99 WINTER HABITAT USE OF MOOSE IN CAPE BRETON, NOVA SCOTIA Jason I. Airst and Jason W. B. Power Nova Scotia Department of Lands and Forestry, Wildlife Division, Kentville, Nova Scotia, Canada ABSTRACT: Aerial survey data collected between 2001 and 2020 were used to assess winter habitat use by moose (Alces alces) in the Greater Highland Ecosystem of Cape Breton, Nova Scotia. These data were analyzed using generalized additive mixed models that explored the influence of habitat variables. We compared abundance estimates developed directly from the surveys to those estimated from habitat use. Moose generally occupied the same general area throughout the study despite a marked population decline. Moose favoured areas comprised of greater proportions of coniferous forest showing preference for younger forest, and moose meadows, areas of predominantly coniferous forest but with abnormal or retarded regeneration due to high moose herbivory. Moose occupied areas farther away from roads inferring that moose preferred areas with younger plant forage and lower human access. The use of long-term survey data coupled with related habitat use relation- ships provided a useful approach to assess temporal tends in abundance and habitat use of moose in Cape Breton. ALCES VOL. 57: 99–111 (2021) Key words: aerial survey; Alces alces; Cape Breton; GIS; habitat use; roads. A fundamental objective of wildlife management is to maintain healthy sustain- able wildlife populations (Fryxell et al. 2014). Monitoring population-wide demo- graphics such as abundance and vital rates is one effective approach for managing popu- lations (Williams 2011, Boyce et al. 2012, Fryxell et al. 2014). Another important strategy is to anticipate effects of manage- ment actions on wildlife populations by understanding wildlife-habitat relationships in the context of environmental change (Krausman 1999, Hebblewhite and Merrill 2008). In this study, our primary objective was to assess moose (Alces alces) habitat use in the Greater Highland Ecosystem of Cape Breton, Nova Scotia using 20 years of aerial survey data. Aerial surveys are commonly used to assess large mammal populations active during winter (Gasaway et al. 1986, Kantar and Cumberland 2013). This approach has been used in many jurisdictions, including Nova Scotia, to assess changes in moose populations and to inform management deci- sions (Snaith et al. 2002, Van Beest et al. 2012, Andreozzi et al. 2016). Moose, like other mammals, show seasonal patterns of habitat use (Schwartz and Franzmann 1998, Manly et al. 2002, Van Beest et al. 2012). Therefore, by noting spatial variation in the abundance of moose observed during aerial surveys, we can infer habitat preferences across the landscape (Manly et al. 2002, Van Beest et al. 2012, Andreozzi et al. 2016). However, one must also account for how easily animals are detected during surveys, or their sightability. If not corrected for, sightability can bias survey results (Anderson and Lindzey 1996). In Cape Breton, moose habitat selection is predominantly shaped by a history of WINTER HABITAT USE IN NOVA SCOTIA – AIRST AND POWER ALCES VOL. 57, 2021 100 spruce budworm (Choristoneura fumiferana) infestations with the last major outbreak in the late 1970s (Bridgland et al. 2007, Smith et al. 2015). As a result, Cape Breton has experienced significant change in habitat composition and forest age structure in recent decades, providing a unique opportu- nity to examine habitat use by moose in response to this forest heterogeneity. Further, because road density varies greatly across the Cape Breton landscape, we also examined its potential influence on habitat use. We predicted that moose should be attracted to younger softwood forests and avoid areas nearer roads (Schwartz and Franzmann 1998, Manly et al. 2002, Van Beest et al. 2012). STUDY AREA In Cape Breton, moose were abundant prior to European arrival. However, subse- quent change in land use (i.e., habitat) and hunting pressure caused population decline that continued until the early 1900s when moose were virtually extirpated from the region (Pulsifer and Nette 1995, Davis and Browne 1996). In 1947 and 1948, Parks Canada reintroduced moose to the Cape Breton Highlands National Park (CBHNP) by translocating 18 moose from Elk Island National Park in Alberta (Davis and Browne 1996). This reintroduction was successful and moose gradually spread across the region, and by the mid-1980s the population was approximately 4000 animals (Pulsifer and Nette 1995). In the mid-1970s, a major spruce budworm out- break caused extensive tree mortality spur- ring an abundance of new tree growth (Bridgland et al. 2007, Smith et al. 2015). Taking advantage of this abundant food source, the population rapidly grew in the late-1990s until 2004 when it peaked at just over 8000 animals (Bridgland et al. 2007, Smith et al. 2015). The population gradually declined to ~ 4500 in 2015 and roughly halved again to 2300 animals by 2020 (Smith et al. 2015, Nova Scotia Lands and Forestry 2020a). Moose on Cape Breton Island are almost exclusively found in one region, the Greater Highland Ecosystem (GHE), with limited migration, immigration, and emigration. The GHE makes up the northwestern third of Cape Breton Island (Fig. 1) and is bordered by the Gulf of Saint Lawrence to the west where the land rises rapidly from the ocean to a height of 500 m. The land forms a large plateau that slopes eastward and northward toward the Atlantic Ocean. South of the CBNHP, there is an extensive road network built in the 1980s in conjunction with increased logging. The area has a maritime climate with average winter and summer Fig. 1. Map of Nova Scotia’s Greater Highland Ecosystem (GHE). The map also shows the location of Cape Breton Highlands National Park (CBHNP) and paved and unpaved roads. ALCES VOL. 57, 2021 AIRST AND POWER. – WINTER HABITAT USE IN NOVA SCOTIA 101 temperatures of −5 and 18°C, respectively. Annual precipitation averages 1053 mm of rainfall and 337 cm of snowfall (Environment Canada 2020). The GHE has 3 major land types: boreal forest, Acadian forest, and taiga (Smith et al. 2015) which are character- ized by a mixture of forest types and ages over recent decades (Table 1). Following disturbance events, boreal succession typ- ically starts with rapid growth of trees in the understory favouring faster growing shade-intolerant species such as white birch (Betula papyrifera). These shade intolerant species are eventually overtaken by slower growing shade tolerant conifer species (MacLean and Ostaff 1989, Smith et al. 2010). However, herbivores can affect this successional pattern through their foraging activity, most typically by over-browsing preferred plants (McLaren et al. 2004, Smith et al. 2010). Moose in many areas of the GHE consume nearly all conifer trees before they can grow large enough to escape herbivory. The resultant forest is characterized by open savannahs dominated by remnant white birch, alder (Alnus spp.), black spruce (Picea mari- ana), and herbaceous growth (Smith et al. 2010, 2015), hereafter termed “moose meadows” that are mostly located on the western side of the CBHNP. METHODS Aerial Survey Data were collected as part of 10 aerial surveys conducted over a 20-year period (2001, 2002, 2004, 2006, 2008, 2011, 2013, 2015, 2019, 2020). These helicopter surveys were conducted by 4 personnel: 2 back-seat observers, 1 front-seat recorder, and the pilot. All surveys were completed in 1–2 days during the first week of March and conducted in conditions of adequate visibil- ity and minimal precipitation. Effort was made to maximize sightability of moose by flying at low speed (< 130 km/h) and altitude (< 100 m above ground), and only recording animals within 150 m of either side of the helicopter (Gasaway et al. 1986). This work was completed as part of an effort by Nova Scotia Lands and Forestry to monitor moose in the area and done in partnership with Parks Canada, the Unama’ki Institute of Natural Resources, and the Confederacy of Mainland Mi’kmaq. To survey the area, we divided the GHE into 893 equal-sized survey blocks that were 2 min of longitude (~ 2.5 km) × 1 min of latitude (~1.9 km) large, or ~ 4.7 km2. Data were collected on an east-west transect flight flown over the midline of each block; tran- sect lines were 1 min of latitude apart (~ 1.9 km). Observations (sightings) of moose were made within 150 m of the heli- copter covering 1/6 of the block area; the number of moose observed per transect was recorded. A total of 8674 transects were flown over the 20-year study period. Table 1. Habitat composition of the Nova Scotia’s Greater Highlands Ecosystem in 1999, 2009, and 2020. Habitat type % of land cover 1999 2009 2020 Non-forest 37.3 28.1 28.1 <25 year conifer 12.0 8.7 2.8 25–40 year conifer 8.9 18.2 5.2 >40 year conifer 3.9 5.4 24.5 Uneven year conifer 4.4 4.6 4.4 <25 year deciduous 0.3 0.1 0.2 25–40 year deciduous 0.8 0.8 0.1 >40 year deciduous 9.2 8.0 8.5 Uneven year deciduous 0.5 1.3 1.3 <25 year mixed wood 2.8 1.2 0.6 25–40 year mixed wood 3.2 5.0 1.0 >40 year mixed wood 10.5 9.9 14.6 Uneven year mixed wood 2.2 5.0 4.8 Moose meadow 3.9 3.9 3.9 WINTER HABITAT USE IN NOVA SCOTIA – AIRST AND POWER ALCES VOL. 57, 2021 102 To control for sightability, we created a sightability correction factor (SCF) in each survey year. This required flying 6 equidistant, east-west transect lines over 28–44 survey blocks each year, counting all moose observed; lines were spaced 300 m apart. These same areas were re-flown using a more intense survey regime of 12 flights spaced 150 m apart with half the viewing distance. Moose counts from the two sur- veys were compared to determine if more animals were observed in the more intense survey. These comparisons were averaged to create a SCF in each survey year (Table 2). These same SCFs were used when histori- cal moose abundances were first calcu- lated in the area (Bridgland et al. 2007, Smith et al. 2015, Nova Scotia Lands and Forestry 2020a). Habitat and Surface Features Habitat type underlying survey transects was determined using Nova Scotia’s 1999, 2009, and 2020 Forest Inventories (Nova Scotia Lands and Forestry 1999, 2009, 2020b). The 1999 and 2009 forest invento- ries were based on forestry records and satellite imagery, and the 2020 inventory was based on forestry records and age progres- sion of the 2009 inventory. The 2001–2005 surveys were assessed with the 1999 inven- tory, the 2006–2014 surveys with the 2009 inventory, and those in 2015–2020 with the 2020 inventory. Habitat types were catego- rized according to forest type (deciduous, conifer, mixed) and stand age (< 25 years old, 25–40 years old, > 40 years old, and uneven aged) and included an additional for- est category for moose meadows. All non-for- ested habitats were combined into a single category. The forest inventories were also used to calculate the average percent crown closure within each transect. Additionally, provincial road (distinguishing paved and unpaved) and surface water maps were used to assess abiotic habitat characteristics. Because data collection was restricted to 150 m on either side of transect lines, we created a 150 m buffer from the transect using ArcGIS (version 10.5.1, Environmental Systems Research Institute 2018). We then calculated the percent habitat cover for each habitat type in each transect using the ArcGIS add-in Patch Analyst Version 5.2 (Rempel et al. 2016). To determine the average dis- tance to paved and unpaved roads and sur- face water features in each transect, we first calculated the Euclidian distance of all these features on the landscape. We then averaged these values in each transect to produce a measure of the distance to these features. We repeated the entire process for each survey block to subsequently use this information to determine the likely abundance and distribu- tion of moose across the landscape. Statistical Analysis To determine how site composition affects moose numbers in each transect, we used a generalized additive mixed model (GAMM) with a Poisson distribution using the R package “gamm” (Version 0.2–6, R Table 2. Sightability correction factors (SCF) for each year of aerial moose surveys in Nova Scotia. SCF were calculated based on the average difference between two sets of surveys flown over the same areas using different survey intensities. Year n SCF Variance 2001 36 1.21 0.02 2002 37 1.36 0.05 2004 30 1.12 0.01 2006 37 1.14 0.01 2008 32 1.14 0.01 2011 28 1.30 0.02 2013 36 1.25 0.02 2015 36 1.51 0.05 2019 44 1.48 0.07 2020 38 1.18 0.01 ALCES VOL. 57, 2021 AIRST AND POWER. – WINTER HABITAT USE IN NOVA SCOTIA 103 Core Team 2019, Wood and Scheipl 2020). A Poisson (P) distribution was chosen over a negative binomial (NB) distribution as it yielded a lower Akaike information criteria (AIC) value when the two were compared (AICP = 11,941.9; AICNB = 12,183.3, theta = 10). Generalized additive mixed models allow for both linear and non-linear effects in the model. For variables with non-linear effects, we used spline smoothers and a non-parametric reverse iterative approach to create separate model estimates for sections of the regression line (Wood 2017). The procedure involves assessing how much model fit is improved as more spline smoothers are used to explain the non-linear effect, while simultaneously penalizing the model for each smoother added. The result of this was that an optimal number of smoothers, or effective degrees of freedom (EDF), was chosen for each non-lin- ear effect. This method, however, did not yield a single model estimate for each non-linear effect as multiple smoothers were used when creating these model effects (Wood 2017, Wood and Scheipl 2020). Our objectives were to understand moose habitat use with respect to roads and forest type and age, while controlling for the underlying effect of forest cover; thus, we included crown closure as a non-linear fixed effect. Because survey blocks are spatially autocorrelated, we also used the x and y coordinates of the transect lines as a non-lin- ear effect (Kneib et al. 2009, Wood 2017). The remaining fixed linear effects were per- cent cover for each habitat type and the aver- age distance to roads and surface water features. Survey year was used as a random effect to account for population differences between surveys. For model selection, we fitted a global model and then used stepwise backward selection to remove non-important variables from the model. This involved sequentially removing the variable with the lowest Beta/ SE absolute value until the AIC value stopped declining (Pagano and Arnold 2009). Once all non-important variables were removed form the model, we compared models using their AIC values and estimated model weights. These weights were then used to create model averaged estimates that were unlogged to generate incidence rate ratios (IRRs). An IRR indicates how the rel- ative count of moose changes as you increase an independent variable by 1 unit. In the case of habitat type, one unit was the differ- ence between a site without or devoid of a habitat type (0) or entirely composed of that type (1). For roads and water, this was mea- sured as distance (km) from the feature. To assess if our model was able to accu- rately predict moose abundance in the GHE, we compared historic abundance estimates from this area (historical estimates) to abun- dance estimates based on our model and the habitat and surface feature data from the survey blocks (model estimates). The model estimate included the random intercept for each survey year. We also used the survey block data to determine the road density across the GHE. Because moose abundance was calculated to account for the GHE and not the entire area surveyed with transects, we applied a multiplier of 6 to the model estimates to account for the entire area. We further needed to multiply our estimates by the SCF for each year to account for missed animals. After this adjustment, we compared the observed and estimated abun- dances using a Pearson’s correlation coeffi- cient. We calculated the average abundance in each survey block using the 1999, 2009, and 2020 forest inventories to determine the likely moose distribution during 3 time peri- ods (2001–2005, 2006–2014, and 2015– 2020), and visually assessed these data for temporal and geographic trends in abundance. WINTER HABITAT USE IN NOVA SCOTIA – AIRST AND POWER ALCES VOL. 57, 2021 104 RESULTS Our best supported model identified rela- tionships between habitat use and several categories of forest type and age, as well as distance to water and roads (Table 3). While AIC values indicated an initial best model, (AIC = 11,933.5, variables = 16; Table 4), the removal of uneven aged coniferous for- est (11,934.6) and < 25-year-old deciduous forest (11,934.4) separately, and together (11,935.3), increased the AIC value by < 2 points. Because of the small difference between these four models, we used model averaging based on model weights to gener- ate the final estimates. We found that the probability of obser- vation was highest in moose meadows fol- lowed consecutively by younger coniferous forest types. The incidence rate ratios (95% CI) were 10.32 (8.25, 12.90) in moose meadows, 3.67 (3.02, 4.46) in < 25-year-old conifer forests, 3.16 (2.58, 3.88) in 25–40-year old conifer forests, and 1.30 (0.99, 1.69) in > 40-year old conifer forests (Table 3). Lower selection was found for >25-year-old deciduous for- est, 25–40 year-old mixed wood forest, and > 40-year-old mixed wood forest: incidence ratios were 2.87 (0.98, 8.40), 3.12 (2.17, 4.50), and 4.33 (3.39,5.53), respectively. There was minimal chance of observation in uneven aged deciduous for- est: incidence ratio = 0.45 (0.33, 0.610) (Table 5). Areas with greater average dis- tance to roads had higher observations of moose: paved = 1.05 (1.03, 1.07), unpaved = 1.27 (1.24, 1.30). The opposite was true for areas with a greater average distance to surface water (0.07 (0.05, 0.11) (Table 5). Road density (km of road/km2) was lowest in the CBHNP and highest in the area to the south: north of park = 0.29 (0.07, 0.51), park = 0.15 (0.03, 0.27), south of park = 0.71 (0.70, 0.73). Table 3. Removal of non-informative variable from the habitat use model generated from transects flights over the Greater Highland Ecosystem of Nova Scotia (10 surveys, n = 8674). Variables were sequentially desegrated (bolded) based on their Beta/SE absolute values until the simplified model’s AIC value began to increase. All models were general additive models with Poisson distributions and with year as a random factor. Transect location and % crown closure were treated as non-linear factors. Model # of variables AIC ΔAIC CE, CY, CM, CU, DE, DY, DM, DU, ME, MY, MM, MU, MOO, NF, Paved roads, Unpaved, Water 21 11,941.85 8.35 CE, CY, CM, CU, DE, DY, DU, ME, MY, MM, MU, MOO, NF, Paved roads, Unpaved, Water 20 11,939.90 6.40 CE, CY, CM, CU, DE, DY, DU, ME, MY, MM, MOO, NF, Paved roads, Unpaved, Water 19 11,937.90 4.40 CE, CY, CM, CU, DE, DY, DU, MY, MM, MOO, NF, Paved roads, Unpaved, Water 18 11,935.96 2.46 CE, CY, CM, CU, DE, DY, DU, MY, MM, MOO, Paved roads, Unpaved, Water 17 11,934.31 0.81 CE, CY, CM, CU, DE, DU, MY, MM, MOO, Paved roads, Unpaved, Water 16 11,933.50 0.00 CE, CY, CM, DE, DU, MY, MM, MOO, Paved roads, Unpaved, Water 15 11,934.50 1.00 Habitat types: <25 year conifer (CE), 25–40 year conifer (CY), >40 year conifer (CM), uneven year conifer (CU), <25 year deciduous (DE), 25–40 year deciduous (DY), >40 year deciduous (DM), uneven year deciduous (DU), <25 year mixed (ME), 25–40 year mixed (MY), >40 year mixed (MM), uneven year mixed (MU), moose meadow (MOO), non-forest (NF) ALCES VOL. 57, 2021 AIRST AND POWER. – WINTER HABITAT USE IN NOVA SCOTIA 105 Table 4. Comparison of models generated from flown transects over the Greater Highland Ecosystem of Nova Scotia using AIC to determine which habitat type and surface features best predict moose counts (10 surveys, n = 8674). All models were generalized additive mixed models with Poisson distributions and with year as a random factor. Transect location and % crown closure were both treated as non-linear factors. Model # of variables AIC ΔAIC Weight CE, CY, CM, CU, DE, DU, MY, MM, MOO, Paved roads, Unpaved, Water 16 11,933.5 0 0.38 CE, CY, CM, CU, DU, MY, MM, MOO, Paved roads, Unpaved, Water 15 11,934.4 0.9 0.25 CE, CY, CM, DE, DU, MY, MM, MOO, Paved roads, Unpaved, Water 15 11,934.6 1.1 0.22 CE, CY, CM, DU, MY, MM, MOO, Paved roads, Unpaved, Water 14 11,935.3 1.8 0.15 Null 4 13,077.7 1144.2 0 Habitat types: <25 year conifer (CE), 25–40 year conifer (CY), >40 year conifer (CM), uneven year conifer (CU), <25 year deciduous (DE), uneven year deciduous (DU), 25–40 year mixed (MY), >40 year mixed (MM), moose meadow (MOO) Table 5. Model averaged incidence rate ratio ratios (IRRs) of seeing a moose on a transect in Nova Scotia’s Greater Highland Ecosystem based on its habitat type and average distance from surface features (10 surveys, n = 8674). Results were generated from a generalized additive mixed model with a Poisson distribution, a with year as a random effect. Transect location and % crown closure were treated as non-linear effects. Habitat type IRRs are based on a 100% cover of that habitat type. EDF shows the number of spline smoothers used in each non-linear effect. Linear variables Variables Incidence rate ratio 95% CI (Intercept) 0.22 (0.16, 0.31) <25 year conifer 3.67 (3.02, 4.46) 25–40 year conifer 3.16 (2.58, 3.88) >40 year conifer 2.01 (1.54, 2.63) Uneven year conifer 1.30 (0.99, 1.69) <25 year deciduous 2.87 (0.98, 8.40) Uneven year deciduous 0.14 (0.05, 0.42) 25–40 year mixed wood 3.12 (2.17,4.50) >40 year mixed wood 4.33 (3.39, 5.53) Moose meadow 10.32 (8.25, 12.90) Avg. Distance paved road (per km) 1.05 (1.03, 1.07) Avg. Distance unpaved road (per km) 1.27 (1.24, 1.30) Avg. Distance surface water (per km) 0.07 (0.05, 0.11) Non-linear spline smoothers Variables EDF Chi2 Location (x, y) 24.90 328.28 % Crown closure 3.17 27.39 WINTER HABITAT USE IN NOVA SCOTIA – AIRST AND POWER ALCES VOL. 57, 2021 106 Because the GAM analysis precluded cal- culation of model estimates for the non-linear variables (average percent crown closure and moose group location), we plotted these data to assess potential relationships. Observations increased as average percent crown closure rose to ~17%, and then declined. This decline slowed around 55% but continued to 80%, the upper end of the range (Fig. 2). Moose groups were highly clustered across the landscape, creating distinct high and low abundance areas (Fig. 3). Both these patterns fit with our expectations based on the best model (EDF values; crown closure = 3.17; moose location = 24.90) (Table 5). When we estimated moose abundance for the entire GHE, the abundance estimates based on the model were highly correlated with the historical estimates (R2 = 0.94; Fig. 4). When examining the average moose distribution across the GHE in the 3 time periods (2001–2005, 2006–2013, and 2015–2020), we found that moose occupied the same areas throughout, although distri- bution contracted especially south of CBHNP (Fig. 3). The principal areas of con- sistent occupation were just north of CBHNP, the western side of the CBHNP, and to a lesser extent an area southeast of CBHNP, all protected areas with poor road access and limited forestry. DISCUSSION As we predicted, moose showed higher use and preference for younger softwood for- est and moose meadows. Young coniferous forests provide optimal moose forage, and Fig 2. The relationship between average % crown closure of a transect line and the number of moose seen on that line during flights over the Greater Highland Ecosystem of Nova Scotia (10 surveys, n = 8674). The regression line used was fitted using spline smoothers. a b c Fig. 3. Expected numbers of moose in each survey block in the Greater Highlands Ecosystem between 2001–2005 (a, est. pop. = 6883), 2006–2014 (b, est. pop. = 3563), and 2015–2020 (c, est. pop. = 3202) based on the best transect model and forest inventories in each period. The map also shows the location of Cape Breton Highlands National Park (CBHNP). ALCES VOL. 57, 2021 AIRST AND POWER. – WINTER HABITAT USE IN NOVA SCOTIA 107 moose meadows, although over-browsed, are dominated with early successional species (Schwartz and Franzmann 1998, Andreozzi et al. 2016). Our hypothesis that moose would avoid areas closer to roads was also supported. Roads pose a direct mortality risk to moose due to vehicle collisions and indi- rectly through habitat loss and fragmentation (Forman and Alexander 1998, Beazley et al. 2004, Eldegard et al. 2012). They also increase hunter access, which in turn leads to higher moose mortality (Forman and Alexander 1998, Beazley et al. 2004). One factor we could not fully account for was how moose population changes might have affected habitat use. Moose numbers in the GHE changed dramatically (> 4-fold) during the study from a high of 8000 animals to a low of 1350 (Bridgland et al. 2007, Smith et al. 2015, Nova Scotia Lands and Forestry 2020a). We accounted for the annual variation in habitat use by using a random intercept to account for overall population differences among years, but other subtle and unmodelled differences may have influenced habitat use such as resource (forage) compe- tition. Animals adjust foraging behavior and consumption to maximize caloric intake while minimizing energetic costs (Emlen 1966), which moose often accomplish by foraging on the most abundant vegetation rather than selectively feeding on the most nutritious plants (Wam and Hjeljord 2010, Bjornerass et al. 2012). Overall, this pattern is mediated by resource availability and com- petition (Wam and Hjeljord 2010). Our decision to use a single sightability correction factor each year may have affected the relative accuracy of our surveys because sightability often varies widely among habitat types (Anderson and Lindzey 1996, Quayle et al. 2001, McIntosh et al. 2007). By not accounting for this variation, we likely reduced the accuracy of certain abundance estimates (Quayle et al. 2001). However, since this method was used to cal- culate the historical abundance estimates and we did not have sightability information for each habitat type, we adopted this approach to compare the two data sets and believe our estimates are reasonable. Additionally, it is likely that the crown closure values used in our analysis were not entirely representative of the field conditions during the surveys. Crown closure values came from forest inventories based on con- ditions during the growing season, whereas our surveys occurred during winter. Thus, areas with higher proportions of deciduous forest likely had inflated crown closure val- ues due to the lack of leaves during the win- ter surveys. While only 10% of our site was dominated by deciduous trees (Table 1), the results in certain blocks were presumably biased to a degree. We found that moose abundance increased more rapidly in response to increased distance to unpaved rather than paved roads. This suggests that hunter access, not vehicle collision, was the more import Fig. 4. Comparing the historical moose population estimates for the Greater Highland Ecosystem and estimates based on the habitat use model. Model estimates were extrapolated to the entire survey area and both sets of estimates have had the same set of sightability correction factors applied to them. Results for 2013 were not used to calculate the R2 due to there not being an historical abundance estimate for this year. WINTER HABITAT USE IN NOVA SCOTIA – AIRST AND POWER ALCES VOL. 57, 2021 108 risk factor driving moose avoidance of roads. Other studies have shown that moose avoid areas near roads year-round (Forman and Alexander 1998, Dussault et al. 2007, Van Beest et al. 2012). However, on Cape Breton this pattern might simply reflect the relative location of roads, as unpaved roads are mostly inland where we expect most moose activity to occur, whereas paved roads are largely near the coast where fewer moose are expected (Fig. 1). Further, the timing of the winter survey relative to the autumn hunt- ing season may have some influence on win- ter abundance and location of moose, as animals in easily accessible areas may be har- vested at higher rate. Because moose move little during winter (Wattles and DeStefano 2013), their ability to disperse and recolonize areas is limited until the following spring. Roads also have a cumulative effect on moose populations. Beazley et al. (2004) suggested that areas with road density >0.6 km/km2 in mainland Nova Scotia were incapable of supporting significant moose populations. We found that the average road density was 0.71 km/km2 in the GHE south of CBHNP and most of this area supported low numbers of moose. The same pattern was observed in the moose management area directly south of the GHE, an area with even higher road density and fewer moose (Nova Scotia Lands and Forestry 2020a). If managers seek to increase the number of moose south of the CBHNP, one course of action may be to decommission certain roads in this area, an action proposed in other parts of Nova Scotia to help local moose popula- tions (Beazley et al. 2004). This action would be in accordance with Nova Scotia’s Endangered Species Act (Government of Nova Scotia 1998) as several endangered species live within the area. If full decom- missioning of roads is not feasible, managers could regulate access into areas of the GHE by using gates and signage during the hunting season. This approach has been suc- cessful in maintaining and growing wildlife populations in other jurisdiction with high human access (Cole et al. 1997, Crichton et al. 2004). Our results provide further sup- port for strategic planning and placement of new roads on the landscape. Our predictive maps in the GHE show that moose generally used the same areas throughout the study, although their distribu- tion was expected to contract given the mea- surable population decline. This same pattern was observed during the corresponding aerial surveys (Bridgland et al. 2007, Smith et al. 2015, Nova Scotia Lands and Forestry 2020a), suggesting that factors beyond habi- tat covariates were associated with the decline in the southern population; for exam- ple, the relative abundance of white-tailed deer (Odocoileus virginianus). When deer and moose are sympatric, their general abun- dance pattern is an inverse relationship in that as deer numbers increase, moose abun- dance declines (Snaith et al. 2002), a rela- tionship believed to explain, in part, moose decline elsewhere in Nova Scotia (Pulsifer and Nette 1995, Snaith et al. 2002). This effect is unlikely to be direct competition between the species, rather, the increased spread and abundance of Parelaphostrongylus tenuis, or meningeal worm associated with high deer populations (Anderson 1972, Lankester 2010). White-tailed deer are the principal host of meningeal worm and rarely display symptoms or negative effects, whereas this parasite can be mortal to moose (Lankester 2010). Unfortunately, the prevalence of meningeal worm is unknown on Cape Breton and warrants further study. To conclude, our aerial surveys provided a useful assessment of the annual and long- term winter abundance and habitat use of moose on Cape Breton. We found similarity between population estimates from histori- cal aerial surveys and habitat use models ALCES VOL. 57, 2021 AIRST AND POWER. – WINTER HABITAT USE IN NOVA SCOTIA 109 developed from aerial survey data. Our approach is applicable in many jurisdictions where long-term aerial survey data are avail- able, but more specific GPS-telemetry data from marked individuals are lacking. We encourage further development of similar approaches and use of long-term data sets from various sources to assess wildlife pop- ulations and improve understanding of how change in environmental factors affects moose and other wildlife over time. ACKNOWLEDGEMENTS We would like to acknowledge that this work took place in Unama’ki, the ancestral and unceded territory of the Cape Breton Mi’kmaq. Data collection was a joint effort of many groups including the Nova Scotia Department of Lands and Forestry, Parks Canada, the Unama’ki Institute of Natural Resources, and the Confederacy of Mainland Mi’kmaq. We would like to thank all participants for their help collecting this data. We would especially like to thank J. Bridgeland and M. Lemieux of Parks Canada for their advice early on when developing methods for interpreting the survey data. REFERENCES Anderson, C. R., and F. G. Lindzey. 1996. Moose sightability model developed from helicopter surveys. Wildlife Society Bulletin 24: 247–259. _____, R. C. 1972. The ecological relation- ships of meningeal worm and native cervids in North America. Journal of Wildlife Diseases 8: 304–310. doi: 10. 7589/0090-3558-8.4.304 Andreozzi, H. A., P. J. Pekins, and L. E. Kantar. 2016. Using aerial survey observations to identify winter habitat use of moose in northern Maine. Alces 52: 41–53. Beazley, K. F., T. V. Snaith, F. Mackinnon, and C. David. 2004. Road density and the potential impact on wildlife species such as American Moose in mainland Nova Scotia. Proceedings of the Nova Scotian Institute of Science 42: 339–357. doi: 10.15273/pnsis.v42i2.3610 Bjørneraas, K., I. Herfindal, E. J. Solberg, B. E. Sǣther, B. Van Moorter, and C. M. Rolandsen. 2012. Habitat quality influences population distribution, indi- vidual space use and functional response in habitat selection by a large herbivore. Oecologia 168: 231–243. doi: 10.1007/ s00442-011-2072-3 Boyce, M. S., P. W. Baxter, and H. P. Possingham. 2012. Managing moose harvests by the seat of your pants. Theoretical Population Biology 82: 340– 347. doi: 10.1016/j.tpb.2012.03.002 Bridgland, J., T. Nette, C. Dennis, and D. Quann. 2007. Moose on Cape Breton Island, Nova Scotia: 20th century demo- graphics and emerging issues in the 21st century. Alces 43: 111–121. Cole, E. K., M. D. Pope, and R. G. Anthony. 1997. Effects of road management on movement and survival of Roosevelt elk. Journal of Wildlife Management 61: 1115–1126. doi: 10.2307/3802109 Crichton, V., T. Barker, and D. Schindler. 2004. Response of a wintering moose population to access management and no hunting – a Manitoba experiment. Alces 40: 87–94. Davis, D., and S. Browne. 1996. The natural history of Nova Scotia. Nova Scotia Museum, Halifax, Nova Scotia, Canada. dussault, c., j. p. ouellet, c. laurian, r. courtois, m. poulin, and l. breton. 2007. Moose movement rates along highways and crossing probability mod- els. Journal of Wildlife Management 71: 2338–2345. doi: 10.2193/2006-499 Eldegard, K., J. T. Lyngved, and O. Hjeljord. 2012. Coping in a human-dominated landscape: trade-off between foraging and keeping away from roads by moose (Alces WINTER HABITAT USE IN NOVA SCOTIA – AIRST AND POWER ALCES VOL. 57, 2021 110 alces). European Journal of Wildlife Research 58: 969–979. doi: 10.1007/ s10344-012-0640-4 Emlen, J. M. 1966. The role of time and energy in food preference. American Naturalist 100: 611–617. doi: 10.1086/282455 Environment Canada. 2020. Grand Etang Nova Scotia historic weather data. Government of Canada. (accessed December 2020). Environmental Systems Research Institute. 2018. ArcGIS. Version 10.5.1. Redlands, California, USA. Forman, R. T., and L. E. Alexander. 1998. Roads and their major ecological effects. Annual Review of Ecology, Evolution, and Systematics 29: 207–231. doi: 10.1146/annurev.ecolsys.29.1.207 Fryxell, J. M., A. R. Sinclair, and G. Caughley. 2014. Wildlife ecology, con- servation, and management. John Wiley and Sons, New York, New York, USA. Gasaway, W. C., S. D. Dubois, D. J. Reed, and S. J. Harbo. 1986. Estimating moose pop- ulation parameters from aerial surveys. Institute of Arctic Biology, University of Alaska, Fairbanks, Alaska, USA. Government of Nova Scotia. 1998. Endangered Species Act. Ratified 3 December 1998. (accessed December 2020). Hebblewhite, M., and E. Merrill. 2008. Modelling wildlife–human relationships for social species with mixed-effects resource selection models. Journal of Applied Ecology 45: 834–844. doi: 10.1111/j.1365-2664.2008.01466.x Kantar, L. E., and R. E. Cumberland. 2013. Using a double-count aerial survey to estimate moose abundance in Maine. Alces 49: 29–37. kneib, t., t. hothorn, and g. tutz. 2009. Variable selection and model choice in geoaddative regression models. Biometrics 65: 626–634. doi: 10.1111/ j.1541-0420. 2008.01112.x krausman, p. r. 1999. Some basic princi- ples of habitat use. Idaho Forest, Wildlife and Range Experimental Station Bulletin 70: 85–90. Lankester, M. W. 2010. Understanding the impact of meningeal worm, Parelaphostrongylus tenuis, on moose populations. Alces 46: 53–70. Maclean, D. A., and D. P. Ostaff. 1989. Patterns of balsam fir mortality caused by an uncontrolled spruce budworm outbreak. Canadian Journal of Forest Research 19: 1087–1095. doi: 10.1139/x89-165 Manly, B. F. J., L. L. Mcdonald, D. L. Thomas, T. L. Mcdonald, and W. P. Erickson. 2002. Resource selection by animals: statistical analysis and design for field studies. Second edition. Kluwer Academic Publishers, Boston, Massachusetts, USA. McIntosh, T. E., R. C. Rosatte, J. Hamr, and D. L. Murray. 2007. Development of a sightability model for low-density elk populations in Ontario, Canada. Journal of Wildlife Management 73: 580–585. doi: 10.2193/2007-550 Mclaren, B. E., B. A. Roberts, N. Djan-Chékar, and K. P. Lewis. 2004. Effects of overabundant moose on the Newfoundland landscape. Alces 40: 45–59. Nova Scotia Lands and Forestry. 1999. Forest inventory – second round. Forestry Division, Truro, Nova Scotia, Canada. ______. 2009. Forest inventory – third round. Forestry Division, Truro, Nova Scotia, Canada. ______. 2020a. Winter 2020 Cape Breton Moose survey. Wildlife Division, Kentville, Nova Scotia, Canada. ______. 2020b. Forest inventory – third round aged forward. Forestry Division, Truro, Nova Scotia, Canada. pagano, a. m., and t. w. arnold. 2009. Detection probabilities for ground-based breeding waterfowl surveys. Journal of Wildlife Management 73: 392–339. doi: 10.2193/2007-411 https://climate.weather.gc.ca https://climate.weather.gc.ca https://nslegislature.ca/legc/bills/57th_1st/3rd_read/b065.htm https://nslegislature.ca/legc/bills/57th_1st/3rd_read/b065.htm https://nslegislature.ca/legc/bills/57th_1st/3rd_read/b065.htm ALCES VOL. 57, 2021 AIRST AND POWER. – WINTER HABITAT USE IN NOVA SCOTIA 111 Pulsifer, M. D., and T. L. Nette. 1995. History status and present distribution of moose in Nova Scotia. Alces 31: 209–219. Quayle, J. F., A. G. Machutchon, and D. N. Jury. 2001. Modeling moose sightabil- ity in south-central British Columbia. Alces 37: 43–55. R Core Team. 2019. R: a language and envi- ronment for statistical computing. Version 3.6.1. R Foundation for Statistical Computing, Vienna, Austria. Rempel, R. S., D. Kaukinen, and A. P. Carr. 2016. Patch analyst and patch grid 5.2. Center for Northern Forest Ecosystem Research, Ontario Ministry of Natural Resources, Thunder Bay, Ontario, Canada. Schwartz, C. C., and A. W. Franzmann. 1998. Ecology and Management of the North American Moose. Smithsonian Institution Press, Washington, DC, USA. Smith, C., K. F. Beazley, P. Duinker, and K. A. Harper. 2010. The impact of moose (Alces alces andersoni) on forest regen- eration following a severe spruce bud- worm outbreak in the Cape Breton Highlands, Nova Scotia, Canada. Alces 46: 135–150. Smith, R. M. Smith, P. Paul, and C. Bellemore. 2015. Hyperabundant moose management plan for North Mountain, Cape Breton Highlands National Park. Parks Canada. snaith, T. V., K. F. beazley, F. Mackinnon, and P. Duinker. 2002. Preliminary habitat suitability analysis for moose in mainland Nova Scotia, Canada. Alces 38: 73–88. Van Beest, F. M., B. Van Moorter, and J. M. Milner. 2012. Temperature-mediated habitat use and selection by a heat- sensitive northern ungulate. Animal Behaviour 84: 723–735. doi: 10.1016/j. anbehav.2012.06.032 Wam, H. K., and O. Hjeljord. 2010. Moose summer and winter diets along a large scale gradient of forage availability in southern Norway. European Journal of Wildlife Research 56: 745–755. doi: 10.1007/s10344-010-0370-4 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. Williams, B. K. 2011. Adaptive manage- ment of natural resources – framework and issues. Journal of Environmental Management 92: 1346–1353. doi: 10. 1016/j.jenvman.2010.10.041 Wood, S. 2017. generalized additive mod- els: an introduction with R. Second Edition. Chapman and Hall/CRC Press, London, United Kingdom. _____, and scheipl, f. 2020. Generalized additive mixed models using “mgcv” and “lme4.” Version 0.2–6. (accessed December 2020). https://cran.r-project.org/web/packages/gamm4/gamm4.pdf https://cran.r-project.org/web/packages/gamm4/gamm4.pdf https://cran.r-project.org/web/packages/gamm4/gamm4.pdf