Attempts to understand wildlife population dynamics and implement management practices benefit from understanding animal behavior traits. In Alaska, USA, grizzly bear (Ursus arctos) behavior is important to understand because the species is an apex predator, exerts top-down population effects, and is a cornerstone species across complex landscapes. Our objectives were to examine how environmental and anthropogenic factors affect behavior patterns in grizzly bears in south-central Alaska. We hypothesized that, for a given sex, the time spent by bears resting, foraging, and traveling are similar and show consistent seasonal shifts as resource availability changes throughout their nondenning season. Additionally, we hypothesized that males spend more time traveling than do females because of differences in home range sizes, metabolic needs, and the rearing of cubs associated with females. We used a hidden Markov model (HMM) to test our hypotheses and examine how various dynamic, static, and temporal variables affected bear (n = 12) behavior during the summers of 2021–2022. Males spent the most time foraging and the least time resting while females spent the most time resting and the least time foraging. These patterns shifted as the summer progressed with increases in traveling and foraging and decreases in resting for both sexes. Bears were found to be most likely in a resting state at hotter temperatures and more likely to be traveling at colder temperatures. Additionally, bear behaviors deviated depending on elevation, whereby bears were foraging at higher elevations and resting or traveling at lower elevations. Our findings indicate that male–female differences in behavior are seasonally dependent, being similar in autumn and opposite during the postdenning period. In addition, we see evidence that changes in environmental conditions, such as warming, can have direct effects on behavior. Further studies should explore how future environmental and anthropogenic factors such as predicted changes in climate and increases in land-use changes can affect bear behavior and subsequent demographic effects.
Effective management or conservation of any animal species benefits from an understanding of that species' behavior in natural environments. Studies of this type contextualizes how animals move and react to environmental variability, such as inter-specific and intraspecific interactions and how they learn about and use their habitat (Manning and Dawkins 2012, Breed and Moore 2022). Both extrinsic and intrinsic factors can affect animal behavior. Externally, behaviors can be modified by environmental conditions, resource availability, other animals, or anthropogenic disruptions (Alcock 1975). Internally, behaviors may change as a result of the complex interactions of hormones, the animal's circadian rhythm, hunger, or fear (Rubenstein and Alcock 2019). Documenting these behavior traits are becoming increasingly important as landscapes are undergoing major changes, such as vegetation shifts due to warming (Myers-Smith et al. 2020) and or snowfall patterns or fire frequency (Hessilt et al. 2024) that may in turn alter the resource base for co-occurring ungulate and carnivore populations (Stanek et al. 2010, Brockman et al. 2017).
Quantifying animal behavior and how external or internal factors can affect wildlife may provide useful insights to wildlife managers (Sutherland 1998). For example, a study by Shier (2006) showed that kin behavior resulted in translocated black-tailed prairie dog (Cynomys ludovicianus) family groups being 5 times more likely to survive and reproduce than were translocated nonfamily groups because of kin behavior. Additionally, by quantifying animal behavior and movement patterns, managers can better understand certain movement drivers such as seasonal changes in vegetation, prey availability, presence of competitors, and the distribution of possible mates. Expanding the understanding of animal behavior is especially important because of rapid land use changes and disturbances; examples include large-scale resource extraction, large wildfires, and extreme weather events, such as excessive autumn or spring snow events and the ongoing rapid warming of the Boreal and Arctic regions (e.g., Taylor et al. 2017, Naidoo and Burton 2020, Beery et al. 2021, Rantanen et al. 2022).
Recent advances in global navigation satellite systems (GNSS; Tomkiewicz et al. 2010, Wilmers et al. 2015, Johansson et al. 2016) combined with the increase in computing power and open-source statistical software to analyze these data sets have allowed for extensive improvements in the study of animal movement and behavior (Seidel et al. 2018, Joo et al. 2020). Many different types of movement analyses can be done with high-resolution GNSS data (e.g., utilization distributions, resource selection functions, step selection functions, etc.) with hidden Markov models (HMMs) among the most inferentially powerful. Hidden Markov models have been used widely in human biology from speech recognition (e.g., Inoue et al. 2011, Wu et al. 2014) to DNA sequencing (e.g., Chao et al. 2013, Wu et al. 2021), and in recent years have been applied to animal tracking data (e.g., van Beest et al. 2019, Beumer et al. 2020, Farhadinia et al. 2020, Saldanha et al. 2023). A HMM is a state-space model that allows for the interpretation of hidden behavior states based on observed movement data collected with high-frequency GNSS (Zucchini et al. 2016, Leos-Barajas and Michelot 2018). Importantly, HMMs allow for the interpretation of behaviors in relation to both internal and environmental factors (McClintock et al. 2020), such as the level of hunger in leopards (Panthera pardus) or distance to the nearest road in common genets (Genetta genetta; Farhadinia et al. 2020, Ferreira et al. 2022).
Here, we describe an application of HMMs to GNSS data collected from a terrestrial omnivore in south-central Alaska, USA, the grizzly bear (Ursus arctos). Recent studies of grizzly bears using GNSS collar data and cameras (e.g., Deacy et al. 2016, Brockman et al. 2017, Hertel et al. 2019) have revealed a suite of states that constitute major patterns of behavior. Bears typically partition their time between resting, foraging, and traveling punctuated by event-based behaviors, such as hunting and mating periods (Brockman et al. 2017). The proportion of time individuals spend in these behavior states can, however, vary, reflecting life-history traits and individual knowledge or experience (e.g., young or mature bears, lone females, females with cubs; Lamb et al. 2020). Male grizzly bears are known for having larger home ranges than female bears (Mace et al. 1996, Mangipane et al. 2018) and so may spend more time traveling or traveling faster than do females. Landscape properties, environmental factors, or anthropogenic disturbances may also influence the frequency or cumulative time that an individual resides in these behavior states. Pigeon et al. (2016b) found that food availability and weather events affected when grizzly bears enter and exit their dens. Additionally, spring snow cover can dictate grizzly bear habitat selection following den emergence (Berman et al. 2019). Anthropogenic disturbances, such as vehicle traffic, road density, and human outdoor recreation have also resulted in deviations from typical bear behavior (e.g., Northrup et al. 2012, Ladle et al. 2018, Parsons et al. 2021). Ladle et al. (2019) found that grizzly bears alter their behaviors in response to human outdoor trail and recreation use, leading to a potential decrease in preferential foraging for bears. Additionally, a study by Gibeau et al. (2002) revealed that female adult grizzly bears are vulnerable to anthropogenic disturbances and will choose to avoid humans over selecting high-quality habitats.
Our objectives are to examine how a suite of dynamic, static, and temporal factors affect behavior patterns in grizzly bears. We hypothesize that within a given sex, the time spent by grizzly bears resting, foraging, and traveling are similar between individuals and show consistent seasonal shifts as resource availability changes throughout their nondenning season. Additionally, we hypothesize that male grizzly bears will spend more time in a traveling behavior state than will females because of differences in home range sizes and metabolic needs. An understanding of how grizzly bears respond to environmental and anthropogenic factors could improve management and conservation initiatives by increasing the ability to predict how individuals and populations respond to environmental or anthropogenic changes.
Study area
This study focuses on fine-scale behaviors of grizzly bears in south-central Alaska, primarily in Game Management Unit 13A (GMU 13; Fig. 1). This unit encompasses most of the Nelchina Basin, which is approximately 11,200 km2. This area is characterized by cold winters (January average –29°C), warm summers (July average 19°C), and little precipitation (yearly average of 28.4 cm; Walton et al. 2013). GMU 13 is bounded by the Chugach Mountains to the south, the Talkeetna Mountains to the west, the Alaska Range to the north, and the Wrangell and St. Elias Mountain ranges to the east. Its geographic locations and topographic variability allow the Nelchina Basin to support a diverse assemblage of natural communities (Barrett and Christensen 2011). Forest communities are dominated by white spruce (Picea glauca), black spruce (P. mariana), and dwarf birch (Betula nana). Riparian communities are composed of riverine willows (Salix alaxensis and S. hastata) with subalpine communities comprising various dwarf ericaceous and berry producing shrubs (Cassiope spp., Empetrum spp., Rhododendron spp., Vaccinium spp., and Arctostaphylos spp.; Skoog 1968). Other large mammals include American black bears (U. americanus), wolves (Canis lupus), Canada lynx (Lynx canadensis), wolverines (Gulo gulo), Dall's sheep (Ovis dalli), moose (Alces alces), and caribou (Rangifer tarandus).
Fig. 1.
Location of the study area in south-central Alaska, USA. The grizzly bear (Ursus arctos) study took place within Game Management Unit 13 in Alaska during the summers of 2021–2022. Lines indicate individual grizzly bear movements throughout their nondenning season.
The Nelchina Basin is considered to be high-quality grizzly bear habitat. In addition to abundant berries and herbaceous resources, the Nelchina caribou herd, estimated at 44,500 in 2020, is an important food source for grizzly bears in the area, as are the 5 Pacific salmon species (Oncorhynchus keta, O. kisutch, O. nerka, O. tshawytscha, O. gorbuscha) present in the major rivers of GMU 13. More details on the topography, climate, and plant and animal communities of the study area can be found in Skoog (1968).
This GMU provides significant food resources for Alaskans and is easily accessible to recreational and subsistence hunters; therefore, game populations in this area are intensively managed by the Alaska Department of Fish and Game (ADFG [Brockman et al. 2017]). In addition, grizzly bears in GMU 13 can be legally harvested throughout the entire year with a bag limit of 1 bear per regulatory year. During 2008–2013 an average of 141 bears were harvested within GMU 13 per year (Harper and McCarthy 2015). Additionally, bear baiting is allowed within the unit from 15 April to 30 June every year. Typical bear bait in this area consists of dry dog food, popcorn, and stale donuts or pastries (H. Hatcher, ADFG, personal communication, 2022).
Methods
Capture and collaring of grizzly bears
Grizzly bears (n = 12; 7 females, 5 males) were captured and fitted with GNSS collars (Vertex Plus Iridium GPS [Global Positioning System] collars) from October 2019 to May 2021. We located bears using a fixed-wing aircraft and immobilized them via a dart system with Telazol® (tiletamine hydrochloride and zolazepam hydrochloride; Fort Dodge Animal Health, Overland Park, Kansas) from an R-44 helicopter (Robinson Helicopter Company, Torrance, California, USA). We immobilized bears using 3–10-mL darts with 1 1/8-inch–1 1/2-inch needles. While the bears were immobilized, we recorded their sex, took multiple morphometric measurements, fit a GPS collar, and collected tissue samples (i.e., hair, serum, and blood). We fitted collars for each animal tightly enough to deter bears from removing them while still allowing for growth. Collars were programmed with an electronic drop-off to release the collar at 160 weeks postdeployment. We lip-tattooed all newly handled bears with a unique identifier. All chemical capture, marking, animal handling, and sample collection were completed using protocols approved by ADFG Institutional Animal Care and Use Committee (IACUC protocol no. 0094-2019-66) and the protocol outlined in the ADFG-Division of Wildlife Conservation bear capture manual.
Collars were programmed to collect and store locations, ambient temperature, and elevation every 4 hours at the time of deployment, and then hourly starting in June of 2021. We truncated stored location data to 4-hour intervals and uploaded them via satellite, allowing animal locations to then be downloaded and remotely monitored. When mortalities occurred, as indicated by lack of collar movement for 24 hours, we investigated them as quickly as possible to determine the cause of death and to retrieve collars. Additionally, we monitored and observed any females with cubs by fixed-wing aircraft twice per month throughout the nondenning period to evaluate cub survival.
Processing of location, environmental, and anthropogenic data
To obtain hourly location data, collars had to be retrieved from the field and the data manually downloaded. We processed and screened hourly location data for any impossible movements or inaccurate locations following the methods of D'Eon et al. (2002), and removed locations during bear denning season (Table 1). Any missing locations due to collars failing to retrieve precise locations (NAs; 0.02% of location data) were linearly interpolated using the package “adehabitatLT” in Program R (Calenge 2006, R Core Team 2020, RStudio 2020).
Table 1.
Details of the 12 grizzly bears (Ursus arctos) from south-central Alaska, USA, that are included in the hidden Markov model from the summers of 2021–2022. * Indicates a female with cubs.
Covariates considered for the analysis represented environmental and anthropogenic variables that are known from research in the area to influence bear movements and behaviors (e.g., Brockman et al. 2017, Rogers 2021). Variables included a range of static, dynamic, and temporal covariates that characterized the landscape well and were readily available (Table 2). Static covariates included land-cover type, elevation, terrain ruggedness, hillshade, and distance to roads. We derived and calculated static covariates using digital elevation models (DEM; 100-m resolution [U.S. Geological Survey 2017]), Alaska Vegetation and Wetland Composite (30-m resolution [Alaska Center for Conservation Science 2017]), and Alaska Department of Transportation and Public Facilities functional classification shapefiles (Alaska Department of Transportation and Public Facilities 2013). These factors can influence food availability and quality, spring snow depth and cover, and potential avoidance of certain areas (e.g., roads; Ciarniello et al. 2007, Boulanger and Stenhouse 2014, Berman et al. 2019). Temperature, a dynamic covariate, was derived from individual GPS collars. Temperature plays an important role in the physiology of bears, what resources are available to them, and can influence how bears behave (Pigeon et al. 2016a). Temporal covariates included time of day, Julian day, month, and whether or not legal moose or caribou hunting were open within the game management unit. Temporal covariates can reflect things like food availability, abundance, and visibility in addition to mating season, hormone levels, and the ability to move across the landscapes (Munro et al. 2006, van Manen et al. 2019). We extracted all covariates and processed them in Program R (version 4.0.3 [R Core Team 2020, RStudio 2020]) or ArcGIS Pro (version 2.8.1 [Esri 2020; Table 2]).
Table 2.
Summary of the covariate types, descriptions, and significance that were used in the hidden Markov model on high-resolution grizzly bear (Ursus arctos) Global Positioning System (GPS) location data during the summers of 2021–2022 in south-central Alaska, USA.
Statistical analysis (HMM)
We analyzed individual bears' behaviors via a HMM as implemented in the Program R moveHMM package (Michelot et al. 2016). A HMM is a time series model with 2 components. The first component consists of a series of observations (e.g., GNSS location data), and the second component is an underlying nonobservable state sequence (i.e., hidden state—what they are doing). The hidden state is serially correlated and can take on one of a number of finite, discrete values. Every observation is derived from a hidden state via a probability distribution. A hidden state (St) determines which probability distribution generates the observation (Zt), so the hidden state results in the observations. The fundamental assumptions of HMMs are that hidden state processes exist but are not observed, yet result in the observations. Another important assumption is that the observation is independent given the underlying state. This independence means that it is likely that the probability distribution selected at a given time (e.g., T + 1) is the same as the previous one. The hidden state sequence is constructed via a Markov chain, which has 2 components: the initial state distribution, and the collection of transition probabilities among the states. The initial state distribution (Sk = 1) is the process that the system is in during the first time-step and is calculated as a probability vector (δ(k)), of length N, whose superscript k = 1 denotes the first time-step and whose elements, δ(i k), denote the probability of being in a state i at time-step k = 1:
Transition probabilities give the probability of succession between states from time-step k to k + 1 and are calculated as
where c estimates the probability of transitioning to state j if a state is currently in state i. Transition probabilities from the elements of the transition matrix can be used to derive all future states if given the initial state and an unchanging transition matrix, C. Here, the ith row and jth column of C give the transition probability of a state changing from S = i to S = j in a single time-step. The diagonal elements give the probability of remaining in a certain state during one time-step:
For more details regarding HMMs, see Zucchini et al. (2016).
We modeled behavior states using the function fitHMM() in the moveHMM package (Michelot et al. 2016) through the time series of step lengths and turning angles of the location data. We modeled step lengths with a gamma distribution and modeled turning angles with a von Mises distribution. We selected a 3-state model among 3 models: 2-state, 3-state, and 4-state. The Akaike Information Criterion (AIC) favored the 4-state model, but we decided on the 3-state model because it is more biologically feasible and interpretable, such as those used by Beumer et al. (2020) to describe behaviors in high-arctic muskoxen (Ovibos moschatus) and Farhadinia et al. (2020) to examine behavior and decision making of leopards. We used the Viterbi algorithm (i.e., a recursive scheme to decode behavior states [McClintock et al. 2020]) to interpret the behavior states from the fitted models.
We based behavior interpretations of the 3 states on step length and turning angles. Consisting of very short step lengths and minimum turning angles, State 1 was considered a “resting” state. State 2 was considered a “foraging” state because of its relatively short steps and many turning angles to account for less persistent directionality. With longer step lengths and fewer turning angles showing a more persistent direction, State 3 was determined to be a “traveling” state (Table S1, Supplemental material).
To investigate the influence of environmental and anthropogenic covariates on behavior states, we built each covariate into the model as an additive variable. We tested all covariates for collinearity, with only those with correlations of |r|, 0.7 included and standardized in Program R to allow for a better model fit prior to analysis. We evaluated numerical covariates for normality and log-transformed them (if strictly positive) or square-root-transformed them (if containing negatives) if skewed to reduce their leverage. We separated categorical covariates into binary predictors and assessed them independently from one another. We first considered single covariates to select candidate variables based on AIC values compared with the null model AIC. Covariates that reduced AIC by .2 were retained for further analysis (Arnold 2010). We followed this forward selection process using AIC to evaluate the remaining covariates. We estimated stationary state probabilities for each covariate via a multinomial logit link to examine how a certain covariate could affect switching from one state to another. In addition, we calculated the time spent in each behavioral state based on the stationary state probabilities.
To ensure model convergence and fit, we applied many initial starting values (i.e., 30) as vectors of step and angle distributions, and selected those with the highest log-likelihood. A zero-inflation parameter accounted for step lengths of zero. We assessed goodness-of-fit via pseudo-residuals of the step lengths from the fitted model (Michelot et al. 2016, Zucchini et al. 2016). We checked the adequacy of the model via the normality of the pseudo-residual plots and by comparing the qq-plots of the pseudo-residuals against the theoretical quantiles (Michelot et al. 2016). We ran 2 separate models based on sex.
Results
Allocation among Markov behavioral states
We tracked 5 male and 7 female grizzly bears for an average of 247 days (range = 37–381) with a total number of fixes averaging 3,178/individual (range = 870–4,941; Table 1). Based on AIC, the HMM results supported 7 variables (time of day, Julian day, terrain ruggedness, elevation, distance from road, temperature, and hunting season; Table 3) and 3 behavior states (Table S1) for both males and females (Fig. 2). State 1, interpreted as a “resting” state, was characterized by very short step lengths (near-zero) with a mean turning angle near 180° and with mid concentration (i.e., a shape parameter of a Von Mises distribution) values. State 2, interpreted as a “foraging” state, was characterized by medium step length with a mean turning angle near zero, but with a more uniform distribution of concentration than the other states. State 3, interpreted as a “traveling” state, was characterized by long step length with more turning angles concentrated near zero indicating purposeful directed movements. Our examination of the qq-plots (Figs. S1 and S2, Supplemental material) of the pseudo-residuals against the theoretical quantiles for both male and female step lengths and turning angles, suggested that the model goodness of fit was adequate.
Table 3.
AIC (Akaike Information Criterion) and ΔAICa for a 3-state hidden Markov model with different covariates for grizzly bears (Ursus arctos) during the summers of 2021–2022 in south-central Alaska, USA. The model in bold was the most supported.
Fig. 2.
Histograms of step lengths and turning angles overlaid with the colored behavior states as estimated by a hidden Markov model for male (n = 5) and female (n = 7) grizzly bears (Ursus arctos) during the summers of 2021–2022 in south-central Alaska, USA. The dotted line shows the full model.
When all summer (Jun–Sep) observations were pooled by sex, males and females both spent roughly equal (29–35%) time in each behavior state (Fig. 3), although overall males spent the most time in the foraging state (35%) while females spent the most time in the resting state (38%). Males spent the least amount of time in a resting state (32%) while females spent the least amount of time in a foraging state (29%). Males spent more time moving (foraging or traveling 68% of the time, or odds = 2.1) than did females (62% time, odds = 1.6). Thus, the odds ratio suggests that males were 1.3 times (= 2.1/1.6) more likely to move than were females. There was, however, variability among individuals in the sample, with some females spending more time foraging and traveling than did some males (Fig. S3, Supplemental material).
Fig. 3.
Proportions of time spent in each behavior state for grizzly bears (Ursus arctos) during the summer (Jun–Sep) of 2021 and 2022 in south-central Alaska, USA.
Overall, bears showed a bimodality in their activities over the course of a day (Fig. S4, Supplemental material) and this pattern shifted over the course of the summer, with bears spending more time traveling and foraging as the summer progressed. On average, the bears were most active (i.e., not resting) in the early morning to afternoon (hours 6 to 13) and late evening (hours 19 to 23). Similarly, bears rested over 2 time periods, after midnight for about 5 hours (hours 0 to 5) and again in the late afternoon (hours 14 to 18). Males and females differed in their seasonal trends of behavior. As the summer progressed (i.e., from May to Aug) the time males spent traveling increased by 89%, foraging increased by 44%, and resting decreased by 69% (Fig. 4). For females, there was a 66% increase in traveling, a 41% decrease in resting, and foraging stayed consistent as the summer progressed (Fig. 5). The trend of decreasing resting led to a virtual loss of the bimodality of resting such that most of the day from hours 5 to 22 was spent either foraging or traveling for the months of July–September.
State occupancy in relation to environmental and anthropogenic variables
The probability of bears being in 1 of 3 behavior states was best explained by the full model (Table 3) where variables time of day and Julian day were the most important variables as determined by AIC in the forward selection process, followed by terrain ruggedness, elevation, distance from road, temperature, and hunting season. These variables differed in their effect on behavior by sex for some but not all variables.
For both sexes, as terrain ruggedness increased, the probability of being in a foraging or resting state increased and the probability of being in a travelling state decreased (Fig. S5, Supplemental material). Elevation was the fourth most important variable in relation to the state probability for both sexes (Table 3). As elevation increased the probability of being in a traveling state increased for both males and females. At lower elevations, both sexes were most likely to be in a foraging state. Elevation did not have a significant effect on the resting state (Fig. 6). Males and female bears were most likely to be in a foraging state (Fig. 6; 54% probability for males; 50% for females) while at their average elevation (Fig. S6, Supplemental material; mean = 1,119 m above sea level [asl] for males; mean = 878 m asl for females). For both sexes, the distance from the nearest road was the fifth most important variable on the state probability (Table 3). The distance to the nearest road affected male bear behavior more than female behavior (Fig. 6). When males were the closest to roads, they were most likely to be in a traveling state and least likely to be resting. At the furthest distance from roads, males were most likely to be foraging and least likely to be resting. At the average (Fig. S6; mean = 6.2 km) and further distances from roads, male bears were most likely to be foraging and least likely to be resting. For females, the distance to the nearest road did not have a large impact on their behavior (Fig. 6). When at the average distance (Fig. S6; mean = 4.1 km) from roads or closer, females were most likely to be in a foraging or traveling state and least likely to be resting. For both sexes, temperature was the sixth most important variable in relation to the state probability (Table 3). As the temperature increased, foraging and resting states increased while traveling decreased (Fig. 6). At the lowest temperatures (–10°C; Fig. S6) bears of both sexes were most likely to be in a traveling state and least likely to be resting (Fig. 6). At the average temperature (17°C for both sexes; Fig. S6) both males and females were most likely to be in either a foraging or traveling state (Fig. 6). Lastly, for both sexes, hunting season was the seventh most important variable in relation to the state probability (Table 3). Hunting season did not have a large effect on male behavior, whereas females were most likely to be in a traveling state during hunting season and most likely to be in a foraging state when it was not hunting season (Fig. S5).
Fig. 6.
Stationary state probability (mean and 95% confidence interval) of activity state occupancy as a function of elevation, temperature, and distance to road for male (left) and female (right) grizzly bears (Ursus arctos) during the summers of 2021–2022 in south-central Alaska, USA. State 1 is resting, State 2 is foraging, and State 3 is traveling.
Discussion
Implementation of HMMs with high-frequency GPS location data allowed us to explore and gain insights pertaining to movements of grizzly bears in the 11,000-km2 Nelchina Basin of south-central Alaska. We were able to successfully infer 3 distinct behavior states (i.e., resting, foraging, and traveling) for 12 grizzly bears. Based on our dynamic, static, and temporal covariates we found that time of day, Julian day, elevation, distance to roads, and temperature affected grizzly bear behaviors more than did terrain ruggedness or timing of hunting seasons.
Results here show that overall grizzly bears spent approximately equal amounts of time in each behavior state, but females spent around 13% more time resting and 18% less time foraging than did males. This may be due to males' larger size and consequent greater metabolic needs that require more time spent foraging (Dahle and Swenson 2003, Bjornlie et al. 2014). Male bears also tend to have larger home ranges, display farther dispersal distance behaviors, and roam farther to find mates than do females, all of which could contribute to their increased foraging time (McLellan and Hovey 2001, Zedrosser et al. 2007, McArt et al. 2009, Edwards and Derocher 2015).
All bears showed a bimodal activity pattern throughout the day (i.e., most active in the morning and/or early afternoon and late evening. These patterns have been documented in bears before via high-frequency GPS location data (e.g., Heard et al. 2008, Donatelli et al. 2022) and are suspected to be either a thermoregulatory strategy or possible avoidance of human activity (Ordiz et al. 2017). By being most active during the morning and late evening, bears can avoid the hottest parts of the day and maintain thermoneutrality by seeking shade or water bodies and decreasing activity (Pigeon et al. 2016a). Furthermore, by being active during the morning and late evening bears can avoid an influx of potential anthropogenic disturbances. Within our study area anthropogenic disturbances (e.g., hiking, camping, hunting) have been increasing and resulting in an increase in bear–human encounters (Harper and McCarthy 2015). By selectively choosing when to be active bears may be able to avoid both heat and human disturbances.
As the summer progressed, all bears showed a decrease in resting and an increase in foraging and traveling. We hypothesize this is due to bears' seasonally accelerating need to forage sufficiently to accumulate fat reserves in preparation for hibernation, a process known as hyperphagia (Jansen et al. 2019). During hyperphagia, bears can consume up to 20,000 kcal/day and spend much of their time searching for and consuming food (Schulte et al. 2021, Pérez-Girón et al. 2022), leading to less time resting and more time foraging and traveling. Additionally, as the summer progresses a variety of different food resources start to become available, including an assortment of berry species, spawning salmon, and abundant herbaceous forage. The wide variety and abundance of different food resources available to the bears as the summer progresses could explain the increase in activity as seen with Kodiak brown bears (U. a. middendorffi) tracking the phenology of spawning salmon (Deacy et al. 2016) and American black bears in Montana, USA, tracking photosynthetic green-up (Mace et al. 1999). Furthermore, as the summer progressed, hours 22 to 5 were predominantly spent resting by both females and males (Figs. 4 and 5). We hypothesize this is due to decreasing daylight as the summer progresses. Being most active throughout the daytime allows for more efficient foraging and traveling (McLellan and McLellan 2015).
Elevation, temperature, and distance to roads had a strong influence on the 3 behavior states for grizzly bears. At lower elevations (,1,000 m asl) both male and female bears were most likely to be foraging; as elevation increased traveling increased rapidly and foraging decreased. Miller et al. (1997) found that bears in our study area rarely used higher elevation sites (.1,524 m asl). However, we found that bears here, and males in particular, used higher elevations more than previously thought. We hypothesize that male use of these higher elevations areas (mean = 1,119 m asl, range = 600–2,100 m asl) is likely influenced by multiple factors, such as seasonal caribou availability and potential caribou gut piles left over from hunters. Caribou generally move to higher elevations as summer progresses, thereby avoiding heat and biting insects while foraging on vegetation responding to snowmelt. Bears may be following this important prey base (Ion and Kershaw 1989). Additionally, at higher elevations, there is a decrease in vegetation forage diversity and abundance so bears may travel longer distances to obtain food while at higher elevations or perhaps travel more to get back down to lower elevations. At mid elevations (,1,400 m asl) both sexes were more likely to be in a foraging state. Mid elevations in GMU 13 tend to support a variety of berry-producing shrubs and tundra forbs for consumption, as well as habitat of moose and small mammals, especially Arctic ground squirrels (Urocitellus parryii; Skoog 1968).
The role of ambient air temperature on behavior was suggested by the observation that at temperatures .20°C bears were most likely to be in a foraging or resting state, switching to either traveling or foraging at temperatures,20°C. Pigeon et al. (2016a) found that habitat selection by grizzly bears in Alberta, Canada, was greatly influenced by ambient temperatures and when temperatures were at the warmest bears chose locations that would provide thermal shelter while also providing foraging opportunities. Additionally, temperature can play an important role and can act as a proxy for forage and vegetation availability, its phenology, and how it changes with increasing temperatures. When plants are exposed to increasing temperature their phenology can accelerate, leading to shorter growing periods and potentially less quality forage for bears (Hollister et al. 2005, Mann et al. 2021). However, these temperature effects can depend on prior year snow conditions (Welker et al. 2005, Kelsey et al. 2023).
Further, we found that colder temperatures affected female behavior more than male behavior. Female bears were 22% more likely to be in a traveling state at extremely cold temperatures than were males. We hypothesize this is due to larger body size and greater body fat in males than females (Cameron et al. 2020), therefore enabling males to maintain a constant body temperature more readily than females.
We found that the distance to the nearest road affected male bear behavior more than female behavior. When male bears were closest to roads, they were more likely to be in a traveling state. McLellan and Shackleton (1988) found that young female bears used habitats that were closer to roads than those used by older male bears. In our study sample, 2 younger female bears heavily used areas around a major highway. When female bears were nearer roads, they were more likely to be in a foraging or traveling state and least likely to be resting. Roads can be an attractant for some bears because they can provide movement corridors and quality forage, such as a higher frequency of ants, berries, and sedges (Roever et al. 2008) and/or ungulate road kills providing scavenging opportunities. Roads can also pose a barrier for some bears and can alter behaviors and movements in areas with high road densities or high traffic volumes (Waller and Servheen 2005).
Caribou and moose hunting season was included in the full model but did not have a large impact on bear behavior. Moreover, hunting itself added little to the best model fit with a ΔAIC,10 among models where the min(AIC) .50,000, resulting in a weak effect on bear behavior states. Our results did show that females during hunting season were most likely to be in a traveling state while males were most likely to be in a foraging state. The probability of being in a resting state decreased for females based on whether it was hunting season or not but did not affect the male's resting state. Many studies have been done on Scandinavian brown bears that have shown that bears will avoid areas that are associated with hunting and that hunting seasons can result in different foraging strategies for bears (Ordiz et al. 2012, Hertel et al. 2016, Lodberg-Holm et al. 2019). GMU13 receives a high-density influx of hunters with approximately 4,050 caribou and 850 moose being harvested each year (Robbins 2019, Hatcher and Robbins 2021), which may alter bear behaviors and preferred foraging areas.
Strengths, limitations, and future research
Our findings from this analysis depict the power of high-frequency location data in movement models of grizzly bears and are the first of their kind for bears in south-central Alaska depicting seasonal and sex-based movement differences. Although the limited number of individual animals used in this analysis does not fully capture population-wide trends nor individual patterns in this bear population, our sample size was able to distinguish important behavior discoveries for an apex predator. A larger sample size would, however, provide greater detail of among-individual, between-sex, and age-based variability than our study can report. Additionally, employing GNSS collars that have a video camera attached (e.g., the Lotek InSight Video Camera Module; Lotek Wireless, Newmarket, Ontario, Canada) could verify how well the model is able to parse out various behavior states. Another meaningful addition to this analysis would be to include a measure of primary plant productivity (i.e., Normalized Difference Vegetation Index, NDVI; Pettorelli et al. 2005). High-frequency NDVI data paired with the GNSS location data may be able to provide insights into how bears react to plant phenology on a landscape and fine-scale level (Kelsey et al. 2023), including lag effects of berry-producing plants (e.g., Stetz et al. 2019). Lastly, an ongoing challenge within GMU 13 is to ensure the sustainability of wildlife populations in an area of intensive human use and to quantify anthropogenic disturbances with higher spatial and temporal resolution. There is an extensive off-road vehicle (ORV) trail system throughout GMU 13, but the extent and usage of the trails by either bears or recreationalists, including hunters, is currently unknown. Many recreationists and hunters access areas in GMU 13 by boat, airplane, and ORV (Alaska Department of Fish and Game 2015). High-use road systems have been shown to deter bears from high-quality foraging areas (Mace et al. 1996) leading to potentially harmful impacts on body condition and survival (Boulanger et al. 2013). A recent study by Kearney et al. (2021) in the Yellowhead Bear Management Area in Alberta used the combination of a “network-based” approach via social media and an “image-based” approach via satellite imagery to estimate human use of unpaved roads within critical grizzly bear habitat. Adopting a similar strategy in GMU 13 would prove beneficial to quantifying human use on the landscape and could provide similar insights on how grizzly bears respond to anthropogenic stressors.
Our findings show that GNSS-data can inform HMMs and suggest how environmental and anthropogenic factors affect grizzly bear behavior. With current climate predictions and increases in human disturbances and landscape use, knowing how bears react to various factors will provide a fuller understanding of grizzly bear behavior ecology and so inform management and conservation with future implications.
Acknowledgments
We thank the Alaska Department of Fish and Game, JMW's UArctic Research Chairship, and the University of Alaska Anchorage, Biological Sciences, for funding this project. Additionally, we thank C. Heun, R. Sattler, J. Holyoak, K. Wall, H. Hatcher, J. Hepler, N. Jensen, and J. Mortenson for the collection of data. We would also like to thank the many pilots in this project for their knowledge and expertise—in particular, T. Cambier, J. Lee, M. Meekin, H. McMahan, L. Williams, and E. Finch. Lastly, we would like to thank S. Crimmins for his early review of the manuscript; and the reviewers, Associate Editor, and Editor-in-Chief for their thoughtful comments and suggestions.
This research was funded in part by JMW's UArctic Research Chairship, the National Science Foundation Major Research and Instrumentation award (0953271) to JMW to establish the Environment and Natural Resource Institutes' Stable Isotope Lab, which undertook our δ13C and δ15N analyses, thanks to J. Ferguson and University of Alaska Anchorage's laboratory analyses (IRBNet #1661273-1). Field data collection, grizzly bear captures and monitoring, and associated expenses were supported by Pittman-Robertson funds from the Federal Aid in Wildlife Restoration grant AKW-R-11-2019 and Alaska Department of Fish and Game's Fish and Game Fund.
Literature cited
Appendices
Supplemental material
Fig. S1. Hidden Markov model diagnostic output to determine whether the model properly fit for male grizzly bears.
Fig. S2. Hidden Markov model diagnostic output to determine whether the model properly fit for female grizzly bears.
Fig. S3. Proportion of time spent in each behavior state for individual grizzly bears during their nondenning season in south-central Alaska, USA. * Indicates a female with cubs.
Fig. S4. State time budget averaged for male (n = 5) and female (n = 7) grizzly bears during their nondenning season in south-central Alaska, USA, and the frequency of being in 1 of 3 behavior states during a given hour of the day.
Fig. S5. Stationary state probability (mean and 95% confidence interval) of activity state occupancy as a function of Julian day (JDAY), Terrain Ruggedness, and Hunting Season for male (left) and female (right) grizzly bears in south-central Alaska, USA. State 1 is resting, State 2 is foraging, and State 3 is traveling.
Fig. S6. Histograms of the variables included in the hidden Markov model. Red line is showing the mean of the variables for male (left) and female (right) grizzly bears.
Table S1. Starting parameters for hidden Markov model for step lengths and turning angles for male and female grizzly bears in south-central Alaska, USA. State 1 is resting, State 2 is foraging, State 3 is traveling.
