Journal of Mammalogy, 98(1):94–105, 2017

DOI:10.1093/jmammal/gyw135
Published online September 17, 2016

Summer habitat selection by Dall’s sheep in Wrangell-St. Elias

National Park and Preserve, Alaska
Gretchen H. Roffler,* Layne G. Adams, and Mark Hebblewhite
United States Geological Survey, Alaska Science Center, 4210 University Drive, Anchorage, AK 99508, USA (GHR, LGA)
Wildlife Biology Program, Department of Ecosystem Sciences and Conservation, College of Forestry and Conservation,
University of Montana, Missoula, MT 59812, USA (GHR, MH)
Present address of GHR: Alaska Department of Fish and Game, Division of Wildlife Conservation, 802 3rd Street, Douglas,
AK 99824, USA
* Correspondent: gretchen.roffler@alaska.gov

Sexual segregation occurs frequently in sexually dimorphic species, and it may be influenced by differential habitat
requirements between sexes or by social or evolutionary mechanisms that maintain separation of sexes regardless
of habitat selection. Understanding the degree of sex-specific habitat specialization is important for management
of wildlife populations and the design of monitoring and research programs. Using mid-summer aerial survey
data for Dall’s sheep (Ovis dalli dalli) in southern Alaska during 1983–2011, we assessed differences in summer
habitat selection by sex and reproductive status at the landscape scale in Wrangell-St. Elias National Park and
Preserve (WRST). Males and females were highly segregated socially, as were females with and without young.
Resource selection function (RSF) models containing rugged terrain, intermediate values of the normalized
difference vegetation index (NDVI), and open landcover types best explained resource selection by each sex,
female reproductive classes, and all sheep combined. For male and all female models, most coefficients were
similar, suggesting little difference in summer habitat selection between sexes at the landscape scale. A combined
RSF model therefore may be used to predict the relative probability of resource selection by Dall’s sheep in
WRST regardless of sex or reproductive status.

Key words:  alpine ungulate, Dall’s sheep, habitat selection, Ovis, reproductive status, resource selection function, sexual segregation

In sexually dimorphic ungulates, male and female adults tend to
display spatial sexual segregation with mixing occurring only
during the mating season (Bowyer 1984; Main et al. 1996; Kie
and Bowyer 1999). Sexual segregation in ungulates has been
researched extensively (Bowyer et al. 1996; Main and Coblentz
1996), and although multiple mechanisms have been proposed
(Barboza and Bowyer 2000; Ruckstuhl and Neuhaus 2002;
Main 2008), a universal explanation has not been accepted
(Bowyer 2004) and debate continues over the validity of cer-
tain hypotheses (Bowyer and Kie 2004). Furthermore, within
study systems more than 1 mechanism of sexual segregation
may be present, as they are not mutually exclusive (Bonenfant
et al. 2004; Bowyer 2004; Ruckstuhl 2007), or different mecha-
nisms may be present at varying spatial scales (Bowyer et al.
1996). Spatial segregation may be influenced by different
habitat requirements between the sexes (Main et al. 1996) or
by social mechanisms that maintain separation (Cransac et al.
1998; Ruckstuhl 1998) in the absence of habitat-based segrega-
tion (Conradt 2005).
Most hypotheses relating sexual segregation to habitat use
rely on body size dimorphism as the underlying factor explain-
ing differential use of habitat by males and females due to dif-
ferences in vulnerability to predation, reproductive strategies, or
metabolic requirements and digestive capabilities (Barboza and
Bowyer 2000; Ruckstuhl 2007). The 2 most widely accepted
and applied hypotheses are: 1) the predation risk hypothesis
(Bleich et al. 1997; Ruckstuhl and Neuhaus 2002; Neuhaus
et al. 2005), which is based on differences in strategies between
sexes to maximize reproductive success (Main and Coblentz
1996); and 2) the gastrocentric hypothesis (Barboza and
Bowyer 2000), which distinguishes between the greater diges-
tive efficiency of larger- relative to smaller-sized ungulates
(but see Clauss et al. 2007), while recognizing that larger indi-
viduals (males) have greater absolute metabolic requirements

Published by Oxford University Press on behalf of the American Society of Mammalogists. This work is written by (a) US Government
employee(s) and is in the public domain in the US.

94

Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949

by guest
on 23 April 2018
 ROFFLER ET AL.—HABITAT SELECTION BY DALL’S SHEEP 95
(Van Soest 1982; Bowyer 2004). The gastrocentric hypothesis
also predicts that reproductive status may influence digestive
efficiencies of females (Barboza and Bowyer 2000). Of the
hypotheses for sexual segregation related to social factors,
the most commonly used has been the activity budget hypoth-
esis (Ruckstuhl 1998; Ruckstuhl and Neuhaus 2002; but see
Yearsley and Pérez-Barbería 2005; Pérez-Barbería et al. 2007).
This hypothesis provides within-group synchrony of foraging
patterns as an explanation for individuals associating more
closely with their own sex. Other hypotheses of sexual segrega-
tion related to social factors include social preference (inclina-
tions to associate with same-sex individuals—Bon and Campan
1996; Yearsley and Pérez-Barbería 2005; Pérez-Barbería et al.
2007) and sexual segregation as a tactic to avoid aggression in
mixed-sex groups (Weckerly 2001).
Because ungulates occupy distinct habitat niches according
to sex, males and females behave like different species for much
of the year (Kie and Bowyer 1999; Bowyer 2004). Sexual seg-
regation in ungulates often is most pronounced during summer
(Festa-Bianchet 1988; Main et al. 1996; Bleich et al. 1997),
especially during and following the birthing period (Cransac
et al. 1998; Bonenfant et al. 2004). Females of alpine ungulate
species, particularly those with young of the year, spend more
time constrained to rugged escape terrain in summer (Rachlow
and Bowyer 1998; Hamel and Côté 2007; Walker et al. 2007),
even when doing so involves sacrificing access to high-quality
forage (Main and Coblentz 1996; Bleich et al. 1997; Rachlow
and Bowyer 1998). In contrast, males are predicted to select
habitats with both high-quality and high-quantity forage to
increase breeding opportunities through amassing larger body
size and secondary sexual characteristics, despite potentially
increased exposure to predation (Main et al. 1996; Bleich et al.
1997; but see Mysterud 2000). Thus, males are expected to use
gentler terrain at lower elevations (Festa-Bianchet 1988; Corti
and Shackleton 2002). Sexual segregation also may be particu-
larly distinct in alpine ungulates during summer because of the
altitudinal cline associated with forage availability in moun-
tainous environments (Albon and Langvatn 1992). During the
time of green-up and peak biomass, forage availability has high
spatial heterogeneity relative to the rest of the year, emphasiz-
ing differences in foraging strategies and capacities between
males and females (Festa-Bianchet 1988; Bischof et al. 2012).
Females are expected to track emerging, high-quality plants
across elevational gradients to extend their access to nutritious
forage (Hebert 1973), whereas males are expected to exploit
areas of lower-quality but higher-quantity forages at lower ele-
vations (Main 2008).
Habitat selection may be modeled via resource selection
functions (RSFs), which define the proportional probability of
use of a given habitat covariate by an organism (Manly et al.
2002). Differential habitat use by sex has been evaluated for
alpine ungulates including bighorn sheep (Ovis canadensis
spp.—Bleich et al. 1997; Festa-Bianchet 1988; Ruckstuhl 1998;
Schroeder et al. 2010), thinhorn sheep (Ovis dalli spp.—Corti
and Shackleton 2002), mouflon sheep (Ovis gmelini—Cransac
et al. 1998), Soay sheep (Ovis aries—Ruckstuhl et al. 2006),
Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949
by guest
on 23 April 2018
Tibetan argali (Ovis ammon hodgsoni—Singh et al. 2010),
mountain goats (Oreamnos americanus—White 2006; Festa-
Bianchet and Côté 2008; Shafer et al. 2012), Cantabrian cham-
ois (Rupicapra pyrenaica parva—Pérez-Barbería et al. 1997),
and alpine chamois (Rupicapra rupicapra—Nesti et al. 2010;
Unterthiner et al. 2012). Habitat selection has been modeled
using RSFs for Stone’s sheep (Ovis dalli stonei), the southern
subspecies of thinhorn sheep (Walker et al. 2007), but not previ-
ously for northern subspecies (Dall’s sheep, Ovis dalli dalli).
Whereas Stone’s sheep models may provide insights into habitat
selection by Dall’s sheep, they were based solely on seasonal
habitat use of females, so differences between sexes cannot be
assessed. Furthermore, the Stone’s sheep habitat measured in
Walker et al. (2007) was at lower latitude, influenced by fire
disturbance, and more forested in contrast to habitat evaluated
in the current study, which is heavily glaciated, contains more
rugged terrain, and extends to higher elevations. Other habitat
studies of Dall’s sheep provide valuable insights into behavior,
seasonal range characteristics, and forage selection and quality
(Hoefs et al. 1975; Hoefs 1984; Frid 1997; Rachlow 1998; Corti
and Shackleton 2002; Weir et al. 2008); however, no studies pre-
viously evaluated habitat selection using RSFs, and few studies
addressed sexual differences in habitat selection more generally.
Our overarching objective was to determine differences in
habitat selection between Dall’s sheep males and females, as
well as between reproductive and nonreproductive females.
Because male and female mountain sheep in general are
thought to overlap rarely except during the breeding season
in late November (Geist 1971), we predicted a high degree of
sexual segregation in Dall’s sheep during summer. We hypothe-
sized that males and females select different habitats because of
differences in strategies to maximize reproduction, nutritional
status, and fitness. Furthermore, we presumed that reproduc-
tive status of females would influence differential selection of
habitat covariates due to variation in nutritional requirements
and predator avoidance. To assess differences in habitat use, we
developed summer RSFs for Dall’s sheep in Wrangell-St. Elias
National Park and Preserve (WRST) for males and females
(with and without young) separately and collectively. We pre-
dicted that males would use a wider range of elevations and
vegetation types, and they would select less steep and rugged
terrain than females. We also predicted that females, in particu-
lar those with young, would sacrifice access to high-quality for-
age in favor of safety (Festa-Bianchet 1988; Bleich et al. 1997;
Rachlow and Bowyer 1998) and therefore they would select
more rugged terrain and more sparsely vegetated areas than
males. Finally, we predicted significant sexual segregation in
Dall’s sheep during summer, which we assessed by testing for
patterns of segregation among male and female sheep among
groups. We also tested for differences in group composition of
females with and without young.
Materials and Methods
Study area.—Dall’s sheep in WRST have been estimated to
number 11,000–14,500 sheep (Schmidt and Rattenbury 2013),
 96 JOURNAL OF MAMMALOGY
equivalent to 15% of the population of this subspecies (Festa-
Bianchet 2008), and occur over approximately 28,000 km2 of
sheep habitat across multiple mountain ranges (Fig. 1). The
major mountain ranges of WRST are distinguished by unique
environmental characteristics and accordingly host varying
densities of sheep, ranging between 0.2 and 1.1 sheep/km2
(Strickland et al. 1992; Terwilliger 2005). The Chugach Range
in the southern portion of the study area (Fig. 1) contains low
sheep densities due to high snowfall influenced by a maritime
climate, whereas the St. Elias and Wrangell Ranges in central
WRST represent a transitional climate zone and support mod-
erate densities of sheep. The Nutzotin Range (Fig. 1) occurs at a
lower elevation and forms the northern portion of the study area,
with a drier and colder interior climate supporting high densi-
ties of sheep (Terwilliger 2005). The extent of the study area
was based on 34 existing sheep count units originally defined
by Alaska Department of Fish and Game in 1949 (Terwilliger
2005). Potential predators of Dall’s sheep in the study area
included grizzly bears (Ursus arctos), wolverines (Gulo gulo),
and golden eagles (Aquila haliaetus) that commonly hunt in
the alpine areas used by sheep, and wolves (Canis lupus), black
bears (U. americanus), and coyotes (C. latrans) that tend to
occur at lower elevations but hunt in alpine areas (Mitchell
et al. 2014).
Survey data.—We used observations of sheep groups
(n = 19,243 sheep in n = 2,877 groups) from aerial surveys con-
ducted during 15 June–8 August, 1983–2011 as sample units to
characterize summer habitat selection. Annual survey data were
collected at irregular intervals by various means including a total
count by helicopter (1983–1984—Singer 1984), fixed-wing
stratified random sampling combined with double sampling via
helicopter (1990–1992—Strickland et al. 1992), fixed-wing cen-
suses of specific count units (1994–2005—Terwilliger 2005),
and fixed-wing distance sampling (2010–2011—Schmidt and
Fig. 1.—Wrangell-St. Elias National Park and Preserve study area and classified sheep locations (n = 2,602) during summer (June–August),
1983–2011. Available habitat is represented by the 7-km radius buffer around all survey locations (n = 2,877).
Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949
by guest
on 23 April 2018
Rattenbury 2013). Observers classified and enumerated sheep
into various sex and age categories including young, females
(juvenile males included with females because they were not
distinguishable from females by horn size and were generally
located with female bands up to 2 years of age—Geist 1971;
Strickland et al. 1992), and males (> 1/4 horn curl). Group
sizes ranged from 1 to 235 and included males (n = 1,018
groups; mean = 3.7 sheep; range = 1–25), females (n = 1,353
groups; mean = 7.8 sheep; range = 1–180), mixed-sex (n = 231
groups; mean = 11.7 sheep; range = 2–155), and unclassified
(n = 275 groups; mean = 6.9 sheep; range = 1–235) groups.
Female groups were larger than male groups by survey year
(t17 = 4.026, P < 0.001). Excluding years when unclassified
groups made up > 50% of observations (1985, 1991, 1992),
the mean proportion of males, females, and young of the total
sheep surveyed each year was fairly consistent (males = 0.23,
SE = 0.02; females = 0.53, SE = 0.01; young = 0.13, SE = 0.01).
All research activities complied with guidelines established by
the American Society of Mammalogists (Sikes et al. 2016).
Segregation by sex and female reproductive class.—We
used the sexual segregation and aggregation statistic (SSAS—
Bonenfant et al. 2007) to test for patterns of segregation among
male and female sheep groups, as well as segregation by female
reproductive class (e.g., with and without young) with respect
to males and to each other, compared to random associa-
tion:
N k
X ×Y
SSAS = 1 - ×å i i (1)
X × Y i =1 N i
where X = the total number of individuals sampled in class I,
Y = the total number of individuals sampled in class II, N = X +
Y, k = the total number of sampled groups, Xi = the number of
class I individuals in the ith group, Yi = the number of class II
 ROFFLER ET AL.—HABITAT SELECTION BY DALL’S SHEEP 97

individuals in the ith group, and Ni = Xi + Yi. Values of SSAS
vary between 0 (complete aggregation) and 1 (complete segre-
gation) and estimate deviation between observed and expected
distributions of males and females. We identified 95% confi-
dence intervals (CIs) of SSAS for expected random association
of sexes with a randomization procedure (10,000 iterations).
Development of habitat selection models.—We estimated
resource selection models and compared selection coefficients
(βi) between sexes and female reproductive classes. We first
used a classification and regression tree with the rpart pack-
age (CART—Breiman et al. 1984) to reveal structure within the
data and select a set of candidate habitat covariates (Table 1;
Supplementary Data SD1). Using these covariates, we then
estimated summer habitat selection of WRST sheep with RSFs
(Boyce and McDonald 1999; Boyce et al. 2002; Manly et al.
2002) using generalized logistic regression (GLM—Hosmer
and Lemeshow 2000) to compare habitat at used and available
sampling locations of each sheep group. We evaluated RSFs
at the 2nd-order scale (landscape—Johnson 1980), and our
design corresponded to surveys with population level informa-
tion about use and availability within seasonal home ranges
(e.g., Manly et al.’s [2002] design I at the population level).
The used–available design we adopted cannot estimate a true
probability of resource selection (the intercept is meaning-
less and determined by the used:available sampling ratio), but
it still yields useful predictions about relative resource selec-
tion (Johnson et al. 2006; Lele et al. 2013). We estimated
coefficients of selection (βi) to determine relative probability of
use with the exponential approximation of the logistic regres-
sion model as:
W = exp(β1 x1 + β2 x2 +  + β p x p ) (2)
where W is an index of probability of use of a given site and
β1 is the selection coefficient of resource variable X1 (Manly
et al. 2002; Johnson et al. 2006; Lele and Keim 2006). We
defined used habitat by group survey locations of classified
sheep (n = 2,602) and available habitat as the area within a
7-km radius buffer established around all survey locations,
including unclassified groups (n = 2,877; Fig. 1). We chose
the 7-km buffer because it represents the radius of a 157-km2
circle, the area estimated to be the mean 95% adaptive kernel
home range size for radiocollared males and females in central
Alaska (Burch and Lawler 2001). We selected points to repre-
sent available habitat randomly at a density of approximately 1
point per 0.73 km2 (mean density of WRST sheep from 1949
to 2002—Terwilliger 2005) to produce 13,200 locations from
which we drew our sample of available points.
We developed RSF models for 1) all sheep, 2) males, and
3) females, with subcategories of a) females with young, and b)
females without young. For each RSF analysis, we accounted
for the number of sheep in the appropriate sex or female repro-
ductive class within each observed group by weighting each
group by the number of: 1) males and females, 2) males,
3) females, 3a) females with young, and 3b) females without

Table 1.—Habitat covariates included in summer resource selection models for Dall’s sheep in Wrangell-St. Elias National Park and Preserve
(WRST), 1983–2011. The 1st set of values in the range column describes habitat defined as available, and the 2nd set of values describes used
habitat.

Variable group Variable name Abbreviation Range/description

Biotic
 Landcovera,b Conifer–hardwood CH Picea, Populus spp.
Tall shrub TSH Alnus, Salix, Betula spp., > 1.5 m
Low shrub LSH Alnus, Salix, Betula spp., 20 cm–1.5 m
Dwarf shrub DSH Ericaceous, Dryas spp., < 20 cm
Herbaceous HERB Graminoids, sedges, forbs
Sparse SPRS Lichens, mosses
Barren BAR Talus, snow, ice
 Normalized difference vegeta- NDVI NDVI 0–1.0/0–0.9
tion index
Abiotic
 Terrain Elevation ELV 442–4,843 m/500–3,247 m
Curvature CRV −11 to 14/−9 to 14
Vector ruggedness measure RUG 0–0.61/0–0.53
Slope SLP 0–84°/0–80°
 Aspect East E 46–135°
North N 315–45°
South S 136–226°
West W 227–315°
Flat F < 1° slope
 Environmental Mean summer precipitation PREC.S 12.7–86.0 cm/12.7–54.9  cm
Mean summer temperature TEMP.S −15.9°C to 13.3°C/−4.3°C to 12.8°C
Mean annual precipitation PREC.A 29.5–275.6 cm/29.5–264.2  cm
Mean annual temperature TEMP.A −9.9°C to 1.1°C/−9.8°C to 1.0°C
Solar radiation SOL 1,040–165,407 WH/m2/1,040–128,450 WH/m2
a
Categorical variable, other variables continuous.
b
Vegetation types in WRST within landcover categories following Viereck et al. (1992).

Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949

by guest
on 23 April 2018
 98 JOURNAL OF MAMMALOGY
young; unclassified groups were not included. To assess non-
linear effects and to identify informative biotic and abiotic
covariates to include in the multivariate models, we first used
univariate logistic regression. We additionally evaluated non-
linear distributions of covariates using semiparametric gener-
alized additive models (GAMs—Hastie and Tibshirani 1990)
because they allow for nonlinear response shapes. A linear
response is indicated when the estimated degrees of freedom
(edf) of the smooth function = 1, whereas edf > 1 is evidence of
a nonlinear response. Nonlinear covariates were incorporated
as polynomials in the GLMs (Hosmer and Lemeshow 2000).
Because our objective was to test for differences in habitat
use among sheep classes, we used a constrained model selec-
tion approach to compare coefficients consistently. Unless the
same set of parameters is used in a model, coefficients are not
comparable because values are adjusted with different combi-
nations of covariates (McCullagh and Nelder 1989; Hosmer
and Lemeshow 2000). Therefore, using covariates identified as
influential, we first constructed a suite of competing a priori
multivariate logistic candidate models for each class to predict
Dall’s sheep habitat selection. Because of the high number of
potentially nonindependent sample units, we used the conser-
vative Bayesian information criterion (BIC—Schwarz 1978)
to rank models based on BIC weights (wi) and calculate evi-
dence ratios (i.e., ratio between BIC weights) to assess strength
of evidence for a given model relative to a competing model
(Anderson and Burnham 2002), with evidence ratio values >
10 indicating strong support. To determine relative importance
of covariates, we ranked them based on the sum of their BIC
weights (∑wi) from all models where the covariate was present
following Burnham and Anderson (2002:167–168). We then
selected a consistent set of the highest ranked habitat covari-
ates to build comparative RSF models for each class. For each
sheep class, we randomly drew a balanced number of available
points (same number as used points) and performed 100 itera-
tions to estimate βi values and SE. We also used generalized
linear mixed effects models to account for the random effect of
survey year and any potential methodological differences over
time. Using BIC model selection criteria, we did not find differ-
ences between the top fixed effects and mixed effects models,
and variance around the random effect (year) was low; thus, we
present fixed effects results.
Habitat covariates.—We conducted spatial and statistical
analyses in the R statistical program (R Project version 3.0.1)
and ArcMap 10 (ESRI 2011). Models included biotic and abi-
otic habitat covariates previously reported to be influential
to Dall’s sheep resource selection (Table 1). Thinhorn sheep
favor rugged terrain, steep slopes, ridges, and mid-elevations
(Geist 1971; Rachlow and Bowyer 1998; Walker et al. 2007);
thus, we considered elevation and topographic covariates. We
resampled GIS data to a 60-m2 cell resolution for analysis. We
derived elevation, slope, curvature, and terrain ruggedness for
each pixel across the study area using a 60-m2 digital eleva-
tion model and Spatial Analyst (ESRI 2011). We calculated the
curvature index, which characterized the overall concavity or
convexity of a pixel in relation to its 3 × 3 pixel neighborhood,
Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949
by guest
on 23 April 2018
with positive values representing convex features such as ridges
and mountain tops, and negative values signifying concave fea-
tures such as valley bottoms. We used the vector ruggedness
measure script for ArcGIS to calculate an index of ruggedness
between 0 and 1 for each point based on vector dispersion of
terrain aspect and angles in the 3 × 3 pixel neighborhood in
comparison to the 3-dimensional vector normal to each focal
pixel (Sappington et al. 2007). Thinhorn sheep favor south-fac-
ing aspects (Summerfield 1974; Walker et al. 2007); thus, we
included aspect and solar radiation, which incorporates slope
and latitude (Keating et al. 2007). We used the Solar Radiation
tool in Spatial Analyst to calculate insolation at 14-day intervals
spanning the summer survey dates. Insolation measures of the
units of watt hours per square meter (WH/m2) were based on
the angle and aspect of the sun at a given latitude and elevation.
Thinhorn sheep generally avoid forested valleys (Walker
et al. 2007), and Dall’s sheep in WRST typically occupy
areas with open vegetation (Terwilliger 2005); therefore, we
included landcover covariates. Landcover classifications were
based on attributes defined using multispectral image analysis
of Landsat TM5 and TM7 data in combination with supporting
field data and made available at the 28.5-m2 pixel resolution by
Stumpf (2008). Vegetation categorizations followed the Alaska
Vegetation Classification system (Viereck et al. 1992). We
used level II vegetation types and collapsed “mixed conifer–
hardwood,” “conifer,” and “hardwood” into 1 category (coni-
fer–hardwood). For models containing categorical covariates,
reference categories were used for comparison and subsumed
into the intercept (Menard 2002); hence, for models containing
landcover, tall shrub was used as a reference category because
it maximized the contrast statistically.
Sheep distribution is influenced by availability of adequate
forage (Hoefs and Cowan 1979), and sheep density estimates
in WRST previously were positively correlated with normal-
ized difference vegetation index (NDVI—Terwilliger 2005).
We used NDVI to characterize relative primary productivity
of forage biomass (Pettorelli et al. 2005; Hebblewhite et al.
2008). We obtained NDVI values using 16-day composites
at a 250-m2 resolution from Moderate Resolution Imaging
Spectroradiometer (MODIS) NASA satellites (Jenkerson et al.
2010). We used the mean maximum annual NDVI value for
each location during 2001–2011 (which, on average, occurred
19 July) to represent Dall’s sheep forages.
Precipitation influences sheep habitat use positively by
increasing primary productivity of summer range (Bunnell
1978; Hoefs and Cowan 1979; Hik and Carey 2000) or nega-
tively by limiting movements and access to forage in winter
(Nichols 1978; Burles and Hoefs 1984; Rachlow and Bowyer
1998). Therefore, we considered effects of both mean annual
and seasonal precipitation and temperature. Mean annual tem-
perature and precipitation were obtained from the PRISM
Climate Group (2009), reflecting the climatological period
during 1971–2000, and we obtained downscaled historical sea-
sonal (June–August) estimates of temperature and precipitation
decadal means from Scenarios Network for Alaska and Arctic
Planning (SNAP 2013) at the 771-m2 pixel spatial scale.
 ROFFLER ET AL.—HABITAT SELECTION BY DALL’S SHEEP 99

We screened for collinearity and did not include significantly Results

correlated habitat covariates (r ≥ 0.7) from a Pearson’s correla-
Sexual segregation.—Males and females were highly segre-
tion matrix (Hosmer and Lemeshow 2000) or with variance
gated among groups (observed SSAS = 0.88, 95% CI of expected
inflation factors (VIF) > 10 (McCullagh and Nelder 1989).
SSAS = 0.57–0.62). When considered separately, females with
We further assessed VIF for the global model (all covariates
young (observed SSAS = 0.94, 95% CI of expected SSAS = 0.42–
included) and combinations of potentially collinear covariates
0.53) and females without young (observed SSAS = 0.87, 95%
(McCullagh and Nelder 1989). We examined collinear vari-
CI of expected SSAS = 0.58–0.62) also displayed high levels of
ables and retained the habitat covariate that explained a greater
segregation from males. Females with and without young were
portion of the deviance. Mean summer temperature and eleva-
segregated among groups (observed SSAS = 0.82, 95% CI of
tion were strongly correlated (r = −0.936), and mean sum-
expected SSAS = 0.54–0.61) compared to random association.
mer temperature had a VIF value of 8.48; thus, we removed it
Habitat selection.—Eight habitat covariates, including biotic
from further consideration. Comparisons of other habitat vari-
(NDVI, open landcover [barren, sparse]), and abiotic (eleva-
ables did not exceed the correlation threshold for exclusion
tion, ruggedness, slope, mean annual precipitation, and mean
from analyses, and VIF values in multivariate models were
summer precipitation) covariates (Tables 2 and 3), were con-
within the acceptable tolerance range. The GAMs indicated
sistently retained in best approximating models for all sheep
nonlinearity for several habitat covariates (NDVI, elevation,
classes. These 8 covariates made up the constrained model,
ruggedness, slope, mean annual precipitation, mean summer
which was the top-ranking model for all sheep, all females,
precipitation) and thus were included as 2nd-order polynomi-
and females without young classes (Table 3). For these sheep
als in the GLMs.
classes, the 2nd-ranked models were 7.62–9.54 ΔBIC units
Model validation.—To assess the predictive capacity of
from the top model, and the BIC weights for top models were
used–available design RSF models, we used a 5-fold cross-
high (wi = 0.98–0.99), providing support for the top model as
validation procedure (Boyce et al. 2002). We reconstructed the
the most parsimonious for these classes. In contrast, the con-
top model for each sheep class using a random 80% subset
strained model ranked 2nd in the male class to a model differ-
of the data (training set) and used the remaining data (20%)
ing by 1 covariate (dwarf shrubs). The dwarf shrub landcover
for evaluation (test set). We compared the area-adjusted fre-
covariate was not significant (β = 0.206, P < 0.167), and the
quencies of test data predictions within 10 ranked bins to
95% CI overlapped 0 (95% CI = −0.03 to 0.44). For males, the
the observed bin rank for each RSF model. A strong posi-
constrained model differed by 8.54 ΔBIC units from the top
tive correlation, measured with Spearman’s rank correlation
model and had low BIC weight (wi = 0.01 compared to 0.98
coefficients (rs), indicated the model had good predictive per-
for the top model). The most parsimonious constrained model
formance based on a higher proportion of used locations in
ranked 3rd for the females with young class. The top model for
the top-ranked RSF bins. To test for differences in resource
this class included elevation as a linear covariate and only con-
selection among sheep classes, we first compared coefficients
tained the sparse landcover covariate (and not barren), whereas
and 95% CIs between sex-specific models. We then assessed
the 2nd-ranked model differed from the constrained model by
the ability of each top model for a particular sheep class to pre-
the inclusion of mean annual temperature.
dict resource selection of all other sheep classes by iteratively
Examination of selection coefficients of categorical land-
calculating rs between the area-adjusted frequencies of each
cover covariates in the global model revealed that sheep selected
model training set and the test set.

Table 2.—Akaike weights (wi) for covariates for each Dall’s sheep class, average weights, and average covariate rank of importance in models
of the relative probability of summer resource selection, Wrangell-St. Elias National Park and Preserve, Alaska, 1983–2011.

Covariatea All Males Females Females with young Females without young Average Akaike weight, wi Average rank
ELV 1 1 1 1 1 1 1
NDVI 1 1 1 1 1 1 1
PREC.A 1 1 1 1 1 1 1
PREC.S 1 1 1 1 1 1 1
RUG 1 1 1 1 1 1 1
SLP 1 1 1 1 1 1 1
SPRS 0.99 0.1 0.99 1 0.01 0.8 2
BAR 0.71 0.99 0.98 0.98 0.01 0.73 3
TEMP.A 0.31 0.97 0.03 0.02 0 0.26 4
LSH 0.28 0.02 0.02 0.02 0 0.07 5
HERB 0.21 0.01 0.02 0.02 0 0.05 6
DSH 0.03 0.01 0.01 0 0 0.01 7
TSH 0.1 0.01 0.01 0.01 0 0.03 7
CH 0 0.01 0 0 0 0 8
a
ELV = elevation (m); PREC.A = mean annual precipitation (cm); PREC.S = mean summer precipitation (June–August, cm); SLP = slope (degrees); RUG = rug-
gedness index; NDVI = normalized difference vegetation index; SPRS = sparse vegetation; BAR = barren vegetation; TEMP.A = mean annual temperature;
LSH = low shrub; HERB = herbaceous; TSH = tall shrub; DSH = dwarf shrub; CH = conifer–hardwood.

Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949

by guest
on 23 April 2018
 100 JOURNAL OF MAMMALOGY

open habitat types and avoided closed landcover types (conifer–
hardwood forests, tall shrubs, and low shrubs; Supplementary
Data SD2). Among the topographic covariates evaluated, rug-
gedness had the highest selection coefficient values (Table 4),
with 95% CI = 5.75–8.47 for the all sheep models. Elevation
had a significant nonlinear effect, with Dall’s sheep largely
selecting mid-elevations (approximately 1,200–2,000 m). Dall’s
sheep selected areas with intermediate NDVI values (0.2–
0.6; Supplementary Data SD3). The probability of selection
decreased linearly among all sheep with increasing mean annual
precipitation (range: β = −0.001 to −0.003, P < 0.001) but had
a nonlinear relationship with summer precipitation, as sheep
disproportionally selected areas between approximately 20 and
40 cm of precipitation per year (Supplementary Data SD3).
Sex-specific and female reproductive class models demon-
strated that whereas the same habitat covariates were impor-
tant predictors for Dall’s sheep summer resource selection,
degree of selection for specific covariates varied among
groups. Rugged terrain was an important habitat covariate for
all sheep; however, females without young selected the most
rugged terrain of all classes (Fig. 2). Elevation was not sig-
nificant (P = 0.097) in the constrained model for females with
young (Table 4). Females with young also had a higher prob-
ability of selecting areas with higher summer precipitation
than other sheep classes (Fig. 2). All sheep classes selected
sparse vegetation, but females had the highest coefficients of
selection (β = 0.407, P < 0.001) and males had the lowest
coefficients (β = 0.150, P < 0.001; Table 4). Despite some
model uncertainty in the male and female with young classes,
internal validation of resource selection models for all sheep
classes using resampled training data performed well in the
k-fold cross validations (rs = 0.951–0.997, P < 0.001; Table 3).
Discussion
Sex classes of Dall’s sheep were highly segregated among
groups but displayed similar patterns of habitat selection at
the landscape scale. In general, topography, open landcover

Table 3.—Fit of the constrained model of the relative probability of summer resource selection by each Dall’s sheep class in Wrangell-St. Elias
National Park and Preserve, Alaska, 1983–2011. BIC = Bayesian information criterion.

Group Ka BICb ΔBICc wid Model ranke Evidence ratiof Ng rsh

All sheep 14 82,588.45 0 0.99 1 125.64 15,387 0.997
Males 14 28,156.30 8.54 0.01 2 0.01 14,449 0.988
All Females 14 57,197.02 0 0.98 1 99.00 14,776 0.988
Females w/ young 14 49,826.80 47.73 0.27 3 0.79 14,069 0.988
Females w/out young 14 16,586.85 0 0.98 1 99.78 13,917 0.951
a
Number of model parameters.
b
BIC = −2 LL + K * log n, where n is the sum of used and available locations.
c
ΔBIC = BICi − min BIC (model that minimizes BIC), presented for the top model relative to competing models in each group (not comparable among groups).
d
BIC weight = percentage of total weight from all candidate models that can be attributed to the specific model;

∑
R
wi = e(− i BICi ) r
e(− i BICi )
e
Model rank among candidate models based on ΔBIC.
f
Ratio between BIC weights for the top model in a group and the next highest ranked competing model.
g
Number of groups included in the model.
h
Spearman’s rank correlation coefficient for k-folds procedure averaged from 5 partitions.

Table 4.—Parameter estimates presented as selection coefficients (β1) and SEs for covariatesa in the summer resource selection function model
for Dall’s sheep in Wrangell-St. Elias National Park and Preserve, Alaska, 1983–2011. Where 95% CI did not overlap 0, β1 values are in bold.

Covariate All sheep Males All females Females with young Females without young
β1 ± SE β1 ± SE β1 ± SE β1 ± SE β1 ± SE
ELV 1.00 × 10−3 ± 1.00 × 10−4 2.00 × 10−3 ± 1.00 × 10−4 1.00 × 10−3 ± 1.34 × 10−5 1.00 × 10−3 ± 1.78 × 10−5 4.00 × 10−3 ± 1.00 × 10−3
ELV2 1.00 × 10−4 ± 8.61 × 10−6 7.31 × 10−6 ± 1.78 × 10−5 1.00 × 10−4 ± 3.17 × 10−9 1.00 × 10−4 ± 4.32 × 10−9 1.00 × 10−4 ± 2.83 × 10−5
PREC.A −2.00 × 10−3 ± 2.06 × 10−9 −1.00 × 10−3 ± 4.50 × 10−9 −2.00 × 10−3 ± 6.67 × 10−6 −2.00 × 10−3 ± 8.94 × 10−6 −3.00 × 10−3 ± 7.26 × 10−9
SLP 5.80 × 10−2 ± 3.00 × 10−3 5.60 × 10−2 ± 6.00 × 10−3 4.50 × 10−2 ± 4.00 × 10−3 3.90 × 10−2 ± 4.00 × 10−3 7.90 × 10−2 ± 9.00 × 10−3
SLP2 1.00 × 10−4 ± 1.00 × 10−4 1.00 × 10−4 ± 2.81 × 10−4 1.80 × 10−4 ± 3.00 × 10−6 1.07 × 10−4 ± 3.73 × 10−6 1.12 × 10−4 ± 5.52 × 10−6
RUG 0.71 × 101 ± 2.75 × 10−1 0.41 × 101 ± 5.38 × 10−1 0.77 × 101 ± 3.44 × 10−1 0.85 × 101 ± 3.86 × 10−1 0.52 × 101 ± 6.69 × 10−1
RUG2 −1.81 × 101 ± 8.09 × 10−1 −1.17 × 101 ± 0.16 × 101 −1.91 × 101 ± 0.10 × 101 −2.20 × 101 ± 0.12 × 101 −1.15 × 101 ± 0.18 × 101
BAR 3.40 × 10−2 ± 2.60 × 10−2 7.40 × 10−2 ± 5.10 × 10−2 1.12 × 10−1 ± 3.30 × 10−2 4.80 × 10−2 ± 3.70 × 10−2 3.84 × 10−1 ± 7.10 × 10−2
SPRS 3.02 × 10−1 ± 2.80 × 10−2 1.50 × 10−1 ± 5.50 × 10−2 4.07 × 10−1 ± 3.50 × 10−2 4.06 × 10−1 ± 3.80 × 10−2 3.72 × 10−1 ± 7.50 × 10−2
PREC.S 3.30 × 10−2 ± 1.00 × 10−3 3.00 × 10−2 ± 3.00 × 10−3 2.70 × 10−2 ± 1.00 × 10−3 2.70 × 10−2 ± 1.00 × 10−3 5.00 × 10−2 ± 4.00 × 10−3
PREC.S2 1.00 × 10−4 ± 1.00 × 10−4 2.35 × 10−7 ± 2.08 × 10−7 1.00 × 10−6 ± 1.45 × 10−7 1.22 × 10−7 ± 1.63 × 10−7 1.02 × 10−6 ± 4.92 × 10−7
NDVI 0.83 × 101 ± 1.28 × 10−1 0.76 × 101 ± 2.61 × 10−1 0.78 × 101 ± 1.61 × 10−1 0.80 × 101 ± 1.76 × 10−1 0.68 × 101 ± 3.59 × 10−1
NDVI2 −0.79 × 101 ± 1.53 × 10−1 −0.70 × 101 ± 3.06 × 10–1 −0.75 × 101 ± 1.95 × 10–1 −0.75 × 101 ± 2.12 × 10–1 −0.63 × 101 ± 4.43 × 10−1
a
ELV = elevation (m); PREC.A = mean annual precipitation (cm); PREC.S = mean summer precipitation (June–August, cm); SLP = slope (degrees); RUG = rug-
gedness index; NDVI = normalized difference vegetation index; SPRS = sparse vegetation; BAR = barren vegetation; LSH = low shrub; HERB = herbaceous;
TSH = tall shrub; DSH = dwarf shrub; CH = conifer–hardwood.

Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949

by guest
on 23 April 2018
 ROFFLER ET AL.—HABITAT SELECTION BY DALL’S SHEEP 101

Fig. 2.—Relative probability of Dall’s sheep resource selection by classes (all sheep, males, females with young [FwY], and females without
young [FwoY]), calculated using the summer resource selection function model, Wrangell-St. Elias National Park and Preserve, 1983–2011.

Fig. 3.—a) Predicted relative probability of use values for Dall’s sheep summer resource selection functions. b) Sample map with classified aerial
survey locations (n = 2,602), 1983–2011, Wrangell-St. Elias National Park and Preserve, Alaska.

Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949

by guest
on 23 April 2018
 102 JOURNAL OF MAMMALOGY
types, and intermediate values of NDVI were the most impor-
tant habitat resources for Dall’s sheep. We conclude that at the
landscape scale, a combined RSF model may be used to predict
the relative probability of resource selection by Dall’s sheep
during summer, despite the importance of sexual segregation at
the group level, because habitat selection was largely consistent
among sex and reproductive status classes.
Despite the overall similarity in resource selection between
sexes, selection of certain habitat covariates differed somewhat
among classes. Sex and reproductive status of females affected
selection of rugged habitat, as females with young had a higher
probability of selecting intermediate rugged terrain, whereas
females without young selected for a broader range of rugged-
ness values. This pattern may be due to reproductive females
selecting the optimal habitat for predator evasion, as intermedi-
ate rugged terrain may provide sufficient safety from preda-
tors and simultaneously maneuverable topography for young.
The “trade-off” phenomenon wherein reproductive females
sacrifice forage quantity for safety from predators has been
documented in Dall’s sheep and other ungulates (Rachlow and
Bowyer 1998; Main and Coblentz 1996; Corti and Shackleton
2002). Our results indicate that although reproductive females
selected sparse vegetation categories, within those areas they
selected higher-quality forage than nonreproductive females or
males, reflected by higher selection coefficients for interme-
diate values of NDVI. NDVI is inversely correlated to forage
quality over the duration of the growing season and positively
associated with forage biomass (Hebblewhite et al. 2008); thus
intermediate values are generally associated with the high-
est amount of digestible biomass during mid-summer. Also,
intermediate NDVI values may better predict open, sparsely
vegetated habitats utilized by sheep than the landcover classifi-
cation from satellite imagery we had available. Consequently,
we cannot conclude that reproductive Dall’s sheep females in
our study area forgo high-quality forage in favor of safety. They
seem to select high forage quality and avoid predation risk at
the same time, which may suggest risk and forage are not cor-
related at this landscape scale.
The probability of selection for all sheep classes increased in
areas of relatively high summer precipitation, but in particular
for females with young. Dall’s sheep also consistently selected
areas of lower mean annual precipitation, a pattern indicating a
complex relationship with precipitation that varies throughout
the year. In summer, precipitation translates to elevated pri-
mary productivity, which is beneficial to sheep (Bunnell 1978;
Hoefs and Cowan 1979; Hik and Carey 2000), whereas during
winter, precipitation in the form of snow limits food accessi-
bility and inhibits movement, with a negative effect on sheep
(Nichols 1978; Burles and Hoefs 1984; Rachlow and Bowyer
1998). Sheep have strong fidelity to seasonal ranges (Geist
1971), and winter range generally consists of a smaller subset
of summer range (Hoefs and Cowan 1979). It is uncommon for
sheep to disperse or to make large movements between seasonal
ranges (Geist 1971; Hoefs and Cowan 1979; Festa-Bianchet
1991). Thus, individuals appear to select areas within their
annual range representing favorable conditions throughout the
Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949
by guest
on 23 April 2018
year. Precipitation appears to be an important driver of Dall’s
sheep distribution, and further investigations of winter habitat
use, in particular measurements of snow depth and consistency
at the spatial scale of daily and seasonal habitat selection, would
provide insight into limiting effects of winter precipitation.
The top-ranked model revealed that many landscape features
are important to Dall’s sheep, with topography, vegetation,
and precipitation the most explanatory for resource selection.
Similar to Stone’s sheep (Walker et al. 2007), Dall’s sheep
select some terrain features consistently but appear to express
plasticity in selection of habitat attributes within their seasonal
range, possibly serving as an adaptive strategy. Results of this
study suggest that terrain features, in particular the rugged-
ness index, played a large role in predicting the probability
of sheep habitat use. This landscape feature will remain static
despite changes in climactic conditions, which is favorable for
long-term persistence of the species. However, abiotic covari-
ates such as landcover and precipitation will change over time.
Climate models predict a shift from open vegetation classes in
alpine ecosystems to an upward expansion in shrub communi-
ties both in elevation and in latitude (Myers-Smith et al. 2011).
As evidenced by this study, Dall’s sheep avoid tall shrubs, and
thus an increase in extent of this vegetation community could
result in reduction and fragmentation of habitat. Maintenance
of population connectivity among favorable habitats is crucial
for preserving genetic diversity (Epps et al 2006); thus, pre-
dicted changes in habitat fragmentation are an important con-
sideration for long-term population viability. However, the
simultaneous receding of glaciers at the upper elevational lim-
its currently used by Dall’s sheep will result in expansion of
habitat, although the net habitat gain will depend on rates of
shrub expansion and glacier recession.
Resource selection may depend on scale (Bowyer et al. 1996;
DeCesare et al. 2012), because ecological processes often func-
tion at different spatial and temporal scales. Our analysis tested
differences in habitat selection by sex and reproductive class
of mountain sheep at the landscape scale, i.e., a broader spa-
tial scale than previous investigations, and thus provides new
knowledge on this topic. However, we cannot rule out sex-based
differences in resource selection at finer spatial and temporal
scales. Previous work based on more fine-scale observations
indicated different habitat use by males and females (Bleich
et al. 1997; Corti and Shackleton 2002; Schroeder et al. 2010).
We used the 2nd-order scale (seasonal home ranges within a
population range—Johnson 1980) to assess RSFs and found
that at this scale, differences in resource selection patterns
between sexes were not pronounced. Comparing our results to
previous work based on fine-scale observations suggests that
habitat-based theories of sexual segregation that character-
ize males and females as occupying different realized habitat
niches are likely more operational at smaller spatial scales.
Fine-scale location or habitat attribute information, such as that
from GPS collar-based studies or direct habitat assessments,
also might reveal greater differences in habitat selection, as
probability of use can vary across spatial scales. For manage-
ment and ecological investigations at the landscape level, our
 ROFFLER ET AL.—HABITAT SELECTION BY DALL’S SHEEP 103
work is among the first such study of Dall sheep, and our results
at the 2nd-order scale show minimal habitat selection conse-
quences of sexual segregation.
Differences in daily activity patterns, preferences for individu-
als to aggregate with their own sex, and avoidance of intrasex-
ual aggression can lead to sexual segregation in the absence of
any clear differences in habitat selection between sexes. Strong
sexual segregation was the major distinguishing characteristic of
Dall’s sheep rather than clear differences among sheep classes
in use of habitat features. Despite little mixing at the group level
between sexes, males and females selected the same types of
habitats during summer when differences in habitat selection
due to varying reproductive strategies or digestive capacities are
expected to be most distinct. Therefore, the ecological processes
that result in sexual segregation in Dall’s sheep appear to be more
consistent with social rather than habitat factors.
Evident segregation and limited differences in habitat selec-
tion among sexes in mountain sheep may be a consequence of
their evolution in alpine habitats. The dispersal theory of Geist
(1971) suggests that because sheep are habitat specialists (i.e.,
they require escape terrain with open landcover), and suitable
habitats are patchy and small in comparison to the surrounding
uninhabitable low-elevation forests, there is little evolutionary
advantage to dispersal. As a result, mountain sheep inherit their
home ranges through learned behavior, are gregarious through-
out the year, and have a complex social structure. Moose (Alces
alces), in contrast, are highly mobile with a marked ability
to disperse (Geist 1971) and have the ability to exploit more
diverse habitat niches in heterogeneous forests and early suc-
cessional vegetation. Thus, habitat segregation documented
in cervids (e.g., Oehlers et al. 2011) could be influenced by
the variety of habitats these species may exploit, in contrast to
ovids, for which habitat segregation has been documented less
often (Cransac et al. 1998; Ruckstuhl 1998; Singh et al. 2010)
but for which sexual segregation is common.
Understanding patterns of habitat use based on sex or repro-
ductive status can have far-reaching implications for manage-
ment and conservation efforts (Bleich et al 1997; Bowyer 2004;
Rubin and Bleich 2005). Yet, results of this study indicate that
a combined RSF model is robust for predicting the relative
probability of Dall’s sheep resource selection and may be used
for modeling connectivity at landscape scales (Fig. 3), and the
advantages of developing separate sex-specific or reproductive
status-specific models are limited.
Supplementary Data
Supplementary Data SD1.—Classification and regression
tree methods.
Supplementary Data SD2.—Selection coefficients of categorical
landcover coefficients in the Dall’s sheep global summer resource
selection function model presented by classes (males, females,
females with young, and females without young), Wrangell-St.
Elias National Park and Preserve, Alaska, 1983–2011.
Supplementary Data SD3.—Distributions of continuous habi-
tat covariates used during summer by male (blue) and female
Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949
by guest
on 23 April 2018
(green) Dall’s sheep compared to available (pink) habitat,
Wrangell-St. Elias National Park and Preserve, Alaska, 1983–
2011. Distributions are smoothed using kernel density estimation.
Acknowledgments
This work was supported by funding from the United States
Geological Survey, the National Park Service, and the University
of Montana. The National Park Service and the Alaska Department
of Fish and Game, Wildlife Conservation Division provided sur-
vey data and valuable insights. We especially thank J. Lawler,
J. Putera, R. Schwanke, and M. Terwilliger. D. Gregovich,
D. Gustine, J. Pearce, C. Bonenfant, and K. White provided con-
structive criticism of previous drafts of this work. Any use of
trade, firm, or product names is for descriptive purposes only and
does not imply endorsement by the U.S. Government.
Literature Cited
Albon, S. D., and R. Langvatn. 1992. Plant phenology and the ben-
efits migration in a temperate ungulate. Oikos 65:502–513.
Anderson, D. R., and K. P. Burnham. 2002. Avoiding pitfalls
when using information-theoretic methods. Journal of Wildlife
Management 66:912–918.
Barboza, P. S., and R. T. Bowyer. 2000. Sexual segregation in dimor-
phic deer: a new gastrocentric hypothesis. Journal of Mammalogy
81:473–489.
Bischof, R., L. E. Loe, E. L. Meisingset, B. Zimmermann, B. Van
Moorter, and A. Mysterud. 2012. A migratory northern ungu-
late in the pursuit of spring: jumping or surfing the green wave?
American Naturalist 180:407–424.
Bleich, V. C., R. T. Bowyer, and J. D. Wehausen. 1997. Sexual
segregation in mountain sheep: resources or predation? Wildlife
Monographs 134:3–50.
Bon, R., and R. Campan. 1996. Unexplained sexual segregation in
polygamous ungulates: a defense of an ontogenetic approach.
Behavioural Processes 38:131–154.
Bonenfant, C., et al. 2004. Multiple causes of sexual segregation in
European red deer: enlightenments from varying breeding phenol-
ogy at high and low latitude. Proceedings of the Royal Society of
London, B. Biological Sciences 271:883–892.
Bonenfant, C., J. Gaillard, S. Dray, A. Loison, M. Royer, and
D. Chessel. 2007. Testing sexual segregation and aggregation: old
ways are best. Ecology 88:3202–3208.
Bowyer, R. T. 1984. Sexual segregation in southern mule deer.
Journal of Mammalogy 65:410–417.
Bowyer, R. T. 2004. Sexual segregation in ruminants: definitions,
hypotheses, and implications for conservation and management.
Journal of Mammalogy 85:1039–1052.
Bowyer, R. T., and J. G. Kie. 2004. Effects of foraging activity on sex-
ual segregation in mule deer. Journal of Mammalogy 85:498–504.
Bowyer, R. T., J. G. Kie, and V. Van Ballenberghe. 1996. Sexual seg-
regation in Black-tailed deer: effects of scale. Journal of Wildlife
Management 60:10–17.
Boyce, M. S., and L. L. McDonald. 1999. Relating populations to
habitats using resource selection functions. Trends in Ecology &
Evolution 14:268–272.
Boyce, M. S., P. R. Vernier, S. E. Nielsen, and F. K. Schmiegelow.
2002. Evaluating resource selection functions. Ecological
Modelling 157:281–300.
 104 JOURNAL OF MAMMALOGY
Breiman, L., J. H. Friedman, R. Olshen, and C. J. Stone. 1984.
Classification and regression trees. Wadsworth International
Group, Belmont, California.
Bunnell, F. L. 1978. Horn growth and population quality in Dall
sheep. Journal of Wildlife Management 42:764–775.
Burch, J., and J. Lawler. 2001. Ecology and demography of Dall’s
sheep in Yukon-Charley Rivers National Preserve: identifying criti-
cal Dall’s sheep habitat and habitat use patterns. Yukon-Charley
Rivers National Preserve, Alaska. National Park Service, Technical
Report NPS/AR/NRTR-2001/39:1–82.
Burles, D. W., and M. Hoefs. 1984. Winter mortality of Dall sheep,
Ovis dalli dalli, in Kluane National Park, Yukon. Canadian Field-
Naturalist 98:479–484.
Burnham, K.P., and D. R. Anderson. 2002. Model selection and
multi-model inference: a practical information-theoretic approach.
2nd ed. Springer, Heidelberg, Germany.
Clauss, M., A. Schwarm, S. Ortmann, W. Streich, and J. Hummel.
2007. A case of non-scaling in mammalian physiology? Body
size, digestive capacity, food intake, and ingesta passage in mam-
malian herbivores. Comparative Biochemistry and Physiology,
A. Molecular and Integrative Physiology 148:249–265.
Conradt, L. 2005. Definitions, hypotheses, models and measures in the
study of animal segregation. Pp. 11–32 in Sexual segregation in ver-
tebrates: ecology of the two sexes (K. E. Ruckstuhl and P. Nehaus,
eds.). Cambridge University Press, Cambridge, United Kingdom.
Corti, P., and D. M. Shackleton. 2002. Relationship between preda-
tion-risk factors and sexual segregation in Dall’s sheep (Ovis dalli
dalli). Canadian Journal of Zoology 80:2108–2117.
Cransac, N., J.-F. Gerard, M. L. Maublanc, and D. Pepin. 1998. An
example of segregation between age and sex classes only weakly
related to habitat use in mouflon sheep (Ovis gmelini). Journal of
Zoology 244:371–378.
DeCesare, N. J., et  al. 2012. Transcending scale dependence in
identifying habitat with resource selection functions. Ecological
Applications 22:1068–1083.
Epps, C. W., P. J. Palsbøll, J. D. Wehausen, G. K. Roderick, and D.
R.  McCullough. 2006. Elevation and connectivity define genetic
refugia for mountain sheep as climate warms. Molecular Ecology
15:4295–4302.
ESRI (Environmental Systems Research Institiute). 2011. ArcGIS
Desktop: Release 10. Redlands, California.
Festa-Bianchet, M. 1988. Seasonal range selection in bighorn sheep:
conflicts between forage quality, forage quantity, and predator
avoidance. Oecologia 75:580–586.
Festa-Bianchet, M. 1991. The social system of bighorn sheep:
grouping patterns, kinship and female dominance rank. Animal
Behaviour 42:71–82.
Festa-Bianchet, M., and S. D. Côté. 2008. Mountain goats: ecol-
ogy, behavior, and conservation of an alpine ungulate. Island Press,
Washington, D.C.
Frid, A. 1997. Vigilance by female Dall’s sheep: interactions between
predation risk factors. Animal Behaviour 53:799–808.
Geist, V. 1971. Mountain sheep. University of Chicago Press,
Chicago, Illinois.
Hamel, S., and S. D. Côté. 2007. Habitat use patterns in relation to
escape terrain: are alpine ungulate females trading off better forag-
ing sites for safety? Canadian Journal of Zoology 85:933–943.
Hastie, T., and R. J. Tibshirani. 1990. Generalized additive models.
Chapman and Hall, London, United Kingdom.
Hebblewhite, M., E. Merrel, and G. McDermid. 2008. A multi-scale
test of the forage maturation hypothesis in a partially migratory
ungulate population. Ecological Monographs 78:141–166.
Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949
by guest
on 23 April 2018
Hebert, D. M. 1973. Altitudinal migration as a factor in the nutri-
tion of bighorn sheep. Ph.D. dissertation, University of British
Columbia, Vancouver, British Columbia, Canada.
Hik, D. S., and J. Carey. 2000. Cohort variation in horn growth of
Dall sheep rams in the southwest Yukon, 1969–1999. Pp. 88–100
in Proceedings of the Biennial Symposium of the Northern Wild
Sheep and Goat Council, Whitehorse, Yukon, 31 May–4 June 2000
(J. Carey, ed.). Northern Wild Sheep and Goat Council, Wyoming
Game and Fish Department, Cody.
Hoefs, M. 1984. Productivity and carrying capacity of a subarctic
sheep winter range. Arctic 37:141–147.
Hoefs, M., and I. M. Cowan. 1979. Ecological investigations of a
population of Dall sheep (Ovis dalli dalli Nelson). Syesis 12:1–81.
Hoefs, M., I. M. Cowan, and V. J. Krajina. 1975. Phytosociological
analysis and synthesis of Sheep Mountain, southwest Yukon
Territory, Canada. Syesis 8:125–228.
Hosmer, D. W., and S. Lemeshow. 2000. Applied logistic regression.
2nd ed. John Wiley and Sons, New York.
Jenkerson, C. B., T. Maiersperger, and G. Schmidt. 2010. eMODIS:
a user-friendly data source. U.S. Geological Survey Open-File
Report 2010-1055:1–10.
Johnson, D. H. 1980. The comparison of usage and availability mea-
surements for evaluating resource preference. Ecology 61:65–71.
Johnson, C. J., S. E. Nielsen, E. H. Merril, T. L. McDonald, and M.
S. Boyce. 2006. Resource selection functions based on use – avail-
ability data: theoretical motivation and evaluation methods. Journal
of Wildlife Management 70:347–357.
Keating, K. A., P. J. P. Gogan, J. M. Vore, and L. R. Irby. 2007. A
simple solar radiation index for wildlife habitat studies. Journal of
Wildlife Management 71:1344–1348.
Kie, J. G., and R. T. Bowyer. 1999. Sexual segregation in white-tailed
deer: density-dependent changes in use of space, habitat selection,
and dietary niche. Journal of Mammalogy 80:1004–1020.
Lele, S. R., and J. L. Keim. 2006. Weighted distributions and estimation
of resource selection probability functions. Ecology 87:3021–3028.
Lele, S. R., E. H. Merrill, J. Keim, and M. S. Boyce. 2013. Selection,
use, choice and occupancy: clarifying concepts in resource selec-
tion studies. Journal of Animal Ecology 82:1183–1191.
Main, M. B., and B. E. Coblentz. 1996. Sexual segregation in Rocky
Mountain mule deer. Journal of Wildlife Management 60:497–507.
Main, M. B., F. W. Weckerly, and V. C. Bleich. 1996. Sexual seg-
regation in ungulates: new directions for research. Journal of
Mammalogy 77:449–461.
Main, M. B. 2008. Reconciling competing ecological explanations
for sexual segregation in ungulates. Ecology 89:693–704.
Manly, B. F. J., L. L. McDonald, D. L. Thomas, T. L. McDonald,
and W. P. Erickson. 2002. Resource selection by animals: statisti-
cal design and analysis for field studies. 2nd ed. Kluwer, Boston,
Massachusetts.
McCullagh, P., and J. A. Nelder. 1989. Generalized linear models.
2nd ed. Chapman and Hall, London, United Kingdom.
Menard, S. 2002. Applied logistic regression analysis. 2nd ed. Sage
Publications, London, United Kingdom. Vol. 106.
Mitchell, C. D., R. Chaney, K. Aho, J. G. Kie, and R. T. Bowyer.
2014. Population density of Dall’s sheep in Alaska: effects of pred-
ator harvest? Mammal Research 60:21–28.
Myers-Smith, I. H., et  al. 2011. Shrub expansion in tundra ecosys-
tems: dynamics, impacts and research priorities. Environmental
Research Letters 6:045509.
Mysterud, A. 2000. The relationship between ecological segregation
and sexual body size dimorphism in large herbivores. Oecologia
124:40–54.
 ROFFLER ET AL.—HABITAT SELECTION BY DALL’S SHEEP 105
Nesti, I., M. Posillico, and S. Lovari. 2010. Ranging behaviour
and habitat selection of Alpine chamois. Ethology Ecology and
Evolution 22:215–231.
Neuhaus, P., K. E. Ruckstuhl, and L. Conradt. 2005. Conclusions
and future directions. Pp. 395–402 in Sexual segregation in ver-
tebrates: ecology of the two sexes (K. Ruckstuhl and P. Neuhaus,
eds.). Cambridge University Press, Cambridge, United Kingdom.
Nichols, L. 1978. Dall sheep reproduction. Journal of Wildlife
Management 42:570–580.
Oehlers, S. A., R. T. Bowyer, F. Huettmann, D. K. Person, and W.
B. Kessler. 2011. Sex and scale: implications for habitat selection
by Alaskan moose Alces alces gigas. Wildlife Biology 17:67–84.
Pérez-Barbería, F. J., M. Olivan, K. Osoro, and C. Nores. 1997.
Sex, seasonal and spatial differences in the diet of Cantabrian
chamois Rupicapra pyrenaica parva. Acta Theriologica 42:37–46.
Pérez-Barbería, F. J., E. Robertson, R. Soriguer, A. Aldezabal,
M.  Mendizabal, and E. Pérez-Fernández. 2007. Why do polyg-
ynous ungulates segregate in space? Testing the activity budget
hypothesis in Soay sheep. Ecological Monographs 77:631–647.
Pettorelli, N., J. O. Vik, A. Mysterud, J.-M. Gaillard, C. J. Tucker,
and N. C. Stenseth. 2005. Using the satellite-derived NDVI to
assess ecological responses to environmental change. Trends in
Ecology and Evolution 20:503–510.
PRISM Climate Group, Oregon State University. http://prism.oregon-
state.edu. Accessed 1 December 2009.
Rachlow, J. L., and R. T. Bowyer. 1998. Habitat selection by
Dall’s sheep (Ovis dalli): maternal trade-offs. Journal of Zoology
245:457–465.
Rubin, E. S., and V. C. Bleich. 2005. Sexual segregation: a necessary con-
sideration in wildlife conservation. Pp. 379–391 in Sexual segregation
in vertebrates: ecology of the two sexes (K. Ruckstuhl and P. Neuhaus,
eds.). Cambridge University Press, Cambridge, United Kingdom.
Ruckstuhl, K. 1998. Foraging behaviour and sexual segregation in
bighorn sheep. Animal Behaviour 56:99–106.
Ruckstuhl, K. E. 2007. Sexual segregation in vertebrates: proximate and
ultimate causes. Integrative and Comparative Biology 47:245–257.
Ruckstuhl, K. E., A. Manica, A. D. C. MacColl, J. G. Pilkington,
and T. H. Clutton-Brock. 2006. The effects of castration, sex
ratio and population density on social segregation and habitat use
in Soay sheep. Behavioral Ecology and Sociobiology 59:694–703.
Ruckstuhl, K. E., and P. Neuhaus. 2002. Sexual segregation in ungu-
lates: a comparative test of three hypotheses. Biological Reviews of
the Cambridge Philosophical Society 77:77–96.
Sappington, J. M., K. M. Longshore, and D. B. Thompson. 2007.
Quantifying landscape ruggedness for animal habitat analysis: a
case study using bighorn sheep in the Mojave Desert. Journal of
Wildlife Management 71:1419–1426.
SNAP. 2013. Historical monthly and derived precipitation and tem-
perature products. University of Alaska. http://www.snap.uaf.edu/
data.php. Accessed 1 June 2013.
Schmidt, J. H., and K. L. Rattenbury. 2013. Reducing effort while
improving inference: Estimating Dall’s sheep abundance and
composition in small areas. Journal of Wildlife Management
77:1048–1058.
Schroeder, C. A., R. T. Bowyer, V. C. Bleich, and T. R. Stephenson.
2010. Sexual segregation in Sierra Nevada Bighorn sheep,
Ovis canadensis sierrae: ramifications for conservation. Arctic,
Antarctic, and Alpine Research 42:476–489.
Schwarz, G. 1978. Estimating the dimension of a model. Annals of
Statistics 6:461–464.
Downloaded from https://academic.oup.com/jmammal/article-abstract/98/1/94/2628949
by guest
on 23 April 2018
Shafer, A. B. A., J. M. Northrup, K. S. White, M. S. Boyce, S.
D.  Côté, and D. W. Coltman. 2012. Habitat selection predicts
genetic relatedness in an alpine ungulate. Ecology 93:1317–1329.
Sikes, R. S., and The Animal Care and Use Committee of the
American Society of Mammalogists. 2016. Guidelines of the
American Society of Mammalogists for the use of wild mammals
in research and education. Journal of Mammalogy 97:663–688.
Singer, F. J. 1984. Some population characteristics of Dall sheep in six
Alaska National Parks and Preserves. Pp. 1–10 in Proceedings of the
Biennial Symposium of the Northern Wild Sheep and Goat Council,
Whitehorse, Yukon, 30 April 1984 (M. Hoefs, ed.). Northern Wild
Sheep and Goat Council, Wyoming Game and Fish Department, Cody.
Singh, N. J., C. Bonenfant, N. G. Yoccoz, and S. D. Côté. 2010. Sexual
segregation in Eurasian wild sheep. Behavioral Ecology 21:410–418.
Strickland, D., L. McDonald, K. Taylor, K. Jenkins, and J. Kern.
1992. Estimation of Dall sheep numbers in the Wrangell-St.
Elias National Park and Preserve. Pp. 237–255 in Proceedings of
the Biennial Symposium of the Northern Wild Sheep and Goat
Council, Cody, Wyoming, 27 April–1 May 1992 (J. Emmerich and
W. G. Hepworth, eds.). Northern Wild Sheep and Goat Council,
Wyoming Game and Fish Department, Cody.
Stumpf, K. 2008. Wrangell-St. Elias National Park and Preserve land-
cover mapping project. National Park Service, Natural Resource
Technical Report NPS/WRST/NRTR—2008/095:1–118.
Summerfield, B. L. 1974. Population dynamics and seasonal movement
patterns of Dall sheep in the Atigun Canyon area, Brooks Range,
Alaska. Ph.D. dissertation, University of Alaska, Fairbanks, Alaska.
Terwilliger, M. L. N. 2005. Population and habitat analyses for
Dall’s sheep (Ovis dalli dalli) in Wrangell-St. Elias National Park
and Preserve. M.S. thesis, University of Alaska, Fairbanks, Alaska.
Unterthiner, S., F. Ferretti, L. Rossi, and S. Lovari. 2012. Sexual
and seasonal differences of space use in Alpine chamois. Ethology
Ecology and Evolution 24:257–274.
Van Soest, P. J. 1982. Nutritional ecology of the ruminant. O & B
Books, Inc., Corvallis, Oregon.
Viereck, L. A., C. T. Dyrness, A. R. Batten, and K. J. Wenzlick.
1992. The Alaska vegetation classification. U.S. Forest Service,
General Technical Report PNW-GTR-286:1–278.
Walker, A. B. D., K. L. Parker, M. P. Gillingham, D. D. Gustine, and
R. J. Lay. 2007. Habitat selection by female Stone’s sheep in relation
to vegetation, topography, and risk of predation. Ecoscience 14:55–70.
Weckerly, F. W. 2001. Are large male Roosevelt elk less social
because of aggression? Journal of Mammalogy 82:414–421.
Weir, J. N., S. F. Morrison, and D. S. Hik. 2008. Linking foraging
behavior to population density: an assessment of GMM models for
Dall sheep. Ecological Modelling 211:396–402.
White, K. S. 2006. Seasonal and sex-specific variation in terrain use
and movement patterns of mountain goats in Southeastern Alaska.
Pp. 183–194 in Proceedings of the Biennial Symposium Northern
Wild Sheep and Goat Council, Kananaskis, Alberta, 2–6 April
2006 (B. Wishart and M. Pybus, eds.). Northern Wild Sheep and
Goat Council, Wyoming Game and Fish Department, Cody.
Yearsley, J. M., and F. J. Pérez-Barbería. 2005. Does the activity
budget hypothesis explain sexual segregation in ungulates? Animal
Behaviour 69:257–267.
Submitted 26 February 2016. Accepted 26 July 2016.
Associate Editor was Christine Maher.