Range contraction and increasing isolation of a polar bear subpopulation in an era of sea‐ice loss

Abstract Climate change is expected to result in range shifts and habitat fragmentation for many species. In the Arctic, loss of sea ice will reduce barriers to dispersal or eliminate movement corridors, resulting in increased connectivity or geographic isolation with sweeping implications for conservation. We used satellite telemetry, data from individually marked animals (research and harvest), and microsatellite genetic data to examine changes in geographic range, emigration, and interpopulation connectivity of the Baffin Bay (BB) polar bear (Ursus maritimus) subpopulation over a 25‐year period of sea‐ice loss. Satellite telemetry collected from n = 43 (1991–1995) and 38 (2009–2015) adult females revealed a significant contraction in subpopulation range size (95% bivariate normal kernel range) in most months and seasons, with the most marked reduction being a 70% decline in summer from 716,000 km2 (SE 58,000) to 211,000 km2 (SE 23,000) (p < .001). Between the 1990s and 2000s, there was a significant shift northward during the on‐ice seasons (2.6° shift in winter median latitude, 1.1° shift in spring median latitude) and a significant range contraction in the ice‐free summers. Bears in the 2000s were less likely to leave BB, with significant reductions in the numbers of bears moving into Davis Strait (DS) in winter and Lancaster Sound (LS) in summer. Harvest recoveries suggested both short and long‐term fidelity to BB remained high over both periods (83–99% of marked bears remained in BB). Genetic analyses using eight polymorphic microsatellites confirmed a previously documented differentiation between BB, DS, and LS; yet weakly differentiated BB from Kane Basin (KB) for the first time. Our results provide the first multiple lines of evidence for an increasingly geographically and functionally isolated subpopulation of polar bears in the context of long‐term sea‐ice loss. This may be indicative of future patterns for other polar bear subpopulations under climate change.


| INTRODUCTION
Conservation of wildlife populations across dynamic landscapes requires information on population structure and boundaries for establishing a valid ecological and demographic basis for determining population abundance, rates of population change, and the impacts of human activities (Thomas, 2010;Weinbaum, Brashares, Golden, & Getz, 2013). In an era of climate change, barriers to dispersal are quickly being altered, which can facilitate functional or genetic transfer between animals where it was once absent, or lead to increasing isolation because corridors of exchange are no longer available (Ching Chen, Hill, Ohlemuller, Roy, & Thomas, 2011;Rubidge et al., 2012).
Concurrently, many species are shifting to higher latitudes and higher elevations in response to a warming earth (Loarie et al., 2009). Shifts or reductions in ranges will likely result in reduced genetic diversity and in extreme cases, genetic impoverishment that may reduce population viability (Yannic et al., 2013). This can have rapid and sweeping impacts on conservation of large mobile species, especially those subject to subsistence harvests .
The Arctic is losing sea ice at a rapid rate (Overland & Wang, 2013). This large-scale habitat change is predicted to have multiple and significant consequences for ice-dependent species . A key question is whether loss of sea ice will reduce or enhance dispersal, given sea ice serves as both a platform for movement and a physical barrier. In the case of cetaceans, loss of sea ice has opened new habitats and resulted in positive population responses (George, Druckenmiller, Laidre, Suydam, & Person, 2015); however, this has generally not been the case for other species that rely on sea ice for key aspects of their life history, such as polar bears (Ursus maritimus; Amstrup & Gardner, 1994;PBSG 2010;Regehr et al., 2016).
Contrasting patterns may emerge depending on a species' dispersal ability and how changes in sea-ice extent, formation, or break-up structure habitat (e.g., Travis et al., 2013).

Polar bears are found throughout the circumpolar Arctic in 19
subpopulations that differ in productivity, size, and degree of connectedness with adjacent subpopulations (PBSG 2010). In areas with multiyear sea ice, some polar bears may remain on sea ice yearround, however, in areas where the annual sea ice melts completely in summer most bears spend up to several months on land, largely fasting until freeze-up (e.g., Stirling, Lunn, & Iacozza, 1999;Stirling & Parkinson, 2006).
The Baffin Bay (BB) subpopulation of polar bears is located in the seasonal ice ecoregion where sea ice forms and disappears each year (Amstrup, Marcot, & Douglas, 2008;Taylor et al., 2001; Figure 1). In the 1990s, analyses of movements from satellite-collared bears (Taylor et al., 2001), genetic analyses (Paetkau et al., 1999), and recaptures and harvest recoveries of individually marked bears (Taylor & Lee, 1995;Taylor et al., 2001) indicated limited interchange among BB and adjacent subpopulations including Davis Strait (DS) and Lancaster Sound (LS; Taylor et al., 2001). However, BB bears could not be genetically differentiated from the adjacent Kane Basin (KB) subpopulation to the north (Paetkau et al., 1999). As those studies, the BB subpopulation has been considered a distinct demographic unit with its dynamics largely driven by intrinsic rates of reproduction and mortality rather than exchange between neighboring subpopulations (SWG 2016).
The BB subpopulation currently supports a subsistence harvest of approximately 134 bears per year, which provides a nutritional, cultural, and economic resource to Inuit communities (PBSG 2010). Declining sea ice in Baffin Bay has resulted in a lengthening of the ice-free season by >12 days per decade during the period 1979-2014 (Stern & Laidre, 2016). There is evidence for positive correlation between seaice availability and BB subpopulation reproductive rates (SWG, 2016). long-term sea-ice loss. This may be indicative of future patterns for other polar bear subpopulations under climate change.

K E Y W O R D S
animal movements, Arctic, contraction, isolation, polar bear, range, sea ice F I G U R E 1 Distribution of capture locations of adult female polar bears fitted with satellite collars in Baffin Bay during the 1990s and 2000s Furthermore, recent studies suggest low apparent survival for male polar bears in BB, which may reflect either increased natural mortality or emigration (SWG, 2016).
We conducted an assessment of the extent to which sea ice trends have affected the distribution and boundaries of the BB subpopulation and discuss results in the context of other polar bear subpopulations facing sea-ice loss, as well as near-term management and long-term conservation planning. We used home range estimators from satellite telemetry to delineate subpopulation range boundaries, movement data from collared adult females and from the recapture or harvest of marked individuals to assess latitudinal shifts or changes in immigration/emigration, and genetic analyses to assess relationships between BB and adjacent subpopulations. Our objective was to explore multiple lines of evidence with respect to the delineation of the BB subpopulation in an era of rapid sea-ice loss. We discuss the implications of the results for effective conservation and management strategies and identify where additional research and monitoring will be needed in the future.

| Study area
The boundaries of the BB subpopulation encompass ~1 million km 2 , covering approximately 656,000 km 2 of ocean as well as portions of Baffin Island, all of Bylot Island in Nunavut/Canada, and parts of West and Northwest Greenland (66°N to 77°N;Taylor et al., 2005) ( Figure 1). During late spring and summer break-up, sea ice recedes from Greenland westward across Baffin Bay; the last remnants of ice typically occur off the coast of Baffin Island. Most polar bears remain on the sea ice as it recedes and then come ashore to spend the icefree period on Baffin and Bylot Islands (Taylor et al., 2005). An unknown but probably small number of bears remain on land in Melville Bay, NW Greenland throughout the ice-free period (Born, 1995;Born, Heilmann, Holm, & Laidre, 2011;SWG, 2016).

| Handling and collaring
Adult female polar bears (AF) were captured in NW Greenland on the fast and pack ice between mid-March and mid-April 2009-2013 in Baffin Bay. Searches for bears occurred between 70° 22′ N and 76° 20′ N, to a maximum distance of ca. 150 km from the coast and included areas at glacier fronts. Bears were located and captured from a helicopter using standard chemical immobilization techniques (Stirling, Spencer, & Andriashek, 1989). Field estimates of age and reproductive status were recorded, with age later estimated by counting cementum growth layers on a premolar tooth extracted during capture (Calvert & Ramsay, 1998 Figure 1), which provided information on geographic location, internal transmitter temperature, and activity. These data were combined with a historical data set of 43 AF that were collared within the BB subpopulation management boundaries (PBSG, 2010) between 1991 and 1995 (Ferguson, Taylor, & Messier, 1997;Taylor et al., 2001). The majority of the 1990s tags were deployed during the ice-free season in fall on Baffin Island (with the exception of n = 9 deployed in spring in NW Greenland).

| Data filtering and sub-sampling
Locations and transmitter status were collected via the Argos Location Service Plus system (Toulouse, France). The quality of each location was assigned by ARGOS with location qualities of 0-3 estimated to have errors of 1.5 km or less and those categorized as "A," "B," or "Z" had no predicted accuracy. Unrealistic and poor quality locations were removed using a speed and angle filter in R version 2.13.2 (R Development Core Team 2012) using the package "argosfilter" (Freitas, Kovacs, Lydersen, & Ims, 2008). Positions exceeding a maximum between-location travel velocity (10 km/hr based on previous movement studies of polar bears, Laidre et al., 2013) and angle (measured from the track between three successive locations; set to the default) were removed by the filtering algorithm. The resulting locations for each bear were reduced to a single position per day to reduce autocorrelation bias, standardize temporal sampling, and address the effects of variable duty cycling among PTTs. The first, best quality location within the period of peak satellite passage was selected to obtain a daily position for each PTT. Daily positions, after filtering and optimal daily position selection, only consisted of ARGOS qualities 1-3. Distances between successive daily positions were calculated as the great circle route and used to compute minimum daily displacements.
As a result of variable experimental objectives in both decades, different duty cycles were used for tags in an effort to extend battery life or gather information from specific time periods. Collars in the 2000s were programmed to transmit during one 6-hr period each day on 4-day intervals. The 1990s collars were programmed to transmit on varying and intermittent intervals, ranging from 1 to 6 days. We  subsampled the 1990s data and created a strict 4-, 5-, or 6-day interval time series for each individual to best match the 2000s data. This ensured that serial autocorrelation was consistent among decades.
Telemetry data were divided into seasons: spring (March-July, which included the peak of sea-ice coverage and initiation of seaice break-up), summer (August-October, which included the end of break-up and the on-land period), and winter (November-February, which included the freeze-up period and time when bears returned to the sea ice). All periods when collared females were denning (use of maternity and shelter dens) were identified (Escajeda et al., 2018) and removed from analyses. Bears with <3 locations were removed from analyses when transmitter failure occurred immediately after deployment.

| Comparison of 1990s and 2000s BB satellite telemetry data
Polar bears collared in this study ranged over the entire Baffin Bay region (SWG, 2016). All captures occurred within the bounds of the BB subpopulation management unit (PBSG, 2010) and bears moved back and forth freely between Canada and Greenland during both periods ( Figure 1). However, seasonal and geographic differences in capture locations occurred in our study between the two periods of fieldwork.
In the 1990s, it was possible to deploy collars on AFs on both the We then quantified what overall range of latitudes on land on Baffin Island was used by polar bears in both decades. Finally, we subset the 1990s dataset to conduct interdecadal analyses on range size using a subsample of 1990s bears (n = 9) collared in spring on the sea ice in Greenland compared with range size of bears collared in the same area in the 2000s (see Figure 1). This essentially replicated some analyses we present in this manuscript (but with a substantially reduced sample size in the 1990s).

| Monthly and seasonal kernel density estimates
We estimated the geographic area characterized by a high probability of use by collared AF polar bears in Baffin Bay using a fixed kernel density approach (Worton, 1989). Kernel density estimators provide a nonparametric probability of using a given point in space and are reliably used to define the utilization distribution, or home range, for marine and terrestrial wildlife (Kie et al., 2010). We calculated Gaussian bivariate normal kernel density estimates for each decade (n = 2), month (n = 12), and season (n = 3) for all bears in the sample. Kernel density estimates (KDEs) were calculated using the "bkde2D" function in "KernSmooth" R package (Wand, 1994;Wand & Jones, 1995). As the sample size of collared bears slightly differed between the 1990s and 2000s (Table 1), we generated random samples of equal size (n = 38) from the pool of AF bears in each decade. We sampled bears with replacement 1,000 times for each monthly and seasonal KDE and calculated the area of the 95% contour polygon (bounding 95% of the KDE surface volume). We produced a mean and bootstrapped standard error (SE) for monthly and seasonal home ranges, and used the "intersect" tool in ArcGIS to calculate overlap in home ranges between decades. We also estimated the proportion of home range overlap between the 1990s and the 2000s (Fieberg & Kochanny, 2005) based on the bootstrapped mean. We used a cell size of 6 km and bandwidth of 50 km (approximately 50% of the 4-day movement step of AFs in this study) to calculated KDEs. We also examined whether there has been a distributional shift by calculating the median seasonal latitude values for AFs in the 1990s and 2000s of the KDE and of the pooled location data. We compared changes in median latitude and longitude across decades with t tests at a significance level of α = 0.05.

| Movements across subpopulation boundaries
Using each complete individual AF bear trajectory as a single sample, we calculated the number of days each bear spent in BB during the tracking period. We examined the departures from BB using telemetry data by estimating the fraction of bears that crossed BB management boundaries up to 300 days after capture. Specifically, we calculated the number of days since capture until each polar bear left BB or until the end of their observation period for those that were not observed to leave. For each decade, statistical methods for censored event times were used to construct "survival" curves (Kaplan-Meier) to characterize the distribution of first exit times, to estimate departure probabilities from BB, and test for differences among decades (log-rank test of equality) with α = 0.05. We considered a departure to be at least 30 days long, thus bears that departed BB but returned in <30 days were not included in the estimates. We summarized which subpopulation bears departed to, the departure month, and compared this across decades. We also tested whether 1990s capture season (spring or fall) impacted the time until departure from BB by assessing capture season as a factor.

| Harvest recoveries and recaptures of marked bears
We evaluated spatial patterns in live recaptures and dead recoveries of marked bears. We included bears that were marked in spring (April-May) or fall (August-October) during two periods when BB subpopulation mark-recapture studies were undertaken; 1991and 2009-2014. From 1991, 881 polar bears were captured, physically marked, and tissue samples were collected (Taylor et al., 2005). We also updated previously reported information on long-term fidelity and patterns of harvest recovery in BB (Taylor & Lee, 1995;Taylor et al., 2001;Peacock et al., 2012)  Given the loss of sea ice that has occurred in both areas (Stern & Laidre, 2016) and the opportunity presented by having three intensive periods of mark-recapture that were relatively close in time, we used these data to assess movement between these two subpopulations. In particular, we examined whether there was evidence that DS bears had moved northward, in concert with sea-ice loss, across the BB-DS subpopulation boundary. For this assessment,

| Radio telemetry
Adult females in all reproductive states were fitted with collars in the 1990s (n = 43) and 2000s (n = 38) ( Table 1). Collars deployed be-

| Subpopulation ranges
In the 1990s, monthly 95% bivariate kernel range sizes varied be-

| Movement between subpopulations
Overall, a significantly lower fraction of collared polar bears left BB in the 2000s than in the 1990s (χ 2 = 6.8, df = 1, p = .009). The reductions in numbers of bears departing BB across subpopulation boundaries were due to fewer bears moving south into DS and fewer bears moving west into LS. In the 1990s, 14 of 43 (33%) radio-collared bears moved south into DS, and 12 of 43 (28%) bears moved west into LS.
In the 2000s, movements occurred to these areas but at a significantly lower rate; three of 38 (8%) tagged bears moved to DS, and three of 38 (8%) tagged bears moved to LS. In the 2000s, five of 38 (13%) bears also moved north directly from BB into KB, and an additional two moved to KB after first moving to LS. In the 1990s, n = 2 of 43 bears moved to KB and both initially moved into LS.

| Harvest recoveries and recaptures of marked bears
Of the 1,250 bears marked in BB (2011BB ( -2013

| Genetic connectivity between Baffin Bay and other subpopulations
The average expected heterozygosity, based on eight polymorphic microsatellites, for 402 polar bears sampled during winter-spring  (Table S1) nor was a linkage disequilibrium observed between all pairs of loci within the dataset. Population structure estimated using multilocus F ST statistics was generally low, although tests were statistically significant after sequential Bonferroni correction (Rice, 1989) at α = 5% level (Table 4). Pairwise F ST values showed low but significant differentiation for BB between KB, DS, and LS when comparing the winter and spring samples for bears of all ages and sexes, adults of both sexes, and adult females (Table 4). There were no significant differences for adult males and subadults (both sexes). These results provide the first evidence for a genetic difference between BB and KB, in contrast to previous studies by Paetkau et al. (1999), Peacock et al. (2015), and Malenfant et al. (2016). The discriminant analysis of principal components (DAPC, Jombart et al., 2010), separated BB and KB from LS and DS, supporting the grouping observed using GENELAND (Figures S1-S4, Table 2). Of note, Bayesian clustering method in STRUCTURE did not separate BB and KB, possibly due to low F ST values (see Appendix S1).

| DISCUSSION
Over the past three decades, there has been a significant decline in sea ice in Baffin Bay and surrounding areas (Stern & Laidre, 2016), which has directly translated to loss of habitat available to the BB polar bear subpopulation. We document that BB subpopulation has become increasingly isolated through contraction of seasonal ranges and reductions in emigration across subpopulation boundaries. We suggest that this is a consequence of ongoing loss of sea ice in Baffin Bay and reflects climate impacts that are likely to occur for other subpopulations in the seasonal sea-ice ecoregion (Amstrup et al., 2008).

| Satellite telemetry
The reliability of inferences about population structure based on movement data is dependent on the extent to which sampled individuals represent the subpopulation of interest. This is a function of sampling strategy, timing, and sample size. In this study, we investigated the movement of AFs and interpret the data as representative for the subpopulation as a whole. The movements of individual polar bears have been studied using satellite telemetry for decades (e.g., Amstrup, Durner, Stirling, Lunn, & Messier, 2000;Born, Wiig, & Thomassen, 1997;Laidre et al., 2013Laidre et al., , 2015Wiig, Born, & Pedersen, 2003), and inference has almost exclusively been based on adult females because of the challenges of instrumenting adult males (Wiig et al., 2017). Our analyses, which also included harvest recoveries of both sexes, did not indicate vastly different movement between males and females.
Previous studies in BB also indicated that both sexes utilize the same habitats in spring (Laidre et al., 2013). We therefore suggest our satellite telemetry data are a representative index of broad population movement patterns.
Our analyses addressed differences in sampling between decades to ensure that the 1990s and 2000s data, which were collected within the boundaries of BB but different areas and seasons, were comparable.
Bears collared in West Greenland in the 2000s used nearly the entire Baffin Island coastline in fall and were distributed over the same capture region and same range of latitudes where females in the 1990s were collared, with the exception of the southern point around Cape Dyer.
Further, subsample analyses of a small number of bears collared in spring in a restricted area of West Greenland in the 1990s (n = 9) provided the same significant results on summer range reduction as the larger dataset.
Overall, in our study, sample sizes for each decade were roughly equivalent (1990s, n = 43; 2000s, n = 38), sampling durations were similar (6to 7-year tracking periods in each decade), and collars in both decades transmitted up to 3 years. Finally, in both decades, collar deployments were distributed over multiple years and over broad geographic areas within BB management boundaries (Figure 1).
We found that seasonal ranges of adult females in BB became significantly smaller, by a third to a half, between the 1990s and 2000s.
This range contraction coincided with a northward shift during the on-

| Recovery of marks
Although the use of tag recoveries is a relatively coarse means of as- Our findings imply that BB bears exhibit a high degree of short-term fidelity to this geographically defined unit, which is consistent with estimates of site fidelity derived from mark-recapture analyses (SWG, 2016

| Movements between Baffin Bay and Davis Strait
The boundary between the BB and DS subpopulations is spanned by pack ice in spring that provides a continuous platform for bears to move between subpopulations (Stern & Laidre, 2016). Nevertheless, consistent differences between BB and DS polar bears have been shown through genetics (Paetkau et al., 1999;Peacock et al., 2015; this study), movements (Taylor et al., 2001), and diet (Thiemann, Iverson, & Stirling, 2008), suggesting a functional boundary between them. Here, we investigated if there has been an increased exchange of polar bears across this boundary, which could contribute to exchange between subpopulations. Since 1990, marking effort (i.e., number of unique individuals marked) in BB and DS was equivalent to 41% (BB 1991(BB -1997, 70% (DS 2005(DS -2007, and 44% (BB 201144% (BB -2013 of estimated abundance at the time of marking (Peacock et al., 2013;SWG, 2016

| Genetic connectivity between subpopulations
Previous analysis of polar bear population structure using microsatellites showed no significant genetic differences between BB and neighboring KB (Paetkau et al., 1999), though studies have found that bears from BB-KB differed genetically from the LS and DS subpopulations (Paetkau et al., 1999;Peacock et al., 2015;Malenfant et al., 2016). In contrast, we found low but significant F ST estimates between winter-  (Paetkau et al., 1999;Peacock et al., 2015;Malenfant et al., 2016).
Future sea-ice loss is expected to fragment polar bear subpopulations, increase isolation, alter gene flow, and disrupt population boundaries (Derocher, Lunn, & Stirling, 2004;Kutschera et al., 2016;Sahanatian & Derocher, 2012). Genetic variation in polar bears is relatively low, and genetics is a relatively insensitive means of identifying, defining, or detecting change in subpopulation structure, especially when compared to individual-based telemetry and tag recoveries.
Time-lag effects in genetic composition are influenced by the relatively long generation length for polar bears (~11.5 years; Regehr et al., 2016), and several generations may be needed to demonstrate increased genetic isolation as a consequence of climate-induced range contraction.
Collectively, our findings from multiple lines of evidence suggest that the current BB subpopulation boundaries continue to be relevant for harvest management and subpopulation monitoring. However, an important consideration is that the subpopulation range is contract- ing and there appears to be reduced connectivity with subpopulations to the west and south. Our results provide evidence that the BB subpopulation is becoming increasing isolated, with reduced range sizes, shifts northward, and reduced emigration across subpopulation boundaries. These changes are almost certainly driven by loss of sea ice (Stern & Laidre, 2016). In general, telemetry and genetics share major advantages including standardized objective methodologies for delineating demographic structure allowing for broad inferences on movements and for calculating immigration and emigration rates. In our study, both telemetry and tag recovery data provided early signals of range shifts and changes in individual movements between local populations. Although our genetic analyses had relatively coarse resolution, primarily because the markers used were intended for mark-recapture and not dispersal, the results indicated a differentiation between the BB and KB subpopulations. More detailed genetic analyses would be worthwhile to explore this further. Sea-ice loss is predicted to continue and information from telemetry and marked or recovered bears through long-term monitoring programs (i.e., Laidre et al., 2015) will be critical to understanding increased fragmentation among polar bear subpopulations.
One key question unanswered is how loss of sea ice will alter dispersal patterns of apex predators, either by increasing genetic exchange as barriers to dispersal are removed or by increasing isolation as platforms for genetic transfer disappear. In some cases, different patterns may emerge within a species, depending on their dispersal ability and how loss of habitat redistributes subpopulations. For example, in areas around the Arctic Basin where sea ice converges or diverges (Amstrup et al., 2008), polar bears from multiple subpopulations may become aggregated or even mix. However, in the seasonal ice polar bear ecoregion (Amstrup et al., 2008), where annual sea ice exerts strong control on movements and directly influences the amount of time and area bears have to use and interact, the loss of sea ice is likely to make subpopulations more isolated. These findings and Bold value is significant at p < .05 after sequential Bonferroni correction (Rice, 1989).