INTRODUCTION
Knowledge of habitat selection by wildlife populations is central to environmental mitigation of industrial developments affecting ecological systems. Development of radiotelemetry technologies allows collection of substantial amounts of animal-use data (Cagnacci et al. Reference Cagnacci, Boitani, Powell and Boyce2010), while high-resolution aerial and satellite imagery enable improved representations of landscape complexity. Coupling Geographic Information Systems (GIS) with statistical modelling provides a powerful framework to inform management and conservation (Strickland & McDonald Reference Strickland and McDonald2006).
Resource selection functions (RSFs) have been widely used in studies of wildlife habitat selection (Boyce & McDonald Reference Boyce and McDonald1999; Manly et al. Reference Manly, McDonald, Thomas, McDonald and Erickson2002), but their predictive accuracy is dependent on the quality of animal-use data and input habitat layers (Morehouse & Boyce Reference Morehouse and Boyce2013). In addition, scale can influence RSF outputs (Boyce Reference Boyce2006), and modelling habitat selection is not straight-forward, particularly for wide-ranging, rare species with high inter-individual variability (Cristescu & Boyce Reference Cristescu and Boyce2013; Nielsen et al. Reference Nielsen, Shafer, Boyce and Stenhouse2013). With many conservation decisions requiring spatially-explicit baseline information for comparison of trends and impacts, RSFs could potentially provide a framework for predicting relative probability of animal response to land-use change along with identifying the direction of the response. However, such methods are difficult for dynamic landscapes that are characteristic of industrial sites (Johnson & Boyce Reference Johnson and Boyce2004; Johnson et al. Reference Johnson, Boyce, Case, Cluff, Gau, Gunn and Mulders2005), and are possibly site-specific. Further testing is required to assess the utility of using these statistical tools for predicting changes in habitat selection under conditions of changing landscapes at appropriate scale.
Surface (open-pit) mining provides an extreme example of landscape change as a result of human activity. While habitat is an important consideration in mine reclamation planning, more knowledge is required on the effects of mining on wildlife populations during and post mining, particularly for wide-ranging mammals of conservation concern. Most current knowledge on response of large/medium-bodied mammals to mining comes from studies on ungulates, which show varying response to mines ranging from avoidance to selection (Merrill et al. Reference Merrill, Hemker, Woodruff and Kuck1994; Bristow et al. Reference Bristow, Wennerlund, Schweinsburg, Olding and Lee1996; Weir et al. Reference Weir, Mahoney, McLaren and Ferguson2007; Bleich et al. Reference Bleich, Davis, Marshal, Torres and Gonzales2009; Blum et al. Reference Blum, Stewart and Schroeder2015). The effects of mining on omnivores and carnivores remain largely unknown despite expansion of this industry in the distribution ranges of many of these species. For example, in Alberta, Canada, mining occurs within the grizzly bear's (Ursus arctos) range, with the species threatened primarily due to human-caused mortality (AGBRT 2008). Grizzly bear population persistence is dependent on availability of suitable habitats that provide sufficient foods and are safe from humans (Nielsen et al. Reference Nielsen, Stenhouse and Boyce2006). Industrial activities can change the spatial distribution of foods potentially placing bears at risk, for example, if natural foods become available close to humans (Roever et al. Reference Roever, Boyce and Stenhouse2008) or if bears seek human-sourced foods on industrial sites (McLellan Reference McLellan1990). Reproductive classes and individual bears can vary in their behaviour in relation to extractive industries such as logging (Roever et al. Reference Roever, Boyce and Stenhouse2008) and oil and gas (Laberee et al. Reference Laberee, Nelson, Stewart, McKay and Stenhouse2014).
Management of grizzly bears and other omnivores or carnivores relies on habitat modelling often at large scales such as province, state, region or management area (Nielsen et al. Reference Nielsen, Stenhouse and Boyce2006; Mace et al. Reference Mace, Waller, Manley, Ake and Wittinger2008). Such scales can be appropriate for broad assessments but might not adequately incorporate finer-scale, site-specific conditions, particularly for highly dynamic landscapes such as those associated with mines. Anticipating grizzly bear response to active mining and mine closure can facilitate land-use planning that will allow persistence and possibly enhancement of bear populations on industrially-modified landscapes (Johnson & Boyce Reference Johnson and Boyce2004; Johnson et al. Reference Johnson, Boyce, Case, Cluff, Gau, Gunn and Mulders2005).
We used empirical data and a GIS framework to illustrate the application of habitat selection modelling for understanding wildlife response to industrial disturbance. We assessed grizzly bear habitat selection, differentiating individuals within reproductive classes (males, females, females with cubs) and comparing selection between active mining and post mining. We expected low selection of actively mined areas due to operational disturbances, and higher selection following mine closure because of reclamation to wildlife habitat. We also investigated potential differences in habitat selection inside and outside mines, using models derived at local and regional scales. If patterns of habitat selection were similar between scales, this would suggest that broad-scale models can be sufficient for site-specific management. We expected bears to avoid active areas or their proximity because of disturbance and to also avoid inactive areas due to lack of bear foods. Reclaimed (grassland) areas were expected to be used in proportion to availability as a trade-off between attractiveness of herbaceous foods planted as part of reclamation and potential risks associated with little hiding cover. We anticipated that bears would select undisturbed areas as these represented natural habitats.
METHODS
Study area
The study area was located in west-central Alberta at the boundary between the eastern slopes of the Rocky Mountains and Foothills. The predominant natural vegetation cover is conifer forest, with mixed conifer–deciduous forest at lower elevations. Shrub cover is found above tree line (~1800 m) and along river corridors with barren lands occurring naturally at elevations exceeding shrub range. The area includes Luscar and Gregg River open-pit metallurgical coal mineral surface leases (MSLs) and a 1 km buffer around these leases (Fig. 1). The spatial extent reflects the size of the mining zone of influence considered locally in land-use planning as well as the 4-h step length (distance between two consecutive GPS relocations) of collared bears in the study (mean ± SE: 1008 ± 63 m; see ‘Grizzly bear data’ section). Public access is strictly regulated on MSLs and occurs along designated motorized and non-motorized trails. For either trail type, human-use data were not collected for this study.
Figure 1 Study area for grizzly bear habitat selection on and near open-pit mines in west-central Alberta, Canada. The area includes two MSLs and adjacent 1 km buffer. Dark gray shading indicates forest whereas light gray shading indicates barren land, grass or shrub. Satellite photo is a Landsat orthoimage, year circa 2006.
Data were partitioned chronologically into two sampling periods to reflect changing landscape conditions and bear responses during (1999–2003) and post (2006; 2008–2010) mining. During-mining industrial operations involved removal of natural vegetation and soil, blasting to create coal extraction pits, mechanized shovelling to extract coal and overburden, as well as haul-truck traffic. Post-mining land reclamation efforts and related activities included sloping, soil placement and seeding. Reclamation activities occurred over shorter time spans and with smaller machinery than active mining and did not involve blasting. As mining did not occur simultaneously over the extent of MSLs and some areas remained undisturbed, the mined landscape consisted of active (mining ongoing), inactive (previously active; mining stopped), reclaimed or unaltered patches. We recognize that pooling data across years within each sampling period might not capture variability in mining activity within a given year. Partitioning was based on best available habitat and bear data (see below).
Grizzly bear data
Grizzly bear GPS radiocollar data were collected by the Foothills Research Institute Grizzly Bear Program (Hinton, Alberta, Canada) and the University of Alberta (Edmonton, Alberta, Canada). Capture methods for radiocollar deployment included darting from a helicopter, culvert trapping and leg-hold snaring and were approved by the University of Saskatchewan and University of Alberta. GPS radiocollars used were Televilt GPS-Simplex (Televilt, Sweden), Telus UHF (Followit, Sweden) and ATS (Advanced Telemetry Systems, USA). After setting the constraint of a minimum of 50 telemetry relocations per individual bear within the study area in each sampling period (Leban et al. Reference Leban, Wisdom, Garton, Johnson, Kie, Millspaugh and Marzluff2001), seven adult bears (nmales = 2; nfemales = 3; nfemales with cubs = 2) were included in analyses during active mining, and nine adult bears after mine closure (nmales = 2; nfemales = 4; nfemales with cubs = 3). Bear G040 was designated as having cubs in certain years, and solitary in other years, with a similar switch in reproductive class for bear G023. Considering that the density of this threatened grizzly bear population is 4.79 individuals/1000 km2 (Boulanger et al. Reference Boulanger, Stenhouse, Proctor, Himmer, Paetkau and Cranston2005), we were able to monitor with radiocollars a large proportion of the bears. All radiocollar data were rarefied to 4-h collar fix rate to minimize potential bias related to bear habitat selection at multiple scales (Ciarniello et al. Reference Ciarniello, Boyce, Seip and Heard2007). The final dataset contained 1291 relocations during mining and 2514 locations after mine closure (Table S1).
Study design and variables
The study followed a use-available design (Johnson et al. Reference Johnson, Nielsen, Merrill, McDonald and Boyce2006), wherein habitat features at bear GPS fixes (‘use’ locations) were compared with those at random locations (‘available’ locations). Available locations were generated at a density of 30 spatially-referenced points/km2 (Northrup et al. Reference Northrup, Pitt, Muhly, Stenhouse, Musiani and Boyce2012) at two spatial scales: individual bear annual home range, and study-area extent. Sampling intensity was higher for home-range areas that were used repeatedly across years. Home ranges were delineated using the least squares cross-validation procedure for fixed-kernel home ranges, based on 95% of GPS locations from each bear, and clipped to study-area extent.
Grizzly bear habitat-related variables (Table 1) were available in GIS format from the Foothills Research Institute Grizzly Bear Program, Teck and Alberta Environment and Sustainable Resource Development. We updated layers to reflect annual changes in landscape features associated with mining development, based on interpretation of orthorectified aerial photography (2001; 2004; 2007; 2010) and a SPOT image (2004). Land-cover categorization (30 m × 30 m) was reclassified to forest, shrub, grassland and barren land. We calculated distance to the nearest water course and habitat edge, defined as the boundary between forest and another land-cover type. Mining-specific covariates included a dummy variable (1/0) for tree island (original tree patch left undisturbed during mining, area range = 118−307 600 m2); distance to nearest mining pit high wall (designed to represent bighorn sheep escape terrain [MacCallum & Geist Reference MacCallum and Geist1992]); distance to the nearest active mine haul road; and distances to different mine disturbance types including active, inactive and reclaimed land. We also calculated distance to the nearest public road, motorized trail and non-motorized trail.
Table 1 Predictor variables included in RSF analyses for grizzly bear habitat selection on and near Luscar and Gregg River open-pit coal mines in west-central Alberta, Canada.
GIS procedures were carried out in ArcGIS v.9.2 (ESRI, Redlands, USA), using the Spatial Analyst Extension, Home Range Extension for ArcGIS (Rodgers & Carr Reference Rodgers and Carr2007), and Hawth's Analysis Tools for ArcGIS (Beyer Reference Beyer2004).
Statistical modelling
We contrasted habitat features at grizzly bear GPS radiocollar and random (available) locations using logistic regression, with the binary response variable (1/0) coding bear use and available locations. The model structure was,
$$\begin{equation}
w\left( x \right) = {\rm{exp}}({\beta _1}{x_1} + \ {\beta _2}{x_2} + \ldots \ {\beta _n}{x_n})
\end{equation}$$
where w(x) is the relative probability of selection by a bear based on predictor variables xi
and estimated coefficients β
i
, with
$i = \ \overline {1,\ n} $
. Following the information-theoretic approach to model selection (sensu
Burnham & Anderson Reference Burnham and Anderson2002), we created 26 a priori candidate models to predict grizzly bear habitat selection on and in the immediate vicinity of MSLs. The models reflected hypotheses for grizzly bear habitat selection, including broad habitat characteristics, ungulate and herbaceous foods, mining-specific features and human access (Table S2). We constructed a correlation matrix for all independent variables and to minimize collinearity we excluded highly correlated variables (|r| > 0.7) from the same model structure. For all continuous variables we tested the influence of non-linearities on model maximum log-likelihood estimation by including squared terms for covariates representing distance metrics. We used robust standard errors in STATA v.11.2 (StataCorp, College Station, USA) to account for heteroskedastic distribution of regression error terms. To assess model adequacy, we computed percentage deviance explained (hereafter, DE) for each model.
We estimated separate models for each bear during and post mining as delineated by closure of mining operations. We did not average models because covariate combinations generally varied between top models among bears in the same reproductive class and sampling period, as well as between supported models (∆AICc < 2) for each bear (Cade Reference Cade2015). The coefficients for the top models for each bear were used to generate predictive RSF surfaces at the home-range and study-area availability scales, separately for during and post mining. Prior to mapping, predicted relative probabilities were transformed in GIS using:
$$\begin{equation}
Tw\left( x \right) = \frac{{w\left( x \right)}}{{1 + w\left( x \right)}}
\end{equation}$$
where w(x) is the prediction from eqn (1).
Predictive accuracy
To assess predictive accuracy of top models for each bear we used k-fold cross-validation (Boyce et al. Reference Boyce, Vernier, Nielsen and Schmiegelow2002). In this approach k was equal to n – 1, with n representing the number of individual bears of a given reproductive class and sampling period (Wiens et al. Reference Wiens, Dale, Boyce and Kershaw2008). Logistic regression coefficients were estimated iteratively for k subsets of data withheld for model training, across all predictor variables present in the top model for the corresponding reference bear. The β i estimates obtained from model training were used in eqn 1 in conjunction with predictor variables xi corresponding to the reference bear (model testing). This framework allowed us to assess how population-level data could be used to predict relative probability of selection by any one bear.
Site-specific and regional RSF predictions
Predictions were compared within vs. outside MSLs, as constrained by the 1 km buffer. We did not test for differences in predictions between these areas because we considered all RSF values (i.e., we did not sample). We thereby simply compared the estimates from the complete pixel census by plotting the mean RSF values for each reproductive class, sampling period and availability scale.
We further contrasted our results with those from regional-scale grizzly bear habitat selection models for pooled adult females with/without cubs, which included our study area extent (Nielsen Reference Nielsen2005). Our site-specific models rendered annual relative probabilities of habitat selection. The regional models were, however, seasonal (3×) and we included all seasons for comparisons. Because regional models were based on data collected during active mining at the home-range scale of availability, comparisons with fine-scale models were restricted to this sampling period and scale of availability. All regional surfaces were clipped to study area extent and we applied a similar procedure to the one for our site-specific models to assess differences in mean RSF scores within vs. outside MSLs.
RESULTS
There was substantial variability among individual bears, with AICc support received for diverse hypotheses related to broad habitat characteristics, ungulate and herbaceous distribution, mining-specific features and human access (Appendix S1). We report main patterns focusing on instances where selection was consistent in terms of sign of parameter estimate and confidence intervals not overlapping zero for at least 50% of monitored individuals within a reproductive class.
Habitat features
Barren land was avoided by males and solitary females as well as by females with cubs during but not post mining (Tables 2 and 3, and Appendix S2). Shrub was avoided by males and solitary females post mining but selected by these reproductive classes (at study-area availability scale for males) during mining. Grassland was selected by males and females with cubs during mining. Males selected areas of intermediary ruggedness. Males and solitary females selected edge proximity or areas far from edges after mining, whereas females with cubs selected such areas during mining. Males selected intermediary distance to riparian areas during mining.
Table 2 Individual grizzly bear habitat selection on mined lands and adjacent area, west-central Alberta, Canada, with availability drawn at home-range level. Coefficients for which the confidence intervals did not overlap zero have an asterisk (*). No reporting of coefficients refers to the specific variable(s) not being included in the top model. Forest land-cover type was withheld as reference category. During mining: 1999–2003; after (post) mining: 2006 and 2008–2010. ‘+’ corresponds to positive selection coefficient whereas ‘–’ denotes negative selection. Note that for distance variables ‘–’ implies selection of areas close to the respective feature.
Table 3 Individual grizzly bear habitat selection on mined lands and adjacent area, west-central Alberta, Canada, with availability drawn at the study-area level. Coefficients for which the confidence intervals did not overlap zero have an asterisk (*). No reporting of coefficients refers to the specific variable(s) not being included in the top model. Forest land-cover type was withheld as reference category. During mining: 1999–2003; after (post) mining: 2006 and 2008–2010. ‘+’ corresponds to positive selection coefficient whereas ‘–’ denotes negative selection. Note that for distance variables ‘–’ implies selection of areas close to the respective feature.
Mine features
Females with cubs selected tree islands on mined land after mining (Tables 2 and 3, and Appendix S2). Males and solitary females (the latter at home-range availability scale) selected areas either close or far from high walls after mining, with proximity to walls selected by females with cubs during mining at home-range availability scale. During mining, males selected areas at intermediary distances from mine haul roads (home-range availability). After mining, solitary females and females with cubs also selected intermediate distances from haul roads (study-area availability).
Human access
Males selected areas close or far from public roads during mining (home-range availability) but intermediary distances to public roads after mining (Tables 2 and 3, and Appendix S2). Solitary females selected areas far or at intermediary distances from public roads after mining (study-area availability). Areas close or far from public roads were selected by females with cubs during mining (home-range availability). Males during mining and solitary females after mining (the latter at home-range availability scale) selected areas near or far from motorized trails. However this pattern switched for solitary females after mining when study-area availability was considered, with selection for intermediate distance to motorized trails apparent. After mining, males avoided non-motorized trails, whereas solitary females selected areas at intermediary distance from such trails. Such intermediary distances were also selected by females with cubs during mining.
Ranking of both home-range and study-area availability models showed that top models for post mining generally received more substantial support (w AICc > 0.9) compared to during mining models. Post-mining models also explained more deviance than models from active (during) mining (Table S3). Models had variable predictive accuracy, with substantial differences occurring in relation to bear reproductive class and sampling period (Appendix S3 and Fig. S1). Differences in predictive accuracy within reproductive class in relation to scale of availability were most pronounced for males and solitary females. For these two reproductive classes, home-range availability models had more distinctive predictive differences between sampling periods compared to study-area availability models. Overall, habitat models for solitary females (during mining) and males and females with cubs (post mining) had the highest predictive power for selection (eqn 1) across all reproductive classes and sampling periods.
Mean RSF scores on vs. outside MSLs differed within reproductive class irrespective of scale of availability (Fig. 2). Males and solitary females had higher mean RSF values within a 1 km buffer outside MSLs compared to inside MSLs during mining with the opposite pattern observed post mining. Females with cubs, however, had consistently higher RSF scores within MSLs regardless of sampling period (Fig. 2).
Figure 2 Mean relative probability of grizzly bear habitat selection inside (I) and outside (O) coal MSLs in west-central Alberta, Canada. Reproductive classes include males (△), solitary females (⋄) and females with cubs (□). Filled symbols correspond to during mining whereas symbols with no fill represent post mining. Predictions are based on study area – specific availability drawn at the home-range (a) and study-area level (b); and on regional scale availability drawn at the home-range level for three separate seasons (c; Nielsen Reference Nielsen2005). For regional availability, RSF calculations pooled females and females with cubs, and regional RSF surfaces were only available during mining.
Mean RSF scores for the regional models (during mining) also differed, being higher within a 1 km buffer outside MSLs compared to inside MSLs (Fig. 2). One exception occurred for males in the third (fall) season, when RSF scores were similar on MSLs and the 1 km buffer. The spatial location of high RSF score pixels showed major differences between regional and fine-scale models. In particular, the northern block on Gregg River MSL, northern sector of Luscar MSL and extreme south-eastern sector of the latter MSL were highly selected by males and some solitary females (Fig. 1 and Appendix S4) based on the fine-scale home-range during-mining models, as compared to the regional models. Differences between spatial predictions were even more obvious for females with cubs, which based on the top fine-scale home-range during-mining models selected the southern block on Gregg River MSL, central, central-northern and eastern sectors of Luscar MSL (Fig. 1 and Appendix S4), in contrast with regional model predictions.
DISCUSSION
We illustrated how habitat features on a dynamic landscape can be linked to animal telemetry data across different time periods, enabling spatial visualization of wildlife habitat selection. The RSF approach enables retrospective assessments of animal response to landscape change and the mapping outputs are informative for land-use planning and environmental impact mitigation.
Mining activity at the two mines under study created variance in behavioural response of bears. Higher selection of mining areas by males and solitary females after mine closure suggests that active mining deters these reproductive classes, but individuals are able to colonize reclaimed mines in our study system. Selection of MSLs during and post mining by females with cubs could be a reflection of potentially avoiding males at the active mining stage, and trading security for forage once mines were reclaimed. Although human-caused habitat disturbance and fragmentation have been suggested to affect the ability of females with cubs to elude infanticidal males (Fernández-Gil et al. Reference Fernández-Gil, Swenson, Granda, Naves, Perez, Dominguez, Ordiz and Delibes2010), no infanticide data were available for our study system. Alternatively, selection patterns documented for females with cubs could be a result of increased tolerance of human activities associated with mining, to acquire foods in mining areas.
Barren land avoidance by all reproductive classes is likely connected to scarcity of grizzly bear foods on this land-cover category. Differences in selection within the shrub land class before and after mining are possibly related to selection of secure cover provided by shrubs during mining and less need for such cover following mine closure. Selection of grassland by males and females with cubs probably occurred because reclaimed herbaceous material provides forage (Cristescu et al. Reference Cristescu, Stenhouse and Boyce2015). Selection of areas of intermediary ruggedness by males also may be indicative of foraging, with Hedysarum spp. roots frequently associated with ruggedness (Hamer & Herrero Reference Hamer and Herrero1987). Edge selection across reproductive classes as well as tree island selection by females with cubs could be associated with forest proximity enabling escape to vegetation cover when encountering threats (Nielsen et al. Reference Nielsen, Boyce and Stenhouse2004a ), or opportunities for bear consumption of ungulates (Cristescu et al. Reference Cristescu, Stenhouse and Boyce2014). The predicted selection of proximity to high mine walls is likely related to the distribution of individuals within MSLs and may not necessarily reflect hunting of ungulates. Bighorn sheep use high walls on mines to evade predators (MacCallum & Geist Reference MacCallum and Geist1992) and this species forms only a small proportion of bear diet (Cristescu et al. Reference Cristescu, Stenhouse and Boyce2015).
Avoidance of public roads by males is in accordance with studies documenting grizzly bear response to roads (McLellan & Shackleton Reference McLellan and Shackleton1988). While females have been shown to select areas near public roads (Graham et al. Reference Graham, Boulanger, Duval and Stenhouse2010; Roever et al. Reference Roever, Boyce and Stenhouse2010), we showed that solitary females avoided roads whereas females with cubs selected proximity of these roads. Selection of roaded areas by females with cubs might be an outcome of sexual segregation manifested due to male avoidance (Wielgus & Bunnell Reference Wielgus and Bunnell1994). Selection of areas located at intermediary distance from active haul roads by all reproductive classes is a novel finding. The avoidance of non-motorized trails by all reproductive classes suggests a loss in grizzly bear habitat effectiveness associated with human recreation, which may become problematic if recreationists use MSLs extensively following mine closure. Although data on recreational activity levels along these trails were unavailable, carnivore avoidance of human-use trails has been documented (Muhly et al. Reference Muhly, Semeniuk, Massolo, Hickman and Musiani2011) and might relate to unpredictability of non-motorized traffic. Response of bears to motorized traffic was less conclusive, with such traffic possibly more predictable to wildlife because of long-range detection of engine noise.
Irrespective of scale of availability and bear reproductive class, post-mining habitat selection models had greater DE and received more support than during mining models. This finding suggests that modelling habitat selection for reclaimed mines is easier than modelling habitat selection on highly dynamic industrial landscapes such as those supporting active mining operations. Predictive accuracy varied between individuals within reproductive class possibly in connection to different life history strategies, which have been demonstrated for grizzly bears (Northrup et al. Reference Northrup, Pitt, Muhly, Stenhouse, Musiani and Boyce2012), as well as more broadly for other species (Bolnick et al. Reference Bolnick, Svanback, Fordyce, Yang, Davis, Hulsey and Forister2003).
We showed that broad-scale habitat modelling does not necessarily reflect site-specific conditions influencing grizzly bear habitat selection. Spatially, site-specific RSFs showed high selection for certain areas within MSLs during mining, with the largest difference compared to regional predictions recorded for females with cubs (Appendix S4). We suspect that differences between regional and fine-scale predictions are mainly caused by improved GIS layers and adequate model covariates representative of site-specific conditions. For example, much of the area within MSLs was classified as barren land in the regional-level land-cover layer, whereas our fine-scale land-cover layer included grassland, which is present on MSLs and grazed by bears following reclamation (Cristescu et al. Reference Cristescu, Stenhouse and Boyce2015). Furthermore, our model-ranking procedure suggested that fine-scale factors such as human access can be key predictors of bear habitat selection. Detailed knowledge of the area reflected in realistic GIS layers, as well as understanding of the target species’ ecology, are essential, especially when scientific outputs are to be used in environmental conservation and management decisions. We suggest that habitat modelling on highly dynamic industrial landscapes such as those with surface mining operations be carried out using best available fine-scale layers (see also Morehouse & Boyce Reference Morehouse and Boyce2013).
While our results are likely species and site specific, the approach outlined herein can be used as an example for future studies that aim to assess wildlife responses to mining and other industrial activities, enabling land-use planning strategies for wildlife conservation in industrially-modified landscapes. An important improvement for future studies would be to collect data prior to industrial disturbance, in addition to during and after operational activity, allowing understanding of wildlife response at multiple temporal phases. Sampling a large proportion of the target wildlife population is important and will affect results and interpretation. Our study has relatively small sample sizes especially within reproductive class. However given the low density of grizzly bears in the region (Boulanger et al. Reference Boulanger, Stenhouse, Proctor, Himmer, Paetkau and Cranston2005) we are confident that we monitored a substantial proportion of individuals.
Following reclamation, grizzly bears of all reproductive classes selected mine sites under study. The documented selection of tree islands by females with cubs emphasizes the importance of preserving natural habitat patches on mined areas to minimize long-term displacement and promote colonization of mines. Because reclaimed MSLs under study contain ungulate and herbaceous foods that attracted bears of all reproductive classes, such industrially-disturbed sites could potentially serve as local refugia for bears. However, not all open-pit mines may contain the abundance of foods that our study sites have.
As open-pit coal mines are reclaimed in this area, and these lands are opened for public use it would be important to undertake grizzly bear mortality risk assessments for identifying key areas for protection and management. Such an assessment has been performed regionally (Nielsen et al. Reference Nielsen, Herrero, Boyce, Mace, Benn, Gibeau and Jevons2004b ), but mining site-specific conditions likely do not reflect region-wide mortality risk because access restrictions and hunting prohibition on MSLs likely decrease bear mortality. Adaptive management will be necessary for conserving grizzly bears and other wildlife on and around mine sites in response to stakeholder pressures regarding future land use.
ACKNOWLEDGEMENTS
We thank Katie Yalte, Jerome Cranston and Julie Duval for assistance with aerial photo and GIS layer acquisition. Karen Graham supplied much of the bear GPS radiocollar data, based on efforts of multiple capture teams. Curtis Brinker, Beth MacCallum and Leo Paquin shared their rich site-specific experience which helped with GIS layer updating. We thank Teck Coal Ltd and Sherritt for granting access on their mine leases to Foothills Research Institute and University of Alberta personnel. Steven Hilts reviewed an earlier draft of this manuscript. The editor and anonymous reviewers made excellent constructive suggestions that greatly improved the manuscript. This analysis was funded by Teck Coal Ltd, Sherritt, and Foothills Research Institute Grizzly Bear Program and its sponsors. The research was funded in part by two mining companies (Teck Coal Ltd and Sherritt). Research findings were transparent and funders had no role in study design, reporting or decision to submit the manuscript for publication. The research involved capturing adult grizzly bears for GPS radiocollar deployment. Capture methods were approved by the University of Saskatchewan and University of Alberta Animal Care and Use Committee for Biosciences (558804). The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional guides on the care and use of wildlife in research.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0376892916000217
