Drivers of spatial dynamics of a large ungulate (American Bison, Bos bison): forage productivity or site fidelity? Authors: Kate A Asmus, Carter G Kruse, Dustin H Ranglack, John F Benson Published: 2026 ABSTRACT AbstractOptimal foraging theory predicts that animals should reduce energy expenditure and maximize fitness by establishing home ranges in areas with abundant, high-quality forage. However, ungulates may not select optimal foraging habitat when behavior is driven by site fidelity. On Vermejo Park Ranch, New Mexico, managers translocated American Bison (hereafter, bison) to high elevations to reduce historically heavy grazing in lower elevations and to distribute the ecological effects of grazing across the broader landscape. We hypothesized that: (i) home range size is driven by forage abundance and quality; but that (ii) site fidelity can result in suboptimal selection of home ranges in areas with lower forage abundance and quality. We estimated space use and resource selection of free-roaming GPS-collared bison occupying lower elevation sites of historical use and bison translocated to higher elevations to evaluate drivers of home range size and selection. Home range size was strongly negatively influenced by forage abundance. However, without translocation, bison exhibited fidelity to areas of historical use and selected seemingly suboptimal home ranges with lower forage biomass. Conversely, bison that established home ranges following translocation to higher elevations selected home ranges with greater forage biomass. Greater forage biomass enabled bison to use smaller home ranges, increasing spatial efficiency and cohesion. Bison on Vermejo that were not translocated by managers established home ranges at lower elevations despite the availability of higher quantity and quality (nitrogen yield) of forage at higher elevations. Thus, home range selection appeared to be guided by site fidelity rather than foraging optimally across the larger landscape. For managed ungulates, translocation may facilitate access to superior resources and eventually begin to restore knowledge of migratory routes and seasonally abundant forage. Optimal foraging theory predicts that animals should make movement and space use decisions that allow them to achieve energetic efficiency and maximize fitness (MacArthur and Pianka 1966; Pyke et al. 1977). One prediction of optimal foraging suggests that animals should acquire resources needed for survival and reproduction within the smallest area possible (Mitchell and Powell 2004). Indeed, by using smaller home ranges in areas of abundant, high-quality forage, herbivores may optimize energy intake and reduce energetic costs of travel (Fryxell 1991; Fryxell et al. 2008; Ranglack and du Toit 2015). Increased forage availability can also relax intragroup competition, allowing herbivores to increase group cohesion and size (Ranglack and du Toit 2015; Pinacho-Guendulain and Ramos-Fernández 2017). Conversely, in areas of lower forage availability, animals may need to expand their home ranges and further reduce intraspecific competition to meet foraging requirements (Tufto et al. 1996; Kreling et al. 2021). Thus, the size and selection of a home range may have important implications for spatial and foraging efficiency, which can strongly influence survival and reproduction (McLoughlin et al. 2006, 2007). Animals may not always select resources optimally when behavior is driven by site fidelity and when they have limited knowledge of resource conditions across the broader landscape (Merkle et al. 2015a; Kreling et al. 2021). Site fidelity, or the tendency of animals to return to previously visited locations can increase fitness by allowing animals to efficiently acquire resources, reduce movement costs, and lower predation risk (Stamps 1995; Gehr et al. 2020; Morrison et al. 2021). However, site fidelity is likely guided by spatial memory of past experiences (Merkle et al. 2019; Cameron et al. 2020; Rheault et al. 2021), which may cause animals to favor previously visited, lower quality sites over the use of higher quality sites, resulting in suboptimal habitat selection (Tracz et al. 2010; Kreling et al. 2021; Merkle et al. 2022). Evidence suggests that both site fidelity and migration are learned behaviors, passed to offspring and conspecifics through social learning and the transmission of knowledge from generation to generation (Sweanor and Sandegren 1988; Nelson and Mech 1999; Jesmer et al. 2018). If social information and spatial memory are lost from a population, as is often the case following restoration efforts, it may take decades for the behavior to reemerge, compromising the ability of individuals to respond adaptively to spatial variation in resources (Jesmer et al. 2018; Kreling et al. 2021). Johnson (1980) proposed a hierarchical ordering of resource selection decisions made by animals from coarser to finer spatial scales that reflect changes in both use and availability. The most important factors limiting fitness should drive selection at coarser spatial scales (e.g., home range selection; Rettie and Messier 2000). With the second order of resource selection, an animal selects its home range from the larger landscape, which then constrains selection at finer spatial scales within the home range (Johnson 1980). Evaluating the home range of an animal provides ecologists with valuable information about how important resources influence space use decisions and resource selection across biologically relevant time periods (Powell and Mitchell 2012). Despite its importance, estimating second-order resource selection has proven challenging for ecologists because it is difficult to accurately perceive what is truly available to an animal across the larger landscape (Hutto 1990; Garshelis 2000; McDonald et al. 2005). Consequently, ecologists often use management units (e.g., park boundaries) or spatial extents defined by sampling considerations (e.g., area used by all study animals) to represent the larger landscape (Boyce et al. 2003; DeCesare et al. 2012). Indeed, many studies estimate a “population-level” range using locations from all study animals to define the larger landscape (e.g., DeCesare et al. 2012), but there are inevitably animals in the population that were not monitored who moved beyond this area. Thus, defining the larger landscape arbitrarily or from sampling home ranges may under- or over-represent the extent of habitat available to animals choosing home ranges (Garshelis 2000; McDonald et al. 2005). However, situations in which landscape-level movement of animals is constrained provide opportunities for ecologists to better understand the processes by which animals select home ranges from clearly defined larger landscapes. These situations could include islands (Thalmann et al. 2016), urban parks surrounded by anthropogenic barriers, or reserves and private land with fenced boundaries (Kohl et al. 2013; Brockman et al. 2022; Sawatzky et al. 2026). Large herbivores are often confronted with a scale-dependent tradeoff between forage quantity and quality, since these forage characteristics are commonly inversely related (Fryxell 1991). Given that higher quality forage is often scarce and widely dispersed (Fryxell 1991), at coarse spatial scales, large herbivores may select home ranges with a higher quantity of forage to ensure sufficient forage abundance needed to meet high intake requirements and to avoid depleting resources (van Beest et al. 2010; Frank et al. 2016; Kaszta et al. 2016; Dupke et al. 2017; Gutierrez et al. 2026). Conversely, at finer-spatial scales within the home range, herbivores should select patches of high-quality forage to maximize nutrient intake (Senft et al. 1987; Bailey et al. 1996; van Beest et al. 2010; Kaszta et al. 2016). Yet, studies have shown that large herbivores may select both higher forage quantity and quality at large spatial scales (Kaszta et al. 2016; Gutierrez et al. 2026), as these forage characteristics are not always inversely related. Indeed, in some systems, interactions between grazing, adequate precipitation, and moderate temperatures may result in areas of both high forage biomass and high plant nutritive value (McNaughton 1985; Marshal et al. 2005; Schoenecker et al. 2022; Mainetti et al. 2023; Musto et al. 2025). Over 97% of American Bison (Bos bison; hereafter, bison) are managed as captive herds (Ranglack et al. 2023), providing opportunities to evaluate second-order resource selection within landscapes defined by fenced boundaries. Bison usually make movement and space use decisions based on forage conditions (Merkle et al. 2014, 2015a), with forage productivity often driving selection at coarser spatial scales and forage quality often driving selection at finer spatial scales (Senft et al. 1987; Bailey et al. 1996; van Beest et al. 2010). Thus, home range selection by bison should be driven primarily by forage availability, especially given that the Gray Wolf (Canis lupus), the primary predator of bison aside from humans, was extirpated from most of its historical range (Li 2000). Most captive bison herds reside on landscapes <50 km2 (Freese et al. 2007; Gates et al. 2010; Rogers 2021), severely limiting their space use patterns (Ritson 2019). Summer home ranges for free-ranging bison in several large, unfenced landscapes in North America ranged from 137 to 1270 km2 (Merkle et al. 2015a; Ritson 2019), although natural landscape barriers may have prevented bison from ranging further. Vermejo Park Ranch (hereafter, Vermejo) in northeastern New Mexico encompasses 2,270 km2, providing adequate and appropriate space to allow bison to maintain natural movement and space use decisions (Sanderson et al. 2008). Historically, previous ranch owners restricted bison to lower elevations on Vermejo following their restoration to the landscape in the 1920s. Beginning in the late 1990s, ranch managers allowed bison to range freely on Vermejo, but bison continued to limit their space use to lower elevations despite potentially greater forage abundance and quality at higher elevations associated with greater precipitation, cooler temperatures, and more mesic habitats (Legler 2010). In 2016, managers started translocating bison to underused, higher elevation areas on Vermejo to reduce grazing pressure at lower elevations and to disperse the ecological impact of bison across the landscape (e.g., Sanderson et al. 2008; Kohl et al. 2013; Geremia et al. 2019; Musto et al. 2025). Bison may maintain fidelity to sites at lower elevations to avoid energetic costs and uncertainty of migration. Alternatively, if greater forage abundance and quality exist at higher elevations, site fidelity could represent suboptimal resource selection driven by familiarity and lack of spatial knowledge of available resources. We tracked captive, but largely free-roaming bison on Vermejo with global positioning system (GPS) telemetry in lower elevation sites of historical use (late 1990s to 2023), as well as higher elevation sites following translocations to test 2 hypotheses. First, in accordance with optimal foraging theory, we hypothesized that space use by herbivores is primarily driven by forage abundance and quality, such that they should use the smallest possible areas within which they can acquire sufficient resources to maximize energy efficiency (H1; MacArthur and Pianka 1966; Mitchell and Powell 2007). Thus, we predicted that bison home range size decreases with increasing forage abundance (P1). Given the importance of forage quality on ungulate movement decisions, we also predicted that bison home ranges would be smaller in areas with higher forage quality (P2). We predicted that bison increase pairwise spatial cohesion (i.e., distance between collared individuals) and group size when using smaller home ranges in areas with greater forage abundance and quality due to relaxed intragroup competition (P3). Second, we hypothesized that site fidelity can lead to suboptimal home range selection (H2). Here, we predicted that bison in areas of historical use select home ranges with lower forage biomass, whereas bison translocated to higher elevations select home ranges with higher forage biomass, relative to what is available across Vermejo (P4). Our study examines the interplay between optimal space use and site fidelity for herbivores by evaluating space use with and without management intervention. Methods Study area We studied bison on Vermejo Park Ranch, a 2,270 km2 landholding of Turner Enterprises, Inc., located in northeastern New Mexico (Fig. 1). Between the 1920s and 1980s, bison were translocated to Vermejo from 3 locations in the United States: (i) Yellowstone National Park, Wyoming; (ii) Philmont Scout Ranch, New Mexico; and (iii) Maxwell Wildlife Refuge, Kansas. During our study from 2022 to 2023, the population remained stable at approximately 1,200 bison. Northwestern Vermejo is defined by the Sangre de Cristo Mountains. The Park Plateau, comprising the central and eastern portions, encompasses the largest part of Vermejo and is primarily distinguished by the broad grasslands of Castle Rock and Van Bremmer Park as well as woodlands, narrow valleys, and escarpments (Fig. 1). Elevation rises abruptly from the western edge of the Park Plateau to a steep rock face, known as the Wall (Fig. 1). In southwestern Vermejo, the Park Plateau changes abruptly into low relief and shortgrass prairies (i.e., winter range; Fig. 1). Elevation on Vermejo ranges from approximately 1,800–4,000 m. Annual precipitation on Vermejo varies considerably based on elevation, with higher elevations receiving 635–889 mm and lower elevations receiving 305–381 mm (Legler 2010). While Vermejo was largely undeveloped during the study, natural gas extraction occurred throughout the ranch resulting in an extensive dirt road network. Elk (Cervus canadensis), Mule Deer (Odocoileus hemionus), Pronghorn (Antilocapara americana), and feral horses (Equus caballus) occurred throughout most of the study area. Elk were the most numerous ungulate species, reaching an estimated 7,200–7,500 individuals in both 2022 and 2023 (Lance Bernal, Vermejo Park Ranch, personal communication). Mountain Lion (Puma concolor) and American Black Bear (Ursus americanus) were the only other large mammals on Vermejo. Vermejo Park Ranch in northeastern New Mexico, USA, where we studied home ranging behavior by bison in April–September 2022–2023. Shown are vegetation sampling sites throughout Vermejo color coded by area, major rivers and creeks, 2 prominent topographic features, the location of interior fencing, and the range of elevation throughout Vermejo. Field methods and bison management In February 2022 and 2023, Vermejo personnel worked the bison through corrals and into a hydraulic squeeze chute to perform routine health checks. While bison were physically restrained in the squeeze chute, we deployed GPS collars (Vectronic Aerospace, Iridium Vertex Plus model, Berlin, Germany) on 35 pregnant adult (>4 years old) female bison, programmed to collect 1 GPS location every hour. Adult female and offspring survival most strongly influence population growth of ungulates (e.g., Gaillard et al. 1998; Raithel et al. 2007). Thus, evaluating adult female space use and movements relative to valuable resources is important for informing management of ungulate populations. In an attempt to distribute the collars across rather than within subgroups within the bison herd, we deployed collars on every 25th female bison to mitigate the potential lack of independence of individual movement behavior due to sociality in our analyses. During 2023, if a collared bison entered the squeeze chute and was not pregnant, we removed the collar and placed it on the next pregnant bison since females that were not pregnant were removed from the herd. Therefore, we had 17 bison with 2 years of data and 18 bison with 1 year of data. We deployed collars on bison in accordance with a protocol approved by the University of Nebraska-Lincoln Institutional Animal Care and Use Committee (project ID: 2382). From December through April, Vermejo managers rotated the herd throughout fenced pastures on the 243 km2 winter range. We did not include movements during this period (December–April) in our analyses. From May through November, managers allowed the herd to freely roam the remaining 2,027 km2 of Vermejo. Within the lower elevations, managers occasionally used low-stress stockmanship to encourage bison away from heavily grazed areas. This management was infrequent (<10 times in 2 years), typically brief in duration (<2 h), and resulted in relatively modest movement by bison (typically <5 km). Thus, although managers occasionally influenced bison movement, we do not believe this strongly influenced our results. Fencing on the boundary of Vermejo kept bison within the perimeter. Aside from the fenced pastures on the winter range, managers used fence in only 2 locations in the interior of Vermejo. The first location was at the bottom of Van Bremmer Canyon, which separated the winter range from the rest of Vermejo (i.e., Bremmer fence; Fig. 1). The second location was in a large open meadow approximately 3,000 m elevation northeast of Costilla (i.e., drift fence; Fig. 1). This two-strand electrical wire drift fence spanned approximately 400 m across the meadow and stopped abruptly at the tree line, acting as a deterrent but not a barrier as bison could easily walk around the fence. Though bison were allowed to range freely throughout Vermejo, they often concentrated their use within lower elevations in Castle Rock and Van Bremmer (Fig. 1). To reduce grazing pressure in the lower elevations and to disperse the ecological impacts of bison use more evenly throughout Vermejo, ranch managers translocated bison in semitruck trailers to 2 underused locations in 2022 and 2023. In 2022, managers translocated 900 bison (14 collared) to the Canadian River before calving began in mid-April and 300 bison (7 collared) to Costilla following calving in mid-May (Fig. 1). Upon arrival to the Canadian River, managers transferred some of the bison (8 collared) to a soft-release paddock until they were accustomed to the area (approximately 2 weeks). Once habituated, managers released these bison at the end of April to roam freely throughout Vermejo. Collared bison that were soft-released in the Canadian River returned to lower elevations within 5–25 days. A second group of bison translocated to the Canadian River was immediately released (hard-released; 6 collared) from the semitruck trailers in mid-April. All hard-released collared bison left the Canadian River and returned to lower elevations within 3–5 days. Bison translocated to Costilla were kept in a soft-release paddock and then released at the end of May. Bison in Costilla established home ranges and remained in the area until the end of August and then returned to lower elevations. In 2023, managers translocated 100 bison (0 collared) to the Canadian River before calving in April and 300 bison (11 collared; 4 that were also translocated to Costilla in 2022) to Costilla after calving in mid-June. Bison in both areas were held in soft-release paddocks for approximately 2 weeks and then released to roam freely. In 2023, uncollared bison in the Canadian River returned immediately to lower elevations after being soft-released, while bison in Costilla established home ranges and stayed until mid-August and then returned to lower elevations. In 2023, managers moved the remaining 800 bison (21 collared) from the winter range to the bottom of Cerrososo Canyon in lower elevations using all-terrain vehicles (Fig. 1). These bison moved from Cerrososo Canyon into areas historically (late 1990s to 2023) used by bison during spring and summer either up the canyon to Castle Rock or over the ridgeline into Van Bremmer Canyon. In subsequent sections, we refer to areas where bison established home ranges in the lower elevations (relative to higher elevation in Costilla) as historical areas, as these are areas they have continuously used in the past (from the late 1990s to 2023). We refer to areas where bison were translocated to as Costilla (established home ranges) and the Canadian River (abandoned following translocation). Estimating forage availability via remote sensing We estimated forage availability in historical areas, Costilla, and the Canadian River using remote sensing to evaluate the influence of forage quantity on space use, home range selection, and translocation success. We used 30 m resolution herbaceous aboveground biomass estimates from the Rangeland Analysis Platform (RAP; kg/km2; 16-day temporal resolution) as an index of forage production (Robinson et al. 2019; Jones et al. 2021). RAP uses land surface data (continuous vegetation cover), satellite imagery (absorbed solar radiation), and meteorological data (temperature and vapor-pressure deficit) to estimate production accumulated every 16 days according to plant functional type at a spatial resolution of 30 m (Robinson et al. 2019; Jones et al. 2021). Using Google Earth Engine, we extracted all 16-day biomass images from between the start and end dates of the monitoring period for home ranges (home range estimation details below) and averaged these across the calving (1 April–30 June) and breeding (1 July–30 September) seasons within historical areas and Costilla. We merged the home ranges to form 1 polygon for both areas, extracted the mean biomass values at each pixel, and averaged these values across the composite home ranges to obtain estimates of forage biomass within both areas. Bison did not establish home ranges in the Canadian River, so we estimated a minimum convex polygon, with all GPS data collected from bison during the periods they remained in the area (3–25 days) following translocation within which we estimated forage biomass in the Canadian River using the RAP data as described above. Estimating forage productivity and quality via vegetation sampling We also used vegetation sampling to estimate forage productivity and quality at multiple sites within each of the 3 areas of interest (historical areas, Costilla, and the Canadian River). Given the labor-intensive nature of vegetation sampling, it was not possible to estimate forage productivity and quality at a similar resolution (30 m pixels) as the RAP data. Nonetheless, the estimates from our vegetation sampling allowed us to: (i) ensure the trends in RAP forage biomass estimates were consistent with data sampled directly from our study area; and (ii) estimate forage quality within the areas occupied by bison. Although we could not include the forage quality estimates directly in our models of home range size and resource selection, they did allow us to test our prediction that bison would use smaller home ranges in areas with higher forage quality (P2) by comparing home range size between areas of Vermejo with higher and lower forage quality. We randomly selected 9 sites in historical areas, and 6 sites each in Costilla and the Canadian River following methodology of Schoenecker et al. (2022) and Musto et al. (2025; Fig. 1). At each site, we randomly placed 5 approximately 1 m2 movable grazing cages which created a protected subplot (inside cage) and an unprotected subplot (directly outside cage). These small cages protected 1 m2 areas from grazing for short periods, allowing us to account for ungulate consumption in our estimates of annual aboveground net primary productivity (ANPP). We sampled herbaceous biomass 3 times in 2023 (end of June, end of August, and end of September) and 2 times in 2024 (end of June and end of August). To estimate ANPP in the 3 areas, we harvested all herbaceous biomass within 0.25 m2 quadrat rings inside and outside of grazing cages, separated and discarded standing dead material (previous year’s growth) from live material (current year’s growth), sorted plants by functional group (graminoid, forb, and shrub), and oven dried the vegetation before weighing the samples (McNaughton 1985; Bonham 1989; McNaughton et al. 1996). We used the sum of positive and negative increments to estimate forage productivity in the presence of grazers (Bonham 1989; additional details in Supplementary Data SD1). Within each site, we took a portion of plant material from each of the 5 samples outside of grazing cages to form a single, composite sample of graminoids and forbs each per site (i.e., 1 sample per site per year). We created these composite samples from the June sampling period in both years equating to n = 42 graminoid samples and n = 42 forb samples total. We sent the samples to a laboratory (Ward Laboratories, Inc., Kearney, Nebraska, USA) to obtain an estimate of nitrogen concentration (%N). To estimate nitrogen yield (measured in g/m2) for each sample, we multiplied %N by ANPP for each site and functional group (Schoenecker et al. 2022; Musto et al. 2025). Thus, we generated estimates of forage productivity (ANPP) and quality (nitrogen yield) for historical areas and translocation areas (Costilla, Canadian River) from vegetation sampling. We used separate linear mixed-effects models with response variables of productivity and N yield of graminoids and forbs and included independent variables for area and year. Within each model, we included a random intercept for site to account for the lack of independence between data collected from the same site in different years. Here, we investigated potential differences in the response variables between historical areas and both translocation sites by including dummy-coded variables for Costilla and the Canadian River (with historical areas withheld as the reference group). For each response variable, we considered separate models with: (i) the main effects of area and year with no interactions; and (ii) 2-way interactions between area and year. Space use analyses We estimated home ranges for bison on Vermejo with 99% adaptive local convex hulls (a-LoCoH; Getz et al. 2007) using the “sf” package (Pebesma 2018) in R version 4.3.0. We estimated the a parameter as the distance between the farthest 2 locations in each input dataset (Getz et al. 2007). We estimated separate home ranges during the calving (1 April–30 June) and breeding (1 July–30 September) seasons with GPS telemetry data collected for all collared bison that remained in either of the 2 areas (i.e., historical or Costilla) for ≥30 days. These were the 2 prominent periods when bison were left to roam freely throughout Vermejo and when biological demands (i.e., calf rearing and breeding) may influence movement and space use decisions (Meagher 1973). For bison that established home ranges following translocation, we estimated home ranges starting the day after they were released from the soft release. Exploratory analyses suggested that home range size was related to number of monitoring days (Supplementary Data SD2 and SD3). Therefore, we re-estimated home ranges using a standardized number of monitoring days (n = 34 days) within each season to ensure differences in home range size were attributed to biological variables of interest rather than number of monitoring days. We estimated calving home ranges from 28 May to 30 June in both years and both areas. We estimated breeding home ranges for bison in Costilla from 1 July to 3 August in both years. We estimated breeding home ranges for bison in historical areas from 27 August to 30 September in both years, which includes the home range estimates for bison after returning to lower elevations from Costilla. We used linear mixed-effects models (“lmer” function in “lme4” R package; Bates et al. 2015) to evaluate whether home range size varied relative to forage biomass, season, and area. Within each model, we included random intercepts for an individual to account for the lack of independence of multiple home ranges for the same individuals (n = 35 bison) across seasons and years (1–4 seasonal home ranges per bison; Bolker et al. 2009). We estimated forage biomass within seasonal home ranges as described above using the 16-day RAP biomass estimates (Robinson et al. 2019; Jones et al. 2021). Forage biomass varied across seasons and areas (Fig. 2a). Thus, we considered univariate models with predictors of forage, season, and area to avoid correlation and redundancy in predictor variables included in the same models. While we were interested in evaluating the influence of other natural and anthropogenic variables on home range size, forage biomass was correlated with other landscape variables (|r > 0.5|). We ran an overall analysis that included data from both seasons and both areas to evaluate drivers of space use for bison across Vermejo and to evaluate whether there were differences in home range size between historical areas and Costilla. We also ran analyses to evaluate factors influencing home range size for bison in historical areas and Costilla separately. We compared the fit of the null model and each univariate model for the different analyses (i.e., overall, historical, and Costilla) using the corrected Akaike’s Information Criteria (AICC; Burnham and Anderson 2002). We calculated the difference (Δ) in AICC values (lower values indicate better fit) and concluded that there was strong empirical support for a model if ΔAICC < 10 (following model selection guidance for mixed effects models; Bolker et al. 2009). Estimates in historical areas, Costilla, and the Canadian River on Vermejo Park Ranch, New Mexico, USA of (a) forage biomass (kg/km2; mean with 95% confidence intervals) estimated using the Rangeland Analysis Platform 16-day biomass, (b) forage productivity (g/m2; mean with 95% confidence intervals) predicted from a linear-mixed effects model with independent variables for area and year, and (c) herbaceous plant nitrogen (N) yield (g/m2; mean with 95% confidence intervals) of graminoids and forbs predicted from linear mixed-effects models with independent variables of area and year. Estimating spatial cohesion, group size, and density We estimated spatial cohesion between pairs of collared bison as the proportion of time pairs of collared bison within historical areas and Costilla spent within different (Euclidean) distances of each other using telemetry data collected simultaneously (within 5 min) for both individuals in the pair (Benson and Patterson 2015). Thus, our use of the term cohesion reflects spatial proximity between pairs of collared bison within each area, consistent with previous definitions in the animal social ecology literature (Kappeler and van Schaik 2002; Aureli et al. 2008; Benson and Patterson 2015; Benson et al. 2025). Importantly, our definition of spatial cohesion does not imply understanding the precise nature of interactions between individuals, simply that the individuals were sharing time and space (Benson et al. 2025). We present the mean proportion of simultaneous locations in which bison within historical areas and Costilla were separated by <50 m, <100 m, <250 m, and <500 m during the calving and breeding seasons. We acknowledge that these estimates provide information about pairwise cohesion within bison social groups, rather than estimates of cohesion of entire groups. In 2022, we conducted visual observational surveys to estimate group size in historical areas and Costilla. We located all collared bison at least once a week using GPS telemetry and very high frequency transmitters during May–July and conducted surveys between dawn and dusk. Once we located a collared bison, we recorded all collared individuals in the group, total group size, and the number of males, females, and calves in the group. To estimate bison density, we first estimated separate minimum convex polygons (MCPs) within historical areas and Costilla during both the calving and breeding season using GPS data from all collared individuals in R using the “st_convex_hull” function in the “sf” package (Pebesma 2018). These MCPs provided estimates of the areas occupied by bison in historical areas and Costilla during each season. Next, we divided the annual bison abundance estimates (acquired from ranch managers) for each area in each season by the year-, area-, and season-specific MCPs to estimate bison density (bison/km2) for both seasons in 2022 and 2023 within historical areas and Costilla. Resource selection models We investigated the second order of resource selection by evaluating selection of home ranges by bison from the larger landscape across Vermejo defined by the fenced boundaries (Johnson 1980). Specifically, we compared resource conditions at locations distributed across bison home ranges to those distributed across Vermejo. We used home ranges estimated with all available data within historical areas and Costilla during the calving (1 April–30 June) and breeding (1 July–30 September) seasons. We estimated resource conditions at the center of every fourth 30 m2 pixel within the home range of each bison (use) and across Vermejo (availability; Benson 2013). We estimated Euclidean distances from the centroid of all systematically distributed used and available pixels to the center of the closest pixels for natural land cover types (Supplementary Data SD4; Conner et al. 2003; Benson 2013; Dewitz 2023). We modified the National Landcover Database land cover layer by combining similar habitat types to create a layer with 3 broad habitat classes: grasslands, forests, and wetlands (Supplementary Data SD4; Dewitz 2023). We estimated elevation and slope using digital elevation models from the 3D Elevation Program and classified all used and available pixels accordingly (Supplementary Data SD4; U.S. Geological Survey 2021). We estimated forage biomass within seasonal home ranges as described above using the 16-day RAP biomass estimates (Supplementary Data SD4; Robinson et al. 2019; Jones et al. 2021). We paired each home range with the corresponding seasonal forage estimates and classified all used and available pixels. We included all variables in our models as correlation between predictor variables was modest or low (r < 0.41). For all continuous variables, we rescaled data by subtracting the mean from each value and dividing by 2 SDs to produce more interpretable regression coefficients (Gelman 2008). We modeled resource selection in a use-availability framework with generalized linear mixed models (GLMMs) using the “glmmTMB” R package with a binary (0 = available, 1 = used) response variable (family = binomial, link = logistic; Brooks et al. 2017). We constructed a separate set of models within the calving and breeding seasons using all home ranges from both areas and considered models with: (i) all resource variables and no interactions; (ii) interactions between specific resource variables (forage, elevation) and a dummy-coded variable for area (historical = 0, Costilla = 1) plus slope, and distance to grasslands, forests, and wetlands as main effects; and (iii) the null model (Supplementary Data SD5). Within each model, we included random intercepts for individual and year, with year nested in individual, to account for unbalanced samples sizes of locations between individuals, the lack of independence of multiple home ranges for the same individuals across years, and to appropriately pair the year-specific used and available data within our models (Gillies et al. 2006). We also included random slope terms for all resource variables as recommended by Muff et al. (2020) to estimate appropriately conservative estimates of uncertainty that accounted for variation in resource selection among individual bison. We also included second-order resource selection results from models without random slopes (Supplementary Data SD6 and SD7) to illustrate that our inference about bison resource selection remained unchanged between models with and without random slopes. Lack of independence of data from multiple individuals within social groups is also a potential concern for animal resource selection studies (Hebblewhite and Merrill 2008). Bison social structure is characterized by fission–fusion dynamics where individuals frequently leave and join groups, making it difficult to assign bison to specific groups with certainty (Lott and Minta 1983; Fortin et al. 2009; Merkle et al. 2015b; Bresnan et al. 2025) and account for sociality in resource selection models. Nonetheless, as noted above, we attempted to mitigate departures from independence by deploying collars on every 25th bison to minimize placement of collars within family groups. We also quantified the distance between collared individuals on an hourly basis and found that pairs of collared individuals in both areas of our study spent the majority of time > 100 m apart (Merkle et al. 2015b; Table 2). Thus, we believe the potential lack of independence between social bison in our models was not a serious problem and did not strongly influence our results. Within each season, we compared the fit of the model with no interactions, the model with an area interaction, and the null model using AICC (Burnham and Anderson 2002). We calculated the differences (Δ) in AICC values (lower values indicate better fit) and concluded that there was strong empirical support for models if ΔAICC < 10 (Bolker et al. 2009). We considered the model with area × resource interactions to evaluate whether bison used resources differently when establishing home ranges in historical areas and Costilla. If we found differences, we ran separate models for each area within each season to provide clearer interpretation of selection and avoidance relative to availability without the presence of interactions (Benson et al. 2021). We inferred that bison selected or avoided resource variables when the 95% confidence intervals surrounding the fixed beta coefficients did not overlap 0. Below, we use the terms selection and avoidance to indicate that: (i) home range locations were significantly closer to (selection) or farther from (avoidance) distance-based resource variables (grasslands, wetlands, forests) than were locations available throughout Vermejo; or (ii) values of classification-based resource variables (forage biomass, elevation) within home ranges were significantly greater (selection) or lower (avoidance) than locations available throughout Vermejo. Results Forage availability and quality Forage abundance estimates from remotely sensed RAP data were highest in both seasons and years in Costilla (calving 2022 = 7,670.0 kg/km2 ± 22.3; calving 2023 = 7,741.7 kg/km2 ± 25.6; breeding 2022 = 11,152.5 kg/km2 ± 30.7; breeding 2023 = 7,095.0 kg/km2 ± 23.0), where bison established home ranges following translocation, compared with historical areas (calving 2022 = 3,599.0 kg/km2 ± 3.1; calving 2023 = 5,775.7 kg/km2 ± 7.1; breeding 2022 = 8,126.2 kg/km2 ± 7.0; breeding 2023 = 6,212.9 kg/km2 ± 5.7) and the Canadian River (calving 2022 = 3,896.1 kg/km2 ± 5.5; calving 2023 = 4,600.0 kg/km2 ± 7.3; breeding 2022 = 7,333.7 kg/km2 ± 10.9; breeding 2023 = 4,804.0 kg/km2 ± 7.9; Fig. 2a). Forage abundance estimated from remotely sensed RAP data was lowest in the Canadian River, the site abandoned by bison following translocation, during most seasons and years (Fig. 2a). The only exception was during calving in 2022, when forage abundance estimated from remote sensing was slightly higher in the Canadian River than historical areas, but considerably lower than Costilla (Fig. 2a). Productivity estimated from vegetation sampling was significantly higher in Costilla (2023 = 67.3 g/m2 ± 9.3; 2024 = 50.7 g/m2 ± 6.1) than in historical areas (β = 21.20, 95% CI [9.77, 32.62]; 2023 = 43.2 g/m2 ± 5.2; 2024 = 32.4 g/m2 ± 3.7) but did not differ significantly between the Canadian River (2023 = 45.7 g/m2 ± 5.6; 2024 = 44.0 g/m2 ± 3.5) and historical areas (β = 7.11, 95% CI [−4.31, 18.54], n = 42; Fig. 2b). Productivity was significantly higher in 2023 compared to 2024 (β = 9.85, 95% CI [1.60, 18.10]; Fig. 2b). There were no significant interactions between area and year in terms of their influence on productivity. Forage productivity estimates obtained from vegetation sampling were comparable to forage abundance estimates obtained from remote sensing, as both methods indicated that forage quantity was higher in Costilla compared to historical areas (Fig. 2). N yield of graminoids was significantly higher in Costilla (2023 = 0.88 g/m2 ± 0.13; 2024 = 0.63 g/m2 ± 0.07) compared to historical areas (β = 0.30, 95% CI [0.15, 0.46]; 2023 = 0.49 g/m2 ± 0.06; 2024 = 0.41 g/m2 ± 0.04) and did not significantly differ between the Canadian River (2023 = 0.45 g/m2 ± 0.08; 2024 = 0.50 g/m2 ± 0.07) and historical areas (β = 0.02, 95% CI [−0.13, 0.18], n = 42; Fig. 2c). There were no significant differences in N yield of graminoids between 2023 and 2024 across all sites combined (β = 0.10, 95% CI [−0.01, 0.20]; Fig. 2c). There were no significant interactions between area and year in terms of their influence on N yield of graminoids. N yield of forbs did not differ significantly between Costilla (2023 = 0.19 g/m2 ± 0.04; 2024 = 0.23 g/m2 ± 0.06) and historical areas (β = 0.001, 95% CI [−0.13, 0.13]; 2023 = 0.18 g/m2 ± 0.06; 2024 = 0.24 g/m2 ± 0.10) or the Canadian River (2023 = 0.26 g/m2 ± 0.05; 2024 = 0.19 g/m2 ± 0.02) and historical areas (β = 0.02, 95% CI [−0.12, 0.15], n = 42; Fig. 2c). There were no significant differences in N yield of forbs between 2023 and 2024 across all sites combined (β = −0.01, 95% CI [−0.12, 0.09]; Fig. 2c). There were no significant interactions between area and year in terms of their influence on N yield of forbs. Space use Variation in home range size for all collared bison was influenced more by forage biomass than by area or season (Supplementary Data SD8; n = 102 home ranges from 35 bison). Home range size decreased with increasing forage biomass (β=−35.8, 95% CI [−42.6, −29.0]; Fig. 3). There were no differences between calving and breeding season (β=−2.8, 95% CI [−12.8, 7.2]), but home ranges established in Costilla (calving = 10.1 km2, 95% CI [9.8, 10.5]; breeding = 13.7 km2, 95% CI [12.1, 15.4]) were significantly smaller than home ranges established in historical areas (β=−37.8, 95% CI [−46.8, −28.8]; calving = 46.7 km2, 95% CI [37.7, 55.8]; breeding = 53.0 km2, 95% CI [47.3, 58.6]; Table 1). Within historical areas, variation in home range size was also explained best by forage biomass (Supplementary Data SD8; n = 79 home ranges from 35 bison). As forage biomass increased, home range size decreased (β=−28.4, 95% CI [−35.8, −21.0]). Within Costilla, the model with the strongest support was the null model, suggesting that neither forage biomass nor season substantially influenced home range size (Table 1; Supplementary Data SD8; n = 14 home ranges from 7 bison). The negative linear relationship (β=−35.8, 95% CI = −42.6, −29.0) between home range size (km2) and forage biomass (kg/km2; estimated from RAP data) for bison on Vermejo Park Ranch, New Mexico, USA, May–September 2022–2023. This model evaluated home ranges across both seasons (calving and breeding) and both areas. Table 1Mean home range (HR) sizes (km2) for bison on Vermejo Park Ranch, New Mexico, USA, May–September 2022–2023, that established home ranges in historical areas and Costilla during the calving and breeding seasons.AreaSeason n HR (x–)95% CIHR (range)Forage (x–)95% CI Historical Calving3246.737.7–55.816.2–96.16605.16091.9–7118.3 Costilla Calving710.19.8–10.59.8–10.78433.88355.9–8511.6 Historical Breeding4753.047.3–58.619.1–90.67797.57277.5–8317.5 Costilla Breeding1613.712.1–15.49.5–18.810042.49218.6–10866.2We estimated home ranges using a standardized number of monitoring days (34 days). n, number of home ranges; x–, mean; CI, 95% confidence intervals of home ranges; forage x–, mean forage biomass (kg/km2), CI, 95% confidence intervals for forage biomass. Mean home range (HR) sizes (km2) for bison on Vermejo Park Ranch, New Mexico, USA, May–September 2022–2023, that established home ranges in historical areas and Costilla during the calving and breeding seasons. We estimated home ranges using a standardized number of monitoring days (34 days). n, number of home ranges; x–, mean; CI, 95% confidence intervals of home ranges; forage x–, mean forage biomass (kg/km2), CI, 95% confidence intervals for forage biomass. Table 2Mean proportion of simultaneous Global Position System telemetry locations collected from bison within the same area (i.e., historical or Costilla) where bison were separated by <50 m, <100 m, <250 m, and <500 m on Vermejo Park Ranch, New Mexico, USA, 2022–2023.Proximity (m, calving)Proximity (m, breeding)Area<50<100<250<500<50<100<250<500 Historical 0.060.130.260.360.040.080.140.19 Costilla 0.220.390.610.720.180.360.630.78 Mean proportion of simultaneous Global Position System telemetry locations collected from bison within the same area (i.e., historical or Costilla) where bison were separated by <50 m, <100 m, <250 m, and <500 m on Vermejo Park Ranch, New Mexico, USA, 2022–2023. Spatial cohesion, group size, and density Collared bison in Costilla spent a greater proportion of time in close proximity to one another during simultaneous locations than collared bison in historical areas at all estimated distances (Table 2). In fact, the mean proportion of time spent within 50, 100, 250, and 500 m of other collared bison was 2.0–4.5 times greater in Costilla compared to historical areas across both seasons (Table 2). The mean distance between collared bison during simultaneous locations was 525.6 m (SD = 847.2, n = 7, range = 0.0–4,740.5) in Costilla and 5,639.0 m (SD = 9,660.6, n = 24, range = 0.0–42,736.4) in historical areas during the calving season. The mean distance between collared bison during simultaneous locations was 489.1 m (SD = 975.4, n = 13, range = 0.0–9,472.3) in Costilla and 7,621.9 m (SD = 7,416.6, n = 27, range = 0.0–30,019.7) in historical areas during the breeding season. Mean group size from May to July was larger and less variable for bison in Costilla (x¯ = 195 bison, SD = 76, n = 28, range = 36–292) compared to historical areas (x¯ = 154 bison, SD = 147, n = 70, range = 2–669). Bison density was higher in Costilla than in historical areas during both the calving (Costilla = 12.0 bison/km2 [2022], no estimate for 2023; historical = 1.7 bison/km2 [2022], 2.0 bison/km2 [2023]) and breeding seasons (Costilla = 11.0 bison/km2 [2022], 4.4 bison/km2 [2023]; historical = 2.1 bison/km2 [2022], 1.6 bison/km2 [2023]) in both years of the study. Resource selection During both calving and breeding seasons, there was strong empirical support for differences in second-order resource selection between Costilla and historical areas as the models with area × resource (forage, elevation) variable interactions outperformed the models without interactions (calving ΔAICC = 1,537.4; breeding ΔAICC = 88,482.6) and the null models (calving ΔAICC = 215,095.6; breeding ΔAICC = 651,527.5; Supplementary Data SD5). During both the calving and breeding season, bison in historical areas selected home ranges with lower forage biomass relative to what was available across Vermejo, while bison in Costilla selected home ranges with higher forage biomass relative to what was available across Vermejo (Table 3; Fig 4; Supplementary Data SD9). Bison selected higher elevations when establishing home ranges in Costilla, while bison in historical areas avoided higher elevations (Table 3; Fig 4; Supplementary Data SD9). Bison in both Costilla and historical areas avoided steeper slopes when selecting home ranges (Table 3; Fig 4; Supplementary Data SD9). Bison in Costilla and historical areas selected grasslands and wetlands when establishing home ranges (Table 3; Fig 4; Supplementary Data SD9). Bison in Costilla avoided forests, while bison in historical areas selected forests (Table 3; Fig 4; Supplementary Data SD9). Coefficients from second-order resource selection models for bison in historical areas and Costilla during (a) calving season (historical n = 25 bison; Costilla n = 7 bison) and (b) breeding season (historical n = 35 bison; Costilla n = 13 bison) on Vermejo Park Ranch, New Mexico, USA, April–June 2022–2023. Forage, elevation, and slope are classification-based meaning that coefficients >0 indicate selection, <0 indicate avoidance. All other variables are distance-based, so coefficients <0 indicate selection, >0 indicate avoidance. Table 3Mean values used (± SD) by bison within home ranges on Vermejo Park Ranch, New Mexico, USA, April–September 2022–2023 in historical areas and Costilla during the calving (n historical = 25 bison; n Costilla = 7 bison) and breeding seasons (n historical = 35 bison; n Costilla = 13 bison).HistoricalCostillaCalvingMeanSDMeanSD Forage (kg/km2) 5222.93316.18451.02507.8 Elevation (m) 2446.6144.52954.157.1 Slope (%) 6.86.05.14.2 Grasslands (m) 206.2250.377.383.9 Wetlands (m) 374.1305.6146.7144.6 Forests (m) 79.6122.5201.8182.7 Breeding MeanSDMeanSD Forage (kg/km2) 7523.94142.19533.74971.8 Elevation (m) 2353.0168.32988.293.6 Slope (%) 7.55.76.84.8 Grasslands (m) 168.7206.098.3113.3 Wetlands (m) 436.6322.4179.7163.3 Forests (m) 88.9130.5139.1165.6 Mean values used (± SD) by bison within home ranges on Vermejo Park Ranch, New Mexico, USA, April–September 2022–2023 in historical areas and Costilla during the calving (n historical = 25 bison; n Costilla = 7 bison) and breeding seasons (n historical = 35 bison; n Costilla = 13 bison). Discussion Bison on Vermejo used smaller home ranges where a critical resource (forage) was both abundant and of higher quality with respect to nitrogen concentration, illustrating the fundamental role of forage productivity in allowing ungulates to reduce energetic costs and maximize space efficiency (Anderson et al. 2005; Fryxell et al. 2008). However, only bison in Costilla selected home ranges based upon the spatial distribution of the highest quantity of forage from the larger available landscape following translocation to higher elevations. Indeed, bison that remained in historical areas avoided higher forage biomass when selecting home ranges. These findings suggest that the spatial distribution and home range establishment of bison across Vermejo is governed by site fidelity rather than an optimal foraging strategy, consistent with work studying bison in Canada (Merkle et al. 2015a; Sawatzky et al. 2026) and Barren-ground Caribou (Rangifer tarandus granti) in Alaska, USA (Cameron et al. 2020). If animals do not overcome an important limiting factor (e.g., predation, forage productivity) when establishing home ranges, they are expected to continue to exhibit selection or avoidance of that factor at finer spatial scales (Rettie and Messier 2000; Hebblewhite and Merrill 2009). Since bison in historical areas on Vermejo did not select forage abundance when selecting home ranges from the larger landscape, theory predicts that they should prioritize selecting forage abundance and quality at finer spatial scales (Rettie and Messier 2000). Thus, site fidelity not only affects coarse-scale selection of home ranges but also likely influences spatial efficiency and movement decisions within home ranges (Kreling et al. 2021). Our first hypothesis was supported as forage abundance drove home range size for bison on Vermejo. Mitchell and Powell (2004) proposed 2 strategies used by animals to optimize their fitness when establishing home ranges: resource maximizing and area minimizing. Our findings support the area-minimizing strategy (Mitchell and Powell 2004, 2007), as bison used space efficiently by using smaller home ranges in areas of increased forage biomass and quality (supporting P1 and P2). Forage biomass strongly influenced home range size within the historical areas at lower elevations, but not within Costilla where there was little variation in forage biomass or home range size. Bison in Costilla used smaller home ranges, which allowed them to spend more time closer together and remain in larger groups over smaller areas, indicating higher levels of spatial cohesion and reduced intergroup movement (supporting P3). Conversely, bison in historical areas with lower forage biomass occurred in smaller and more variable groups and exhibited lower levels of spatial cohesion and greater distances between pairs of collared bison. Foraging competition should be reduced in areas with higher forage productivity and quality, allowing animals to aggregate in larger groups and feed closer together while using less space (Beauchamp et al. 1997; Pinacho-Guendulain and Ramos-Fernández 2017; Bond et al. 2019). Thus, forage availability shaped both space use (Tufto et al. 1996; Knüsel et al. 2019) and conspecific density (Coe et al. 1976; Pettorelli et al. 2009), as greater forage biomass at higher elevations allowed bison to congregate in larger, more spatially cohesive groups over smaller areas, similar to findings with other ungulates (Isvaran 2007; Bercovitch and Berry 2010). Our second hypothesis was also supported as bison at lower elevations selected home ranges with lower forage biomass, whereas bison at higher elevations selected home ranges with higher forage biomass, relative to availability across Vermejo (supporting P4). Consequently, without translocation, home range selection by bison on Vermejo appears to be influenced more strongly by site fidelity than the spatial variation in forage availability (Merkle et al. 2015a; Sawatzky et al. 2026). Similarly, forage availability did not influence variation in home range size within Costilla, providing further support that site fidelity plays a primary role in home range establishment and efficiency for bison on Vermejo. Traveling long-distances to high elevation sites would be energetically expensive, such that if migration does not provide a substantial fitness advantage, residency may be favored (Middleton et al. 2013; Courtemanch et al. 2017; Soriano-Redondo et al. 2023). Indeed, the straight-line distance from historical areas to Costilla is approximately 14 km with 500 m of elevation gain. Thus, site fidelity to lower elevations could result in a net energetic gain if energetic costs of migration are higher than maintaining larger home ranges in lower elevations with reduced forage conditions (Stamps 1995; Powell 2000; Gehr et al. 2020). Future research quantifying energy acquisition and expenditure of bison across sites with variation in forage quality and abundance during both migration and residency would help clarify whether site fidelity to poorer quality, lower elevation sites is truly maladaptive. Site fidelity is reliant on the ability of an animal to return to previously visited sites using knowledge that was developed and retained from previous experiences (Dalziel et al. 2008; Wolf et al. 2009; Merkle et al. 2014). Bison were restored to Vermejo via translocation in the 1920s meaning that they had no spatial memory regarding local environmental conditions. Previous owners of the ranch confined bison to lower elevations up until the late 1990s (> 70 years), as Vermejo was predominately used to produce a large herd of domestic cattle (Bos taurus; typically 4,000 to 5,000 but peaking at 19,000; Adams et al. 1979). As a result, bison on Vermejo have a history of using lower elevation sites and only recently (∼6–10 years ago) started to develop knowledge of additional sites via translocations. Migration originates when multiple generations acquire knowledge about the spatiotemporal variation in forage and relay that information to offspring and conspecifics (Jesmer et al. 2018). Thus, translocated animals require several decades of spatial learning and sharing of knowledge before migration may re-emerge (Jesmer et al. 2018). Given that individual bison make foraging decisions using knowledge gained from past experiences and information learned from group members (Merkle et al. 2014, 2015b), translocating bison to higher elevations on Vermejo should accelerate spatial learning and the transmission of forage patch information to bison at lower elevations. Yet, retaining accurate long-term information that can be relayed to conspecifics may require repeated experiences, as new memories may degrade the accuracy of old memories or old memories may interfere with the storage of new memories (Bouton 1993; Dukas 1999; Spencer 2012). Therefore, translocating the same bison each year may allow bison to retain spatial knowledge of high elevation sites more accurately and for longer periods (Spencer 2012), potentially promoting migration in the future. This pattern may also partially explain the higher levels of spatial cohesion in Costilla (in addition to reduced competition associated with abundant forage), as bison that were translocated for the first time likely gain information about the environment by following bison with previous experience in Costilla (Merkle et al. 2015b). We found, with data from remote sensing and vegetation sampling, that both forage abundance and quality were higher in Costilla—where bison established home ranges following translocation—relative to historical areas and the Canadian River (Fig. 2). Importantly, our strongest inference is based on forage quantity because our estimates of forage quantity provided finer spatial and temporal resolution (30 m every 16 days from RAP) than our forage quality data estimated from vegetation sampling only. However, our forage quality estimates also support our hypothesis, though at a coarser scale. Forage quantity and quality are often negatively correlated in space (Fryxell 1991; Hebblewhite and Merrill 2008), but several studies have found that areas with the highest forage quantity also corresponded with the highest forage quality (Kaszta et al. 2016; Schoenecker et al. 2022; Gutierrez et al. 2026; Musto et al. 2025). Our results also indicate that forage quantity and quality were positively related on Vermejo based upon our vegetation sampling, but it is possible that if we sampled forage quality at a finer scale quantity and quality of forage would not be as strongly linked. Forage abundance and quality were likely greater at higher elevations in Costilla due to higher precipitation, cooler temperatures during the growing season, and more mesic habitats (Legler 2010), but additional work is needed to explore these hypotheses. Thus, without translocation efforts, establishment of home ranges by bison likely represented suboptimal habitat selection, as areas of higher forage abundance and quality in Costilla were left unused. However, research suggests that bison can enhance forage productivity and quality through intense grazing in some landscapes (Geremia et al. 2019; Schoenecker et al. 2022; Musto et al. 2025), such that future work should evaluate whether concentrated grazing by bison on Vermejo might improve forage conditions in lower elevations. Nevertheless, managers can continue to aid home range selection for bison by translocating them to Costilla and other sites where forage availability and quality are higher. In addition, these translocations have important implications for high-elevation grassland ecosystems by allowing bison to expand their ecological role as large aggregate grazers to underused areas (e.g., Sanderson et al. 2008; Kohl et al. 2013; Geremia et al. 2019; Musto et al. 2025). However, not all translocations resulted in home range establishment on Vermejo. Indeed, managers translocated bison to 2 sites on Vermejo, yet bison only established home ranges in Costilla where productivity and quality were higher than historical areas, while they abandoned the Canadian River where productivity and quality were similar to historical areas. Managers translocating ungulates should use remote sensing or vegetation sampling to evaluate resources available in release sites, such as forage abundance and quality, prior to translocation to increase the success of postrelease fidelity. In addition to forage quantity and quality, our results suggest that the timing and type of translocations may have influenced postrelease fidelity for bison on Vermejo. Bison translocated to the Canadian River prior to calving left these areas quickly and returned to historical areas where they typically give birth, whereas bison translocated to Costilla after giving birth established home ranges and remained until late August. Given that many ungulates have preferred calving grounds (Cameron et al. 2020), translocating bison after they have given birth or holding bison in soft releases until they have given birth may help prevent bison from abandoning translocation sites to return to areas where they historically produce calves. In 2022, bison hard-released in the Canadian River returned immediately to historical areas, whereas some bison soft-released in the Canadian River stayed for nearly a month, perhaps suggesting that soft releases are more effective than hard releases (Parker et al. 2008; Bryan 2018). However, in 2023, all bison returned almost immediately to historical areas despite being soft-released in the Canadian River. Therefore, bison may abandon translocation sites even following a soft release if forage productivity or quality is not superior to historical areas or to return to preferred calving grounds. Translocations similar to those in our study may increase spatial knowledge of animals that may eventually promote migration, though long-term research is needed to fully evaluate the success of translocations with respect to the cultural transmission of information on Vermejo (Jesmer et al. 2018). Although forage availability strongly influences home range size, large herbivores may not always select home ranges optimally (Kreling et al. 2021) or in accordance with ecological theory (Powell 2000; Mitchell and Powell 2004). Our findings indicate that suboptimal selection of home ranges with respect to forage abundance by bison was driven by site fidelity shaped by limited spatial knowledge of landscape-level resource availability derived from decades of historical occupation of lower elevation sites. As ecosystems become increasingly modified, site fidelity, though once beneficial, may have negative consequences on fitness for animals that remain loyal to deteriorated sites (Merkle et al. 2022). Recognizing the drivers of home range establishment and site fidelity can help ecologists identify appropriate conservation actions to allow populations to persist and maintain important behavioral traits. Animals that display lower levels of fidelity may easily adapt to changing landscapes (Morrison et al. 2021; Milligan et al. 2023), but it may be critical to employ management efforts (e.g., translocation) for animals that exhibit strong site fidelity and are less behaviorally flexible (Tracz et al. 2010; Wyckoff et al. 2018). Our work highlights the multifaceted nature of home range establishment by large ungulates, suggesting that past experiences or spatial knowledge should be considered alongside important limiting resources like forage availability and quality. Supplementary Material