71 MODELING MOOSE HABITAT USE BY AGE, SEX, AND SEASON IN VERMONT, USA USING HIGH-RESOLUTION LIDAR AND NATIONAL LAND COVER DATA Joshua Blouin1, Jacob Debow1, Elias Rosenblatt1, Cedric Alexander2, Katherina Gieder3, Nicholas Fortin3, James Murdoch4, and Therese Donovan5 1Vermont Cooperative Fish and Wildlife Research Unit, Rubenstein School of Environment and Natural Resources, University of Vermont, 302 Aiken Center, 81 Carrigan Drive, Burlington, Vermont 05405, USA; 2Vermont Fish and Wildlife Department, 374 Emerson Falls Rd, St Johnsbury, Vermont 05819, USA; 3Vermont Fish and Wildlife Department, 271 North Main Street, Suite 215, Rutland, Vermont 05701, USA; 4Wildlife and Fisheries Biology Program, Rubenstein School of Environment and Natural Resources, University of Vermont, 302 Aiken Center, 81 Carrigan Drive, Burlington, Vermont 05405, USA; 5U.S. Geological Survey, Vermont Cooperative Fish and Wildlife Research Unit, Rubenstein School of Environment and Natural Resources, University of Vermont, 302 Aiken Center, 81 Carrigan Drive, Burlington, Vermont 05405, USA. ABSTRACT: Moose (Alces alces) populations have experienced unprecedented declines along the southern periphery of their range, including Vermont, USA. Habitat management may be used to improve the status of the population and health of individuals. To date, however, Vermont wildlife managers have been challenged to effectively use this important tool due to the lack of fine-scale information on moose space use and habitat characteristics. To assess habitat use, we combined more than 40,000 moose locations collected from radio-collared individuals (n = 74), recent land cover data, and high resolution, 3-dimensional lidar (light detection and ranging) data to develop Resource Utilization Functions (RUF) by age (mature and young adult), season (dormant and growth), and sex. Each RUF linked home range use to average habitat conditions within 400 m or 1 km of each 30 m2 pixel within the home range. Across analyses, the top RUF models included both composition (as measured through the National Land Cover Database) and structure (as measured through lidar) vari- ables, and significantly outperformed models that excluded lidar variables. These findings support the notion that lidar is an effective tool for improving the ability of models to estimate patterns of habitat use, especially for larger bodied mammals. Generally speaking, female moose actively used areas with proportionally more regenerating forest (i.e., forage < 3.0 m) and more mature forest (i.e., canopy structure > 6.0 m), while males actively used more high elevation, mixed forest types. Further, moose exhibited important seasonal differences in habitat use that likely reflect temporal changes in energetic and nutritional requirements and behavior across the year. Moose used areas with proportionally more regenerating forest (i.e., forage < 3.0 m) during the growth period and female moose had strong posi- tive associations with lidar-derived canopy structure during the growth (but not the dormant) period. Ultimately, the resultant maps of habitat use provide a means of informing management activities (e.g., the restoration or alteration of habitats to benefit moose) and policies around land use that may contribute to population recovery. ALCES VOL. 57: 71–98 (2021) Key words: Alces alces, habitat suitability, land cover, lidar, moose, resource utilization function INTRODUCTION Moose (Alces alces) have experienced declines in many regions along the southern periphery of their distribution in North America (Murray et al. 2006, Jones et al. 2019). In Vermont in the northeastern United BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 72 States, moose have declined to the point of requiring new management approaches to improve the status of the population and health of individuals. Habitat management may be an important tool to achieve this end, yet an assessment of fine-scale habitat use by moose in this region is lacking. Habitat selection is the process in which an animal chooses or selects a particular habitat, given a range of options (Johnson 1980, Beyer et al. 2010). The selection of resources by individuals is thought to occur in a hierarchical manner, from a more coarse- scale examination of where a species selects their geographic range (first-order selec- tion), to the home range of an individual within their geographical range (second- order), and finally to more fine-scale exam- ination of the habitat components within home ranges (third-order, Johnson 1980). Regardless of the available habitat options, habitat selection decisions result in patterns of habitat use, which can be quantified at these same scales (Johnson 1980, Beyer et al. 2010) and is the focus of this study. Common approaches to assess first-and sec- ond order habitat use involve detection or non-detection of unmarked individuals across space and time, and the probability a species will exist at a given location (MacKenzie et al. 2017). Although such modeling approaches are valuable in describing use of locations across a defined geographic extent, they often lack detail regarding resource use by individuals within their home range (i.e., third order) which can provide a more nuanced and finer-scale depiction of the rel- ative importance of specific habitat resources. Resource utilization functions (RUFs) allow estimation of wildlife habitat use within individual home ranges (Marzluff et al. 2004), and RUFs relate home range utilization distribution (UD) to its underly- ing resources, where the UD is a probability distribution that describes an individual’s pattern of space use. RUF models provide the importance of habitat variables within the home range, given the home range has been selected. Population-level RUFs can be derived by aggregating individual RUFs, thus providing an understanding of which habitat variables are most heavily used in aggregate. For example, Marzluff et al. (2004) used this approach to document how Steller’s Jays (Cyanocitta stelleri) respond to different land cover amounts and types, and Amelon et al. (2014) investigated resource use by eastern red bats (Lasiurus borealis) during the maternity season. RUF models and maps of habitat use that can be produced from them provide important information that can guide decision-making in wildlife management. However, many challenges exist in building models that are of utility, especially for spe- cies that occur in a diversity of habitats. Evaluating habitat composition alone (i.e., landcover type) may be insufficient as finer-scale structural features (i.e., the height of vegetation) within habitats may also influ- ence use. For instance, during summer or the vegetative growth period in the northeastern United States (May-Sep), moose consume large quantities of deciduous tree/shrub leaves and stems during their productive state when they recover physically from winter, grow, lactate, and attain pre-winter condition (fat reserves); thus, areas of forest regeneration are highly used and preferred habitat (Schwartz and Renecker 2007). Moreover, moose may alter daily behavior and resource use to avoid thermal stress during summer (Montgomery et al. 2019). Despite its seasonally lower nutritional value, moose continue foraging buds and stems of deciduous regeneration during winter, and commonly consume bal- sam fir (Abies balsamea) (Schwartz et al. 1988). As snow deepens, moose employ ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 73 energy-conserving behaviors by limiting movement and activity and seeking refuge under canopied coniferous forest (Dussault et al. 2004). Such life history complexities suggest that coarse scale landcover maps such as the National Land Cover Database (NLCD) alone may be insufficient in estimating RUFs useful for management purposes, par- ticularly for a species that depends on dif- ferent forest age classes seasonally. Lidar (light detection and ranging) is a remote sensing tool used to generate precise, three-dimensional information about the height of vegetation across the landscape (Lefsky et al. 2002). Lidar data are acquired from a fixed-wing aircraft that projects a beam of light towards earth’s surface, which is reflected and captured by a sensor. The resulting three-dimensional point cloud records not only the x- and y-coordinates indicating the horizontal location of each point, but also a z-coordinate fixing the ver- tical location of the point relative to sea level. The point cloud thus captures both ground and aboveground features, including impervious land cover such as buildings and semi-pervious objects such as foliage and branches associated with vegetation (Lefsky et al. 2002). Lidar, combined with NLCD, allows a description of habitat use within home ranges that features combinations of critical habitat, such as structural forage and cover from solar radiation or deep snow. Indeed, studies from other regions (e.g., Isle Royale National Park, Michigan U.S.A, and western Finland) have shown the benefits of lidar in quantifying specific structural con- ditions used by moose (e.g., Verissimo 2012, Melin et al. 2016). However, lidar data are expensive to produce, and it behooves resource managers in Vermont and the northeastern U.S. to understand if including lidar data in the mapping of habitat use will increase the success of management activities that rely on it (i.e., the value of information; Howard 1966). Declines in moose health, survival, and fecundity presents concerns for Vermont and for regional moose populations as Vermont provides important connectivity between populations in Maine, New Hampshire, New York (U.S.) and southern Quebec, Canada (Kretser et al. 2011). Although direct management through har- vest provides one means of influencing moose density and individual health (Boertje et al. 2019), indirect management through the protection, alteration, or cre- ation of important habitat provides another that may benefit moose. As such, reliable RUF models provide a crucial tool to inform future stewardship for one of the most char- ismatic and culturally important wildlife species of the region. We used data from GPS radio-collared moose and multi-scale land cover data to examine third-order habitat use in Vermont. The objectives of this study were to 1) describe general home range habitat charac- teristics by moose age (mature and young adult), season (dormant and growth), and sex, 2) develop RUF models to determine patterns of utilization within home ranges and their underlying resource variables by age, sex, and season, and 3) apply RUF models to map habitat use and assess the importance of habitats across the study area. METHODS Study Area The study occurred in Essex County in northeastern Vermont, USA, in state wildlife management units E1 and E2 that covered 1,738 km2 (Fig. 1, mean latitude = 44.77°; mean longitude = -71.74°). This area was selected due to the relatively high density of moose and its importance for Vermont’s broader moose population (Pearman- Gillman et al. 2020, VFWD 2020). BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 74 The study area has history of logging and mixed geography, characterized by extensive bogs and softwood swamps, and young, intermediate, and mature forest stands. Elevation ranged between 200 to 1000 m. High elevation (> 800 m) forest was dominated by red spruce (Picea rubens) and balsam fir. Intermediate elevations (300 – 800 m) consisted primarily of maple (Acer spp.), birch (Betula spp.), and beech (Fagus grandifolia), while the lowland swamps and bogs were dominated by bal- sam fir, red spruce, black spruce (P. mari- ana), poplar (Populus spp.), paper birch (B. papyrifera), and alder (Alnus spp.). Vermont experiences 4 distinct seasons including summer (Jun-Aug), fall (Sep-Nov), winter (Dec-Feb), and spring (Mar-May). These seasons were broadly categorized as the dormant season (Oct-Apr in which deciduous trees lacked leaves) and the growth season (May-Sep in which leaves were present). Between 2017 and 2019, the average tempera- ture during the growth season was 15.3 °C, with average precipitation ranging between 100 and 110 cm (NCDC 2019). The average temperature during the dormant season was -3.8 °C, with an average snowfall ranging between 220 and 250 cm (NCDC 2019). RADIO-COLLARING To estimate home ranges, 126 moose were captured and fitted with GPS radio-collars. Capturing occurred via helicopter in January of 2017 (n = 30 calves [<1 yr], 30 adult [≥  1  yr]  females),  2018  (n  =  30  calves,  Fig. 1. The location of the study area in northeastern Vermont, USA (1,738 km2), encompassing Essex County (Wildlife Management Unit E). GPS radio-collars were attached to moose (n = 126) in the area and monitored from 2017 to 2019. The study area was bounded by the Canada-U.S. border to the north, New Hampshire to the east, VT-Route 2 to the south, and VT-Route 114 to the west. Triangles indicate the capture location for all moose throughout the study; red triangles indicate the capture location for female moose (n = 74), while circles indicate where male moose were captured (n = 52). The map on the right shows the study area in relation to other northeastern states (ME = Maine, NH = New Hampshire, and MA = Massachusetts). ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 75 6 adult females), and 2019 (n = 30 calves). One radio-collared adult female moose from a New Hampshire study immigrated into the study area in 2018 and was included in our study. Each captured moose was fit- ted with a Survey Globalstar V7.1 GPS and VHF radio-collar (VECTRONIC Aerospace GmbH, Berlin, Germany). A GPS location of the moose was transmitted from each collar every 13 h. As highly mobile animals, moose can traverse the entirety of their home range within this time span, ensuring that points provided a representative sample of habitat use (Fieberg 2007). All capture, han- dling, and radio-collaring procedures were reviewed and approved by the University of Vermont Institutional Animal Care and Use Committee (IACUC protocol #17-035). Objective 1: General Home Range Characteristics We estimated each individual’s home range separately for the growth season and the dor- mant season across each year (2017–2019) to determine individual habitat use. To avoid bias in measuring habitat use by potentially unhealthy or unrepresentative individuals (e.g., an individual that was limited in its movements due to an infestation of winter ticks [Dermacentor albipictus] or brain worm [Parelaphostrongylus tenuis]), we removed GPS collar locations as follows. First, if an adult died during the study, GPS coordinates for the two-week period leading up to the mortality event were removed to avoid potentially overestimating habitat use due to behavior associated with brain worm (Lankester 2010). Second, we removed all GPS locations from individuals < 1 year old to ensure that locations were associated with adult age classes. Third, GPS coordinates transmitted from beyond the boundaries of our study area were removed. After filtering, home range analyses were based on 40,451 locations of 74 moose (219 total home ranges). We used the kern- elUD function in the R package, adehabi- tatHR (Calenge 2011) to estimate 95% fixed kernel home ranges for each moose during each year. We used the ad hoc method to select the smoothing factor (h) for each home range. This method tends to outperform the reference and least-squares cross-validation derived smoothing factors by providing robust UD estimates even with autocor- related data and limiting the number of dis- joint home range polygons or ‘islands’ that can result from under smoothing (Kie 2013). Within each defined home range, we identified habitat variables important to moose including broad land cover types (deciduous forest, coniferous forest, mixed forest, wetland, developed, and open) from National Landcover Data (USGS 2016a), spatial layers describing anthropogenic influences such as snowmobile trails (VTANR 2019), and terrain characteristics such as elevation models (VCGI 2002) (Table 1). Lidar data were used to characterize for- est age structure within each home range (Table 1). The raw lidar point cloud data (.laz/.las format) for the Connecticut River Basin (which encompassed our study area) was downloaded from the U.S. Geological Survey’s open source National Geospatial Program (USGS 2016b). We used lidar point cloud data obtained in early November 2016 (USGS 2016b) and extracted to our study area at the 10 m2 resolution. The lidar returns were extracted and summarized as the pro- portion of total returns in each 10 m2 pixel that fell within 5 vegetative height classifi- cations (open, shrub, forage, cover, and can- opy) associated with important life requisites of moose (Table 1, and see Schwartz and Renecker 2007). We estimated the average kernel home range area for both male and female moose for each season. Additionally, moose were BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 76 categorized as either “mature adult” or “young adult” (mature = mature at capture, young adult = calf at capture), under the assumption that female adults were not cap- tured as yearlings. No mature male moose were captured during the study, so all males were classified as “young adults”. We used the cellStats function in the R raster package (Hijmans and Van Etten 2012) to calculate the average habitat composition (as measured by NCLD) and structure met- rics (as measured by lidar) of each home range, as well as for the overall study area. For each NLCD and lidar variable, we used a Table 1. Covariates used to develop Resource Utilization Distribution models describing habitat use for radio-collared moose in Vermont, USA. Covariate Name Description Data Source Reference P_Open Proportion of each home range that was defined as “open” (vegetation between 0.00 - 0.02 m) at a 10 m2 resolution. Lidar 2016 USGS 2016b P_Shrub Proportion of each home range that was defined as “shrub” (vegetation between > 0.02 - ≤ 2.0 m) at a 10 m2 resolution. Defined because of its potential importance to moose as a food source, but also to winter ticks as they tend to quest (or seek a host) within this height range. Lidar 2016 USGS 2016b P_Forage Proportion of each home range that was defined as potential “forage” (vegetation ≤ 3.0 m) or vegetation that was within  reach of moose at a 10 m2 resolution. Lidar 2016 USGS 2016b P_Cover Proportion of each home range that was defined as “cover” (vegetation between > 3.0 - < 6.0 m) at a 10 m2 resolution. Lidar 2016 USGS 2016b P_Canopy Proportion of each home range that was defined as “canopy” (vegetation > 6.0 m) at a 10 m2 resolution. Defined because of its potential importance to moose as a source of protection for thermal stress or shelter during periods of deep snow. Lidar 2016 USGS 2016b P_Wetland Proportion of each home range defined as “wetland” forest (NLCD emergent and woody wetland classifications were combined to represent general wetlands) at a 30 m2 resolution. NLCD 2016 USGS 2016a P_Deciduous Proportion of each home range defined as “deciduous” forest (> 75% of the tree species shed foliage simultaneously in response to seasonal change) at a 30 m2 resolution. NLCD 2016 USGS 2016a P_Evergreen Proportion of each home range defined as “evergreen” forest (> 75% of the tree species maintain their leaves all year) at a 30 m2 resolution. NLCD 2016 USGS 2016a P_Mixed Proportion of each home range defined as “mixed” forest (neither deciduous nor evergreen species are > 75% of total tree cover) at a 30 m2 resolution. NLCD 2016 USGS 2016a P_OpenWater Proportion of each home range defined as “open water” (areas of open water, < 25% cover of vegetation or soil) at a 30 m2 resolution. NLCD 2016 USGS 2016a P_Developed Proportion of each home range defined as “developed” (indication of impervious surfaces, covering all definitions in NLCD legend to represent all development) at a 30 m2 resolution. NLCD 2016 USGS 2016a P_SnoMoTrails Proportion of each home range defined as “snowmobile trail” (defined by ANR trail maps) at a 30 m2 resolution. VCGI 2019 VTANR 2019 Elevation A measure of the elevation (m) in the study area/100. VCGI 2002 VCGI 2002 ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 77 linear mixed-effects model (Bates et al. 2015) to determine if the average home range variable differed by sex, season, and age. We included moose identification number as a random effect, as the same moose had multi- ple home ranges across the study period. Objective 2: Resource Utilization Function (RUF) Models by Age, Sex, and Season We estimated a utilization distribution (UD) for each kernel home range; a three- dimensional probability map identifying peaks (frequently-used areas) and valleys (less-used areas) that explains where an individual was most likely to occur within the home range boundary (Worton 1989). We used the kernelUD function in the R package adehabitatHR (Calenge 2011) to estimate 99% kernel UDs for each moose during each season (dormant and growth) and year (2017–2019). UD pixel probabilities were converted to percentiles for interpretation and analysis (Marzluff et al. 2004, Donovan et al. 2011), where pixels of lower UD probabilities (i.e., valleys) had low percen- tiles while the highest UD probabilities (i.e., peaks) had percentiles closer to 100. RUF predictor variables included habi- tat and terrain variables that may define the UD peaks and valleys for each home range (Table 1). For each variable, we used the focal function in the R raster package (Hijmans and Van Etten 2012) to create layers at the 400 m and 1 km scales, in which each pixel’s value was the mean value across the given scale, centered on the pixel itself. For example, while a single NLCD pixel (30 m2) may indicate deciduous forest, its focal value at the 400 m scale provides the proportion of deciduous forest within a 400 m of that cell. This effectively smoothed the resource level at each pixel, allowing analysis of habitat use to more accurately reflect how a moose may perceive and use resources on the landscape (i.e., daily move- ment patterns; Wattles and DeStefano 2013). We extracted the underlying raster value for each cell in an individual’s UD. Due to computing constraints, large home ranges with more than 500 UD locations were reduced to 500 points by randomly selecting 5 points from each UD percentile. Each record (row) of the dataset was assigned a weight based on a GPS transmission collar bias study (see S1). Weights were computed as the inverse of the estimated transmission rate, such that records with a low probabil- ity of successful transmission had higher weights. A model set of 9 linear regression mod- els was used to relate home range space use to the resource attributes (Table 2). To avoid multicollinearity in any given model, highly correlated variables (r > 0.5) were ultimately dropped from consideration as explanatory variables in RUF models. Each RUF model estimated how likely moose were to use a given part of their home range as a function of the smoothed NLCD and lidar variables; UD percentile was used as the response vari- able. The model set included models that estimated habitat use for moose with the NLCD and lidar variables smoothed at 400 m, NLCD at 1 km and lidar at 400 m, and NLCD at 400 m and lidar at 1 km. We also examined how well NLCD variables only or lidar variables only (at both smooth- ing scales) estimated habitat use (Table 2). Each RUF model in the set was run with both unstandardized explanatory vari- ables and standardized explanatory variables (z-scores), and with and without the GPS weights (4 analyses of the model set per home range). For each analysis, the models were ranked using Akaike’s Information Criterion (AIC) to identify the best-fit model (Burnham and Anderson 2002). The AIC ranks were averaged across the individual home range models to identify BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 78 a single, best model to estimate habitat use for moose by age, sex, and season. The pop- ulation level RUF by age, sex, and season was estimated by averaging the regression coefficients across individuals (Marzluff et al. 2004). As such, individual animals were treated as the experimental unit, result- ing in unbiased model coefficients for popu- lation level RUFs (Millspaugh et al. 2006). Following Marzluff et al. (2004), the top-ranked, unstandardized, weighted mod- els were used to create maps of habitat use, while the top-ranked, standardized models were used to identify the most important variables across moose. Objective 3: Mapping Moose Habitat Use We created habitat use rasters for each age, sex, and season using the averaged unstan- dardized coefficients from analyses where GPS radio-collar bias study weights were applied. We used the overlay function in the raster package (Hijmans et al. 2015) to compute a score estimating habitat use by moose for each 10 m2 cell within the study area by multiplying each resource variable in the raster stack by its corre- sponding averaged unstandardized coeffi- cient. The resultant maps contained cell values that described habitat use for moose by age, season, and sex. RESULTS Objective 1: General Home Range Characteristics We calculated kernel home ranges for 57 females and 17 males (Table 3). Average home range area was nearly the same during the dormant season for both mature and young female moose (40 and 39 km2, respectively; Table 3). Female home range areas increased during the growth season for both mature and young individuals (+ 4 and + 19 km2, respectively). Male home Table 2. The 9 Resource Utilization Functions (models) analyzed for each radio-collared moose (n = 74) in northeastern Vermont, USA. Models include combinations of NLCD (National Land Cover) and Lidar (light detection and ranging) variable classes smoothed at two scales (400 m and 1 km). The formula defines the utilization distribution percentiles (PCT100) as a function of the different resources within each model (100th percentile representing core-use areas and 1st percentile representing valleys or areas of least-use). Model Formula K 1. Intercept PCT100 ~ 1 1 2. NLCD (400m) & Lidar (400m) PCT100 ~ nlcd_evergreen_400m + nlcd_mixed_400m + nlcd_ wetland_400m + forage_400m + canopy_400m + Elevation 7 3. NLCD (1km) & Lidar (1km) PCT100 ~ nlcd_evergreen_1km + nlcd_mixed_1km + nlcd_wetland_1km + forage_1km + canopy_1km + Elevation_s 7 4. NLCD (400m) & Lidar (1km) PCT100 ~ nlcd_evergreen_400m + nlcd_mixed_400m + nlcd_ wetland_400m + forage_1km + canopy_1km + Elevation 7 5. NLCD (1km) & Lidar (400m) PCT100 ~ nlcd_evergreen_1km + nlcd_mixed_1km + nlcd_wetland_1km + forage_400m + canopy_400m + Elevation_s 7 6. Lidar (400m) PCT100 ~ forage_400m + cover_400m + canopy_400m 4 7. Lidar (1km) PCT100 ~ open_1km + forage_1km + canopy_1km + Elevation 5 8. NLCD (400m) PCT100 ~ nlcd_evergreen_400m + nlcd_mixed_400m + nlcd_ wetland_400m + nlcd_open_400m + Elevation 6 9. NLCD (1km) PCT100 ~ nlcd_evergreen_1km + nlcd_mixed_1km + nlcd_wetland_1km + nlcd_open_1km + Elevation 6 ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 79 Ta bl e 3. M ea n ho m e ra ng e si ze a nd p ro po rti on o f v er tic al st ru ct ur e an d fo re st ty pe fo r m oo se in V er m on t, U SA b y se x, a ge (m at ur e an d yo un g) , a nd a cr os s t w o se as on s fo r v eg et at io n (d or m an t a nd g ro w th ) f ro m 2 01 7 to 2 01 9. T he “ Fr eq ” ro w in di ca te s th e nu m be r o f h om e ra ng es a na ly ze d fo r t he g iv en s ex , s ea so n, an d ag e. T he s tu dy a re a m ea n co lu m n pr ov id es th e ov er al l p ro po rti on o f e ac h ha bi ta t f ea tu re fo r t he e nt ire s tu dy a re a. A N O VA re su lts in di ca te th e ef fe ct o f ea ch co va ria te (s ex , s ea so n, an d ag e) o n ea ch h ab ita t v ar ia bl e. V al ue s i n pa re nt he se s a re st an da rd d ev ia tio ns fo r h om e r an ge h ab ita t p ro po rti on s a nd p ro ba bi lit ie s fo r A N O VA re su lts . Se x Fe m al e Fe m al e Fe m al e Fe m al e M al e M al e A N O VA F- St at (p -v al ue ) Se as on D or m an t D or m an t G ro w th G ro w th D or m an t G ro w th A ge M at ur e Yo un g M at ur e Yo un g Yo un g Yo un g Fr eq 37 16 33 19 15 17 H R A re a (k m 2 ) 40 .5 4 (5 1. 33 ) 39 .9 6 (4 3. 84 ) 44 .0 4 (5 7. 92 ) 58 .9 3 (5 3. 37 ) 10 0. 37 (1 14 .5 6) 11 1. 02 (1 66 .0 1) St ud y A re a M ea n In te rc ep t Se x Se as on A ge P_ O pe n 0. 33 (0 .0 4) 0. 33 (0 .0 4) 0. 34 (0 .0 3) 0. 34 (0 .0 3) 0. 32 (0 .0 4) 0. 34 (0 .0 3) 0. 35 F = 89 10 .8 0 (0 .0 0) F = 0. 00 (0 .9 8) F = 53 .6 1 (0 .0 0) F = 0. 01 (0 .9 3) P_ Sh ru b 0. 17 (0 .0 2) 0. 17 (0 .0 2) 0. 17 (0 .0 2) 0. 17 (0 .0 2) 0. 17 (0 .0 2) 0. 17 (0 .0 1) 0. 16 F = 58 72 .0 3 (0 .0 0) F = 0. 31 (0 .5 8) F = 2. 74 (0 .1 0) F = 1. 59 (0 .2 1) P_ Fo ra ge 0. 20 (0 .0 3) 0. 20 (0 .0 3) 0. 20 (0 .0 3) 0. 20 (0 .0 3) 0. 20 (0 .0 2) 0. 19 (0 .0 1) 0. 18 F = 53 25 .2 8 (0 .0 0) F = 0. 36 (0 .5 5) F = 6. 27 (0 .0 1) F = 1. 58 (0 .2 1) P_ C ov er 0. 10 (0 .0 2) 0. 10 (0 .0 2) 0. 10 (0 .0 2) 0. 10 (0 .0 2) 0. 10 (0 .0 2) 0. 10 (0 .0 1) 0. 09 F = 31 11 .8 0 (0 .0 0) F = 0. 08 (0 .7 8) F = 25 .2 6 (0 .0 0) F = 0. 30 (0 .5 9) P_ C an op y 0. 37 (0 .0 5) 0. 37 (0 .0 5) 0. 36 (0 .0 4) 0. 36 (0 .0 4) 0. 38 (0 .0 3) 0. 37 (0 .0 2) 0. 37 F = 79 35 .8 2 (0 .0 0) F = 0. 06 (0 .8 1) F = 4. 82 (0 .0 3) F = 0. 89 (0 .3 5) P_ D ec id uo us 0. 41 (0 .1 4) 0. 41 (0 .1 4) 0. 46 (0 .1 1) 0. 46 (0 .1 1) 0. 42 (0 .1 3) 0. 45 (0 .1 0) 0. 44 F = 12 63 .5 0 (0 .0 0) F = 0. 14 (0 .7 1) F = 28 .1 2 (0 .0 0) F = 2. 42 (0 .1 2) P_ Ev er gr ee n 0. 13 (0 .1 1) 0. 13 (0 .1 1) 0. 11 (0 .0 7) 0. 11 (0 .0 7) 0. 13 (0 .0 7) 0. 11 (0 .0 5) 0. 12 F = 18 1. 24 (0 .0 0) F = 0. 33 (0 .5 7) F = 18 .8 2 (0 .0 0) F = 0. 05 (0 .8 2) P_ M ix ed 0. 28 (0 .0 8) 0. 28 (0 .0 8) 0. 27 (0 .0 7) 0. 27 (0 .0 7) 0. 29 (0 .0 7) 0. 28 (0 .0 6) 0. 28 F = 13 57 .0 8 (0 .0 0) F = 0. 53 (0 .4 7) F = 3. 43 (0 .0 6) F = 1. 21 (0 .2 7) P_ W et la nd 0. 08 (0 .0 9) 0. 08 (0 .0 9) 0. 06 (0 .0 6) 0. 06 (0 .0 6) 0. 09 (0 .1 0) 0. 07 (0 .0 7) 0. 07 F = 78 .2 6 (0 .0 0) F = 0. 33 (0 .5 7) F = 16 .8 7 (0 .0 0) F = 0. 27 (0 .6 0) P_ O pe nW at er 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 2) 0. 01 F = 39 .1 2 (0 .0 0) F = 0. 77 (0 .3 8) F = 1. 96 (0 .1 6) F = 0. 00 (0 .9 5) P_ D ev el op ed 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 02 F = 29 4. 75 (0 .0 0) F = 0. 18 (0 .6 7) F = 0. 04 (0 .8 5) F = 0. 31 (0 .5 8) P_ Sn oM oT ra ils 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 01 (0 .0 1) 0. 02 F = 25 0. 44 (0 .0 0) F = 0. 80 (0 .3 7) F = 3. 21 (0 .0 7) F = 1. 42 (0 .2 4) BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 80 ranges were much larger than female home ranges during the dormant (100 km2) and growth (111 km2) seasons. Like females, male home ranges were larger during the growth season (Table 3). The overall composition of home ranges for males and females, both mature and young, did not vary significantly (Fig. 2). The mean cover type and vertical structure values within home ranges were similar to the habitat composition of the study area. In terms of NLCD composition, the average home range was largely deciduous and mixed forest, but forest composition differed slightly between the dormant and growth seasons, with higher proportion of deciduous forest during the growth season (e.g., 0.46 for females) than the dormant season (e.g., 0.41 for females; Fig. 2). The percentage of development (including snowmobile trails) in the average home range was very low due to undeveloped nature of the study area. Unlike the NLCD, the vertical forest struc- ture of home ranges as measured through lidar remained relatively constant across sea- sons (~35% canopy and ~ 33% open forest), Fig. 2. ANOVA results estimating habitat composition of moose (n=74) across sex and season (dormant and growth) during 2017–2019 in northeastern Vermont, USA. Lidar covariates described the vertical structure of the forest (P_Open, P_Shrub, P_Forage, P_Cover, and P_Canopy), while the NLCD (P_Deciduous, P_Mixed, P_Evergreen, P_Wetland, P_Developed, P_OpenWater) and VCGI (P_SnoMoTrails) covariates described the forest type. The y-axis is the home range mean proportion (± SD) of each habitat feature across all moose. The horizontal line indicates the average proportion of each habitat feature across the entire study area. ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 81 but there was a slight increase in the overall proportion of “open” from the dormant sea- son to the growth season (Fig. 2). Objective 2. Resource Utilization Functions Each RUF model in the Table 2 model set was run with both unstandardized explana- tory variables and standardized explanatory variables (z-scores), and with and without the GPS weights (S1), resulting in 4 analyses of the model set per home range. Across analyses and across moose, the top ranked model was the model in which both NLCD and lidar resources were measured at 1 km2 (Fig. 3). The models that were least sup- ported in estimating moose habitat use included the intercept model and those that included only NLCD habitat types or only lidar vertical structure at the 400 m scale. There was very little variation in the rank- ings of the models with or without radio- collar bias weights applied, and little variation in rankings between models where coefficients were standardized versus unstandardized (Fig. 3). Average adjusted R-squared values ranged from 0.00 (intercept model) to 0.34 (top model) across all models (Fig. 4). As expected, the top model (NLCD and lidar at Fig. 3. Average AIC ranking for each RUF model for moose in northeastern Vermont, USA. For each home range, 9 RUF models were evaluated and ranked with AIC approaches. Bars indicate the average ranking across individuals by age, sex, and season (dormant and growth). The lowest AIC ranking on average describes the top model for estimating habitat use. RUF models were evaluated for each group with and without radio-collar bias weights (S1). BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 82 1 km2) was on average the best fit model for both male and female moose during the dor- mant period and the growth period. Female moose habitat use was more consistent than male, with mature females being the most consistent (Fig. 4). The top model estimated UD percen- tiles as a function of the proportion of ever- green, mixed, and wetland forests (within 1 km2), as well as forage and canopy struc- ture (within 1 km2), and elevation (model 3, Table 2). Some of the NLCD and lidar hab- itat variables in this top model were cor- related with other variables that were intentionally excluded from the model function (Fig. 5). For example, deciduous forest was relatively widespread in the study area and meaningful to moose habitat use but was not included in any model as it was strongly negatively correlated with evergreen forest. Therefore, we interpreted strong negative effects of evergreen forest in a model as positive effects of deciduous forest; this was the case during the growth period for all moose. The average standardized betas from the top model indicated the relative importance of resources within moose home ranges (Fig. 6). During the growth season, the amount of vegetation classified as forage (height ≤ 3.0 m) within 1 km of a given loca- tion within a home range was highly import- ant in shaping the UD surface, especially for females. This pattern held for young females Fig. 4. Adjusted R-Squared values across the models in the RUF model set for moose in northeastern Vermont, USA. The average adjusted R-squared values (y-axis) show how well each of the models (x-axis) estimated habitat use; error bars are standard errors. No mature male moose were radio- collared for the study, thus all males are categorized as “young adult”. ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 83 during the dormant season (Fig. 6). Forage habitat was strongly associated with use for young male moose as well, but higher eleva- tions and the proportion of mixed forests were of greater importance in estimating UD percentile during both the dormant and the growth seasons. All moose had a negative association with evergreen forests, much more so during the growth period. Areas of an individual’s home range that were classi- fied as wetland habitat typically had low UD percentiles, except during the dormant sea- son for females and the growth period for males, where there was increased use. The canopy classification (vegetation > 6.0 m) was a positive predictor of use for young female moose during the growth period. Notably, young females did not heavily use forest classified as coniferous (mixed or evergreen classifications) during the dor- mant season, instead opting for high forage structure in deciduous forest (Fig. 6). Objective 3. Mapping Moose Habitat Use We used the unstandardized (Fig. 7), weighted coefficients to create maps of hab- itat use within the study area (WMUs E1 & E2) for mature and young adult females during the dormant and growth seasons (S2 and S3) and for young adult males during the dormant and growth seasons (S4). As reflected by the RUF coefficients, high qual- ity “hotspots” differ for male vs. female and young vs. mature moose. Maps of habitat use for mature female moose during the growth period are presented for the top model (NLCD & lidar at 1 km2) and NLCD (1 km2) only to illustrate differences (Fig. 8). DISCUSSION To date, wildlife managers in Vermont have been challenged to effectively employ habi- tat management as a tool for managing moose populations due to the lack of fine- scale information on moose space use and Fig. 5. Correlations between 2016 National Land Cover Database (NLCD) and lidar variables used to describe habitat use for moose in northeastern Vermont, USA. Darker cells represent highly correlated variables, while the lighter cells indicate variables that were less correlated. BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 84 habitat characteristics in Vermont. Our results help to fill these knowledge gaps by identifying potential “hotspots” indicating areas throughout the landscape of high use by moose. This knowledge may inform wildlife management about the consider- ation of the spatial distribution of optimal habitats and which forest composition and structure to conserve, modify, or create to promote healthier and more persistent popu- lations of moose across the region. We examined patterns of habitat use using RUFs based on error-corrected GPS collar data and multi-scale habitat and land cover information. No home range or UD estimation is without error (Powell and Mitchell 2012), but our results appear to be consistent with what is known about moose ecology in the Northeast (Crossley and Gilbert 1983, Healy et al. 2018). For all sea- sons (dormant and growth), sexes (male and female), and ages (mature and young adult), Fig. 6. The relative effects of habitat variables used by sex, age and season for moose in northeastern Vermont, USA. Values represent average standardized regression coefficients across all moose in a given category from the top model. Error bars represent standard error (± SE) for the coefficient estimates. ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 85 the most supported RUF models included both coarse-scale habitat composition (derived from NLCD) and fine-scale struc- ture (derived from lidar) at the 1 km2 scale. The two least supported models examined habitat use at the 400 m scale using NLCD or lidar data solely, indicating that the combina- tion of composition and structure better explained habitat use. AIC rankings order model support, but the ranking in and of themselves do not convey how much better the top model was relative to the others. AIC weights and ΔAIC metrics convey these dif- ferences and suggest that the NLCD + lidar model at the 1 km2 scale was superior to other models under consideration. For example, for adult females during the dormant season, the “NLCD (1 km2) and lidar (1 km2)” per- formed significantly better than the “NLCD (1 km2)” model (mean model weights = 0.54 Fig. 7. The effects of habitat variables used by sex, age, and season for moose in northeastern Vermont, USA. Values represent average unstandardized regression coefficients across all moose in a given category from the top model. Error bars represent standard error (± SE) for the coefficient estimates. BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 86 and 0.15 respectively; median model weights = 0.67 and 0.00 respectively; mean AIC dif- ference = -24.5; median AIC difference = -10.6). These findings support the notion that lidar is an effective tool for improving the ability of models to estimate patterns of habitat use, especially for larger bodied mammals, as other lidar-based studies have focused mainly on birds (e.g., Bradbury et al. 2005). Moose exhibited important seasonal dif- ferences in habitat use that likely reflect tem- poral changes in energetic and nutritional requirements and behavior associated with certain social activities across the year (Schwartz and Renecker 2007). For instance, during the growth period, moose maximize food intake to meet high energy demands, including lactation, rearing of young, and the accumulation of fat (Shively et al. 2019, Pekins 2020). We found that moose used areas with proportionally more regenerating forest (i.e., forage < 3.0 m) during the growth period, which included areas of recent tim- ber harvest, indicating the importance of focusing on regenerating forest patches. Moose also actively used regenerative forest in the dormant season (although to a lesser Fig. 8. Habitat use by mature female moose during the growth season using the top NLCD (1 km2) and lidar (1 km2) model and the NLCD (1 km2) only model, based on a Resource Utilization Function in northeastern Vermont, USA. Each map pixel provides a z-score, where 0 indicates average use. Positive scores indicate higher use than average, and negative scores indicate lower use than average. ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 87 extent), presumably to cope with energetic demands associated with heavy tick loads over the winter (Pekins 2020), thus further underlining the essential importance of this habitat type to managers. Additionally, our results indicated mature and young female moose had strong positive associations with lidar-derived can- opy structure during the growth (but not the dormant) period, which suggests that enhancing canopy cover may be an import- ant consideration for managing female habi- tat use. Given that the Vermont population is near the southern range limit for the species, thermal stress may affect habitat use more so than in other regions (Montgomery et al. 2019, Alston et al. 2020). In contrast to females, elevation was an important predic- tor of habitat use for young males in both the dormant and growth periods, indicating that the trade-offs between food and temperature may differ by sex or that other factors may be driving use. Although elevation only var- ied by 800 m, higher elevation can be cooler than lower areas and potentially offset the costs of not using canopied habitats. Certainly moose thermoregulate behavior- ally (e.g., Montgomery et al. 2019, Alston et al. 2020), but caution should be used when drawing conclusions about population-level responses from thermal stress (Lowe et al. 2010, Pekins 2020). In terms of landcover, it is important to emphasize that the effect sizes from our models are relative to the intercept, which encompasses deciduous landcover, the most dominant landcover class in our study area (28% mixed, 44% deciduous, and 12% evergreen). Young male moose used mixed forest types much more than young female moose, which complements other studies of habitat use in New England (Leptich and Gilbert 1989). Mixed forest cover was also used more by mature female moose in the dormant season, thus emphasizing the potential importance of this resource for breeding females. Both evergreen and wet- land had negative or no effect on habitat use for both sexes and seasons relative to deciduous  forest  (the  modelʼs  intercept).  We expected stronger evergreen and wet- land signals, as it is well documented that moose seek wetlands in summer and coniferous forest for shelter and as a source of forage in winter (Timmermann and McNicol 1988). Our results have implications for moose management in this region where declines in moose health, survival, and fecundity have been linked to heavy parasite loads, chief among them winter ticks (Jones et al. 2019, Pekins 2020, Debow et al. 2021). During our study, winter tick epizootics occurred in 2 of 3 years (2018 and 2019), in which calf mor- tality exceeded 50% and cause of death was attributed to atrophy due to winter tick infes- tation (Jones et al. 2019, Debow et al. 2021). As a single-host parasite, the prevalence of winter ticks in our study area is largely shaped by the distribution of moose in the fall (when ticks attach to moose) and in the spring (when ticks detach from moose to lay eggs on the forest floor). Thus, areas that are heavily used by moose in both fall and spring may actively promote the host-parasite cycle (Healy et al. 2018), highlighting the concept that habitat use should not be confused with habitat quality (Van Horne 1983). Blouin et al. (2021) found that habitat selection decisions made by adult females during the fall questing period can influence whether their calves survive or not the following spring. Reducing the impacts of ticks will largely involve efforts to reduce the abun- dance and distribution of ticks on the land- scape, which may be accomplished by reducing moose density through harvest (Ellingwood et al. 2020) or habitat manipu- lation that aims to reduce high-density BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 88 moose congregations that may support high tick densities (VFWD 2020). Given this and the need to promote healthy moose on the landscape, the spatial distribution of heavily used habitats (e.g., the amount and distribu- tion of regenerating forest) warrants man- agement attention (Johnson et al. 2002). Previous research indicates that home ranges of young females often overlap with that of the parent cow, leaving questions about how highly distributed optimal habitat on the landscape (i.e., timber harvests) would ulti- mately impact dispersal behaviors and local densities (Cederlund et al. 1987, Johnson et al. 2002). Future studies on moose habitat use may examine the comparative values of various habitat management options in best reducing high congregations of moose on the landscape while critically providing females with high quality forage and refuge habitat. Several caveats remain. We did not eval- uate the composition of forest species that are used within home ranges or the nutri- tional landscape (Schrempp et al. 2019) as tree and shrub species vary in nutritional value (Timmermann and McNicol 1988). Failure to recover adequate fat depots during late summer-autumn can predispose individ- uals to mortality (Schrempp et al. 2019), most notably parasitism by winter ticks regionally (Pekins 2020). Further, locally high seasonal density of moose (browsing) can measurably impact forage quality on the landscape (Persson et al. 2005, Bergeron et al. 2011), and is an important consider- ation in habitat management. Ultimately, wildlife and forest manag- ers must balance multiple objectives when developing resource management plans. Maintaining healthy populations of moose while also considering other societal and biological objectives, such as maximizing biodiversity and forest sustainability, can be complex. Structured decision approaches, as suggested by Franklin et al. (2020), can help elucidate tradeoffs. Our aim was to provide science-based information that identified habitat use and importance to moose (and presumably winter ticks) that may be incorporated into decision-making and management frameworks. Ultimately, accurate maps of critical habitats for moose will provide managers guidance when inte- grating moose population and forest man- agement goals. ACKNOWLEDGMENTS Funding was provided by the Vermont Fish and Wildlife Department in cooperation with the U.S. Fish and Wildlife Service Division of Wildlife and Sportfish Restoration – Wildlife Restoration Program, the University of Vermont, Vermont Fish and Wildlife Cooperative Research Unit, Safari Club International Foundation, and Vermont Electric Company. Veterinarian consultation was funded and provided by the Northeast Wildlife Disease Cooperative. We are also grateful for the support of the U.S. Fish and Wildlife Service National Wildlife Refuge System – Nulhegan Division (S. Agius), and land access from Weyerhaeuser Company, LandVest Inc., and many private landowners. We thank T. Smith, C. Lampart, L. Rossier, D. Hotchkiss, M. Luther, and E. Nichols for fieldwork assistance and S. MacFaden for lidar analy- sis. We appreciate the comments provided by B. Mosher, S. McKay, and two anony- mous reviewers, as they improved the clar- ity of the manuscript. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorse- ment by the U.S. Government. The Vermont Cooperative Fish and Wildlife Research Unit is jointly supported by the U.S. Geological Survey, University of Vermont, Vermont Fish and Wildlife Department, and Wildlife Management Institute. ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 89 REFERENCES Alston, J. M., M. J. Joyce, J. A. Merkle, and r. A. Moen. 2020. Temperature shapes movement and habitat selection by a heat-sensitive ungulate. Landscape Ecology 35: 1961–1973. AMelon, s. k., F. r. thoMpson III, and J. J. MIllspAugh. 2014. Resource utilization by foraging eastern red bats (Lasiurus borealis) in the Ozark region of Missouri. The Journal of Wildlife Management 78: 483–493. BAtes, D., M. MAechler, B. Bolker, s. WAlker, r. h. B. chrIstensen, h. sIngMAnn, B. DAI, g. grothenDIeck, p. green, and M. B. Bolker. 2015. Package ‘lme4’. Convergence 12: 2. Bergeron, D. h., p. J. pekIns, h. F. Jones, and W. B. leAk. 2011. Moose browsing and forest regeneration: a case study in north- ern New Hampshire. Alces 47: 39–51. Beyer, h. l., D. t. hAyDon, J. M. MorAles, J. l. FrAIr, M. heBBleWhIte, M. MItchell, and J. MAtthIopoulos. 2010. The interpretation of habitat preference metrics under use–availability designs. Philosophical Transactions of the Royal Society B: Biological Sciences 365: 2245–2254. BlouIn, J. A., J. DeBoW, e. rosenBlAtt, J. hInes, c. AlexAnDer, k. gIeDer, n. FortIn, J. MurDoch, and t. DonovAn. 2021. Moose Habitat Selection and Fitness Consequences During Two Critical Winter Tick Life Stages in Vermont, Untied States. Frontiers in Ecology and Evolution. 9: 204. doi: 10.3389/fevo.2021.642276 BoertJe, r. D., g. g. Frye, and D. D. young Jr. 2019. Lifetime, known-age moose reproduction in a nutritionally stressed population. Journal of Wildlife Management 83: 610–626. BrADBury, r. B., r. A. hIll, D. c. MAson, s. A. hInsley, J. D. WIlson, h. BAlzter, g. Q. AnDerson, M. J. WhIttInghAM, I. J. DAvenport, and p. e. BellAMy. 2005. Modelling relationships between birds and vegetation structure using airborne LiDAR data: a review with case studies from agricultural and woodland envi- ronments. Ibis 147: 443–452. BurnhAM, k. p., and D. r. AnDerson. 2002. A Practical Information-Theoretic Approach. Model Selection and Multi- model Inference, 2nd edition. Springer, New York, New York, USA. cAlenge, c. 2011. Home range estimation in R: the adehabitatHR package. Office national de la classe et de la faune sau- vage: Saint Benoist, Auffargis, France. ceDerlunD, g., F. sAnDegren, and k. lArsson. 1987. Summer movements of female moose and dispersal of their off- spring. Journal of Wildlife Management 51: 342–352. crossley, A., and J. R. gIlBert. 1983. Home range and habitat use of moose in northern Maine. Transactions of the Northeast Section of the Wildlife Society 40: 67–75. DeBoW, J., J. BlouIn, e. rosenBlAtt, k. gIeDer, W. cottrell, J. MurDoch, and t. DonovAn. 2021. Effects of Winter Ticks and Internal Parasites on Moose Survival in Vermont, USA. Journal of Wildlife Management. In press. DonovAn, t. M., M. FreeMAn, h. ABouelezz, k. royAr, A. hoWArD, and r. MIckey. 2011. Quantifying home range habitat requirements for bobcats (Lynx rufus) in Vermont, USA. Biological Conservation 144: 2799–2809. 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. ellIngWooD, D., p. J. pekIns, h. Jones, and A. r. MusAnte. 2020. Evaluating moose (Alces alces) population response to infestation level of winter ticks (Dermacentor albipictus). Wildlife Biology. doi: 10.2981/wlb.00619 FIeBerg, J. 2007. Kernel density estimators of home range: smoothing and the BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 90 autocorrelation red herring. Ecology 88: 1059–1066. FrAIr, J. l., 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 telem- etry data. Philosophical Transactions of the Royal Society B: Biological Sciences 365: 2187–2200. _____, 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 collar bias in habitat selection studies. Journal of Applied Ecology 41: 201–212. FrAnklIn, o., A. krAsovskIy, F. krAxner, A. plAtov, D. shchepAshchenko, s. leDuc, and B. MAttsson. 2020. Moose or spruce: A systems analysis model for managing conflicts between moose and forestry in Sweden. BioRxiv https://doi. org/10.1101/2020.08.11.241372. heAly, c., p. J. pekIns, l. kAntAr, r. g. congAlton, and s. AtAllAh. 2018. Selective habitat use by moose during critical periods in the winter tick life cycle. Alces 54: 85–100. hIJMAns, r., and J. vAn etten. 2012. Geographic analysis and modeling with raster data. R Package version 2: 1–25. _____, _____, J. cheng, M. MAttIuzzI, M. suMner, J. A. greenBerg, o. p. lAMIgueIro, A. BevAn, e. B. rAcIne, and A. shortrIDge. 2015. Package ‘ raster’. R package. horne, J. s., e. o. gArton, and k. A. sAger- FrADkIn. 2007. Correcting home-range models for observation bias. Journal of Wildlife Management 71:996–1001. hoWArD, r. A. 1966. Information value the- ory. IEEE Transactions on systems sci- ence and cybernetics 2:22–26. Johnson, D. D., r. kAys, p. g. BlAckWell, and D. W. MAcDonAlD. 2002. Does the resource dispersion hypothesis explain group living? Trends in Ecology & Evolution 17: 563–570. Johnson, D. h. 1980. The comparison of usage and availability measurements for evaluating resource preference. Ecology 61: 65–71. Jones, h., p. pekIns, l. kAntAr, I. sIDor, D. ellIngWooD, A. lIchtenWAlner, and M. o’neAl. 2019. Mortality assessment of moose (Alces alces) calves during suc- cessive years of winter tick (Dermacentor albipictus) epizootics in New Hampshire and Maine (USA). Canadian Journal of Zoology 97: 22–30. kIe, J. g. 2013. A rule-based ad hoc method for selecting a bandwidth in kernel home- range analyses. Animal Biotelemetry 1: 1–13. kretser, h., M. glennon, M. schWArtz, and k. pIlgrIM. 2011. Evaluating genetic connectivity and re-colonization dynam- ics of moose in the Northeast. Wildlife Conservation Society, Saranac Lake, New York, USA. lAnkester, M. W. 2010. Understanding the impact of meningeal worm, Parelaphostrongylus tenuis, on moose populations. Alces 46: 53–70. leFsky, M. A., W. B. cohen, g. g. pArker, and D. J. hArDIng. 2002. Lidar remote sensing for ecosystem studies: Lidar, an emerging remote sensing technology that directly measures the three-dimen- sional distribution of plant canopies, can accurately estimate vegetation structural attributes and should be of particular interest to forest, landscape, and global ecologists. BioScience 52: 19–30. leptIch, D. J., and J. r. gIlBert. 1989. Summer home range and habitat use by moose in northern Maine. Journal of Wildlife Management 53: 880–885. loWe, s. J., B. r. pAtterson, and J. A. schAeFer. 2010. Lack of behavioral responses of moose (Alces alces) to high ambient temperatures near the southern periphery of their range. Canadian Journal of Zoology 88: 1032–1041. MAckenzIe, D. I., J. D. nIchols, J. A. royle, k. h. pollock, l. BAIley, and https://doi.org/10.1101/2020.08.11.241372 https://doi.org/10.1101/2020.08.11.241372 ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 91 J. e. hInes. 2017. Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence. Elsevier, Cambridge, Massachusetts, USA. MArzluFF, J. M., J. J. MIllspAugh, p. hurvItz, and M. s. hAnDcock. 2004. Relating resources to a probabilistic measure of space use: forest fragments and Steller's jays. Ecology 85: 1411–1427. MAzerolle, M. 2017. Model selection and multimodel inference based on (Q) AIC (c). R package version 2.1–1. http://cran.r-project.org/web/packages/ AICcmodavg/index.html. MelIn, M., J. MAtAlA, l. MehtätAlo, J. pusenIus, and p. pAckAlen. 2016. Ecological dimensions of airborne laser scanning—analyzing the role of forest structure in moose habitat use within a year. Remote Sensing of Environment 173: 238–247. MIllspAugh, J. J., r. M. nIelson, l. McDonAlD, J. M. MArzluFF, r. A. gItzen, c. D. rIttenhouse, M. W. huBBArD, and s. l. sherIFF. 2006. Analysis of resource selection using uti- lization distributions. Journal of Wildlife Management 70: 384–395. MontgoMery, r. A., k. M. reDIllA, r. J. Moll, B. vAn Moorter, c. M. rolAnDsen, J. J. MIllspAugh, and e. J. solBerg. 2019. Movement modeling reveals the complex nature of the response of moose to ambient tempera- tures during summer. Journal of Mammalogy 100: 169–177. 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: 1–30. ncDc (nAtIonAl clIMAte DAtA center). 2019. Climate Data Online: Dataset Directory. Volume 2019. National Climate Data Center. (accessed January 2020). peArMAn-gIllMAn, s. B., J. e. kAtz, r. M. MIckey, J. D. MurDoch, and t. M. DonovAn. 2020. Predicting wildlife dis- tribution patterns in New England USA with expert elicitation techniques. Global Ecology and Conservation 21: e00853. pekIns, p. J. 2020. Metabolic and Population Effects of Winter Tick Infestations on Moose: Unique Evolutionary Circumstance? Frontiers in Ecology and Evolution 8: 176. doi: 103389/ fevo.2020.00176 persson, I.-l., k. DAnell, and r. BergströM. 2005. Different moose densities and accompanied changes in tree morphol- ogy and browse production. Ecological Applications 15: 1296–1305. poWell, r. A., and M. s. MItchell. 2012. What is a home range? Journal of Mammalogy 93: 948–958. pyle, e. 2002. ElevationDEM_DEM24. USGS. Waterbury, Vermont, USA. (accessed January 2020). schreMpp, t. v., J. l. rAchloW, t. r. Johnson, l. A. shIpley, r. A. long, J. l. AycrIgg, and M. A. hurley. 2019. Linking forest management to moose population trends: the role of the nutritional landscape. PloS One 14: e0219128. schWArtz, c. c., M. e. huBBert, and A. W. FrAnzMAnn. 1988. Energy requirements of adult moose for winter maintenance. Journal of Wildlife Management 52: 26–33. _____, and l. renecker. 2007. Nutrition and energetics. Pages 441–478 in A. W. Franzmann and C. C. Schwartz, editors. Ecology and Management of the North American Moose. University Press of Colorado, Boulder, Colorado, USA. shIvely, r. D., J. A. crouse, D. p. thoMpson, and p. s. BArBozA. 2019. Is summer food intake a limiting factor for boreal brows- ers? Diet, temperature, and reproduction http://cran.r-project.org/web/packages/AICcmodavg/index.html http://cran.r-project.org/web/packages/AICcmodavg/index.html https://www.ncdc.noaa.gov/cdo-web/datasets https://www.ncdc.noaa.gov/cdo-web/datasets https://maps.vcgi.vermont.gov/gisdata/metadata/ElevationDEM_DEM24.htm https://maps.vcgi.vermont.gov/gisdata/metadata/ElevationDEM_DEM24.htm https://maps.vcgi.vermont.gov/gisdata/metadata/ElevationDEM_DEM24.htm BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 92 as drivers of consumption in female moose. PLoS One 14: e0223617. teAM, r. c. 2018. R: A Language and Environment for Statistical Computing. Vienna, Austria. (accessed January 2020). tIMMerMAnn, h., and J. McnIcol. 1988. Moose habitat needs. The Forestry Chronicle 64: 238–245. usgs. 2016a. NLCD 2016 Land Cover Conterminous United States. U.S. Geological Survey. Sioux Falls, South Dakota, USA. (accessed January 2020). _____. 2016b. usgs Lidar Point Cloud VT Connecticut-River 2016 N5495E569 LAS 2018: U.S. Geological Survey. (accessed January 2020). vAn horne, B. 1983. Density as a misleading indicator of habitat quality. Journal of Wildlife Management 47: 893–901. VCGI. 2002. Slopes generated from USGS NED based DEM24 data. Eric Pyle, VCGI. Waterbury, Vermont, USA. verIssIMo, l. M. 2012. Landscape and forest structure at moose mortality sites on Isle Royale National Park. A LiDAR based assessment. M.S. Thesis. Michigan Technological University, Houghton, Michigan, USA. vFWD (Vermont Fish and Wildlife Department). 2020. Big Game Management Plan: 2020–2030. Vermont Fish and Wildlife Department, Montpelier, Vermont, USA. VTANR. 2019. VAST Trails hand drawn on USGS Topographic quads by foresters of the Vermont Department of Forests, Parks, & Recreations using orthophotos, survey data, and personal knowledge of the area as references. Agency of Natural Resources, Montpelier, Vermont, USA. WAttles, D. W., and s. DesteFAno. 2013. Space use and movements of moose in Massachusetts: implications for conser- vation of large mammals in a fragmented environment. Alces 49: 65–81. Worton, B. J. 1989. Kernel methods for estimating the utilization distribution in home-range studies. Ecology 70: 164–168. https://www.R-project.org/ https://www.R-project.org/ https://www.mrlc.gov/data/nlcd-2016-land-cover-conus https://www.mrlc.gov/data/nlcd-2016-land-cover-conus https://geodata.vermont.gov/pages/elevation#datasets https://geodata.vermont.gov/pages/elevation#datasets ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 93 S1: GPS RADIO-COLLAR BIAS STUDY RUFs estimated from GPS collar data may suffer potential bias if locations are not observed equally. Successful communica- tion between the radio-collar and satellites may depend on habitat types or terrain (e.g., the likelihood of obtaining a GPS coordinate may be lower in evergreen forest than in an open field). Failure to correct for GPS collar bias may in turn bias RUFs (Frair et al. 2004, Horne et al. 2007, Frair et al. 2010). To account for radio-collar bias, we estimated transmission rates as a function of habitat and terrain by deploying 12 VECTRONIC GPS radio-collars across 60 randomly selected sites that varied by the cover types used in the analysis (deciduous, coniferous, mixed, wetland forests and open habitats). We assumed that habitat cover types affected transmission rate and GPS fix rate (i.e., collection of locations from the GPS satellite system) the same. Cover type of each site was derived from NLCD 2016 land cover classifications at 30 m resolu- tion (USGS 2016a) and confirmed by ground-truthing. Terrain covariates for each site (slope, aspect, and elevation) were also used to account for location bias and derived from U.S. Geological Survey data (Pyle 2002). Sites were stratified by habitat inside a 200 m buffer of VT Rt. 105 from Island Pond, Vermont east to Bloomfield, Vermont and ranged between 268 and 403 m in ele- vation, 0.01 and 12.28 in slope angle degrees, and included all aspect categories (North as < 45° and ≥ 315°, East as ≥ 45°  and < 135°, South as ≥ 135° and < 225°, and  West as ≥ 225° and < 315°).  Collars were fastened to a pole posi- tioned downward (mimicking a standing moose) at each trial site, approximately 1 m off the ground. GPS collars were deployed for a minimum of 2 days; when inactive for >5 h, the collars switched to mortality mode, sending a GPS coordinate every half hour for 6 h (12 total transmis- sions per site, per collar). Each collar was moved to a new, randomly selected site after a minimum of two days until all 60 sites were sampled. Sampling was repeated during two distinct seasons that coincided with the broad seasonal classifications of growth and dormant: “leaf on” (08 July to 29 July 2019) and “leaf off” (08 April to 29 April 2019). Thus, total sample size was: 60 sites * 2 seasons * 12 transmission attempts = 1,440 transmission attempts. We analyzed the probability of a trans- mission being successful (1) or unsuccess- ful (0) as a function of the terrain and cover type variables, while also accounting for the random effect of the individual radio-collar, using mixed-effects logistic regression. Eight models of the covariate data were evaluated to understand effects on transmission probability. Each covariate (cover type, elevation, aspect, slope, sea- son, and intercept) was modelled separately and two models were developed that accounted for combined effects: a model estimating the effect of all covariates together on transmission rate and a model estimating the interaction of season (leaf-on or leaf-off) across cover type variables on transmission. Akaike’s Information Criterion adjusted for small sample size (AICc) and Akaike weights (AICWt) were used to determine the models that best explained the data (Burnham and Anderson 2002). The best-supported model was then applied to objective 3 (analysis of utilization distribu- tions) to correct for collar transmission bias. All analyses were conducted in R (Team 2018). Mixed-effect models were analyzed in the R package lme4 (Bates et al. 2015) and model selection metrics were calcu- lated with the R package AICcmodavg (Mazerolle 2017). BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 94 RESULTS A majority of all GPS collar transmission attempts (76%, n = 1,094 of 1,440) were successful. Of the eight models evaluated, the interaction model was the top-ranking model, carrying 68% of the model weight (Table S1–1). This model examined all landcover and terrain covariates on GPS transmission rates, and further explored the effect of cover type dependent upon season (leaf-off vs. leaf-on). The next best model estimated the effect of all covariates (with- out interactions) on collar transmission probability and carried 31% of the model weight (Table S1–1). In the top model (interaction), the leaf-on period, evergreen forest, mixed forest, and higher elevations had significant negative Table S1–1. Comparison of the 8 mixed effect logistic regression models for GPS transmission bias in northeastern Vermont, USA. Model K LL AICc Delta AICc AICc Wt 1. Interaction 16 -717.24 1466.86 0.00 0.68 2. All 12 -722.10 1468.42 1.55 0.31 3. CoverType 6 -733.74 1479.54 12.68 0.00 4. Slope 3 -747.73 1501.49 34.62 0.00 5. Elevation 3 -760.67 1527.36 60.50 0.00 6. Intercept 2 -765.11 1534.23 67.37 0.00 7. Season 3 -764.66 1535.33 68.47 0.00 8. Aspect 5 -764.73 1539.50 72.63 0.00 Table S1–2. Coefficients (with 95% confidence intervals) of the top mixed effect logistic regression model (interaction) describing the effect of cover types and their interaction with the leaf-on vs. leaf-off period and the effect of terrain characteristics on GPS radio-collar transmission rates. Model Parameter Estimate Std. Error Upper 95 Lower 95 Interaction (Intercept) 4.51 1.00 6.48 2.55 Leaf-On -0.93 0.37 -0.21 -1.65 Evergreen Forest -1.48 0.35 -0.80 -2.16 Mixed Forest -1.00 0.36 -0.31 -1.70 Open -0.25 0.42 0.56 -1.07 Wetland -0.17 0.44 0.69 -1.04 Slope -1.18 0.28 -0.63 -1.72 Aspect (North) 0.35 0.24 0.81 -0.12 Aspect (South) 0.16 0.19 0.54 -0.22 Aspect (West) 0.47 0.24 0.94 0.00 Elevation -0.65 0.26 -0.15 -1.16 Leaf-on: Evergreen Forest 1.34 0.45 2.22 0.45 Leaf-on: Mixed Forest 0.75 0.44 1.61 -0.11 Leaf-on: Open 0.75 0.52 1.77 -0.27 Leaf-on: Wetland 0.53 0.54 1.59 -0.53 ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 95 impacts (P = <0.05) on transmission proba- bility when compared to the intercept ref- erence (leaf-off, east-facing, deciduous) (Table S1–2). The only significant positive effects on transmission rate was when the aspect was west-facing, and during the leaf-on period in evergreen forest (Table S1–2). In the all covariate, interaction model, the average probability of transmission was highest in the open and wetland cover types, at a nearly 90% success rate (Fig. S1–1). The lowest probability of a successful GPS trans- mission was in evergreen forest during the leaf-off period (63% success rate), which we would predict given the denser canopy structure that could impact the relationship of the collar to the satellite. Similarly, trans- mission rates in all cover types except ever- green forest were lower during the leaf-on period (Fig. S1–1). This decrease in success- ful transmissions rates from the leaf-off to the leaf-on period was most significant in the deciduous forest type (87% to 73%, respectively), as would be expected in a fully canopied seasonal forest. We used the predict function from the package lme4 (Bates et al. 2015) to obtain the probability of a successful GPS trans- mission for each UD value, depending on the UD’s covariate values. Figure S1–1. The average probability of transmission of Survey Globalstar V7.1 GPS radio-collars placed in five different habitats in northeastern Vermont, USA during a leaf-on and leaf-off period (2019). Vertical bars represent the range in predicted ping rate for any given category. For example, in the deciduous “leaf off” period, predicted ping rates ranged from 0.75 to 0.93, resulting from collars placed at different elevations, aspects, and slopes. BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 96 S2. Map of habitat use by mature female moose during the dormant and growth seasons in northeastern Vermont, USA. ALCES VOL. 57, 2021 BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT 97 S3. Map of habitat use by young adult female moose during the dormant and growth seasons in northeastern Vermont, USA. BLOUIN ET AL. – MOOSE HABITAT USE IN VERMONT ALCES VOL. 57, 2021 98 S4. Map of habitat use by young adult male moose during the dormant and growth seasons in northeastern Vermont, USA.