Deep snow and human influence affect survival of an ungulate at the northern part of its range

Introduction

Understanding the impact of climate and anthropogenic factors on wildlife populations is important for effective management. According to theory, harsh climate (e.g. deep snow and low temperatures) is considered a limiting factor for populations at high latitude range margins (Louthan et al. 2015; Sirén and Morelli 2020). However, biotic factors such as food or habitat availability can sometimes buffer the negative effects of harsh abiotic stress and allow populations to persist (or even expand ranges) at high latitudes (Sirén and Morelli 2020). Consequently, it is imperative to understand the relative contribution of abiotic and biotic factors influencing populations along range margins. This is especially important for game species in high-latitude regions where pre-harvest quotas need to be established on the basis of the severity of winter conditions during the previous year (Norton et al. 2021). As such, developing models that best predict sensitive demographic parameters (e.g. survival) can improve wildlife management and allow for a better understanding of the impacts of climate and habitat change on game populations (Weiskopf et al. 2019).

White-tailed deer (Odocoileus virginianus) are subject to severe winter conditions in the northern parts of their range (Fieberg et al. 2008; Kautz et al. 2020; Norton et al. 2021). Although deer are adapted to tolerate some harsh winter conditions (e.g. reduced movement during snowy periods; Moen 1976), severe winters can result in significant decreases in winter survival for northern populations (Fuller 1990; Whitlaw et al. 1998; Dumont et al. 2000; Patterson and Power 2002; Norton et al. 2021; Lasharr et al. 2023). Deep snow makes movement more energetically costly, and low temperatures make it more difficult for deer to maintain body temperature (Moen 1976). There may also be some threshold effects with regards to snow depth and temperature and a mediating effect from anthropogenic factors. For example, Norton et al. (2021) found that cumulative days above 30.5 cm snow depth and days above 0°C influenced survival in forested areas and cumulative days above 0°C influenced survival in farmlands, where there was more access to food. To account for winter survival, wildlife managers must have a solid understanding of how different variables affect winter survival of deer, including abiotic factors (e.g. snow depth, temperature) and biotic factors such as habitat availability and artificial feeding, or food provided by humans to improve survival or viewing opportunities (Norton et al. 2021; Lasharr et al. 2023).

Northern deer populations often migrate to deer wintering areas, or yards, to adjust to snow conditions and food availability (Nelson 1995; Van Deelen et al. 1998; Ditchkoff and Servello 2002; Sabine et al. 2002). The habitat within these natural deer wintering areas is often characterized by mature softwoods with high levels of canopy closure and interspersed or adjacent hardwood browse (Morrison et al. 2003; DelGiudice et al. 2013). These closed-canopy softwood areas intercept snow, resulting in lower snowpack at ground level, which facilitates establishment of trail networks, and serves as thermal and wind cover (Morrison et al. 2003; Lefort et al. 2007). In Maine, and in other northern regions, deer may have historically used natural winter cover (DelGiudice et al. 2013; Simons-Legaard et al. 2018) but now often frequent towns and developed areas where they have access to compacted snow, refuge from predation, and artificial feed (Dawe et al. 2014; Fortin et al. 2015). In particular, artificial feeding is widespread in some northern regions, but the extent to which it provides a survival advantage is context dependent (Tarr and Pekins 2002; Milner et al. 2014; Fortin et al. 2015).

Some studies have used a winter severity index (WSI) to describe multiple winter weather variables that affect winter survival (DelGiudice et al. 2002; Kautz et al. 2020). For example, DelGiudice et al. (2002) found that a WSI, which accounted for the number of days with snow depth of ≥38 cm and ambient temperature lower than −17.7°C during 1 November−31 May, best predicted winter survival, but snow depth alone was nearly as predictive. Kautz et al. (2020), by using the same WSI as DelGiudice et al. (2002), found the best fit model of winter survival to include WSI, cumulative snow-free days, and body mass at capture. The number of snow free days was the most impactful variable, followed by WSI and body mass. Contrastingly, Garroway and Broders (2005) found that snow depth rather than rainfall and temperature best predicted winter survival. These WSIs are often used by wildlife management agencies to estimate and account for winter survival rates and develop pre-harvest quotas (Norton et al. 2021).

Historically, Maine Department of Inland Fisheries and Wildlife (MDIFW) estimated winter mortality rates by using a WSI model developed in the 1990s from dead deer surveys in deer wintering areas and deer density estimates from pellet group counts (Maine Department of Inland Fisheries and Wildlife (MDIFW) 2007). This modeling work (hereafter ‘historical WSI model’) related snow and temperature conditions with observations of dead deer in deer wintering areas from mid-December through April. Specifically, the WSI was calculated using the following formula:

where SD = snow depth in open canopy forests, SK = deer sinking depths in open canopy forests, Tm = long-term mean temperature, and Ts = current season mean temperature are multiplied by 33.3. MDIFW would then predict winter mortality rates by exponentiating 2.29 by 0.222 × WSI for each region and winter (Maine Department of Inland Fisheries and Wildlife (MDIFW) 2007). In summary, the WSI formula combined with the predictions of winter mortality rates constituted the historical WSI model. The estimates of winter mortality were then used to adjust doe removal rates when developing recommendations for antlerless harvest. Because the estimates were not specific to adult females, it was not ideal to use them to adjust doe removal rates and estimates specific to adult females were needed. Further, the estimates relied on site visits, which were labor intensive and costly. For example, using the historical WSI model developed in the 1990s, Maine collected weekly snow depth, deer sinking depth, and temperature data at 26 sites across the state over a 20-week period to produce annual winter mortality estimates, and because some of these data required weekly site visits, they were very costly and time-consuming to maintain.

Our primary objectives were to determine what factors most influenced winter survival in adult female white-tailed deer in Maine and to develop a predictive equation allowing for the estimation of winter survival. We also examined differences in winter survival rates between deer living in traditional deer wintering areas and those exposed to anthropogenic influences, primarily in the form of artificial feeding. Our primary hypotheses were that winter temperature and snowpack variables would influence winter survival (Table 1). We predicted that colder and snowier winters would be negatively correlated with survival and that deer exposed to anthropogenic influences would have higher survival rates (Table 1). We also evaluated the WSI that Maine has used for predicting winter survival as a predictor variable on deer survival and compared it with other winter variables. Last, we compared predictions of overwinter deer survival across the entire state of Maine for a seven-winter period (2015–2021) between the top model and the historical WSI model that MDIFW has used to predict winter mortality rates since the 1990s.

Materials and methods

Study area

We captured and radio-collared deer in four Wildlife Management Districts (WMDs) in central and northern Maine (WMDs 1, 5, 6, 17; Fig. 1). Our WMD 1 study site extended from the town of Allagash, ME (47°05′N, 69°02′W), south along the Allagash River to the Ben Glazier Brook deer wintering area in T14 R12 WELS, Maine (46°55′N, 69°14′W). Landcover in WMD 1 was primarily Laurentian–Acadian Northern Hardwood Forest and Acadian Low Elevation Spruce-Fir-Hardwood Forest (Ferree and Anderson 2013). Deer in WMD 1 were collared in the Ben Glazier Brook deer wintering area as well as around the town of Allagash. Deer in the town of Allagash had access to artificial winter feed, primarily in the form of lucerne hay bales placed by a regional conservation group. Winter snow depth in open cover averaged 53.6 cm, and winter temperature averaged −7.9°C. Our WMD 5 study site was centered around the Scraggly Lake Maine Public Reserved Land in T7 R8 WELS, Maine (46°14′N, 68°45′W). Deer in WMD 5 were collared around Scraggly, Millimagassett, and Millinocket lakes. Landcover was primarily Laurentian–Acadian Northern Hardwood Forest and Acadian Low Elevation Spruce–Fir–Hardwood Forest (Ferree and Anderson 2013). These deer did not have access to artificial feed in the winter. Winter snow depth in open cover averaged 50.0 cm, and winter temperature averaged −6.1°C. Our WMD 6 study site was geographically broad, ranging from Caribou, ME (46°51′N, 68°00′W), south to Houlton, ME (46°07′N, 67°50′W). Deer were collared widespread throughout WMD 6, and landcover in the study site was primarily Laurentian–Acadian Northern Hardwood Forest, Acadian Low Elevation Spruce-Fir-Hardwood Forest, and mixed agriculture (Ferree and Anderson 2013). Winter snow depth in open cover averaged 50.0 cm, and winter temperature averaged −6.1°C. Our WMD 17 study site was geographically broad ranging from Parkman, ME (45°08′N, 69°25′W), south to Canaan, ME (44°45′N, 69°33′W). Deer were collared throughout WMD 17, and landcover was primarily Laurentian-Acadian Pine-Hemlock-Hardwood Forest, Laurentian-Acadian Northern Hardwood Forest, Acadian Low Elevation Spruce-Fir-Hardwood Forest, and mixed agriculture (Ferree and Anderson 2013). Winter snow depth in open cover averaged 24.1 cm, and winter temperature averaged −3.6°C. The WMD 6 and WMD 17 study sites were in more developed parts of the state with higher road densities, abundant edge habitat, and widespread artificial feeding of deer in winter. There was no allowable antlerless harvest in WMDs 1 and 5 during the study. Annual antlerless harvest was 0.01–0.03 adult females/km² in WMD 6 and 0.14–0.43 adult females/km² in WMD 17 during the study. Predators on adult deer in Maine are primarily coyotes and bobcat or lynx, depending on location (N. Bieber, unpubl. data, Major and Sherburne 1987; Fuller 2004).

Fig. 1.

Study area map of the four wildlife management districts (1, 5, 6, 17) where we evaluated overwinter survival of adult female white-tailed deer (Odocoileus virginianus) from winter 2015–2016 to winter 2021–2022. The points are locations of the weather stations where we collected environmental data to evaluate relationships between winter variables and survival.

Results

Survival analyses

The top survival model, as determined by the lowest LOO-IC score, included closed-canopy snow depth and ANTHRO status (cdep + ANTHRO); all other models, which included weather variables and ANTHRO status, had low predictive power and credible intervals of variables overlapped zero (Table 3). The GOF test indicated that the top model fit the data well, with 100% of the Pareto-k estimates being <0.7. Further, Rhat values were <1.1 and the MCMC chains mixed well, indicating convergence. Snow depth under closed-canopy forests had a negative effect on survival (Table 3, Fig. 2a). Also, deer living close to artificial feeding and developed areas (i.e. ANTHRO) had higher survival rates, yet there was some uncertainty in this relationship as credible intervals slightly overlapped zero (Table 3, Fig. 2a). Survival varied by WMD, with the lowest and highest overwinter survival rates in the northernmost (WMD 1: ϕ = 0.66, CI = 0.49–0.81) and southernmost (WMD 17: ϕ = 0.85, CI = 0.73–0.93) regions respectively (Fig. 2b, Table S2). Overwinter survival varied annually, with the lowest estimate during the 2016–2017 winter (ϕ = 0.72, CI = 0.54–0.85) and the highest estimate during the 2020–2021 winter (ϕ = 0.82, CI = 0.67–0.93) (Fig. 2c, Table S2).

Table 3.Results of model evaluation using leave-one-out information criterion (LOO-IC) and the mean, standard deviation (s.d.), and lower and upper 95% credible intervals (CI) of the model predictor variables.

Model	LOO-IC	Parameter	Mean	s.d.	Lower CI	Upper CI
cdep + ANTHRO	553.08	cdep	−0.41	0.15	−0.72	−0.11
		ANTHRO	0.52	0.29	−0.04	1.08
osnk + ANTHRO	554.60	osnk	0.09	0.14	−0.16	0.37
		ANTHRO	0.51	0.29	−0.07	1.08
csnk + ANTHRO	555.14	csnk	0.04	0.13	−0.21	0.30
		ANTHRO	0.51	0.29	−0.05	1.10
odep + ANTHRO	555.30	odep	0.03	0.14	−0.24	0.32
		ANTHRO	0.49	0.30	−0.09	1.07
cm25.4 + ANTHRO	555.50	cm25.4	0.05	0.13	−0.21	0.31
		ANTHRO	0.50	0.29	−0.07	1.08
d0C + ANTHRO	555.67	d0C	−0.05	0.13	−0.30	0.19
		ANTHRO	0.51	0.29	−0.06	1.07
atmp + ANTHRO	555.73	atmp	0.04	0.13	−0.21	0.29
		ANTHRO	0.51	0.29	−0.07	1.07
wsi + ANTHRO	555.79	wsi	−0.14	0.15	−0.43	0.14
		ANTHRO	0.51	0.29	−0.07	1.07
d45.7 + ANTHRO	555.91	d45.7	−0.08	0.15	−0.38	0.22
		ANTHRO	0.53	0.29	−0.04	1.10
dneg17.8C + ANTHRO	555.92	dneg17.8C	−0.19	0.15	−0.49	0.10
		ANTHRO	0.52	0.28	−0.05	1.08

Note: variables highlighted in bold had CIs that did not overlap zero.

Fig. 2.

Predictive plots from the top model (cdep + ANTHRO), indicating (a) the relationship between under-canopy snow depth and ANTHRO status on overwinter (mid-December–April) survival of adult female white-tailed deer (Odocoileus virginianus), (b) estimates of overwinter survival for each wildlife management district, and (c) estimates of overwinter survival for each winter of the study. Points and lines are posterior estimates (means), and error bars and colored bands are 95% credible intervals respectively.

Model comparison

The predictive equation from our top model (cdep + ANTHRO) that we compared with the historical WSI model described above was

ϕw,tlog˜it(α3.578+βwmd[w]+βwinter[t]+βcdep(-0.398)×Cover Depthw,t+βANTHRO(0.526)×ANTHROStatusw,t)9

where overwinter survival (ϕ) within a WMD (w) during a winter (t) is the inverse logit of the sum of the intercept (α) and slopes (β; here cdep and ANTHRO), given the unique cover depth and ANTHRO status values within a WMD (w) and winter (t). Overwinter survival, as a function of snow cover depth and ANTHRO status, was predicted at the population level by exponentiating the number of bi-weekly survey occasions in a winter (n = 9). Note, that the random effects (WMD and winter) were retained for making comparisons between approaches for the WMDs and winters that we sampled but were marginalized from the likelihood to make comparisons across the entire state (i.e. WMDs other than 1, 5, 6, 17) for each winter.

The historical WSI model predicted higher overwinter survival for the majority of WMD and winter combinations that coincided with our study (Fig. 3), than were predictions of overwinter survival from the top model using radio-collared deer (cdep + ANTHRO). We also compared predictions in other WMDs across the state and found a similar pattern; the historical WSI model predicted higher survival rates with less differences among annual mean survival rates than did the top model (Supplementary Figs S1–S6). One common trend was that the lowest survival rates were predicted for the winter of 2018–2019, yet there was high uncertainty associated with predictions from the top model (cdep + ANTHRO) (Fig. 3).

Fig. 3.

Predictions and 95% credible intervals (error bars) of overwinter (mid-December–April) survival from the top model (cdep + ANTHRO status), which used known fates of radio-collared, adult female white-tailed deer (Odocoileus virginianus), for a seven-winter period in four wildlife management districts (WMD) compared with predictions from the historical WSI model developed during the 1990s that used a winter severity index (WSI) to estimate overwinter survival. Note that the WSI model does not include estimates of uncertainty.

Discussion

Our study has provided a long-term evaluation of factors influencing adult female white-tailed deer survival at the northern end of their range. We showed that winter weather has a strong impact on survival of deer, yet the availability of anthropogenic resources (e.g. food availability) can ameliorate mortality rates, supporting theory on factors that influence populations along high latitude range margins (Sirén and Morelli 2020). These results are similar to other studies of northern ungulates (DelGiudice et al. 2002; Kautz et al. 2020; Norton et al. 2021) and indicate that it is critical to monitor relevant winter weather variables to adjust harvest quotas for future seasons and use radio-collared animals for developing such metrics.

Snow depth under closed canopy cover had a strong negative effect on deer survival compared with other snow and temperature variables. The negative effect of deep snow on deer is well-documented (Robinette et al. 1952; Fuller 1990; DelGiudice et al. 2002; Lasharr et al. 2023). However, we were surprised that other snow and temperature variables, especially those that represent thresholds (e.g. number of days with >45.2 cm of snow), were not impactful, especially because our study included several long and severe winters. Deep snow, under closed-canopy forests, restricts the ability of deer to access food (Dumont et al. 1998, 2005) and avoid predators (Witt et al. 2012; Moratz et al. 2018; Kautz et al. 2020). Indeed, predation was the primary cause of mortalities, especially by canids (Table 2). Further, the amount and quality of deeryards have declined in northern Maine since the 1970s as a result of the logging practices of the past half-century that were primarily in response to a large spruce-budworm (Choristoneura fumiferana) outbreak (Simons-Legaard et al. 2018). Consequently, exceptionally deep snow in closed-canopy forests, including deer wintering areas, can result in higher overall mortality and potentially explains this finding.

In contrast, anthropogenic influences provided a survival advantage for deer. Previous work has indicated that artificial feeding (in the form of deer winter feed stations) and byproducts from agriculture can have a positive impact on survival during harsh winters (Baker and Hobbs 1985; Norton et al. 2021) and our results align with theoretical expectations. For instance, positive biotic factors such as food and habitat availability can ameliorate harsh climate for populations along high latitude range margins and allow for population persistence (Sirén and Morelli 2020). However, there are numerous unintended consequences of feeding deer, including increased risk of disease transmission (Dunkley and Cattet 2003; Murray et al. 2016), habitat degradation around feeding sites (Inslerman et al. 2006; Milner et al. 2014), incidental deer–vehicle collisions (Dunkley and Cattet 2003), increased competition among deer (Dunkley and Cattet 2003; Milner et al. 2014), and sickness or death from rumenitis and/or rumen acidosis (Wobeser and Runge 1975) or enterotoxemia (Rideout 2003). There are preferred alternatives to artificial feeding, and timing timber harvests to benefit deer in nearby wintering areas (e.g. Maine Department of Inland Fisheries and Wildlife (MDIFW) 2012) is one such approach.

While we found a positive effect from anthropogenic influence on deer survival, it is difficult to determine whether artificial feeding was the primary contributor. Artificial feeding tends to occur near human developed areas, and survival may be affected by human development in other ways. For example, predator success and hunting methods may differ from more natural environments (Fleming and Bateman 2018), deer may have access to trails that have compacted snow (e.g. snowmobile trails; Richens and Lavigne 1978), and deer may encounter vehicles more frequently (Fudge et al. 2007). Indeed, we did record some road-kill mortalities in the developed areas, so the influence of humans is not always positive for deer and could have been the source of high variance associated with this variable (see Table 3). Clearly, more study is warranted to better understand which anthropogenic factors have a negative and positive impact on deer survival.

One of the motivations of this study was to understand the limitations of the historical WSI model as a monitoring tool. Prior work in our region has relied on a WSI to monitor deer survival; however, compared with the top model, the WSI performed poorly and explained little variation in deer survival. Additionally, our comparison of the predictive equation from the top model and the historical WSI model (that did not use radio-collared deer), indicated that the latter overpredicted survival for all WMDs and that there was little variation among winters. Our findings of WSIs performing poorly are in line with a recent study in Wisconsin, which used a large sample size of radio-collared deer (Norton et al. 2021), yet in contrast to previous work that found WSIs to have high predictive power (DelGiudice et al. 2002; Kautz et al. 2020). WSIs can be valuable because they capture the cumulative impacts of snow and temperature that deer experience over an entire winter (Dawe and Boutin 2012). Other variables that capture winter length (e.g. number of days above freezing), instead of the weather within a shorter time frame, can also have a strong impact on survival (Kautz et al. 2020; Norton et al. 2021). Although our deepest snowpack primarily occurred later in the winter, we did not find support for variables associated with winter length. It is important to note that our applications are specific to adult female deer, so additional research is needed to assess factors, including WSIs, that influence juvenile and adult male survival.

Our study has several management implications and caveats worth noting. First, estimating winter survival rates in areas with low deer densities and limited antlerless harvest (e.g. northern Maine) may be less important for controlling deer populations if the range of possible management actions is very limited, such as in areas with restricted or no antlerless harvest opportunities. However, monitoring survival in these areas may be beneficial because mild winters will likely increase population sizes and provide hunting opportunities and a greater range of possible management actions. Further, keeping densities somewhat lower in areas where deer overlap with moose (Alces alces) could be an important management strategy to reduce the transmission of diseases that negatively affect moose (e.g. Parelaphostrongylus tenuis). Our study also indicated that it will be important to continue monitoring snow depth in closed-canopy forests. Although MDIFW historically collected weekly snow depth data, the number and distribution of the weather stations were sparsely distributed (see Fig. 1). Future monitoring could benefit from increased sampling to better capture the variation among regions, particularly in regions of greater deer densities and a greater range of possible management actions but where deer are still subject to occasional severe winters; however, adding more snow stations in each WMD would be prohibitively costly because snow depth data are collected manually on a weekly basis (Maine Department of Inland Fisheries and Wildlife (MDIFW) 2007). One solution would be to use remote cameras to record snow depths (Sirén et al. 2018) because these can be deployed in a variety of conditions to evaluate forest-snow dynamics (Sirén et al. 2018; Breen et al. 2023). Further, recent developments in cellular and satellite cameras and automated snow depth extraction from snow stakes (Breen et al. 2024) could greatly increase the efficiency of monitoring snowpack. Additionally, both monitoring of snow and demographic parameters (abundance, survival, and recruitment) can be achieved using cameras (Sirén et al. 2024) and allow for efficient large-scale monitoring of climate and deer populations. We suggest exploring these options for developing cost-effective and robust approaches for monitoring deer populations in northern regions.

Supplementary material

Supplementary material is available online.