Environmental heterogeneity influences animal behavior (Wiens and Milne 1989, Crist et al. 1992, With 1994) with selection for food resources depending partly on how the properties of those resources vary in space (Heinrich 1979). The response of an animal to variation in food abundance and distribution is contingent on the amount of resource heterogeneity that it detects in its surroundings, which is consistent with its environmental ‘grain (Levins 1968). Optimal foraging theory proposes that animals will seek to acquire food resources at the lowest energetic cost, thereby maximizing efficiency and fitness (Charnov 1976). This assumes that animals have perfect knowledge of the heterogeneity of these resources (Rapport 1991); however, this information is typically incomplete (Pyke 1984) as it is constrained by their grain size. Generally, grain size increases with body size (With 1994, Ritchie 1998, Mech and Zollner 2002), suggesting that large mammals would perceive their environment at a relatively broad spatial scale beyond that of the local patch.

As experimental scales should be dictated by the organism and phenomenon of study (Wiens et al. 1986, Addicott et al. 1987), examination of selection for food resources (Nielsen et al. 2010) should therefore not only consider local supply, but also incorporate landscape-level resource estimates to better reflect the amount of environmental heterogeneity large mammals perceive. Grizzly (brown) bears Ursus arctos are habitat generalists with a diverse, omnivorous diet (Hamer et al. 1991, Mattson et al. 1991, McLellan and Hovey 1995, Munro et al. 2006) that enables them to adjust their foraging behavior based on annual and seasonal food availability (Jonkel and Cowan 1971, Bunnell and Tait 1981, Deacy et al. 2017). Resource abundance and distribution therefore affect habitat use (Jonkel and Cowan 1971, Nielsen et al. 2004a, 2010), and accounting for this heterogeneity by acknowledging their grain size could provide valuable insight into the landscape factors most relevant for habitat selection.

Resource selection functions (RSFs) are statistical tools for evaluating animal habitat selection that estimate relative probability of use given particular environmental site characteristics (Manly et al. 2002). RSF models that have been developed for grizzly bears previously (Nielsen et al. 2002, Moe et al. 2007, Goldstein et al. 2010, Peters et al. 2015) have mainly evaluated the effects of environmental variables, such as habitat cover type and elevation, measured at the local patch-level. Questions of spatial scale, which are essential for testing the influence of environmental heterogeneity, have been largely overlooked in the context of grizzly bear habitat selection, although these have been investigated for other animals such as ungulates (Boyce et al. 2003, Anderson et al. 2005). The few studies that have directly considered scale have focused on the extent of the landscape available for bear use (Nielsen et al. 2004a, Ciarniello et al. 2007), rather than the spatial scale at which properties of the resource units themselves were estimated. Grizzly bear habitat selection models also seldom incorporate food resource attributes as explanatory variables, despite food-probability models often explaining bear selection more effectively than those that are habitat-oriented (Nielsen et al. 2003). Buffaloberry occurrence in particular, along with that of a few other key food items, significantly predicts bear foraging activity (Nielsen et al. 2010). Spatial variation of food resources appears to be more influential for grizzly bear habitat use than temporal availability (Mangipane et al. 2018); however, the importance of food resource heterogeneity relative to other food properties has not been explored, although it has been demonstrated to strongly affect grizzly bear foraging behavior (Searle et al. 2006).

Canada buffaloberry Shepherdia canadensis (L.) Nutt. is one of the primary fruit resources for grizzly bears in the Canadian Rocky Mountains, where it comprises a major component of their summer and early fall diet (Munro et al. 2006). The fruit is especially crucial during hyperphagia, when bears increase their food consumption to build body fat reserves in preparation for winter denning (Nelson 1980). Grizzly bear selection for buffaloberry has been assessed using predictions of shrub occurrence (Nielsen et al. 2003, 2010), but given the dioecious habit of this species and that only female plants bear fruit, occurrence does not correspond to the availability of the food resource itself. Although local density of grizzly bears is correlated with buffaloberry fruit abundance (Nielsen et al. 2017), selection for buffaloberry fruit resources has not been fully examined. An understanding of seasonal habitat use of this important resource is valuable for informing the conservation and management of grizzly bear populations (Boyce et al. 2002), including the threatened population in the province of Alberta (ASRD 2010).

Figure 1.

Location of the study area near Hinton, Alberta, Canada (53°24′41″N, 117°33′50″W) defined by a buffaloberry fruit density model (fruit/900 m²) by Nielsen et al. (2017). Fruit density is symbolized based on quantile classification.

The focus of this paper is to evaluate the role of landscapelevel food resource heterogeneity in grizzly bear selection for buffaloberry patches by comparing patch- and landscape-level foraging strategies. Buffaloberry heterogeneity as defined here reflects both fruit distribution and variability in patch quality, represented by fruit density, which are properties that contribute to spatial heterogeneity (Kotliar and Wiens 1990). We know that buffaloberry heterogeneity is significantly related to the heterogeneity of conifer forest canopy (Denny and Nielsen 2017), but the implications of these shrub (resource) patterns for bears have not yet been studied.

Our objectives were therefore to: 1) identify the spatial scale at which grizzly bears select buffaloberry fruit resources; 2) determine whether patch- or landscape-level foraging strategies explain resource use; and 3) assess the importance of resource heterogeneity in structuring habitat selection. We hypothesize that: 1) landscape-level foraging strategies will be more supported than those of the patch-level because grizzly bears perceive their surroundings at spatial scales beyond that of the patch, in accordance with their environmental grain; and 2) the heterogeneity of the food resource (buffaloberry) will further explain grizzly bear resource selection because changes in food availability influence foraging behavior.

Methods

Study area

We defined the study area by the spatial extent of a buffaloberry fruit density model (see Nielsen et al. 2017 for details of the modelling approach) that covered 19 952 km² of managed, conifer-dominated forest located near the town of Hinton (53°24′41″N, 117°33′50″W) in the Rocky Mountains foothills of west-central Alberta, Canada (Fig. 1). Elevation ranges from approximately 850–3200 m with dominant tree species being lodgepole pine Pinus contorta, white spruce Picea glauca and trembling aspen Populus tremuloides. Active natural resource extraction and human activity related to the forestry, mining, and energy (oil and gas) industries, along with recreational use, have produced a variety of anthropogenic disturbances and early seral forests.

Table 1.

Eight candidate hypotheses and associated foraging strategies describing grizzly bear selection for buffaloberry fruit resources at the patch- and landscape-levels.

Comparison of patch- and landscape-level foraging strategies

We developed eight a priori candidate hypotheses describing grizzly bear selection for buffaloberry based on fruit resource properties proposed to influence foraging strategies at the patch- and landscape-levels (Table 1). Patch-level hypotheses suggested bears would prioritize the local quality of the resource and direct proximity to a patch, while those of the landscape-level focused on the average quality of the resource, variability in patch quality, and fruit distribution. In particular, the resource heterogeneity hypothesis considered an interaction between variability in patch quality and fruit distribution, where it was hypothesized that greater contrast between patches could promote selection for areas with a widespread fruit distribution by facilitating the assessment of patch quality. We fit a mixed-effects logistic regression model representing each hypothesis to the set of ‘used’ and ‘available’ locations. Multicollinearity among explanatory variables was evaluated to ensure that Kendall rank correlation coefficients did not exceed a threshold of |0.7| and variance inflation factors for linear terms were < 5. We tested variable transformations using AIC and consequently applied log₁₀-plus-one transformations to all variables except elevation and distance to nearest patch. After determining the final model structures, we compared support for the eight hypotheses with AIC. Coefficients were also used to predict effects of both individual variables and interaction terms on the relative probability of grizzly bear use of a site.

Figure 2.

Effects of variability in buffaloberry patch quality (standard deviation of fruit density), mean fruit density, and fruit distribution (proportion of the landscape with buffaloberry fruit present) on the odds of grizzly bear use (selection) of a site, across spatial scales (moving window radii) from 172 m to 5488 m. Dashed blue line represents an odds ratio of 1 which would indicate no effect of the explanatory variables.

Table 2.

Comparison of the eight candidate foraging hypotheses describing grizzly bear selection for buffaloberry fruit resources at the patchand landscape-levels. Standardized beta (β) coefficients for model variables are displayed with standard errors in square brackets and odds ratios italicized. Dashes indicate where variables are not applicable.

Results

Comparison of patch- and landscape-level foraging strategies

Landscape-level foraging strategies were generally more supported than those of the patch-level, the highest ranked of which was proximity to patch. The heterogeneity hypothesis was most supported overall and represented all of the model weight (Table 2). For this hypothesis, variability in patch quality demonstrated a strong positive effect on grizzly bear habitat selection with a one standard deviation increase in variability raising the odds of bear use by 66%, compared to 16% for an equivalent increase in fruit distribution (Table 2). The relative probability of bear use was predicted to increase more than six-fold between areas of no variability in patch quality and those with an average landscape-level variation of 400 fruit per 900 m², the latter representing the approximate yield of an average shrub in a typical season (Fig. 3). Probability of habitat use gradually increased 1.5-fold from a landscape with no buffaloberry to one completely occupied by shrubs (Fig. 4). The interaction between variability in patch quality and fruit distribution was substantially supported over the individual variability and distribution hypotheses (ΔAIC scores of 25.4 and 749.3, respectively). Although within the heterogeneity hypothesis these two variables were positively related to bear habitat use, their interaction was negative, indicating decreased use of areas with a more widespread fruit distribution when variability in patch quality was high (Table 2). The relative probability of bear use was predicted to remain low across most observed levels of fruit distribution until the variability in patch quality exceeded about 100 fruit per 900 m² (2.0 on a log₁₀-transformed scale; Fig. 5). Relative probability of bear use appeared highest at moderate fruit distributions and high variability in patch quality, demonstrating a tradeoff between these two factors. Predictions were constrained to conditions found within the study area (Fig. 5) after further inspection revealed that extreme tradeoffs could not occur, since both very low fruit distribution and very high variability in patch quality was not possible (Fig. 6). Overall, habitat selection was highest in the upper foothills and subalpine regions of the study area (Fig. 6).

Figure 3.

Predicted effect of the variability in buffaloberry patch quality (standard deviation of fruit density; fruit/900 m²), estimated at a spatial scale of 1887 m (window radius), on the relative probability of grizzly bear use (selection) of a site, based on the heterogeneity foraging hypothesis which was most supported in AIC comparison. Dashed lines indicate 95% confidence intervals.

Figure 4.

Predicted effect of buffaloberry fruit distribution (proportion of the landscape with buffaloberry fruit present), estimated at a spatial scale of 1887 m (window radius), on the relative probability of grizzly bear use (selection) of a site, based on the heterogeneity foraging hypothesis which was most supported in AIC comparison. Dashed lines indicate 95% confidence intervals.

Discussion

We observed that buffaloberry heterogeneity best explained grizzly bear selection for fruit resources. Bears selected for areas with a wider distribution of buffaloberry fruit and greater variability in patch quality, thus providing both a higher probability of shrub encounter and greater contrast between resource patches. Landscape-level foraging strategies were more supported than those of the patch-level, pointing to the large environmental grain of grizzly bears. Specifically, bears were most influenced by buffaloberry properties estimated at a spatial scale of 1887 m or 1.89 km (window radius) which corresponds to an average travel distance over 5.5 h, suggesting they may synthesize information on environmental conditions they encounter while moving through their home range. A similar scale of 1.69 km was used to measure average buffaloberry fruit supply when comparing this to the local abundance of bears (Nielsen et al. 2017).

Variability in patch quality emerged as an especially meaningful landscape-level buffaloberry property, based on its inclusion in the two most supported hypotheses. This implies that contrast between adjacent resource patches promotes selection by facilitating the assessment of quality and enabling bears to more efficiently identify high-quality patches. Even the addition of a modest amount of variability, equivalent to the fruit yield of an average shrub, was shown to increase bear habitat use (Fig. 3). Less support was found for the landscape resource quality hypothesis, which highlighted areas with a broad distribution of high-quality buffaloberry patches that were thus characterized by greater landscape homogeneity. Grizzly bear utilization of buffaloberry resources may therefore be contingent not only on locating high-quality patches, but also recognizing them as such, suggesting contrast could act as a cue for exploitation.

Figure 5.

Predicted effect of the interaction between variability in buffaloberry patch quality (standard deviation of fruit density) and fruit distribution (proportion of the landscape with buffaloberry fruit present) on the relative probability of grizzly bear use (selection) of a site, based on the heterogeneity foraging hypothesis which was most supported in AIC comparison. Axes displayed with log₁₀-plus-one transformed values to reflect variable transformations in the model. Predictions were constrained to conditions found within the study area by generating a minimum convex polygon representing data for all ‘used’ and ‘available’ locations in the model.

The positive effect of landscape variability on grizzly bear habitat selection is also likely related to their generalist habit (Hamer et al. 1991, Mattson et al. 1991, McLellan and Hovey 1995, Munro et al. 2006) and distinct habitat needs for different diel activities (Munro et al. 2006). Fruit density is a function of environmental and demographic factors that influence shrub growth and fruit production, such as local canopy cover (Hamer 1996, Nielsen et al. 2004b) and density (Johnson and Nielsen 2014). Greater variability in fruit density suggests the presence of a broader range of site types and increased environmental heterogeneity, which promotes plant species richness (Kreft and Jetz 2007). A higher diversity of complementary food resources, which affect local bear abundance (Nielsen et al. 2017), could provide animals with the opportunity to substitute these more readily and optimize macronutrient intake (Coogan et al. 2014). Landscape heterogeneity may thus contribute to grizzly bear foraging success by facilitating dietary flexibility, and helping to buffer against inter-annual fluctuations in the productivity of key food species (Krebs et al. 2009).

Previous work has also found food resource heterogeneity to be important for grizzly bear foraging behavior (Searle et al. 2006) and habitat use (Mangipane et al. 2018). Searle et al. (2006) conducted feeding trial experiments where the spatial arrangement of patches was manipulated and residence times of resource patches examined. Surrounding spatial context in the patch hierarchy was shown to affect residence time, with models accounting for this broader heterogeneity 34-times more supported than those limited to the patch-level.

The combination of both fruit distribution and variability in patch quality in the heterogeneity hypothesis suggests that bears respond to multiple aspects of their food resources simultaneously, implying that they compromise between different foraging strategies (Table 1) when selecting resources (Senft et al. 1987). This adaptability may be particularly necessary in our study area where anthropogenic disturbance caused by resource extraction is prevalent (Gaulton et al. 2011), requiring that bears navigate a dynamic environment. The interaction between variability in patch quality and fruit distribution was important to include, as the heterogeneity hypothesis received considerably more support than either of the individual hypotheses. Contrary to our expectations, the interaction was observed to be negative; this may relate to the elevation gradient and associated geographic separation of areas with a widespread fruit distribution and those with high variability in fruit density, which were restricted to high elevations in the west. Fruit distribution presumably demonstrates greater control over grizzly bear resource selection in the low-lying eastern portion, but its effect may decline with proximity to the mountains as it is overwhelmed by the comparatively stronger influence of variability. Bears in our study area may thus face a tradeoff between selecting for fruit distribution and variability in patch quality, partly due to the effect of terrain on buffaloberry properties.

Figure 6.

Buffaloberry fruit distribution (proportion of the landscape with buffaloberry fruit present), variability in buffaloberry patch quality (standard deviation of fruit density), and grizzly bear habitat selection based on the heterogeneity foraging hypothesis for the study area near Hinton, Alberta, Canada (53°24′41″N, 117°33′50″W). Elevation was not included as a variable in prediction of habitat selection to focus on the effects of buffaloberry properties. Symbolization is based on quantile classification for ease of display; however, areas with ‘very high’ variability in patch quality, in particular, may be over-represented due to a low frequency of extremely high values.

Our findings suggest human activities that shift the distribution and quality of food resources and generate landscape heterogeneity could promote grizzly bear habitat selection. Forest harvesting, for example, has been found to increase bear habitat use (Nielsen et al. 2004a, Stewart et al. 2013) through the creation of canopy openings which foster the growth of various food plants (Nielsen et al. 2004b, Munro et al. 2006) and significantly affect understory heterogeneity (Denny and Nielsen 2017). The effects of disturbance on food resource heterogeneity are not straightforward to predict, however, as they depend on a number of factors including initial site conditions and disturbance type. Proposed developments should therefore be evaluated on an individual basis to anticipate possible consequences for landscape patterns of food resources and bear foraging behavior, as well as mortality risk (Boulanger and Stenhouse 2014).

Conclusions

These results emphasize the value in considering environmental heterogeneity and spatial scale in studies of animalresource interactions. Landscape-level estimates of resource properties should be incorporated into resource selection frameworks, especially for large-bodied, vagile species which typically perceive and respond to their surroundings at broad spatial scales. We recommend that researchers test the sensitivity of habitat selection to resource spatial scale, as we have here, to determine that which is most appropriate prior to proceeding with further analysis. This step will contribute to reducing the error associated with adopting an arbitrary experimental scale (Wiens 1989), as the conventional patchlevel focus may obscure the factors affecting animal space use (Morrison et al. 2006). Accurately representing the level of spatial perception of study organisms will reflect the influence of food resource heterogeneity and ensure valid conclusions regarding foraging behavior to inform the conservation and management of wildlife species at risk.