OUP user menu

Yellowstone Pronghorn Alter Resource Selection After Sagebrush Decline

Sylvanna J. Boccadori, P. J. White, Robert A. Garrott, John J. Borkowski, Troy L. Davis
DOI: http://dx.doi.org/10.1644/07-MAMM-A-173.1 1031-1040 First published online: 15 August 2008


Animals alter resource use as conditions change and these changes could have significant consequences on demography. We generated a logistic resource selection probability function for habitat use and estimated percent composition of diets by pronghorn (Antilocapra americana) in Yellowstone National Park during winter to assess if selection of sagebrush (Artemisia) has changed since the 1960s, corresponding with diminishing sagebrush and increasing seral species such as rabbitbrush (Chrysothamnus). We located 37 radiocollared adult females in 1,440 different groups during December 1999 through March 2005 and compared habitat attributes to 9,347 randomly selected points from the winter range. Pronghorn preferred greasewood (Sarcobatus vermiculatus) and selected grasslands in comparison to the sagebrush cover type. The relative selection against sagebrush may stem from a substantial decrease in this plant type on the winter range by the 1960s due to intense browsing by congregated ungulates. The percent composition of sagebrush in the winter diets of pronghorn decreased from 67% during 1985–1988 to <10% during 2000–2001, while rabbitbrush increased from 5% to 60%. These findings and the results of companion time-series analyses suggest the winter range for Yellowstone pronghorn may not support larger populations (>500) of pronghorn for sustained periods, as occurred historically. Similar effects from habitat fragmentation and degradation face managers of migratory ungulates worldwide.

Key words
  • Antilocapra americana
  • Artemisia
  • diet
  • habitat
  • pronghorn
  • regression
  • resource
  • sagebrush
  • selection
  • Yellowstone

Habitat selection requires making trade-offs between the acquisition of resources, avoidance of predators, and minimization of energetic costs (Werner and Anholt 1993). Thus, large herbivores should select habitats that allow them to avoid or reduce the detrimental effects of the most important limiting factors such as avoiding areas where the likelihood of encountering a predator is high or using habitats where thermal conditions are enhanced (Dussault et al. 2004, 2005). In turn, the exploitation of resources could result in altered plant-herbivore equilibria by suppressing the growth and recruitment of plants (Caughley 1981; Sinclair 1977). In response, large herbivores should alter resource use as conditions change and these changes, in turn, could have significant consequences on demography (Dussault et al. 2005).

Logistic regression has become an increasingly popular model for studies of habitat selection by wildlife (Manly et al. 2002). Resource selection probability functions (RSPFs) can be fitted when both used and unused locations were sampled. However, if true unused locations are unknown, and instead a random sample of available locations was collected, then logistic regression procedures cannot be used to estimate the intercept of the exponential RSPF. Rather, this resource selection function can only provide relative probabilities of use for a given set of predictor variable levels (Keating and Cherry 2004). Lele and Keim (2006) applied the method of simulated maximum likelihood (Robert and Casella 1999) to estimate the parameters of an RSPF. In turn, the resulting selection probability estimates can be used to estimate the relative risk, which is the ratio of the probability of use for 1 set of covariate levels relative to the probability of use for a 2nd set of covariate levels. In addition, the RSPFs fitted by this method do not suffer the drawbacks of the exponential RSPF, including biased parameter estimates if the random sample of available locations contains both unused and used locations and a constraint that parameter values must yield a negative exponent in the function for all values of the covariate predictors (Lele and Keim 2006).

Selection of habitats by pronghorn (Antilocapra americana) has been well documented for grassland, shrub-steppe, and desert biomes (Yoakum 2004a). Pronghorn are long-legged cursors, adapted to great speed and endurance running over uneven, rolling terrain to outdistance predators (Byers 1997). Thus, they are associated most frequently with flat to rolling terrain < 10% in grade and with low, relatively open vegetation that allows broad visibility and mobility (Yoakum 2004a). They also use topography (e.g., depressions), rocky areas, and taller plants for thermal, hiding, and bedding cover of young. Pronghorn are selective feeders and prefer to eat a diversity of shrub and forb species, with graminoids composing a minor portion of the diet. During autumn and winter, however, pronghorn in shrub-steppe habitats frequently browse on shrubs, which are higher in several nutrients (e.g., protein) and more available than most forbs or grasses (Yoakum 2004b). Thus, ranges with a variety of vegetative communities are preferred, although winter distribution is generally restricted to lower-elevation, windswept areas where food is exposed and snows are generally <15 cm deep. Loss of habitat is the biggest dilemma facing pronghorn (Yoakum 2004a).

There are serious concerns about the viability of pronghorn in Yellowstone National Park because low abundance (<300 individuals) and apparent isolation have increased their susceptibility to random, naturally occurring catastrophes (National Research Council 2002). This low abundance may stem from reductions in the density and productivity of big sagebrush (Artemisia tridentataHouston 1982; Singer and Renkin 1995; Wagner 2006; Wambolt and Sherwood 1999), the staple winter food of pronghorn during 1930–1990 (Barmore 2003; Murie 1940; O'Gara 1968; Singer and Norland 1994). Our objectives were to generate logistic RSPFs of habitat use, estimate percent composition of diets, and evaluate if selection of sagebrush types by pronghorn has changed since the late 1980s, corresponding with diminishing sagebrush and increasing seral species such as rabbitbrush (ChrysothamnusWagner 2006). This information was essential for developing effective management strategies to conserve this native species of special concern that retains 1 of only 2 pronghorn migrations remaining in the greater Yellowstone region (White et al. 2007b).

Materials and Methods

Study area.—Pronghorn in Yellowstone inhabited foothills, mountain slopes, and valley bottoms along the Gardiner, Lamar, and Yellowstone rivers in the northern portion of Yellowstone National Park, Wyoming, and adjacent areas of Montana (Boccadori 2002). Pronghorn congregated on a 30-km2 winter range near Gardiner, Montana, during November through March. Approximately 70% of these pronghorn then migrated 15–50 km east to summer at higher elevations, whereas the rest stayed on the winter range year-round (White et al. 2007b). The climate was characterized by short, cool summers and long, cold winters with a mean annual temperature of 1.8°C. Mean

annual precipitation varied from 25 to 35 cm as elevation increased from 1,500 m in river drainages to 3,400 m on mountains. Average snow–water equivalents (i.e., amount of water in snow) ranged from 2 to 30 cm along this elevational gradient (Farnes et al. 1999). Large-scale fires during 1988 burned approximately 30% of the summer range of pronghorn, but none of the winter range (Scott and Geisser 1996). Extreme drought conditions existed in northwestern Wyoming during 1998–2004 (Palmer Drought Severity Index; National Climatic Data Center, Asheville, North Carolina).

The winter range for pronghorn was primarily open grassland–sagebrush steppe with interspersed upland grasslands, wet meadows, old agricultural fields and pastures, agricultural fields on private land, and nonvegetated areas (Boccadori 2002; Savage 2005). Vegetation composition of the winter range of pronghorn was surveyed using methods described by Daubenmire (1959) and Canfield (1941), as discussed in Boccadori (2002:13–14). Grassland cover type composed 48% of the range and occurred on any slope and exposure below 1,850 m. Dominant grass species were Sandberg bluegrass (Poa secunda), Idaho fescue (Festuca idahoensis), prairie junegrass (Koeleria macrantha), and bluebunch wheatgrass (Elymus spicata). Other dominant species were prickly pear cactus (Opuntia polyacantha), fringed sage (Artemisia frigida), alyssum (Alyssum desertorum), and sandwort (Arenaria hookeri). Most frequent species were Sandberg bluegrass, fringed sage, and sandwort. Herbaceous plant cover was low (9%) relative to other cover types and a moderate to high proportion of ground surface was bare soil, gravel, or rock.

Old fields cultivated, irrigated, or both before park acquisition in 1932 composed 13% of the winter range. This cover type also included an area in the park planted in alfalfa (Medicago sativa) during the early 1900s. Current pastureland lies outside the north boundary of the park on private land. These 2 types were combined because the vegetation structure and plant types were similar. Both types occurred below 1,650 m on level terrain. The dominant plant species was crested wheatgrass (Agropyron cristatum), with some annual wheatgrass (Agropyron triticeum) and alyssum. The most frequent species (98%) was crested wheatgrass. Herbaceous plant cover was moderate (18%) relative to other cover types.

Grassland–sagebrush mix cover type comprised 10% of the winter range and occurred between 1,500 and 1,850 m elevation. The overstory was dominated by Wyoming big sagebrush (Artemisia tridentata wyomingensis), with varying amounts of rubber rabbitbrush (Chrysothamnus nauseosus) and green rabbitbrush (C. viscidiflorus; 0.2–4.5% cover). The understory was dominated by Sandberg bluegrass, prairie junegrass, annual wheatgrass, and alyssum. The most frequent species were Sandberg bluegrass and prairie junegrass. Total herbaceous canopy cover was moderate (16%) relative to other cover types. Shrub canopy cover was low (8%) relative to other shrub cover types; 2% of the canopy was composed of dead sagebrush, and 1% of dead rubber rabbitbrush.

Sagebrush cover type composed 20% of the winter range and occurred between 1,500 and 1,850 m elevation in and out of the park. The overstory was dominated by basin big sagebrush (Artemisia tridentata tridentata) and Wyoming big sagebrush. The understory was dominated by Sandberg bluegrass, prairie junegrass, and bluebunch wheatgrass; these species also were the most frequent. Live shrub canopy cover was among the highest (14%) relative to other shrub cover types; 4% of the canopy was composed of dead sagebrush. Herbaceous plant cover was moderate (18%) relative to other cover types.

Rabbitbrush cover type composed 3% of the range and occurred below 1,650 m elevation on level terrain. Rubber rabbitbrush and green rabbitbrush were the dominant overstory shrubs. The understory was dominated by crested wheatgrass, alyssum, dandelion (Taraxacum officinale), and stickseed (Lappula redowskii). The most frequent herbaceous species were crested wheatgrass, alyssum, and stickseed. Shrub canopy cover was among the highest (15%) relative to other shrub cover types; 7% of the canopy was composed of dead rabbitbrush species. Herbaceous plant cover was high (23%) relative to other cover types.

Greasewood cover type composed <1% of the winter range and occurred below 1,650 m elevation on level terrain. Greasewood (Sarcobatus vermiculatus) was the dominant overstory shrub. The understory was dominated by annual wheatgrass and Sandberg bluegrass. The most frequent herbaceous species were annual wheatgrass, Sandberg bluegrass, and alyssum. Shrub canopy cover was among the highest (17%) of the other shrub cover types; 7% of the canopy was composed of dead greasewood. Herbaceous plant cover was moderate (15%) relative to other cover types. Other cover types composed 5% of the winter range and included agricultural fields planted in alfalfa, riparian areas, and coniferous forests of Douglas-fir (Pseudotsuga menziesii) at higher elevations.

Habitat use.—We used net guns from helicopters to capture and radiocollar (Telonics, Mesa, Arizona; Lotek Wireless, Newmarket, Ontario, Canada) 37 adult female pronghorn 1–8 years old during February 1999, March 2000, and February 2004. Capture and handling protocols followed guidelines approved by the American Society of Mammalogists (Gannon et al. 2007). Telemetry homing techniques (White and Garrott 1990) were used to visually locate 11–23 marked animals (i.e., 7–16% of total females) during December through March of 2000–2005. We located marked animals in 1,446 different groups and obtained a mean of 86 locations per animal (SE = 8 locations, minimum = 12 locations, maximum = 153 locations). Although individuals were repeatedly sampled, we pooled data across winters and assumed that telemetry locations provided a simple random sample of use by the population. This assumption seemed reasonable because large numbers of individuals were sampled and pronghorn were gregarious (i.e., location data represent larger groups rather than single animals).

We assumed that habitat use by pronghorn was non-destructive and any location could be revisited multiple times. For each telemetry location, we let (X1, X2, …, X8) denote the vector of environmental covariates representing resources that could be used by pronghorn, where X1 was elevation in thousands of feet, X2 was slope in tens of degrees, X3 was solar radiation index (SRI), and X4X8 were dummy variables indicating if the cover type was old fields–current pasture, grassland–sagebrush mix, sagebrush, rabbitbrush, and grease-wood, respectively. The “other” cover type was excluded from the modeling process because of the small number (n = 6) of observations in this category. The reference cover type was grassland, which corresponded to setting each of the 5 dummy variables to 0. We assigned attribute values to use points under the assumption telemetry locations were strictly accurate. Any violation of this assumption would introduce additional noise to our analyses, thereby making the modeling of the selection probabilities less precise. However, locations were not systematically biased. In the post hoc exploratory analysis, X1 and X2 were centered by subtracting the mean elevation and mean slope, respectively, to minimize collinearity problems due to the inclusion of square terms in alternate models.

The spatial extent of habitats available to pronghorn was defined by aggregating relocation points across winters and constructing a polygon of the winter range with ArcView Geographic Information System (GIS—Environmental Systems Research Institute 1998). We assumed that each pronghorn had a priori knowledge of the winter range and could travel throughout this area in the time elapsed between relocations. We generated a similar vector of covariates (X1, X2, …, X8) for 9,347 available locations randomly selected from the winter range using the Hawth's Analysis Tool extension in ArcGIS (Hooge et al. 1999). We used geographic information system coverages of vegetation type, elevation, solar radiation, and slope to obtain attribute data on available and used locations. We classified each location into 1 of the 7 vegetation cover types based on a geographic information system coverage with a minimum polygon size of 0.6 ha (Boccadori 2002:12–14).. Fenced areas and bodies of water were considered unavailable to pronghorn. The accuracy of the vegetation cover map based on ground verifications (n = 1,779) was 73% for rabbitbrush and 88–100% for the other cover types (Boccadori 2002:27).

Elevation and slope were derived from a 10-m digital elevation model (Spatial Analysis Center, Yellowstone National Park, Wyoming). Solar radiation affects microsite conditions such as temperature, snow depth, and vegetation (Watson et al. 2006) and was indexed as a function of slope, aspect, and latitude (Keating et al. 2007). Index values were greatest for south-facing, moderate slopes and least for steep, northern aspects.

Our data set was comprised of use and availability data rather than use and nonuse data. Thus, we applied the simulated maximum-likelihood method developed by Lele and Keim (2006) in this contamination setting to estimate an RSPF for pronghorn. Using the simulated maximum-likelihood method, we fitted logistic RSPFs π(x; β) = exp(xβ)/[1 + exp(xβ)], where x is a set of the covariate values in the logistic RSPF and β is the vector of parameters corresponding to covariates plus an intercept. Our goal was to estimate the logistic RSPF π(x; β) or, equivalently, to estimate the vector β of parameters.

To generate the simulated maximum likelihood estimates (Lele and Keim 2006), we let x1, x2, …, x1,440 be vectors of elevation, slope, SRI, and cover type associated with the 1,440 locations of groups with radiocollared pronghorn. Based on these observations, the log-likelihood function for the logistic RSPF was: Embedded Image where fA(x) is the multivariate density function of covariates for the population of available pronghorn locations, P(β) is the expected probability of pronghorn use given x under fA(x), and π(x; β) is the logistic RSPF defined earlier. The analytical forms of both fA(x) and P(β) were not known, but fA(x) was approximated from our large sample of model covariates associated with the 9,347 random locations of availability. From this empirical distribution, we could estimate P(β) for any fixed parameter vector β using Monte-Carlo simulation methods and, also, obtain Monte-Carlo estimates of the log-likelihood function for any fixed β using: Embedded Image where xj*, j =1, 2, …, 9,347 is the large simple random sample taken with replacement from the distribution fA(x). We used the Nelder–Mead numerical optimization method to maximize the function l̂(β; x1,x2, …, xn) and obtain the maximum-likelihood estimate of β. Asymptotic standard errors and log-likelihood function values were generated using modified R-code (S. Lele, University of Alberta, Edmonton, Alberta, Canada, pers. comm.), which, in turn, can be used to generate Akaike's information criterion corrected for small sample size (AICc) values. Because the maximum-likelihood method generates asymptotically normal estimators, confidence intervals also were generated.

We fitted a set of 15 a priori logistic RSPFs corresponding to the possible linear combinations of the 4 covariates and evaluated them using AICc (Burnham and Anderson 2002). A linear structure was equivalent to assuming that probability of use changed with the covariate in a strict logistic fashion. We also examined 8 post hoc logistic RSPF models that included 1 or more of the following effects: elevation × slope, elevation × cover type, and slope × cover type interactions and quadratic effects for elevation or slope. In addition, we compared the empirical cumulative distribution functions of elevation and slope across cover types for the telemetry location (pronghorn use) data and the random (available) location data. A cumulative distribution function for elevation or slope is a plot of the cumulative proportion of elevations or slopes less than or equal to any given elevation or slope. If resource selection by pronghorn was not related to elevation or slope, then the cumulative distribution functions for use and availability would be similar within each plot.

Diet composition.—We estimated the botanical composition of diets of pronghorn using microscopic examination of fresh fecal material (Sparks and Malachek 1968). Fecal pellets (1 pellet/pile) were collected from adult female pronghorn each week during January through March, 2000 and 2001, and randomly assembled into 2 (2001) or 3 (2000) monthly composites of 12 pellets each. Samples were oven-dried at 70 C and ground to 1-mm size in a Wiley Mill (Thomas Scientific, Swedesboro, New Jersey). Staff at the Composition Analysis Laboratory, Fort Collins, Colorado, prepared 5 slides from each composite and examined 20 fields per slide for a total of 100 fields per composite sample. They identified plant species to at least the genus level using epidermal cell tissue fragments. Monthly means were calculated as percent relative density based on the average of 3 composites per month during 2000 and 2 composites per month during 2001. Highly digestible species such as sagebrush taxa may be underestimated because we did not adjust results for differential digestibility (Sparks and Malachek 1968; Striby et al. 1987).


The best a priori logistic RSPF model included elevation, slope, and cover type (Table 1): Embedded Image with corresponding parameter estimates and standard errors: b0 = 1.075 ± 0.880, b1 = ‒1.329 ± 0.138, b2 = ‒0.172 ± 0.026, b3 = ‒0.049 ± 0.077 (old fields–current pasture), b4 = ‒0.284 ±0.110 (grassland–sagebrush mix), b5 = ‒1.789 ± 0.320 (sagebrush), b6 = 0.303 ±0.116 (rabbitbrush), and b7 = 1.152 ± 0.206 (greasewood). Model effect weights (wp) were 1.00 for elevation and slope, 1.00 for vegetation cover type, and 0.21 for SRI. No other model was within 2.6 ΔAICc values of this model having AICc = ‒633.7.

View this table:
Table 1

A priori and exploratory approximating logistic resource selection probability function models for winter habitat selection by pronghorn (Antilocapra americana) in Yellowstone National Park during 1999–2005. Abbreviations are K (number of model parameters), ΔAICc (Akaike's information criterion corrected for small sample size), wi (AIC model weight), and SRI (solar radiation index). AICc was ‒633.7 for the top a priori model and ‒1,059.0 for the top post hoc model.

A priori models
Elevation + slope + cover type800.77
Elevation + slope + SRI + cover type92.60.21
Elevation + cover type77.30.02
Elevation + SRI + cover type840.90.00
Slope + cover type7129.60.00
Slope + SRI + cover type8134.40.00
Elevation + slope + SRI4159.80.00
Elevation + slope3178.90.00
Elevation + SRI3190.80.00
SRI + cover type7220.40.00
Cover type6278.60.00
Slope + SRI3381.40.00
Exploratory models: elevation + slope + cover type
Elevation2 + slope2 + elevation × cover type + slope × cover type2000.98
Elevation2 + slope2 + elevation × slope + elevation × cover type + slope × cover type217.90.02
Elevation2 + slope2 + slope × cover type1522.60.00
Elevation2 + slope2 + elevation × slope + slope × cover type1631.50.00
Elevation2 + elevation × cover type + slope × cover type1955.50.00
Elevation2 + slope2 + elevation × cover type1579.40.00
Elevation2 + slope2 + elevation × slope + elevation × cover type1684.60.00
Slope + elevation × cover type + slope × cover type19200.20.00

The best post hoc model included cover type, linear, and quadratic effects for elevation and slope, and elevation × cover type, and slope × cover type interactions (Table 1): Embedded Image with parameter estimates b0 = ‒8.552 ± 0.454 (SE), b1 = ‒2.384 ± 0.477, b2 = ‒0.085 ± 0.031, b3 = ‒0.083 ± 0.173 (old fields–current pasture), b4 = 0.080 ± 0.113 (grassland–sagebrush mix), b5 = ‒1.685 ± 0.259 (sagebrush), b6 = ‒0.543 ± 0.372 (rabbitbrush), b7 = ‒1.856 ± 1.613 (greasewood), b8 = ‒13.603 ± 1.817, and b9 = ‒0.063 ± 0.017. Ci were dummy variables with C3 = 1 for old fields–current pasture, C4 = 1 for grassland–sagebrush mix, C5 = 1 for sagebrush, C6 = 1 for rabbitbrush, and C7 = 1 for greasewood. Thus, the elevation × cover type and slope × cover type interaction effects were, respectively, be3 = 0.969 ± 0.678, bs3 = ‒0.499 ± 0.133 (old fields–current pasture), be4 = 4.152 ± 0.869, bs4 = 0.047 ± 0.074 (grassland–sagebrush mix), be5 = 0.665 ± 1.226, bs5 = 0.261 ± 0.111 (sagebrush), be6 = ‒0.155 ± 1.441, bs6 = ‒0.755 ± 0.182 (rabbitbrush), and be7 = ‒8.334 ± 3.951, bs7 = ‒1.116 ± 0.791 (greasewood). Model effect weights (wp) were 1.00 for all linear, interaction, and quadratic effects, except for the elevation × slope interaction with weight = 0.02. No other exploratory model was within 7.9 ΔAICc values of this model with AICc = ‒1,059.

The lowest elevations and slopes have negative-coded values because elevation and slope were centered by their means. Thus, multiplication of a negative coefficient by a low (centered) slope or elevation yields a positive value. For example, the estimate of ‒8.334 for the elevation × (greasewood) cover type interaction effect means that for centered values of elevation, there is a very large positive effect at lower elevations. In fact, all elevations at use locations for greasewood were less than the mean elevation, implying that all centered elevations will be less than 0. Thus, the probability of use will tend to be larger within the greasewood cover type because of the lower elevations. As a result, interpretation of a single parameter estimate should not be made without taking into account related effects in the model. For example, the estimate for the elevation coefficient should not be interpreted without taking into account the estimates of the quadratic elevation effect b8 and the elevation × cover type interaction effects be3, be4, be5, be6, and be7, which also depend on the elevation.

For all cover types except sagebrush, pronghorn tended to select elevations and slopes that were in the low to middle range relative to the elevations and slopes that were available (Figs. 1 and 2). However, they tended to avoid the very lowest elevations within the grassland, old fields–current pasture, sagebrush, and rabbitbrush cover types. The cumulative distribution function plots of elevation for grassland indicated that the use locations tended to occur at lower elevations relative to what was available. In particular, 84% of use locations occurred at elevations < 1,707 m, whereas only 58% of available locations were < 1,707 m in the grassland cover type. Conversely, there was no strong evidence that resource selection based on elevation within the sagebrush cover type differed from what was available because the cumulative distribution function plots for the grassland–sagebrush mix were similar.

Fig. 1

Empirical cumulative distribution function (CDF) of elevation (feet) across cover types for pronghorn (Antilocapra americana) location (use) and random (available) data in Yellowstone National Park, Montana and Wyoming.

Fig. 2

Empirical cumulative distribution function (CDF) of slope across cover types for pronghorn (Antilocapra americana) location (use) and random (available) data in Yellowstone National Park, Montana and Wyoming.

Major forage items in the winter diet included 4 woody taxa: rabbitbrush; Gardner saltbush (Atriplex gardneri); winterfat (Eurotia lanata); and Rocky Mountain juniper (Juniperus-scopulorum; Table 2). Two forbs (granite gilia [Leptodactylon pungens] and madwort [Alyssum]) were consistently used, but graminoids were only used in March.

View this table:
Table 2

Percent relative density of epidermal cell tissue fragments in feces from adult female pronghorn (Antilocapra americana) in Yellowstone National Park during winters 2000 and 2001 using microscopic analysis. Monthly means were based on 3 composites of 12 pellets per month during 2000 and 2 composites per month during 2001. Winter means were based on all 15 composites.

Food itemJanuaryFebruaryMarchWinter
Bouteloua gracilis<1<121
Oryzopsis hymenoides<1<131
Bromus, Festuca, and Phleum<1<121
Total graminoids<1<1114
Eriogonum caespitosum1<10<1
Leptodactylon pungens2322
Other forbsa11<11
Total forbs5555
Shrubs and trees
Artemisia frigida7424
Artemisia tridentata4323
Atriplex gardneri6867
Ceratoides lanata9546
Juniperus scopulorum6656
ShepherdiaEleagnus, Pinus,
Total shrubs and trees92938288
  • a Antennaria, Aster, Astragalus, Cerastium, Descurainia, Oenothera, Suaeda, and Taraxacum.


The negative parameter estimates in the best a priori model indicated pronghorn in Yellowstone National Park selected lower elevations and slopes on their winter range. However, these relationships exhibited concave downward curvature (i.e.,

negative quadratic parameters) in the best post hoc model, suggesting probability of use peaked at some intermediate values of these covariates. Pronghorn selected the grassland reference type in comparison to sagebrush. Comparisons with old fields–current pasture, grassland–sagebrush mix, rabbit-brush, and greasewood cover types depend on elevation and slope.

The selection for greasewood, which naturally occurs at lower elevations relative to available elevations, is probably of little biological consequence because it composed <1% of the winter range and <0.5% of the winter diet. During feeding in 2000 and 2001, pronghorn did not show preference for particular cover types (Boccadori 2002), likely because they select for individual plants rather than cover types (Byers 1997; Dirschl 1963; Schwartz and Nagy 1976; Schwartz et al. 1977; Yoakum 2004a). None of the cover types on the winter range were extremely productive, as evidenced by the low percent canopy cover of herbaceous plants and shrubs (10–38%— Boccadori 2002). Thus, it is unlikely that even the most productive cover type could support sustained feeding by pronghorn and they met their nutritional needs by using a diversity of cover types.

The relative selection against sagebrush in comparison to the grassland reference type was surprising because other studies in the shrub-steppe biome have documented the importance of sagebrush to pronghorn in winter (Amstrup 1978; Bayless 1969; Severson and May 1967; Sundstrom et al. 1973; Yoakum 2004a). However, Barmore (2003) reported that pronghorn in Yellowstone selected for xeric grassland and avoided sagebrush and old fields during 1968–1970 when approximately 100–125 pronghorn (3 individuals/km2) occupied the winter range. Singer and Norland (1994) repeated this study during 1986–1988 when approximately 500 pronghorn (7–10 individuals/km2) and > 1,000 elk (Cervus elaphus) occupied the winter range. They also found apparent selection against sagebrush, although confidence intervals overlapped zero slightly (‒0.19, 0.0008). This selection against a cover type that is highly important to other populations of pronghorn may reflect the decreased abundance and quality of sagebrush on the winter range of pronghorn by the 1960s.

Evidence indicates that the production, germination, and survival of sagebrush on the winter range of pronghorn has been declining since the early 1900s in response to browsing by elk, mule deer (Odocoileus hemionus), and pronghorn. Rush (1932) and Cahalane (1943) reported losses of big sagebrush on the winter range, concurrent with increases in less-palatable rabbitbrushes. These declines in sagebrush were attributed to excessive levels of browsing by pronghorn (Kittams 1950) and 3–4 decades of use by a large elk herd (Wagner 2006). As a result, the National Park Service and State of Montana implemented aggressive population reduction efforts during 1932–1968, decreasing pronghorn counts from 800 to <200 and elk counts from 11,000 to <4,000 (Houston 1982). However, these reductions did not reduce browsing or improve the declining condition of sagebrush or other shrubby species on the winter range due to the continued concentration of ungulates in this wintering area (Houston 1982; Wagner 2006). A moratorium on culling in Yellowstone National Park was instituted in 1969 and elk numbers increased rapidly under this management regime to >19,000 by 1988 (Cole 1971; Coughenour and Singer 1996). Pronghorn numbers remained at approximately the postculling level for the next 15 years, before irrupting to 600 animals during 1983–1991 and then crashing to approximately 200 animals during 1992–1995 (White et al. 2007a).

Singer and Renkin (1995) reported that numbers of Wyoming big sagebrush decreased 43% and cover decreased 29% on the winter range of pronghorn between 1957 and 1990. Ungulate browsing has continued to restrict sagebrush heights and size on the winter range of pronghorn during recent decades, with almost no recruitment of seedlings (Wagner 2006). Browsed plants have thinner leaves, lower shrub production, and lower numbers of seedheads, which result in little regeneration (Hoffman and Wambolt 1996). Intense browsing also has provided subdominant sprouting shrubs (e.g., rabbitbrush) an opportunity to increase on the winter range of pronghorn because they sustain prolonged herbivory better than non-sprouters such as sagebrush (Wambolt and Sherwood 1999). Rabbitbrush is known to benefit when associated species are preferred forages and overutilized (Young and Evans 1978).

Temporal and spatial comparisons of diets lend further evidence of declining habitat condition on the pronghorn winter range. Browse species generally compose the dominant component of pronghorn diets during winter in the shrub-steppe biome, with big sagebrush being predominant (Bayless 1969; Cole and Wilkins 1958; Medcraft and Clark 1986; Severson and May 1967). Rabbitbrush is often consumed during autumn, but rarely during winter (Amstrup 1978; Cole and Wilkins 1958). Murie (1940) found that big sagebrush was the staple winter food of pronghorn in Yellowstone during the 1930s. Rabbitbrush and greasewood were eaten, but much less. Similarly, rumen and feces contents from pronghorn in Yellowstone during 1962–1970, and 1986–1988, respectively, were composed mainly of sagebrush (59–67%) with very little rabbitbrush (4–5%—Singer and Norland 1994). Consumption of big sagebrush was generally higher than for other browse species (O'Gara 1968). However, the percent composition of sagebrush in winter diets decreased to <10% by 2000–2001, while rabbitbrush increased to approximately 60%.

Despite the decline in sagebrush and change in diet selection, pronghorn in Yellowstone maintained relatively high reproductive rates (1.8 young per female) and survival rates of adult females (0.86). Recruitment is relatively low (<25 fawns per 100 adult females), but natal mortality appears to be due primarily to predation rather than malnutrition (Scott 2004). The persistence of high demographic rates despite decreased food supplies during the leanest period of the year may be possible because numbers of pronghorn are well below ecological carrying capacity after the crash during 1992–1995 (White et al. 2007a) and numbers of elk have decreased >50% since that time (White and Garrott 2005). Thus, per capita resources were not as limiting. The behavioral flexibility of pronghorn in Yellowstone also enables them to make dynamic and rapid changes in migratory tendencies in response to changing conditions (White et al. 2007b).

Our findings and those of companion time-series analyses (White et al. 2007a) suggest that the winter range for pronghorn in Yellowstone may not support larger populations (>500) of pronghorn for sustained periods, as occurred historically. This apparent reduction in carrying capacity due to decreased sagebrush is worrisome because migration routes to historic wintering habitat outside the park have been fragmented by development, fencing, and other land-use practices (Caslick 1998; Scott 2004; White et al. 2007b). The National Park Service has developed plans in conjunction with restoration experts to reestablish native vegetation dominated by big sagebrush–bluebunch wheatgrass to areas once tilled for agriculture and now supporting invasive alien species. Park personnel also are working with the United States Forest Service, State of Montana, private landowners, and conservation organizations to improve connectivity between the park and historic winter ranges to the north. Similar problems and remedial actions are facing managers of migratory ungulates worldwide (Berger 2004; Hebblewhite et al. 2006; Johnson et al. 2005; Schaller 1988; Thirgood et al. 2004).


This study was funded by the Bernice Barbour Foundation, Montana State University, National Park Service, and Yellowstone Park Foundation. We thank K. Keating for providing information and assistance to generate the SRI; S. Lele for providing R code used to generate parameter estimates, log-likelihood function values, and standard errors; S. Cater and A. Rodman for geographic information system analyses; T. Blackford for assistance with figure preparation; J. Byers, Hawkins & Powers Aviation, Helicopter Capture Services, W. Maples, Montana Aircraft, and M. Robinson for capture and aerial telemetry support; and K. Tonnessen and the Rocky Mountains Cooperative Ecosystem Studies Unit for facilitating cooperative funding agreements. We are grateful to the numerous technicians and volunteers that assisted with this project.


  • Associate Editor was Martin B. Main.

Literature Cited

View Abstract