multiscale models of habitat use by mule deer in winter · research article multiscale models of...

Research Article

Multiscale Models of Habitat Use by MuleDeer in Winter

PRISCILLA K. COE,1 Oregon Department of Fish and Wildlife, 1401 Gekeler Lane, La Grande, OR 97850, USA

DARREN A. CLARK,2 Oregon Department of Fish and Wildlife, 1401 Gekeler Lane, La Grande, OR 97850, USA

RYAN M. NIELSON,3 Western EcoSystems Technology, 415W. 17th Street, Suite 200, Cheyenne, WY 82001, USA

SARA C. GREGORY, Oregon Department of Fish and Wildlife, 61374 Parrell Road, Bend, OR 97702, USA

JACQUELINE B. CUPPLES,4 Oregon Department of Fish and Wildlife, 1401 Gekeler Lane, La Grande, OR 97850, USA

MARY JO HEDRICK, Oregon Department of Fish and Wildlife, 53447 Highway 31, Summer Lake, OR 97640, USA

BRUCE K. JOHNSON,1 Oregon Department of Fish and Wildlife, 1401 Gekeler Lane, La Grande, OR 97850, USA

DEWAINE H. JACKSON, Oregon Department of Fish and Wildlife, 4192N Umpqua Highway, Roseburg, OR 97470, USA

ABSTRACT Mule deer (Odocoileus hemionus hemionus) populations have experienced widespread declines inmuch of western North America and alteration or loss of habitat could be contributing to these declines.Consequently, understanding habitat features that are important to mule deer is necessary for effectivemanagement of the species and their habitat. From 2005–2012 we radio-marked 452 mule deer with globalpositioning system collars across 9 distinct winter ranges to evaluate winter habitat use along the east slope ofthe Cascade Range in south-central Oregon, USA. Using data from 357 mule deer across 9 analysis areas, wedeveloped regional habitat use models for mule deer on winter range at 3 spatio-temporal scales: herd rangebased on 100% minimum convex polygons around deer locations, home range based on 90% kernel densityestimates of deer locations, and foraging range based on locations obtained within 2 hours of sunrise or sunsetwithin the boundaries of the home range scale.We assessed habitat use of mule deer using a generalized linearmodel with a negative-binomial link function. We validated our models with locations from an independentdataset of 95 deer that wintered within 8 of our analysis areas. Model validation indicated that regionalmodels for all spatio-temporal scales predicted probability of use moderately to very well. At all spatio-temporal scales, predicted use by mule deer was greater in areas with forest canopy cover. These findings callinto question large-scale removal of western juniper (Juniperus occidentalis) in Oregon and other portions ofwestern United States to enhance habitat for sagebrush-dependent wildlife species. Across all spatio-temporal scales, we documented increased probability of use of areas farther from roads open to motorizedvehicle use by mule deer, highlighting the importance of limiting motorized vehicle use and access on muledeer winter range.We also documented increased use of areas with lower snow depth and on moderate slopesby mule deer. Our models can be used in land management planning to spatially predict mule deerdistribution and probability of use under alternative management scenarios that affect forest canopy cover ormotorized vehicle use on roads while accounting for other physical features of the landscape. Additionally,our models can be used to develop juniper removal projects to mitigate effects on mule deer winter rangehabitat while benefiting other wildlife. � 2018 The Wildlife Society.

KEY WORDS habitat use, juniper, Juniperus occidentalis, mule deer, multiscale, Odocoileus hemionus, Oregon, roads,winter range.

Mule deer (Odocoileus hemionus hemionus) populationsthroughout the western United States have experienced

decades-long population fluctuations over the past century(Wallmo 1981, Unsworth et al. 1999), including populationsin Oregon (Cliff 1939, Ebert 1976, Salwasser 1979, Peeket al. 2002). Mule deer populations in Oregon peaked atapproximately 575,000 in the 1960s (Workman and Low1976) and declined to nearly 230,000 by early 2015 (OregonDepartment of Fish and Wildlife [ODFW] 2003; ODFW,unpublished data). Several hypotheses have been proposed toexplain population declines of mule deer (Ballard et al. 2001,Peek et al. 2001, Hurley et al. 2011, Shallow et al. 2015,Johnson et al. 2017). One compelling argument is that

Received: 1 September 2017; Accepted: 9 March 2018

1Retired2E-mail: [email protected] Address: Eagle Environmental, Inc. 30 Fonda Road, Santa Fe,NM 87508.4Present Address: U.S. Fish andWildlife Service, 3502Highway 30, LaGrande, OR 97850.

The Journal of Wildlife Management; DOI: 10.1002/jwmg.21484

Coe et al. � Mule Deer Winter Habitat Use 1

increased forest canopy cover due to wildfire suppression anddeclining timber harvest has reduced available forageresources for mule deer, reducing mule deer nutritionalcondition, and concomitantly contributing to populationdeclines (Peek et al. 2001, 2002). Although effects of reducednutritional conditions on summer range may influencepopulation performance of mule deer, it is unknown whetherchanges to forest canopy cover on winter range is beneficial ordetrimental to mule deer. For example, forest canopy cover isan important habitat component for mule deer during winter(Leckenby and Adams 1986, D’Eon and Serrouya 2005,Anderson et al. 2012, Webb et al. 2013, Smith et al. 2015).In contrast, increased western juniper (Juniperus occidentalis)encroachment may reduce availability of forage (Rowlandet al. 2011, Witt and Patterson 2011, Bombaci and Pejchar2016) on winter range and be detrimental to mule deer.Most mortality of mule deer in northern climates, outside

of hunting, occurs in winter as a result of weather, predation,or starvation (Wallmo 1981, Ballard et al. 2001, Forresterand Wittmer 2013). Adult mule deer have adapted to losingweight in winter (Torbit et al. 1985), but during severewinters, adult survival declines, and given the high sensitivityof ungulate population growth to adult survival (Gaillardet al. 1998) populations also decline (Loveless 1967, Gilbertet al. 1970, Hurley et al. 2011). This is particularly true ifsufficient winter range habitat is not available (Wallmo 1981,Bergman et al. 2015). Further, fawn survival has increased inareas where supplemental forage was provided during winter(Baker and Hobbs 1985, Bishop et al. 2009), indicating thata lack of abundant and high quality forage on winter rangecould be a factor limiting mule deer population growth.Consequently, identifying habitat use patterns of mule deeron winter range is important for identifying habitat featurescritical to overwinter survival of mule deer.In the western United States, most mule deer winter range

occurs on publicly owned land managed by the Bureau ofLand Management (BLM; Cox et al. 2009) or the UnitedStates Forest Service (USFS). For example, in Oregon, BLMadministers 33% of mule deer winter range including >50%of our study area. Land management activities, includinglivestock grazing (Robinette et al. 1952, Willms et al. 1979),timber harvest (Bergman et al. 2014, Bombaci and Pejchar2016), energy development (Johnson et al. 2017, Sawyeret al. 2017), and recreational activities (Wisdom et al. 2004,Stankowich 2008) could affect mule deer distributions andpopulation dynamics. Additionally, industrial and residentialdevelopment on or near public lands is increasing, resultingin reduced winter range area and increased disturbance todeer (Sawyer et al. 2006, Kline et al. 2010, Duncan andBurcsu 2012). Identifying and protecting important habitatfeatures of mule deer habitat could mitigate the potentialnegative effects of human development on mule deer.In many parts of the western United States, including

central Oregon, fire suppression activities and disruptions tohistorical disturbance regimes has resulted in increaseddensity of western juniper in shrub-steppe plant communi-ties, causing declines in understory productivity of shrubs,grasses, and forbs (Rowland et al. 2011, Bombaci and Pejchar

2016). As a result, intensive harvest of western juniper onpublic and private lands has increased to provide habitat forsagebrush-dependent species, or to enhance conditions forgrowth of shrubs, forbs, and grasses that provide nourish-ment to a wide range of wild and domestic animals (Bombaciand Pejchar 2016). However, the effects of vegetationmanipulations that reduce forest canopy cover on mule deerwinter range and subsequent effects on habitat use, survival,and rate of population change is not well understood and infact could be detrimental given forest cover may be animportant component of mule deer winter range (Leckenbyand Adams 1986).Assessments of habitat use can provide robust indictors of

animal distribution in heterogeneous environments and maybe directly linked to animal abundance (Manly et al. 2002,Boyce et al. 2016). Further, habitat use models may beextended to a mapping environment and used in spatialanalyses to estimate changes in animal distribution whenhabitat conditions change because of environmental orhuman perturbations. This will allow land managers to betterunderstand and predict effects of timber harvest, fuelstreatments, changes in road access and motorized use, orincreased human developments on animal behavior anddistribution. Scale (i.e., extent, timing, and resolution) caninfluence inference about ecological patterns and processes(Wiens 1989, DeCesare et al. 2012) and, consequently,models developed at several spatio-temporal scales may beuseful for understanding underlying biological phenomenonand effects of different management applications (Senft et al.1987, Prokopenko et al. 2017). Consequently, we developedmodels at 3 spatio-temporal scales for estimating habitat useof mule deer on winter range. Our objectives were to identifyhabitat features that were related to habitat use on mule deerwinter range at 3 spatio-temporal scales (i.e., herd range,home range, and foraging scales) to apply at a regional leveland develop predictive maps and models that could be usedby land managers to inform future land management.

STUDY AREAWe conducted our study on 9 winter ranges east of thecrest of the Cascade Range in the high desert region ofsouth-central Oregon and northern California, USA from418N–458N and 1208W–1228W (Fig. 1; Table 1) from2006–2012. Mule deer winter range occurred primarily insagebrush steppe in proximity to mid-elevation forests(Zalunardo 1965, Dealy 1971). These wintering areasoccurred <1,500m elevation where mule deer concentratedat high densities during mid-winter. Climate in this regionwas strongly influenced by an orographic rain-shadow effectof the Cascade Mountains that restricted moisture fromPacific storms. Climate was temperate, with hot and drysummers and cold, wet winters with increased snowaccumulation at higher elevations. On mule deer winterranges, during January to March, 2006–2012 mean monthlyminimum temperature was �7.68C to �2.48C and meanmonthly cumulative precipitation was 19.7mm to 92.5mm(Oregon State University PRISM Climate Group 2012).Topography of winter ranges is relatively flat with gently

2 The Journal of Wildlife Management � 9999()

rolling hills ranging in elevation from 587–2,192m. Otherungulate species present in our study area included elk(Cervus canadensis) and small, disjunct populations ofbighorn sheep (Ovis canadensis). Potential predators of muledeer included cougar (Puma concolor), black bear (Ursusamericanus), coyote (Canis latrans), and bobcat (Lynx rufus).All mule deer populations in our study were huntedannually. Authorized hunting was restricted to harvest ofantlered animals through a limited permit system withhunts occurring annually during a 12-day hunt fromapproximately 1–12 October. Hunting of antlerless muledeer was not authorized during our study. Dominant landuse on public lands in our study area was livestock grazingduring summer with limited timber harvest that primarilyoccurred through removal of western juniper for habitat

improvement projects. On private lands, land use wasdominated by livestock grazing and agriculture operations.Winter ranges consisted of forested and non-forested

ecological systems (Dealy 1971, Leckenby 1978). Forestedecological types were dominated primarily by western juniperat lower elevations, and ponderosa pine (Pinus ponderosa),lodgepole pine (Pinus contorta), and Douglas-fir (Pseudotsugamenzesii) at higher elevations (U.S. Geological Survey 2001).Understory vegetation in ponderosa pine–Douglas-fir plantassociations consisted of bitterbrush (Purshia tridentata),greenleaf manzanita (Arctostaphylus patula), and snowbrushceanothus (Ceanothus velutinus); graminoids were dominatedby western needlegrass (Achnatherum occidentale), Ross’ssedge (Carex rossii), and bottlebrush squirreltail (Elymuselymoides; Franklin and Dyrness 1973). Non-forested

Figure 1. Study area depicting year-round minimum convex polygon of locations of 452 adult mule deer in south-central Oregon and northern California,USA, 2006–2012. Hatched lines represent herd range analysis areas used to model mule deer winter habitat use. Solid gray depicts designated mule deer winterrange as defined by the Oregon Department of Fish and Wildlife.


ecological types were dominated by Inter-Mountain BasinBig Sagebrush Steppe and Inter-Mountain Basin BigSagebrush Shrubland in the north and by Columbia PlateauLow Sagebrush Steppe and Columbia Plateau Steppe andGrassland in the south (U.S. Geological Survey 2001). Shrubspecies consisted primarily of Wyoming big sagebrush(Artemisia tridentata wyomingensis), low sagebrush (A. t.arbuscula), mountain sagebrush (A. t. vasyana), antelopebitterbrush (Purshia tridentata), gray rabbitbrush (Chrys-othamnus nauseosus), and yellow rabbitbrush (C. viscidiflorus).

METHODSIn our analysis of mule deer habitat use, we generally followedthe analysis approach of Rowland et al. (2018) used to assessregional habitat use of elk across multiple study areas. Ouroverall objectivewas similar toRowland et al. (2018) in thatwewere interested in assessing habitat use across multiple studyareas to develop a regional habitat use model for mule deer onwinter range. We used mule deer locations from 9 distinctwinter ranges to model intensity of use in relation to habitatcharacteristics using negative-binomial regression (Nielsonand Sawyer 2013), which is considered an analysis of habitatuse. By modeling use as a continuous variable, we consideredhabitat use in a probabilistic manner that relies on the relativeamount of time spent by the animals in each sampling unit(NielsonandSawyer2013).This is different than ananalysis ofhabitat selection (Manly et al. 2002), which Lele et al.(2013:1185) define as “strictly a binary decision, withoutcomes of use or nonuse of a resource unit.” The modelingapproach we used reveals where animals are on the landscapeand the relative amountof time spent by themin each samplingunit, rather than presence or absence of animals as typicallyreported in use-availability studies (Manly et al. 2002,Nielsonand Sawyer 2013).We considered ourmodeling to represent apopulation-scale or regional analysis; at this scale, spatialvariation in habitat features should account for differences inmule deer habitat use (Gaillard et al. 2010). Consequently, ourapproach allowed us to make regional inference on mule deerhabitat use that may be useful for landscape-level landmanagementplanning,whichwas theultimateobjectiveof ourresearch.

Mule Deer Capture and MonitoringFrom 2005 to 2011, we captured adult female mule deerduring winter from 9 distinct wintering herds within 7Wildlife Management Units (WMU) using net guns firedfrom a helicopter (Jacques et al. 2009). We attempted torandomly sample animals in proportion to winteringdensities (�1 collar/150 deer) to obtain a representativesample of the mule deer population in south-central Oregon.For winter captures, we physically restrained mule deerduring capture and released them following attachment of aglobal positioning system (GPS) collar (Coe et al. 2015). Wealso captured deer on summer range in a systematic effort tocollar 1 deer/100 km2. For summer captures, we used Clovertraps (Clover 1954) baited with alfalfa to attract deer or wechemically immobilized deer using darts fired from projectileguns. We immobilized deer with a combination of Telazol,xylazine, and ketamine, with Tolazine used as an antagonistfor xylazine (Kreeger et al. 2002). Summer capture methodsdiffered from winter because deer were widely dispersed inforested areas during summer making helicopter capturelogistically challenging. During all captures we restraineddeer and blindfolded them to reduce stress; we then ear-tagged, sexed, and aged deer and collected blood, tissue, andfecal samples. Wildlife veterinarians from ODFW approvedand supervised capture operations, which followed theGuidelines of the American Society ofMammalogists for theuse of wild mammals in research (Sikes and Gannon 2011).We fitted adult deer with GPS collars programmed to

record a location every 4 hours and self-release after 52–72weeks (Coe et al. 2015). We assessed winter habitat use ofmule deer using locations obtained between 1 January–31March because this was the time period when deer were non-migratory and concentrated on winter ranges. We calculatedfix success as the number of GPS locations obtained dividedby the number attempted. Mean fix success for GPS collars 1January–31March was>93% so we did not adjust for habitatinduced observation bias (Friar et al. 2004) in our assessmentof habitat use by mule deer.

Defining Analysis AreasWedefined each distinct wintering population ofmule deer asan analysis area (n¼ 9; Fig. 1). We calculated the 100%

Table 1. Name of wildlife management unit (WMU), estimated number of mule deer inWMU, name of analysis area, and number of individual radio-collaredadult female mule deer used in model development and model validation to assess habitat use of mule deer on winter range in south-central Oregon, USA,2006–2012.

WMU Mule deer populationa Analysis area name Radio-collars used in model development Radio-collars used in model validation

Metolius 8,400 Metolius 54 5Paulina 13,800 North Paulina 27 11

South Paulina 47 13Wagontire 4,000 Wagontire 30 6Fort Rock 11,000 Fort Rock 43 8Silver Lake 9,400 Silver Lake 60 25Interstate 7,000 East Interstate 32 0

West Interstate 50 15Klamath 5,000 Klamath 16 12

Total 359b 95

a Oregon Department of Fish and Wildlife, unpublished data.b Two deer occurred in different winter ranges in different years so total number of deer used in model development was 357.


minimum convex polygon (MCP) of all locations of allsampled deer in a distinct winter range to define our 9 analysisareas.We retained only the portion of theMCP that occurredwithin designated winter range as defined by ODFW and wedefined this largest geographical extent as a herd range.Winterrange designated by ODFW delineates areas where mostwintering populations of mule deer concentrate fromDecember–April while excluding high elevation sites withincreased snow depth and large tracts of lands converted toagriculture or other human development. We excluded areasoutside of defined winter range from the analysis because wewere interested in assessing habitat use by mule deer withinhabitats that were not dominated by human use and wheremost of the deer population resides during winter.We did notexclude private land from our analysis areas if they occurredwithin areas designated as winter range by ODFW.We identified 3 spatio-temporal scales to assess habitat use

of mule deer (Fig. S1, available online in SupportingInformation). Our first spatio-temporal extent was the herdrange, defined as the 100% MCP created from all deerlocations obtained during winter (1 Jan–31 Mar) clipped todefined winter range boundaries (Fig. S1). Next, wecalculated the 90% bivariate kernel density estimate(KDE) from all locations of all deer within a herd range(Fig. S1). We used a 90% KDE to ensure we captured areaswhere deer spent most of their time while restricting theareas infrequently used by wintering deer within their herdrange.We defined each 90%KDE as the home range analysisarea to provide a finer scale spatial extent of habitat use ofmule deer on winter range. Finally, we selected locationsfrom within the home range analysis area boundaries whendeer were likely to be actively foraging, which we defined aslocations that occurred between 0600–1000 and 1400–1800(Ager et al. 2003). We defined this spatio-temporal scale asthe foraging range. At all 3 spatio-temporal scales, weincluded a buffer of 4 km around each analysis boundary toensure we captured the influence of habitat features beyondthe analysis areas for distance-based covariates (e.g., distanceto roads) to which mule deer may respond.

Habitat VariablesWe developed an a priori set of 31 landscape characteristics(i.e., covariates; Table S1, available online in SupportingInformation) that may influence winter habitat use of muledeer based on existing literature (Leckenby 1978, Peek et al.2001, D’Eon and Serrouya 2005, Anderson et al. 2012,Webb et al. 2013) and expert opinion of wildlife managerswithin our study areas. Several landscape characteristics (e.g.,forest canopy cover) had multiple geographic informationsystem (GIS) data sources available, which increased thenumber of potential covariates to consider in our analysis ofwinter habitat use by mule deer to 41.We grouped covariatesinto 4 categories to predict mule deer winter habitat use:vegetation, topographic, human disturbance, and weather.We defined categories to distinguish those variables thatcould be influenced by managers (vegetation, humandisturbance) from those that could not (topographic,weather).

Habitat Use Model DevelopmentWe used our 9 individual analysis areas as replicates ingenerating a regionalwinter habitat usemodel formule deer insouth-central Oregon at 3 spatio-temporal scales. Thisapproach is analogous to that of using individual animals asthe primary sampling units when creating a population-levelmodel for a single study site (Sawyer et al. 2006, Fieberg et al.2010).We followed the approach ofRowland et al. (2018) andused a 4-step process to develop regional winter mule deerhabitat use models for all 3 spatio-temporal scales by 1)measuring covariates within systematically selected circularsampling units within each analysis area and spatio-temporalscale, 2) estimating the relative frequency of use in thesampling units for all collared mule deer within each analysisarea duringwinter, 3)modeling the relative frequency of use asthe response variable in a generalized linear model (GLM)using a negative-binomial habitat-use model (Nielson andSawyer 2013), and 4) averaging coefficients across analysisareas to generalize habitat relationships and develop a regionalwinter habitat use model at each spatio-temporal scale.After identifying our 9 analysis areas, and defining the

extent of each spatio-temporal scale of analysis, werandomly selected a starting point and systematically placednon-overlapping 350-m diameter circular sampling unitswithin the boundaries of each analysis area (Fig. S1, inset).The size of the sampling units was large enough to detectchanges in animal movements while providing counts ofanimal locations that approximated a negative-binomialdistribution (Nielson and Sawyer 2013). In addition, weensured that the size of the sampling units exceeded theinherent error in GPS locations and covariate layersconsidered during modeling (Nielson and Sawyer 2013).Next, we summarized habitat variables within these circularsampling units by calculating the point value, mean,distance to, or proportion within the circle of habitatvariables, creating 41 independent covariates for consider-ation in model development at all 3 spatio-temporal scales.We conducted Pearson’s pairwise correlation tests formulti-collinearity among covariates prior to modeling. Ifcovariates were highly correlated (i.e., |R|> 0.60), we eitherdropped 1 of the 2 related covariates or did not include them inthe same model. When deciding to drop a covariate fromconsideration because of multi-collinearity, we retained thecovariate thathadeithermoremanagementutility orwaseasierto interpret biologically. We also investigated histograms ofcovariates to identify substantial differences in distributions ofcovariates among analysis areas. These differences couldindicate potential problems in either identifying a commonrelationship between mule deer winter habitat use and thatcovariate, or applying our final regional model for winterhabitat use across all analysis areas. Thus, if we identified amarked difference in distributions among analysis areas, wedropped the covariate from further consideration prior tomodeling. We further analyzed combinations of covariates orbiologically meaningful quadratic relationships within eachcovariate category prior to model development.We used the number of locations observed within each

sampling circle as the dependent variable in a GLM to assess


habitat use for each analysis area. We used the natural log ofthe number of locations obtained from a particular analysisarea as an offset term (McCullagh and Nelder 1989) in theGLM to estimate probability of use for all radio-collareddeer in an analysis area as a function of a linear combinationof predictor covariates. Inclusion of the offset term scales theresponse to ensure modeling of intensity of use (e.g., 0.01,0.03, . . .) instead of integer counts (e.g., 0, 1, 2, . . .; Nielsonand Sawyer 2013). We used a negative-binomial model(Nielson and Sawyer 2013) because our dependent variablewas count data and we estimated probability of use for eachanalysis area and each spatio-temporal scale as:

ln E ti=totalð Þ½ � ¼ b0 þ b1x1i þ b2x2i . . . þ bpxpi; ð1Þ

where ti is the number of GPS locations from all mule deerwithin sampling circle i, total is the total number of locationsfor all deer in the analysis area, b0 is an intercept term, b1, b2,. . ., bp are coefficients to be estimated, x1i, . . . xpi are thevalues of p covariates measured in sampling circle i, and E[.]denotes the predicted probability of use. We conducted ouranalysis in R (R version 2.11.1, www.r-project.org, accessed6 Jul 2017) using the NB2 formulation available in theMASS contributed package (Venables and Ripley 2002).Model selection.—We followed the model selection

approach of Rowland et al. (2018), when determining thebest regional model across our 9 analysis areas for each of our3 spatio-temporal scales (Fig. S2, available online inSupporting Information). After reducing our list ofcandidate covariates for use in model development throughassessing correlations and histograms of covariate distribu-tions, we fit a univariate model for each covariate withmultiple data sources to determine which data source bestexplained habitat use. We then fit univariate models for eachremaining covariate and assessed consistency of estimatedcoefficients across analysis areas. We removed covariatesfrom further consideration when estimated coefficients wereinconsistent at 3 or more analysis areas. This approachensured that patterns of habitat use across analysis areas werenot attributable to errors in underlying GIS data layers, ordifferences in habitat features among analysis areas, but wereactual differences in habitat use.We then used a 2-stage information-theoretic approach

(Burnham and Anderson 2002) when developing our finalmodels. First, after eliminating correlated covariates orcovariates with inconsistent results across analysis areas, wefit a set of models within each covariate category todetermine which models best explained mule deer habitatuse. We then ranked models by Akaike’s InformationCriterion (AIC) values, where a rank of 1 indicated themodel had the lowest AIC at a particular analysis area. Wethen summed ranks of models within each covariate categoryacross analysis areas to identify the consistently best modelfor that category (i.e., the lowest AIC rank). For example, if amodel ranked first in 6 of 9 analysis areas and second in theother 3 analysis areas, its summed rank would be 12([6�1]þ [3�2]) and if a second model ranked first in 3 of 9analysis areas and second in the remaining 6 analysis areas, its

summed rank would be 15 ([3�1]þ [6�2]). In this scenario,the model with the lower summed rank would be retained forfurther model development. Using this approach, rather thanthe AIC values, gives equal weight to each data set inidentifying the best model across analysis areas. Further, thesummed ranks allow a qualitative assessment of differencesand similarities across analysis areas. After identifying thebest models within each category, we created all possiblemodel combinations, excluding interaction terms, using thebest summed rank models from vegetation, human distur-bance, topographic, and weather categories. We then rankedthis model set using the summed ranks procedure outlinedabove to determine our best rankedmodel across all 9 analysisareas. We retained the best ranked model (i.e., lowestsummed rank) at each spatio-temporal scale for interpreta-tion and model validation at a regional scale.After identifying our best ranked model, we estimated

regional habitat use coefficients by calculating the geometricmean of the coefficients across our 9 analysis areas for each ofour 3 spatio-temporal scales of analysis. Our regional habitatuse models represent relative probability of use becausepredictions from the regional model represent geometricmeans of analysis area probabilities rather than trueprobabilities (Nielson and Sawyer 2013). To estimate 95%confidence intervals for regional coefficients in our habitatuse models, we bootstrapped (Manly 2006) the primarysampling units (i.e., our 9 analysis areas) 1,000 times and re-estimated regional model coefficients for each sample. Weused the central 95% of the 1,000 estimates for eachcoefficient as the estimated confidence interval in ourregional models. This approach should result in moreconservative estimates than standard confidence limits formeta-analysis of ecological data (Adams et al. 1997).The model ranking approach we used allowed us to give

equal weight to each analysis area in developing the finalregional habitat use model. This approach also allowed us tohighlight similarities and differences among analysis areas.Although each analysis area model represents a measure ofprobability of use, the regional model, based on the averageof coefficients from individual analysis areas, representsrelative probability of use; predictions from the regionalmodels are geometric means of study-area probabilitiesrather than true probabilities (Nielson and Sawyer 2013).Although our approach allowed us to identify the bestcandidate model across analysis areas, it did not ensure thatthe candidate model contained covariates that had consistentrelationships with mule deer habitat use in each individualanalysis area. This approach is referred to as meta-replication(Johnson 2002) and allowed us to identify the mostconsistent habitat use relationships across our large modelingarea comprised of multiple analysis areas. Further, it allowedus to report effects that were spatially robust and to estimateeffect size at a higher precision than was possible whenanalyzing analysis areas individually (Borenstein et al. 2009).By combining information from 9 distinct analysis areas, wecreated a broader, more robust understanding of mule deerhabitat use during winter and developed a regional inferencespace useful for management applications.


http://www.r-project.org

Model validation.—We evaluated predictive performanceof our regional models using an independent sample ofanimals (Johnson et al. 2006). We withheld our independentsample of adult female mule deer (n¼ 95) from modeldevelopment and it did not contribute information topredicted habitat use patterns by mule deer in our 9 analysisareas. For each spatio-temporal scale of analysis, we used thebest ranked regional model to calculate probability of use for90-m grid cells in each of our 9 analysis areas. We calculatedpredicted use within each grid cell as:

Prk ¼ exp b1X 1 þ . . .þ bpX p

� �k; ð2Þ

where Prk was the probability of use at grid cell k, and ß1, . . .ßp were the estimated global model coefficients, p was thenumber of covariates, and X1, . . . Xp were the values of eachcovariate at grid cell k. We calculated observed use withineach grid cell as:

Obk ¼ locsk=X

locsð Þ; ð3Þ

where Obk was the proportion of total locations (locs) at gridcell k.For model validation, predicted probability of use for each

cell was normalized across an analysis area to sum to 1, sortedfrom 0 to 1, and assigned to 20 equal-area bins. Thus,prediction bin 1 had the lowest 5% of predicted values on thegrid, and bin 20 had the highest 5%. We calculated aSpearman rank correlation coefficient (Johnson 2001)between bin rank and the number of observed mule deerlocations within each bin for our independent sample of muledeer. We further compared expected (bin rank) withobserved (number of observations from summer-caughtdeer) using linear regression (Johnson et al. 2006). Aregression line with an intercept of zero and a slope of 1.0would indicate that predictions were proportional tointensity of use. We performed all statistical analyses in R.

RESULTSWe used 193,130 locations from 357 adult female mule deerto assess habitat use of mule deer on winter range across 3

spatio-temporal scales. Mean herd range size was 932 km2

(range¼ 549–2,017 km2; Table S2, available online inSupporting Information). Mean home range size was 194km2 (range¼ 140–623 km2; Table S2). After assessinghistograms of covariate distributions among study areasand collinearity between covariates, we retained 17 covariatesfor consideration in model development (Table S3, availableonline in Supporting Information).We estimated regional winter habitat use models for each

of 9 analysis areas and for each spatio-temporal scale using amean of 21,460 (range¼ 11,714–46,878) locations peranalysis area. Estimates of predicted use varied spatiallyacross our 9 analysis areas and by spatio-temporal scale (Figs.S3–S11, available online in Supporting Information). Ourbest ranked regional model to predict winter habitat use atthe herd range included 6 covariates: distance to forest,proportion shrubland, snow depth, aspect, slope, anddistance to open roads (Table 2). At the herd range scale,intensity of use declined as snow depth and distance fromforest increased (Fig. 2). Intensity of use was greatest onmoderate slopes (15–25%) and with increasing distance fromroads open to motorized vehicle access (Fig. 2). We alsoobserved a decline in intensity of use as proportion ofshrubland increased, and an increase in intensity of use onnorth-facing aspects (Fig. 2), but these relationships hadminimal influence on intensity of use by mule deer and 95%confidence intervals for estimated coefficients overlapped 0.At the home range scale, our best ranked regional model

included 4 covariates: proportion of area classified as forest,snow depth, slope, and distance to road open to motorizedvehicle use (Table 3). Intensity of use by mule deer at thisspatial scale was highest when the proportion of forestedarea was between 0.50 and 0.70 (Fig. 3). As observed at theherd range scale, intensity of use was greater at lower snowdepths and moderate slopes (10–30%; Fig. 3). Intensity ofuse increased as distance to roads open to motorized vehicleuse increased (Fig. 3), but this relationship was weak and95% confidence intervals for the estimated coefficientoverlapped 0.At the foraging range scale, our best ranked regional model

included 4 covariates: proportion of forested area with 1–20%

Table 2. Herd range coefficients for analysis area models of mule deer habitat use during winter in south-central Oregon and northern California, USA, 2006–2012. Regional estimates are the geometric mean of coefficients from individual analysis areas.

Analysis area InterceptDistance toforest (m)

Proportionshrubland

Proportionshrubland2

Mean snowdepth (cm)

Meancosineaspect

Slope(%)

Slope2

(%)

Distance toopen road

(km)

Distance toopen road2

(km)

Metolius �9.327 �0.002 2.919 �1.863 �0.017 0.330 �0.034 0.000 0.518 0.135North Paulina �8.395 0.000 �2.546 1.340 �0.033 0.410 0.084 �0.002 �0.851 0.082South Paulina �9.580 �0.001 �0.463 0.348 �0.004 0.331 0.103 �0.004 0.306 0.027Wagontire �10.942 �0.001 �0.419 1.094 �0.080 0.451 0.180 �0.005 0.487 �0.067Fort Rock �8.594 �0.003 3.105 �2.849 �0.006 0.057 �0.031 0.001 �0.136 0.035Silver Lake �8.903 �0.001 �0.634 1.061 �0.005 0.271 0.013 �0.001 0.197 �0.030East Interstate �10.346 �0.001 2.679 �2.147 �0.007 0.393 0.076 �0.001 1.616 �0.572West Interstate �10.795 0.000 1.371 �2.559 0.003 �0.080 0.180 �0.007 1.563 �0.365Klamath �9.371 �0.009 �0.989 0.275 �0.002 �0.683 0.011 �0.000 1.613 �0.412Regional �9.584 �0.002 0.558 �0.589 �0.017 0.164 0.065 �0.002 0.590 �0.130Lower 95% CI �0.003 �0.446 �1.485 �0.031 �0.064 0.024 �0.003 0.202 �0.267Upper 95% CI �0.001 1.599 0.213 �0.005 0.338 0.104 �0.001 1.062 �0.007


canopy cover, proportion of forested area with >20% canopycover, slope, and distance to roads open to motorized vehicleuse (Table 4). At this spatio-temporal scale, intensity of usewas higher in areas with lower tree canopy cover (0–20%;Fig. 4). In contrast, intensity of use declined in areas withforest canopy cover >20% (Fig. 4). Although the 95%confidence interval for the effect of forest canopy cover>20% did not overlap 0, this effect was relatively weakcompared to other covariates in the model. Similar to theherd range and home range scales, intensity of use at theforaging range scale was greatest in areas with moderateslopes (10–30%) and farther from roads open to motorizedvehicle use (Fig. 4).We were unable to evaluate predictive performance in the

East Interstate analysis area because the validation data setdid not include any deer that wintered in this area. Mean

Spearman rank correlation coefficients (rs) across analysisareas were 0.72 for herd range, 0.46 for home range, and 0.47for foraging range models (Fig. S12a–S12c, Table S4,available online in Supporting Information). Application ofour regional models to individual analysis areas indicated ourregional models performed better at some analysis areas thanothers. At the herd range scale, predicted and observed usewas highly correlated (rs� 0.75) at 5 of 8 analysis areas,moderately correlated (0.75> rs> 0.50) at 2 analysis areas,and negatively correlated (rs¼�0.50) at 1 analysis area(Table S4). At the home range scale, predicted and observeduse was highly correlated (rs� 0.75) at 1 of 8 analysis areas,moderately correlated (0.75> rs> 0.50) at 2 analysis areas,and marginally or weakly correlated (rs< 0.50) at theremaining 5 analysis areas (Table S4). At the foragingrange scale, predicted and observed use was highly correlated

Figure 2. Marginal plots from an analysis of mule deer winter habitat use in south-central Oregon and northern California, USA, 2006–2012. Plots are for theregional herd range model and show intensity of use in relation to individual model covariates, when all other covariates included in the model are held at theirmedians.


(rs� 0.75) at 1 of 8 analysis areas, moderately correlated(0.75> rs> 0.50) at 3 analysis areas, and marginally orweakly correlated (rs< 0.50) at 3 analysis areas, andnegatively correlated at 1 analysis area (Table S4). Acrossour 3 spatio-temporal scales of analysis, Wagontire, SilverLake, and Klamath had consistently high validation scores,whereas Metolius, North Paulina, and Fort Rock had thelowest.

DISCUSSIONPerhaps the primary determinant of distribution of mule deerduring winter is snow depth, with depths >46 cm effectivelyprecluding use (Gilbert et al. 1970). Our results supportedthis conclusion; one of the strongest relationships weobserved in our analysis was decreased use of areas with

increased snow depth in the herd range and home rangemodels. This pattern of increased use in areas with decreasedsnow depth is likely influenced by energy costs and foragingefficiency. For example, increased snow cover and depthreduces forage availability and foraging efficiency ofungulates (Wickstrom et al. 1984). Further, energyexpenditure from locomotion increased in a log-linearmanner with increased snow depth (Parker et al. 1984).Consequently, it was not surprising that mule deer used areaswith lower snow depth within our analysis areas. We did notobserve a relationship between snow depth and habitat use bymule deer in our foraging range models. We contend we didnot observe this relationship because patterns of habitat useat coarser scales (e.g., herd range and foraging range) alreadyaccounted for deer using areas with reduced snow cover.

Table 3. Home range coefficients for analysis areas in models of mule deer habitat use during winter in south-central Oregon and northern California, USA,2006–2012. Regional estimates are the geometric mean of coefficients from individual analysis areas.

Analysis area Intercept Proportion forest Proportion forest2 Mean snow depth (cm) Slope (%) Slope2 Distance to open road (km)

Metolius �8.110 1.884 �1.610 �0.006 0.007 �0.000 0.251North Paulina �7.782 1.717 �1.002 �0.014 0.030 0.000 �0.781Wagontire �9.498 1.566 �1.546 �0.030 0.095 �0.002 0.238South Paulina �8.347 0.362 �0.472 �0.001 �0.007 0.000 0.136Fort Rock �7.586 1.633 �1.400 �0.001 �0.018 0.000 �0.163Silver Lake �8.239 1.826 �1.314 �0.002 �0.034 0.000 �0.054East Interstate �8.528 2.476 �2.525 0.002 0.064 �0.001 0.188West Interstate �8.878 0.346 �0.347 0.000 0.160 �0.006 0.352Klamath �8.398 3.573 �1.817 �0.003 0.006 �0.001 0.097Regional �8.374 1.709 �1.337 �0.006 0.034 �0.001 0.029Lower 95% CI 1.255 �1.715 �0.011 0.003 �0.002 �0.154Upper 95% CI 2.232 �0.997 �0.001 0.067 �0.000 0.185

Figure 3. Marginal plots from an analysis of mule deer winter habitat use in south-central Oregon and northern California, USA, 2006–2012. Plots are for theregional home range model and show intensity of use in relation to individual model covariates, when all other covariates included in the model are held at theirmedians.


Across our 3 spatio-temporal scales of analysis, wedocumented several associations between forest cover andhabitat use by mule deer. Interpretation of these patternsvaried across spatial and temporal scales and likelyhighlighted the importance of mule deer balancing theneeds of thermal and hiding cover with foraging opportunity.At the herd range scale, we documented increased intensityof use by deer in areas closer to forest cover and at the homerange scale, increased intensity of use was associated withareas composed of 60–70% forest. We contend this indicatesmule deer were using areas at these spatial scales that allowedthem to seek thermal or hiding cover. In contrast, at theforaging range scale, we documented increased use of areas

with low forest canopy cover (i.e., 1–20%) and decreasedintensity of use of areas with >20% forest canopy cover.Areas with lower canopy cover likely had increased foragebecause herbaceous understory vegetation decreases substan-tially where juniper cover is >20% (Miller et al. 2000, Wittand Patterson 2011). The need for shrubs and associatedvegetation for forage (Leckenby 1977) and tree cover forthermal and hiding (Leckenby and Adams 1986) necessitatesthe use of forested and unforested areas during winter (Kieet al. 2002). We contend mule deer in our study areaconcentrated use in areas that provided sufficient thermaland hiding cover while also providing foraging opportunitiesin areas with relatively low forest canopy cover (<20%).

Table 4. Foraging range coefficients for analysis area models of mule deer habitat use during winter in south-central Oregon and northern California, USA,2006–2012. Regional estimates are the geometric mean of coefficients from individual analysis areas.

Analysis area InterceptProportion canopy cover

1–20%Proportion canopy

cover >20%Slope(%)

Slope2

(%)Distance to open

road (km)Distance to open

road2 (km)

Metolius �8.221 0.654 0.007 0.003 �0.000 �0.126 0.198North Paulina �7.711 0.895 0.565 0.015 �0.000 �0.677 0.004Wagontire �9.721 0.426 �0.007 0.090 �0.003 0.795 �0.266South Paulina �8.234 0.106 �0.226 0.005 �0.001 �0.519 0.306Fort Rock �7.879 0.518 0.197 �0.004 0.000 0.565 �0.334Silver Lake �8.389 0.879 0.356 �0.020 0.000 �0.145 0.074East Interstate �8.634 0.630 �0.083 0.064 �0.001 0.887 �0.361West Interstate �9.823 0.660 0.183 0.170 �0.006 2.295 �0.915Klamath �9.421 1.944 1.673 0.021 �0.001 2.015 �0.803Regional �8.670 0.746 0.296 0.038 �0.001 0.566 �0.233Lower 95% CI 0.517 0.041 0.009 �0.003 0.012 �0.459Upper 95% CI 1.042 0.618 0.074 �0.000 1.149 �0.001

Figure 4. Marginal plots from an analysis of mule deer winter habitat use in south-central Oregon and northern California, USA, 2006–2012. Plots are for theregional foraging range model and show intensity of use in relation to individual model covariates, when all other covariates included in the model are held attheir medians.


Forest cover is an important component of mule deer winterhabitat use throughout the inter-mountain west in Colorado,Oregon, Washington, British Columbia, and Idaho (Leck-enby and Adams 1986, Carson and Peek 1987, D’Eon andSerrouya 2005, Anderson et al. 2012, Smith et al. 2015).Perhaps the most compelling function of forest cover is that itprovides important energetic benefits formuledeer.Because oftheir low ratio of body mass to surface area, mule deer have asmall thermal neutral zone and may have increased needs forthermal cover, particularly during winter (Raedeke and Taber1982). Consequently, mule deer in our study likely used forestcover to regulate body temperature during cold winterconditions (Leckenby and Adams 1986). Further, forestcanopy intercepts snowfall and influences snow accumulationpatterns so that snowdepth is lower in forested areas (Pomeroyet al. 1998). Mule deer use areas with forest cover to mediateeffects of snow (Mysterud and Østbye 1999) and this patternwas present in our study area.Predation is often a proximate cause of mortality in mule

deer populations (Bleich and Taylor 1998, Ballard et al.2001, Forrester and Wittmer 2013) and forest cover may beused by mule deer in an effort to reduce risk of predation(Mysterud and Østbye 1999). For example, forest cover mayallow increased concealment and escapement from pred-ators (Bowyer and Bleich 1984, Jedrzejewska et al. 1994).Consequently, minimization of predation risk may be analternative or supplemental explanation to the patterns wedocumented of increased use of forest cover by mule deer onwinter range. However, we lacked data on mule deerpredators within our study areas to test this hypothesis, sowe can not explicitly link use of forest cover to reducedpredation risk.In our analysis areas, western juniper was the primary tree

species and given our associations between mule deer habitatuse and forest cover, juniper is likely an important habitatfeature for mule deer winter range. Western juniper has beenexpanding within shrub-steppe communities in Oregon since1900 (Miller and Rose 1999). In Oregon, juniper removal hasoccurred on>165,000 ha from2010–2015 in efforts to benefitsage-grouse (Centrocercus urophasianus; Natural ResourcesConservation Service 2015, Sage-Grouse ConservationPartnership 2015) and other sagebrush-dependent species;however, evidence to support any positive or detrimentalbenefits of juniper removal on mule deer habitat use ordemography is lacking. Leckenby (1978) suggested that clear-cutting juniper on winter range could be detrimental to muledeer because thermal and hiding cover would be removed.Increases in juniper cover across mule deer winter range,however, may have negatively affected forage availability formule deer because abundance of shrubs, forbs, and grassesdecline in juniper stands (Miller et al. 2000, Peek et al. 2001).Past research has indicated that nutritional resources duringwinter can influence survival and population dynamics ofmuledeer (BakerandHobbs1985,Bishopetal.2009,Bergmanetal.2014) and if increases in juniper cover have reduced foragequantity and quality formule deer, this could have population-level consequences. Given the lack of knowledge on effects ofjuniper treatments on mule deer habitat use and demography,

future research to elucidate effects of juniper removal on muledeer is warranted.One of the strongest patterns we observed across all 3 spatio-

temporal analyses was a consistent avoidance of roads open tomotorized vehicle use, which has also been documented inWyoming, Colorado, and southeast Alaska (Sawyer et al.2006, Webb et al. 2013, Northrup et al. 2015, Gilbert et al.2017). In contrast, in Idaho, where mule deer shared winterrange with elk, mule deer were closer to roads during winter,whereas elk used areas farther from roads (Stewart et al. 2010).This pattern likely occurs because of competitive interactionsbetween the mule deer and elk, where mule deer are displacedby elk to areas closer to roads (Johnson et al. 2000). In ouranalysis areas, noor very fewelk co-occurredwithmule deer onwinter range (ODFW, unpublished data), which allowed us toestablish clearer inference regarding effects of roads open tomotorized vehicle use onmule deerwinter habitat use. Inotherareas, ungulates demonstrated avoidance of roads open tomotorized access and open areas (e.g., meadows, non-forest)during the day but used areas closer to both at increasedfrequencies at night (Ager et al. 2003, Crosmary et al. 2012,Bonnot et al. 2013, Northrup et al. 2015). Given theassociations we documented between use of forest cover andhabitat use bymule deer, forest covermay be used bymule deerto minimize the effects of human disturbance in our studyareas. Forest cover allows increased concealment and escape-ment from human hunting (Kufeld et al. 1988) andminimization of disturbance from other forms of humanrecreation (Wisdom et al. 2004), which aligns with ourfindings that mule deer spent less time in areas close to roadsopen to motorized vehicle use.We view our findings of mule deer avoidance of roads open

to motorized vehicle use to be linked to human disturbance,and not the physical features of the road. The strongavoidance of roads open to motorized vehicle use, andassociated human disturbance, by ungulates observed in ourstudy and others (Rowland et al. 2000, Sawyer et al. 2006,Coe et al. 2011, Webb et al. 2013, Gilbert et al. 2017)suggests a substantial portion of what may be suitable habitatmay no longer be used (Rogala et al. 2011). In a meta-analysis of the effects of flight responses of ungulates tohuman disturbance, it was found that ungulates, includingmule deer, had a strong avoidance and increased movementrates in response to human activity, especially for huntedpopulations (Stankowich 2008) like our study populations insouth-central Oregon. Consequently, human disturbanceassociated with motorized vehicle use on roads may causemule deer to increase movement rates, leading to a netreduction in their energy budget, which could ultimatelyhave demographic consequences (Robinette et al. 1973,Bishop et al. 2009, Tollefson et al. 2010). Fortunately, giventhat ungulates likely respond to the disturbance associatedwith motorized vehicle use of roads and not the road itself,this situation can be rectified through seasonal or permanentroad closures. Finally, increased densities of roads open tomotorized vehicle use are likely to increase human access andrates of poaching (Cole et al. 1997). In our study areas,poaching represented 3.1% of adult female mule deer


mortality (Mulligan 2015). Population growth rates ofungulates are highly sensitive to changes in adult femalesurvival (Gaillard et al. 1998, 2000) and the rate of poachingobserved in our study areas could be contributing toreductions in population growth rates.At all 3 of our spatio-temporal analysis scales, we observed a

strong association of mule deer using areas of moderate (10–30%) slopes and at the herd range scale showed a weakrelationship with increased use on northern aspects.Prevailing winter winds in our analysis areas were fromthe south and deer typically seek shelter from wind whengusts are>40 km/hour and temperatures<�108C (Loveless1967, Wood 1988). Topography can effectively serve asthermal cover for deer by providing a windbreak.Most of ouranalysis areas were relatively flat (�x slope across the 9 analysisareas was 8.4%) and we contend deer increased use onnorthern aspects and moderate slopes within our analysisareas as protection from wind, given the scarcity oftopographic complexity on the landscape.Our overarching objective was to assess habitat use patterns

of mule deer across multiple analysis areas to documenthabitat use patterns on a regional scale. Using this approach,our regional models were able to accurately predict habitatuse in most analysis areas and we identified several robustpatterns of habitat use of mule deer on winter range that areinformative to managers. Despite the benefits of ourmodeling approach, validation results across our 3 spatio-temporal scales and 9 individual analysis areas were notperfect. There are several possible reasons for poor validationresults in any particular analysis area: difference inphysiographic characteristics, differences in the proportionof shrub or other covariates, or small validation sample sizes.No one reason stands out among these options. The mostlikely explanation for low correlation between predicted andobserved use in any of the analysis areas is that predictorvariable(s) important at an individual analysis area scale werenot identified prior to our modeling efforts or were notincluded in the regional model. Given the geographic areaand diversity of conditions experienced by mule deer in ouranalysis areas, this was not unexpected, and it highlights thecomplexity of modeling habitat use patterns for species acrosslarge areas.

MANAGEMENT IMPLICATIONSWe encourage wildlife and land managers to employpredictive models developed here and other areas in aGIS to assess effects of proposedmanagement actions such astimber harvest, juniper removal, or closures of roads open tomotorized vehicle use in mule deer winter range. Thesemodels and models developed elsewhere for mule deer willallow effective assessment of alternative managementpractices such as habitat treatments and changes tomotorized vehicle access to quantify the degree to whichmanagement practices may be beneficial or detrimental.Our models identified 2 robust patterns of mule deer winter

habitat use that are under direct influence of land managers:increased use of areas with forest cover and decreased use ofareas with roads open to motorized vehicle access. We

encourage land managers to implement permanent orseasonal road closures on mule deer winter range tominimize disturbance and displacement of mule deer. Wealso recommend managers attempt to retain forested areaswithin mule deer winter ranges to increase use by mule deer.Based on our results showing an association between forestcover and winter habitat use by mule deer, habitatmanagement that benefits species such as sage-grouse(e.g., widespread juniper removal) have the potential toreduce use of these areas by mule deer during winter.Consequently, we recommend managers implement projectsthat consider and integrate the needs of multiple species. Forexample, where mule deer and sage-grouse co-occur,mitigation of effects of removal of tree cover to benefitsage-grouse could occur through strategic planning tominimize vegetation disturbance in areas that have otherattributes associated with increased probability of use bymule deer but not of sage-grouse (e.g., moderate slopes).Consideration of multiple vegetation treatment prescriptions(e.g., thinning vs. complete juniper removal) will also allowmanagers to identify alternative management practices thatmitigate or benefit predicted habitat use by mule deer inconjunction with balancing the needs of other sagebrush-dependent species. Although snow depth is a factor outsidethe control of wildlife managers, our models could be used toidentify areas that have a high probability of use by mule deerduring winters with heavy snowfall. During these years, muledeer populations are likely to be nutritionally stressed andhave increased risk of overwinter mortality. Our models canbe used to identify these areas and assign a high level ofprotection from future land management changes to ensurethe long-term conservation of mule deer winter range.

ACKNOWLEDGMENTSBrim Aviation and Leading Edge assisted with capture andplacing GPS collars on mule deer. The Klamath Tribesprovided some of theGPS collar data used in our analysis andwe appreciate their cooperation. Numerous staff fromODFW assisted with capture and monitoring of muledeer including A. N. Larkins, G. A. Bjornstrom, K. M.Drake, Z. Turnbull, D. Immell, and J. Orr. We thank 2anonymous reviewers for providing comments on earlierversions of this manuscript. Funding for this research wasprovided by Federal Aid in Wildlife Restoration Grant W-102-R, ODFW, and the Oregon Department of Transpor-tation.

LITERATURE CITEDAdams, D. C., J. Gurevitch, and M. S. Rosenberg. 1997. Resampling testsfor meta-analyses of ecological data. Ecology 78:1277–1283.

Ager, A. A., B. K. Johnson, J. W. Kern, and J. G. Kie. 2003. Daily andseasonal movements and habitat use by female Rocky Mountain elk andmule deer. Journal of Mammalogy 84:1076–1088.

Anderson, E. D., R. A. Long, M. P. Atwood, J. G. Kie, T. R. Thomas, P.Zager, and R. T. Bowyer. 2012. Winter resource selection by female muledeer Odocoileus hemionus: functional response to spatio-temporal changesin habitat. Wildlife Biology 18:153–163.

Baker, D. L., and N. T. Hobbs. 1985. Emergency feeding of mule deerduring winter: tests of a supplemental ration. Journal of WildlifeManagement 49:934–942.


Ballard,W. B., D. Lutz, T.W. Keegan, L. H. Carpenter, and J. C. DeVos Jr.2001. Deer-predator relationships: a review of recent North Americanstudies with emphasis on mule and black-tailed deer. Wildlife SocietyBulletin 29:99–115.

Bergman, E. J., C. J. Bishop, D. J. Freddy, G. C. White, and P. F. Doherty.2014. Habitat management influences overwinter survival of mule deerfawns in Colorado. Journal of Wildlife Management 78:448–455.

Bergman, E. J., P. F. Doherty, G. C.White, andD. J. Freddy. 2015. Habitatand herbivore density: response of mule deer to habitat management.Journal of Wildlife Management 79:60–68.

Bishop, C. J., G. C. White, D. J. Freddy, B. E. Watkins, and T. R.Stephenson. 2009. Effect of enhanced nutrition on mule deer populationrate of change. Wildlife Monographs 172:1–28.

Bleich, V. C., and T. J. Taylor. 1998. Survivorship and cause-specificmortality in five populations of mule deer. Great Basin Naturalist58:265–272.

Bombaci, S., and L. Pejchar. 2016. Consequences of pinyon and juniperwoodland reduction for wildlife in North America. Forest Ecology andManagement 365:34–50.

Bonnot, N., N. Morellet, H. Verheyden, B. Cargnelutti, B. Lourtet, F.Klein, and A. M. Hewison. 2013. Habitat use under predation risk:hunting, roads and human dwellings influence the spatial behaviour of roedeer. European Journal of Wildlife Research 59:185–193.

Borenstein, M., L. V. Hedges, J. P. T. Higgins, and H. R. Rothstein. 2009.Introduction to meta-analysis. John Wiley and Sons, Ltd, West Sussex,United Kingdom.

Bowyer, R. T., and V. C. Bleich. 1984. Effects of cattle grazing on selectedhabitats of southern mule deer. California Fish and Game 70:240–247.

Boyce, M. S., C. J. Johnson, E. H. Merrill, S. E. Nielsen, E. J. Solberg, andB. Moorter. 2016. Can habitat selection predict abundance? Journal ofAnimal Ecology 85:11–20.

Burnham, K. P., and D. R. Anderson. 2002. Model selection andmultimodel inference: a practical information theoretic approach.Springer-Verlag, New York, New York, USA.

Carson, R. G., and J. M. Peek. 1987. Mule deer habitat selection patterns innorthcentral Washington. Journal of Wildlife Management 51:46–51.

Cliff, E. P. 1939. Relationship between elk and mule deer in the BlueMountains of Oregon. North American Wildlife Conference 4:560–569.

Clover, M. R. 1954. A portable deer trap and catch-net. California Fish andGame 40:367–373.

Coe, P. K., B. K. Johnson, M. J. Wisdom, J. G. Cook, M. Vavra, and R. M.Nielson. 2011. Validation of elk resource selection models with spatiallyindependent data. Journal of Wildlife Management 75:159–170.

Coe, P. K., R. M. Nielson, D. H. Jackson, J. B. Cupples, N. E. Seidel, B. K.Johnson, S. C. Gregory, G. A. Bjornstrom, A. N. Larkins, and D. A.Speten. 2015. Identifying migration corridors of mule deer threatened byhighway development. Wildlife Society Bulletin 39:256–267.

Cole, E. K., M. D. Pope, and R. G. Anthony. 1997. Effects of roadmanagement on movement and survival of Roosevelt elk. Journal ofWildlife Management 61:1115–1126.

Cox,M.,D.W.Lutz, T.Wasley,M. Fleming, B. B.Compton, T.Keegan,D.Stroud, S. Kilpatrick, K. Gray, and J. Carlson. 2009. Habitat guidelines formule deer: Intermountain West ecoregion. Mule Deer Working Group,Western Association of Fish and Wildlife Agencies, Boise, Idaho, USA.

Crosmary, W.-G., M. Valeix, H. Fritz, H. Madzikanda, and S. D. Cot�e.2012. African ungulates and their drinking problems: hunting andpredation risks constrain access to water. Animal Behaviour 83:145–153.

D’Eon, R. G., and R. Serrouya. 2005. Mule deer seasonal movements andmultiscale resource selection using global positioning system radiotelem-etry. Journal of Mammalogy 86:736–744.

Dealy, J. E. 1971. Habitat characteristics of the Silver Lake mule deer range.USDA Forest Service Research Paper PNW-125. Pacific NorthwestForest and Range Experiment Station, U.S. Department of Agriculture,Forest Service, Portland, Oregon, USA.

DeCesare, N. J., M. Hebblewhite, F. Schmiegelow, D. Hervieux, G. J.McDermid, L. Neufeld, M. Bradley, J. Whittington, K. G. Smith, andL. E. Morgantini. 2012. Transcending scale dependence in identifyinghabitat with resource selection functions. Ecological Applications22:1068–1083.

Duncan, J. A., and T. Burcsu. 2012. Landscape development and mule deerhabitat in Central Oregon. Pages 85–102 in B. K. Kerns, A. J. Shlisky, andC. J. Daniel, technical editors. Proceedings of the First Landscape State-

and-Transition Simulation Modeling Conference PNW-GTR-869:85-101. U.S. Department of Agriculture, Forest Service, Pacific NorthwestResearch Station, Portland, Oregon, USA.

Ebert, P. N. 1976. Recent changes in Oregon’s mule deer population andmanagement. Proceedings of the 56th Annual Conference of the WesternAssociation of State Game and Fish Commisioners 56:408–414.

Fieberg, J., J. Matthiopoulos, M. Hebblewhite, M. S. Boyce, and J. L. Frair.2010. Correlation and studies of habitat selection: problem, red herring oropportunity? Philosophical Transactions of the Royal Society B:Biological Sciences 365:2233–2244.

Forrester, T. D., and H. U. Wittmer. 2013. A review of the populationdynamics of mule deer and black-tailed deer Odocoileus hemionus in NorthAmerica. Mammal Review 43:292–308.

Franklin, J. F., and C. T. Dyrness. 1973. Natural vegeation of Oregon andWashington. U.S. Department of Agriculture, U.S. Forest Service,Portland, Oregon, USA.

Friar, J. L., S. E. Nielsen, E. H. Merrill, S. R. Lele, M. S. Boyce, R. H.Munro, G. B. Stenhouse, and H. L. Beyer. 2004. Removing GPS collarbias in habitat selection studies. Journal of Applied Ecology 41:201–212.

Gaillard, J.-M., M. Festa-Bianchet, and N. G. Yoccoz. 1998. Populationdynamics of large herbivores: variable recruitment with constant adultsurvival. Trends in Ecology & Evolution 13:58–63.

Gaillard, J. M., M. Festa-Bianchet, N. G. Yoccoz, A. Loison, and C. Toigo.2000. Temporal variation in fitness components and population dynamicsof large herbivores. Annual Review of Ecology and Systematics31:367–393.

Gaillard, J.-M., M. Hebblewhite, A. Loison, M. Fuller, R. Powell, M.Basille, and B. Van Moorter. 2010. Habitat–performance relationships:finding the right metric at a given spatial scale. Philosophical Transactionsof the Royal Society B: Biological Sciences 365:2255–2265.

Gilbert, P. F., O. C. Wallmo, and R. B. Gill. 1970. Effect of snow depth onmule deer in Middle Park, Colorado. Journal of Wildlife Management34:15–23.

Gilbert, S. L., K. J. Hundertmark, D. K. Person, M. S. Lindberg, and M. S.Boyce. 2017. Behavioral plasticity in a variable environment: snow depthand habitat interactions drive deer movement in winter. Journal ofMammalogy 98:246–259.

Hurley, M. A., J. W. Unsworth, P. Zager, M. Hebblewhite, E. O. Garton,D.M.Montgomery, J. R. Skalski, and C. L.Maycock. 2011. Demographicresponse of mule deer to experimental reduction of coyotes and mountainlions in southeastern Idaho. Wildlife Monographs 178:1–33.

Jacques, C. N., J. A. Jenks, C. S. Deperno, J. D. Sievers, T. W. Grovenburg,T. J. Brinkman, C. C. Swanson, and B. A. Stillings. 2009. Evaluatingungulate mortality associated with helicopter net-gun captures in thenorthern Great Plains. Journal of Wildlife Management 73:1282–1291.

Jedrzejewska, B., H. Okarma, W. Jedrzejewski, and L. Milkowski. 1994.Effects of exploitation and protection on forest structure, ungulate densityand wolf predation in Bialowieza Primeval Forest, Poland. Journal ofApplied Ecology 31:664–676.

Johnson, B. K., J. W. Kern, M. J. Wisdom, S. L. Findholt, and J. G. Kie.2000. Resource selection and spatial separation of mule deer and elk duringspring. Journal of Wildlife Management 64:685–697.

Johnson, C. J., S. E. Nielsen, E. H. Merrill, T. L. McDonald, and M. S.Boyce. 2006. Resource selection functions based on use-availability data:theoretical motivation and evaluation methods. Journal of WildlifeManagement 70:347–357.

Johnson, D. H. 2001. Validating and evaluating models. Pages 105–119 inT. M. Shenk, and A. B. Franklin, editors. Modeling in natural resourcemanagement. Island Press, Washington D.C., USA.

Johnson, D. H. 2002. The importance of replication in wildlife research.Journal of Wildlife Management 66:919–932.

Johnson, H. E., J. R. Sushinsky, A. Holland, E. J. Bergman, T. Balzer, J.Garner, and S. E. Reed. 2017. Increases in residential and energydevelopment are associated with reductions in recruitment for a largeungulate. Global Change Biology 23:578–591.

Kie, J. G., R. T. Bowyer, M. C. Nicholson, B. B. Boroski, and E. R. Loft.2002. Landscape heterogeneity at different scales: effects on spatialdistrubution of mule deer. Ecology 83:530–544.

Kline, J. D., A. Moses, and T. Burcsu. 2010. Anticipating forest and rangeland development in central Oregon (USA) for landscape analysis, with anexample application involving mule deer. Environmental Management45:974–984.


Kreeger, T. J., J. M. Arnemo, and J. P. Raath. 2002. Handbook of wildlifechemical immobilization. Wildlife Pharmaceuticals, Fort Collins,Colorado, USA.

Kufeld,R.C.,D.C.Bowden,andD.L.Schrupp.1988. Influenceofhuntingonmovements of female mule deer. Journal of Range Management 41:70–72.

Leckenby, D. A. 1977.Management of mule deer and their habitat: applyingconcepts of behavior, physiology, and microclimate. Western Proceedings57th Annual Conference Western Association of State Game and FishCommissioners 57:206–217.

Leckenby, D. A. 1978. Mule deer occupancy of plant communities on asouth-central Oregon winter range. OregonWildlife Research ReportW-70-R. Oregon Department of Fish and Wildlife, Portland, USA.

Leckenby, D. A., and A. W. Adams. 1986. A weather severity index on amule deer winter range. Journal of Range Management 39:244–248.

Lele, S. R., E. H. Merrill, J. Keim, and M. S. Boyce. 2013. Selection, use,choice and occupancy: clarifying concepts in resource selection studies.Journal of Animal Ecology 82:1183–1191.

Loveless, C. M. 1967. Ecological characteristics of a mule deer winter range.Technical Publication Number 20. Colorado Game, Fish and ParksDepartment, Denver, USA.

Manly, B. F. J. 2006. Radomization, bootstrap andMonte Carlo methods inbiology. Third edition. Chapman & Hall, Boca Raton, Florida, USA.

Manly, B. F. J., L. L. McDonald, D. L. Thomas, T. L. McDonald, andW. P. Erickson. 2002. Resource selection by animals: statistical design andanalysis for field studies. Second edition. Kluwer Academic Publishers,Norwell, Massachusetts, USA.

McCullagh, P., and J. A. Nelder. 1989. Generalized linear models.Chapman and Hall, London, United Kingdom.

Miller,R.F., andJ.A.Rose.1999.Firehistoryandwestern juniper encroachmentin sagebrush steppe. Journal of Range Management 52:550–559.

Miller, R. F., T. J. Svejcar, and J. A. Rose. 2000. Impacts of western juniperon plant community composition and structure. Journal of RangeManagement 53:574–585.

Mulligan, E. M. 2015. Survival rates and cause-specific mortality of muledeer in south-central Oregon. Thesis, Oregon State University, Thesis,Corvallis, USA.

Mysterud, A., and E.Østbye. 1999. Cover as a habitat element for temperateungulates: effects on habitat selection and demography. Wildlife SocietyBulletin 27:385–394.

Natural Resources Conservation Service. 2015. Sage Grouse Initiative 2.0Investment Strategy, FY 2015–2018. http://www.sagegrouseinitiative.com/wp-content/uploads/2015/08/SGI2.0_Final_Report.pdf. Accessed25 Aug 2017.

Nielson, R. M., and H. Sawyer. 2013. Estimating resource selection withcount data. Ecology and Evolution 3:2233–2240.

Northrup, J.M.,C.R.Anderson, andG.Wittemyer. 2015.Quantifying spatialhabitat loss from hydrocarbon development through assessing habitatselection patterns of mule deer. Global Change Biology 21:3061–3970.

Oregon Department of Fish and Wildlife. 2003. Oregon’s mule deermanagement plan. OregonDepartment of Fish andWildlife, Portland, USA.

Oregon State University PRISM Climate Group. 2012. http://prism.oregonstate.edu. Accessed 24 Aug 2017.

Parker, K. L., C. T. Robbins, and T. A. Hanley. 1984. Energy expendituresfor locomotion by mule deer and elk. Journal of Wildlife Management48:474–488.

Peek, J. M., B. Dennis, and T. Hershey. 2002. Predicting population trendsof mule deer. Journal of Wildlife Management 66:729–736.

Peek, J. M., J. J. Korol, D. Gay, and T.Hershey. 2001. Overstory-understorybiomass changes over a 35-year period in southcentral Oregon. ForestEcology and Management 150:267–277.

Pomeroy, J., J. Parviainen, N. Hedstrom, and D. Gray. 1998. Coupledmodelling of forest snow interception and sublimation. HydrologicalProcesses 12:2317–2337.

Prokopenko, C. M., M. S. Boyce, and T. Avgar. 2017. Extent-dependenthabitat selection in a migratory large herbivore: road avoidance acrossscales. Landscape Ecology 32:313–325.

Raedeke, K. J., and R. D. Taber. 1982.Mechanisms of population regulationin western Washington forests for Cervus and Odocoileus. TransactionsInternational Congress of Game Biology 14:69–79.

Robinette, W. L., C. H. Baer, R. E. Pillmore, and C. E. Knittle. 1973.Effects of nutritional change on captive mule deer. Journal of WildlifeManagement 37:312–326.

Robinette, W. L., O. Julander, J. S. Gashwiler, and J. G. Smith. 1952.Winter mortality of mule deer in Utah in relation to range condition.Journal of Wildlife Management 16:289–299.

Rogala, J.K.,M.Hebblewhite, J.Whittington,C.A.White, J.Coleshill, andM.Musiani. 2011. Human activity differentially redistributes large mammals inthe Canadian Rockies National Parks. Ecology and Society 16:24.

Rowland, M. M., M. J. Wisdom, R. M. Nielson, J. G. Cook, R. C. Cook,B. K. Johnson, P. K. Coe, J. M. Hafer, B. J. Naylor, D. J. Vales, R. G.Anthony, E. K. Cole, C. S. Danilson, R. W. Davis, F. Geyer, S. Harris,L. L. Irwin, R. McCoy, M. D. Pope, K. Sager-Fradkin, and M. Vavra.2018. Modeling elk nutrition and habitat in western Oregon andWashington. Wildlife Monographs in press.

Rowland, M. M., L. H. Suring, R. J. Tausch, S. Geer, and J. J. Wisdom.2011. Dynamics of western juniper woodland expansion into sagebrushcommunities in central Oregon. Natural Resources and EnvironmentalIssues 16:1–11.

Rowland, M. M., M. J. Wisdom, B. K. Johnson, and J. G. Kie. 2000. Elkdistribution and modeling in relation to roads. Journal of WildlifeManagement 64:672–684.

Sage-Grouse Conservation Partnership. 2015. The Oregon Sage-GrouseAction Plan. Governor’s Natural Resources Office, Salem, Oregon, USA.

Salwasser, H. J. 1979. Ecology and management of the Devil’s GardenInterstate deer herd and range. Dissertation, University of California,Berkeley, USA.

Sawyer, H., N. M. Korfanta, R. M. Nielson, K. L. Monteith, and D.Strickland. 2017. Mule deer and energy development: long-term trends ofhabituation and abundance. Global Change Biology 23:4521–4529.

Sawyer, H., R. M. Nielson, F. Lindzey, and L. L. McDonald. 2006. Winterhabitat selection of mule deer before and during development of a naturalgas field. Journal of Wildlife Management 70:396–403.

Senft, R. L., M. B. Coughenour, D. W. Bailey, L. R. Rittenhouse, O. E.Sala, and D. M. Swift. 1987. Large herbivore foraging and ecologicalhierarchies. Bioscience 37:789–799.

Shallow, J. R., M. A. Hurley, K. L. Monteith, and R. T. Bowyer. 2015.Cascading effects of habitat on maternal condition and life-historycharacteristics of neonatal mule deer. Journal of Mammalogy 96:194–205.

Sikes, R. S., and W. L. Gannon. 2011. Guidelines of the American Societyof Mammalogists for the use of wild mammals in research. Journal ofMammalogy 92:235–253.

Smith, S. M., P. R. Krausman, and G. Painter. 2015. Winter habitat use bymule deer in Idaho and Montana. Northwestern Naturalist 96:50–70.

Stankowich, T. 2008. Ungulate flight responses to human disturbance: areview and meta-analysis. Biological Conservation 141:2159–2173.

Stewart, K. M., R. T. Bowyer, J. G. Kie, and M. A. Hurley. 2010. Spatialdistributions of mule deer and North American elk: resource partitioningin a sage-steppe environment. American Midland Naturalist 163:400–412.

Tollefson, T. N., L. A. Shipley, W. L. Myers, D. H. Keisler, and N.Dasgupta. 2010. Influence of summer and autumn nutrition on bodycondition and reproduction in lactating mule deer. Journal of WildlifeManagement 74:974–986.

Torbit, S. C., L. H. Carpenter, D. M. Swift, and A. W. Alldredge. 1985.Differential loss of fat and protein by mule deer during winter. Journal ofWildlife Management 49:80–85.

U.S. Geological Survey. 2001. National Gap Analysis Program Land CoverData � version 2. U.S. Geological Survey, Reston, Virginia, USA.

Unsworth, J. W., D. F. Pac, G. C.White, and R. M. Bartmann. 1999. Muledeer survival in Colorado, Idaho, and Montana. Journal of WildlifeManagement 63:315–326.

Venables, W. N., and B. D. Ripley. 2002. Modern applied statistics with S.Springer, New York, New York, USA.

Wallmo, O. C. 1981. Mule and black-tailed deer of North America.University of Nebraska Press, Linclon, USA.

Webb, S. L., M. R. Dzialak, K. L. Kosciuch, and J. B. Winstead. 2013.Winter resource selection by mule deer on theWyoming-Colorado borderprior to wind energy development. Rangeland Ecology and Management66:419–427.

Wickstrom,M. L., C. T. Robbins, T. A. Hanley, D. E. Spalinger, and S. M.Parish. 1984. Food intake and foraging energetics of elk and mule deer.Journal of Wildlife Management 48:1285–1301.

Wiens, J. A. 1989. Spatial scaling in ecology. Functional Ecology 3:385–397.


http://www.sagegrouseinitiative.com/wp-content/uploads/2015/08/SGI2.0_Final_Report.pdf

http://www.sagegrouseinitiative.com/wp-content/uploads/2015/08/SGI2.0_Final_Report.pdf

http://prism.oregonstate.edu

http://prism.oregonstate.edu

Willms, W., A. McLean, R. Tucker, and R. Ritcey. 1979. Interactionsbetween mule deer and cattle on big sagebrush range in British Columbia.Journal of Range Management 32:299–304.

Wisdom,M.J.,A.A.Ager,H.K.Preisler,N. J.Cimon, andB.K. Johnson.2004.Effects of off-road recreation on mule deer and elk. Transactions of the 69thNorth American Wildlife and Natural Resources Conference 69:531–550.

Witt, C., and P. L. Patterson. 2011. Big sagebrush in pinyon-juniperwoodlands: using forest inventory and analysis data as a management toolfor quantifying and monitoring mule deer habitat. Natural Resources andEnvironmental Issues 16:27.

Wood, A. K. 1988. Use of shelter by mule deer during winter. PrairieNaturalist 20:15–22.

Workman, G. W., and J. B. Low. 1976. Mule deer decline in the West: asymposium. Utah Agricultural Experiment Station, Logan, USA.

Zalunardo, R. A. 1965. The seasonal distribution of a migratory mule deerherd. Journal of Wildlife Management 29:345–351.

Associate Editor: Kelly Proffitt.

SUPPORTING INFORMATIONAdditional supporting information may be found in theonline version of this article at the publisher’s website.


multiscale models of habitat use by mule deer in winter · research article multiscale models of...

Documents