Potential for population expansion by black bears in Alabama

Abstract

Human development and lack of habitat can impede spatial population growth for many species. The amount of habitable area available to American black bears has been drastically reduced, especially in the southeastern United States, causing some populations to face possible extirpation. Nonetheless, some Black Bear populations have begun to recover and recolonize portions of historic Black Bear range, despite human-dominated landscapes. The relationship between Black Bear population expansion and human development is especially apparent in Alabama. Our study makes predictions of the potential for population range expansion of black bears in Alabama. We created resource selection models for black bears via a Geographic Information System and location data from GPS-collared black bears. Our models suggested that there are areas of Alabama that could provide opportunities for population growth, allowing bear populations to expand toward their historic distribution. Understanding potential bear population expansion in Alabama could help to inform wildlife managers who are seeking to enhance bear populations and prepare for potential bear population growth in the state and elsewhere in the United States.

Locations from GPS-collared black bears were used to create potential bear population range expansion maps for Alabama. The prediction maps indicate that there are areas of the state that have high relative probability of use, and thus high potential for bear population expansion throughout Alabama.

Historic population ranges of many animal species have been reduced as a result of over harvest, habitat modification, and human settlement (Woodroffe 2000). While the effects of overharvest often can be reversed by simply allowing the population to grow once again, the population effects of habitat loss typically cannot be overcome unless lost habitat is restored and animals can reach these areas. Thus, due to alterations to the landscape, many species have experienced permanent reductions in their geographic range (hereafter, range). The effects of habitat alteration, disturbance, and human development across the landscape are especially apparent in large carnivore species because of their potential for human conflict, extensive home ranges, and greater spatial requirements to meet energetic needs compared to herbivores (Woodroffe 2000). Thus, when the conservation goal for a large carnivore species is to expand its range and restore population size, human development and lack of habitat can critically impede those population expansion efforts.

In the early half of the 20th century, Black Bear (Ursus americanus) populations across the United States significantly decreased in range and size in part due to loss of habitat and the expansion of human populations (Cowan 1970; Woodroffe 2000; Scheick and McCown 2014). The decrease in bear populations was particularly dramatic in the southeastern United States, where bear populations experienced an 80% reduction in range (Pelton and van Manen 1997; Scheick and McCown 2014). Like many large mammals, black bears have extensive home ranges and prefer large tracts of forested land with minimal human presence (Clark et al. 1993; Tri et al. 2016). Consequently, increased human development caused a decrease in available Black Bear habitat (Cowan 1970). However, more recently, many Black Bear populations appear to have experienced a resurgence (Bales et al. 2005; Sollmann et al. 2016; Draper et al. 2017). For example, bear populations in Florida, which were once considered threatened at the state level, have experienced an increase in population growth from 2003 to 2015 (Humm et al. 2017). Other bear populations in the southeastern United States and beyond have experienced similar expansions in both size and range (Bales et al. 2005; Sollmann et al. 2016; Draper et al. 2017). Yet, some local Black Bear populations still appear to experience limitations to their growth and expansion, e.g., northern Georgia (Little et al. 2017) and Missouri (Sollmann et al. 2016). Consequently, more information is needed to understand how black bears use human-dominated landscapes and how black bears expand their population range considering the near-ubiquitous presence of humans.

An example of Black Bear populations whose growth and spatial expansion may be limited due to habitat loss and human encroachment are those found in the state of Alabama. While historically native throughout the state, black bears were nearly extirpated from Alabama in the early part of the 20th century (Cowan 1970; Scheick and McCown 2014). For several decades, the only remaining black bears consisted of a small population around the Mobile River Basin (MRB). However, this population has been highly genetically isolated and does not appear to be connected with any other nearby Black Bear populations in neighboring states (Draper et al. 2017). Putatively, the isolation of the MRB population may be a result of the high level of human development and inadequate habitat around Mobile and the surrounding area.

Conversely, a new population of black bears has recently repatriated northern Alabama (NAL) when immigrant bears from northern Georgia settled in the area (Draper et al. 2017). The NAL population appears to be growing quickly and expanding further into Alabama (Draper et al. 2017; H Leeper, personal observation). The rapid expansion of the NAL population could be facilitated by the lower level of human development in the area compared to MRB. Ultimately, however, more information is needed on how black bears use landscapes in Alabama and what potential exists for bear populations to expand spatially in the state.

In this study, we created habitat selection models for black bears, fit to location data from GPS-collared black bears in Alabama via a Geographic Information System (GIS). We then extrapolated these models to generate predictions about the relative probability of habitat use of various landscapes available in Alabama and the potential for Black Bear expansion throughout the rest of the state. We hypothesized that there would be lower potential for bear population expansion into habitat surrounding the MRB population because of higher human density and a more urbanized landscape compared to NAL, and that there would be higher potential for bear population expansion into habitat surrounding the NAL population because of lower human density compared to MRB and a more rural landscape with large tracts of undeveloped, forested land cover. Ultimately, understanding how and where bear populations may expand in the state should be useful to wildlife managers who seek to enhance bear populations and prepare for bear population growth in the state and elsewhere in the United States.

Methods

We created second-order habitat selection models for black bears in Alabama using ArcMap 10.6 (ESRI, Inc., Redlands, California). Second-order selection is defined as the home-range that an animal selects within the geographical range of that species (Johnson 1980). Two different models were created for the population around the MRB and for the north Alabama population (Fig. 1) because northern and southern bear populations appear to be from 2 different subspecies (U. a. americanus and U. a. floridanus, respectively) that are more closely connected to bear populations in other states than each other (Draper et al. 2017). Additionally, we had no reason to think that a model fit to data from the southern coastal plains would describe bear habitat use in the northern mountains and vice versa.

Fig. 1.

Black bear populations in Alabama. Locations are from GPS-collared bears from 2014 to 2020 in used locations from the MRB population and the northern Alabama (NAL) population.

The MRB population (31°12′N, 88°06′W) is centered in Mobile and Washington counties in the southwestern corner of Alabama. Dominant habitat types as determined from the 2016 National Land Cover Database (NLCD; Dewitz 2019) included mixed forest (“deciduous nor evergreen species are greater than 75% of total tree cover”), evergreen forests (“≥75% of the tree species maintain their leaves all year”), and woody wetlands (“areas where forest or shrubland vegetation accounts for greater than 20% of vegetative cover and the soil or substrate is periodically saturated with or covered with water”). Elevation ranged from 0 m (sea level) to about 117 m above sea level (a.s.l.), and slope ranged from zero degrees to about 10 degrees. Conversely, the NAL population (34°30′N, 85°36′W) is centered in DeKalb and Cherokee counties, including the federally managed Little River Canyon National Preserve and DeSoto State Park. In addition, many large tracts of privately managed hunting land and stands managed for timber harvest covered the landscape. Dominant habitat types in NAL included deciduous forests (“≥75% of the tree species shed foliage simultaneously in response to seasonal change”), evergreen forests, and pasture and hay (“areas of grasses, legumes, or grass-legume mixtures planted for livestock grazing or the production of seed or hay crops, typically on a perennial cycle”), with steep topography and mountainous terrain (Dewitz 2019). Elevation ranged from about 164 to 600 m a.s.l., and slope ranged from zero degrees to about 35 °C. Human population density was higher in MRB (12,852 people/sq km) than in NAL (5,351 people/sq km), creating a landscape mosaic and higher proportion of bears occupying suburban areas (U.S. Census Bureau 2010; Dewitz 2019).

Variables considered for inclusion in the models included proximity to water, proximity to primary and secondary roads, land cover type, elevation, slope, aspect, stewardship, and human density. Proximity to primary and secondary roads (TIGER/Line Shapefile) was considered (as both categorical and continuous variables) based on previous studies that showed tendency of a bear to avoid roads because of human activity, motorists, and increased perceived danger (Reynolds-Hogland and Mitchell 2007). To create a binary categorical road variable, roads were buffered by 800 m (Reynolds-Hogland and Mitchell 2007; Atwood et al. 2011), in which bear locations were classified as either being within or outside of the buffered areas. Separately, to create a continuous variable, the shortest distance to roads was calculated for each bear location using the Euclidean Distance tool in the Spatial Analyst toolbox in ArcMap 10.6. Similarly, proximity to water (U.S. Geological Survey 2020) was tested as both categorical and continuous variables. To create a binary categorical variable, water was buffered by 600 m based on previous studies that showed tendency of a bear to select for areas within 600 m of water (Clark et al. 1993; van Manen and Pelton 1997; Atwood et al. 2011; Tri et al. 2016). The shortest distance to water was also calculated for each bear location to create a continuous variable using the Euclidean Distance tool in the Spatial Analyst toolbox in ArcMap 10.6. Land cover types were derived from the NLCD (Dewitz 2019). Slope and aspect were both derived from elevation data from the Consortium for Spatial Information (Reuter et al. 2007). Elevation in Alabama ranges from −18 m a.s.l. to 733 m a.s.l., and slope in Alabama ranges from 0 °C to about 45 °C (Reuter et al. 2007). Aspect was categorized into 4 cardinal directions. Stewardship of each location was classified into 4 groups, from “1” being highly protected and managed for biodiversity to “4” having no known mandate for biodiversity protection (USGS 2018). Human density was calculated as the number of people per square kilometer using data from the U.S. Census Bureau (2010). Roads, water, land cover, stewardship, and human density variables had a 30 m resolution; elevation, slope, and aspect had a resolution of about 89 m. For our analysis and extrapolated maps, we converted all layers to 89 m resolution.

We fit the habitat selection models to location data from GPS-collared black bears collected from 2014 to 2020 (Fig. 1). Bears were handled following guidelines from the American Society of Mammalogists (Sikes et al. 2016). Scientific Collection Permits were acquired from the Alabama Department of Conservation and Natural Resources (2021074581268680, 2019110186068680, 2018085664868680, 2015076620068680, and 2014082460268680), and handling protocols were approved by Auburn University’s IACUC (2014-2561, 2018-3266, and 2021-3889). Adult black bears (>1.5 yr old based on body size, body condition, and tooth condition) were trapped from May through November using corn, pastries, and commercial bear attractants inside a Cambrian-style trap, which has a counter-weighted door that swings down and latches when the trigger is pulled. Captured black bears were anaesthetized intramuscularly using a jab stick (Dan-Inject, Austin, TX), with 4 mg/kg of body weight of Telazol (Fort Dodge Laboratories, Fort Dodge, IA; Hebblewhite et al. 2003; Garrison et al. 2007; Tri et al. 2016). Bears were fitted with a motion-sensitive GPS collar (Vectronics, Berlin, Germany; Telonics, Inc., Mesa, AZ). Collars were designed to enter mortality mode after a 12-h stationary period. Collar locations were acquired every 1 to 2.5 h. Iridium satellite connections built into the collars allowed remote monitoring of movements without the need to relocate the bear and download the GPS locations from the collar. We determined collar accuracy by measuring the distance of acquired GPS fixes from a single collar that was placed at a known location. Any 1 GPS location from a collar could be up to about 86 m from the true location, although most locations were within about 20 m.

Used locations were divided into the MRB population and NAL population. A minimum convex polygon (MCP) was created by pooling the locations from all bears within each population and then buffered by 1 km to represent the area of habitat that was putatively “available” to bears for use during the study period. A sample of random locations from the available habitat equal to the number of used locations was generated within each MCP (n = 92,851 for MRB, n = 40,765 for NAL). We followed a Design II sampling protocol, whereby used points were collected at the individual level and available points were sampled at the population level (Manly et al. 2002). Habitat attributes from each GIS layer were then extracted for both used and available locations. Any land cover types that had fewer than 5 used locations were removed from the analysis due to problems they would cause with model convergence. Instead, we considered those land cover types as unusable. In MRB, high-intensity development and cultivated crops were removed from the analysis because they had fewer than 5 used locations in those land cover types. In NAL, low-intensity development, medium-intensity development, high-intensity development, and barren land were removed from the analysis because they had fewer than 5 used locations in those land cover types. Similarly, human densities above which there were no used locations were removed from the analysis; we considered those higher human densities as unusable by bears. In MRB, densities above 1,576.15 humans per square km; and in NAL, densities above 134.78 humans per square km were removed from the analysis. The MRB population did not have any locations (used or available) with a stewardship ranking of 1.

Logistic regression was used to build a model for each dataset using the ‘glm’ function in R (MRB and NAL; R Core Team 2020). Models were built using a carefully directed backwards stepwise procedure (Hosmer and Lemeshow 2000). Specifically, a model containing all terms was fit to the data. Originally, the binary, categorical variables for proximity to water and proximity to roads were included in the model. We compared this full model to a similar one in which the continuous proximity to water variable was included instead of the binary water variable using Akaike Information Criterion (AIC; Burnham and Anderson 2002). We similarly compared the full model to a similar one in which the continuous proximity to road variable was included instead of the binary road variable using AIC. Subsequently, a directed approach was used to remove model terms; the most nonsignificant variable was removed from the model until all variables remaining in the model were significant (P < 0.05). Categorical variables, such as land cover type, aspect, and stewardship were either kept in the model or removed, rather than attempting to combine categories that were not significantly different because trying to combine nonsignificant terms would have likely been very difficult, if not impossible, in such complicated models. Variance inflation among variables was generally low (all VIF < 1.55 for MRB and < 1.54 for NAL). Thus, we considered multicollinearity of negligible to no concern, especially given our sample sizes (see below).

Parameters from the models generated via the logistic regression were used to create models with a Poisson form that should be proportional to the probability of habitat use (i.e., resource selection functions (RSF); Manly et al. 2002; Keating and Cherry 2004; Johnson et al. 2006). Specifically, the model had the form:

$${\displaystyle \begin{array}{l}{\displaystyle Y=\mathrm{e}\mathrm{x}\mathrm{p}({\beta }_1{X}_1+\cdots +{\beta }_k{X}_k)}\end{array}}$$

where Y is proportional to the probability of use, X is the habitat attribute, and β is the coefficient estimate for that habitat attribute generated via the logistic regression for k parameters. The RSF models were used to generate potential population expansion maps via the Raster Calculator in ArcMap 10.6. Because values of some of the continuous variables state-wide exceeded those observed within the MCPs, we created 2 versions of the population expansion maps for each population: in one version, we allowed the model to extrapolate predictions, such that the parameter values continued with changing X outside the range of continuous habitat attribute values observed within that population’s MCP (i.e., extrapolated version). In the other version, we identified the minimum and maximum values of each continuous habitat attribute within the MCP, and any areas of the state that had values for those habitat attributes outside of the MCPs were set to those minimum and maximum values for the purposes of generating predictions (i.e., truncated version). The minimum and maximum values for each continuous variable within each MCP and across the state are provided in Table 1.

Table 1.

Minimum and maximum values of continuous variables for black bears in the MRB, northern Alabama (NAL), and Alabama.

Variable	MRB		NAL		Alabama
	Min	Max	Min	Max	Min	Max
Distance to water	0	3,780	0	2,363	0	5,398
Distance to roads	0	12,720	0	8,742	0	18,816
Elevation	0	117	164	599	−18	733
Slope	0	9.5	0	35.4	0	44.8
Human density	0	12,852	0	5351	0	400,833

Distances and elevation are in m a.s.l. Slope is in degrees. Human density is number of people per square km.

The predictive ability of each model within the respective bear study area was evaluated using the k-fold cross-validation method (Johnson et al. 2006). Specifically, we randomly divided the data from each area (MRB and NAL) into 10 equally sized groups (folds). For each of 10 evaluations, a different group was designated as the testing dataset, whereas the remaining 9 groups were combined to represent the training dataset. The logistic regression model chosen for that area during the procedures outlined previously was fit to each training dataset. The coefficients from each logistic regression were then used to create models with a Poisson form, as described above. The Poisson models, or RSF’s, were used to assign an index of probability of use (hereafter referred to as “relative probability of use”) values to pixels in a GIS, with an extent equal to the appropriate MCP described previously. The pixels were ranked by RSF value and equally divided into 10 bins by number of pixels in the MCP. For example, the lowest 10% of the ranked RSF values comprised the first bin, the second lowest 10% of ranked RSF values comprised the second bin, and so on. We then calculated the expected number of used locations within each bin via the utilization function described in Johnson et al. (2006). Finally, we compared the expected number of used locations with the observed number of used locations from the testing dataset within each bin via linear regression and chi-square tests. This process was repeated a total of 10 times, with a different group designated as the testing dataset and the remaining 9 groups combined to form the training dataset for each iteration. The subsetting was done by removing points from all bears randomly. Finally, we compared the similarity of both MRB models (extrapolated and truncated) to both NAL models by standardizing the scale of each raster, then plotting the difference in values between each raster. For example, the extrapolated versions of both the MRB and NAL models were re-scaled from 0 to 1. Then, one raster was subtracted from the other in order to determine the difference in cell values. While we recognize that the MRB and NAL models cannot be compared directly, this analysis was simply used to visually compare how similar the MRB and NAL model predictions were to each other.

Results

We acquired location data from 48 different bears in Alabama: 29 (10 M:19 F) bears in the MRB population and 19 (5 M:14 F) bears in the NAL population. The distribution of these bears adequately represented current Black Bear range in Alabama (Draper et al. 2017; H Leeper, personal observation; C Seals, personal observation). The number of locations per bear ranged from 150 to 8043.

We found that the model to predict relative probability of Black Bear use of a location, and thus potential for population expansion, was best fit by the global model (all variables) for both the MRB and NAL datasets. Variables in the global model included distance to water, distance to roads, land cover type, elevation, slope, aspect, stewardship, and human density (Tables 2 and 3). AIC scores indicated that both distance to water and distance to roads were better treated as continuous (Euclidean distance) rather than categorical (buffered) variables (distance to water for MRB, ∆ AIC = 3,236; for NAL, ∆ AIC = 2,299; distance to roads for MRB, ∆ AIC = 1,560; for NAL, ∆ AIC = 10,188). Coefficient estimates for the models are listed in Tables 2 and 3. However, we caution that our goal in the analysis was prediction, not inference, and thus care should be used in interpretation of coefficient estimates, especially given collinearity present in the variables (see methods; Arif and MacNeil 2022). The RSF models indicated that some areas of the state have higher probability of use, while other areas of the state have lower probability of use, and thus low potential for population expansion for both models (Figs. 2 and 3).

Table 2.

Coefficients from logistic regression models for black bears in the MRB.

Region	Variable	Coefficient	Confidence interval	P
MRB	Distance to water	0.00139	(0.00136–0.00141)	<0.0001
MRB	Distance to roads	−0.000110	(−0.000115 to −0.000105)	<0.0001
MRB	Barren ground	−2.15	(−2.53 to −1.80)	<0.0001
MRB	Deciduous forest	0.43	(0.21–0.66)	0.00014
MRB	Developed, low intensity	−1.79	(−2.02 to −1.57)	<0.0001
MRB	Developed, medium intensity	−1.83	(−2.18 to −1.49)	<0.0001
MRB	Developed, open space	−0.74	(−0.85 to −0.64)	<0.0001
MRB	Emergent herbaceous wetland	0.24	(0.12–0.35)	<0.0001
MRB	Evergreen forest	0.50	(0.45–0.55)	<0.0001
MRB	Grassland/herbaceous	−0.21	(−0.28 to −0.15)	<0.0001
MRB	Pasture/hay	−0.67	(−0.81 to −0.53)	<0.0001
MRB	Shrub/scrub	0.19	(0.13–0.25)	<0.0001
MRB	Woody wetland	0.65	(0.60–0.70)	<0.0001
MRB	Elevation	−0.061	(−0.062 to −0.060)	<0.0001
MRB	Slope	0.16	(0.15–0.18)	<0.0001
MRB	North	0.0024	(−0.029–0.034)	0.88
MRB	West	−0.056	(−0.088 to −0.024)	0.00053
MRB	East	0.052	(0.021–0.082)	0.00093
MRB	Stewardship 2	−3.94	(−5.45 to −2.75)	<0.0001
MRB	Stewardship 4	1.93	(1.31–2.62)	<0.0001
MRB	Human density	−0.0045	(−0.0048 to −0.0041)	<0.0001

The reference group for land cover type is mixed forest, and for stewardship is a ranking of 3.

Table 3.

Coefficients from logistic regression models for black bears in northern Alabama (NAL).

Region	Variable	Coefficient	Confidence interval	P
NAL	Distance to water	0.00132	(0.00127–0.00137)	<0.0001
NAL	Distance to roads	0.00052	(0.00051–0.00053)	<0.0001
NAL	Cultivated crops	1.27	(1.12–1.41)	<0.0001
NAL	Deciduous forest	0.099	(0.040–0.16)	0.00098
NAL	Developed, open space	−1.26	(−1.42 to −1.10)	<0.0001
NAL	Emergent herbaceous wetland	0.39	(−0.96–1.68)	0.57
NAL	Evergreen forest	0.085	(0.074–0.16)	0.032
NAL	Grassland/herbaceous	0.087	(−0.035–0.21)	0.16
NAL	Pasture/hay	−0.85	(−0.95 to −0.75)	<0.0001
NAL	Shrub/scrub	0.099	(−0.020–0.22)	0.10
NAL	Woody wetland	0.014	(−1.11–0.98)	0.98
NAL	Elevation	0.0062	(0.0060–0.0064)	<0.0001
NAL	Slope	0.036	(0.031–0.041)	<0.0001
NAL	North	−0.050	(−0.11–0.0083)	0.093
NAL	West	−0.11	(−0.16 to −0.051)	0.00020
NAL	East	0.057	(0.0011–0.11)	0.046
NAL	Stewardship 1	1.13	(0.77–1.47)	<0.0001
NAL	Stewardship 2	−1.61	(−2.0085 to −1.22)	<0.0001
NAL	Stewardship 4	−0.12	(−0.48–0.21)	0.49
NAL	Human density	−0.031	(−0.033 to −0.029)	<0.0001

The reference group for land cover type is mixed forest, and for stewardship is a ranking of 3.

Fig. 2.

RSF for MRB, with relative probabilities of use extrapolated beyond the environmental attribute values of the MRB MCP (top row), and with relative probabilities of use calculated using truncated environmental attribute values from the MRB MCP (bottom row). The rightmost panel shows bear locations overlayed. RSFs were calculated as a function of distance to water, distance to roads, land cover type, elevation, slope, aspect, stewardship, and human density.

Fig. 3.

RSF for northern Alabama (NAL), with relative probabilities of use extrapolated beyond the environmental attribute values of the NAL MCP (top row), and with relative probabilities of use calculated using truncated environmental attribute values from the NAL MCP (bottom row). The rightmost panel shows bear locations overlayed. RSFs were calculated as a function of distance to water, distance to roads, land cover type, elevation, slope, aspect, stewardship, and human density.

When we combined the results from each fold, the cross-validation analysis of the model for MRB indicated that the regression line (Fig. 4a) had a slope of 0.89 (0.87 to 0.90; 95% C.L.), which was significantly different from 0, indicating that the model was better than a random or neutral model. However, the slope was significantly different from 1, suggesting that the model does not provide predictions that are proportional to the probability of use. The estimate of the intercept was 91.22 (66.94 to 115.49; 95% C.L.), which was significantly different from 0 (P < 0.0001); a result that again indicates that the model does not provide predictions that are proportional to the probability of use (Johnson et al. 2006). The chi-square test indicated significant deviation between observed and expected values (X²₉₉ = 3190.02, P < 0.0001). Although the cross-validation suggested that our predictions were slightly biased, we note that the general relationship of increasing expected number of used locations with increasing observed number of locations was very strong (R² = 0.99; Fig. 4a). Thus, we posit that our model is valid and useful for estimating, at least qualitatively, relative probability of use.

Fig. 4.

Cross-validation linear regression for a) MRB and b) northern Alabama (NAL) using 100 expected and 100 observed locations. For MRB, the slope of the line was 0.89 (0.87–0.90; 95% C.L.), and the intercept was 91.22 (66.94–115.49; 95% C.L.). For NAL, the slope of the line was 0.946 (0.939–0.954; 95% C.L.), and the intercept was 21.53 (14.93–28.12; 95% C.L.).

When we combined the results from each fold, the cross-validation analysis of the model for NAL indicated that the regression line (Fig. 4b) had a slope of 0.946 (0.939 to 0.953; 95% C.L.), which was significantly different from both 0 and 1. Additionally, the estimate of the intercept, 21.53 (14.93 to 28.12; 95% C.L.), was significantly different from 0 (P < 0.0001). Similarly, the chi-square test indicated significant deviation between observed and expected values (X²₉₉ = 765.64, P < 0.0001). As with results from our MRB model, however, we note that the general relationship of increasing expected number of used locations with increasing observed number of locations was very strong (R² = 0.9986; Fig. 4b). Nonetheless, based on results of cross-validation, we posit that our models are valid and useful for estimating relative probability of use.

The analysis of similarity indicated that the extrapolated MRB model is very similar to the extrapolated NAL model, and likewise, the truncated MRB model is very similar to the truncated NAL model (Fig. 5). Most areas of the state were predicted to have a similar relative probability of use, regardless of whether the MRB or NAL model was used.

Fig. 5.

Analysis of similarity between both versions (extrapolated and truncated) of each model (MRB and NAL). Similarity was calculated as the difference in standardized raster values between the extrapolated MRB model and extrapolated NAL model (left panel), and the truncated MRB model and truncated NAL model (right panel).

Discussion

Counter to our hypothesis that there is relatively low potential for population expansion surrounding the MRB population, our RSF appears to show high relative probability of use throughout the southern portion of the state (Fig. 2). Furthermore, our model suggests that there are substantial areas of the state that have the same relative probability of use as areas that are currently used by bears in the MRB. Conversely, the NAL model suggests relatively low potential for population expansion around the NAL population as indicated by low relative probability of habitat use (Fig. 3), which differs from our hypothesis that there would be higher potential for population expansion around the NAL population because of lower human density and larger tracts of undeveloped, forested land. However, the NAL model does indicate that there are other areas of the state further from the current NAL population range that appear to have the same relative probability of use as areas that are currently used by bears in NAL. Thus, both RSFs indicate that portions of Alabama could support spatial population growth. However, currently the MRB population does not appear to be growing spatially (Draper et al. 2017; C Seals, personal observation). Our model suggests that the MRB population does not appear to be limited by a lack of habitat surrounding the population, so there must be other factors that are hindering population expansion. For example, a lack of numeric population growth would likely limit spatial growth as well. Additionally, a lack of connectivity with nearby bear populations in neighboring states could also be limiting the MRB population expansion (Draper et al. 2017). In the NAL population, while it does appear to be expanding spatially, the growth is slow. The slow spatial growth could be explained by the fact that this population has newly recolonized the area, and thus not yet reached carrying capacity (Sinclair 1992; Frary et al. 2011). As population density increases, individuals may be pushed out of the current population range, thus expanding the NAL population throughout Alabama. Although male bears would likely disperse out of current NAL bear range prior to the population reaching carrying capacity, females must also disperse for the population to expand. However, females tend to disperse at a lower rate compared to males (Alston et al. 2022), which could also explain the slowed or delayed range expansion by the NAL population. Both Black Bear populations in Alabama should continue to be monitored for indicators of spatial and numeric population expansion throughout the state in order to prepare for the future of managing the species. Understanding population growth of black bears in Alabama can help managers predict if or when a population could support a harvest season or predict when human-bear interactions are likely to increase.

Regardless of whether the model made predictions for the rest of the state by extrapolating continuous variables to fit habitat attribute values outside those observed in the MCP, or the model made predications for the rest of the state by truncating habitat attribute values to those observed in the MCP, the RSF maps appear to show similar trends in the relative probability of use throughout Alabama (Fig. 5). These results suggest that our RSFs are fairly robust. We note that we did explore possible means of improving model fit to the data, including nonlinearities (polynomial) relationships in continuous variables and even the use of generalized additive models. While such model complexities did significantly improve the fit to the data in all cases (H Leeper, personal observation), they did not overcome the problems encountered in cross-validation. Thus, we did not include such model complexities in our final RSFs because we were concerned about overfitting the models since they would be extrapolated to other areas of the state. Ultimately, regardless of the exact method or model used, our results indicate that there are areas of the state that have similar probability of use to areas currently occupied by bears, which could mean that those areas have the potential to support spatial and numeric population expansion of the species. However, habitat is only useful to bears if they can access it; if there are barriers, bears cannot disperse to new areas of the state.

A previous study that analyzed Black Bear resource selection at the first-order scale (the geographic range of a species; Johnson 1980) in order to identify potential corridors and barriers in Alabama also generated similar predictions as this study about relative probability of habitat use (see Fig. 2 in Leeper and Steury 2021). In south Alabama, all maps predicted higher relative probability of use in areas surrounding the Mobile and Tensaw Rivers and their tributaries, as well as areas in and around the Conecuh National Forest on the Florida-Alabama Border. Similarly, in northern Alabama all models predicted higher relative probability of habitat use in the mountainous areas of northeastern Alabama—including Jackson County—areas along Lookout Mountain, and areas in and around Talladega National Forest (Fig. 3) as found in Leeper and Steury (2021). Notably, in Leeper and Steury (2021), bears sightings from the public (often of bears in areas that typically don’t have bears) were used to fit the resource selection function. Thus, the fact that all models generate similar relative probability of use maps, despite very different data sources, also suggests that our predictions are robust and our maps may do a good job of predicting relative probability of use across Alabama. However, while there are similarities between the predictions of our first- and second-order selection models, each set of models should serve a slightly different purpose. The first-order selection models are likely more accurate at predicting dispersal areas and potential corridors for young males that are leaving their natal territory. The second-order selection models are likely more accurate for predicting population expansion outside of current range. Yet the fact that the models make similar predictions suggests that habitats that bears use for dispersal may be similar to habitats that bears occupy long term. Previous studies in bears and other species have also found similarities between habitats occupied by resident populations and habitats used for dispersal (Fattebert et al. 2015; Myers and Young 2018; Sanz-Pérez et al. 2018). However, there is also evidence to suggest that habitats occupied by resident populations differ from habitats used for dispersal (Zeller et al. 2012). Therefore, we emphasize that our predictive models are only hypotheses, and the relative value of different habitat attributes for dispersal and residency needs to be verified with further research.

While we caution against using coefficient estimates from our RSFs for making inference about the importance of habitat attributes for bear habitat selection (Arif and MacNeil 2022), we note that both of our models indicated that bears avoided areas with higher human density (Tables 2 and 3). Furthermore, we did not observe a single bear use of high-intensity development areas (and no low- and medium-intensity development areas were used in NAL). However, as human densities continue to increase in Alabama and elsewhere and developed areas expand further into current bear population range, bears may be required to adapt and become more comfortable around areas with a heavy human presence. Such changes in habitat use will be especially important in the MRB if bear populations persist, as human populations in the area are growing quite rapidly and expanding into areas currently occupied by bears (i.e., Saraland, AL; U.S. Census Bureau 2010; T Steury, personal observation). Additionally, just as bear habitat use may change in response to increasing human densities, bear habitat use may change as bear density increases as well. The NAL population is newly recolonized, and the area has likely not yet reached carrying capacity. As the bear population continues to grow in number and increase in density, resource use and allocation will likely begin to shift (Sinclair 1992; Frary et al. 2011). Understanding more about how resource selection changes in response to changing bear and human population density could help in making accurate predictions about the spatial expansion of the species. Such knowledge will be especially important given that as bear (and other animal) populations grow and expand outwards from protected areas with low human density, population ranges may begin to encroach on more human-dominated landscapes, which increases the chances of negative human-wildlife interactions.

Author contributions

Hannah Jane Leeper (Data curation, Formal analysis, Investigation, Methodology, Project administration, Visualization, Writing—original draft), Todd Steury (Conceptualization, Funding acquisition, Methodology, Project administration, Supervision, Validation, Writing—original draft), and Chris Seals (Conceptualization, Data curation, Methodology)

Funding

Funding for this project was provided by the Alabama Department of Conservation and Natural Resources, and we especially thank Traci Wood.

Conflict of interest.

None declared.