Movement decisions driving metapopulation connectivity respond to social resources in a long-lived ungulate, bighorn sheep (Ovis canadensis)

Abstract

The spatial availability of social resources is speculated to structure animal movement decisions, but the effects of social resources on animal movements are difficult to identify because social resources are rarely measured. Here, we assessed whether varying availability of a key social resource—access to receptive mates—produces predictable changes in movement decisions among bighorn sheep in Nevada, the United States. We compared the probability that males made long-distance ‘foray’ movements, a critical driver of connectivity, across three ecoregions with varying temporal duration of a socially mediated factor, breeding season. We used a hidden Markov model to identify foray events and then quantified the effects of social covariates on the probability of foray using a discrete choice model. We found that males engaged in forays at higher rates when the breeding season was short, suggesting that males were most responsive to the social resource when its existence was short lived. During the breeding season, males altered their response to social covariates, relative to the non-breeding season, though patterns varied, and age was associated with increased foray probability. Our results suggest that animals respond to the temporal availability of social resources when making the long-distance movements that drive connectivity.

This article is part of the theme issue ‘The spatial–social interface: a theoretical and empirical integration’.

1. Introduction

Metapopulation connectivity has consequences for fitness-relevant processes such as gene flow, demography and pathogen transmission [1–4]. Movement decisions that create connectivity result from intrinsic motivations modulated by extrinsic environmental attributes [5]. Intrinsic attributes, such as sex, age and physiological condition, affect an individual’s motivation to make specific movements [6]. Extrinsic attributes, including both biophysical attributes like access to food, water and shelter, and social attributes like competition, predation and breeding opportunity, interact with intrinsic motivators to inform when and where individuals move [5–7].

To accurately predict or effectively manipulate connectivity, we need to link movement decisions to the environmental attributes that drive them. To date, most connectivity models have focused on how the biophysical environment drives movement decisions. However, focusing on the biophysical environment without also exploring the social environment’s role may produce an incomplete assessment of connectivity, especially in systems where temporally dynamic social environments may drive seasonal shifts in connectivity patterns.

Here, we explore how a temporally dynamic attribute of the social environment, breeding opportunity, affects the long-distance exploratory movements (forays) that give rise to connectivity in bighorn sheep (Ovis canadensis). We explore this effect across three ecoregions where the duration of the breeding season varies. We hypothesize that access to breeding opportunities will inform ram movements. Specifically, we postulate that: (i) intrinsic attributes affecting an individual’s ability to exploit breeding opportunities will affect how the individual responds to resource availability; and (ii) when breeding opportunities are not continuously available through time, the drive to acquire breeding opportunities increases as the window for acquisition decreases. If present, variation in social resource availability and the intensity of its exploitation could drive regional variation in temporal patterns of connectivity.

2. Study system

Bighorn sheep are habitat specialists that range across most of western North America. Long-distance exploratory movements, referred to here as ‘forays’ [8,9] and sometimes referenced as ‘breeding migrations’ [10], are hypothesized to arise primarily from males (rams) searching for breeding opportunities [11–15]. While forays predominantly occur during the breeding season, they can occur throughout the year [16].

Our focus is on bighorn sheep living in the state of Nevada, United States (figure 1). In this part of their range and throughout the intermountain west, bighorn sheep live in distinct, topographically defined mountain ranges, leading to spatially structured, demographically independent herds connected through relatively rare inter-herd movements [17,18]. Nevada is home to three ecoregions, each with distinct geology, climate, hydrology and vegetation [19]. The Northern Basin and Range ecoregion consists of isolated high elevation mountain ranges with hot summers and cold winters and is the coolest and wettest ecoregion. The Central Basin and Range ecoregion has more connected mountain ranges with hot summers and warm winters and intermediate precipitation levels. The Mojave Basin and Range ecoregion has scattered low-elevation mountain ranges with hot summers and warm winters and is the warmest and driest ecoregion. Bighorn sheep occupy all three ecoregions, with the California bighorn sheep subspecies (O.c. californiana) [20] occupying the Northern ecoregion and the desert bighorn sheep subspecies (O.c. nelsoni) inhabiting both the Central and Mojave ecoregions. While a few populations are remnants of historic native herds, many have been established by translocation over the last half-century [21].

Figure 1.

Map of Nevada bighorn herds coloured by ecoregion. The Northern ecoregion is populated by California bighorn sheep (O.c. californiana), while the other two ecoregions are populated by desert bighorns (O.c. nelsoni).

The timing and duration of the bighorn sheep breeding season varies between ecoregions, probably owing to environmental constraints on lambing. Lambing is usually timed so that lactation costs to females (ewes) peak near maximum plant nutrient availability, while still being early enough to allow for sufficient lamb growth prior to winter or periods of intense drought [22,23]. The window of time meeting both of these requirements is fairly limited in the Northern ecoregion but expands to cover more of the year moving south. Since bighorn sheep gestation is very consistent across individuals and populations [24–26], breeding follows a parallel pattern across Nevada’s ecoregions: breeding opportunities are briefly available for rams in the Northern ecoregion, available for an intermediate period in the Central ecoregion and available for a longer period in the Mojave ecoregion [23,27,28].

Ram reproductive success depends on the local competitive environment [29]. Rams establish a largely age-associated linear dominance hierarchy at the beginning of breeding season [10,15,30], and while rams reach sexual maturity around 1.5 years of age, a small number of dominant rams are responsible for the majority of mating events [31–35]. High-ranking rams protect and mate with ewes during their period of oestrus, which lasts approximately 2 days [29]. Rams typically only tend one ewe at a time, so a single group of ewes can support multiple non-competing rams if multiple ewes are in oestrus simultaneously. Lower ranking rams can attempt to mate by out-running the tending ram in a coursing chase or by switching groups in an effort to improve their relative ranks [29].

We modelled how the temporal availability of mating opportunities mediated the probability of rams foraying across three distinct ecoregions in Nevada. We predicted that: (i) the probability of foraying will increase during the breeding season, enabling rams to better exploit mating opportunities; (ii) during the breeding season, the magnitude of change in foray probability will be greatest when the breeding season is brief and rams have a limited time to maximize breeding opportunities; (iii) a ram’s internal state, measured by age, will modulate the magnitude of their response to mating opportunities based on their relative potential to successfully breed; and (iv) rams will be less likely to foray with increasing availability of breeding opportunities, measured by the number of ewes and sex ratio within the population.

While similar questions have been studied in detail in several intensively studied populations of this same species [10,29,36], here we attempt to assess them at a much larger spatial scale using methods and models analogous to approaches used in studies of the biophysical environment. We hope that by casting these questions under the same methodological umbrella used for biophysical drivers of movement, we can help bridge the gap currently dividing the spatial–social interface.

3. Methods

(a). Global positioning system data collection and processing

Between 2010 and 2023, 220 bighorn rams were captured and fitted with global positioning system (GPS) collars as part of Nevada Department of Wildlife’s (NDOW’s) standard herd monitoring (92 rams in the Northern ecoregion, 79 in the Central ecoregion and 49 in the Mojave ecoregion). Collars were deployed for 10–1789 days (mean = 426 days), collecting 84 878 days of location data in total with collar fix rates ranging from 1 to 24 h. GPS data from these collars were organized into individual trajectories of consecutive fixes. Trajectories were split into separate ‘bursts’ when more than 36 h elapsed between consecutive fixes using the adehabitatLT R package [37], and bursts with fewer than 10 fixes were excluded from analyses.

(b). Identification of forays

We fit hidden Markov models (HMMs) to identify exploratory movements using the MomentuHMM R package [38]. HMMs identify and classify fixes into latent behavioural states by segmenting GPS trajectories based on persistent patterns of step lengths and turning angles (e.g. [39,40]). Because forays are multi-day events [8–10,16], we used a 24 h fix rate, which allowed us to maximize the number of individuals included in the analysis without constraining our ability to pick up the behaviour of interest [41].

Bighorn sheep tend to forage in one patch for several days and then travel to a different patch. As such, we defined initial parameter values for three daily behavioural states, treating step lengths as arising from gamma distributions and turning angles as arising from Von Mises distributions. Our a priori expectation was that the states would reflect: (i) a foraging state with small step lengths (gamma distribution with a mean of 200 m and standard deviation of 200 m); (ii) a transiting state (moving between foraging patches) with medium step lengths (gamma distribution with a mean of 2000 m and standard deviation of 2000 m); and (iii) an exploratory state with long step lengths (gamma distribution with a mean of 5000 m and standard deviation of 2000 m). Initial values for turning angles were uniform for all states (Von Mises distribution with a mean of π and a concentration of 0.1). Owing to variation in the timing of the breeding season, we fitted separate HMMs for each ecoregion, pooling all individuals within the ecoregion [42]. To account for seasonal variability in foray occurrence, we allowed state transition probabilities to vary according to a cyclic cubic spline that forced continuity between the beginning and end of the calendar year [43,44]. We first fitted the HMMs without the time-varying component and then used those estimates as initial parameter values for models that included the cyclic cubic spline fit to week of year.

Owing to variation in fix rates between trajectories, we used the momentuHMM’s multiple imputation approach and MIfitHMM function [45]. A continuous-time correlated random walk model was used to simulate 10 datasets with locations imputed to adjust for missing fixes and to standardize fix rate to a 24 h interval [38,46]. An HMM was fitted to each simulated dataset. To prevent state label switching between simulations, we used a pseudo-design matrix [38] that required foraging movements to have shorter mean step lengths than transitional movements and transitional movements to have shorter mean step lengths than foray movements. State-dependent step-length and turning angle distribution estimates and standard errors were pooled across imputations. For each imputation, the Viterbi algorithm was used to assign the most likely sequence of behavioural states [47], and the final state sequence was based on each step’s most frequent Viterbi assignment across imputations.

(c). Social model covariate development

Ram ages were estimated at capture by counting horn annuli and updated annually to account for ageing. Annual measurements of number of ewes and sex ratio (number of ewes : number of rams) were extracted from aerial surveys conducted by NDOW when available [48]. Distributions of covariate values and the number of missing covariates are shown in the electronic supplementary material, figure S1 and table S1. The timing of the breeding season for each ecoregion was determined by compiling published rut and lambing records with area-biologist expertise (electronic supplementary material, figure S2). We defined the breeding season as lasting from 25 October to 15 December (seven weeks) in the Northern ecoregion, from 21 August to 30 November (14 weeks) in the Central ecoregion and from 1 July to 30 November (21 weeks) in the Mojave ecoregion. All herds within an ecoregion were assumed to have the same breeding season. Rut status was captured through an indicator variable that took on the value 1 during the breeding season and 0 otherwise.

(d). Social model construction and interpretation

We connected the probability of rams being on foray to age and attributes of the social environment using a Bayesian framework. The probability that the Viterbi algorithm assigned each step to the foray state was modelled as a binary response using a logit link function. Predictors included rut status, sex ratio, number of ewes and linear and quadratic effects of age. To explore how breeding opportunities modulated the role of the social environment, we included interactions between rut status and sex ratio, number of ewes and age. Finally, we included a hierarchical effect on ram identity (ID) to account for pseudoreplication of data within each individual. Continuous covariates were centred and scaled globally prior to inclusion in the models (electronic supplementary material, table S2). Ultimately, we fitted the model below separately within each ecoregion:

$${\displaystyle \mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}(\mathrm{P}\mathrm{r}[\mathrm{f}\mathrm{o}\mathrm{r}\mathrm{a}\mathrm{y}])={\mathit{\beta }}_0+{\mathit{\beta }}_{\mathrm{R}\mathrm{u}\mathrm{t}}{\mathit{I}}_{\mathrm{R}\mathrm{u}\mathrm{t}}+{\mathit{\beta }}_{\mathrm{E}\mathrm{w}\mathrm{e}}\mathrm{E}\mathrm{w}\mathrm{e}+{\mathit{\beta }}_{\mathrm{E}\mathrm{w}\mathrm{e}\mathrm{x}\mathrm{R}\mathrm{u}\mathrm{t}}(\mathrm{E}\mathrm{w}\mathrm{e}\ast {\mathit{I}}_{\mathrm{R}\mathrm{u}\mathrm{t}})+{\mathit{\beta }}_{\mathrm{S}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}}\mathrm{S}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}+{\mathit{\beta }}_{\mathrm{S}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{x}\mathrm{R}\mathrm{u}\mathrm{t}}(\mathrm{S}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\ast {\mathit{I}}_{\mathrm{R}\mathrm{u}\mathrm{t}})+{\mathit{\beta }}_{\mathrm{A}\mathrm{g}\mathrm{e}}\mathrm{A}\mathrm{g}\mathrm{e}+{\mathit{\beta }}_{\mathrm{A}\mathrm{g}\mathrm{e}\mathrm{x}\mathrm{R}\mathrm{u}\mathrm{t}}(\mathrm{A}\mathrm{g}\mathrm{e}\ast {\mathit{I}}_{\mathrm{R}\mathrm{u}\mathrm{t}})+{\mathit{\beta }}_{\mathrm{A}\mathrm{g}{\mathrm{e}}^2}\mathrm{A}\mathrm{g}{\mathrm{e}}^2+{\mathit{\beta }}_{\mathrm{A}\mathrm{g}{\mathrm{e}}^2\mathrm{x}\mathrm{R}\mathrm{u}\mathrm{t}}(\mathrm{A}\mathrm{g}{\mathrm{e}}^2\ast {\mathit{I}}_{\mathrm{R}\mathrm{u}\mathrm{t}})+{\mathit{b}}_{\mathrm{r}\mathrm{a}\mathrm{m}\mathrm{I}\mathrm{D}}}$$

where b _{ram ID}~N(0, σ _ram).

We assigned all regression coefficients Gaussian priors centred at 0 with a variance of 10 and estimated missing covariates as part of the model’s joint posterior [49]. Because the covariates were standardized prior to inclusion, we took missing covariate values to arise from a Gaussian distribution centred at 0 and with variance of 1. We placed a half-t prior distribution of the standard deviation of ram-specific random intercepts and took those intercepts to arise from a normal distribution centred at 0 [50].

The models were fitted using Markov chain Monte Carlo samplers implemented in the R package NIMBLE [51,52]. Models were fitted with three chains consisting of 1 000 000 steps, of which the first 900 000 steps were discarded as burn-in. Chains were thinned so that every tenth step contributed values to the posterior. We assessed convergence using Gelman–Rubin statistics, and we used posterior predictive checks implemented in the DHARMa package in R [53] to assess compliance with model assumptions and temporal autocorrelation. We exported baseline and interaction-adjusted values for each effect as derived quantities (e.g. β _Age and β _Age|Rut). Covariate effects were interpreted as the multiplicative change in the odds of being on foray given a one standard deviation increase in the covariate. To facilitate coefficient interpretation, age effects were interpreted as the probability of being on foray at a given age, assuming ecoregion-specific mean sex ratios and ewe counts. Breeding opportunity effects were interpreted as the probability of foraying at a given number of ewes, assuming ecoregion-specific mean ram counts and the associated sex ratio.

4. Results

(a). Hidden Markov model fit and foray identification

State-dependent step length and turning angle distributions were similar among ecoregions (electronic supplementary material, figure S3 and table S3), suggesting broadly conserved movement behaviours. Overall, 17.0% of steps in the Northern ecoregion (n = 6196 steps), 23.4% of steps in the Central ecoregion (n = 6539 steps) and 11.8% of steps in the Mojave ecoregion (n = 2432 steps) were assigned to the foray state. See the electronic supplementary material, figure S4 for an example movement trajectory coded by the most likely state assignments.

(b). Social model fit and convergence

The Bayesian models converged with individual and multivariate Gelman–Rubin diagnostics of less than 1.05. Posterior predictive checks showed neither over- nor under-dispersion, nor problematic deviation in the residuals (electronic supplementary material, figure S5).

(c). Effect of breeding season on foray probability

The proportion of steps assigned to the foray state changed throughout the year. The probability of being on foray was greatest during and immediately prior to each ecoregion-specific breeding season (figure 2). The maximum observed probability of being in the foray state was 0.55 in the Northern ecoregion, 0.50 in the Central ecoregion and 0.29 in the Mojave ecoregion. The odds of foraying increased during the breeding season in all ecoregions (figure 3a ; Northern ecoregion: exp(β _Rut) = 2.93, 95% posterior credible interval (PCI) = (2.48, 3.47); Central ecoregion: exp(β _Rut) = 4.37, 95% PCI = (3.84, 4.95); Mojave ecoregion: exp(β _Rut) = 4.02, 95% PCI = (3.44, 4.66)).

Figure 2.

Estimated probability that a ram’s movement was assigned to the foray state over the course of the year for the Northern, Central and Mojave ecoregions. The breeding season for each ecoregion is highlighted in grey.

Figure 3.

Effect of rut, social resources and age on foray behaviour. (a) The multiplicative changes in the odds of foray associated with rut and one standard deviation increases in sex ratio (ewe : ram) and number of ewes for the non-breeding and breeding seasons. Lines show 95% posterior credible intervals. (b) The effect of age on the probability of being on foray during the non-breeding and breeding seasons when sex ratio and number of ewes are set to ecoregion-specific mean values. Polygons cover 95% posterior credible intervals.

(d). Effect of age on foray probability

The probability of being on foray decreased with age in the Northern and Mojave ecoregions, but age had limited effect on foray probability in the Mojave ecoregion during the non-breeding season (figure 3b ; Northern: exp(β _Age) = 0.61, 95% PCI = [0.57, 0.68], exp(β _Age ²) = 0.85, 95% PCI = [0.83, 0.88]; Central: exp(β _Age) = 0.69, 95% PCI [0.63, 0.76], exp(β _Age ²) = 1.19, 95% PCI [1.13, 1.24]; Mojave: exp(β _Age) = 1.02, 95% PCI [0.88, 1.21], exp(β _Age ²) = 1.07, 95% PCI [1.03, 1.13]). Conversely, age had a positive effect on foray probability in all ecoregions during the breeding season, with middle-aged to older rams most likely to foray in the Northern and Central ecoregions (Northern: exp(β _Age|Rut) = 1.53, 95% PCI [1.33, 1.72], exp(β _Age ² _|Rut) = 0.84, 95% PCI [0.79, 0.89]; Central: exp(β _Age|Rut) = 1.26, 95% PCI [1.13, 1.40], exp(β _Age ² _|Rut) = 1.21, 95% PCI [1.06, 1.19]; Mojave: exp(β _Age|Rut) = 2.37, 95% PCI [2.03, 2.81], exp(β _Age ² _|Rut) = 0.81, 95% PCI [0.77, 0.85]).

(e). Effect of breeding opportunities on foray probability

During the non-breeding season, the effect of the social environment on foray probability varied between ecoregions (figure 3a ; electronic supplementary material, figure S6). Sex ratio had no effect on foray probability in the Northern and Central ecoregions (Northern: exp(β _SRatio) = 0.99, 95% PCI = [0.89, 1.10]; Central: exp(β _SRatio) = 1.12, 95% PCI = [0.88, 1.42]) but increasing the ratio of ewes to rams decreased foray probability in the Mojave ecoregion (exp(β _SRatio) = 0.75, 95% PCI = [0.65, 0.87]). In the Northern ecoregion, the number of ewes in the population had a positive effect on foray probability (exp(β _Ewe) = 6.28, 95% PCI = [3.96, 10.29]). In the Central ecoregion, the number of ewes in the population had a negative effect on foray probability (exp(β _Ewe) = 0.44, 95% PCI = [0.36, 0.52]). In the Mojave ecoregion, the number of ewes in the population had no effect on foray probability (exp(β _Ewe) = 0.90, 95% PCI = [0.72, 1.08]).

During the breeding season, increasing the ratio of ewes to rams increased foray probability in the Northern and Central ecoregions (Northern: exp(β _SRatio|Rut) = 1.47, 95% PCI = [1.34, 1.62]; Central: exp(β _SRatio|Rut) = 2.16, 95% PCI = [1.77, 2.68]) but had no effect in the Mojave ecoregion (exp(β _SRatio|Rut) = 1.01, 95% PCI = [0.87, 1.17]). The number of ewes in the population had a positive effect on foray probability in the Northern ecoregion (exp(β _Ewe|Rut) = 2.47, 95% PCI = [1.40, 4.30]) but a negative effect in the Central and Mojave ecoregions (Central: exp(β _Ewe|Rut) = 0.67, 95% PCI = [0.54, 0.80], Mojave: exp(β _Ewe|Rut) = 0.29, 95% PCI = [0.21, 0.38]).

5. Discussion

Foray decisions among the bighorn sheep analysed here responded to both an intrinsic motivator, age, and a social resource, breeding opportunity. Rams were more likely to foray when breeding opportunities were present during the rut, and as the window of breeding opportunity shortened, the probability of being in the foray state during the breeding season increased (figure 2). During the breeding season, older rams were more likely to foray in all ecoregions. We did not see a similar pattern during the non-breeding season, suggesting that the drive to encounter breeding opportunities is modulated by an individual’s ability to successfully breed (figure 3b ). Rams showed altered responses to social covariates during the breeding season, suggesting that they optimize their movement behaviours in response to the availability of breeding opportunities, measured through the ewe : ram ratio and the total numbers of ewes in the population, but different patterns emerged between ecoregions (figure 3a ; electronic supplementary material, figure S6).

Our finding that rams were more likely to foray during the breeding season is consistent with previous studies from across the bighorn sheep range [12,13,16]. Evidence from the HMM was consistent with the hypothesized negative relationship between breeding season length and foray probability. The inverse relationship between foray probability and duration of rut is similar in form to expectations from a functional response, where a response varies according to the resource availability [54]. Functional responses are widely studied in the context of biophysical attributes in habitat selection studies, and the role of biophysical resource availability on movement is an area of active empirical work (e.g. [55]). We look forward to its greater integration into studies of the social environment. We found a consistent effect of age with older rams more likely to foray during the breeding season. This result supports previous findings from herds in the southern reaches of the bighorn sheep range [16]. However, fine-scale studies of intensively monitored populations in more northerly settings, where ram presence was recorded daily for the duration of the breeding season, suggested young- to middle-ranking [56] or middle-aged [10] rams were most likely to leave the herd home range on breeding migrations. This discrepancy may be owing to our inclusion of all long-distance exploratory movements in the foray state, agnostic to the ram’s location on the landscape, or it could reflect a fundamentally different spatial scale of inquiry (our study covers an area roughly 3500 times the size of the intensively monitored populations, and therefore, the foray events we identify may reflect fundamentally different biology than those documented at finer scales). For example, owing to the effect of dominance hierarchy on mating opportunities (e.g. [29]), middle-aged rams may engage in more forays that depart from the herd home range to increase the probability of encountering oestrus ewes that are untended by higher ranking rams. By contrast, older, higher ranking rams may not need to foray outside the herd home range, since their rank ensures that they are more likely to obtain a tending role in any group they encounter and instead make long-distance movements within the herd home range in search of oestrus females. Alternatively, these results may point towards behavioural differences between rams in northern and southern latitudes.

Two general patterns emerged in how rams responded to the social environment. In the Northern and Central ecoregions, the number of ewes in the population affected foray probability consistently throughout the year. While we found opposing effects, with Northern ecoregion rams foraying more while Central ecoregion rams foray less as the number of ewes increases, the temporal consistency suggests that number of ewes may be a proxy for an unexplored static social resource that warrants further investigation. During the breeding season, rams forayed more as the ewe : ram ratio increased in both ecoregions. In the Mojave ecoregion, we found the opposite pattern. During the non-breeding season, the number of ewes in the population had no effect and sex ratio had a small effect on foray probability. During the breeding season, we found no effect of sex ratio and a negative effect of number of ewes on foray probability.

These responses to ewe : ram ratio on foray probability are inconsistent with previous studies in northern populations which found rams were less likely to go on breeding migrations as herds became more female biased [10,56]. These discrepancies, in conjunction with the reduced overall probability of being in the foray state during the breeding season in the Mojave ecoregion, relative to Northern and Central ecoregions, may signal a change in how rams perceive and respond to the availability of breeding opportunities as the duration of the breeding season increases in the southern portion of the bighorn range.

(a). Implications for bighorn sheep connectivity

Because increased probability of forays probably leads to increased connectivity, our results suggest that connectivity may be dynamic through time, with peak connectivity occurring during the breeding season. Analysing how foray probabilities shifted throughout the year allowed us to detect a switch between older rams foraying more during the rut and younger rams foraying more throughout the rest of the year. These different timings have different implications of fitness: when older rams foray, they are more likely to add to gene flow (because they are the most likely to breed successfully). On the other hand, younger rams are more likely to foray when gene flow is not possible, creating opportunities for disease transmission. In addition, the overall lower probability of foraying in the Mojave ecosystem suggests that they might be buffered from rapid disease spread.

(b). Study limitations

The set of potential drivers of movement within this study was deliberately constrained to focus narrowly on drivers within the social environment. However, myriad other factors could be relevant, including the full ensemble of biophysical drivers, as well as anthropogenic factors like hunting pressure. While a complete exploration of those forces is beyond our current scope, we would welcome a more comprehensive analysis that incorporates biophysical, social and anthropogenic drivers in the future. Moreover, fine-scale patterning of sex ratios and group sizes could have substantial bearing on foray activities, particularly within bighorn sheep populations. However, we believe those fine-scale dynamics are best studied within the context of the rich ethology and evolutionary ecology study systems that are well established for this species (e.g. [10,29]).

Our results are also limited to the data commonly collected by state management agencies—namely, GPS collars deployed on a subset of the population and population-wide estimates of herd attributes. Thus, it is possible that the rams studied here disproportionately reflect particular behavioural or life-history traits. In particular, our dataset included only 13 animals over the age of 8 years. This could be owing to a bias in which rams were collared, or it might reflect actual reductions in life expectancy in the arid environments we studied. If the age structure of our sampled rams is not representative of the full structure within the population, our results could be biased (probably towards an overestimation of overall foray occurrence outside rut and an underestimation of overall occurrence within rut). This could ultimately lead to skewed assumptions about a population-wide behaviour.

In addition, our social resource metrics were collected at a coarse scale (once per year at the herd level), with some information from some herd-years missing (electronic supplementary material, table S2), and the metrics are subject to a measurement error. While the approach we took, in which we directly estimate missing covariate values as dimensions of the joint posterior, is thought to be very robust [57], this undoubtedly adds some uncertainty to our analyses. Our results are not cross-validated, in an effort to channel all of the limited foray data available towards inference. Out-of-sample validation will be important, and while this is something we are working towards with data from other jurisdictions, the predictive performance of these results remains to be determined.

Finally, data from some of our rams were subject to significant temporal autocorrelation (electronic supplementary material, figure S7). Examination of autocorrelation function plots suggests a more complex correlation structure than could be accounted for within our modelling framework. Not accounting for this structure could have constricted the PCIs associated with some of our estimated effects.

6. Conclusion

Social drivers of movement are notoriously difficult to measure, particularly at the large spatial and temporal scales important for connectivity of a large, long-lived animal like bighorn sheep. This is in large part because the configuration of the social environment depends on the locations of a set of constantly moving individuals and may be idiosyncratic to a specific animal (i.e. a ‘good’ social environment for an animal of high rank might simultaneously be a ‘bad’ social environment for a lower ranking individual) [58]. Measurements like daily density surveys are time-consuming and often cost-prohibitive to obtain. Therefore, an outstanding challenge is to find ways of assessing the role of social factors in informing animal movements in systems beyond the model systems of behavioural ecology using data routinely collected by management agencies. Our results show that GPS collar-based studies have the potential to address these questions but will require collar deployment schedules designed to accurately describe population-level effects.

Ethics

All animal handling was conducted in accordance with approved Nevada Department of Wildlife bighorn sheep capture protocols.

Data accessibility

Code is available on Github [59]. Data are available via the Dryad Digital Repository [60].

Supplementary material is available online [61].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

L.E.R.: conceptualization, data curation, formal analysis, methodology, visualization and writing—original draft; M.C.: conceptualization, formal analysis, resources and writing—review and editing; K.R.M.: conceptualization, funding acquisition, investigation, project administration, resources, supervision and writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by grants provided by the Nevada Department of Wildlife through a USFWS Wildlife Restoration Grant, the Utah State University Ecology Center and the Wild Sheep Foundation. Additional support was provided by the Utah Agricultural Experiment Station Project no. 1427.