Introduction

This analysis provides scientific support for the U.S. Fish and Wildlife Service (FWS) regarding management decisions at the National Elk Refuge (NER) in Jackson, Wyoming. The NER has fed most of the Jackson elk herd for part of the winter for more than a century. Cervus elaphus canadensis (Erxleben, 1777; elk) abundance has changed over time, but the average number fed at the NER each winter during 2017–22 was approximately 7,500 elk, 70 percent of the Jackson elk herd (Wyoming Game and Fish Department [WGFD], 2022b). Previous research has shown that winter feeding operations create large, dense aggregations of animals that facilitate the transmission of multiple pathogens, including Brucella abortus (which causes brucellosis; Cross and others, 2007), Psoroptes spp. (scabies species; Samuel and others, 1991), and Pasteurella multocida (which causes pasteurellosis; Franson and Smith, 1988). Chronic wasting disease (CWD), which is always fatal to hosts, affects ungulates (including elk, Alces alces [Linnaeus, 1758; moose], and deer) and can lead to population declines (Edmunds and others, 2016). In 2020, the first CWD-positive elk was detected in Grand Teton National Park. In Wind Cave National Park, CWD has reached prevalence as high as 18–29 percent in elk (Sargeant and others, 2021). Chronic wasting disease is spread directly through animal-to-animal contact and indirectly through environmental exposure to prions, the malformed PrP proteins that are the infectious agent of CWD (Miller and others, 2004). Prions can persist for years, if not decades in the environment. Accordingly, CWD could have a larger effect on the Jackson elk herd than that observed in other populations because feeding at the NER greatly increases elk contact rates and elk densities (Galloway and others, 2017; Janousek and others, 2021).

The FWS is writing an environmental impact statement and revising its Bison and Elk Management Plan (U.S. Fish and Wildlife Service and National Park Service, 2007) in consideration of the disease risk to the herd. A structured decision-making (SDM) process, facilitated by the U.S. Geological Survey, provides scientific support for this revision (this volume, chap. A, Cook and others, 2025). Briefly, the management alternatives under consideration include (1) continue feeding, (2) immediate cessation of feeding (hereafter referred to as no feeding), (3) increase elk harvest for a 5-year period to reduce the NER winter elk population to 5,000 animals, followed by feed cessation (hereafter referred to as increase harvest), (4) attempt to lower the number of elk to 5,000 by reducing the feed provisioned to elk at the NER for a 5-year period, followed by feed cessation (hereafter referred to as reduce feeding), and (5) continue feeding until CWD prevalence in the Jackson elk herd is estimated to be 3 percent, followed by feed cessation (hereafter referred to as disease threshold).

This analysis predicted future elk space-use (the areas that elk will occupy) during a 20-year period in the vicinity of Jackson, Wyoming, under the proposed alternatives. The overall approach closely followed previous research in the Bridger-Teton National Forest (Cook and others, 2023), wherein a resource selection function (RSF) was fit to elk global positioning system (GPS) collar data. The RSFs were used to estimate the per capita probability of use across the study area for various population segments of elk at monthly timesteps (Cotterill and Graves, 2024). Elk abundance estimates were derived from a dynamic population and disease simulation model (hereafter, elk CWD model) under the management alternatives (this volume, chap. B, Cross and others, 2025) and projected across the study area using the RSFs. The elk CWD model, in conjunction with the RSFs, formed the basis for the calculation of what we refer to as our measurable attributes, the metrics of interest to decision-makers defined during the SDM process (this volume, chap. A, Cook and others, 2025).

Most of the future alternatives under consideration included novel conditions for which scant or no relevant elk GPS collar data exist. For example, winter feeding operations at the NER have never been experimentally discontinued to allow researchers to observe elk behavior and population dynamics in the absence of feeding. However, decades of research in this study area provided insights into the potential future behavior of elk under the no feeding alternative. Relevant situations included (1) a mild winter with substantial forage resources during which no feeding occurred at the NER, (2) attempts to reduce the duration of feeding by initiating feeding later in the season and ending feeding earlier in the season, and (3) one occasion where underestimating the number of elk in the NER caused elk to be fed reduced rations (Janousek and others, 2021; FWS, 2019). We used a formal expert-elicitation process to collate these insights to parameterize the elk space-use models. Expert judgment is a quantitative expression of an expert’s belief, based on relevant system knowledge, and when elicited by a formal process designed to minimize cognitive biases, can result in reliable predictions when empirical information is impossible or infeasible to collect (Adams-Hosking and others, 2016; Martin and others, 2012; O’Hagan and others, 2006; Runge and others, 2011; Speirs-Bridge and others, 2010). The resulting models integrated the expert-provided information from the Wildlife Subject Matter Expert Team (WSMET; this volume, chap. A, Cook and others, 2025) with elk collar data such that model predictions closely matched expected outcomes from the expert elicitation process.

The analysis area under consideration was delineated by the WSMET and consists of two contiguous regions, the Jackson Elk Herd Unit (JHU) and the Northern Fall Creek Elk Herd Unit (NFCHU). Refer to appendix C1 for a map of the analysis area (fig. C1.2). The JHU matches the Jackson Elk Herd Unit as designated by the Wyoming Game and Fish Department and generally extends north from State Highway 22 and Jackson to encompass Grand Teton National Park, the National Elk Refuge, and other lands up to the southern Yellowstone National Park boundary and east to the Continental Divide (WGFD, 2022a). The NFCHU is the northern third of the Fall Creek Elk Herd Unit (WGFD designation; WGFD, 2022a; refer to this volume, chap. A, fig. A1 in Cook and others, 2025). The WSMET included the NFCHU to consider potential effects from elk movements under alternatives with reduced feeding. Public entities, including the U.S. Department of Agriculture Forest Service, the National Park Service, the FWS, and the State of Wyoming, manage much of the analysis area. However, in addition to the NER, other low-elevation areas that ungulates use in winter tend to be private property, mostly developed for agricultural or residential purposes. Of immediate concern was the potential conflict resulting from negative effects elk could have on private property, livestock, and sensitive vegetation communities if the NER were to cease winter feeding operations. The measurable attributes evaluating these negative effects consisted of the estimated elk use of private property, the Brucella abortus transmission risk from elk to cattle on private property, and elk use of core Populus tremuloides Michx. (quaking aspen), Populus angustifolia E. James (narrowleaf cottonwood), and Salix L. (willow) sensitive vegetation communities.

Methods

The NER has never experimentally discontinued feeding, and therefore, we could not fit models to the full suite of conditions that characterized the management alternatives. Yet, a primary goal of this work was to make quantitative predictions of elk redistribution if the NER ceases feeding elk in winter. To bridge this knowledge gap, a hybrid approach of fitting a resource selection function to data was used where possible, and then predictions were adjusted under the alternatives based on information about current and future elk movements from the expert-elicitation process.

An RSF was first estimated based on a use-availability design (Johnson and others, 2006), which measures where an animal has been compared to where an animal could have been. This RSF predicted space-use patterns that closely matched recent conditions and expert knowledge using data for two elk subpopulations: (1) elk fed during winter and (2) elk not fed in winter. To match the structure of the elk CWD model, monthly probability of use was predicted based on elk GPS collar locations. Fed and unfed population segments were pooled for May through November because the WSMET believed summer range selection patterns were not significantly different for these population segments. Winter was defined as December through April, and we further differentiated between average and severe winters. Severe winters were characterized as winters during which snowfall patterns concentrated more elk on feedgrounds (locations where wildlife agencies feed elk in winter), the NER had to provision larger amounts of feed to elk, and winter elk calf mortality exceeded 5 percent even with additional feed. These criteria were met in approximately 25 percent of recent observed years (this volume, chap. B, Cross and others, 2025). These combinations of variables created 27 monthly elk location datasets, consisting of 7 summer month (May–November) datasets and 20 winter month datasets (December–April). Each of the 5 winter months had fed and unfed population segments and both average and severe winter conditions (5[months]×2[fed, unfed]×2[average, severe winter]=20 datasets). The RSFs were fit to each dataset of elk locations separately to produce 27 predictive maps, each representing the study area as a grid of 1,020 by 1,020-meter (m) cells with values corresponding to the broad-scale RSFs’ per capita probability of use. Thus, the values of the cells in a predictive map sum to one. We estimated the number of elk in each cell by multiplying the number of elk from the elk CWD model (this volume, chap. B, Cross and others, 2025) by the matching predictive map. To express monthly elk use in daily figures, we calculated elk days by multiplying the number of elk estimated to use a given area by the number of days in each month. For alternatives that included feed cessation at the NER, we developed a set of rules for modifying the predictive maps once feeding stopped to match the elk CWD model assumptions and expert judgment (this volume, chap. B, Cross and others, 2025).

The overarching goal of the analysis was to calculate how elk distribution changes influenced measurable attributes, indicating issues of concern to managers. The measurable attributes, identified during the SDM process, can be used to evaluate the performance of the alternatives against one another. Most attributes reflected concerns at the scale of the study area. The reduce feeding alternative included a provision that elk exclosures (fenced areas that exclude elk) be built in the NER to protect sensitive areas—vegetation communities, including quaking aspen, narrowleaf cottonwood, and willow—during winter. The NER is significantly smaller than the full study area, and the proposed exclosures were small relative to the resolution of the broad-scale RSF across the full study area. To evaluate the local effects of exclosures on the quaking aspen, narrowleaf cottonwood, and willow species in the NER, a second (fine-scale) RSF was estimated at a 30 by 30 m resolution for the NER during winter. The broad-scale RSF projected the numbers of elk in the NER, and the fine-scale RSF distributed the numbers of elk across the NER.

Broad-Scale Resource Selection Function

State Highway 22 likely creates a partial barrier to elk movement between the two regions of our study area (the Jackson and Fall Creek Elk Herd Units). Elk rarely crossed this highway in the GPS dataset; however, there were limited data available for elk in the Fall Creek Elk Herd Unit, and the WSMET agreed that more elk are likely to move across this boundary if the NER stops feeding. The elk CWD model accounted for this movement between regions (this volume, chap. B, Cross and others, 2025). The spatial models in this chapter subsequently used the projected number of elk in each region in all simulations of the alternatives from Cross and others (this volume, chap. B, 2025) and assumed that elk remained within those herd units. The habitat covariates used by Cook and others (2023) from Maloney and others (2020) were also used in this study, namely the annual integrated Normalized Difference Vegetation Index, elevation, density of roads within a 2.5 kilometers (km) buffer of the location, maximum snow water equivalent, and the percentage of forest and percentage of native herbaceous cover within a 2.5 km buffer of the location (Cross and others, 2023; Maloney, 2020). Other habitat covariates included the distance to feedgrounds, calculated as the distance from the centroids of the four feeding areas on the NER and State feedgrounds in the JHU and NFCHU. With feedback from the WSMET, two continuous variables for aspect were added, where northness was the cosine-transformed radians and eastness was the sine-transformed radians. Other variables added based on WSMET feedback included the location’s distance to water (Dewitz and U.S. Geological Survey, 2021; Yan and others, 2019); distance to grass and distance to barren ground (Dewitz and U.S. Geological Survey, 2021); distance to agriculture (Forest Service, written commun., 2023; Cook, 2005; Teton County, Wyoming, 2023); and average minimum January temperature (Thornton and others, 2022). For fed elk in winter months, covariates included a binary feedground raster map for the NER; all raster maps were created at or resampled to a 1,020-m resolution (Cotterill and Graves, 2024). This resolution required fewer modifications to the original GIS data, was computationally practical, and was deemed appropriately precise considering multiple sources of uncertainty in the elk CWD model and our spatial analysis.

Elk habitat selection coefficients, $\beta$, were estimated by employing a use-availability design and logistic regression. For each cell, l, we computed the resource selection probability function (Johnson and others, 2006), w, as the following equation:

$w_{ljk}=\text{exp}\left({\beta }_{0jk}+x_{1ljk}{\beta }_{1ljk}+\dots +x_{nljk}{\beta }_{nljk}\right)$

,

where

x

is a vector of mean values of each environmental variable within each cell,

j

is the month, and

k

is an indicator variable for fed (k=1) or unfed elk (k=2).

Equation C1 estimated the per capita probability of use in each cell, U(x_l), as the following equation:

$U\left(x_{ljk}\right)=w\left(x_{ljk}\right)/\sum w\left(x_{ljk}\right)$

,

(C1)

and equation C2 estimated the number of elk, n, in each cell as the following equation:

$n_{ljkt}=N_{jkt}U\left(x_{ljk}\right)$

,

(C2)

where

N_jkt

is the predicted number of fed or unfed elk in month j and year t from the elk CWD model.

For measurable attributes associated with the number of elk days in a particular location, the number of elk estimated in each cell was multiplied by the number of days in the month.

Because the NFCHU region had limited data (from as few as three elk; table C1), we assumed elk in both regions respond similarly to habitat and supplemental feed availability and estimated habitat selection coefficients using data combined across the regions. Each monthly elk GPS dataset contributing to these models was randomly thinned to 6,072 locations, the smallest number of locations in any dataset, because the number of GPS locations varied greatly across datasets, and the total number of cells in the study area was only 6,000 at the 1,020-m resolution. Standardizing the sample size across datasets provided comparable estimates across models in terms of effect magnitude, because the ratio of used-to-available points influences model intercepts. Next, within each region, the per capita probability of elk space-use map was estimated using equation C3:

$U\left(x_{ljkR}\right)=w\left(x_{ljkR}\right)/\sum w\left(x_{ljkR}\right)$

,

(C3)

where

R

refers to the region, either the JHU or NFCHU.

One of the covariates in x, the binary raster map of the NER only exists in the JHU, but we reasoned that the three State feedgrounds in the NFCHU hold a similar level of attraction for fed elk during winter. Therefore, in the per capita probability of elk use calculations for winter-fed elk in the NFCHU, the NER covariate was replaced with a binary raster map representing South Park, Horse Creek, and Camp Creek feedgrounds. Further information supporting this assumption can be found in appendix C2.

Defining available habitat for RSF studies is fraught even under ideal circumstances (Northrup and others, 2013; Warton and Shepherd, 2010), but frequently employed methods include defining availability according to seasonal elk space-use based on home-range or kernel-density estimator methods (Manly and others, 2002). In our case, the GPS dataset likely does not include a representative sample of unfed elk in the system, particularly when divided into monthly timesteps and differing levels of winter severity. As a result, following those conventions could omit habitats occupied by segments of the elk population in our study area, a factor that was considered when the expert elicitation process defined the study area (this volume, chap. A, Cook and others, 2025). Another common convention in use-availability RSF designs is to sample availability at 10 times the number of used locations (Lowrey and others, 2021), but this convention is inadequate when the resolution of the considered areas is coarse (in other words, areas have fewer cells to sample). Most of these concerns are motivated by ensuring accurate estimation of specific habitat selection coefficients and reducing bias from autocorrelated habitat features (Northrup and others, 2013), some of which can be improved through even sampling (for example, a census) of the landscape to represent the full availability of habitats (Benson, 2013). In this analysis, no inference is based on individual selection coefficients, nor is there an inference to habitat selection of elk outside of our study system. The primary goal of the RSF was to predict elk space-use that closely matched observed elk selection patterns and expert judgment in the study area or in segments of the elk population where no comparable observational data existed. To fully represent the available conditions, one available point per cell was sampled in the study area (number of cells in the study area and number of available locations used in the model [n]=6,000).

Experts predicted that once the NER stops feeding elk, NER fed-elk will (1) move to State feedgrounds in the Gros Ventre River drainage, (2) become unfed-elk, or (3) move to the NFCHU. The elk CWD model moves elk to separate population segments to match the proportions of elk that experts identified as likely to transition to each segment (refer to this volume, chap. B, Cross and others, 2025). No further adjustments to the RSF were needed to accommodate the transition of NER elk to the NFCHU because this area had separate predictive maps. To implement the remaining changes for elk-space use predictions, we adjusted the probability of elk use in the NER and two transition areas. The following are cells identified to include in transition areas: (1) areas in the Gros Ventre River drainage surrounding State feedgrounds predicted to be used by fed elk and (2) other winter range areas away from feedgrounds predicted to be used by unfed elk (Native Winter Range Transition Area; NWRTA). The Gros Ventre Transition Area (GVTA) consisted of cells with greater than 0.0001 per capita probability of use for fed elk in February in the Gros Ventre River drainage and comprised 82.2 square km. The NWRTA consisted of cells with greater than 0.0001 per capita probability of use for unfed elk in February. The NWRTA included areas in the Gros Ventre River drainage farther from feedgrounds, private property in the area around Jackson, Wyoming, and other areas near the NER (encompassing 131 square km), excluding the GVTA and NER (app. C3). The RSF initially overestimated elk space-use in the GVTA and underestimated NER elk space-use compared to the elk CWD model. Because of these over and underestimations, a portion of GVTA elk was moved to the NER during winter months. Then, under the alternatives where feeding on the NER was discontinued, portions of predicted fed elk space-use in the NER were transitioned to the GVTA or NWRTA in accordance with expert judgment and the elk CWD model once feeding stopped (fig. C1). As feeding stoppage varied by alternative, this transition was made at various time points according to the alternative: (1) under the no feeding alternative, the NER elk transitioned to the GVTA or NWRTA in the first year; (2) under the increase harvest and reduce feeding alternatives, the elk transitioned at the beginning of the sixth year; (3) under the disease threshold alternative, the elk transitioned at the beginning of the third year because this year was when most elk CWD model simulations predicted that the 3 percent prevalence threshold of CWD in the Jackson elk herd was met. The proportions of fed elk being assigned onto or away from the NER during winter did not change according to winter severity. However, the average and severe winter RSFs made different predictions for the initial distributions of elk in the winter months before the relocation.

Results

The broad-scale RSF predicted more fed elk in the NER and State feedgrounds in midwinter compared to December and April when more animals are moving toward or away from feedgrounds in the JHU (fig. C2) and NFCHU (fig. C3) portions of the study area. Predicted elk space-use of the NER by fed and unfed Jackson elk closely matched the GPS locations (fig. C4). Predicted elk using the NER, State feedgrounds located in the Gros Ventre River drainage, and native winter range closely matched numbers from the elk CWD model during the 20 years of simulations across alternatives (figs. C5, C3.2, C3.3). Information about the summer RSF predictions can be found in appendix C4 (figs. C4.1, C4.2).

During the 20 years of simulations, the number of elk days on private property and sensitive vegetation communities reflected the changes in abundance projected in chapter B (this volume, Cross and others, 2025). However, the relative performance of alternatives varied among measurable attributes. The continue feeding alternative produced median estimates with the fewest cumulative elk days on private property and the least brucellosis risk from elk to cattle, although the interquartile ranges from 100 simulations overlapped across most alternatives, especially the increase harvest and reduce feeding alternatives (fig. C6). The no feeding and disease threshold alternatives produced the highest median estimates for these two measurable attributes. In terms of elk days in sensitive vegetation communities, the continue feeding alternative ranked well for narrowleaf cottonwood and willow when calculated with the broad-scale RSF but poorly for quaking aspen (fig. C7). The four alternatives that featured feeding cessation on the NER projected less impact on quaking aspen, narrowleaf cottonwood, and willow in the NER than the continue feeding alternative. Following feed cessation, the fine-scale winter RSF for the NER predicted lower elk densities in the National Elk Refuge. Further information can be found in appendix C5 (fig. C5.1).

There was substantial variation in the projected measurable attributes throughout the 100 simulations of each alternative. In terms of limiting the cumulative year-round elk space-use of private property, the continue feeding alternative ranked highest in 64–83 percent of simulations across the other alternatives, although the increase harvest and reduce feeding alternatives out-performed the continue feeding alternative in 36 and 35 percent of simulations, respectively (fig. C8). Similarly, although the continue feeding alternative ranked highest in limiting brucellosis transmission risk from elk to cattle (expressed as the number of Brucella-caused abortions on private properties used for overwintering cattle), some of the alternatives outperformed the continue feeding alternative in a substantial proportion of simulations, including the reduce feeding alternative (better in 39 percent of simulations), the increase harvest alternative (33 percent), and even the no feeding alternative (25 percent; fig. C8).

When summed through the end of the first five years of implementation, the increase harvest and reduce feeding alternatives (the alternatives that reduce NER elk abundance) outperformed the continue feeding alternative in terms of private property use by elk and were similar to continue feeding in terms of brucellosis risk (refer to app. C6 for more information).

In the matter of elk negatively affecting sensitive vegetation communities in winter, the continue feeding alternative frequently performed worst (fig. C9). Across core quaking aspen, narrowleaf cottonwood, and willow areas, the increase harvest alternative outperformed the continue feeding alternative in 98, 52, and 42 percent of simulations, respectively, across the full study area. Within the NER boundary, the continue feeding alternative performance was the worst in nearly all cases. The 5-year rankings of the sensitive area measurable attributes supported all other alternatives compared to continue feeding, particularly for the NER (refer to fig. C6.2).

Lastly, we summarized the mean annual sums of these measurable attributes in the first and last years of the simulations to provide some additional context (table C2).

There was substantial variation among the 100 simulations when viewed at any given timestep across each alternative. However, an overall pattern was clear: measurable attributes associated with elk space-use decreased in time across all alternatives because of the projected declines in elk abundance from Cross and others (this volume, chap. B, 2025).

Summary

Cervus elaphus canadensis (Erxleben, 1777; elk) abundance was projected to decline during the 20-year period across the alternatives, including the continue feeding alternative. Depending on the alternative, most abundance changes resulted from disease effects, management actions taken to reduce feeding or elk abundance directly, or a combination of those factors (this volume, chap. A, Cook and others, 2025). The measurable attributes reflected human-elk conflicts on private property, including prospective elk-to-cattle brucellosis spillover risk and effects on sensitive vegetation communities in winter (December–April). These effects were projected to decrease when abundance declined, and in general, the magnitude of these effects in year 20 was roughly half of what was estimated in year 1. However, the exact numbers depend on the alternative. On average, the continue feeding alternative predicted the smallest elk population at the end of year 20 because of the high chronic wasting disease (CWD) mortality (this volume, chap. B, Cross and others, 2025). Continue feeding is also predicted to concentrate more elk at the National Elk Refuge (NER) compared to other alternatives, and the NER contained a relatively large proportion of core Populus tremuloides Michx. (quaking aspen) areas. As a result, continue feeding is projected to result in fewer elk days on private property and reduced brucellosis spillover risk to cattle, but more elk days on quaking aspen areas across the full study area and more elk days on sensitive vegetation communities in the NER. Although the continue feeding alternative had the lowest projection for cumulative risk of brucellosis transmission from elk to cattle, fully contextualizing this risk is difficult without putting the risk into economic terms, which required information that was unavailable at the time of this study; estimates of the cost of brucellosis spillover events to cattle producers in the Jackson Elk Herd Unit have not been documented. We predict differences among the alternatives of about 19–20 elk abortions on cattle properties over the full 20-year simulation period. However, brucellosis risk decreased over time under all alternatives (by roughly half) because elk numbers decreased. A constant seroprevalence rate was assumed, but if elk densities decrease as elk abundance decreases, then brucellosis seroprevalence may also be expected to decline. The total number of predicted elk abortions on winter cattle properties across the alternatives may be less than or nearly equal to historical elk abortion numbers, which did not result in reported cattle Brucella infections in the Jackson Elk Herd Unit.

A fraction of the Jackson elk herd migrates beyond the extent of the study area as defined in this analysis, typically onto adjacent public lands including Yellowstone National Park. Between May and November, 7–12 percent of the global positioning system (GPS) locations in our dataset fell outside the study area. These areas are beyond the authority of the cooperating agencies that contributed to the structured decision-making process, and for the purposes of calculating the measurable attributes in this analysis, only the number of elk days on private properties was likely affected by this decision. Whereas the number of elk days on private property was calculated across all months of the year, other attributes were limited to winter months. Calculation of the sensitive vegetation community measurable attributes was also spatially constrained to core elk winter range. By restricting the summer range, the summer use of private property inside the Jackson Elk Herd Unit was potentially overestimated, although this effect could be similar across alternatives. Thus, although the absolute numbers would change if the study area expanded, this change should not affect the relative ranking of the alternatives against one another.

The GPS data used in this modeling effort were originally collected for a variety of purposes and included mostly female elk captured on feedgrounds. Thus, incorporating expert knowledge was important in developing an appropriate and useful resource selection function (RSF), wherein the numbers of elk predicted in important areas generally match expert knowledge. We accommodated shifts in forage resources using monthly predictions, but forage may decline quickly in some areas where large numbers of elk transition, particularly in low forage years, potentially yielding subsequent movements. Although small areas in these monthly RSFs differ from known use patterns, these differences are consistent across alternatives and should minimally affect rankings across alternatives.

We incorporated as much model complexity as was allowable given our dataset. One limitation was the paucity of unfed-elk collar data, which is not surprising in a region where approximately 90 percent of the elk are fed during the winter (Wyoming Game and Fish Department [WGFD], 2022b) but also makes it challenging to predict what will happen without any feeding. Another limitation was the lack of feedback among models that approximated space-use and elk numbers. The elk CWD model informed our space-use models, but the results from the space-use models did not inform the elk CWD model. If the NER stops feeding elk and Bison bison (Linnaeus, 1758; bison) in winter, some redistribution of animals is expected. The details of that redistribution likely depend on factors not considered in the RSF, including the fidelity of individuals to seasonal ranges and shifting predator densities. For instance, an implicit assumption of habitat selection modeling is that future elk space-use will look like past elk space-use. Yet, the proportion of long-distance migrants in the Jackson elk herd decreased by approximately 40 percent over the 34-year period ending in 2012 (Cole and others, 2015). If this trend continues, we may be underestimating elk use of private properties across the 20-year simulations, and it is unclear how the feedground management alternatives may impact this declining trend in migration.

This chapter presents results based on all available GPS data from 287 female elk from 2006–23. From those GPS data, resource selection functions were estimated for fed and unfed elk across different management alternatives under severe and average winters at a 1,020-meter resolution on a monthly basis over 20 years (27 different predictive maps). We also did this on a 30-meter basis for NER-specific measurable attributes. Formal expert elicitation was used for those aspects that were not predictable given the available data and merged those with quantitative models while accounting for uncertainties. Finally, we summarized the results according to five measurable attributes representing the objectives of the NER. While there are areas for future improvement, these maps provide the best available information to decision-makers on this difficult management decision.