ability to locate carrion (Ruzicka & Conover, 2012). Wolverine density, distribution, dispersal, and reproduction have all been connected to snow (Copeland et al., 2010; McKelvey et al., 2011; Mowat et al., 2020; Schwartz et al., 2009), and olfactory processes are also affected by snow. Snow crystals readily absorb and retain volatile organic compounds, which are released by carcass decomposition (Herbert et al., 2006; Verheggen et al., 2017). Wolverine have also been shown to select deeper, denser snow for travel (Glass et al., 2021) and prefer travel routes along drainage bottoms (Heinemeyer et al., 2019). Seasonal timing affects the detection of other carnivores (Lamb et al., 2016) and likely wolverine as well. Bait stations are typically deployed in winter, when alternative food sources are fewer, hibernating bears pose less risk to researchers, and cold temperatures aid DNA preservation (Mulders et al., 2007). Fisher et al. (2013) noted higher wolverine occurrence at bait stations in late winter, but since wolverine readily revisit traps once they have been captured (Barrueto et al., 2020; Mowat et al., 2020; Royle et al., 2011), the link between this seasonal effect and the detection of new individuals is unclear. Wolverine maternal denning occurs from approximately February to April (Inman, Magoun, et al., 2012; Magoun & Copeland, 1998), and females with young have restricted movements during this period, especially in the first few weeks after parturition (Aronsson et al., 2023; Landa et al., 1998). This likely decreases female encounters with bait stations, and detections, during this period. 3 of 22 Understanding how long animals take to arrive at bait stations is useful for gauging species response to sites and determining the minimum length of time for deployment. As scavengers, wolverine have numerous avian and mammalian competitors that may decrease the potential energetic gain per carcass. Hence, rapid arrival times would be advantageous. Robinson et al. (2017) reported a median time to detection of 29 days but did not assess factors affecting latency. Individual identification also depends on the collection and genotyping of hair samples, although run pole setups may circumvent this requirement by identifying individuals through photographs of distinct ventral patterns (Magoun et al., 2011; Royle et al., 2011). Genetic analysis protocols and error checking produce consistent results, but poor sample quality can preclude successful genotyping (Kendall et al., 2009; Paetkau, 2003). Robinson et al. (2017) found that seasonally colder temperatures during winter retained adequate DNA quality over long trap deployments for multiple species, but sun exposure and frequent melt–freeze cycles in late winter and spring may reduce DNA quality (Stetz et al., 2014). The effect of hair type, numbers, and environmental exposure on genotyping success has been examined for black (Ursus americanus) and grizzly bears (Ursus arctos) and American marten (Martes americana; Lamb et al., 2016; Mowat & Paetkau, 2002; Stetz et al., 2014; Tredick et al., 2007), but not for wolverine. To aid continued research and monitoring efforts for wolverine, and make effective use of available resources, we used a modeling approach to evaluate the factors affecting individual detection for a network of bait stations across western Canadian mountain ranges, the first time this has been done on a large scale for wolverine. We examined the effects of trap type, habitat, sampling schedule, and sample quality on individual wolverine detections. Six different projects spanning seven winter and spring seasons across a wide range of wolverine densities provided our dataset. Our objective was to assess the most effective methods for noninvasive hair trapping of wolverine and provide recommendations for maximizing detection in future work. METHODS Study area The ~74,000 km2 study area included the Columbia and Rocky Mountain ranges and foothills of southeast BC and southwestern Alberta (Figure 1). All of the study area was within the wolverines’ contemporary range in 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE KORTELLO ET AL. F I G U R E 1 Wolverine winter sampling areas in southeast British Columbia and southwest Alberta, Canada, showing sampling area boundaries, bait station locations, and the number of different individuals detected. BYKNP, Banff, Yoho, and Kootenay National Parks and nearby crown lands; MRGNP, Mount Revelstoke and Glacier National Parks and nearby crown lands. western Canada, as defined by recent harvest records (Lofroth & Ott, 2007; Webb et al., 2013) but wolverine density varied considerably, generally increasing from south to north and east to west (Mowat et al., 2020). Elevations ranged from 400 to >3000 m, with large regional variation in seasonal precipitation. At lower elevations, forests consisted of western red cedar (Thuja plicata), western hemlock (Tsuga heterophylla), Douglas fir (Pseudotsuga menziesii), ponderosa pine (Pinus ponderosa), lodgepole pine (Pinus contorta), trembling aspen (Populus tremuloides), and western larch (Larix occidentalis). At higher elevations, Engelmann spruce (Picea engelmannii) and subalpine fir (Abies lasiocarpa) forests give way to treeless alpine meadows, rock, and ice (MacKillop & Ehman, 2016). Major transportation corridors crossed the study area latitudinally, including the TransCanada highway (Hwy 1) in the northern portion of the study area and the Crowsnest highway (Hwy 3) in the south. Smaller highways and human settlements were located in low elevation valleys. Extensive forest harvesting has occurred throughout the area and mining was widespread historically, less so currently. Both industries left a legacy of road networks extending to subalpine areas in many watersheds. Winter recreation (snowmobile use, ski resorts, helicopter or snowcat-access skiing, ski lodges, backcountry skiing) was prevalent in the area. Regulated trapping of wolverine occurred on crown lands in Alberta and BC during the years sampled. Parks and protected areas made up ~25% of the land base. Sampling We used noninvasive techniques in winter and spring to collect hair samples from wolverines in six sampling areas (Figure 1, Table 1) between 2011 and 2018. Sampling occurred on public lands in habitat consistent with wolverine use. Each study area was partitioned 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 4 of 22 5 of 22 T A B L E 1 Wolverine sampling effort and detection success across six sampling areas in southeastern British Columbia and southwestern Alberta, Canada, between 2011 and 2018. Sampling area Year Area sampled (km2) No. sites sampled No. sessions 8400 44 3 Setup Tree Bait Beaver Camera present Yes Total no. individuals (F/M) BYKNP 2011 22 (9/13) Kananaskis 2012 4500 34 4 Tree Beaver Yes 4 (2/2) MRGNP 2011, 2012, 2014, 2015 3900 33 3 Tree Beaver or domestic animal Yes 26 (15/11) South Rocky Mountains 2014–2016 21,400 163 3 Tree Beaver Yes 23 (15/8) South Columbia Mountains 2012–2016 25,000 228 2 Tree Ungulate or beaver Rarely 40 (24/16) North Columbia Mountains 2018 9900 25 2 Runpole Beaver Yes 25 (17/8) Note: Year was the year at the end of the sampling winter. Within a sampling area, different years sampled different subranges or areas. Sessions were the number of intervals that the bait stations were checked during the deployment, and detections were the total number of times wolverine was detected at different sites. We only include data for the first detection of each individual at a site, but some individuals were detected at multiple sites. Abbreviations: BYKNP, Banff, Yoho, and Kootenay National Parks and nearby crown lands; F, female; M, male; MRGNP, Mount Revelstoke and Glacier National Parks and nearby crown lands. into 10 × 10 km or 12 × 12 km cells that approximated the minimum size of a female home range. One or, rarely, two bait sites were placed in each cell, and bait sites were checked between two and four times over the course of the winter at approximately three- to four-week intervals. Hair traps prior to 2018 followed Fisher et al. (2013), consisting of a tree wrapped in barbwire, with bait nailed above the wire ~2 m from the ground or snow surface. We refer to these as tree traps because trees were the primary structural component (Figure 2). Most tree setups included a camera attached to a second nearby tree to photograph animals that visited the site, except in the south Columbia mountains, where few cameras were deployed. In 2018, we used a different configuration, with a 1.2 m horizontal piece of lumber attached to a tree ≥1 m from the snow or ground (Magoun et al., 2011). A frame was mounted to the distal end of the lumber and bait was positioned on a cable strung between that tree and another tree, ~4 m away, such that the wolverine must climb onto the frame to access the bait, exposing its ventral side. A camera was mounted to photograph distinct individual markings on the wolverine’s throat and chest. Alligator clips tethered to the frame collected hair samples when triggered and the run pole support tree was wrapped in barbwire. For more details refer to Magoun et al. (2011). We refer to this as a run pole setup. For bait we used a beaver carcass (Castor canadensis), ungulate head or quarter. In Mount Revelstoke and Glacier National Parks (MRGNP), meat from domestic animals, a pig head or beef roast, was used during 2014, and beaver otherwise. We used a commercial scent-based long-distance lure (Caven’s Gusto Long Distance Call, Minnesota Trapline Products, Pennock, MI, USA), smeared on a rag and suspended from a tree branch at all sites. During each check the barbwire was examined for hairs and the bait replenished if necessary. Hair was collected and stored in paper envelopes in a dry environment. For our evaluation of genotyping success from wolverine hair samples, we used additional hair sample data from a concurrent study in the United States that was adjacent to the study area for the detection analysis. This study used beaver bait and tree setups with 30 caliber bronze gun brushes arranged in two concentric circles, at 30 and 45 cm below the bait, instead of barbwire, to remove hair (Robinson et al., 2017). Hair samples were sent to Wildlife Genetics International (WGI) in Nelson, BC, for microsatellite genotyping. Only samples that had ≥1 guard hair with a root or ≥5 underfur were selected for analysis, and we used up to 10 guard hairs or approximately 30 underfur in an extraction when available. DNA was extracted using QIAGEN DNeasy Tissue kits, following the manufacturer’s instructions (Qiagen, Toronto, ON). Species identification was based on a sequence-based analysis of a segment of the mitochondrial 16S rRNA gene (Johnson & O’Brien, 1997). For samples that yielded wolverine DNA, WGI utilized multilocus genotyping, consisting of a ZFX/ZFY sex marker, and seven additional microsatellite markers for individual identification. Individual identifications were error-checked, following established rules that yield very low error 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE KORTELLO ET AL. F I G U R E 2 (a) Tree bait station setup for noninvasive hair snagging of wolverine. Cameras were present in Banff, Yoho, and Kootenay National Parks and nearby crown lands, Mount Revelstoke and Glacier National Parks and nearby crown lands, Kananaskis and almost all south Rockies sites, but were used only occasionally in south Columbia. (b) Runpole bait station setup used in north Columbia. rates (Kendall et al., 2009; Paetkau, 2003). The studies in the national parks were analyzed at the Rocky Mountain Research Station laboratory in Missoula, Montana, using nearly identical methods. One sample from each individual was reanalyzed at WGI, to verify individual identities were comparable between the labs (Mowat et al., 2020). To preclude learned behaviors affecting the independence of our observations for our analysis of wolverine detection at bait stations, we only used bait stations deployed in areas where wolverine were thought to be naïve; that is, we included only data from the first of multiple years of sampling and only included the first detection of an individual at a particular bait station within years. The sample unit for the genotyping analysis was the hair sample collected from the snagging device and this potentially included multiple detections of the same individual. Detection variables We used variables related to timing, habitat, and methodology. To examine the broad-scale effect of the timing of bait station deployment on detection, we assigned a value for the week of year that a bait station was first deployed (FIRSTWEEK), sequentially, with the last week of November as week 1, and increasing with subsequent 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 6 of 22 weeks to 31 March, week 18, and included this in our models. Our habitat variables considered vegetation, snow, and anthropogenic factors. In montane environments, wolverine occupy upper elevation forests and females den near treeline (Inman, Packila, et al., 2012; Krebs et al., 2007; Magoun & Copeland, 1998). We predicted that bait stations situated near open, alpine terrain would have increased scent dispersal and higher detections than those in closed forest habitat. We separated mapped vegetation communities into two broad zones: (1) alpine, including rock, alpine tundra and grassland, alpine parkland and woodland ecosystems; and (2) forest, grouping Engelmann spruce–subalpine fir typical of subalpine areas and lower elevation forests of pine, Douglas fir, cedar, and hemlock. We calculated the percent area of alpine vegetation types in a 10-km radius from the sampling location to index the influence of alpine versus forest on detection (ALPINE). To assess the influence of snow cover and depth on wolverine detection, we measured late winter snow depth (SNOW) at each site when it was last checked in April or early May. This period is the onset of spring melt, when snow depths were highly variable among sites. We predicted that the amount of snow at each bait station would positively influence detection. Landscape topographic position influences movement (Heinemeyer et al., 2019) and scent dispersal (Conti et al., 2020), conceivably affecting detection. From a sampling standpoint, topographic position can also affect the difficulty in deploying and accessing a site. For our categorical topography variable, we classified trap locations as valley bottom, ridgetop, mid-slope, or on a pass, if the site was located at or near the apex of major watersheds or terrain breaks (TOPO: valley, ridge, mid-slope, pass). Passes concentrate travel routes between watersheds and funnel air movement. We expected higher detection on passes than other topographic features. Stewart et al. (2016) found that wolverine behaviors at bait stations varied with human footprint on the landscape. Wolverine in areas with high human use were less inclined to climb the tree to the bait. We used Human Influence Index (HUMANII) mapping as a generalized measure of anthropogenic disturbance (http://sedac.ciesin. columbia.edu/data/set/wildareas-v2-human-influenceindex-geographic) to test this hypothesis in our area. We created a measure of human disturbance by calculating the mean HUMANII in a 10-km radius window around each site. In steep montane habitats, safety and access limit the location of traps on the landscape. All bait station locations required avalanche terrain avoidance. Traps 7 of 22 deployed using trucks or snowmachines were limited by proximity to plowed or unplowed roads, respectively, while traps deployed using helicopters were limited by the availability of safe landing sites (openings in forest). Ski access was limited by distance (typically <6 km from roads). However, Heinemeyer et al. (2019) found that wolverine avoided areas with human recreational disturbance, both motorized and nonmotorized, and Scrafford et al. (2018) found that wolverine avoided roads. We compared detection among bait stations with ski, truck, snowmachine and helicopter access (ACCESS: ski, truck, snowmachine, heli). We predicted that helicopters would access terrain with less human disturbance, and this would result in more wolverine detections. We recognize that access is related to the HUMANII index described in the previous paragraph, at a different scale. For variables related to methodology, we considered both bait type and trap configuration. Bait is a critical component of hair trap stations. Although wolverine are opportunistic scavengers, fat content, odor, and familiarity might all potentially affect the attractiveness of bait and hence detection. We compared detection when using beaver, ungulate heads, domestic meats (beef roasts and pig heads) and ungulate quarters, in a categorical variable BAIT. The two different categories of trap configurations for our bait stations, run pole and tree, comprised our SETUP variable. In tree setups, cameras (CAMERA) may or may not be used, but their use provides additional information on wolverine presence (Fisher & Bradbury, 2014) and could conceivably affect the search effort for cryptic hair samples. However, the presence of a camera is an integral part of run pole setups, and in this context can provide individual identification independent of hair samples. Since wolverine may visit a bait station without leaving hair, we predicted higher detection for run pole bait stations versus the reference tree setups. Analytic approach We approached the question of wolverine detection from several different angles. First, we modeled methodological and spatial factors affecting individual detection at bait stations. Next, we examined fine-scale temporal variation in detection over the onset of reproductive denning in females. Then we evaluated the same detection factors as predictors of LTD because we felt that it might provide insight into the behaviors driving detection. Finally, we assessed the impact of sample quality (type and number of hairs, length of exposure to environmental conditions) on genotyping success and compared sample quality between different hair-snagging devices. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE Individual detection Although SECR models allow the estimation of covariates for detection probability concurrent with density, models become intractable with large numbers of covariates. Instead, we modeled variables affecting wolverine detection using generalized linear models (GLM; Poisson errors, log link). Our response variable was the number of unique individuals (0–4) identified at each bait station, scaled by the offset terms, DENSITY and DURATION. These terms account for unbalanced sampling between trap sites with respect to both underlying densities and the length of time that traps were deployed, as both are known to functionally influence detection. We used the estimates of wolverine density (DENSITY) presented in Mowat et al. (2020), taking the mean density in a 10-km radius window. Since the onset, duration and number of sessions among sampling areas was highly asynchronous, we measured the total number of days that a station was operational (DURATION) and used this as a second offset term in all detection models. We used residual plots and checked for over- or under-dispersion to determine the most appropriate type of error distribution. We ran nested models to evaluate sex-based differences, using the number of unique males or unique females. All analyses were conducted in software package R version 4.1.1 (R Core Team, 2021). We checked for potential nonlinear relationships between detection and its covariates by plotting univariate response curves. Where a nonlinear relationship was suggested, we transformed the original data and selected the form of the covariate having the lowest corrected Akaike information criterion for small sample sizes (AICc) in subsequent analyses (Burnham & Anderson, 2002). The variables HUMANII, ALPINE, and SNOW were log10-transformed to normalize their relationship with the number of individuals detected. A small constant (0.0001) was added to each to facilitate handling of zeros. We assessed collinearity and multicollinearity using Spearman’s correlation coefficients and variance inflation factors (VIF) using cutoff values of ρ > 0.6 or VIF > 3, respectively (Glasser & Winter, 1961; Zuur et al., 2010). Comparisons between bivariate and continuous covariates used a Pearson’s point-biserial correlation, to avoid tied values. For pairs of correlated covariates, we retained the one with the most direct causal link to the response. When this was not clear, we used the covariate with the lowest AIC value. We calculated AICcs and log likelihoods for univariate models of all variables. Our global model included ALPINE, SNOW, TOPO, CAMERA, ACCESS, BAIT, FIRSTWEEK, and SETUP, with offset terms DENSITY KORTELLO ET AL. and DURATION. We constructed candidate models by evaluating all subsets of the global model using the R package MuMIn (Barton, 2016; Doherty et al., 2012). We used AICc weights and log-likelihoods to compare the relative support for each model, and regarded all models within two AICc points of the top-ranked model as having empirical support, while possibly containing uninformative parameters (Arnold, 2010; Burnham & Anderson, 2002). We calculated parameter estimates for the top-ranked model and plotted effects in the original scale, with non-focal variables held at typical values using the R package Effects (Fox, 2003; Fox & Weisberg, 2018; Fox & Weisberg, 2019). We report beta coefficients (β) and SEs for variables of interest. Since variables with weak effects may still help to improve sample methods, we noted effect sizes for variables in other models within two AIC points of the top-ranked model. Temporal comparison To assess fine-scale and sex-specific temporal effects of denning on individual identification at bait sites that were deployed and monitored in asynchronous time intervals, we used an index of relative detection (RDI; Lamb et al., 2016). We calculated an RDI for each day and site sampled. This was accomplished by taking the number of new individuals detected at a site and dividing by the length of the session where the bait site was operational to obtain a daily detection probability. Individuals previously detected at the site were not counted again. We accounted for variation in wolverine density across the sampled landscape by further dividing this number by the sex-specific density at that location (from Mowat et al., 2020). For example, two new females identified over a 30-day sampling session with a female density of 0.7/1000 km2 would derive a value of 0.095 for each of the 30 days that the trap was operational in that session. We then averaged these daily detections across all sites (Lamb et al., 2016) and calculated SEs for each day. To increase temporal resolution, we removed any sessions that were >50 trap nights. Since some bait stations were deployed in November and others remained operational until the end of May, we truncated the analysis period, selecting date intervals to maximize the number of sampling sites that were operating simultaneously. We compared male and female relative detection over two time periods, pre-denning, from 9 January to 9 February; and early denning, from 10 February to 6 April. To control for nontemporal factors that may affect detection, we only included data from sites that were monitored continuously through each time interval, so that the variation in the relative 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 8 of 22 9 of 22 detection of sites was a function of time only, not the addition or removal of more or less successful sites. obtained from six bait stations outside the study area provided the gun brush comparison (Robinson et al., 2017). Latency to detection RESULTS Using a similar approach to the detection models, we used the subset of bait stations that included a wildlife camera to calculate a LTD, the time (in days) from when the bait station was first established to the first detection of a wolverine. This measure was the response variable in a Gamma-distributed (log link) GLM with a similar set of covariates as our detection analysis. For latency, including DURATION was nonsensical, we used the correlated FIRSTWEEK to reflect the timing of sampling onset. We included DENSITY as a covariate rather than as an offset term, because, unlike detections, we did not expect latency was a proportion of density. We added a new variable, SEX (male, female, unknown). We predicted that the same variables associated with detection would also influence LTD times. We installed 527 sample sites and detected between zero and four individuals at each site for a total of 151 individual wolverine, 90 females and 61 males, across 6 winters. We examined 10 variables potentially related to individual detection and LTD. Some variables had collinear relationships. The variables ALPINE and HUMANII were correlated with DENSITY (0.81, −0.61) and unstable in models containing DENSITY. As density was a key component of the individual detection model framework, we did not consider either ALPINE or HUMANII further in the detection analysis. ACCESS was only weakly correlated with HUMANII (−0.4). For the latency analysis we included DENSITY, HUMANII, and ALPINE as variables, but not in the same model. DURATION was highly correlated with FIRSTWEEK (0.97), we used DURATION in all detection models for its strong causal link. The categorical variable CAMERA was confounded with BAIT and SETUP, in that the majority of tree setups that lacked cameras also had bait other than beaver, while all run pole setups necessarily had cameras. As the effect of CAMERA was not significantly different from zero in models that also contained SETUP and BAIT, whereas coefficients on the other two variables changed little, we considered that the effect of CAMERA on individual detection, independent of SETUP and BAIT, was negligible and dropped this variable. We simplified some of our categorical variables after running univariate models. For ACCESS, the only significant contrast to the reference level snowmachine was heli; grouping ski and truck with snowmachine resulted in a heli versus ground comparison for that variable. For BAIT, beaver performed better than other levels, and significantly better than ungulate heads. We grouped ungulate head, ungulate quarter, and domestic meat into a reference level, resulting in a beaver versus other comparison for BAIT. The variable TOPO did not generate any significant contrasts or logically justifiable groupings between categories. Genotyping success We evaluated genotyping success rates for wolverine guard hair and underfur hair samples as a function of the number of hairs in the sample (HAIRCOUNT), the length of the interval between trap checks (DAYS_OUT), and the month the samples were collected (MONTH_COLLECTED; January as the first month), using the binomial family of general linear models. We used MONTH_COLLECTED as an indication of increasing sun exposure, temperatures, and increasing frequency of melt/freeze cycles. We included an interaction term for DAYS_OUT and MONTH_COLLECTED because we expected that colder months would show less deterioration of genetic material for a given period of time. We log10-transformed HAIRCOUNT because we expected increasing hair numbers to have an asymptotic effect on genotyping success. This analysis only used data from samples at WGI, where HAIRCOUNT information was available. To compare the effectiveness of different hair snag devices in capturing hair samples, we used coefficients from these models to calculate a guard hair equivalent (GHE) for underfur and mixed samples (Lamb et al., 2016), using DAYS_OUT = 30 and MONTH_COLLECTED = 1 as reference levels. Since hair samples were sometimes collected from snow, tree pitch, or branch stubs, we used GHE as our response on a set of samples from sites in north Columbia (all years, 2018–2020) where both barbwire and alligator clips were used simultaneously and the location of the hair sample was explicitly recorded. Additional data Individual detection The variable set for predicting individual detection included TOPO, SNOW, SETUP, BAIT, and ACCESS, scaled using DENSITY and DURATION as offsets. The highest ranked model predicting individual detection included variables related to both habitat and methods: 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE TABLE 2 KORTELLO ET AL. Model selection results to compare covariates predicting wolverine detection at bait stations. Model df LL AICc ΔAICc ωi TOPO 1 4 −414.51 837.01 13.18 0.00 Log(SNOW) 2 2 −410.33 824.69 0.85 0.23 Model set Habitat Method Combined factors Variables SETUP 3 2 −413.07 830.17 6.34 0.00 BAIT 4 2 −415.47 834.95 11.12 0.00 CAMERA + SETUP + BAIT 5 4 −412.07 832.22 8.39 0.00 ACCESS 6 2 −413.97 831.97 8.14 0.00 Log(SNOW) + SETUP 7 3 −408.90 823.83 0.00 0.35 Log(SNOW) + SETUP + ACCESS 8 4 −408.75 825.58 1.75 0.15 Log(SNOW) + SETUP + BAIT 9 4 −408.77 825.62 1.78 0.14 Log(SNOW) + ACCESS 10 3 −409.82 825.69 1.85 0.14 Note: Wolverine density and sampling duration are included in all models as offset terms. TOPO-contrasted topographical locations, valley, ridge, mid-slope and pass; SNOW was the snow depth (in centimeters) at each site when it was visited in April or early May; SETUP-contrasted baited trees wrapped in barbwire with sites having suspended bait and run poles with frames fixed to trees and integrated with cameras for ventral photographs; BAIT-contrasted sites baited with beaver versus other baits (ungulate heads, ungulate quarters, and domestic meat); CAMERA-contrasted sites with and without remote cameras present; and ACCESS-contrasted sites visited from the ground (by snowmachine, ski, or truck) with helicopter-accessed sites. Bold values indicate the model with the lowest AICc. Abbreviations: AICc, corrected Akaike information criterion for small sample sizes; LL, log-likelihood; ΔAICc, difference in AICc from the top model; ωi, Akaike weight for the model. SNOW and SETUP (Table 2). The habitat variable SNOW was a strong and consistent predictor of detection, indicating a positive effect of the amount of late winter snow (Figure 3) present on the ground at the site. For variables related to sampling methods, run pole setup methods appeared to be more effective than tree setups. A typical run pole setup detected 0.38 ± 0.07 (SE) individuals per site, compared to 0.27 ± 0.02 individuals per site for tree setups. Effect sizes for BAIT and ACCESS were small (ACCESS: heli β = 0.08 ± 0.14, p = 0.6, model 8; BAIT: beaver β = −0.07 ± 0.15, p = 0.6, model 9) and not significant. The variable TOPO did not appear in the top models. Sex-specific analyses generated top-ranked models similar to the combined-sex analysis; however, both snow and run pole setups were more important for female than for male detections (Table 3). Run pole setups detected more females (0.22 ± 0.05 per site) and a higher proportion of females (61%) than tree setups (0.14 ± 0.01 females per site; 55%). Temporal comparison For the pre-denning period, 9 January to 9 February, 118 sites were monitored continuously in four sampling areas. Male and female RDI were similar with no discernable temporal trend (Figure 4). For the first 2 months of the reproductive denning period, 10 February to 6 April, 195 sites in six sampling areas were in continuous operation. Temporal trends show diverging RDI in males and females beginning in late February and continuing through most of March, with male RDI generally increasing and females having both lower RDI and lower variability in RDI. Female RDI approached male RDI again by late March. Latency to detection We were able to evaluate LTD for 132 sites in a camera-monitored subset of our sites that retained consistency with the full dataset with respect to factors predicting individual detection. All LTD data necessarily had at least one detection; we were interested in how long it took for that detection to happen. Average time to detection at bait stations was ~32 days (range 1–95). The top combined model for latency included HUMANII and ACCESS (Table 4). Bait stations located in areas with low human influence were visited much more quickly than areas with anthropogenic impact (β = 0.07 ± 0.02; model 12). Additionally, bait stations set up using helicopters had shorter latency times than those accessed by ground travel (Table 5, Figure 5). For example, we estimated 10.6 ± 2.4 days for LTD in helicopter-accessed sites with no human infrastructure within 10 km, while ground-accessed sites having very minimal infrastructure nearby (HUMANII = 1) had an LTD of 35.0 ± 3.2 days. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 10 of 22 11 of 22 b) a) No. individuals detected No. individuals detected 0.35 0.30 0.25 0.20 0.15 0.10 0.50 0.45 0.40 0.35 0.30 0.25 0.05 0 tree 50 100 150 200 250 300 350 runpole Setup type Late winter snowpack (cm) F I G U R E 3 Covariates derived from the best multivariate model (model 7, Table 2, Table 3) predicting wolverine detection success at bait stations, back-transformed to original scale. Shaded areas indicate 95% CI. The response, number of individuals, is scaled by wolverine DENSITY, measured within a 10-km radius of the bait station, and DURATION, the number of days that a trap was operational. (a) SNOW was the snow depth (in centimeters) at each site when it was visited in April or early May. (b) SETUP-contrasted baited trees wrapped in barbwire with sites having suspended bait and run poles with frames fixed to trees and integrated with cameras for ventral photographs. TABLE 3 Parameter estimates and significance levels for variables appearing in the top wolverine detection model (model 7 in Table 2). Both sexes Female Male Model variable Estimate (SE) p Estimate (SE) p Estimate (SE) p (Intercept) −6.16 (0.15) <0.0001 −6.34 (0.22) <0.0001 −5.93 (0.19) <0.0001 Log(SNOW) 0.08 (0.03) 0.0133 0.09 (0.05) 0.0501 0.06 (0.04) 0.1260 SETUP:runpole 0.33 (0.19) 0.0783 0.44 (0.23) 0.0631 0.17 (0.31) 0.5950 Note: SNOW was the snow depth (in centimeters) at each site when it was visited in April or early May; and SETUP-contrasted baited trees wrapped in barbwire with sites having suspended bait and run poles with frames fixed to trees and integrated with cameras for ventral photographs. Other variables in competing models with AIC < 2 had weak effects that were also potentially of interest from a study design perspective. Latency times for sites with beaver as bait had a nonsignificant advantage (β = −0.25 ± 0.32; model 13) over domestic bait, and LTD had a nonsignificant declining trend throughout the sampling season, indicated by the FIRSTWEEK variable (β = −0.02 ± 0.02; model 14), such that the typical site set up in late November was visited 10 days later (42.3 ± 6.1 days) than those in March (32.2 ± 7.8 days). We did not find strong differences between sexes, but females averaged shorter latency times (27.6 ± 3.2 days) than males (30.3 ± 3.7 days) and wolverine of unknown sex (i.e., caught on camera but no viable genetic sample or definitive ventral photographs) took the longest (33.8 ± 3.6 days). Surprisingly, DENSITY, despite being negatively correlated with HUMANII, did not have a significant effect on latency (β = −0.12 ± 0.08; model 2) and did not appear in the top models. The top variables predicting detection, SNOW and SETUP, were also poor predictors of LTD, and so was topography (TOPO). Genotyping success With the exception of run pole setups, successful individual identification depended on the hair samples used for genetic analysis. We obtained 4642 genetic samples. Of these, 742 hair samples were wolverine, suitable for genotyping, and 529 of these generated successful wolverine genotypes. On average, sites with wolverine DNA detected had 3.6 samples per site that contained wolverine hair and 1.5 samples per site with successful genotypes. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE KORTELLO ET AL. F I G U R E 4 Relative detection index for wolverine bait sites with SE bars for male and female wolverine, for 118 sites monitored continuously in the pre-denning period (9 January—9 February) and 195 sites monitored continuously during the onset of the denning season (10 February–6 April). TABLE 4 Model selection results to compare covariates predicting wolverine latency to detection at bait stations. Model set Variable Model df LL AICc ΔAICc ωi Null NULL 1 2 −587.83 1179.8 19.2 0 Density Log(DENSITY) 2 3 −586.87 1179.9 19.3 0 Sex SEX 3 4 −585.93 1180.2 19.6 0 Habitat Method Timing Combined Log(ALPINE) 4 3 −587.70 1181.6 21 0 Log(SNOW) 5 3 −587.41 1181.0 20.4 0 TOPO 6 5 −587.35 1185.2 24.6 0 Log(HUMANII) 7 3 −580.86 1167.9 7.3 0.01 ACCESS 8 3 −579.43 1165.1 4.5 0.05 SETUP 9 3 −581.51 1169.2 8.6 0.01 BAIT 10 3 −586.95 1180.1 19.5 0 FIRSTWEEK 11 3 −580.26 1166.7 6.1 0.02 ACCESS + Log(HUMANII) 12 4 −576.12 1160.6 0 0.48 ACCESS + BAIT + Log(HUMANII) 13 5 −575.86 1162.2 1.6 0.22 ACCESS + FIRSTWEEK + Log(HUMANII) 14 5 −575.95 1162.4 1.8 0.20 Note: ALPINE was the percent area of alpine vegetation within a 10-km radius of the sampling location, SNOW was the snow depth (in centimeters) at each site when it was visited in April or early May; TOPO-contrasted topographical locations, valley, ridge, mid-slope and pass; HUMANII was an index of human use; ACCESS-contrasted sites visited from the ground (by snowmachine, ski, or truck) with helicopter-accessed sites; SETUP-contrasted baited trees wrapped in barbwire with sites having suspended bait and run poles with frames fixed to trees and integrated with cameras for ventral photographs; BAIT-contrasted sites baited with beaver versus other baits (ungulate heads, ungulate quarters, and domestic meat); and FIRSTWEEK was the week of the year the bait stations were deployed, sequentially, with the last week of November as week 1. Bold values indicate the model with the lowest AICc. Abbreviations: AICc, corrected Akaike information criterion for small sample sizes; LL, log-likelihood; ΔAICc, difference in AICc from the top model; ωi, Akaike weight for the model. We were able to evaluate the effect of hair type, number of hairs, sampling interval length, and month collected on genotyping success for 500 wolverine samples. To be conservative, we constrained our analysis to values of DAYS_OUT <60 and MONTH_COLLECTED <6, representative of typical sampling times and intervals; however, the dominant trends in our results were reinforced when we also included longer and later outlier events. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 12 of 22 13 of 22 We found strong relationships between genotyping success and the number of hairs, the length of the sampling interval, and the month the hairs were collected, for both hair types (Table 6). Guard hairs were much more likely to generate a genotype (Figure 6). At median levels of DAYS_OUT (30 days) and MONTH_COLLECTED (4, April), the probability of a successful genotype for one guard hair was 55 ± 5%. Two guard hairs increased the probability to 69 ± 3%, equivalent to approximately 30 underfur hairs (68 ± 6%, Figure 6). Ten guard hairs had a probability of 91 ± 2%. We observed declining genotype success over the length of the sampling interval, and this effect became more pronounced as the season progressed. Genotyping success for five guard hairs declined negligibly for different length sampling periods that ended in January, from 92 ± 3% at 30 days to 90 ± 2% at 60 days. In contrast, genotyping success dropped more substantially over T A B L E 5 Parameter estimates for variables appearing in the top model predicting latency to detection. Model variable Estimate (SE) p (Intercept) 3.555 (0.0917) <2e-16 Log(HUMANII) ACCESS:Helicopter 0.074 (0.0240) 0.002750 −0.520 (0.1471) 0.000574 Note: HUMANII was an index of human use; and ACCESS-contrasted sites visited from the ground (by snowmachine, ski, or truck) with helicopter-accessed sites. DISCUSSION Optimizing individual detection facilitates estimates of density and the ability to monitor population trends b) 40 Latency to detection (days) Latency to detection (days) a) equivalent sampling periods that ended in May, from 80 ± 4% at 30 days to 50 ± 18% at 60 days. For underfur hairs this decline was even steeper, so that even with 30 underfur hairs the probability of a successful genotype after 60 days, ending in May, was 11 ± 9%. In 120 run pole sites having both barbwire and alligator clips on the same setup, the number of possible wolverine hair samples we got from each snagging device was roughly equal (barbwire = 761, alligator clips = 763), but we were unable to make a similar comparison with gun brushes. In samples identified to wolverine, barbwire collected more hair, 4.9 ± 0.2 GHE per sample (n = 182), compared with 4.2 ± 0.9 GHE per sample (n = 16) for gun brushes, and 3.1 ± 0.3 GHE per sample (n = 131) for alligator clips. However, gun brush samples had the highest proportion of underfur hairs, averaging 71 ± 12% underfur per sample, whereas alligator clips had 58 ± 0.4% underfur per sample and barbwire was the lowest at 47 ± 0.3% underfur per sample. Since underfur hair DNA quality declined more rapidly with environmental exposure, barbwire samples tended to have the highest genotyping success, 74 ± 3%, followed by alligator clips, 63 ± 4%, and gun brushes, 50 ± 13%. On average, female wolverine left better hair samples (5.3 ± 0.3 GHE) than males (4.3 ± 0.3 GHE). 35 30 25 20 15 10 0 2 4 6 8 10 12 Human influence index 14 45 40 35 30 25 20 ground helicopter Access type F I G U R E 5 Covariates from the top model predicting wolverine latency to detection at bait stations, back-transformed to original scale. Shaded areas indicate 95% CI. (a) Human Influence Index (HUMANII), measured within a 10-km radius of the bait station. (b) ACCESS, access to the bait station, that is, ground (snowmachine, ski, or truck) versus helicopter access. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE TABLE 6 KORTELLO ET AL. Parameter estimates for variables affecting sample genotyping success of wolverine guard hairs and underfur. Guard hair Model variable (Intercept) Log(HAIRCOUNT) DAYS_OUT:MONTH_COLLECTED Underfur Estimate (SE) p Estimate (SE) p 1.302 (0.459) 0.005 −0.605 (1.010) 0.549 0.899 (0.166) 6.09e-8 0.950 (0.342) 0.0054 −0.009 (0.004) 0.012 −0.015 (0.005) 0.0008 Note: HAIRCOUNT was the number of hairs in the genetic sample; DAYS_OUT was the length of sampling interval; and MONTH_COLLECTED was the month of the year, January = 1, February = 2, etc. b) 1.0 0.8 0.6 0.4 0.2 0.0 5 10 15 20 25 30 Probability of a successful genotype Probability of successful genotype a) 20 1.0 60 20 40 60 MONTH = 1 MONTH = 2 MONTH = 3 MONTH = 4 MONTH = 5 0.8 0.6 0.4 0.2 0.0 20 40 60 No. guard hairs 20 40 60 20 40 60 Length of sampling interval c) d) 20 1.0 0.8 0.6 0.4 0.2 0.0 5 10 15 20 25 30 Probability of a successful genotype Probability of successful genotype 40 1.0 40 60 20 40 60 MONTH = 1 MONTH = 2 MONTH = 3 MONTH = 4 MONTH = 5 0.8 0.6 0.4 0.2 0.0 No. underfur hairs 20 40 60 20 40 60 20 40 60 Length of sampling interval F I G U R E 6 Factors affecting genotyping success of wolverine hair samples for guard hair: (a) hair quantity and (b) length of sampling interval by month of year (January = 1, February = 2, etc.), and for underfur hair: (c) hair quantity and (d) length of sampling interval by month of year (January = 1, February = 2, etc.). Laboratory analysis spanned a range of 1–10 hairs for guard hairs and 5–30 hairs for underfur; values were extrapolated to aid comparison. for wolverine and other low-density species. This information is critical for the implementation and evaluation of conservation measures. Using a large dataset of noninvasive genetic sampling sites with broad geographic extent and variable wolverine density, we demonstrated that several factors related to methods, habitat, and seasonal timing are important to wolverine individual detection and LTD at bait stations. We also provided information on factors affecting genotyping success, including hair type and exposure, and the effect of different hair snag devices on sample quality. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 14 of 22 Individual detection The configuration of bait station components was an important predictor of success. Magoun et al.’s (2011) run pole setup was more effective at detecting individuals than the tree or pole setup first described by Mulders et al. (2007). Ventral photographs facilitate the identification of individuals that do not leave hair samples, and can also be cross-referenced between sites or sessions having better hair samples. The run pole setup had higher detection for females than the tree setups. The proportion of females detected by run poles, 61%, closely matched the spatially explicit capture–recapture estimated proportion of females in the population, 62% (Mowat et al., 2020), and was greater than the proportion detected by tree setups, 55%, suggesting that females are overrepresented in the animals missed by tree setups. The run pole setup confers other advantages including the identification of sex and breeding condition and reductions in genetic laboratory costs once a particular ventral pattern is linked with a particular genotype. Run poles can be left in place for multi-year studies, and re-baiting is easy. Disadvantages include more materials and complexity, with longer setup times, higher initial costs, and difficulties in transporting materials and tools. Run poles also require cameras and incur the cost of analyzing photos. In montane environments, access can preclude randomized placement of bait stations. Most studies opt for a systematic deployment within grid cells of varying dimensions (Fisher et al., 2013; Lukacs et al., 2020; Robinson et al., 2017), although clustered sampling designs may allow field workers to sample inaccessible areas more efficiently (Clark, 2019; Efford & Boulanger, 2019). While we did not find an effect of access method or topography on detection, we did find that snow at bait stations mattered, even when we controlled for density effects. Wolverine have been linked strongly with snow presence (Copeland et al., 2010; Inman, Magoun, et al., 2012). Although the broad-scale extent of multiyear spring snow predicted wolverine density in our study area (Mowat et al., 2020), we also showed that local snow amounts affected wolverine detection, independently of density. Our results suggest that the presence of snow improved detection more than snow depth, because detection increased little after about 40 cm snow depth. The importance of this variable may be related to higher encounter rates resulting from fine-scale habitat use. Snow is used for reproductive dens and food caches (Inman, Magoun, et al., 2012; Magoun & Copeland, 1998) and Glass et al. (2021) showed that wolverine selected deeper, denser 15 of 22 snow for travel routes. Snow also retains scent, and this might provide a “scent pathway” facilitating encounters with bait stations. Future workers should select sites and sampling periods that will remain snow covered. As befits a generalist scavenger, wolverine were not selective about bait. However, beaver had a weak trend for shorter latency times, compared to other baits. Beaver are generally considered a good bait for winter trap sets because they are fatty and remain pliable and pungent even in cold temperatures, and in many places the skinned carcasses can be purchased from local trappers. Temporal comparison The timing of bait station deployment affected detection. Other studies have noted a trend toward increasing species-level detections over late winter (Fisher et al., 2013; Kortello et al., 2019), but our fine-scale temporal comparison indicated a reduction in daily RDI for females at the onset of the maternal denning period, in approximately mid-February (Inman, Magoun, et al., 2012; Magoun & Copeland, 1998). Wolverine kits are altricial, and this observation is consistent with limited foraging movements by the female beyond the den during the early denning period (Aronsson et al., 2023). Similarly, our observed difference in RDI was more pronounced in the early denning period, and female RDI approached male RDI by late March. Male detection generally increased through the winter, and late winter had the highest RDI. Trends toward increasing detection and decreasing LTD suggest that sampling later in the winter, into April or May, would generate higher detection for both male and female wolverine although care must be taken to choose sites that will retain snow during this time and bear encounters become a concern. Future analysts should test for lower detection probabilities for females from February to mid-March. Latency to detection LTD for bait stations averaged 32 days, rather a long period given potential losses from other scavengers. We observed many baits reduced to skeletons by other scavengers over this length of time. Wolverine use olfaction to detect carcasses, and scent plumes travel rapidly as a function of air movement and diffusion processes (Farrell et al., 2002). Since our traps averaged 7–8 km apart, and wolverine travel upward of 20 km/day, essentially covering the area of their annual home range over a similar 32-day period of time (Copeland, 1996; Inman, Packila, et al., 2012), this latency suggests either poorly 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE directed search efforts with respect to bait stations or, perhaps initially, active avoidance. The latter is supported by the poor relationship between density and LTD since, even under a random search pattern, more wolverine in an area would be expected to translate to faster encounter rates with bait stations. Naïve latency does not preclude a subsequently positive trap response, as detected in previous noninvasive baited wolverine studies (Barrueto et al., 2020; Mowat et al., 2020; Royle et al., 2011), because trap response is a learned behavior and bait is a substantive reward. Wolverine avoidance of human activity has been noted previously (Barrueto et al., 2022; Heinemeyer et al., 2019; Scrafford et al., 2018), and our results suggest a strong effect of anthropogenic influence on LTD. Our HUMANII variable mapped generalized human influence at broad scales, and we suggest that ACCESS method is another proxy for human use and influence at a finer resolution, in that helicopters are typically reserved for remote areas where difficulty in access provides little alternative. Our latency-modeling results indicated that remote sites without human influence were visited almost a month earlier than sites more proximal to human infrastructure, suggesting increased wariness in human-modified landscapes. These results are consistent with Stewart et al.’s (2016) observations of changes in wolverine behavior around bait stations associated with anthropogenic influence. Many other carnivores show behavioral avoidance of human presence, presumably due to risk associations (Muhly et al., 2011; Ordiz et al., 2021; Suraci et al., 2019). For wolverine, such behaviors might also be related to avoidance of competitive interactions with species associated with anthropogenically modified environments such as coyotes, rather than human presence per se (Chow-Fraser et al., 2022; Heim et al., 2017). In general, future workers should try to set their sites in areas removed from human disturbance. We found a weak trend toward lower LTD as winter progressed. This could reflect snow conditions and ease of travel in compacted, late winter snowpacks, although as previously mentioned, given the proximity of bait stations and the average daily travel distances of wolverine, we suspect travel times were not a limiting factor. Shorter latency times could also reflect a shift in alternative food availability and a stronger food drive or warming temperatures which leads to stronger smelling baits. Although little is known about annual energy budgets for wolverine, in late winter rodent availability is low, food caches may have become depleted and wolverine may be more dependent on less reliable resources—ungulates in poor condition and carrion caught in late winter avalanche cycles (Inman, Magoun, et al., 2012). Persson (2005) showed that female KORTELLO ET AL. wolverine are food limited, with reproductive success tied to food availability. Certainly, for reproductive females, after the initial period of restricted activity around parturition, there will be increasing energetic demands associated with lactation (Blecha et al., 2018; Clutton-Brock, 1991; Gittleman & Thompson, 1988) that could reduce their caution around bait stations. This is consistent with the weakly shorter time to detection for females. Our SETUP variable was not a good predictor of latency, suggesting that potential disturbance impacts associated with trap setup (i.e., noise, human presence, odor) were consistent between trap types, despite the higher amounts of infrastructure and longer initial construction times required for run poles. In general, the important variables in latency models were not the same as the important variables in detection models, suggesting that a long sampling interval can overcome wolverines’ initial reluctance to visit bait stations, especially if it extends past late March. Since longer trap deployments will generally lead to more detections, we suggest future workers leave traps out for several months and remind the reader that we checked most sites monthly and added new bait and lure when needed. Genotyping success While visiting bait stations, wolverine leave highly variable amounts of hair behind. The number of hairs in most samples is much fewer than most bear studies because wolverine hair have deep roots attached to the skin, and unlike bear hair sampling, sampling occurs in winter when natural shedding is minimal (Mowat et al., 2017). To reduce lab costs for genotyping, hair samples are frequently subsampled, thereby increasing the possibility that an individual may be missed. Conversely, generating two genotypes for the same animal at the same sampling period adds costs but no add information. Wolverine hair has higher genotyping success than that of grizzly bears, wolves, or pine marten (Lamb et al., 2016; Mowat & Paetkau, 2002; Stenglein et al., 2011); one guard hair provided a genotype more often than not, while underfur had a much lower likelihood of success. A negative effect of environmental exposure is well documented in studies of summer bear DNA genotyping success (Lamb et al., 2016; Stetz et al., 2014), but like Robinson et al. (2017), we found little loss in genotyping success rates for samples left for long periods in midwinter. However, as the season progressed toward spring, successive months showed an increasingly negative relationship between genotyping success and the number of days in the sampling interval. This loss was more pronounced for underfur than for guard hairs. Consequently, 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 16 of 22 we suggest future workers should plan to check traps at 20- to 30-day intervals if they are deployed during late winter or spring. However, since LTD can easily be as long as a month, access may be difficult, and little is gained from checking a trap that has not been visited yet, workers might consider longer intervals between trap checks in midwinter. Although many factors can potentially influence whether a snag device is successful in capturing hair, wolverine hair samples obtained from barbwire tended to be larger, have proportionately more guard hair, and were more likely to generate an individual genotype than those from either gun brushes or alligator clips. However, we point out that our sample from gun brushes was relatively small and may not be representative. Different snag devices may confer other advantages. For example, gun brushes facilitated rapid collection of samples during trap checks. Additionally, alligator clips in the run pole setup allow hair capture in a specific clip to be documented on camera and associated with a particular individual and chest pattern, even if more than one animal visits the site and leaves hair (Magoun et al., 2011). This can save on genetic analysis costs. We recommend subsampling based first on sample quality. Samples with two or more guard hairs are most likely to generate a genotype. Next, we suggest field staff use the location of different samples on the tree or run pole and camera data to help decide whether more than one animal may have visited the site and hence whether it is worth running more samples from that site visit. Monitoring wolverine using noninvasive capture– recapture is a tremendous improvement over snow tracking or harvest data for abundance estimation (Mowat, 2001). Yet, the resources required to access large areas of wolverine habitat are considerable and detection success is variable. To maximize individual detection within a selected area (e.g., predetermined grid cell), we recommend using run pole and camera bait station setups deployed in areas of persistent snow. Longer deployments had higher success, but we caution that detection for females may be lower temporarily during the reproductive denning period. We suggest using longer deployment times in areas that experience human influence, to account for longer LTD. Barbwire tends to provide better quality hair samples and laboratory subsampling protocols for genetic analysis should consider that even one guard hair from a wolverine has a reasonable chance of yielding a genotype, but use of more guard hairs can greatly increase genotyping success. Longer sampling intervals between trap checks will negatively affect genotyping success in warmer months. Consideration of these factors should provide better individual detection rates and thus, estimates of density to inform conservation measures for wolverine. 17 of 22 We attempted to control for systematic differences among sampling protocols and study areas using density and trap duration; however, we acknowledge the potential for unexamined confounding factors. Given the observed responses to snow, there may have been annual climatic effects on detection that we were unable to assess. Additionally, wolverine residency status within home ranges potentially affects behavior around bait stations and this could shift the context of our observations on detection and latency. We sampled wolverine in a montane environment at the southern end of their range in North America. The increase in detection that we observed in late winter may not occur, or may occur later, in more northern environments. Similarly, the use of run poles in arctic environments may be more difficult due to the lack of trees. Our study assessed factors affecting the initial capture of individual wolverine. However, recaptures are essential for density estimation. We assumed that the food reward associated with bait stations will reduce the caution and selectivity of wolverine in subsequent encounters with similar structures, facilitating recaptures. CONCLUSIONS We can make several suggestions to help field biologists to improve inventory results in the future: 1. Detections are higher in areas with greater wolverine density, but for a spatial design, it is important to sample the range of densities in a study area, especially if one intends to use a model to extrapolate density to unsampled areas. 2. Sample in areas that will remain snow covered through spring and away from human influence. 3. Run pole set-ups to improve individual detection, especially for females, as well as generating other data such as breeding condition. Tree setups trade off somewhat lower detection for less cost and effort to implement. 4. Beaver as bait may confer a slight advantage in the shorter term, but other baits have similar success. Larger baits last longer so are preferable. 5. While longer trap periods lead to greater detections, trapping later in the winter, and perhaps into early spring (May), has the potential to generate more captures and increase the chance of detecting females that are raising young that winter. 6. Analysts may want to allow for sex-based detection probabilities and accommodate lower detection for females during February to mid-March. 7. Longer deployments (>35 days) are necessary for detection in human-influenced areas. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE 8. Genotyping success was greater than 50% with one guard hair, and nearly 70% when two guard hairs are included in the sample. Studies that generate many samples may prefer to only genotype samples with two or more guard hairs to reduce genetic analysis costs. 9. Monthly, or more frequent (20–30 day), trap checks yield better DNA quality during late winter and spring, but longer trap check intervals could be used in midwinter with negligible quality loss. 10. Barbwire yielded more individual genotypes per sample than either alligator clips or gun brushes, but a combination of barbwire and alligator clips confers some advantages. In instances where multiple wolverine visited a bait station, the visual cue of displaced alligator clips can reduce genetic costs by allowing samples to be linked via a photograph to individuals. A U T H O R C ON T R I B U T I O NS Garth Mowat, Andrea Kortello, Doris Hausleitner, Mirjam Barrueto, and Anthony Clevenger conceived the ideas for this paper. Andrea Kortello led the analysis and writing. All authors provided data and input that shaped the analysis and manuscript. A C K N O WL E D G M E N T S Funding was provided by the British Columbia Ministry of Forests, the Columbia Basin Trust, the Fish and Wildlife Compensation Program, Parks Canada, Western Transportation Institute-Montana State University, Woodcock Foundation, Wilburforce Foundation, Liz Claiborne Art Ortenberg Foundation, Volgenau Foundation, Cross Foundation, United States Fish and Wildlife Service, Great Northern Landscape Conservation Cooperative, Alberta Conservation Association, Mountain Equipment Co-op, McLean Foundation, Patagonia, Cameron Plewes, Alberta Sport Parks Recreation and Wildlife Foundation, National Geographic Society, Disney Conservation Fund, Norcross Foundation, Bow Valley Naturalists, Yellowstone to Yukon Conservation Initiative, Lake O’Hara Lodge, Alpine Club of Canada, The Wolverine Foundation, the University of Calgary, Natural Sciences and Engineering Research Council of Canada (NSERC), British Columbia Habitat Conservation Trust Foundation, Forest Enhancement Society of British Columbia, Helicat Canada, K3 Cat Ski, and an anonymous donor. Substantial in-kind support was received from Mike Wiegele Helicopter Skiing, Selkirk Tangiers Helicopter Skiing, Whitetooth Helicopters, Mustang Powder, BC Provincial Parks, Idaho Fish and Game and Idaho Panhandle National Forest, U.S. Fish and Wildlife Service Wildlife and Sport Fish and Restoration Program, Teck Coal, MD of KORTELLO ET AL. Ranchlands, Alberta Environment and Parks-Fish and Wildlife, and Wildsight. Mount Revelstoke and Glacier National Park and Banff, Yoho, Kootenay National Parks wolverine bait station data were provided by Parks Canada Agency. We thank many dedicated field technicians, volunteers, and partner staff. The north Columbia work would not have been possible without support by M. Musiani, A. Jacob, B. Sayer, E. Ortner, M. Kosior, D. Moore, and J. Effa. Pilots S. Ward, J. Parker and D. Hawrys, D. DeBie, K. Lancour, and M. Morison provided safe and efficient helicopter transportation. Staff at Wildlife Genetics International and Rocky Mountain Research Station conducted all genetic analysis. We thank two anonymous reviewers for their useful comments on the manuscript. C O N F L I C T O F I N T E R E S T S T A TE M E N T The authors declare no conflicts of interest. DA TA AVAI LA BI LI TY S T ATE ME NT Data supporting this research are sensitive and are not publicly available due to the conservation status of wolverine in Canada. Wolverine bait station data and genetic data were compiled from numerous sources under data sharing agreements that did not include public dispersal of data. Ownership resides with the following researchers and agencies and data are available to qualified researchers. For data requests regarding the south Columbia wolverine bait stations and associated genetic data, please contact Andrea Kortello (kortello@yahoo.com), Grylloblatta Ecological Consulting or Doris Hausleitner (dorishaus78@ gmail.com), Seepanee Ecological Consulting. For north Columbia bait stations and genetic data, please contact Mirjam Barrueto (mirjam.barrueto@ucalgary.ca), Department of Biological Sciences, University of Calgary. For south Rockies and Banff, Yoho, Kootenay Nation Park data, please contact Anthony Clevenger (apclevenger@gmail. com), Western Transportation Institute, Montana State University. For Kananaskis data, please contact Nikki Heim (heimnikki@gmail.com), Nikki Heim Consulting. Mount Revelstoke and Glacier National Park portions of the data were obtained under license from Parks Canada Agency. Contact Mount Revelstoke and Glacier National Parks Ecological Integrity Monitoring Coordinator, Lisa Larson (lisa.larson@pc.gc.ca). Contact Michael Lucid (michael@selkirkwildlife.com), Selkirk Wildlife Science LLC, for wolverine genetic data from Idaho. Other datasets used for this research included Human Influence Index mapping (https://sedac.ciesin.columbia. edu/data/set/wildareas-v2-human-influence-index-geographic/ maps) and Biogeoclimatic Ecosystem Classification (BEC) Zone mapping (https://catalogue.data.gov.bc.ca/ dataset/f358a53b-ffde-4830-a325-a5a03ff672c3). 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 18 of 22 ORCID Andrea Kortello https://orcid.org/0000-0001-8047-3331 Michael Lucid https://orcid.org/0000-0003-0777-6365 R EF E RE N C E S Arnold, T. W. 2010. “Uninformative Parameters and Model Selection Using Akaike’s Information Criterion.” The Journal of Wildlife Management 74(6): 1175–78. https://doi.org/10. 1111/j.1937-2817.2010.tb01236.x. Aronsson, M., H. Andrén, M. Low, and J. Persson. 2023. “Wolverine Denning Behaviour and Its Implications for Monitoring Denning Females.” Wildlife Biology 2023(4): e01079. https://doi.org/10.1002/wlb3.01079. Barrueto, M., A. Forshner, J. Whittington, A. Clevenger, and M. Musiani. 2022. “Protection Status, Human Disturbance, Snow Cover and Trapping Drive Density of a Declining Wolverine Population in the Canadian Rocky Mountains.” Scientific Reports 12: 17412. https://doi.org/10.1038/s41598-022-21499-4. Barrueto, M., M. Sawaya, and A. Clevenger. 2020. “Low Wolverine (Gulo gulo) Density in a National Park Complex of the Canadian Rocky Mountains.” Canadian Journal of Zoology 98(5): 287–298. https://doi.org/10.1139/cjz-2019-0165. Barton, K. 2016. “MuMIn: Multi-Model Inference.” R Package Version 1.15.6. https://CRAN.R-project.org/package=MuMIn. Blecha, K. A., R. B. Boone, and M. W. Alldredge. 2018. “Hunger Mediates Apex Predator’s Risk Avoidance Response in Wildland–Urban Interface.” Journal of Animal Ecology 87(3): 609–622. https://doi.org/10.1111/1365-2656.12801. Burnham, K. P., and D. R. Anderson. 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. New York, NY: Springer-Verlag. https://doi.org/10. 1007/b97636. Chow-Fraser, G., N. Heim, J. Paczkowski, J. P. Volpe, and J. T. Fisher. 2022. “Landscape Change Shifts Competitive Dynamics between Declining at-Risk Wolverines and Range-Expanding Coyotes, Compelling a New Conservation Focus.” Biological Conservation 266: 109435. https://doi.org/10. 1016/j.biocon.2021.109435. Clark, J. D. 2019. “Comparing Clustered Sampling Designs for Spatially Explicit Estimation of Population Density.” Population Ecology 61: 93–101. https://doi.org/10.1002/1438390X.1011. Clutton-Brock, T. H. 1991. The Evolution of Parental Care. Princeton, NJ: Princeton University Press. https://doi.org/10. 2307/j.ctvs32ssj. Conner, M. M., J. J. Keane, C. V. Gallagher, T. E. Munton, and P. A. Shaklee. 2016. “Comparing Estimates of Population Change from Occupancy and Mark-Recapture Models for a Territorial Species.” Ecosphere 7(10): e01538. https://doi.org/ 10.1002/ecs2.1538. Conti, C., M. Guarino, and J. Bacenetti. 2020. “Measurements, Techniques and Models to Assess Odor Annoyance: A Review.” Environment International 134: 105261. https://doi. org/10.1016/j.envint.2019.105261. Copeland, J. P. 1996. “Biology of the Wolverine in Central Idaho.” M.S. Thesis, University of Idaho. Copeland, J. P., K. S. McKelvey, K. B. Aubry, A. Landa, J. Persson, R. M. Inman, J. Krebs, et al. 2010. “The Bioclimatic Envelope 19 of 22 of the Wolverine (Gulo gulo): Do Climatic Constraints Limit Its Geographic Distribution?” Canadian Journal of Zoology 88(3): 233–246. https://doi.org/10.1139/Z09-136. Doherty, P. F., G. C. White, and K. P. Burnham. 2012. “Comparison of Model Building and Selection Strategies.” Journal of Ornithology 152(2): 317–323. https://doi.org/10.1007/s10336010-0598-5. Efford, M. 2004. “Density Estimation in Live-Trapping Studies.” Oikos 106(3): 598–610. https://doi.org/10.1111/j.0030-1299. 2004.13043.x. Efford, M. G., and J. Boulanger. 2019. “Fast Evaluation of Study Designs for Spatially Explicit Capture-Recapture.” Methods in Ecology and Evolution 10(9): 1529–35. https://doi.org/10.1111/ 2041-210X.13239. Ellis, M. M., J. S. Ivan, and M. K. Schwartz. 2014. “Spatially Explicit Power Analysis for Occupancy-Based Monitoring of Wolverine Populations in the U.S. Rocky Mountains.” Conservation Biology 28(1): 52–62. https://doi.org/10.1111/cobi.12139. Ellison, A. M., and A. A. Agrawal. 2005. “The Statistics of Rarity.” Ecology 86(5): 1079–80. https://doi.org/10.1890/04-1456. Farrell, J. A., J. Murlis, X. Long, W. Li, and R. T. Cardé. 2002. “Filament-Based Atmospheric Dispersion Model to Achieve Short Time-Scale Structure of Odor Plumes.” Environmental Fluid Mechanics 2: 143–169. https://doi.org/10.1023/A: 1016283702837. Fisher, J. T., and S. Bradbury. 2014. “A Multi-Method Hierarchical Modeling Approach to Quantifying Bias in Occupancy from Noninvasive Genetic Tagging Studies.” The Journal of Wildlife Management 78(6): 1087–95. https://doi.org/10.1002/ jwmg.750. Fisher, J. T., S. Bradbury, B. Anholt, L. Nolan, L. Roy, J. P. Volpe, and M. Wheatley. 2013. “Wolverines (Gulo gulo luscus) on the Rocky Mountain Slopes: Natural Heterogeneity and Landscape Alteration as Predictors of Distribution.” Canadian Journal of Zoology 91(10): 706–716. https://doi.org/10.1139/ cjz-2013-0022. Fisher, J. T., S. Murray, M. Barrueto, K. Carroll, A. P. Clevenger, D. Hausleitner, W. Harrower, et al. 2022. “Wolverines (Gulo gulo) in a Changing Landscape and Warming Climate: A Decadal Synthesis of Global Conservation Ecology Research.” Global Ecology and Conservation 34: e02019. https://doi.org/ 10.1016/j.gecco.2022.e02019. Fox, J. 2003. “Effect Displays in R for Generalised Linear Models.” Journal of Statistical Software 8(15): 1–27. https://doi.org/10. 18637/jss.v008.i15. Fox, J., and S. Weisberg. 2018. “Visualizing Fit and Lack of Fit in Complex Regression Models with Predictor Effect Plots and Partial Residuals.” Journal of Statistical Software 87(9): 1–27. https://doi.org/10.18637/jss.v087.i09. Fox, J., and S. Weisberg. 2019. An R Companion to Applied Regression, 3rd ed. Thousand Oaks, CA: Sage. https:// socialsciences.mcmaster.ca/jfox/Books/Companion/index. html. Gittleman, J. L., and S. D. Thompson. 1988. “Energy Allocation in Mammalian Reproduction.” American Zoologist 28(3): 863–875. https://doi.org/10.1093/icb/28.3.863. Glass, T. W., G. A. Breed, G. E. Liston, A. K. Reinking, M. D. Robards, and K. Kielland. 2021. “Spatiotemporally Variable Snow Properties Drive Habitat Use of an Arctic 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE Mesopredator.” Oecologia 195: 887–899. https://doi.org/10. 1007/s00442-021-04890-2. Glasser, G. J., and R. F. Winter. 1961. “Critical Values of the Coefficient of Rank Correlation for Testing the Hypothesis of Independence.” Biometrika 48: 444–48. https://doi.org/10. 1093/biomet/48.3-4.444. Green, A. M., M. W. Chynoweth, and Ç. H. Şekercioglu. 2020. “Spatially Explicit Capture-Recapture through Camera Trapping: A Review of Benchmark Analyses for Wildlife Density Estimation.” Frontiers in Ecology and Evolution 8: 563477. https://doi.org/10.3389/fevo.2020.563477. Green, P. A., B. Van Valkenburgh, B. Pang, D. Bird, T. Rowe, and A. Curtis. 2012. “Respiratory and Olfactory Turbinal Size in Canid and Arctoid Carnivorans.” Journal of Anatomy 221(6): 609–621. https://doi.org/10.1111/j.1469-7580.2012.01570.x. Heim, N., J. T. Fisher, A. Clevenger, J. Paczkowski, and J. Volpe. 2017. “Cumulative Effects of Climate and Landscape Change Drive Spatial Distribution of Rocky Mountain Wolverine (Gulo gulo l.).” Ecology and Evolution 7(21): 8903–14. https://doi.org/ 10.1002/ece3.3337. Heinemeyer, K., J. Squires, M. Hebblewhite, J. J. O’Keefe, J. D. Holbrook, and J. Copeland. 2019. “Wolverines in Winter: Indirect Habitat Loss and Functional Responses to Backcountry Recreation.” Ecosphere 10(2): e02611. https://doi. org/10.1002/ecs2.2611. Herbert, B. M. J., S. Villa, and C. J. Halsall. 2006. “Chemical Interactions with Snow: Understanding the Behavior and Fate of Semi-Volatile Organic Compounds in Snow.” Ecotoxicology and Environmental Safety 63: 3–16. https://doi.org/10.1016/j. ecoenv.2005.05.012. Inman, R. M., A. J. Magoun, J. Persson, and J. Mattisson. 2012. “The Wolverine’s Niche: Linking Reproductive Chronology, Caching, Competition, and Climate.” Journal of Mammalogy 93(3): 634–644. https://doi.org/10.1644/11-MAMM-A-319.1. Inman, R. M., M. L. Packila, K. H. Inman, A. J. McCue, G. C. White, J. Persson, B. C. Aber, et al. 2012. “Spatial Ecology of Wolverines at the Southern Periphery of Distribution.” The Journal of Wildlife Management 76(4): 778–792. https://doi. org/10.1002/jwmg.289. Johnson, W. E., and S. J. O’Brien. 1997. “Phylogenetic Reconstruction of the Felidae Using 165 rRNA and NADH-5 Mitochondrial Genes.” Journal of Molecular Evolution 44: S98–S116. https://doi.org/10.1007/PL00000060. Keiter, D. A., A. J. Davis, O. E. Rhodes, Jr., F. L. Cunningham, J. C. Kilgo, K. M. Pepin, and J. C. Beasley. 2017. “Effects of Scale of Movement, Detection Probability, and True Population Density on Common Methods of Estimating Population Density.” Scientific Reports 7: 9446. https://doi.org/10.1038/s41598-01709746-5. Kendall, K. C., J. B. Stetz, J. Boulanger, A. C. Macleod, D. Paetkau, and G. C. White. 2009. “Demography and Genetic Structure of a Recovering Grizzly Bear Population.” The Journal of Wildlife Management 73(1): 3–16. https://doi.org/10.2193/2008-330. Kéry, M. 2002. “Inferring the Absence of a Species: A Case Study of Snakes.” The Journal of Wildlife Management 66(2): 330–38. https://doi.org/10.2307/3803165. Kéry, M., B. Gardner, T. Stoeckle, D. Weber, and J. A. Royle. 2011. “Use of Spatial Capture-Recapture Modeling and DNA Data to Estimate Densities of Elusive Animals.” Conservation Biology KORTELLO ET AL. 25(2): 356–364. https://doi.org/10.1111/j.1523-1739.2010. 01616.x. Kortello, A., D. Hausleitner, and G. Mowat. 2019. “Mechanisms Influencing the Winter Distribution of Wolverine Gulo gulo luscus in the Southern Columbia Mountains, Canada.” Wildlife Biology 2019(1): 1–13. https://doi.org/10.2981/wlb.00480. Krebs, J., E. C. Lofroth, and I. Parfitt. 2007. “Multiscale Habitat Use by Wolverines in British Columbia, Canada.” The Journal of Wildlife Management 71(7): 2180–92. https://doi.org/10.2193/ 2007-099. Lamb, C., D. Walsh, and G. Mowat. 2016. “Factors Influencing Detection of Grizzly Bears at Genetic Sampling Sites.” Ursus 27: 31–44. https://doi.org/10.2192/URSUS-D-15-00025.1. Landa, A., J. Tufto, R. Franzén, T. Bø, M. Lindén, and J. E. Swenson. 1998. “Active Wolverine Gulo gulo Dens as a Minimum Population Estimator in Scandinavia.” Wildlife Biology 4(3): 159–168. https://doi.org/10.2981/wlb.1998.018. Lofroth, E., and P. Ott. 2007. “Assessment of the Sustainability of Wolverine Harvest in British Columbia, Canada.” The Journal of Wildlife Management 71(7): 2193–2200. https://doi.org/10. 2193/2007-095. Lukacs, P. M., D. E. Mack, R. Inman, J. A. Gude, J. S. Ivan, R. P. Lanka, J. C. Lewis, et al. 2020. “Wolverine Occupancy, Spatial Distribution, and Monitoring Design.” The Journal of Wildlife Management 84(5): 841–851. https://doi.org/10.1002/jwmg. 21856. MacKenzie, D. I., J. D. Nichols, J. A. Royle, K. H. Pollock, L. Bailey, and J. E. Hines. 2017. Occupancy Estimation and Modeling, 2nd ed. San Diego, CA: Academic Press. https://www.elsevier. com/books/occupancy-estimation-and-modeling/mackenzie/ 978-0-12-407197-1. MacKillop, D. J., and A. J. Ehman. 2016. A Field Guide to the Site Classification and Identification for Southeast British Columbia: The South-Central Columbia Mountains. Land Management Handbook, Vol. 70. Victoria, BC: Province of British https://www.for.gov.bc.ca/hfd/pubs/Docs/Lmh/ Columbia. Lmh70.htm. Magoun, A. J., and J. P. Copeland. 1998. “Characteristics of Wolverine Reproductive Den Sites.” The Journal of Wildlife Management 62(4): 1313–20. https://doi.org/10.2307/3801996. Magoun, A. J., C. D. Long, M. K. Schwartz, K. L. Pilgrim, R. E. Lowell, and P. Valkenburg. 2011. “Integrating Motion-Detection Cameras and Hair Snags for Wolverine Identification.” The Journal of Wildlife Management 75: 731–39. https://doi.org/10.1002/jwmg.107. McKelvey, K. S., J. P. Copeland, M. K. Schwartz, J. S. Littell, K. B. Aubry, J. R. Squires, S. A. Parks, M. M. Elsner, and G. S. Mauger. 2011. “Climate Change Predicted to Shift Wolverine Distributions, Connectivity, and Dispersal Corridors.” Ecological Applications 21(8): 2882–97. https://doi.org/10. 1890/10-2206.1. Morin, D. J., J. Boulanger, R. Bischof, D. C. Lee, D. Ngoprasert, A. K. Fuller, B. McLellan, et al. 2022. “Comparison of Methods for Estimating Density and Population Trends for Low-Density Asian Bears.” Global Ecology and Conservation 35: e02058. https://doi.org/10.1016/j.gecco.2022.e02058. Mowat, G. 2001. “Measuring Wolverine Distribution and Abundance in Alberta.” Alberta Species at Risk Report No. 32. Alberta Sustainable Resource Development. Fish and Wildlife Division. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 20 of 22 https://open.alberta.ca/dataset/163495fa-4880-4295-a51d-eff57 8eea9aa/resource/76e5d93c-daf6-4142-9250-6bcc48e50f02/down load/sar032-measuringwolverinedistributionabundancealbertajan2001.pdf. Mowat, G., A. P. Clevenger, A. D. Kortello, D. Hausleitner, M. Barrueto, L. Smit, C. Lamb, B. Dorsey, and P. K. Ott. 2020. “The Sustainability of Wolverine Trapping Mortality in Southern Canada.” The Journal of Wildlife Management 84(2): 213–226. https://doi.org/10.1002/jwmg.21787. Mowat, G., P. J. Curtis, and D. J. R. Lafferty. 2017. “The Influence of Sulphur and Hair Growth on Stable Isotope Diet Estimates for Grizzly Bears.” PLoS One 12(3): e0172194. https://doi.org/ 10.1371/journal.pone.0172194. Mowat, G., C. Kyle, and D. Paetkau. 2003. “Testing Methods for Detecting Wolverine.” Alberta Species at Risk Report No. 71. Alberta Sustainable Resource Development. Fish and Wildlife Division. https://open.alberta.ca/dataset/88f4cc4f-a407-44d4bcf1-93a78da40840/resource/7caf85c1-d61d-4066-abff-26be6d0 599ed/download/2003-sar071-testingmethodsdetectingwolverinemar2003.pdf. Mowat, G., and D. Paetkau. 2002. “Estimating Marten Martes americana Population Size Using Hair Capture and Genetic Tagging.” Wildlife Biology 8(3): 201–9. https://doi.org/10.2981/ wlb.2002.034. Muhly, T. B., C. Semeniuk, A. Massolo, L. Hickman, and M. Musiani. 2011. “Human Activity Helps Prey Win the Predator-Prey Space Race.” PLoS One 6(3): 1–8. https://doi. org/10.1371/journal.pone.0017050. Mulders, R., J. Boulanger, and D. Paetkau. 2007. “Estimation of Population Size for Wolverine Gulo gulo at Daring Lake, Northwest Territories, Using DNA Based Mark-Recapture Methods.” Wildlife Biology 13(Suppl. 2): 38–51. https://doi.org/ 10.2981/0909-6396(2007)13[38:EOPSFW]2.0.CO;2. Ordiz, A., M. Aronsson, J. Persson, O. Støen, J. E. Swenson, and J. Kindberg. 2021. “Effects of Human Disturbance on Terrestrial Apex Predators.” Diversity 13(2): 68. https://doi.org/ 10.3390/d13020068. Paetkau, D. 2003. “An Empirical Exploration of Data Quality in DNA-Based Population Inventories.” Molecular Ecology 12(6): 1375–87. https://doi.org/10.1046/j.1365-294X.2003. 01820.x. Persson, J. 2005. “Female Wolverine (Gulo gulo) Reproduction: Reproductive Costs and Winter Food Availability.” Canadian Journal of Zoology 83: 1453–59. https://doi.org/10.1139/ Z05-143. Pozzanghera, C. 2015. “Non-Invasive Methods for Obtaining Occupancy Probabilities and Density Estimates of Interior Alaska’s Mesocarnivore Populations.” M.S. Thesis, University of Alaska. https://scholarworks.alaska.edu/handle/11122/ 6384. R Core Team. 2021. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. https://www.R-project.org/. Robinson, L., S. A. Cushman, and M. K. Lucid. 2017. “Winter Bait Stations as a Multispecies Survey Tool.” Ecology and Evolution 7(17): 6826–38. https://doi.org/10.1002/ece3.3158. Royle, J. A., A. J. Magoun, B. Gardner, P. Valkenburg, and R. E. Lowell. 2011. “Density Estimation in a Wolverine Population Using Spatial Capture–Recapture Models.” The Journal of 21 of 22 Wildlife Management 75(3): 604–611. https://doi.org/10.1002/ jwmg.79. Royle, J. A., and J. D. Nichols. 2003. “Estimating Abundance from Repeated Presence–Absence Data or Point Counts.” Ecology 84(3): 777–790. https://doi.org/10.1890/0012-9658(2003)084 [0777:EAFRPA]2.0.CO;2. Ruzicka, R. E., and M. R. Conover. 2012. “Does Weather or Site Characteristics Influence the Ability of Scavengers to Locate Food?” Ethology 118(2): 187–196. https://doi.org/10.1111/j. 1439-0310.2011.01997.x. Sawaya, M. A., A. P. Clevenger, and M. K. Schwartz. 2019. “Demographic Fragmentation of a Protected Wolverine Population Bisected by a Major Transportation Corridor.” Biological Conservation 236: 616–625. https://doi.org/10.1016/j. biocon.2019.06.030. Schaub, M., O. Gimenez, A. Sierro, and R. Arlettaz. 2007. “Use of Integrated Modeling to Enhance Estimates of Population Dynamics Obtained from Limited Data.” Conservation Biology https://doi.org/10.1111/j.1523-1739.2007. 21(4): 945–955. 00743.x. Schwartz, M. K., J. P. Copeland, N. J. Anderson, J. R. Squires, R. M. Inman, K. S. McKelvey, K. L. Pilgrim, L. P. Waits, and S. A. Cushman. 2009. “Wolverine Gene Flow across a Narrow Climatic Niche.” Ecology 90(11): 3222–32. https://doi.org/10. 1890/08-1287.1. Schwartz, M. K., G. Luikart, and R. S. Waples. 2007. “Genetic Monitoring as a Promising Tool for Conservation and Management.” Trends in Ecology & Evolution 22(1): 25–33. https://doi.org/10.1016/j.tree.2006.08.009. Scrafford, M. A., T. Avgar, R. Heeres, and M. S. Boyce. 2018. “Roads Elicit Negative Movement and Habitat-Selection Responses by Wolverines (Gulo gulo luscus).” Behavioral Ecology 29(3): 534–542. https://doi.org/10.1093/beheco/arx182. Slough, B. G. 2007. “Status of the Wolverine Gulo gulo in Canada.” Wildlife Biology 13(s2): 76–82. https://doi.org/10.2981/09096396(2007)13[76:SOTWGG]2.0.CO;2. Stenglein, J. L., L. P. Waits, D. E. Ausband, P. Zager, and C. M. Mack. 2011. “Estimating Gray Wolf Pack Size and Family Relationships Using Noninvasive Genetic Sampling at Rendezvous Sites.” Journal of Mammalogy 92(4): 784–795. https://doi.org/10.1644/10-MAMM-A-200.1. Stetz, J. B., T. Seitz, and M. A. Sawaya. 2014. “Effects of Exposure on Genotyping Success Rates of Hair Samples from Brown and American Black Bears.” Journal of Fish and Wildlife Management 6(1): 191–98. https://doi.org/10.3996/122013JFWM-085. Stewart, F. E. C., N. A. Heim, A. P. Clevenger, J. Paczkowski, J. P. Volpe, and J. T. Fisher. 2016. “Wolverine Behavior Varies Spatially with Anthropogenic Footprint: Implications for Conservation and Inferences about Declines.” Ecology and Evolution 6(5): 1493–1503. https://doi.org/10.1002/ece3. 1921. Suraci, J. P., M. Clinchy, L. Y. Zanette, and C. C. Wilmers. 2019. “Fear of Humans as Apex Predators Has Landscape-Scale Impacts from Mountain Lions to Mice.” Ecology Letters 22(10): 1578–86. https://doi.org/10.1111/ele.13344. Tredick, C. A., M. R. Vaughan, D. F. Stauffer, S. L. Simek, and T. Eason. 2007. “Sub-Sampling Genetic Data to Estimate Black Bear Population Size: A Case Study.” Ursus 18(2): 179–188. 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ECOSPHERE https://doi.org/10.2192/1537-6176(2007)18[179:SGDTEB]2.0. CO;2. Verheggen, F., K. A. Perrault, R. Caparros Megido, L. M. Dubois, F. Francis, E. Haubruge, S. L. Forbes, J. Focant, and P. Stefanuto. 2017. “The Odor of Death: An Overview of Current Knowledge on Characterization and Applications.” Bioscience 67(7): 600–613. https://doi.org/10.1093/biosci/ bix046. Waits, L. P., and D. Paetkau. 2005. “Noninvasive Genetic Sampling Tools for Wildlife Biologists: A Review of Applications and Recommendations for Accurate Data Collection.” The Journal of Wildlife Management 69(4): 1419–33. https://doi.org/10.2193/0022-541X(2005)69[1419: NGSTFW]2.0.CO;2. Walsh, J. R., E. J. Pedersen, and M. J. Vander Zanden. 2018. “Detecting Species at Low Densities: A New Theoretical Framework and an Empirical Test on an Invasive Zooplankton.” Ecosphere 9(11): e02475. https://doi.org/10. 1002/ecs2.2475. Webb, S., D. Manzer, R. Anderson, and M. Jokinen. 2013. “Wolverine Harvest Summary from Registered Traplines KORTELLO ET AL. in Alberta, 1985–2011.” Conservation Report No. T-2013-001. Alberta Conservation Association. https://www.ab-conservation. com/downloads/report_series/wolverine_harvest_in_alberta_ 1985-2011.pdf. Zuur, A. F., E. N. Ieno, and C. S. Elphick. 2010. “A Protocol for Data Exploration to Avoid Common Statistical Problems.” Methods in Ecology and Evolution 1: 3–14. https://doi.org/10. 1111/j.2041-210X.2009.00001.x. How to cite this article: Kortello, Andrea, Doris Hausleitner, Garth Mowat, Mirjam Barrueto, Nicole Heim, Lisa Larson, Michael Lucid, and Anthony Clevenger. 2024. “How to Find a Wolverine: Factors Affecting Detection at Wolverine (Gulo Gulo) Bait Stations in Western Canada.” Ecosphere 15(7): e4929. https://doi.org/ 10.1002/ecs2.4929 21508925, 2024, 7, Downloaded from https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.4929 by SELKIRK COLLEGE, Wiley Online Library on [09/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 22 of 22