Evidence for an Adaptive, Large‐Scale Range Shift in a Long‐Distance Terrestrial Migrant

ABSTRACT

Long‐distance migrations are a striking, and strikingly successful, adaptation for highly mobile terrestrial animals in seasonal environments. However, it remains an open question whether migratory animals are more resilient or less resilient to rapidly changing environments. Furthermore, the mechanisms by which animals adapt or modify their migrations are poorly understood. We describe a dramatic shift of over 500 km in the wintering range of the Western Arctic Herd, a large caribou ( Rangifer tarandus ) herd in northwestern Alaska, an area that is undergoing some of the most rapid warming on Earth. Between 2012 and 2020, caribou switched from reliably wintering in maritime tundra in the southwesternmost portion of their range to more frequently wintering in mountainous areas to the east. Analysis of this range shift, in conjunction with nearly 200 documented mortality events, revealed that it was both broadly adaptive and likely driven by collective memory of poor winter conditions. Before the range shift, overwinter survival in the maritime tundra was high, routinely surpassing 95%, but falling to around 80% even as fewer animals wintered there. Meanwhile, in the increasingly used mountainous portion of the range, survival was intermediate and less variable across years compared to the extremes in the southern winter ranges. Thus, the shift only imperfectly mitigated overall increased mortality rates. The range shift has also been accompanied by changes in seasonal patterns of survival that are consistent with poorer nutritional intake in winter. Unexpectedly, the strongest single predictor of an individual's probability of migrating south was the overall survival of animals in the south in the preceding winter, suggesting that the range shift is in part driven by collective memory. Our results demonstrate the importance and use of collective decision making and memory for a highly mobile species for improving fitness outcomes in a dynamic, changing environment.

The Western Arctic Herd, a large but declining caribou population, has shifted its main wintering range from lichen‐rich coastal tundra in the southwesternmost portion of its range to mountainous areas over 500 km away. Caribou largely abandoned the southern ranges as winter survival worsened, indicating an adaptive response (though not enough to fully mitigate declines). Further, the best predictor of caribou migrating south was collective survival the previous winter, suggesting that decisions were driven by collective memory. Thus, we show how a highly mobile, social species can use collective decision making to improve fitness outcomes in a dynamic, changing environment.

1. Introduction

Large‐scale mobility in animals is a fundamental adaptation to navigating highly variable and dynamic environments (Teitelbaum and Mueller 2019). Long‐distance seasonal migrations are a particularly conspicuous strategy to exploit predictably periodic resources, whether for energetic gain, predator avoidance, or access to reproductive habitat (Avgar, Street, and Fryxell 2014; Fryxell and Sinclair 1988). For many species, seasonal migration is an extremely successful strategy, allowing a far greater number of individuals to inhabit landscapes that might otherwise not be able to support large numbers year round (Fryxell, Greever, and Sinclair 1988; Fryxell and Sinclair 1988). But, as those resources shift in space and time, plasticity of those periodic movements becomes essential for adaptation (Anderson et al. 2013; Gurarie et al. 2021; Xu et al. 2021). Indeed, it remains a relatively open and case‐specific question whether the migratory strategy fundamentally makes a species more or less resilient to environmental disruptions in the environment (Hardesty‐Moore et al. 2018; Moore and Huntington 2008; Xu et al. 2021). While the study of the plasticity of migratory animals has been of theoretical interest (Holt 2003; Holt and Fryxell 2011; Kingsolver, Pfennig, and Servedio 2002), it is especially relevant in light of rapid environmental change—both global warming and anthropogenic development—which is leading to widespread and well‐documented impacts to seasonal migrations for many taxa (Harris et al. 2009; Kauffman et al. 2021; Middleton et al. 2013; Sutherland 1998; Wilcove and Wikelski 2008; Xu et al. 2021).

While long‐term mechanisms by which large‐scale collective movements adapt to changing conditions can be genetically driven (Alerstam, Hedenström, and Åkesson 2003; Anderson et al. 2013), an emerging consensus suggests that for long‐lived, social species, the principal mechanism is cultural (Aikens et al. 2022; Berdahl et al. 2018; Gurarie et al. 2021; Jesmer et al. 2018). In other words, the adaptive mechanism relies on a combination of social interactions and collective memory. For example, Jesmer et al. (2018) showed that migratory propensity emerged over several decades for otherwise naïve, translocated populations of mountain ungulates that collectively learned to exploit a shifting green wave. In another example, reintroduced whooping cranes ( Grus americana ) shortened their migrations based on tactical decisions made by older members of social flocks (Mueller et al. 2013; Teitelbaum et al. 2016). While these studies point to ways in which migration routes emerge or shift over time, they rely on an assumption that the migratory propensity (in the case of the mountain ungulates) and the migration shortening (in the case of the cranes) are adaptive with respect to fitness. Rarely do studies quantify the explicit fitness benefits of behavioral shifts in migration, or point to specific, environmentally driven mechanisms by which specific range shifts might occur. In contrast, Middleton et al. (2013) argued that the persistent, traditional seasonal migration of elk ( Cervus elaphus ) in the face of changes in land use and climate in the Greater Yellowstone Ecosystem had negative effects on survival and recruitment, while Williams et al. (2024) present compelling evidence that another partially migrating population of elk became less migratory under changed predation pressure, with positive population outcomes.

Probably no terrestrial species combines larger seasonal movements in a more seasonal environment than migratory reindeer and caribou (henceforth, simply “caribou”), that is, those subspecies of Rangifer tarandus whose migrations straddle the boreal forest and Arctic tundra across the circumpolar north (Joly et al. 2019). Along with being extremely seasonal, the Arctic is currently undergoing some of the most rapid warming on earth (Rantanen et al. 2022), leading to dramatic changes in vegetation cover (Mekonnen et al. 2021), nutrient cycling (Sarneel et al. 2020), snow conditions (Boelman et al. 2019; Callaghan et al. 2011), and fire regimes (Young et al. 2017), while also experiencing increasing industrial development in the form of mining, oil, and gas exploration and extraction. Caribou populations cycle at multidecadal time scales (Gunn 2003), yet the majority of caribou herds are currently declining and have greatly depressed numbers over the past few decades (Gunn and Russell 2023; Vors and Boyce 2009). There is a great deal of uncertainty as to the causes for the variable population trajectories of the major caribou herds, for example, whether they are driven by predation (Bergerud 1996), fluctuations in forage availability (Gunn 2003), and additional concerns about the potential impacts of a rapidly expanding human footprint (Joly and Klein 2011) and changing climate on their populations (Mallory and Boyce 2017).

Unlike most temperate ungulates, for which winters are often a survival bottleneck (Kautz et al. 2020), seasonal survival patterns of caribou indicate that winter can be a period of relatively lower mortality risk (Gurarie et al. 2020). Uniquely among northern ungulates, fat reserves of body mass—common proxies for body condition—can increase over the winter (Cook et al. 2021; Couturier et al. 2009; Dale et al. 2008). However, the environmental factors that influence caribou survival in winter are complex and poorly understood. For example, deeper snow has been shown to have both positive and negative impacts for survival. Deep snow can impede movements and increase energy expenditures to crater (dig) to reach terricolous forage (Fancy and White 1987). Or, in areas where barren‐ground or tundra‐adapted caribou are generally exposed to less snow than their boreal or mountain counterparts, deep snow can also make animals more susceptible to predation (Joly and Klein 2011). Conversely, another long‐distance migratory population of caribou selected deep snow (Sharma, Couturier, and Côté 2009), and in one nonmigratory boreal population, deep snow was shown to be related to higher survival (Schmelzer et al. 2020). In both cases, this was attributed to evading predation more effectively. Similarly, high winds have an unclear impact: they can scour snow, in some places reducing snow depth, in others increasing snow depth and often hardening it (Collins and Smith 1991). In various places, and to various extents, climate change is likely leading to a generally wetter Arctic (Bintanja and Andry 2017; Langlois et al. 2017), with important potential consequences for snowscapes and wildlife (Boelman et al. 2019). In at least one nonmigratory population of Rangifer, these same warming trends appear to be benefitting populations, whether by dampening population cycling dynamics (Hansen et al. 2019) or simply increasing the availability of high protein forage (Hiltunen et al. 2022). On the other hand, warming‐related icing, in particular rain‐on‐snow events, can create dense layers that are difficult for caribou to penetrate, locking terricolous forage away (Tyler 2010).

Perhaps, the most important adaptation of caribou for surviving winter is plasticity in movement and ability to find and shift seasonal ranges over large spatial scales and across multiple years. They are, therefore, an ideal system in which to study potentially adaptive and environmentally informed mechanisms of large‐scale behavioral shifts. Environmentally driven plasticity in ungulate migration is a widely observed phenomenon (Xu et al. 2021); however, few studies can point directly to a fitness outcome of changing migrations. Without a direct linkage between those shifts and fitness outcomes, it is impossible to determine whether those shifts are, in fact, adaptive.

To demonstrate that a seasonal range shift is an adaptive response to environmental cues requires measuring a fitness outcome (e.g., survival) that depends on a particular behavior (e.g., seasonal range choice) and demonstrating that the behavior change leads to an improvement in that fitness outcome. To further show that this adaptive change is memory‐driven requires, at minimum, demonstrating that past experiences can meaningfully predict future decisions. Here, we present evidence of a large‐scale, adaptive shift in the seasonal range of migratory caribou that is likely driven by collective memory.

2. Material and Methods

2.1. Study Area

The Western Arctic Herd (WAH) is a large caribou herd that ranges over more than 360,000 km² of northwest Alaska (Joly, Cole, and Jandt 2007; see also Figures 1 and 2). Like other large herds, the WAH naturally undergoes oscillations in abundance (Gunn 2003; Joly et al. 2011), peaking most recently at 490,000 individuals in 2003 but steadily declining since that time to about 152,000 caribou in 2023 (A. Hansen, unpublished data). Their vast range includes extensive boreal forest south of the rugged Brooks Range mountains, Arctic tundra north of the range, and alpine tundra within it. Shrubs (Salix spp., Alnus spp., Betula spp.) are found in all these biomes but dominate the landscape in certain ecotones. Elevation ranges from sea level to over 2500 m. The climate varies by region, but winter tends to last 6–8 months, with temperatures dropping below −40°C. Summers are brief, with green deciduous vegetation available from late May through early September and temperatures reaching above 30°C south of the Brooks Range.

FIGURE 1.

Deployment and mortality timelines of adult female collared Western Arctic Herd caribou from 2009 to 2022. Black lines indicate caribou with known mortality events, gray lines correspond to caribou with censored data, whether due to collar failure, drop‐off, or the animal still being alive at the end of the study. The inset map shows the range of the herd in the northwest portion of Alaska.

FIGURE 2.

Seasonal movements by female Western Arctic Herd caribou for biological years 2011–2020, illustrating the high calving ground site fidelity (yellow colors) and highly variable wintering ground behavior (blue colors). Insect harassment season (orange), late summer (red), fall (purple), and spring (green) are also presented. The Kobuk River is the thick white line running east to west. All collaring occurred at Paatitaaq (Onion Portage) prior to 2019. The 2 years in the right panels (2015–2016 and 2020–2021) are larger to highlight the shift in the wintering range.

The WAH's range is roughly bisected by the Kobuk River (Figures 2 and 3), a major east to west flowing river that drains a large basin south of the Brooks Range. Wintering areas north of the Kobuk consist of Arctic coastal plain habitat, rolling foothills, and the Brooks Range mountains. These are characterized by low‐stature arctic tundra vegetation, such as tussock‐forming graminoid herbaceous communities (i.e., Eriophorum spp.), dwarf shrubs (i.e., Betula nana ), moss, and lichens (Gallant et al. 1995). Wintering areas south of the Kobuk are characterized by sub‐Arctic tundra and boreal forest habitats. These are characterized by tussock‐forming graminoid herbaceous communities (i.e., Eriophorum spp.), shrubs (i.e., Salix spp.), deciduous and evergreen forests ( Betula papyrifera , Picea spp., Populus spp.), and lichens (Gallant et al. 1995). Snow is generally shallower in the northernmost wintering areas compared to the more southerly maritime tundra areas. Green up begins later north of the Brooks Range compared to areas to the south. Outside of several isolated human settlements, there is little large‐scale infrastructure within the WAH range. A notable exception is the Red Dog mine, one of the largest zinc and lead mines in the world, located in the western portion of the range north of the Kobuk River and connected to coastal port site by a 90‐km road near the town of Kotzebue.

FIGURE 3.

September photos of Western Arctic Herd caribou swimming the Kobuk River (left) across Paatitaaq (Onion Portage), shown with many caribou (top right) and without caribou (bottom right). In recent years, almost no caribou have crossed the river at this site, which has been used by both caribou and people for millennia (Anderson 1998; Joly and Cameron 2022). Photos: K. Joly.

2.2. Movement and Mortality Data

From 2009 to 2021, adult female caribou from the WAH were captured and fitted with GPS telemetry collars (Telonics TGW‐4680, Mesa, Arizona, USA) following protocols approved by a State of Alaska Institutional Animal Care and Use Committee (IACUC 0040–2017‐40). Only adult (> 2 years old) female caribou were fitted with collars, which collected data at 2‐, 4‐, or 8‐h intervals. To account for unequal relocation intervals, and because the processes we are considering are on large spatial and temporal scales, we used the mean daily location for each individual. On average, around 27 (s.d. 14) caribou were collared annually, ranging from 3 (in 2017) to 52 (in 2021). Since caribou typically survived more than 1 year after collaring, between 39 (in 2009) and 116 collars (in 2021; mean 88, s.d. 24.3) were active in any given year. In total, we collected 1056 caribou‐years of data. Mortality events were determined by the collar not moving but still functioning, transmitting a cluster of positions in nearly the same location.

Wherever possible, we retrieved the collar at those locations. It was unfortunately nearly impossible to determine ultimate causes of mortality, with the exception of the rare reported harvest, as there were typically very few remains aside from hair and bones due to predation, scavenging, and decomposition at the time of collar recovery. Furthermore, the main nominal proximate causes of mortality (e.g., starvation, sickness, predation) are confounded: a sick caribou is often starving (and vice versa) and therefore more susceptible to predation.

For decades, the majority (~80%) of WAH caribou crossed Kobuk River in Kobuk Valley National Park during their fall migration heading to their winter ranges (Figures 2 and 3), and again when migrating northward to calve, often crossing the river at a particular site called Paatitaaq (Onion Portage—Figures 2 and 3) that has been a caribou hunting camp for several millennia (Anderson 1998; Joly and Cameron 2022). From 2009 to 2018, all animals in the study were captured and collared at Paatitaaq as they swam across the river in fall (i.e., September). However, in recent years, an increasing number of animals have remained north of the river throughout the winter season (Figures 2 and 3). Since 2020, all captures shifted to spring (i.e., April) helicopter‐based net gun methods as Paatitaaq became unreliable for captures (Joly and Cameron 2022).

Notable shifts in migration patterns have been detected in recent years (Figure 2), leading to the change in data collection methods described above. For several of our analyses, we contrasted those patterns before and after a cutoff year when those differences were maximized. To quantitatively determine the year at which to make a discrete break between early and late periods, we compared the (a) probability of crossing the Kobuk River and (b) winter and spring survival using potential cutoffs on May 28 every year between 2012 and 2019. We assessed the most significant shift in those processes by computing the mean square contingency coefficient $\varphi =\sqrt{\frac{{\mathrm{\chi }}^2}{n}}$, where $\mathrm{\chi }$ is the chi‐squared test statistic and n is the sample size (Yule 1912), with confidence intervals. The statistic is reported on each of the respective responses after subdividing the data at each possible breakpoint and selected the year where ϕ was highest.

For several analyses, we divided the annual movements of WAH caribou into six ecological seasons following the divisions used in previous studies (Baltensperger and Joly 2019; Joly and Cameron 2022). These were: fall (September 1 to November 30, encompassing the fall migration and the rut), winter (December 1 to March 31), spring (April 1 to May 27, which includes the bulk of the spring migration period [Gurarie et al. 2019]), calving (May 28 to June 14), insect harassment (June 15 to July 14, during which almost the entire WAH aggregates near the coast in pursuit of relief from harassment), and late summer (July 15 to August 31). These seasons correspond mainly to the main movement and behavioral cycles of the caribou, and not necessarily to the climatic seasons (e.g., “winter” in the range is generally considered to be longer than the 4 months—but here refers broadly to the period between the conclusion of the fall migration and the beginnings of the spring migration). While there is considerable interannual phenological variability in caribou movements and behaviors, these divisions are broadly consistent with the annual patterns of these animals.

For analysis of predictors of migration as proxied by crossing the Kobuk River, a further seasonal subdivision was made between “early fall” and “late fall,” separated by October 5, which corresponds to the mean date that animals crossed the river.

2.3. Environmental Covariates

We collected a suite of environmental covariates related to temperature, precipitation, wind, and snow at every caribou location from remotely sensed and modeled sources. Daily maximum temperature and precipitation were obtained from Daymet V3 (Daily Surface Weather and Climatological Summaries, Thornton et al. 2016) at 1‐km resolution; wind speed was obtained from ERA5 Daily Aggregates at 27.83‐km pixel resolution (Copernicus Climate Change Service (C3S) 2017); and snow depth data were obtained from National Centers for Environmental Prediction at 26.98‐km pixel resolution (Mesinger et al. 2006). Every caribou location was annotated with the environmental data using Google Earth Engine (Gorelick et al. 2017) or Movebank Env‐DATA on Movebank (Dodge et al. 2013) depending on the data source hosted on the platform. These environmental covariates were averaged on a seasonal basis (see below) for each individual caribou, and as such provide a broad measure of “coldness,” “windiness,” and “snowiness.”

2.4. Seasonal Patterns of Survival

We focused our survival analyses on estimating the mortality hazard (interpretable as the probability of individual mortality per day) as it varied throughout the year. We estimated the season‐specific hazards—assumed to be constant within each of the six discrete seasons specified above—using parametric survival regression models (Therneau and Grambsch 2000). We also quantified and compared “mortality seasons” (i.e., periods of elevated mortality characterized by a peak date, duration, and intensity) using cyclomort seasonal survival analysis (Gurarie et al. 2020). This analysis allows for the identification of seasonal patterns without a priori determination of those seasons and also provides statistical machinery to estimate and statistically compare models with multiple peaks of mortality that might occur within a year. Both sets of survival analyses were performed for all the data pooled and separately for animals before and after the migration shift, as determined by the analysis described previously.

2.5. Environmental and Behavioral Drivers of Survival

We compared the environmental drivers of winter and early spring survival on either side of the Kobuk River by fitting Cox proportional hazard models for animals that did and did not cross the river against environmental covariates experienced in winter. Because we were fitting this model to a smaller subset of the total mortality data (those that crossed the Kobuk and died in winter and spring), we fitted each of the weather covariates individually. To test whether animal age impacts survival, given the known increase in mortality hazard with age (Prichard, Joly, and Dau 2012), we also fitted models with number of years since collaring as a main effect. To keep the results intuitive in terms of survival (i.e., positive effects positively affect survival), we report the negative of the Cox proportional hazards model coefficient and the corresponding 95% confidence intervals.

2.6. Consequences of Migration Shifts for Survival

Given the shift in winter range areas, we explored the survival consequences of the specific choice of crossing or not crossing the Kobuk River. We assessed whether an individual crossed the Kobuk River in each biological year monitored, using May 28 as the start of the “biological year” to align with the calving season, and determined whether those animals survived the following winter and spring. Because most animals were collared during fall migration, we eliminated each individual's first year of monitoring from the migration analysis (Prichard et al. 2023).

To test whether the collective decision to migrate south was adaptive, we examined the overall proportion of winter and early spring survivors in the population against the proportion of animals that did and did not cross the Kobuk River in a given year. We did this by fitting a binomial GLMM to the probability of an individual survival using only the proportion of individuals that crossed the Kobuk as a predictor. Thus, we fit the following model:

$$\text{logit}\left({\mathrm{s}}_{\mathit{iy}}\right)={\mathrm{\beta }}_c{C}_y+{\mathrm{\beta }}_w{\mathit{W}}_{\mathit{iy}}$$

where s _iy is the probability of individual i surviving winter and spring of year y, C _y is the proportion of all animals that crossed the river in the fall of year y, and W _iy includes any of the environmental covariates in winter (temperature, precipitation, snow depth, wind) that might have an effect on survival. We included all two‐way interactions and took the square root of the snow depth and precipitation variables to account for the long tails in those distributions while retaining the zeroes. If animals have similar mortality rates on either side of the river regardless of the proportion of animals that crossed (i.e., ${\mathrm{\beta }}_c$ = 0), then there is no evidence that the range shift is collective or adaptive. In contrast, if survival south of the river is higher in years when the majority of animals migrate south (${\mathrm{\beta }}_c>0$) and vice versa, then the range shift can be said to be adaptive. Importantly, we fitted this model separately for animals that did and did not cross the Kobuk in a given year.

To estimate the two models (animals that stayed north and went south), we fitted GLMMs with individual survival as a binomial response and random intercepts for individual. We used BIC as a criterion to compare models as it is more conservative than AICc and more reliable in cases with considerable unknown heterogeneity in the process (Brewer, Butler, and Cooksley 2016).

2.7. Causes of Migration Shifts

We performed an analogous analysis but with a (near) inversion of predictors and responses to test the hypothesis that the previous year's collective experience predicted the migratory behavior of animals the following year. For this, we modeled the probability that an individual crosses the river using the proportion of animals that survived as a predictor, but only in the general sector (north or south of the river) where the animal wintered the previous year. Thus, we broke the data into two subsets: those animal‐years that spent the previous winter south of the Kobuk River and those that spent the previous winter north of the river. Then, for each of those groups, we fitted the following model:

$$\text{logit}\left({\mathrm{c}}_{\mathit{iy}}\right)={\mathrm{\beta }}_s{S}_{y-1}+{\mathrm{\beta }}_f{\mathit{F}}_{\mathit{iy}}$$

where c _iy is the probability individual i crossed in year y, S _y‐1 is the proportion of caribou that survived where the caribou spent the previous winter, and F _iy represents a suite of environment variables experienced in early fall, that is, before potentially choosing to cross the Kobuk River (early fall temperature, precipitation, and wind—with two‐way interactions). The assumption behind this analysis was that the memory of conditions south (or north) of the Kobuk River would be predictive of returning and that the most direct measure of “conditions” is survival in the previous year.

As in the previous set of models, we fitted generalized linear mixed effects model with individual crossing of the river as a binomial response, a random intercept for individual, and used BIC as a criterion to compare models, with the same square root transformation on precipitation and snow depth.

Analysis was performed in R v. 4.3.0 (R Core Team 2023). Seasonal patterns in migration were identified with the cyclomort package (Gurarie and Thompson 2020); Cox proportional hazard models were fitted using the survival package (Therneau 2022); generalized mixed effects models were fitted using the glmmTMB package (Brooks et al. 2017), and model selection was facilitated with the MuMIn package (Bartoń 2023). We used Nakagawa's marginal R ² to assess the proportion of variance explained by the fitted models (Nakagawa, Johnson, and Schielzeth 2017; Nakagawa and Schielzeth 2013) and computed those using the performance package (Lüdecke et al. 2021).

3. Results

In total, we analyzed 326 caribou collared between September 7, 2009, and May 5, 2021 (i.e., biological years 2009–2020). The large‐scale winter range shift is illustrated in Figure 2, with an emphasis on the contrast between the winter of 2015–2016, when 69 of 82 animals (84%) wintered on the Seward Peninsula and no animals wintered on the Brooks Range, and 2020–2021, when no animals wintered on the Seward Peninsula and 49 of 65 (75%) wintered on the Brooks Range. The straight‐line distance between these areas is approximately 500 km.

Of the 326 caribou, 196 died during our study and 130 were censored (i.e., collars dropped off or the caribou were still alive at the time of analysis; Figure 1). The break point analysis (Figure 4c,d) showed that the clearest cutoff was in 2016. Specifically, the overall probability of crossing the Kobuk River decreased by 44% (95% CI 36%–52%) in biological years 2017–2020, and the probability of survival in the winter and spring decreased by 9.6% (6.6%–12.5%). We, therefore, classified biological years 2010 through 2016 as “pre‐cutoff” and 2017 through 2020 as “post‐cutoff” in related analyses.

FIGURE 4.

(a) Proportion of female Western Arctic Herd caribou monitored between 2010 and 2020 colored by those that crossed the Kobuk River (blue) and those that did not (orange) and by those that died in the subsequent winter and spring seasons (darker colors). (b) Estimated winter and spring survival by biological year for animals that wintered north (orange) or south (blue) of Kobuk River. Vertical bars are 95% confidence intervals, and areas are proportional to the number of collared individuals that displayed the respective migratory behavior. (c–d) Break‐point analysis comparing the proportional effect of splitting the caribou data before and after various years between 2012 and 2020 on the proportion of animals that (c) migrate/do not cross the Kobuk River and (d) that survive/do not survive. The y‐axis indicates the effect size of a chi‐squared test comparing proportions, with 95% confidence intervals. The cutoff includes all biological years up to and including the one indicated, thus for 2016 “migration” refers to the proportion of animals that crossed the Kobuk River in the fall up to and including fall of 2016; “survival” refers to the proportion of animals that survived in the winter and summer (postmigration) of 2017.

3.1. Interannual Differences in Survival and Migration

The mean annual survival was 0.82 (s.d. 0.06, n = 13 years), with considerable interannual variation: the lowest annual survival rate was 0.69 (61 survived/89 total) in biological year 2017 (May 28, 2017, to May 27, 2018), compared to the highest in biological year 2015 of 0.93 (88/95). The overall daily hazard, which estimates the probability of individual mortality per day accounting for data censoring, was 0.87 (95% CI 0.75–1.01) × 10⁻³ d⁻¹, equivalent to a (postcollaring) life expectancy of 2.71 years (95% CI 2.53–5.67).

Underlying the high level of interannual variation was a significant overall reduction in survival rates over the duration of the study. The pooled daily mortality hazard post‐cutoff was 0.85 (0.70–1.01) × 10⁻³, corresponding to a postcollaring life expectancy of 2.24 years (1.89–2.65), compared to a mean hazard of 0.58 (0.44–0.72) and life expectancy of 3.31 years (2.65–4.17) for animals collared pre‐cutoff (Table 1).

TABLE 1.

Single‐season cyclomort seasonal mortality model estimates (with 95% confidence intervals) for collared female Western Arctic Herd caribou in biological years (June through May) 2009 through 2020, before and after the 28 May 2016 cutoff date (Figure 6).

Parameter	2009–2015	2016–2020
Mean hazard (×10−3)	0.58 (0.44–0.72)	0.85 (0.70–1.01)
Postcollaring life expectancy (years)	3.31 (2.65–4.17)	2.24 (1.87–2.65)
Mortality season peak (date)	July 11 (May 23 to August 30)	March 26 (February 24 to April 26)
Mortality season duration (days)	139 (92–166)	133 (100–156)

There was even greater interannual variation, and a stronger trend, in the proportion of animals that crossed the Kobuk River during the fall migration. At the extremes, 42 of 45 (93%) of collared animals crossed the Kobuk River in 2013, compared to only 6 of 73 (8%) in 2020. Overall, pre‐cutoff, 208 of 278 animals (75%) crossed the Kobuk compared to only 174 of 458 (38%) after 2016.

While overall survival was not significantly influenced by whether individuals crossed the Kobuk River (p = 0.599; logistic regression accounting for significant interannual auto‐correlation and weighted by total number of animals), the interannual variability in survival was greater for animals that did cross versus the animals that did not: animals that wintered north of the Kobuk River had predicted survival point‐estimates between 0.87 and 0.90, while the probability of survival for animals that went south varied between 0.81 and 0.97 (Figure 4b, F‐test comparing variances p < 10e−4).

3.2. Seasonality of Survival

Along with the significant decrease in overall survival, there were highly significant shifts in the patterns of intra‐annual variation before and after the 2016 cutoff (Table 2, Figure 5). Pre‐cutoff, the mortality hazard was lowest (i.e., survival was highest) during winter at 0.47 (s.e. 0.32–0.7) × 10⁻³ d⁻¹ and highest during the calving season and insect‐harassment hazard period 0.71 (~0.3–1.2) × 10⁻³. In contrast, after 2016, the difference in seasonal mortality was marked by higher hazards in spring 2.52 (1.6–4.0) × 10⁻³ and late summer 2.13 (1.2–3.9) × 10⁻³.

TABLE 2.

Season‐specific mortality hazard values (with 95% confidence intervals) for Western Arctic Herd caribou in 2009–2021, pre and post the 2016 cutoff.

Season	Overall hazard (×10−3 days−1)	2009–2015	2016–2020
Winter	0.82 (0.63–1.06)	0.47 (0.32–0.7)	1.84 (1.31–2.59)
Spring	1.05 (0.74–1.49)	0.59 (0.34–1.00)	2.52 (1.59–3.99)
Calving	1.00 (0.51–1.94)	0.72 (0.29–1.75)	1.95 (0.72–5.29)
Insect harassment	0.68 (0.36–1.28)	0.71 (0.35–1.44)	0.58 (0.14–2.40)
Late summer	0.96 (0.63–1.47)	0.62 (0.34–1.13)	2.13 (1.16–3.89)
Fall	0.85 (0.64–1.12)	0.55 (0.37–0.83)	1.62 (1.10–2.38)

FIGURE 5.

Fitted three‐season hazard functions for female Western Arctic Herd caribou, northwest Alaska 2009–2021. The dark orange and purple line indicates the point estimate prediction of the hazard function before the May 2016 cutoff and after it, respectively, while the corresponding shaded areas indicated the 95% prediction intervals. Vertical lines represent the discrete seasonal cutoffs, with summer broken down into calving (“ca.”), insect harassment (“i.h.”), and late summer.

This shift in seasonal patterns is most clearly seen with the continuous‐time seasonal analysis provided by the cyclomort package, which does not rely on a priori seasonal definitions. Comparing fitted models with 0 (constant hazard), 1, 2, or 3 seasons with AIC, the lowest AIC was with the single‐season model that estimated a shift in the main peak of mortality from mid‐summer pre‐cutoff (peak mortality July 9, duration 139 days) to late winter (March 24, duration 133 days) post‐cutoff (Table 1). The two‐ and three‐season models were within 2 AIC values of the one‐season model for post‐cutoff years, suggesting essentially equivalent statistical support; and within 4 for the pre‐cutoff years. We therefore illustrate the three‐season models in Figure 5 to highlight the patterns and the shift between the two periods. In particular, the three‐season model pre‐cutoff revealed the highest mid‐summer peak (June 18, duration 56 days) centered during the insect harassment season, accounting for half the mortalities; a weaker peak in fall, around October 16, during fall migration; and one in later winter (February 19, duration 42 days). In contrast, after 2016, there were two neighboring peaks in late winter (January 31, duration 11 days) and during spring migration (April 5, duration 69 days) and a much more diluted peak in the fall (peak September 29, but with duration 106 days). No summer peak of mortality was detected in more recent years.

3.3. Environmental Predictors of Winter Survival

In all models of winter and spring survival, the inclusion of collar duration (the proxy for age) was not a significant predictor and was therefore excluded from further analyses. Otherwise, environmental covariates were, in nearly all cases, significant predictors of winter and early‐spring survival, but in highly contrasting ways for animals north and south of the Kobuk River (Figure 6). Animals wintering south of the Kobuk River had higher survival when they experienced warmer, windier winters with less daily precipitation, while snow depth, which accumulates and varies more slowly than precipitation, itself was not a significant predictor. In contrast, animals wintering north of the Kobuk River had higher survival when snow was relatively deeper and winds were lower, while precipitation and temperature were not significant. It bears noting that snow depth experienced by caribou was already generally substantially higher north of the Kobuk River (median depth 64 cm [interquartile range 49–83] compared to 44 cm [32–58]), while precipitation, temperatures, and winds were generally similar.

FIGURE 6.

Main effect of winter weather covariates (snow depth, wind speed, precipitation, maximum temperature) on probability of survival of female Western Arctic Herd caribou in winter for animals that wintered north (left panel) or south (right panel) of the Kobuk River. Gray colors represent nonsignificance (p > 0.05), and red and blue represent negative and positive significant effects on survival, respectively.

3.4. Shift in Migration Improved Relative Survival

To test whether migration choice was adaptively related to higher survival rates, we analyzed two groups of caribou‐years: those that crossed the river in a given year (n = 308, unique individual caribou = 144) and those that did not (n = 220, n ind. = 140). South of the Kobuk River, winter and spring survival was highest in those years when a higher proportion of animals overall migrated south (Figure 7), that is, the percentage of collared caribou that crossed the river (C _y ) was a significant and positive predictor of probability of surviving the winter and spring south of the Kobuk River. Specifically, the top models for predicting survival south of the river (Table 3a) included a strong positive effect of C _y (coefficient ${\mathrm{\beta }}_c$ = 3.47, s.e. 1.02, p < 0.001—green‐shaded area in Figure 7b). The other consistently significant predictors included a strong negative winter precipitation effect (consistent with the results in the survival analysis above, Nakagawa marginal R ² values for the top models ranged between 0.20 and 0.30).

FIGURE 7.

Schematic (left panel) and results (right panel) of analysis of the consequences of the migration shift. In the schematic, in a given year, if the choice of where to migrate, that is, whether to cross the Kobuk River (thick blue line), is correlated with survival (gray vs. red silhouettes), then relatively fewer animals will die south of the river when more animals cross the river (thick migration lines, upper panels) and vice versa (bottom panels). In the right panel, the winter and spring survival of female Western Arctic Herd caribou (y‐axis) against proportion of animals that migrated across the Kobuk River that year (x‐axis). Each point represents a distinct biological year (as labeled), orange open circles represent animals that wintered north of the Kobuk River, and green filled circles represent animals that crossed the Kobuk River. The line and corresponding shaded area are the modeled predictions and 95% confidence intervals of the relationship for those that did migrate based on the top fitted model (see text and Table 3a). The size of the circles is proportional to total number, for example, in 2020 only 6 of 73 animals crossed in the Kobuk, whereas in 2013, 42 of 45 animals crossed the Kobuk.

TABLE 3.

Model selection and summary table for generalized mixed effects models predicting survival against migration for animals that (a) did cross the Kobuk River and (b) did not cross the Kobuk. Survival in winter and spring was modeled against proportion of migrants in the preceding fall (Prop. Crossed) and a suite of environmental covariates and two‐way interactions: daily mean precipitation (PR), snow depth (SN, log scaled), maximum daily temperature (TM), and wind speed (WI). The entries under the columns represent coefficient values for the binomial regression (log‐odds slopes). The R ² value is the marginal R ² that does not take individual variation into account.

Model	Prop. Crossed	PR	SN	TM	WI	PR:SN	SN:TM	TM:WI	df	BIC	ΔBIC	Weight	R 2
(a) South of Kobuk
1	3.47	−0.40	0.38	—	—	−0.63	—	—	6	206.3	0.00	0.51	0.28
2	2.13	−0.69	—	—	—	—	—	—	4	209.1	2.79	0.26	0.20
3	3.46	−0.40	0.47	—	0.22	−0.60	—	—	7	211.2	4.92	0.04	0.30
4	3.49	−0.43	0.43	0.12	—	−0.59	—	—	7	211.8	5.45	0.03	0.29
(b) North of Kobuk
1	—	—	0.51	0.16	—	—	0.60	—	5	215.3	0.00	0.29	0.17
2	—	—	0.52	—	—	—	—	—	3	215.5	0.19	0.27	0.07
3	—	—	0.61	0.21	−0.13	—	—	−0.66	6	219.3	4.04	0.04	0.18
4	—	—	—	—	—	—	—	—	2	219.4	4.12	0.04	0.00

In contrast, an individual's probability of surviving north of the Kobuk River was not at all related to the proportion of collared caribou that crossed the Kobuk River that year (Table 3b), and the set of covariates that were included in the top models were quite different from those that predicted survival for the southern migrants, with the most consistently important one being a positive snow depth effect. Overall, the R ² values were substantially lower for the models of animals staying north of the Kobuk (peaking at 0.18), while the fourth best model according to the delta BIC ranking was the null model with no predictors (R ² = 0).

3.5. Previous Year's Survival Predicted Migration

To test whether previous year's survival predicted migration in the subsequent year, we again subdivided the data into two groups: those that had wintered south of the Kobuk River in a given year and were tracked the following year (n = 365, n.ind = 186) and those that had wintered north of the Kobuk in the preceding year (n = 123, n.ind = 84). The probability of crossing the Kobuk River in a given year was very strongly predicted by the collective survival of the previous winter south of the Kobuk, even when accounting for environmental predictors (Figure 8, Table 4a). Specifically, collective survival in the south was included in all top models as a significant positive predictor of individual migration. Among the environmental predictors of crossing the river, all top models according to BIC included positive effects of precipitation and temperature and a negative effect of wind. The marginal R ² values for these models were relatively high, ranging from 0.53 to 0.61 (Table 4). Fitting these top models without previous year's survival dropped the R ² values to between 0.19 and 0.2, that is, nearly two thirds of the variation in individual migration was explained by the collective survival of the preceding year.

FIGURE 8.

Schematic (left panel) and results (right panel) of analysis of the potential causes of migration shifts. In the schematic (upper left panel), if there was higher mortality (red silhouettes) south of the Kobuk River (thick blue line) the previous winter/spring period, we predict that relatively fewer caribou will migrate south of the Kobuk River the following fall (lower left panel). In the right panel, the proportion of animals that migrate across the Kobuk River in a given year (y‐axis) is plotted against the winter/spring survival among those animals that migrated in the previous year (x‐axis). Each circle represents a distinct biological year, with areas reflecting the sample size (i.e., the number of animals that wintered south of the Kobuk in the previous year). The line and shaded area reflect the main effect of the top fitted model (see text and Table 3a).

TABLE 4.

Model selection and summary table for generalized mixed effects models predicting whether an individual caribou crosses the Kobuk River in a given year against the collective survival (Surv.) in the previous winter (a) south of the Kobuk and (b) north of the Kobuk, as well as a set of environmental covariates and two‐way interactions: snow (SN, log scaled), precipitation (PR), and maximum daily temperature (TM). The covariates were measured in the early fall, before October 5, and therefore serve as predictors of crossing. All variables were scaled to facilitate comparisons except for survival (a proportion between 0 and 1). In each of the respective analyses, only those females that spent the previous winter south and north of the Kobuk, respectively, were analyzed as they have some memory of those conditions. As in Table 3, the entries under the columns represent coefficient values (log‐odds slopes) and the R ² value is the marginal R ².

Model	Surv. South	PR	TM	WI	PR.TM	PR.WI	TM.WI	df	BIC	ΔBIC	Weight	R 2
(a) South of Kobuk
1	10.21	1.46	1.07	−0.76	0.51	—	−0.86	8	356.7	0	0.79	0.61
2	10.45	1.32	0.88	−0.62	0.4	0.36	−0.93	9	359.8	3.11	0.17	0.58
3	9.8	1.06	0.59	−0.43	—	0.57	−1.01	8	363.2	6.47	0.03	0.53
(b) North of Kobuk
1	—	−2.38	−1.13	1.04	—	—	—	5	131.6	0	0.30	0.54
2	—	−5.83	—	—	—	—	—	3	132.4	0.86	0.20	0.36
3	—	−2.42	−1.17	0.96	—	—	0.68	6	133.5	1.89	0.12	0.55
4	—	−6.16	—	1.34	—	—	—	4	133.7	2.16	0.10	0.36
5	—	−2.14	−0.64	—	—	—	—	4	134.3	2.7	0.08	0.47
6	—	−2.3	−1.1	0.95	0.54	—	—	6	135.4	3.83	0.04	0.57

In contrast, for those animals that had wintered north in the previous year (n = 83 individuals, 122 caribou‐years), no model within 4 ΔBIC included previous winter's survival as a predictor of staying in the north (Table 4b). Otherwise, the environmental covariates that predicted remaining north for those animals were consistently the inverse of those that predicted animals going south (i.e., more likely to remain north with less precipitation and lower temperatures).

4. Discussion

By analyzing the survival and movements of over 300 caribou over a 12‐year period, we found a large‐scale winter range shift, with the centroids of subsequent winter locations, at times, over 500 km apart. This range shift was accompanied by significant changes in seasonal survival patterns and environmental predictors of survival. We found explicit evidence that the range shift was both adaptive, in the sense of improving an important metric of fitness, and likely driven by collective memory. As far as we are aware, this is the first evidence for a social and cognitive mechanism that helps explain large‐scale behavioral changes for a wide‐ranging terrestrial animal that was also supported by a fitness measure.

Caribou manage to maintain high numbers in extreme Arctic and sub‐Arctic environments that are not only relatively unproductive but also extremely variable. This variability is, most obviously, seasonal. But it is also interannual, with hard‐to‐predict variation in temperatures, precipitation, and snow cover and quality, and spatial, with a single herd's range including coastal tundra, mountains, and boreal forest, as well as smaller‐scaled variation within those biomes. While caribou are aided by an array of physiological adaptations (Klein 1992; Webber et al. 2022), perhaps their greatest tools for navigating the Arctic is a tremendous ability to move (Joly et al. 2019), along with the enhanced information gathering and memory abilities that come from a high level of sociality.

Natural selection for adaptive traits does not operate on a time‐scale fast enough to keep up with rapid, large‐scale changes in environmental conditions that confront long‐lived species in an unpredictable environment (Donelson et al. 2019). Thus, the main mechanism of adaptation is phenotypic plasticity—in particular, behavioral plasticity (Anderson et al. 2013; Donelson et al. 2019; Xu et al. 2021). For wide‐ranging animals, movement‐related plasticity is the most important mechanism of resilience to rapid climate change. However, it is impossible to assess that resilience without a robust and reliable measure of fitness (Xu et al. 2021).

Our analysis of caribou survival data revealed varied and complex patterns: survival depended on season, year, migratory behavior, and various proximal and lagged environmental variables. Nonetheless, we found strong evidence for an adaptive range shift by (a) showing that the range has indeed shifted, (b) collecting a robust measure of fitness in mortality data, and (c) demonstrating that range shift corresponded to improved relative survival measures (i.e., that the choice of new winter range corresponded to increased relative survival against the older alternative). Furthermore, we documented evidence that memory of conditions, as proxied by the overall survival of other animals, is predictive of migration choices in subsequent years. Each of these components are discussed in detail below.

4.1. Choice of Winter Range

We focused our analysis on the choice of winter ranges, which delineate the extrema of migratory caribou ranges. WAH winter ranges can be as far from each other from one winter to the next as they are from the calving grounds (Joly et al. 2021). Once that particular winter range is chosen and winter sets in, movements tend to be more limited and caribou are unlikely to make any further major displacements. The decision of where to winter, made in fall, is therefore vital and more or less committal.

WAH caribou can be found wintering almost anywhere throughout their range. There was, for example, no year when some animals were not found along the coastal plain to the north of the calving range. Nonetheless, the majority of animals in a given year are found in one or another particular area (Joly et al. 2021). Between 2009 and 2016, the large majority of animals crossed the Kobuk River in early fall and migrated to the tundra of the northern portion of the Seward Peninsula or the Nulato Hills region directly east of the peninsula (Joly, Chapin, and Klein 2010; Joly et al. 2021). This appears to have been a largely successful strategy, as overall survival was high and winter survival was especially high. For example, between 2012 and 2015, 242 out of 246 (98.4%) collared individuals that wintered in the Seward Peninsula survived. Migration to the northern Seward Peninsula is among the longest journeys WAH caribou have made, with consequently higher energetic costs. The inference is that in good years, the energetic rewards compensated the costs. It is likely that the success of this strategy is linked to the maritime tundra in that area being rich in terrestrial lichen (Joly, Chapin, and Klein 2010; Joly et al. 2021; Joly and Cameron 2018). Lichen are an essential resource for migratory caribou (Webber et al. 2022) and thought to be important for overwinter adult survival (Joly, Chapin, and Klein 2010). The link between lichen cover and survival is an important topic for future work, especially in relation to how other environmental covariates might impact access to lichen.

Both migration propensity and survival changed substantially after 2016. In 2015–2016, 95 of 108 (88%) collared animals wintered on the Seward Peninsula, whereas between 2019 and 2020 only 4 of 370 (1%) wintered there. While a few animals consistently wintered north and west toward the North Slope coastline, the majority of animals in that latter period moved eastward into the Brooks Range mountains, a much different environment than the southwestern portion of the range. The mountains generally have deeper snow and more rugged terrain than the Seward Peninsula, while the North Slope is marked by sparser shrub cover and less snow. Further underlining the differences in these winter ranges, many of the predictors of survival are inverted in the northern ranges, with deeper snow and less wind more favorable to survival. Higher survival under deeper snow conditions is consistent with observations in other caribou populations, both migratory (Sharma, Couturier, and Côté 2009) and nonmigratory (Schmelzer et al. 2020), and had been attributed to a greater ability to evade predation in deep, soft snow.

The shift in migration patterns was accompanied by an unexpected shift in seasonal mortality patterns. Greater total mortality in late winter, with the seasonal peak right about when the spring migration begins, occurred in the latter portion of our study. High mortality near the end of winter suggests that the caribou generally had a harder time maintaining or increasing their energetic reserves throughout the winter months. The overall lower, but less variable, survival rates north of the Kobuk River suggest an environment where conditions are not as favorable but are at least broadly similar from year to year. Meanwhile, the somewhat atypical (for caribou) late‐winter peak of mortality post‐2016 is more consistent with the food‐limited survival bottleneck typical of temperate ungulates (Kautz et al. 2020; Parker, Barboza, and Gillingham 2009). It is important to note that the body condition of females is also directly related to lower parturition rates (Gerhart et al. 1996, 1997), lower calf birth weights (Skogland 1986), and lower summer calf survival (Cameron et al. 1993). Thus, even if the adult mortality was mitigated, however imperfectly, by wintering in the Brooks Range, there may be an additional, multiplier effect on overall demographics via reduced recruitment (White 1983).

The observed range shift occurred over a time period when the overall population of the Western Arctic Herd has fallen by about 50% (from about 348,000 in 2009 to around 164,000 in census of 2022). An overall smaller population may also be a fundamental driver of a contracted range through some density dependent mechanisms. Indeed, there was an overall shortening of distances traveled; most dramatically in 2019–2020, where the median total distance traveled (for those animals that were tracked calving ground to calving ground) was only 1968km (interquartile range: 1853‐2147, n = 38) compared to 2013–2014 with a median total distance of 3080km (2804‐3196, n = 37). However, the overall range of the WAH caribou did not substantially contract over this period (e.g., the maximum convex polygon, MCP, of all locations in biological year 2015 was, at 156,000 km², in fact smaller than the MCP in 2020: 226,000 km²). Furthermore, the spatial concentration of animals on the lichen‐rich tundra of the Seward Peninsula—when that area was used—was always much higher than that in the Brooks Range. The changes in the winter ranges of the WAH were shifts rather than contraction. While sorting out the cause‐and‐effect relationships of range shifting and total abundance can be difficult, our results point toward the impact of the former on the latter.

4.2. Adaptive Decision Making

Overall winter survival was generally less variable north of the Kobuk River than south. Over a consecutive period of biological years (2016–2020) when the fewest animals crossed the Kobuk River, survival was substantially higher in the north than in the south. Migration across the river could thus be considered in a risk–reward framework, with the rewards generally outweighing the risks in earlier years, and no longer in recent years. From a population‐level fitness perspective, the choice to remain north of the river was clearly advantageous, even if it did not fully mitigate the overall high mortality of recent years.

We posit that caribou learned through individual and social experience that migrating south was risky. This memory‐driven hypothesis is supported by the high predictive power of the previous year's survival on the probability of migrating south. Memory of conditions in the south was by far the greatest single predictor of crossing the river in the following year. Uniquely, those “conditions” were not inferred from available, remotely sensed, environmental covariates, but directly from the survival outcomes of conspecifics that also undertook the longer migration south. Perhaps, the most illustrative example is from the Seward Peninsula. At the start of our study period, survival there was high and caribou continued to migrate south to return to this region in consecutive winters, eventually making it the most used wintering area for the herd. However, for currently unknown reasons, winter survival sharply declined in 2016 (from an overall 94% from 2010 to 2016 to 73% in 2017–2018), after which fewer caribou returned. The Seward Peninsula has been largely abandoned as a winter range since 2018.

There are several important assumptions behind this argument, and some alternative explanations to consider. A major assumption is that an animal can “detect” the high mortality of conspecifics. However—as our other analyses show—mortality is clearly related to broad environmental conditions, which are common to larger areas. A rough winter for one animal is a rough winter for all the animals in the vicinity. More importantly, the strength of the inference is somewhat weakened by the relatively short time series. It could be that there are simply independent, unexplained trends over time occurring: one of worse survival rates on the Seward Peninsula and one of increased movements to the Brooks Range. However, models that also accounted for those effects (e.g., by estimating a trend or that included a random effect for year) still consistently showed a strong positive predictive effect of previous year's survival. Overall, these results are consistent with a growing empirical body of work that suggests that memory may more important than environmental cues when it comes to determining the movements of long‐lived ungulates (Falcón‐Cortés et al. 2021; Gurarie and Avgar 2024; Rheault et al. 2021).

A more in‐depth analysis, reserved for future work, should look at finer‐scaled choices of wintering ranges to account for the considerable remaining variation beyond the separation conveniently provided by the Kobuk River. Our analysis is ultimately quite coarse—separating wintering ranges spatially into “north of Kobuk” and “south of Kobuk” and temporally before and after May 2016. There is great variability within those ranges and time spans, and occasional grouped mortality events which may have significant impacts on all results. As noted previously, within “north of Kobuk,” the central Brooks Range is very different from the North Slope, but it is also quite distinct from the western extent just above Kotzebue Sound, the site of the Red Dog zinc and lead mine—one of the largest in the world—along with a 90‐km road leading to the mine from its coastal port site. This area was the site of several mortality clusters, notably in the winter of 2015–2016. On the south side of the river, in the winters of 2011–2012 and 2016–2017, a number of collared caribou wintered far in the eastern edge of the range going so far as to cross the Koyukuk River, a tributary of the Yukon River. In both cases, mortality was locally very high for reasons that are currently unknown, but that have nothing to do with conditions on the Seward Peninsula 500 km to the west. Nonetheless, and in spite of these potentially distorting nuances, the strong support for the relationships we outlined points to a very real large‐scale phenomena of a memory‐driven, adaptive, large‐scale range shift.

4.3. Collective Decision Making

An individual caribou's decision making is driven, in part, by social decision making and learning. Our results are consistent with the potential of animals to improve their fitness with collective knowledge and collective behaviors. There is considerable evidence, mainly from theoretical studies, that animals in large social groups collectively improve their navigation, nonlocal information, and overall decision making (Berdahl et al. 2018; Gurarie et al. 2021; Martínez‐García et al. 2013). A combination of memory and cue following has been shown to be the most efficient mechanism for helping migratory organisms adapt to dynamic environmental conditions, whether characterized by spatial or temporal trends or increasing stochasticity (Gurarie et al. 2021). Caribou (specifically, the WAH) have been shown to make decisions based on memory‐based anticipation of resource quality, specifically to explain large‐scale fidelity to calving grounds, while refining the habitat selection within those grounds according to finer‐scaled signals of environmental stochasticity and resource availability structure (Cameron et al. 2020).

Observing the actual mechanisms of collective decision making directly in an empirical system, however, is a considerable challenge. The collective nature of caribou is multilayered. Caribou herds can number in the hundreds of thousands, and the calving and postcalving movements to insect‐harassment relief areas (particularly so for the WAH) can involve a large majority of the total population and astonishing spatial aggregation. Certain large‐scale phenomena—such as the start of migration (Gurarie et al. 2019) and parturition timing (Couriot et al. 2023)—are impressive in their temporal synchronization, even over large (continental) spatial scales. But in other times of year (especially winter), caribou herds are widely dispersed. Even then, migratory caribou are rarely found in isolation (Pruitt 1959). Social groupings share the burden of predator vigilance (Baskin and Hjältén 2001; Bøving and Post 1997) and increase access to food through shared cratering yards (Pruitt 1959). Even harder to observe directly is the role that sociality and communication might play for adaptive decision making. There is a widely hewn tradition in Indigenous hunting practices of “letting the leaders pass” (Padilla and Kofinas 2014), which emerges from long‐standing observations of animals in groups following a leader—which is allowed to pass to leave a trail for the other animals to follow. This kind of leader‐following behavior is well documented for domestic reindeer (also Rangifer tarandus ), and the identification and use of lead animals to facilitate herd management is a key tool in the practice of reindeer herding (Baskin 1989). Close observations of wild woodland caribou reveal a complex and fairly stable hierarchy of leadership and dominance (Barrette and Vandal 1986). Given the size of the WAH, around 152,000 individuals in 2023, there may be a hierarchy of leadership or a group consensus that emerges not only on an individual level but on the level of the smaller subgroups or bands.

Further investigation of the mechanisms of information transmission and collective decision making requires closer investigation of interindividual interactions, as well as individual consistency. In systems—like reintroduced whooping cranes (Abrahms et al. 2021; Mueller et al. 2013; Teitelbaum et al. 2016)—where older individuals clearly initiated and led shifts in migrations, studies benefitted from much smaller populations, nearly complete coverage of monitored individuals, and a simpler type of end‐to‐end migration. Nevertheless, sociality and interactions can be explored not only by looking at individual movements but at pair‐wise interactions among all individuals. Furthermore, novel technologies, including animal‐borne cameras and animal‐borne acoustic recorders, can provide direct observations of sociality. Ultimately, understanding the caribou's ability to navigate a vast, heterogeneous, dynamic, and rapidly changing environment may require a novel framework that goes beyond analyzing a myriad of individual movements and assesses the herd collectively as an adaptive meta‐organism, which fluctuates seasonally and interannually in location, density, dispersal, sociality, and spatial memory.

Author Contributions

Eliezer Gurarie: conceptualization, formal analysis, investigation, methodology, software, supervision, visualization, writing – original draft, writing – review and editing. Chloe Beaupré: data curation, formal analysis, investigation, methodology, software, visualization, writing – original draft, writing – review and editing. Ophélie Couriot: data curation, formal analysis, investigation, methodology, software, visualization, writing – review and editing. Matthew D. Cameron: conceptualization, data curation, formal analysis, methodology, validation, writing – review and editing. William F. Fagan: funding acquisition, project administration, supervision, writing – original draft, writing – review and editing. Kyle Joly: conceptualization, data curation, funding acquisition, investigation, project administration, resources, supervision, writing – original draft, writing – review and editing.

Conflicts of Interest

The authors declare no conflicts of interest.

Open Research Statement

GPS data used in this study are owned by the National Park Service and Alaska Department of Fish and Game, both of which restrict the availability of location data for a harvestable species. GPS data are stored in the public repository IRMA and are available from the project leads on reasonable request: https://irma.nps.gov/DataStore/Reference/Profile/2260262.

Data Availability Statement

The data that support the findings of this study are openly available in Dryad at https://doi.org/10.5061/dryad.rfj6q57km. GPS data used in this study are not publicly available due to statutory constraints on sharing of data on harvested wildlife in Alaska and but are available upon reasonable request. Temperature and precipitation data are available from Oak Ridge National Laboratory at https://cds.climate.copernicus.eu/datasets/derived‐era5‐land‐daily‐statistics?tab=overview. Wind data are available from European Union's Earth Observation Programme, Copernicus, at https://climate.copernicus.eu/climate‐reanalysis. Snow depth data are available from the National Oceanic and Atmospheric Administration (NOAA) Physical Sciences Lab at https://downloads.psl.noaa.gov/Datasets/NARR/Dailies/monolevel/.