Land tenure shapes black bear density and abundance on a multi‐use landscape

Abstract Global biodiversity is decreasing rapidly. Parks and protected lands, while designed to conserve wildlife, often cannot provide the habitat protection needed for wide‐ranging animals such as the American black bear (Ursus americanus). Conversely, private lands are often working landscapes (e.g., farming) that have high human footprints relative to protected lands. In southwestern Alberta, road densities are highest on private lands and black bears can be hunted year‐round. On protected lands, road densities are lowest, and hunting is prohibited. On public lands under the jurisdiction of the provincial government (Crown lands), seasonal hunting is permitted. Population estimates are needed to calculate sustainable harvest levels and to monitor population trends. In our study area, there has never been a robust estimate of black bear density and spatial drivers of black bear density are poorly understood. We used non‐invasive genetic sampling and indices of habitat productivity and human disturbance to estimate density and abundance for male and female black bears in 2013 and 2014 using two methods: spatially explicit capture–recapture (SECR) and resource‐selection functions (RSF). Land tenure best explained spatial variation in black bear density. Black bear densities for females and males were highest on parkland and lowest on Crown lands. Sex ratios were female‐biased on private lands, likely a result of lower harvests and movement of females out of areas with high male density. Synthesis and application: Both SECR and RSF methods clearly indicate spatial structuring of black bear density, with a strong influence based on how lands are managed. Land tenure influences the distribution of available foods and risk from humans. We emphasize the need for improved harvest reporting, particularly for non‐licensed hunting on private land, to estimate the extent of black bear harvest mortality.


| INTRODUC TI ON
Increases in the global human population and associated infrastructure development are increasing habitat fragmentation and destruction, and global biodiversity is rapidly decreasing (Benítez-López, Alkemade, & Verweij, 2010). Parks and protected areas often serve as areas of reduced human influence to conserve and protect habitats. While many wildlife species have done well within this framework (e.g., grizzly bear [Ursus arctos] recovery in Yellowstone National Park; van Manen et al., 2016), parks and protected areas have been criticized for preserving scenic beauty rather than biodiversity or connectivity (Jenkins, Van Houtan, Pimm, & Sexton, 2015). In North America, mountainous protected areas have a high proportion of rock and ice, which for many species does not provide adequate foraging opportunities (Joppa & Pfaff, 2009). Further, these areas are commonly of low soil fertility, which, in turn, can result in nutrient-poor areas and increased chances of food shortages (Rogers, 1987). For wide-ranging species, small protected areas do not encompass year-round habitat and the local extinction rate of mammals has been found to be inversely related to spatial extent of the protected area (Newmark, 1995).
Even where protected areas exist, wildlife commonly will use adjacent, unprotected private and public lands for dispersal, migration, foraging, breeding, and overwintering (Berger, 2004). Often, unprotected lands are working landscapes, such as agricultural lands used for farming and ranching. These lands can be productive and attractive to animals for their high-quality forage (Sayre, Carlisle, Huntsinger, Fisher, & Shattuck, 2012), as well as provide food subsidies from agriculture such as stored and standing grain and hay, silage, livestock, dead livestock, and beeyards (Wilson, Madel, Mattson, Graham, & Merrill, 2006). In southwestern Alberta, Canada, private lands are primarily agricultural areas used for cattle ranching, cereal grain, and oilseed farming. While private lands in this area have been shown to be attractive to grizzly bears (Northrup, Stenhouse, & Boyce, 2012), little is known about what drives spatial variation for other large carnivores, including black bears (Ursus americanus).
In North America, black bears are legally harvested throughout much of their range (Garshelis, 1990). In hunted populations, abundance and density estimates are needed to calculate sustainable harvest levels (Williams, Nichols, & Conroy, 2002). Despite low power to detect population trends, harvest data often are the only information biologists have to assess trends or to set harvest objectives (Garshelis & Hristienko, 2006). In Alberta, hunters are allowed to harvest 12% of the estimated provincial population, but the most recent population estimates are 20-30 years old (Gunson & Markham, 1993). In southwestern Alberta, the provincial government derived the minimum number of black bears in permanently occupied habitats by using percent cover of habitat, with minimum black bear densities for an ecoregion extrapolated from an adjacent study area and wildlife management unit (Gunson & Markham, 1993). Using this method, minimum black bear density was estimated to be 52.9 bears/1,000 km 2 (excluding Waterton Lakes National Park [WLNP]), which was low relative to other wildlife management units in Alberta (Gunson & Markham, 1993). Further, no error estimates were calculated so we do not know the extent of variance around the mean. Thus, there is a need to improve both the empirical data and the sophistication of the methods used to estimate density.
Spatially explicit capture-recapture and resource-selection functions (RSF) are two methods that can be used to estimate population density and account for spatiotemporal factors affecting density. SECR models use location information from detection events to estimate the distribution of animal home-range centers and to account for the effects of animal space use and home-range centers on the detection process (Efford, 2004;Royle, Chandler, Sollmann, & Gardner, 2014). These models require large amounts of field data (Czaplewski, Crowe, & McDonald, 1983), which can be time and cost intensive to collect (Royle & Nichols, 2003), and must meet multiple assumptions to avoid biasing parameter estimates and model overparameterization (Choquet, Lebreton, Gimenez, Reboulet, & Pradel, 2009;Fletcher et al., 2012).
Animal densities are usually related to habitat selection , and this relationship has been explored using RSFs (Johnson, Nielsen, Merrill, McDonald, & Boyce, 2006). For example, Boyce and McDonald (1999) associated abundance with RSF scores, which could then be extrapolated to new or unsampled areas. To date, RSF-abundance extrapolations have been applied primarily to theoretical or expanding populations (Boyce & Waller, 2003;Mladenoff & Sickley, 1998). Projections are important to anticipate how habitats could shape population expansions. Because the RSF-abundance extrapolation only requires presence data, it could provide a low-cost alternative to estimating density and abundance for wildlife managers, who often operate on restricted budgets. However, this method has never been applied to a population with concurrent mark-recapture data.
We use SECR models to estimate black bear density from DNA data at rub trees and compare these to RSF-based results. Using non-invasive genetic sampling (NGS) data collected from rub trees, power poles, fence posts, and fence lines (hereafter, rub objects), our objectives were to (a) estimate black bear density using both SECR and RSF methods for the same data, and (b) use information-theoretic methods with SECR and RSF to identify spatial covariates that best explain spatial variation in density. We used black bear NGS data from southwestern Alberta (2013-2014), habitat and human-disturbance covariates, and grizzly bear detection data to account for spatial variation in black bear density. We predicted black bear densities would be highest in protected areas where road densities are lowest hunting is prohibited, and where forested escape terrain from grizzly bears is more abundant. Black bear populations can be female-biased in unhunted populations. This bias is exacerbated in hunted populations, however, because hunters disproportionately select males (Bunnell & Tait, 1980;Miller, 1990). We predicted higher female densities relative to males. | 75 LOOSEN Et aL.

| S TUDY ARE A
The 3,600 km 2 study area is in the southern Canadian Rocky

Mountains and is bounded by Highway 3 to the north, British
Columbia to the west, the United States-Canada border to the south, and Highway 2 to the east (Figure 1). The area includes WLNP, which borders Glacier National Park (GNP), USA. The area is a mix of landcover types: conifer forest (29%), agricultural (22%), native grassland and cultivated fields (16%), shrubland (16%), and deciduous forest (11%). Agriculture is the primary industry (Statistics Canada, MD of Pincher Creek, 2011 Community Profile).
Land management in southwestern Alberta is varied. Private land (1,872 km 2 ; 52% of area) has the highest road density (1.3 km/ km 2 ; Northrup et al., 2012) and is characterized by rough fescue grasslands and agriculture. Crown lands (i.e., public lands under the jurisdiction of the provincial government; 1,204 km 2 ; 34% of area) have lower road density (0.55 km/km 2 ) relative to private land, and licensed black bear hunting occurs in the spring (1 April-31 May) and fall (1 September-30 November). Crown lands are characterized by alpine, montane, and aspen-parkland habitat (Northrup et al., 2012).
Protected areas (511 km 2 ; 14% of area), which includes WLNP and Beauvais Provincial Park, has the lowest road density (0.18 km/km 2 ), and hunting is prohibited. These areas are characterized by alpine, montane, and aspen-parkland vegetation.

| Hair collection and analysis
In 2011 and 2012, we surveyed for and set up rub objects (Figure 1).
We searched for rub objects based on bear travel corridors, roads and trails, local knowledge, and grizzly bear resource-selection maps (Northrup et al., 2012). We established rub objects where we observed fresh hair and attached four barbed-wire segments to each object. We also surveyed barbed-wire fencelines for hair and marked start and end points for resurvey.

Number of rub objects detecting a black bear
Hair collection dates Notes. Data were collected from rub objects (n = 873) in 2013 and 2014. Data from opportunistic samples were grouped into the eighth occasion.

Number of individuals detected
a Number of active sampling stations may vary depending on destruction of a rub tree from windfall or avalanche, access issues because of heavy snow, flooding, or discovery and set up of new rub trees.
( Table 1). We removed all hair during the first visit of each year. The remaining seven visits were collection events, and we collected all hair that was present. We considered a barb or end of a wire a discrete sampling unit, and thus, hair was collected only from the wire.
After collection, we burned remnant hairs to prevent false recaptures. A second data source included "opportunistic" hair samples collected by landowners, Fish and Wildlife officers visiting conflict sites, and field technicians. Opportunistic samples were collected throughout bears' active months (March-December) and were assigned to an eighth sample occasion in 2013 and 2014, as is common with opportunistic samples in traditional mark-recapture (Kendall et al., 2009(Kendall et al., , 2015. Hair samples were stored in coin envelopes and sent to Wildlife Genetics International (WGI; Nelson, British Columbia) for genetic analysis to determine species and individual via multi-locus microsatellite analysis of nuclear DNA (Paetkau, 2003(Paetkau, , 2004. Sex was assigned using the amelogenin marker (Ennis & Gallagher, 1994).
Morehouse and Boyce (2016) explored grizzly bear data from southwestern Alberta using various subsampling strategies and found that subsampling every third hair sample in the genetics laboratory maximized detections of individuals while remaining cost-effective. We used the same approach here. All hair samples were subjected to a three-phase process of first pass, cleanup, and error-check (Paetkau, 2003(Paetkau, , 2004 to establish an eight-locus marker system common to the Rocky Mountains (Paetkau, Calvert, Stirling, & Strobeck, 1995;Sawaya et al., 2012).

| Spatially explicit capture-recapture
Spatially explicit capture-recapture assumes the probability of detection is a decreasing function of the distance between an animal's home-range center and the rub object and parameterizes the following: g 0 is the probability of detection if the trap is located at the animal's home-range center; sigma (σ) is a parameter that describes the spatial scale over which capture probability declines (Efford, 2004). Instead of g 0 , we used lambda 0 (λ 0 ), the cumulative hazard of detection (expected number of detections per unit time at a location and time) which has a more direct relationship with homerange activity than probability. The equation relating λ 0 and g 0 is where g is the probability of detection and d is the distance between trap location and an animal's home-range center (Efford, 2004;Efford, Borchers, & Byrom, 2009). We used a binomial distribution and a hazard half-normal function with a full likelihood to estimate density (D), σ, and λ 0 (Efford, 2004). All analyses were run in program R v.3.2.1 (R Development Core Team 2015) using package "secr" (Efford, 2016).
The area of integration sets the outer spatial limits for which home-range centers can be assigned and should encompass all individuals that could have been exposed to the trap array (Efford & Fewster, 2013;Efford & Mowat, 2014). For both males and females, we calculated the area of integration using three times the root pooled spatial variance (RPSV), which is a 2D measure of dispersion of detections around trap locations (Efford, 2016). We calculated RPSV for each sex and added the largest value for each sex (18 km for males, 13 km for females) as a buffer around the study area.
"Secr" models discretize continuous habitat by using a gridded mask, on which we built spatial models of density. We used a mask with spacing of 2.5 km between grid centroids. We completed a sensitivity analysis, which suggested that our area of integration and spacing were a good compromise between processing time and minimizing bias (minimal change in D or log-likelihood).
Each year, we associated each opportunistic sample with a 7 × 7-km cell centroid (Sawaya et al., 2012;Stetz et al., 2014). Like unstructured scat dog searches with non-fixed trap locations, we defined the spatial extent of the grid based on locations searched by technicians, landowners, and Fish and Wildlife Officers (Thompson, Royle, & Garner, 2012).
Each cell then became a trap location for opportunistic samples, allowing for both "0" and "1" data necessary for mark-recapture. Because we could not quantify search effort for opportunistic samples, we assumed a uniform observation process for encountering hair samples within each grid cell. We believe this to be justifiable because increased effort affects precision, but not accuracy of SECR estimates ( Because "secr" models are computationally intensive, we designed a 2-step modeling approach (Table 2). In step 1, we identified the most parsimonious observation model (i.e., σ and λ 0 ) while holding density constant (D ~ 1). In step 2, we used the most parsimonious observation model as a base model on which to build a full model (i.e., D, σ, and λ 0 ; Table 2). For step 1, we created 17 single-session models for each sex in each year that differed in factors affecting σ and λ 0 ( Table 2). While rub trees offer no lure or bait, there is a potential for individual rub objects to be favored and for individual bears to exhibit variation in rubbing behavior. If rubbing is related to dominance or breeding (Clapham, Nevin, Ramsey, & Rosell, 2012;Lamb et al., 2017), rubbing may be influenced by other bears. The trap-specific behavioral response (bk) allows for a step change after first detection of an individual at a site. We used trap type as a covariate for both σ and λ 0 . We assumed variation in cumulative hazard rates of detection among trap types (rub: trees, power poles, fenceposts, fence: fencelines, and opp: opportunistic). Because bears interact with fencelines and rub trees differently, we would expect to see differential space use between trap types (i.e., a sampling effect).
A bear's decision to rub could be influenced by the bear that rubbed previously (Clapham et al., 2012;Lamb et al., 2017).
Because grizzly bears are dominant over black bears (Herrero, 1978) and can be displaced via interspecific competition (Holm, Lindzey, & Moody, 1999), we created a time-varying index of grizzly bear detection (GB; 1 = detected during previous occasion, 0 = not detected during previous occasion) at each rub object for each sampling occasion (grizzly bear data from Morehouse & Boyce, 2016). Bear use of rub objects varies seasonally , which can influence detection probabilities. We included time trend (T) as a covariate for σ and λ 0 . Last, tree cover provides security to black bears and cover type could influence detection. We included a singular habitat covariate with seven levels: deciduous, coniferous, shrub, grassland, agriculture, mixed forest, barren (30-m resolution; Crown Managers Partnership).
We would expect deciduous and coniferous cover to positively influence detection.
In step 2, we wanted to identify spatial drivers of black bear density. We created eight a priori single-session models for each sex in each year (Table 2). Our objective was to identify variables that best explained black bear density, not all possible covariates.
Normalized Difference Vegetation Index has been correlated with net primary productivity, leaf area index, carbon assimilation, and evapotranspiration (Pettorelli et al., 2011), and has been associated with grizzly bear habitat selection (Northrup et al., 2012).
Bare soil, clouds, and concrete correspond to low or zero NDVI val- For all density covariates, we used the "addCovariates" function, which is a spatial point extraction using the x-y coordinates of the mask grid (Efford, 2016). We used Akaike information criterion corrected for small sample sizes (AICc) to identify the most parsimonious models (Burnham & Anderson, 2002).
Our sampling period extended from June to early November, which is long relative to other NGS bear studies . SECR models assume stationarity in home-range centers and demographic closure, and we would likely be violating this assumption because black bears will increase movements in the fall to acquire enough food resources for winter dormancy. However, SECR models are robust to the closure violation (Efford & Fewster, 2013;Obbard, Howe, & Kyle, 2010). Efford and Mowat (2014) described an inverse and compensatory relationship between σ and λ 0 (or g 0 ), with negligible effect on density. During exploratory modeling, we ran early-season only and full-season models using our grizzly bear data (Morehouse & Boyce, 2016), and while σ and λ 0 changed, density did not.

| Resource-selection functions
To estimate density using RSFs, we (a) defined the area of inference; (b) calculated RSFs for male and female black bears in WLNP, the reference area; (c) associated abundance (N) with habitat selection in the reference area; (d) extrapolated N across the remaining area of inference and calculated density.
We anticipated that rub object locations were biased because surveys were primarily limited to linear features. To quantify this bias, we compared habitat covariates associated with all rub object locations to random locations. We defined the study area as a minimum convex polygon (MCP) bounding all unique rub object locations (n = 873). We buffered the MCP by 2.4 km, so random points could TA B L E 2 Step 1 and 2 candidate SECR models for black bears in southwestern Alberta. We used a hazard half-normal detection function for all models Step number

Model number
Model description  where we detected each sex (used) anytime in 2013 and 2014, to the full set of unique rub objects (available). Although these data were derived from the same dataset used for "secr," RSF data are simple presence/available data, whereas time-varying data were used for "secr." Thus, "secr" and RSF datasets are structured differently, making this exercise possible. We used the same covariates described for the global RSF, as well as grizzly bear use (GBU) which is the sum of unique grizzly bear detections at each rub object, divided by the number of sample occasions each rub object was visited. We used AIC to identify the most parsimonious model for each sex among 11 candidate models (Supporting Information Table S1). We calculated RSFs only within the area of inference (Supporting Information Figure S1). where w(x i ) is the midpoint probability for an RSF bin and A(x i ) is the area for a vector of i habitat variables. For the ith habitat class, we calculated the expected number of bears as N i =N × U(x i ) where N is the estimated population size for WLNP, and density is Fourth, based on density and habitat associations in WLNP, we extrapolated D across our area of inference. We approximated confidence intervals (CI) by extrapolating based on the 95% CI of the abundance estimate for GNP. We used k-fold cross-validation to measure the predictive ability of the RSF (Boyce, Vernier, Nielsen, & Schmiegelow, 2002). We partitioned the data into 10 folds and tested the association between the frequency of presence observations in 10 RSF bin ranks. We did this 10 times and used the averaged Spearman's rank correlation coefficient (r s ) to evaluate the predictive success of each RSF model .

| Hair collection
In 2013

| SECR models
The top detection model from step 1 for males in 2013 and 2014 included λ 0 covariates traptype, bk, and T (Table 3). The σ covariate included T. The top density models for males from step 2 in 2013 and 2014 allowed density to vary by land tenure, with density lowest on Crown lands and highest on protected lands (Table 3).
Relative standard error (RSE) of the density estimate was 8.  Table 4). Detection varied by trap type (Table 4).
The top detection models from step 1 for females in 2013 and 2014 included λ 0 covariates trap type and bk, and σ covariate trap type (Table 4). The top density model for females from step 2 in 2013 included harvest. There was a negative relationship between harvest and black bear density for females (β hunt = −0.27, SE = 0.07; Figure 2,  Figure 2).
For each sex and year, we predicted the density surface at each mask point and used discrete summation to calculate abundance within each land tenure. Abundance estimates indicate female-biased sex ratios on private land (2.3F:1M; Table 5).

| Resource-selection functions
The global RSF indicated rub objects were in areas with high NDVI values, low to mid-elevations, and not in agricultural areas such as cropland and year-round cattle pastures. When the lowest three RSF bins were excluded, the area of inference was reduced to 2,364 km 2 (Supporting Information Figure S1). We removed the lowest three bins because there was a large break between the third and fourth lowest bin, which we interpreted as a natural cutoff for defining the area of inference.  Figure S2).

| D ISCUSS I ON
We generated the first density and abundance estimates for black bears in southwestern Alberta. From SECR models, land tenure best explained spatial variation in male and female black bear density, except females in 2013 where harvest was the best predictor covariate for density. Land tenure was also an important predictor covariate for RSF models. A large-scale covariate like land tenure can be confounded because it may be describing more than just "land tenure." For example, land tenure encompasses several habitat differences such as road density, with protected lands having the lowest road density and private land having the highest road density. However, support for this covariate might be related to the multiplicative effect of road density and harvest intensity, which could explain why the individual covariates "roads" and "harvest density" did not perform better than land tenure (except for 2013 females).
For RSF-derived estimates, male and female density was highest in the reference area, which was consistent with results from SECR modeling. As well, female densities were higher than male densities, meeting our original predictions. On average, male densities in the protected lands were 3.1 times and 2.1 times higher than Crown and private lands, respectively; female densities in the protected lands were 3.4 and 2.3 times higher than Crown and private lands, respectively. While some protected areas in North America, particularly mountain parks, do not contain high-quality habitats (Jenkins et al., 2015;Joppa & Pfaff, 2009), our results suggest that protected lands in our study area contain high-quality habitats because RSF scores and relative densities were highest there. Both RSF and SECR models point to land tenure as the most important predictor of black bear density, followed by harvest density for SECR models, and habitat productivity and recently burned areas for RSF models. Both males and females showed avoidance of recently burned areas, though female avoidance was not significant. Wildfires can produce high-quantity and high-quality bear foods; primary bear foods such as buffalo berry (Shepherdia canadensis), Saskatoon berry (Amelanchier alnifolia), and species of Vaccinium can produce up to 20 times more fruit when comparing adjacent burned and unburned mature forests (Hamer, 1996;Young & Beecham, 1986). Recovery from wildfire depends on site productivity, precipitation levels following the burn, forest type, and intensity of the wildfire, among other factors. Fire can have prolonged negative effects on forest cover, for example, which is important escape terrain for black bears. In turn, black bears have been shown to avoid large, high-intensity burned areas with little protective cover from sympatric grizzly bears, which select for relatively treeless burned areas (McLellan, 2011). Even in areas without grizzly bears, wildfires can have demographic consequences on black bears beyond direct mortalities (Singer, Schreier, Oppenheim, & Garton, 1989), such as reduced cub survival and sex ratios skewed toward males (Cunningham & Ballard, 2004). In our study area, the wildfire that we suspect is driving our top RSF models was a 177-km 2 high-intensity fire that burned on Crown lands in 2003. The Lost Creek Fire continues to have little protective tree cover for black bears, and with high grizzly bear densities for interior populations (Morehouse & Boyce, 2016), this may be a contributing factor to reduced black bear densities on Crown lands.
Sex ratios were strongly female-biased on private lands, while not so on Crown and protected lands (Table 5). We did not predict that there would be sex-specific spatial structuring in our study.
Black bears, and large mammals in general, often exhibit a female bias in un-hunted populations (Clutton-Brock & Iason, 1986), and female black bears drive population growth because one male can impregnate many females (Beston, 2011). Hunting can exacerbate this bias, particularly for bears, where males are disproportionately harvested because of hunter selection for increased body size, legal protection for females with cubs, and larger home-range size for males (Bunnell & Tait, 1980;Garshelis, 1990;Miller, 1990). In our study area, we speculate that skewed sex ratios are a result of females emigrating out of protected lands that have a high density of male black bears, and to an area with lower female harvest relative to Crown lands.
Females will select habitats to minimize predation risk to their offspring (Ruckstuhl & Neuhaus, 2002) or can be excluded from high-quality habitats by males (Craighead, Sumner, & Mitchel, 1995). Because protected lands have the highest male black bear densities of all land tenures, females might be dispersing to areas with a lower density of males where competition for high-quality resources is diminished. In our study area, black bears can be harvested on both Crown and private lands, but we saw higher female harvest on Crown lands ( Figure 6). In Alberta, reporting of black bear harvests is not required and we assume harvest rates re- AICc: Akaike information criterion corrected for small sample sizes; GB: grizzly bear; K: number of model parameters; LL: log-likelihood.
In step 1, we identified the top λ 0 and σ covariates. In step 2, we used the step 1 model as the base model on which to build heterogeneous density models. Models that did not receive any model weight (w i = 0) are not shown here. See Section 2 for variable definitions.
In our study, we found differences in sex ratios between land tenures and this could have long-term consequences for black bears. Because Crown lands have higher harvest rates for black bears, and if there are low recruitment rates, we could expect that black bears on Crown lands would contribute less to population growth (Novaro, Funes, & Walker, 2005). In Alberta, Crown lands are often considered "core" black bear habitats (Webb, Morcos, Allen, & Frame, 2016) and wildlife managers trust these areas to be population sources, rather than population sinks, albeit without empirical data on population trend, abundance, or density. Cautious interpretation is required of our results, however, because without knowing population age structure, differences or changes in sex ratios could be misinterpreted as a population decline, for example, when the population is dominated by a specific age cohort (e.g., subadults; Garshelis, 1990). Despite this caveat, our results indicate spatial structuring of mortality and we suggest further monitoring to assess demographic consequences of high harvest. Improvement to provincial harvest reporting (e.g., required reporting) would help to gain insight, particularly for non-licensed hunting on private land.
Animal densities are usually related to habitat selection , and RSFs can be used to describe this relationship (Johnson et al., 2006;Manly et al., 2002). Our SECR and RSF-derived abundance produced density estimates with generally overlapping confidence intervals (with exceptions). We draw some common conclusions from the concurrent modeling. First, RSF-derived density estimates had smaller 95% CIs than the SECR density estimates. This likely stems from the high variance of ratio estimators (Czaplewski et al., 1983) such as capture-recapture estimators, and the additional parameterizations of σ and λ 0 . RSFs do not account for imperfect detection; they only compare used locations to locations where an animal could have been. In contrast, SECR accounts for animals we did not detect by estimating un-observed bear home-range centers. The accuracy and precision of SECR abundance and density estimates depend on the ability to model factors influencing σ and λ 0 (Whittington & Sawaya, 2015). Thus, it is reasonable that with low cumulative hazard of detection, SECR models would generate larger variance estimates than the RSF models. to identify spatial covariates driving black bear density. Using two parallel methods allowed us to explore differential drivers of density.
While we advocate for the use of SECR models for density estimates, it is not our intent to undermine the utility of RSF models. Many wildlife agencies focus on creating habitat-based models, such as RSFs, which are useful in identifying high-quality or critical habitats. For example, a variation on the RSF-abundance extrapolation was used in British Columbia to estimate grizzly bear abundance (Fuhr & Demarchi, 1990).
In our study, spatial variation in density between RSF and SECR methods were consistent, generally protected lands had highest density and Crown lands had lowest density. Our results reinforce the importance of habitat and land use in estimating population size, even for a generalist species such as the black bear. Further, the reduced laboratory costs for only identifying males and females used in the RSF (vs. identifying individuals for SECR) could be attractive to agencies and institutions with restricted budgets. We acknowledge that the RSF-based method requires an independent abundance estimate for the reference area, and depending on the species, time, distribution, and habitat type(s), these estimates might not exist or be suitable for extrapolation.
Black bear monitoring studies are often spatially and temporally isolated (Beston, 2011). With recent abundance and density estimates for GNP (Stetz et al., 2014), our study adds demographic information to a shared population of black bears but on a multi-use landscape.
The only previous estimate for black bear abundance in southwestern Alberta (Gunson & Markham, 1993) was derived without variance estimates, which precludes comparison and inferences about population changes (Miller, 1990). As a coarse-level comparison, our estimates were in the range of reported interior black bear densities where sympatric with grizzly bears (mean = 164 bears/1,000 km 2 ), although the range of densities is high (range = 9-450/1,000 km 2 ; Mattson, Herrero, & Merrill, 2005). While our black bears density estimates have low relative bias (RSE ranged from 8.5% to 13.5%), precision could be improved with a secondary data source (e.g., hair traps; Boulanger et al., 2008), particularly for females where RSEs were higher than for males. This likely stems from a lower cumulative hazard of detection for females relative to males.
If private land in our study area is acting as a spatial refuge for female black bears, our results suggest the need for active management on Crown lands where black bear densities are lowest. In particular, harvest and management differences by land tenure, such as road densities, should be targeted. For example, grizzly bear densities were higher in British Columbia where motorized vehicle access was restricted (Proctor et al., 2018).

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R S' CO NTR I B UTI O N S
AEL and ATM conceived the ideas and designed methods; AEL and ATM collected the data; AEL analyzed the data; AEL wrote the manuscript; MSB supervised research and analysis.
F I G U R E 5 Spatially explicit capture-recapture (SECR) and resource-selection function (RSF)-derived densities for number of male and female black bears in the southwestern Alberta in 2013 and 2014. For RSF densities, reference area densities were extrapolated from Glacier National Park (Stetz et al., 2014). Error bars represent 95% CI. Density is reported in bears/1,000 km 2 F I G U R E 6 Harvest density (individuals/1,000 km 2 ) for male and female black bears the year prior to non-invasive genetic sampling in southwestern Alberta (2013)(2014). Wildlife management unit (WMU) 400 is on Crown land and WMU 303 and 302 are on private land Black bear rubbing on a tree in southwestern Alberta, Canada. Photo credit: Ryan Peruniak