Our objectives were to describe the denning chronology of brown bears near the northern extent of their North American range and identify links among intrinsic characteristics (e.g., reproductive status, age, and body mass), environmental conditions, and the timing of denning behavior as a necessary first step to identifying the potential consequences of climate change and proposed industrial development on this behavior. We hypothesized that (1) sex, reproductive status, and stored energy resources, or (2) a combination of intrinsic factors and environmental covariates associated with diet and weather would explain the timing of den entry and den exit. We predicted that (1a) gestating females would enter dens earlier and exit dens later and (1b) body fat would explain date of den exit because stored energy would allow longer denning duration. We also predicted that (2a) low ambient temperatures and late snowmelt in spring would be associated with late den exits, and (2b) bears that used salmon in autumn would den later due to increased availability of energy-rich food resources.

 

 

We captured and collared 51 brown bears (32 females and 19 males) between 2014 and 2016 using aerial darting techniques. Capture and handling techniques were in accordance with guidelines of the American Society of Mammalogists (Sikes and Gannon 2011) and were approved by USGS and National Park Service Institutional Animal Care and Use Committees (IACUCs; 2014-1 and 2014A2, respectively). We fitted bears with Telonics (Mesa, AZ) Gen IV Global Positioning System collars. The collars were programmed to obtain a location every 90 minutes between March 15 and November 15 and once a day between November 16 and March 14. A tri-axial accelerometer housed in the collar was programmed to record activity during the five-minute period immediately after a GPS location was recorded.

 

 

We identified den sites following the methods of Sorum et al. (2019a), which involved locating clusters of GPS locations prior to den entry and then visiting a subset of dens for validation (Pigeon et al. 2014). Across 33 validated dens, Sorum et al. (2019a) found that the average den location determined from GPS collars was within 25m (SE = 9m) of the actual den location.

 

The timing of den entry and exit is often inferred from GPS collar data. However, bear behaviors such as lingering near the den site before and after the denning period can decrease the reliability of location data as an indicator of the timing of hibernation. Similarly, some collared bears in dens can continue to record GPS locations in shallow dens or with favorable arrangements of overhead GPS satellites. To address such shortcomings, we used both GPS fix rate and accelerometer activity sensor readings to infer the beginning and end of the denning period. Because the patterns of these variables can be complex due to situations like a bear briefly entering a den, appearing to enter hibernation, then exiting and re-entering a second time, we used a manual graphical technique to select the den entry and exit periods. Abrupt decreases in both GPS fix rates and activity rates were used to infer the beginning of denning, while abrupt increases in both fix rates and activity rates were used to infer den exit timing (see Supporting Information).

 

We estimated denning duration as the number of days between den entrance and exit for each bear. Because two bears moved between multiple dens in a single winter, we identified the interval from the date of entrance at the first den to date of emergence at the last den as the denning period for these individuals (Waller et al. 2012). To understand bear behavior following den emergence, we also calculated the post-emergence period, which we defined as the number of days it took each bear to first move 1km from their dens (Anderson et al. 2024).

 

 

We tested the effect of six intrinsic covariates on the date of den entry and exit: reproductive status and sex (combined), bear age, percent body fat, body mass, den elevation, and a measure of salmon use in September (Table 1). We determined the reproductive status of females via observations during telemetry flights or capture events. We used activity sensors in collars to detect birth events during denning by identifying short spikes of activity, which have been shown to reliably indicate birth (similar to Lemeire et al. 2022 but as in Roberts et al. *in prep*). Females with no cubs observed during at least 2 repeated aerial observations before denning and who did not give birth to cubs in the den were labelled as lone females. Females observed with at least 1 older cub (1- or 2-year-olds) after den exit were classified as females with cub at the time of den entry. To simplify analyses, we generated a variable combining sex and the reproductive status of females (RS), which had 4 classes during den entry (male, solo female, pregnant, female with cub) and 3 classes during den exit (male, solo female, and female with cub). RS information was missing for one individual which was excluded from analyses including RS.

 

Age was estimated by highly experienced biologists from tooth wear at capture (Hilderbrand et al. 2018) and projected forward throughout the study. This method has similar accuracy and is much less invasive than aging based on teeth annuli (Hilderbrand et al. 2018). However, inaccurate age estimates may have persisted and prevented us from detecting a true effect of age on den entrance or exit, but we did not have a way to address this issue. Body mass (to nearest 0.5 kg) was determined by using an electronic load cell (MSI-7200, Measurement Systems International, Seattle, WA, USA; Hilderbrand et al. 2018). Percent body fat was estimated at capture by bioelectric impedance analyses (RJL Systems, Clinton Township, MI, USA; Farley and Robbins 1994, Hilderbrand et al. 1998, Hilderbrand et al. 2018). Den elevation was extracted from a Digital Elevation Model with 20m resolution.

 

We estimated each bear’s use of salmon as a food resource before den entry based on their proximity to salmon-bearing streams. Previous studies in Alaska have found that time spent near streams is a reliable predictor of salmon consumption because bears respond to seasonal changes in salmon availability: 70% of the variation in the percent of assimilated bear diets consisting of salmon was explained by the amount of time GPS collared bears spent near streams (Deacy et al. 2016, 2018). Bears are mostly absent from salmon streams when salmon are not spawning but become ephemerally super abundant during salmon runs (Schindler et al. 2013, Deacy et al. 2016). We used the Alaska Anadromous Streams Catalog (AASC) and field observations to identify the water bodies in north-central Alaska where bears have access to salmon (Figure 1). We included all portions of streams where salmon are considered present. Although Deacy et al. (2016, 2018) used a 50m buffer to identify GPS locations associated with salmon fishing, that research occurred on small high-gradient streams with well-defined channels. Upon viewing satellite imagery with the AASC line features, we identified many clusters of GPS points on off-channel habitat that was up to 1000m from the line feature which defined the main channel. In other areas, the main channel was distant from its associated line feature, possibly because channels tend to migrate in dynamic flood plain rivers. To account for these issues, we considered individuals to be foraging for salmon when their GPS locations were <1000m from a salmon stream. Although some of these locations are likely associated with activities other than catching and eating salmon (e.g. travel, eating other foods), we found that the percent of individual brown bear GPS locations within 1000m of streams in September was positively correlated (r>0.64) with percent of salmon and terrestrial meat in the diet of individual bears documented by Mangipane et al. (2020), which lends support for our approach. Following these procedures, we defined the variable ‘September salmon use’ as the percentage of a bear’s September GPS locations within 1000m of salmon streams.

 

 

We tested the effect of four environmental covariates on the date of den exit (snow depth, mean air temperature, maximum air temperature, and the end of the continuous snow season; Table 1). We also included ‘year’ as a covariate to account for unexplained environmental variation. We did not test for the effect of these variables (except for year) on den entry because in this strongly seasonal system, environmental variables are all highly correlated with date and confounded by the movement of bears. For example, a bear that climbs to higher elevations intending to den may enter a location with deeper snow and cold temperatures, a change resulting from the movement and probably unrelated to the initiation of denning. In contrast, a bear in a den is stationary, which prevents movements from confounding tests for effects of environment on den exit.

 

To test whether den-specific weather conditions affected den exit timing, we attributed mean snow depth, mean surface temperature, and maximum surface temperature to den locations using the track annotation service Env-Data (Dodge et al. 2013) on Movebank (movebank.org, Wikelski et al. 2022). Mean snow depth (m), mean air temperature (C), and maximum air temperature (C) were derived from the National Center for Environmental Predictions’ North American Regional Reanalysis, which has a 3-hour frequency and 0.3-degree spatial resolution (National Centers for Environmental Prediction et al. 2005). To limit temperature variables to periods most likely to influence den exit timing, we averaged the mean and maximum temperatures from March 31- April 9th, the ten days prior to the earliest den exit.  

 

Bilinear interpolation was used for all attributes in Env-DATA. Weather reanalysis data have been found to have good agreement with weather station data in other studies in the Alaskan Arctic (Cameron et al. 2021). We sourced continuous snow season data from the Geographic Information Network of Alaska (GINA) at the University of Alaska Fairbanks, which derived the metric from MODIS Terra Snow Cover Daily L3 Global 500m Grid data (MtOD10A1.v005) from the National Snow and Ice Data center (Lindsay et al. 2015). End of continuous snow season is a measure of the first melt date at a location, which excludes the influence of spring snow fall.