113 YUKON MOOSE: I. SEASONAL RESOURCE SELECTION BY MALES AND FEMALES IN A MULTI-PREDATOR BOREAL ECOSYSTEM Alice M. McCulley, Katherine L. Parker, and Michael P. Gillingham Natural Resources and Environmental Studies, University of Northern British Columbia, Prince George, British Columbia V2N 4Z9 Canada ABSTRACT: Moose (Alces alces) in Yukon experience an extreme range of thermal conditions, highly variable snow depths, natural and anthropogenic disturbances, predation by wolves and griz- zly bears, and hunting pressure. Our objective was to identify variables that best explained hab- itat-selection patterns of moose in south-central Yukon for use in land-use planning and impact assessment. We evaluated selection of land-cover class, elevation, aspect, predation risk, and harvest vulnerability using resource selection functions. We created pooled models for males and females by averaging models for individuals by sex and season. Selection of shrub-dominated land cover highlighted the importance of forage accessibility throughout the year. Selection for elevation, as- pect, and cover changed throughout the year, as influenced by climatic conditions. By selecting mixed cover types during calving and summer, female moose presumably balanced needs for both cover and forage. Males minimized harvest vulnerability during rut. Moose, in general, demon- strated highly variable habitat selection; however, consistent individual responses between sexes supported trends identified by pooled selection coefficients, as well as detected trends among males and females. The greatest amount of individual variation occurred during the growing season and the least amount during late winter, suggesting that climatic factors limited the options available to moose at a critical time of the year. ALCES VOL. 53: 113–136 (2017) Key Words: Alces, boreal, habitat selection, individual variation, predation risk, resource selection Habitat selection is a hierarchical pro- cess in which an animal first chooses a general place in which to live (a habitat or habitats) and then makes subsequent deci- sions about how it moves within the habi- tats and responds to environmental factors (Anderson et al. 2005). Many factors, even those beyond the extent of the home range, influence how animals respond to their en- vironment (Kie et al. 2002, Bowyer and Kie 2006). Effective wildlife management ben- efits from understanding seasonal selection patterns and how animals respond to key habitat variables resulting in those patterns. Moose (Alces alces) use a wide variety of habitats in various successional stages throughout the boreal forest (Kelsall et al. 1977). As with most large herbivores, habi- tat selection is driven by the need to meet nu- tritional requirements with adequate forage and cover, and to minimize mortality risk. Moose feed on a wide variety of plant spe- cies (Miquelle and Jordan 1979), and require large amounts of forage because of their large body size (Renecker and Hudson 1992, Renecker and Schwartz 2007). They adjust foraging behaviour in response to seasonal changes in forage quality and quantity (An- dersen and Saether 1992). Habitat selection by moose is strongly influenced by ambient conditions: both vegetative cover and topog- raphy affect microclimate, snow depth and RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 114 density, and predation risk (Mysterud and Ostbye 1999). Moose use cover and topog- raphy in all seasons to moderate extremes of cold (<-30 °C in winter) and heat stress (Renecker and Hudson 1986), and increase use of cover with increasing snow depth, density, or crusting (Telfer 1970, Van Bal- lenberghe and Peek 1971). Wolves (Canis lupus), grizzly bears (Ursus arctos), and black bears (Ursus americanus) are the pri- mary predators of moose in boreal systems, and deep snow may impair defensive capa- bilities of moose against wolves (Peterson and Allen 1974). The vulnerability of moose to predation is also influenced by their age, size, and body condition, population den- sities of both moose and predators, and the availability of alternative prey (Hayes et al. 2000). Anti-predator behaviour varies with the degree of predation risk, group size, experience, and sex. Moose behaviour and population dynamics are further affected by moose density, which itself is influenced by hunter density, timing of the hunting season, and accessibility (Baskin et al. 2004). In addition to the effects of seasonal changes in forage, climate, and risk, differ- ences in body size and reproductive roles between male and female moose can fur- ther modify habitat selection. Sexual seg- regation, as the differential use of space by the sexes outside of the breeding season, is widespread in sexually dimorphic ungulates such as moose. Most hypotheses for sexual segregation relate to reproductive strategies, sexual dimorphism, and/or social factors (Main et al. 1996). These hypotheses suggest that males should maximize body condition before rut and minimize energy expendi- tures during winter, even if predation risk increases. Larger rumen size allows males to target large quantities of coarse forage and larger body size puts them at less risk of pre- dation. In contrast, females have smaller di- gestive capacity, greater energetic demands of gestation, parturition and lactation, and potentially higher exposure to predation risk. Females should feed more frequently while targeting areas with higher-quality forage in close proximity to cover (Barboza and Bowyer 2000) to meet minimum resource requirements while maximizing security of calves (Main and Coblentz 1990). Relatively few studies have addressed habitat requirements and limiting factors of moose in Yukon, where the distribution of moose reaches some of the most north- ern limits of the species’ range. Moose in Yukon experience extreme thermal condi- tions, highly variable snow depths, natural and anthropogenic disturbances that alter land cover, predation by wolves and griz- zly bears, and hunting pressure (Kelsall and Telfer 1974). Moose are a focal species of many northern communities for subsistence, cultural, economic and recreational values (Timmerman and Rodgers 2005). The overall objective of this study was to identify variables that best explained sea- sonal resource selection patterns of moose in south-central Yukon for subsequent use in land-use planning and impact assessment. Prior to our study, relatively little was known about the distribution, abundance, or habi- tat use of Yukon moose outside of the early winter, post-rut period (Larsen et al. 1989, Gasaway et al. 1992, Florkiewicz and Henry 1994, Boertje et al. 1995, Keith 1995, Mauer 1998, Hayes and Harestad 2000, Hayes et al. 2000). We incorporated topographical attrib- utes, land cover, predation risk from wolves and grizzly bears, and harvest vulnerability into resource selection models. Resource selection functions (RSF) describe the rela- tive selection of attributes used by an animal (Manly et al. 2002) and provide a broad-scale perspective of general selection patterns on the landscape (Boyce and McDonald 1999). They help define which attributes have the greatest influence on habitat use and how ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 115 selection of different attributes can change among seasons. We hypothesized that moose would optimize survival by selecting land- cover classes that minimize energy losses in winter and maximize potential forage intake during the growing season. Because of phys- iological and reproductive differences, we predicted that male and female moose would use different selection patterns to meet these needs. In addition to metabolic demands, moose also faced exposure to mortality risk from predators and hunters. We predicted that females would reduce exposure to pre- dation risk throughout the year, particularly when calves were young, and that males would reduce exposure to harvest risk dur- ing the hunting season. STUDY AREA The South Canol study area in south- central Yukon was 130 km east of White- horse and 100 km north of Teslin, between 60.4743 and 61.9082° N latitude, and 128.9699 and 135.2570° W longitude. Covering approximately 35,000 km2, it ex- tended north from Johnson’s Crossing, east of Lake Laberge, west of Frances Lake, and south of the community of Ross River (Fig. 1). Climate in the South Canol area was charac- terised by short cool summers and long cold winters. Mean annual precipitation ranged from 500–650 mm with most as snow in winter. Mean annual temperature was -3 °C, with a mean January temperature of -20 °C and a mean July temperature of 10 °C (Yukon Ecoregions Working Group 2004). Unlike many other areas of the Yukon, the South Canol area had few wildfires in the past 60 years (Yukon Department of Energy, Mines and Resources 2004). The area was in the Boreal Cordillera Ecozone and includes the Pelly Mountains Ecoregion with small por- tions of the Southern Lakes Ecoregion. The Pelly Mountain Ecoregion is a rolling pla- teau topped by numerous mountain peaks and dissected by small rivers. The South- ern Lakes Ecoregion is characterized by dissected plateaus, rolling hills, and broad Fig. 1. South Canol moose study area located in south-central Yukon, Canada. RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 116 valleys occupied by lakes and rivers (Yukon Ecoregions Working Group 2004). The entire area is within the sporadic discontinuous per- mafrost zone. Shrub and dwarf shrub tundra vegetation occurred above 1,350 m above sea level (a.s.l.), and coniferous and mixed for- ests occurred below 1,350 m a.s.l. In 2007, the average population density was 241 moose/1000 km2 within a 6,735-km2 core portion of the study area, a density 1/3 higher than the Yukon average of 158 moose/1000 km2 (Florkiewicz et al. 2008). There were approximately 22 calves, 18 year- lings, and 76 males for every 100 adult fe- male moose. The area also encompassed the ranges of 5 woodland caribou (Rangifer ta- randus) herds, including the Wolf Lake, Pelly, Carcross, Atlin, and Laberge herds (Yukon Ecoregions Working Group 2004). Stone’s sheep (Ovis dalli stonei) used the Big Salmon Range in the northern part of the study area, and wolves, grizzly bears, and black bears occurred throughout the area. Wolf density was estimated as 8–12 wolves/1,000 km2 (R. Ward, Yukon Department of Environ- ment, pers. comm., Baer 2011), and the griz- zly bear estimate was 15.2 bears/1000 km2 (R. Florkiewicz, Yukon Department of En- vironment, pers. comm.). Black bear density was unknown, but assumed lower than griz- zly bears and with less predatory impact than either wolves or grizzly bears. The South Canol area falls mainly within the traditional territory of the Tes- lin Tlingit First Nation and also includes portions of the Ta’an Kwäch’än, Kwanlin Dun, and Kaska traditional territories. Eight Game Management Subzones (GMS), one big-game outfitting concession, and portions of 17 registered trapping concessions were in the study area. Approximately half of the trapping concessions were operated by local First Nations (T. Boyes, Teslin Tlingit Council, pers. comm.). The average annual reported moose harvest (2002–2006) in the area was approximately 1.7% of the popula- tion (Florkiewicz et al. 2008). This estimate did not include unreported First Nations harvest which was estimated to equal the li- censed harvest. Very little development was present in the study area. The South Canol Road, as a seasonal unpaved highway that extends from Johnson’s Crossing to Ross River, pro- vided access through the eastern portion of the study area (Fig. 1). The Pelly Mountains Ecoregion is considered rich in mineral de- posits (Yukon Ecoregions Working Group 2004), but only one hard-rock mineral claim (Tintina Mines Ltd.) was active during our study. An exploration road was upgraded in 2008 and extended 76 km from the South Canol Road to Red (Slate) Mountain within the study area. This exploration road was ac- cessible by ATV and 4x4 vehicle in summer and by snowmobile in winter. Several small placer-mining operations, some with air- strips and limited roads, were present in the remote northwest portion of the study area. METHODS Animal captures and telemetry data Twenty-seven moose (9 males, 18 fe- males) were captured between 26 February and 27 March 2008 and fitted with global positioning satellite (GPS) collars (15 col- lars: Lotek GPS4400M ARG, New Market, Ontario, Canada; 12 collars: Habit Research Inc., Victoria, British Columbia, Canada). The GPS collars were programmed to ac- quire locations 6–8 times per day over a 24-h cycle and periodically uploaded data to the ARGOS satellite. Location data were downloaded from ARGOS once per month. We used Spatial Viewer (M. Gillingham, un- published Visual Basic program) to examine movement patterns of individual animals and to identify and eliminate errant loca- tion points (i.e., those points that were an improbable distance from previous points) ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 117 that were likely the result of GPS errors. Fifteen collars provided 2 complete years of location data, and 9 collars transmitted for at least one full season. Three collars transmit- ted for less than one season and these were not used in analyses. We retrieved 77,309 valid locations from the 24 collars (8 males and 16 females for our selection analyses) with average fix rates of 88 ± 4% (x− ± SE) for males and 66 ± 7% for females. We defined 5 seasons based on moose life-history and habitat characteristics: Calv- ing (15 May–30 Jun), Summer (1 Jul–14 Aug), Rut (15 Aug–31 Oct), Early Winter (1 Nov–28 Feb), and Late Winter (1 Mar–14 May). These dates generally corresponded with the timing of seasons in other moose studies in Yukon, Alaska, and British Columbia (Larson et al. 1989, Miquelle et al. 1992, Gillingham and Parker 2008a, b). In our analyses we consid- ered 4 groups of individual moose based on sex and reproductive status: we compared males to females, and females without calves to females with calves. Calf status of females was based on aerial surveys during mid-June, October–November, and March from June 2008 through March 2010. Only females of known calf status were used in analyses. For example, if a moose had a calf in June and then again in November (i.e., Calving and Early Winter), we also assumed the calf was present during Summer and Rut. If she was alone in November, however, then she was classified as unknown calf status during Summer and Rut, and with no calf during Early Winter. We conducted similar analyses for each of the 4 groups of moose, including male moose dur- ing the Calving period, to compare selection patterns between sexes at this time of year when the onset of greening occurred. Study design We used resource selection functions (RSFs) to assess habitat selection of moose in the South Canol study area. Coefficients were estimated using logistic regression software (Stata version 12, StataCorp 2011) for the parameters of exponential RSFs with used and available points for individual an- imals (Design 3: Thomas and Taylor 1990, 2006). Used points were the GPS fixes from each radio-collared moose. Availability was identified by selecting 5 random points from within a buffer surrounding each location point. The radius of each buffer was based on the 95th percentile movement distance of each individual in each season. We assumed the individual could have potentially moved anywhere within this buffer over the period represented by the GPS fix. We then used raster remote-sensing software to query the attributes of each used and available point. Attributes for resource selection Land-cover composition.-- We devel- oped a land-cover classification using Earth Observation for Sustainable Development of Forests (EOSD) land-cover information, a digital elevation model (DEM), and Na- tional Topographic Data Base (NTDB) hy- drology information (www.geomaticsyukon. ca). EOSD (circa 2002) was interpreted from Landsat-7 imagery with 25-m resolution and was used to classify land-surface ele- ments (e.g., vegetation, water, rock) (Wulder et al. 2003). Using remote-sensing software (Geomatica 10.3, PCI Geomatics 2009), we combined 26 EOSD cover classes with the above-mentioned data sources to produce 8 land-cover classes relevant to moose ecol- ogy (Table 1). Classes were combined based on similarities in vegetation and elevation. Grouping classes also had the effect of im- proving the accuracy of EOSD data, which approached 75–80% (M. Waterreus, Yukon Department of Environment, pers. comm.). Topographic variables.-- Elevation and aspect were extracted from a DEM using Arc- Map (ArcMap 9.3, ESRI 2006). We entered elevation as a quadratic in all selection models www.geomaticsyukon.ca www.geomaticsyukon.ca RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 118 to be able to discriminate selection for mid-el- evation locations. To reduce the number of categorical variables, we converted aspect into 2 continuous variables: northness and eastness (Gillingham and Parker 2008a, b). Northness (the cosine of aspect) values range from 1.00 to -1.00, indicating north through south aspects. Eastness (the sine of aspect) values range from 1.00 to -1.00, indicating east through west aspects. The values for both northness and eastness must be interpreted to- gether to understand selection for aspect. For example, values near zero for both northness and eastness indicate no selection for aspect, whereas large negative values for both north- ness and eastness indicate selection for south- west aspects. Slopes of ≤ 1° were not assigned an aspect. Predation risk.--Predation risk to moose was defined using RSFs developed for data from GPS-collared wolves and grizzly bears in the Besa-Prophet area of northern British Columbia (Milakovic 2008; Milakovic et al. 2011, 2012). These pred- ator-selection models included elevation, slope, aspect, vegetation type, and fragmen- tation (an index of vegetation diversity). We assumed that risk of predation to moose by wolves and grizzly bears was directly re- lated to selection values for the predators. We generated predation-risk surfaces for moose in the South Canol area as GIS raster layers that defined selection value to wolves or grizzly bears in each season by apply- ing the coefficients from the Besa-Prophet predator-selection models to each 25 x 25-m pixel, based on its topographic and land- cover features. We scaled values from zero to 1 to standardize selection surfaces and to facilitate comparison between seasons. We created 5 seasonal wolf risk surfaces, and 3 seasonal grizzly bear risk surfaces (with no risk during hibernation seasons). In addition to predation risk as a variable in resource selection models, we calculated the average predation risk that each individ- ual moose was exposed to in each season (based on predation-risk values predicted from the predation-selection models (as above) at GPS locations). We set a minimum Table 1. Description of 8 land-cover classes used by radio-collared moose in the South Canol study area of south-central Yukon, Canada. Land-cover Class Description Conifer Spruce, pine, or subalpine fir covering 75% or more of total basal area. Mixed Wood A mix of conifers or deciduous trees with neither exceeding 75% of total basal area. Lowland Shrub Areas below 1,300 m a.s.l. with ≥ 20% vegetative cover of which at least 33% is shrub species. Also includes deciduous trees exceeding 75% of total basal area. Upland Shrub Areas above 1,300 m a.s.l. with ≥ 20% vegetative cover of which at least 33% is shrub species. Also includes deciduous trees exceeding 75% of total basal area. Alpine Areas above 1,300 m a.s.l with ≥ 20% vegetative cover. Includes snow, ice, exposed land, and areas with no data above 1,300 m a.s.l. Excludes Upland Shrub. Lowland Open Areas below 1,300 m a.s.l with ≥ 20% vegetative cover, or exposed land with < 5% vegetation. Excludes Lowland Shrub. Water Lakes, ponds, reservoirs, rivers, streams, or creeks. Riparian Areas within 25 m of small (1-line1) water courses; areas within 100 m of larger water courses (2-line) and water bodies. Includes wetlands. 11-line streams are smaller streams indicated on 1:50,000 maps with a single line, whereas 2-line streams are indicated using 2 lines to delineate the shores of large rivers. ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 119 of 100 location points per individual in each season to be included in calculations. We used a repeated measures 2-way analysis of variance (ANOVA) to investigate whether these predation-risk values differed between sexes and among seasons. Post-hoc anal- yses for significant (α = 0.05) effects were assessed using Bonferroni-corrected con- fidence intervals on marginal means (Mar- gins in Stata version 12, StataCorp 2011). We used 1-way ANOVAs to determine if the average predation risk incurred by females with and without calves differed in each season. Harvest vulnerability.--Harvest infor- mation was collected from 63 First Nations and licensed resident hunters who harvested moose in the study area during the 5 years before moose telemetry locations (i.e., 2004–2008). We collected information about the characteristics of sites where moose were killed. Interviewees were asked to specify on a map where the kill occurred, to comment on proximity to road or water access, and to identify which land-cover class the ani- mal was in, based on examination of several representative photographs of the different land-cover classes (McCulley 2015). We developed a raster surface that de- fined harvest vulnerability to male moose during Rut using the location data collected from 63 hunters, land-cover classes, and NTDB hydrology and road information. This surface was based on a matrix that in- cluded each land-cover class, in combina- tion with distance to both roads and large rivers (> 500 m or < 500 m from each). The number of moose killed, as recorded in the interviews, was entered into each cell of the matrix. Ninety-nine kills were docu- mented in 13 cover-access combinations, with over 80% occurring in the Riparian, Wetland, or Water land-cover classes. To keep harvest vulnerability as a continuous variable, we converted these values into a proportion of the total kills (n = 99). We then assigned the appropriate proportion to each 25 x 25-m pixel based on land-cover class and distance to access. As with predation risk, we calculated the average harvest vulnerability that each individual was exposed to during Rut. Only individual moose with a minimum of 100 lo- cation points were included in calculations. We used a 1-way non-parametric ANOVA (Kruskal-Wallis test because data could not be normalized) to investigate whether av- erage vulnerability to harvest differed (α = 0.05) between males and females. Modeling procedures We used an information-theoretic ap- proach to evaluate seasonal resource se- lection by moose (Burnham and Anderson 2002). First, we developed a set of 10 a priori, ecologically plausible models to describe re- source selection (Table 2). We evaluated the importance of land-cover class, elevation, aspect, predation risk, and harvest vulnera- bility in the models using selection coeffi- cients (β i ) from logistic regression. A set of 6 models was tested on location data from all moose; we ran 4 additional models for male moose during Rut. We used statistical software for all modeling procedures (Stata version 12, StataCorp 2011). Deviation cod- ing (using DESMAT add-in) was used for all categorical variables. To avoid issues of sep- aration, we dropped both used and available points in land-cover classes where there were ≤ 4 used or available points (Menard 2002). Consequently, very strong avoidance of a particular land-cover class may not always be reflected in the final RSFs. We ranked the model sets using Akaike’s Information Cri- terion (AIC c ; Burnham and Anderson 2002) for small sample sizes. Akaike’s weights (w i ) indicate the relative weight of evidence for the top model being the best among the can- didate models. We selected a model as the RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 120 likely top model if w i ≥ 0.95. We used k-fold cross-validation averaged across 5 random subsets and an averaged Spearman’s rank correlation (r s ) to determine the predictive ability of each top model; values of r s > 0.70 indicated good model performance (Boyce et al. 2002). If the top model for each animal had a w i < 0.95, we averaged the selection coefficients (β i ) from the set of top candidate models for which the sum of their respec- tive w i ’s was ≥ 0.95 (Burnham and Anderson 2002). When model averaging was required, averaged coefficients from each component model were weighted by their corresponding w i values; we used a selection coefficient of zero for any parameters not included in an individual’s final model to avoid overem- phasizing the importance of coefficients that were only in some individual models. Our estimates of pooled variance and standard errors (SE) were based on differences be- tween the coefficient for each model being averaged and the average coefficient across models (weighted by w i when averaging within competing models for an individual and weighted equally when averaging across individuals; e.g., Murtaugh 2007). Because each individual has a variance associated with its estimate, however, we also included (in an additive manner) the variance of each coefficient in each model in our calculations. Once we had a single model for each individual, we produced a pooled RSF for males and for females by averaging models across all individuals in that group (using either the top model or an averaged model for each individual as described above) by season. Each model in a sex-season set was equally weighted to avoid over-representa- tion of any individual moose. Models also were developed for females with and without a calf to determine if calf presence affected habitat selection during Calving – a time when newborn calves are most vulnerable. RESULTS Seasonal mortality risk Average exposure to wolf risk was lowest during Early Winter for both male and female moose (Fig. 2A). It was high- est during Rut for females and Late Winter for males. Average exposure to grizzly bear Table 2. Candidate resource selection models for moose in the South Canol study area, Yukon, Canada. M = males, F = females, L = land-cover class, E = elevation, A = aspect, P = predation risk1, H = harvest vulnerability. Model Late Winter Calving Summer Rut Early Winter M F M F M F M F M F L ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ E + E2 + A ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ P ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ L + E + E2+ A ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ L + E + E2+ A + P ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ L + P ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ ¸ H ¸ L + H ¸ L + P + H ¸ L + E + E2+ A + P + H ¸ 1Both wolf and grizzly bear predation risk during Calving, Summer and Rut; only wolf risk during Early and Late Winter. ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 121 risk was lowest during Summer and highest during Rut for both sexes (Fig. 2B); differ- ences between males and females were not significant (wolves: F 1, 88 = 0.07, P = 0.79; bears: F 1, 44 = 0.26, P = 0.61). Predation risk was a function of season (wolves: F 4, 88 = 5.51, P < 0.01; bears: F 2, 44 = 12.08, P < 0.01); wolf risk during Early Winter was lower than during Summer. Calf presence had a significant effect on the wolf risk en- countered by females only during Late Win- ter (F 1, 17 = 6.95, P = 0.02), when females with calves used areas with lower risk than females without calves (Fig. 3A). In spring, females with calves used areas with high bear risk (Fig. 3B), although this was not statistically different from the females with- out calves (F 1, 20 = 3.52, P = 0.08). There also was no difference between male and female moose in exposure to harvest risk during Rut (Kruskal-Wallis: χ2 = 0.91, df = 1, P = 0.34). Seasonal habitat selection The Late Winter models were repre- sented by the fewest moose (5 males, 7 fe- males), and the Calving models were based Fig. 2. Average (x− ± SE) seasonal predation risk by A) wolves and B) grizzly bears to male (n = 8) and female radio-collared moose in the South Canol area of south-central Yukon, Canada. LW = Late Winter, CA = Calving, SU = Summer, RU = Rut, EW = Early Winter. Numbers indicate sample size of females. RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 122 on the most moose (8 males, 13 females). Each seasonal pooled model included eleva- tion, aspect, predation risk, and all 8 land- cover classes; however, both the sign (+/-) and significance of the coefficients varied seasonally (Tables 3, 4). Based on the seasonal pooled models, male moose selected for mid-elevations in all seasons (as indicated by positive β i for elevation and negative β i for elevation2). Females had similar selection patterns ex- cept during Calving and Summer when elevation was not a significant factor and most variation occurred among individuals. Males selected for west aspects (i.e., sig- nificant negative eastness and insignificant northness) during Calving and Early Win- ter (Table 3), and for southeast aspects (i.e., significant positive eastness and negative northness) in Late Winter as did females. Females selected for west aspects during the Rut and northwest aspects in Early Win- ter. Relative to predation risk, both sexes selected habitats that had significant grizzly Fig. 3. Average (x− ± SE) seasonal predation risk by A) wolves and B) grizzly bears to female radio-col- lared moose with and without calves in the South Canol area of south-central Yukon, Canada. LW = Late Winter, CA = Calving, SU = Summer, RU = Rut, EW = Early Winter. Numbers indicate sample size. * indicates significant difference. ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 123 Se as on La te W in te r C al vi ng Su m m er M od el E1 + A 1 + P 1 + L 1 E + A + P + L E + A + P + L C oe f SE + + + - - - n/ a C oe f SE + + + - - - n/ a C oe f SE + + + - - - n/ a El ev at io n (k m ) 10 .8 7 2. 10 2 0 0 3 0 6. 62 1. 49 4 1 1 2 0 20 .2 0 2. 20 5 0 1 1 0 El ev at io n2 (k m ) -6 .2 3 1. 00 1 2 0 2 0 -4 .2 9 0. 71 1 1 1 5 0 -9 .9 5 1. 01 1 0 1 5 0 Ea st ne ss 0. 07 0. 02 2 0 1 2 0 -0 .2 3 0. 03 1 1 1 5 0 0. 00 0. 03 3 0 0 4 0 N or th ne ss -0 .2 8 0. 02 0 1 0 4 0 -0 .0 3 0. 04 1 1 3 3 0 0. 03 0. 02 4 1 1 1 0 W ol f R is k 0. 20 0. 04 3 0 0 2 0 0. 01 0. 07 2 1 2 3 0 0. 00 0. 05 2 1 1 3 0 B ea r R is k 1. 24 0. 13 7 0 1 0 0 0. 05 0. 04 3 1 2 1 0 H ar ve st ri sk C on ife r -0 .4 6 0. 03 0 0 0 5 0 -0 .6 1 0. 05 0 0 1 7 0 -0 .4 7 0. 03 0 0 3 4 0 A lp in e -0 .1 5 0. 04 1 0 0 1 3 -0 .0 7 0. 04 0 0 2 0 6 0. 05 0. 05 2 2 0 1 2 Lo w la nd S hr ub 0. 08 0. 03 2 1 0 2 0 0. 44 0. 05 5 2 0 1 0 -0 .0 5 0. 05 2 2 0 3 0 U pl an d Sh ru b 0. 16 0. 04 3 0 0 1 1 0. 70 0. 12 4 2 1 1 0 0. 72 0. 05 6 0 0 0 1 M ix ed W oo d -0 .0 3 0. 05 2 2 0 1 0 -0 .1 9 0. 06 1 1 2 4 0 -0 .4 0 0. 04 0 1 0 6 0 R ip ar ia n 0. 53 0. 04 4 1 0 0 0 0. 17 0. 06 3 2 1 2 0 0. 13 0. 05 3 1 1 2 0 W at er 0. 08 0. 03 1 0 1 0 3 -0 .7 1 0. 13 0 1 1 2 4 0. 03 0. 02 1 0 0 1 5 Lo w la nd O pe n -0 .2 1 0. 06 1 0 0 2 2 0. 27 0. 09 1 4 0 1 2 -0 .0 2 0. 03 1 1 0 1 4 co ns ta nt -5 .9 9 1. 11 1 1 0 3 0 -4 .5 1 0. 75 1 1 1 5 0 -1 1. 56 1. 19 1 1 1 4 0 1 E = e le va tio n, A = a sp ec t ( ea st ne ss a nd n or th ne ss ), P = pr ed at io n ris k (w ol f a nd g riz zl y be ar ), L = la nd -c ov er c la ss , H = h ar ve st v ul ne ra bi lit y. Ta bl e 3. S el ec tio n co ef fic ie nt s (C oe f ± S E) in p oo le d re so ur ce s el ec tio n m od el s fo r m al e m oo se , c al cu la te d by s ea so n fo r 8 G PS -c ol la re d m al es fr om 2 00 8– 20 10 in th e So ut h C an ol a re a of so ut h- ce nt ra l Y uk on , C an ad a. A s a n in di ca tio n of v ar ia bi lit y am on g in di vi du al s, th e nu m be r o f i nd iv id ua ls th at si gn ifi ca nt ly se le ct ed fo r o r a ga in st e ac h pa ra m et er is sh ow n un de r + + or - -, re sp ec tiv el y. In di vi du al s w ith in si gn ifi ca nt p ar am et er s i n th e m od el a re sh ow n un de r + a nd -; n/ a re fe rs to n um be r o f a ni m al s w ith ou t a p ar am et er . S ig ni fic an t c oe ffi ci en ts a re in b ol d. Ta bl e 3 co nt in ue d. ... RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 124 Se as on R ut Ea rly W in te r M od el E + A + P + L + H E + A + P + L C oe f SE + + + - - - n/ a C oe f SE + + + - - - n/ a El ev at io n (k m ) 10 .2 0 1. 09 7 0 0 1 0 19 .7 4 1. 34 6 1 0 0 0 El ev at io n2 (k m ) -4 .6 6 0. 42 1 0 0 7 0 -8 .6 6 0. 56 0 0 0 7 0 Ea st ne ss 0. 01 0. 02 5 0 1 2 0 -0 .0 9 0. 01 1 1 0 5 0 N or th ne ss 0. 25 0. 02 7 0 0 1 0 0. 03 0. 01 4 0 0 3 0 W ol f R is k -0 .0 7 0. 03 3 0 1 4 0 0. 04 0. 02 2 2 1 2 0 B ea r R is k 0. 09 0. 05 4 0 1 3 0 H ar ve st ri sk -6 .4 7 0. 86 2 0 0 4 2 C on ife r -0 .5 9 0. 02 0 0 0 8 0 -0 .7 8 0. 03 0 0 0 7 0 A lp in e -0 .3 2 0. 05 1 0 2 5 0 -0 .5 1 0. 05 0 1 0 5 1 Lo w la nd S hr ub 0. 00 0. 03 4 0 1 3 0 0. 16 0. 02 5 1 1 0 0 U pl an d Sh ru b 0. 97 0. 03 8 0 0 0 0 0. 34 0. 03 5 0 0 1 1 M ix ed W oo d -0 .3 6 0. 02 0 0 1 7 0 -0 .5 0 0. 04 0 0 2 5 0 R ip ar ia n 0. 48 0. 03 7 0 1 0 0 0. 95 0. 03 7 0 0 0 0 W at er -0 .2 9 0. 04 0 0 1 3 4 0. 43 0. 10 2 0 1 2 2 Lo w la nd O pe n 0. 11 0. 04 2 0 0 2 4 -0 .0 9 0. 03 1 1 1 1 3 co ns ta nt -7 .1 4 0. 69 1 0 0 7 0 -1 2. 73 0. 81 0 0 1 6 0 Ta bl e 3. c on tin ue d ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 125 Ta bl e 4. S el ec tio n co ef fic ie nt s ( C oe f ± S E) in p oo le d re so ur ce se le ct io n m od el s f or fe m al e m oo se , c al cu la te d by se as on fo r G PS -c ol la re d fe m al es fr om 2 00 8– 20 10 in th e So ut h C an ol a re a of so ut h- ce nt ra l Y uk on , C an ad a. A s a n in di ca tio n of v ar ia bi lit y am on g in di vi du al s, th e nu m be r o f i nd iv id ua ls th at si gn ifi ca nt ly se le ct ed fo r o r a ga in st e ac h pa ra m et er is sh ow n un de r + + or - -, re sp ec tiv el y. In di vi du al s w ith in si gn ifi ca nt p ar am et er s i n th e m od el a re sh ow n un de r + a nd -; n/ a re fe rs to n um be r o f a ni m al s w ith ou t a p ar am et er . S ig ni fic an t c oe ffi ci en ts a re in b ol d. Se as on La te W in te r C al vi ng Su m m er M od el E1 + A 1 + P 1 + L 1 E + A + P + L E + A + P + L C oe f SE + + + - - - n/ a C oe f SE + + + - - - n/ a C oe f SE + + + - - - n/ a El ev at io n (k m ) 24 .6 8 4. 99 6 0 1 0 0 -0 .2 1 7. 42 5 3 2 3 0 2. 54 3. 63 4 1 2 3 0 El ev at io n2 (k m ) -1 0. 70 2. 58 0 1 0 6 0 -4 .0 6 4. 12 3 1 4 5 0 -1 .2 7 1. 69 2 3 1 4 0 Ea st ne ss 0. 11 0. 05 3 1 0 3 0 0. 00 0. 05 4 3 1 5 0 0. 02 0. 04 2 1 2 5 0 N or th ne ss -0 .1 7 0. 05 3 0 1 3 0 -0 .0 9 0. 05 5 1 3 4 0 -0 .0 6 0. 04 3 2 1 4 0 W ol f R is k 0. 02 0. 06 2 1 1 3 0 0. 00 0. 14 5 1 2 5 0 0. 00 0. 09 3 2 2 3 0 B ea r R is k 1. 55 0. 37 7 3 1 2 0 0. 07 0. 12 5 0 2 3 0 C on ife r -0 .1 0 0. 04 1 1 0 5 0 -0 .6 1 0. 09 1 1 3 8 0 -0 .1 1 0. 05 3 2 1 4 0 A lp in e 0. 00 0. 00 0 0 1 0 6 0. 21 0. 10 1 3 0 1 8 -0 .3 5 0. 06 0 0 0 2 8 Lo w la nd S hr ub 0. 09 0. 05 5 1 0 1 0 0. 14 0. 09 7 4 1 1 0 0. 04 0. 04 1 3 4 2 0 U pl an d Sh ru b -0 .0 1 0. 02 1 0 1 1 4 0. 54 0. 11 3 3 0 0 7 0. 32 0. 05 3 2 0 0 5 M ix ed W oo d -0 .0 5 0. 05 1 1 2 3 0 0. 13 0. 06 3 4 4 0 2 0. 15 0. 05 5 1 2 1 1 R ip ar ia n 0. 18 0. 06 5 0 0 2 0 -0 .0 1 0. 09 5 4 3 1 0 0. 27 0. 07 6 1 1 2 0 W at er 0. 00 0. 00 0 0 0 0 7 -0 .2 4 0. 07 0 2 2 2 7 -0 .1 6 0. 05 0 1 1 3 5 Lo w la nd O pe n -0 .1 0 0. 02 0 0 0 1 6 -0 .1 5 0. 05 0 0 2 2 9 -0 .1 5 0. 04 0 0 1 2 7 co ns ta nt -1 5. 71 2. 49 0 1 0 6 0 -0 .1 0 3. 47 2 2 3 6 0 -3 .3 9 1. 89 2 2 1 5 0 1 E = e le va tio n, A = a sp ec t ( ea st ne ss a nd n or th ne ss ), P = pr ed at io n ris k (w ol f a nd g riz zl y be ar ), L = la nd -c ov er c la ss , H = h ar ve st v ul ne ra bi lit y. Ta bl e 4 co nt in ue d. ... RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 126 Ta bl e 4. c on tin ue d Se as on R ut Ea rly W in te r M od el E + A + P + L + H E + A + P + L C oe f SE + + + - - - n/ a C oe f SE + + + - - - n/ a El ev at io n (k m ) 10 .3 1 1. 76 5 2 0 3 0 6. 97 0. 69 8 0 0 2 0 El ev at io n2 (k m ) -4 .0 6 0. 72 3 0 2 5 0 -3 .3 0 0. 28 1 1 0 8 0 Ea st ne ss -0 .0 6 0. 02 2 2 0 6 0 -0 .1 1 0. 02 2 1 2 5 0 N or th ne ss -0 .0 4 0. 02 4 0 0 6 0 0. 10 0. 03 7 0 0 3 0 W ol f R is k 0. 50 0. 08 6 2 2 0 0 0. 07 0. 03 4 2 0 4 0 B ea r R is k -0 .0 5 0. 08 5 2 2 1 0 H ar ve st ri sk C on ife r -0 .2 4 0. 03 1 1 2 6 0 -0 .6 4 0. 04 0 0 1 9 0 A lp in e -0 .3 4 0. 08 2 1 1 4 2 -0 .7 9 0. 09 1 2 0 6 1 Lo w la nd S hr ub -0 .0 2 0. 07 7 0 0 3 0 0. 22 0. 03 7 0 2 1 0 U pl an d Sh ru b 0. 42 0. 07 7 1 0 2 0 0. 49 0. 05 6 2 1 1 0 M ix ed W oo d -0 .1 1 0. 06 2 2 2 4 0 -0 .2 4 0. 05 1 1 3 5 0 R ip ar ia n 0. 38 0. 05 7 1 0 2 0 0. 66 0. 03 9 1 0 0 0 W at er -0 .1 4 0. 04 1 1 1 2 5 -0 .1 0 0. 02 0 0 1 2 7 Lo w la nd O pe n 0. 04 0. 04 2 1 1 2 4 0. 39 0. 05 5 0 1 1 3 co ns ta nt -8 .4 9 1. 06 2 1 1 6 0 -5 .3 0 0. 42 1 0 0 9 0 ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 127 bear risk during the Calving season; wolf risk was often insignificant or variable be- tween sexes. Conifer was strongly avoided by both sexes in all seasons. Males also avoided Alpine except in Summer, whereas fe- males selected for Alpine during Calving, but avoided it from Summer through Early Winter. Both sexes selected Upland Shrub areas in almost all seasons. The Lowland Shrub class was selected by both sexes in Early Winter and by males in the 2 sea- sons afterward (Late Winter and Calving). Mixed Wood was almost always avoided by males, but selected by female moose during Calving and Summer. Both sexes usually selected for Riparian areas (al- though females did not during Calving). Female moose selected against Water, and the influence of Water varied across sea- sons for males, with positive selection for frozen waterways in Early and Late Winter. For models developed for females with and without a calf, there were only 7 valid models after k-fold cross-validation (out of 24; 6 models for each animal) to explain habitat selection during Calving for 4 fe- males with known calf status; 6 of these models had AIC c weights < 0.95. Three models would have resulted from model averaging, but given the low sample sizes (2 females with calves, 2 without), we did not pursue further analysis. The only con- sistency observed in those models was se- lection for Lowland Shrub by all 4 females. Additionally, females with calves selected for higher elevations while avoiding Mixed Wood. Females without a calf selected Mixed Wood and avoided Riparian. Variation in seasonal habitat selection among individual moose For the 24 individual moose in our study, 82 final models described sea- sonal habitat selection. In addition to all topographic and predation coefficients, there were at least 3 land-cover classes (i.e., Conifer, Lowland Shrub, Riparian) in all final models. As mentioned above, in the averaging process for sex-specific pooled models, Late Winter had the fewest explanatory models and the fewest indi- viduals; Calving had the most. In general, there were more individual models dur- ing Summer and Rut for males than for females. There were some strong seasonal trends among individual moose within a sex (Tables 3, 4), and female patterns tended to be more variable than males. Responses to east-facing slopes, wolf risk, and the Water and Lowland Open classes were highly variable among individuals. During Late Winter, most individuals selected for Riparian areas and avoided Conifer stands. Most males avoided north-facing slopes, but responses to aspect were highly dis- parate among females. At this time, when snow was presumed to be deepest, most females selected for Lowland Shrub areas, whereas males exhibited more variability. During the Calving season, most individ- uals continued to avoid Conifer and were often in high bear-risk areas. Most males continued to avoid north-facing slopes and selected Lowland and Upland Shrub areas, whereas females demonstrated more variation in habitat use during Calving. In Summer, male moose selected more con- sistently for mid-elevations than during Late Winter and Calving, and avoided Conifer cover. The selection for Conifer and Lowland Shrub classes was more vari- able among individual females in Summer than in other seasons. Not surprisingly, individuals of both sexes displayed most similarities during the breeding season and the post-rut period. During Rut and Early Winter, most individuals selected Upland Shrub and Riparian areas while strongly RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 128 avoiding Conifer. The majority of males avoided Alpine and Mixed Wood during Rut, and most moose (both sexes) selected for Lowland Shrub in Early Winter. DISCUSSION Moose are a keystone species of north- ern boreal forests, playing important roles in predator-prey dynamics and forest succes- sion (Molvar et al. 1993, Danell et al. 1998). The large geographic range of moose is a reflection of their ability to utilize a wide va- riety of habitats. Elevation, aspect, predation risk, and land cover all had significant ef- fects on seasonal habitat selection by moose in the South Canol area. At the resolution of this study, seasonal changes appeared to affect resource selection by moose in the South Canol area more than differences be- tween sexes or reproductive status (females with and without calves). Elevational differences in temperature and soil moisture influence habitat selection through associated changes in vegetation, snow depth, and thermal conditions. In an effort to balance energetic demands, moose respond to the changing quantity and quality of forage species and accessibility to forage and cover that are found along elevation gra- dients (Stumph and Wright 2007). Resource selection for male moose in our study was influenced by elevation in all seasons, and most consistently among individuals in Summer, Rut, and Early Winter. Access to the higher elevations in these seasons (as reflected in use patterns, McCulley et al. 2017b) may play an important role in the survival strategies of male moose in south- central Yukon. Ballard et al. (1991) reported that upland sites had higher quantities of browse, but lower elevation sites had greater availability as winter progressed. In our study area, upland areas may provide simi- lar opportunities for moose to maximize for- age intake in the seasons surrounding Rut, and to build adequate fat reserves before the snowpack forces them to descend in Late Winter. Like males, females also selected for mid-elevations in most seasons, except during Calving and Summer when selection patterns varied among individuals. Oehlers et al. (2011) reported moose calving at lower elevations in southeast Alaska, which contrasted with those in interior Alaska that chose high-elevation birth sites (Bowyer et al. 1999). Female moose in southeast Brit- ish Columbia were categorized as “climb- ers” or “non- climbers” during the Calving season (Poole et al. 2007). These examples, as well as the variation in individual models in our study, highlight the unpredictable na- ture of birth-site selection by female moose during Calving. Similar to elevation, aspect can strongly influence ambient temperature and soil moisture. South-facing slopes receive the most solar radiation and are the first to green up at northern latitudes; north-facing slopes tend to be cooler and moister throughout the year. East and west aspects increase solar insolation at different times of the year. In the South Canol area, we observed that males (group and individually) selected for west aspects during Calving and Early Winter, and females selected northwest as- pects in Early Winter, but with more indi- vidual variation. This selection may reflect a need to minimize heat stress during times of the year when moose have highly insu- lative winter pelage and temperatures may occasionally be relatively warm. Males and females selected southeast aspects during Late Winter, when males in particular must minimize energy losses in order to survive until spring. Similar to our findings in Late Winter, moose in southeast British Colum- bia preferred gentler slopes with high solar insolation during late winter (Poole and Stu- art-Smith 2006), and moose in Montana se- lected south- and west-facing aspects in late ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 129 winter (Langley 1993). Selection of south- facing slopes may allow moose to key in on spring green-up (e.g., Merkle et al. 2016). Bowyer et al. (1999) reported that female moose preferred to give birth on southeast exposures where soils were drier and forage was of higher quality. We defined predation risk by combining habitat information from our study area with seasonal RSFs from GPS-collared wolves and grizzly bears in the Besa-Prophet area of northern British Columbia. These mod- els were assumed to be compatible with our landscape based on similar predator species, climate, and mountainous topography. The predation risk surfaces, however, should be verified with local predator data as they be- come available. The ungulate prey base in the South Canol area is not as diverse as that in the Besa-Prophet area of northern British Columbia, and hence, predators could po- tentially focus more on moose in the South Canol area. Black bears, for which there are no density estimates in Yukon (R. Maraj, Yukon Department of Environment, pers. comm.), were assumed to have a relatively low (although not quantified) impact on adult animals, and this premise also should be verified. Reducing exposure to mortality risk has obvious benefits to individual moose survival (and their calves). Associated with predators, there also can be energetic costs such as reduced foraging efficiency caused by increased vigilance or movement, and by choosing safer habitats that may have less forage (Molvar and Bowyer 1994, White and Berger 2001, Montgomery et al. 2013). Additionally, Kunkel and Pletcher (2000) noted that wolf kill rates increased with in- creasing distance to cover, decreasing road density, increasing trail and stream density, and increasing wolf density. Exposure to predation risk varied season- ally, but not between males and females, yet individual responses were varied. Contrary to our hypothesis, females did not avoid pre- dation risk during the Calving season. Both sexes showed positive selection for areas with higher bear risk during Calving, pre- sumably by taking advantage of areas with earlier green-up that were also frequented by bears (Bastille-Rousseau et al. 2011). During Late Winter, females with calves used areas with significantly less wolf risk than females without calves. Because of their shorter legs, moose calves are more vulnerable in deep snow than adults (Peterson and Allen 1974, Peterson 1977). Females with calves are known to reduce wolf risk by staying closer to cover, which helps make calves less visible and where snow depths may be lower (White and Berger 2001). Male moose in the South Canol area avoided wolf risk during Rut, but their exposure to wolf risk (as indexed by average relative risk; Fig. 2) was slightly higher during Rut than dur- ing the Calving season or Summer. During the moose breeding season, wolf pups (and the associated adults) are more mobile than earlier in the growing season, potentially in- creasing risk to moose (Mech 1970, Mills et al. 2008). During Rut, exposure to bear risk for both sexes was significantly higher than during Summer. This is likely because bears move up in elevation during late sum- mer and fall to target the rich berry crop usu- ally available in subalpine areas. By Early Winter, exposure to bear risk was negligible as bears hibernated. Average exposure to wolf risk (Fig. 2) dropped significantly; the positive selection coefficients in the pooled models for both sexes may indicate simply those areas also frequented by wolves rather than high-risk areas, per se. In Yukon, most harvest targets male moose and most hunting pressure occurs in September (Yukon Department of Envi- ronment 2008).We expected males to select areas with lower harvest vulnerability dur- ing Rut and analysis confirmed that males RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 130 as a group minimized harvest vulnerability. The individual responses, however, were variable and only 50% of males avoided this risk. The selection for less-accessible higher elevations during Rut may have served to reduce exposure to harvest vulnerability as well as predation risk. Because of lim- ited road access, most moose harvested in our study area were harvested at lower el- evations, often on or within 500 m of large waterways. Similarly in Quebec, density of hunting camps, length of large rivers, and surface area of lakes had the greatest effects on harvest vulnerability (Courtois and Beau- mont 1999). Male moose also may make fine-scale adjustments to avoid detection, but such selection or avoidance would be difficult to determine at the resolution (25 m) of our study (Courtois and Beaumont 1999, Laurian et al. 2000). Vulnerability and response to harvest risk can vary depending on age and experience of the moose. In our study, only mature males were collared and individuals likely had several years’ experi- ence avoiding detection by predators or hu- mans. Habitat selection by younger males may differ from our observations. We expected moose to maximize forage intake during the growing season (Belovsky 1978) and minimize energy losses in win- ter, recognizing that male and female moose would likely use different strategies to meet those ends. We assumed a forage-based strategy would be indicated by strong selec- tion for shrub-dominated land-cover classes, whereas selection of Conifer would indicate a greater need for cover. All males avoided Conifer throughout the year, as did females as a group, although individuals had variable responses. The interpretation of strategies used for forage and cover may be confounded by the scale of selection and because Conifer represented a large proportion of the study area (McCulley et al. 2017b). Although Con- ifer was avoided, it was used often because of its abundance on the landscape. Poole and Stuart-Smith (2006) reported that moose selected higher crown closure at the land- scape scale, whereas others have observed selection for conifer stands at finer scales re- lated to thermal cover (McNicol and Gilbert 1978), forage diversity (Peek 1997), and predation risk (Bowyer et al. 1999, Dussault et al. 2004, Bjorneraas et al. 2012). Both sexes selected for shrub-dominated land- cover (i.e., Upland Shrub, Riparian, or Low- land Shrub) throughout most of the year, with highest selection for Upland Shrub during the Calving season through Rut, and for Riparian areas in Early and Late Winter; most individuals also selected for Lowland Shrub from Early Winter through Calving. The elevational gradient encompassed by these communities likely enabled all moose to seasonally target areas with highest food quality and forage biomass. We expected selection patterns of both sexes to be most similar during the breeding season and most different during parturition and lactation (Miquelle et al. 1992, Oehlers et al. 2011). Indeed, during Rut males and females selected strongly for Upland Shrub and Riparian, while avoiding Conifer, Al- pine, and Water. Conversely, during Calv- ing they generally differed in response to Alpine, Riparian, Water, and Lowland Open classes. Interestingly, females did not select for Riparian areas during the Calving season which was surprising given that proximity to water is an important characteristic of birth sites in other areas (Oehlers et al. 2011). We defined riparian zones as areas within 100 m (4 pixels) of large rivers and lakes and only 25 m (1 pixel) from small streams, so per- haps females were able to locate birth sites outside of the riparian zone as we defined it, but still with adequate access to water. Land-cover classification based on sat- ellite imagery is efficient and was particu- larly valuable in our study which spanned ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 131 a remote and isolated area covering 14% of Yukon’s landmass. An existing EOSD land-cover classification for Yukon had a species-appropriate resolution (25 m) and was based on digital imagery collected within 8 years of the study. By merging some of the 24 classes into 8, we reduced the chances for misclassification. The vari- ation encompassed by each land-cover class, however, may have contributed to the variation observed among individuals and could be important in fine-scale selec- tion by moose. Variation in selection patterns may have been confounded by the GPS locations for each individual. Although we could not ad- just for potential bias in fix rates, we reduced bias toward particular individuals with more location fixes by developing selection mod- els per individual, and requiring a minimum of 100 locations per individual per season. We also had the opportunity to monitor a stationary fully-functioning GPS collar for 11 months (after an animal died) to assess GPS field accuracy; mean distance between fixes was 7.9 ± 0.25 m. Given the resolution of the land-cover data, the land-cover class associated with each location point was fairly precise. Some low fix-acquisition rates may have introduced location bias, particularly if missed fixes were more likely to occur in some land-cover classes. Although we were unable to correct for land-cover or terrain characteristics, habitat transition probabilities, or fixes lost across geograph- ical space (Frair et al. 2004, 2010; Nielson et al. 2009), it is most likely that any biases are associated with under-representing the use of closed conifer forests and perhaps north-facing aspects. There was, however, no significant relationship between fix -rate success and the proportion of Coni- fer land cover within annual ranges across collared individuals or among individuals with collars from the same manufacturer. Lowest (18% of annual range) and highest (66%) Conifer cover were both observed with fix rates > 88%. Nonetheless, our habitat selection coefficients should be in- terpreted with some caution in the possible case of alternative biases. In summary, moose in south-central Yukon altered their selection patterns in re- sponse to seasonal changes typical of north- ern boreal forests. Differential seasonal selection was observed within individuals and groups. Differences between the sexes in exposure to risk were not supported by our data, but may occur at finer spatial and temporal scales than in our study. The dif- ferences in the selection coefficients of in- dividual and pooled models underscored the variety of moose-habitat relations in the South Canol area. Pooled models for large non-herding species such as moose may en- compass highly variable behaviour among individuals. The “average” moose may not exist (Gillingham and Parker 2008b), yet in- dividual models effectively demonstrate the variability within the population of interest (McCulley 2015: Appendix F, G). Relative consistency of individual responses by both sexes supports trends identified by pooled selection coefficients, and detects trends be- tween sexes. Most individual variation occurred during the growing season (Calving and Summer) and the least amount during Late Winter, suggesting that climatic factors limited the options available to moose dur- ing the most resource-constrained season. This observation may have important im- plications to moose populations in Yukon, where climate change is occurring at a faster rate than in more southerly moose ranges (ACIA 2005). Moose experience in- creased heat stress, higher parasitic loads, more malnutrition, and greater wolf pre- dation in warmer climates (Rempel 2011), RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 132 with declines reported in certain popula- tions at the southern extent of moose range (e.g., Post et al.1999, Murray et al. 2006). Because moose respond to climatic factors at large scales (Hallett et al. 2004), contin- ual habitat use and population assessments are encouraged to evaluate the broad ex- panse of northern moose populations for which less is known about potential cli- matic impacts. ACKNOWLEDGEMENTS We acknowledge the collaboration among Yukon Government (YG), the Teslin Tlingit Council (TTC), and the University of Northern British Columbia (UNBC). We thank R. Ward (YG) for initiating the proj- ect and all those who participated in moose captures and monitoring of calf survival. We appreciate the efforts of the Yukon Fish and Game Association, F. Johnstone, and R. Ward to collect numerous moose harvest data, and the trust and willingness by South Canol hunters and others to supply sensi- tive and detailed harvest information. We are especially indebted to M. Waterreus and R. Legner with YG for providing geospatial and collar data, to R. Wheate, S. Emmons, and P. Bai at UNBC for assistance in the GIS lab, and to D. Heard for helpful discus- sions on moose management perspectives. We are also grateful to members of the TTC Fish and Wildlife Program, the Tes- lin Renewable Resource Council, and the Laberge Renewable Resources Council for their valuable input. REFERENCES ACIA. 2005. Arctic Climate Assessment. Cambridge University Press, Cambridge, United Kingdom. Andersen, r., and B.-e. sAether. 1992. Interactions between a generalist her- bivore, the moose, and its winter food resources: a study of behavioural varia- tion. Alces Supplement 1: 101–104. Anderson, d. P., M. G. turner, J. d. Forester, J. Zhu, M. s. BoyCe, h. Beyer, and L. stoweLL. 2005. Scale-dependent sum- mer resource selection by reintroduced elk in Wisconsin, USA. Journal of Wildlife Management 69: 298–310. BAer, A. 2011. Wolf survey in the Nisutlin River basin, 2011. Yukon Fish and Wild- life Branch Report TR-11-17, White- horse, Yukon, Canada. BALLArd, w. B., J. s. whItMAn, and d. J. reed. 1991. Population dynamics of moose in south-central Alaska. Wildlife Monographs 114: 5–49. BArBoZA, P. s., and r. t. Bowyer. 2000. Sexual segregation in dimorphic deer: a new gastrocentric hypothesis. Journal of Mammalogy 81: 473–489. BAskIn, L., J. P. BALL, and k. dAneLL. 2004. Moose escape behaviour in areas of high hunting pressure. Alces 40: 123–131. BAstILLe-rousseAu, G., d. FortIn, C. dussAuLt, r. CourtoIs, and J. P. oueLLet. 2011. Foraging strategies by omnivores: are black bears actively searching for un- gulate neonates or are they simply op- portunistic predators? Ecography 34: 588–596. BeLovsky, G. e. 1978. Diet optimization in a generalist herbivore: the moose. Theo- retical Population Biology 14: 105–134. BJørnerAAs, k., I. herFIndAL, e. J. soLBerG, B.-e. sAether, B. vAn Moorter, and C. M. roLAndsen. 2012. Habitat qual- ity influences population distribution, individual space use and functional re- sponses in habitat selection by a large herbivore. Oecologia 168: 231–243. BoertJe, r. d., d. G. keLLyhouse, and r. d. hAyes. 1995. Methods for reducing nat- ural predation on moose in Alaska and Yukon: an evaluation. Pages 505–513 in L. N. Carbyn, S. J. Fritts, and D. R. Seip, editors. Ecology and Conservation of Wolves in a Changing World. Canadian Circumpolar Institute, Occasional Publi- cation No. 35. University of Edmonton, Edmonton, Alberta, Canada. ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 133 Bowyer, r. t., and J. G. kIe. 2006. Effects of scale on interpreting life-history characteristics of ungulates and carni- vores. Diversity and Distributions 12: 244–257. ———, vAn BALLenBerGhe, v., kIe, J. G., and J. A. MAIer. 1999. Birth-site selec- tion by Alaskan moose: maternal strate- gies for coping with a risky environment. Journal of Mammalogy 80: 1070–1083. BoyCe, M. s., and L. L. MCdonALd. 1999. Relating populations to habitats using resource selection functions. Trends in Ecology and Evolution 14: 268–272. ———, P. r. vernIer, s. e. nIeLsen, and F. k. A. sChMIeGeLow. 2002. Evaluating resource selection functions. Trends in Ecology and Evolution 14: 268–272. BurnhAM, k. P., and d. r. Anderson. 2002. Model Selection and Multimodel Infer- ence: Practical Information-Theoretic Approach. Springer-Verlag, New York, New York, USA. CourtoIs, r., and A. BeAuMont. 1999. The in- fluence of accessibility on moose hunting in northwestern Quebec. Alces 35: 41–50. dAneLL, k., t. wILLeBrAnd, and L. BAskIn. 1998. Mammalian herbivores in the bo- real forests: their numerical fluctuations and use by man. Conservation Ecology 2: 1–20. dussAuLt, C., J.-P. oueLLet, r. CourtoIs, J. huot, L. Breton, and J. LAroCheLLe. 2004. Behavioural responses of moose to thermal conditions in the boreal for- est. Ecoscience 11: 321–328. esrI 2006. ArcMap GIS, Release 9. Envi- ronmental Systems Research Institute, Redlands, California, USA. FLorkIewICZ, r., and P. d. henry. 1994. Moose habitat characteristics relative to logging in the Liard River Basin, south- eastern Yukon. Pages 211–216 in Pro- ceedings of International Union Game Biologists XXI Congress, 15–20 August 1993. Halifax, Nova Scotia, Canada. ———, J. MCLeLLAn, s.westover, and r. wArd. 2008. Summary of the South Canol West early-winter 2007 moose survey. Yukon Department of Environ- ment (unpublished report). Whitehorse, Yukon, Canada. FrAIr, J. L., s. e. nIeLsen, e. h. MerrILL, s. r. LeLe, M. s. BoyCe, r. h. Munro, and h. Beyer. 2004. Removing GPS collar bias in habitat selection studies. Journal of Applied Ecology 41: 201–212. ———, J. FIeBerG, M. heBBLewhIte, F. CAGnACCI, n. J. deCesAre, and L. PedrottI. 2010. Resolving issues of imprecise and habitat-biased locations in ecological analyses using GPS te- lemetry data. Philosophical Transac- tions of the Royal Society B: Biological Sciences 365: 2187–2200. GAsAwAy, w. C., r. d. BoertJe, d. v. GrAnGAArd, d. G. keLLeyhouse, r. o. stePhenson, and d. G. LArsen. 1992. The role of predation in limiting moose at low densities in Alaska and Yukon and implications for conserva- tion. Wildlife Monographs 120: 3–59. GILLInGhAM, M. P., and k. L. PArker. 2008a. Differential habitat selection by moose and elk in the Besa-Prophet area of northern British Columbia. Alces 44: 41–63. ———, and ———. 2008b. The impor- tance of individual variation in de- fining habitat selection by moose in northern British Columbia. Alces 44: 7–20. hALLett, t. B., t. CouLson, J. G. PILkInGton, t. h.CLutton-BroCk, J. M. PeMBerton, and B. t. GrenFeLL. 2004. Why large- scale climate indices seem to predict ecological processes better than local weather. Nature 430: 71–75. hAyes, r. d., A. M. BAer, u. wotsChIkowsky, and A. s. hArestAd. 2000. Kill rate by wolves on moose in the Yukon. Cana- dian Journal of Zoology 78: 49–59. ———, and A. s. hArestAd. 2000. Wolf functional response and regulation of moose in the Yukon. Canadian Journal of Zoology 78: 60–66. RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 134 keIth, r. J. 1995. Home range and habitat use by female moose in the southwest Yukon. M.S. The- sis. University of Regina, Regina, Saskatchewan, Canada. keLsALL, J. P., and e. s. teLFer. 1974. Bio- geography of moose with particular reference to western North America. Naturaliste Canadien 101: 117–130. ———, ———, and t. d. wrIGht. 1977. The effects of fire on the ecology of the boreal forest, with particular reference to the Canadian north: a review and se- lected bibliography. Occasional Paper Number 32, Canadian Wildlife Service, Ottawa, Ontario. kIe, J. G., r. t. Bowyer, M. C. nIChoLson, B. B. BoroskI, and e. r. LoFt. 2002. Landscape heterogeneity at differing scales: effects on spatial distribution of mule deer. Ecology 83: 530–544. kunkeL, k. e., and d. h. PLetCher. 2000. Habitat factors affecting vulnerability of moose to predation by wolves in south- eastern British Columbia. Canadian Journal of Zoology 78: 150–157. LAnGLey, M. A. 1993. Habitat selection, mortality and population monitoring of Shiras moose in the North Fork of the Flathead River valley, Montana. Ph.D. Thesis, University of Montana, Mis- soula, Montana, USA. LArsen, d. G., d. A. GAuthIer, and r. L. MArkeL. 1989. Causes and rate of moose mortality in the southwest Yukon. Journal of Wildlife Management 53: 548–557. LAurIAn, C., J.-P. oueLLet, r. CourtoIs, L. Breon, and s. st-onGe. 2000. Effects of intensive harvesting on moose repro- duction. Journal of Applied Ecology 37: 515–531. MAIn, M. B., and B. e. CoBLentZ. 1990. Sexual segregation among ungulates: a critique. Wildlife Society Bulletin 18: 204–210. ———, F. w. weCkerLy, and v. C. BLeICh. 1996. Sexual segregation in ungulates: new directions for research. Journal of Mammalogy 77: 449–461. MAnLy, B. F. J., L. L. MCdonALd, d. L. thoMAs, t. L. MCdonALd, and w. P. erICkson. 2002. Resource Selection by Animals: Statistical Design and Anal- ysis for Field Studies. Second edition. Kluwer Academic Publishers, Boston, Massachusetts, USA. MAuer, F. J. 1998. Moose migration: northeastern Alaska to northwestern Yukon Territory, Canada. Alces 34: 75–81. MCCuLLey, A. M. 2015. Habitat use and selection by male and female moose (Alces alces) in a boreal landscape. M.S. Thesis. University of Northern British Columbia, Prince George, British Co- lumbia, Canada. ———, k. L. PArker, and M. P. GILLInGhAM. 2017b. Yukon moose: II. Range sizes, movement rates, and use of elevation and land cover by males and females. Alces 53: 137–157. MCnICoL, J. G., and F. F. GILBert. 1978. Late winter bedding practices of moose in mixed upland cutovers. Canadian Field- Naturalist 92: 189–192. MeCh, L. d. 1970. The Wolf: The Ecology and Behavior of an Endangered Species. The Natural History Press, Garden City, New York, USA. MenArd, s. 2002. Applied Logistic Regres- sion Analysis. Sage Publications, Inc., Thousand Oaks, California, USA. MerkLe, J. A., k. L. MonteIth, e. o. AIkens, M. M. hAyes, k. r. hersey, A. d. MId- dLeton, B. A. oAtes, h. sAwyer, B. M. sCurLoCk, and M. J. kAuFFMAn. 2016. Large herbivores surf waves of green-up during spring. Proceedings of the Royal Society B 283: 20160456. MILAkovIC, B. 2008. Defining the preda- tor landscape of northeastern British Columbia. Ph. D. Dissertation, Uni- versity of Northern British Colum- bia, Prince George, British Columbia, Canada. ALCES VOL. 53, 2017 MCCULLEY ET AL. – RESOURCE SELECTION BY MOOSE IN YUKON 135 ———, k. L. PArker, d. d. GustIne, r. J. LAy, A. B. wALker, and M. P. GILLInGhAM. 2011. Habitat selection by a focal preda- tor (Canis lupus) in a multiprey ecosys- tem of the northern Rockies. Journal of Mammalogy 92: 568–582. ———, ———, ———, ———, ———, and ———. 2012. Seasonal habitat use and selection by grizzly bears in north- ern British Columbia. Journal of Wild- life Management 76: 170–180. MILLs, k. J., B. r. PAtterson, and d. L. MurrAy. 2008. Direct estimation of early survival and movements in eastern wolf pups. Journal of Wildlife Manage- ment 72: 949–954. MIqueLLe, d. G., and P. A. JordAn. 1979. The importance of diversity in the diet of moose. Proceedings of the North American Moose Conference Workshop 15: 54–79. ———, J. M. Peek, and v. vAn BALLenBerGhe. 1992. Sexual segregation in Alaskan moose. Wildlife Monographs 122: 1–57. MoLvAr, e. M., and r. t. Bowyer. 1994. Costs and benefits of group living in a recently social ungulate: the Alaskan moose. Jour- nal of Mammalogy 75: 621–630. ———, ———, and v. vAn BALLenBerGhe. 1993. Moose herbivory, browse quality, and nutrient cycling in an Alaskan tree- line community. Oecologia 94: 472–479. MontGoMery, r. A., J. A. vuCetICh, r. o. Peterson, G. J. roLoFF, and k. F. MIL- LenBAh. 2013. The influence of winter severity, predation and senescence on moose habitat use. Journal of Animal Ecology 82: 301–309. MurrAy, d. L., e. w. Cox, w. B. BALLArd, h. A. whItLAw, M. s. LenArZ, t. w. Custer, t. BArnett, and t. k. FuLLer. 2006. Pathogens, nutritional deficiency, and climate influences on a declining moose population. Wildlife Monographs 166: 1–29. MurtAuGh, P. A. 2007. Simplicity and com- plexity in ecological data analysis. Ecol- ogy 88: 56–62. Mysterud, A., and e. ostBye. 1999. Cover as a habitat element for temperate un- gulates: effects on habitat selection and demography. Wildlife Society Bulletin 27: 385–394. nIeLson, r. M., B. F. J. MAnLy, L. L. MCdonALd, h. sAwyer, and t. L. MCdonALd. 2009. Estimating habitat selection when GPS fix success is less than 100%. Ecology 90: 2956–2962. 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. PCI GeoMAtICs. 2009. Geomatica, Release 10.3, Richmond Hill, Ontario, Canada. Peek, J. M. 1997. Habitat relationships. Pages 351–375 in A. W. Franzmann and C. C. Schwartz, editors. Ecology and Management of the North Ameri- can Moose. Smithsonian Institute Press, Washington, D. C., USA. Peterson, r. o. 1977. Wolf ecology and prey relationships on Isle Roy- ale. National Park Service Scientific Monograph Series Number 11. US Government Printing Office. Washing- ton, D. C., USA. ———, and d. L. ALLen. 1974. Snow con- ditions as a parameter in moose-wolf relationships. Naturaliste Canadien 101: 481–492. PooLe, k. G., and k. stuArt-sMIth. 2006. Winter habitat selection by female moose in western interior montane for- ests. Canadian Journal of Zoology 84: 1823–1832. ———, r. serrouyA, and k. stuArt-sMIth. 2007. Moose calving strategies in in- terior montane ecosystems. Journal of Mammalogy 88: 139–150. Post, e., r. o. Peterson, n. C. stenseth, and B. e. MCLAren. 1999. Ecosystem consequences of wolf behavioural re- sponse to climate. Nature 401: 905–907. reMPeL, r. s. 2011. Effects of climate change on moose populations: exploring RESOURCE SELECTION BY MOOSE IN YUKON – MCCULLEY ET AL. ALCES VOL. 53, 2017 136 the response horizon through biometric and systems models. Ecological Model- ling 222: 3355–3365. reneCker, L. A., and r. J. hudson. 1986. Seasonal energy expenditures and thermoregulatory responses of moose. Canadian Journal of Zoology 64: 322–327. ———, and ———. 1992. Thermoregula- tory and behavioural response of moose: is large body size an adaptation or con- straint? Alces Supplement 1: 52–64. ———, and sChwArtZ, C. C. 2007. Food habits and feeding behaviour. Pages 403–439 in A. W. Franzmann and C. C. Schwartz, editors. Ecology and Man- agement of the North American Moose. Smithsonian Institute Press, Washing- ton, D. C., USA. stAtACorP. 2011. Stata Statistical Software, Release 12. StataCorp LP, College Sta- tion, Texas, USA. stuMPh, B. P., and r. G. wrIGht. 2007. Effects of willow quality on moose distribution in a montane environment. Alces 43: 129–142. teLFer, e. s. 1970. Winter habitat selection by moose and white-tailed deer. Journal of Wildlife Management 34: 553–559. thoMAs, d. L., and e. J. tAyLor. 1990. Study designs and tests for comparing resource use and availability. Journal of Wildlife Management 54: 322–330. ———, and ———. 2006. Study designs and tests for comparing resource use and availability II. Journal of Wildlife Management 70: 324–336. tIMMerMAnn, h. r., and A. r. rodGers. 2005. Moose: completing and comple- mentary values. Alces 41: 85–120. vAn BALLenBerGhe, v., and J. M. Peek. 1971. Radiotelemetry studies of moose in northeastern Minnesota. Journal of Wildlife Management 35: 63–71. whIte, k. s., and J. BerGer. 2001. Anti- predator strategies of Alaskan moose: are maternal trade-offs influenced by offspring activity? Canadian Journal of Zoology 79: 2055–2062. wuLder, M. A., J. A. deChkA, M. A. GILLIs, J. e. Luther, r. J. hALL, A. BeAudoIn, and s. e. FrAnkLIn. 2003. Operational mapping of the land cover of the for- ested area of Canada with Landsat data: EOSD land cover program. Forestry Chronicle 79: 1075–1083. yukon dePArtMent oF enerGy, MInes And resourCes. 2004. Yukon fire history 1946–2004. (ac- cessed November 2014). yukon dePArtMent oF envIronMent. 2008. Yukon hunting regulations sum- mary 2008–2009. Whitehorse, Yukon, Canada. yukon eCoreGIons workInG GrouP. 2004. Pelly Mountains. Pages 219–226 in C. A. S. Smith, J. C. Meikle, and C. F. Roots, editors. Ecoregions of the Yukon Terri- tory: Biophysical Properties of Yukon Landscapes. PARC Technical Bulletin No. 04-01, Agriculture and Agri-Food Canada, Summerland, British Colum- bia, Canada. www.emr.gov.yk.ca/oilandgas/pdf/2004firehistory2b.pdf www.emr.gov.yk.ca/oilandgas/pdf/2004firehistory2b.pdf