Influence of body mass and environmental conditions on winter mortality risk of a northern ungulate: Evidence for a late‐winter survival bottleneck

Abstract A relationship between winter weather and survival of northern ungulates has long been established, yet the possible roles of biological (e.g., nutritional status) and environmental (e.g., weather) conditions make it important to determine which potential limiting factors are most influential. Our objective was to examine the potential effects of individual (body mass and age) and extrinsic (winter severity and snowmelt conditions) factors on the magnitude and timing of mortality for adult (>2.5 years old) female white‐tailed deer (Odocoileus virginianus [Zimmerman, 1780]) during February–May in the Upper Peninsula of Michigan, USA. One hundred and fifty deer were captured and monitored during 2009–2015 in two areas with varying snowfall. February–May survival ranged from 0.24 to 0.89 (mean = 0.69) across years. Mortality risk increased 1.9% with each unit increase in cumulative winter severity index, decreased 8.2% with each cumulative snow‐free day, and decreased 4.3% with each kg increase in body mass. Age and weekly snow depth did not influence weekly deer survival. Predation, primarily from coyote (Canis latrans [Say, 1823]) and wolves (Canis lupus [L., 1758]), accounted for 78% of known‐cause mortalities. Our results suggest that cumulative winter severity, and possibly to a lesser degree deer condition entering winter, impacted deer winter survival. However, the timing of spring snowmelt appeared to be the most influential factor determining late‐winter mortality of deer in our study. This supports the hypothesis that nutrition and energetic demands from weather conditions are both important to northern ungulate winter ecology. Under this model, a delay of several weeks in the timing of spring snowmelt could exert a large influence on deer survival, resulting in a survival bottleneck.


| INTRODUC TI ON
Identifying limiting factors for animals with seasonally dynamic life histories sometimes requires understanding intra-annual periods of resource scarcity (e.g., Ashmole, 1963). For ungulates living in northern environments, winter is generally a period of negative energy budget when forage provides some energetic intake but most individuals rely heavily on fat stores accumulated during the previous summer and fall to survive until spring green-up (hereafter the nutritional integration model; Mautz, 1978;Parker, Barboza, & Gillingham, 2009). Following this model, an annual survival bottleneck around the time of snowmelt could occur if the intensity and duration of winter are sufficient to exceed the energetic reserves of a substantial portion of the population (Parker et al., 2009).
An important prediction of the nutritional integration model is that the magnitude of late-winter survival bottlenecks is influenced by multiple mechanisms: Winter severity (e.g., depth of snow and temperature) determines rate of energetic expenditure, duration of snow cover determines how long a negative energy budget persists, and body fat reserves carried into the winter from previous foraging seasons determine how much energy is available to lose before succumbing to mortality from starvation or other causes affecting weakened animals (Parker et al., 2009). The importance of late-winter survival for population dynamics of northern ungulates has been recognized (e.g., Clutton-Brock, Price, Albon, & Jewell, 1991;Metz, Smith, Vucetich, Stahler, & Peterson, 2012), yet determining which individual or environmental factors limit wild ungulate populations during late winter remains difficult due to the possible interacting roles of biological (e.g., nutritional status) and environmental (e.g., weather) conditions (Wang et al., 2009).
Determining which winter weather components relate most strongly to deer survival is an important topic for deer managers.
The relationship between winter weather and survival of northern deer has led many natural resource agencies to adopt annual winter severity indexes (WSI) to predict cervid population trends (Verme, 1968;Leckenby & Adams, 1986;Chadwick, 2002;Delgiudice et al., 2002;Duquette et al., 2014). Generally, snow depth and temperature have been considered important predictors of deer mortality, with wind sometimes playing an important role in more open habitats. In addition, spring snowmelt timing may be important when considering winter severity for deer. Ignoring the middle period of winter and considering only the early and late months of winter may result in a better index of weather effects on deer (Verme, 1977), and spring snow depths can influence aspects of northern white-tailed deer ecology including migration behavior (Nelson, 1995), habitat selection (Beier & McCullough, 1990), and natal mortality (Verme, 1977).
The Upper Peninsula of Michigan, USA, has a geographic gradient of snow conditions that is highly variable among years due to the climatic influence of the Great Lakes. Historically, deer population growth in the Upper Peninsula has been linked to variation in winter weather temporally and geographically (Doepker, Beyer, & Donovan, 1995;Leopold, Sowls, & Spencer, 1947). The Upper Peninsula deer population declined due to consecutive severe winters in 1995-1996 and 1996-1997 but did not fully recover over the next 15 years while the recolonizing gray wolf (Canis lupus [L. 1758]) population increased during this same period (Michigan Department of Natural Resources [MDNR], 2010[MDNR], , 2015. The adverse nutritional effects of winter in Michigan can result in substantial deer kill even in the absence of predators (Case & McCullough, 1987), but certain winter weather conditions may facilitate wolf predation of ungulates through either limiting prey mobility (e.g., deep or crusted snow) or by causing nutritional degradation that weakens prey (Mech & Peterson, 2003;Mech, Smith, Murphy, & MacNulty, 2001;Telfer & Kelsall, 1984;Vucetich et al., 2012). These events resulted in uncertainty of how deer in the Upper Peninsula are influenced by winter conditions in the presence of predators, and whether winter kill limits deer population growth.
Our goal was to test whether patterns of deer survival within late winter follow predictions from the nutritional integration model and determine which mechanisms most strongly influence survival. Causespecific mortality of adult female white-tailed deer was investigated in relation to deer age, body mass, age, snow depth, cumulative effects of winter weather (described in Methods section), and cumulative effects of snow-free days in two areas with differing amounts of snowfall. We based our predictions on the hypothesis that deer generally maintain a negative energy balance during winter at northern latitudes, that weather conditions shape the rate of this nutritional decline, and that survival is dependent on conserving energetic stores until spring snowmelt. We predicted that deer mortality risk would increase with greater snow depth, fewer snow-free days during February-May, older and younger (nonprime) age classes, and decreasing body mass. was 65 km northwest of the low-snowfall study area. Mean annual precipitation was 52 cm of rain and 179 cm of snow (1951-1980averages, Michigan Climatology Office, 2013b. Mean January and July temperatures were -13°C and 18°C, respectively (PRISM Climate Group, 2016). Land cover was predominantly deciduous forest (38%), woody wetland (29%), mixed forest (13%), and evergreen forest (6%) (Fry et al., 2011).

| Deer capture and handling
Adult female white-tailed deer were captured during February-April, 2009-2011, in the low-snowfall study area and February-March, 2013-2015 in the mid-snowfall study area. We captured deer primarily using Clover traps (Clover, 1956) baited with shelled corn, alfalfa, and/or molasses, and occasionally used cannon nets. Deer were restrained, blindfolded, and immobilized with an intramuscular injection of ketamine hydrochloride (Putney, Inc., Portland, ME, USA) and xylazine hydrochloride (Lloyd Laboratories, Shenandoah, IA, USA) mixed at a 4:1 ratio and concentration of 100 mg/ml (Duquette, Belant, Beyer, & Svoboda, 2013). For each deer, body mass was recorded and age estimated by extracting a lower incisiform canine to age deer using counts of cementum annuli (Gilbert, 1966; see Nelson[, 2002]  Deer were monitored on a weekly basis using aerial-or groundbased telemetry. When a mortality signal was detected, the date and cause of mortality was determined based on deer remains and sign found at the mortality site. For predation events, evidence at the site (e.g., tracks, scat, canine puncture wounds, and site disturbance) was compared to published reports of predator-specific kills to estimate predator species (Mech, Frenzel, & Karns, 1971;Nelson & Mech, 1986). Mortalities were classified as unidentified predations if the mortality site showed evidence of predation (e.g., blood in surrounding snow, hemorrhaging on hide or tissue), but evidence was insufficient to assign a predator species or evidence of multiple predator species was present. Malnutrition status of mortalities was assessed using rump fat and bone marrow condition (Mech, 2008) or by submitting carcasses for laboratory necropsy by a wildlife pathologist. In 68% of mortalities, investigations occurred <5.3 days after the time of mortality and date of mortality was determined to the nearest half hour using the precise event transmitter code of vaginal implant transmitters. For the remaining 32% of mortality events in which >5.3 days had passed, date of mortality was estimated using a combination of carcass decomposition, snow cover conditions, and telemetry records.  (Delgiudice et al., 2002). From this, a cumulative winter severity index was calculated by summing daily values for each winter beginning 1 November.

| Weather data and deer density estimates
Snow-free days were defined as days from 1 February to 31 May when mean snow depth was <7 cm, a depth that has been linked to behavioral and forage transitions for spring deer (Beier & McCullough, 1990). We predicted that deer mortality risk would respond gradually to snow-free conditions during spring because mass gains for white-tailed deer during spring are gradual (Delgiudice et al., 1992). Therefore, the number of snow-free days was summed daily into cumulative snow-free days from 1 February within each winter, which were then averaged within each weekly survival interval as a time-varying covariate to reflect a possible cumulative effect in the relationship between deer mortality risk and spring snowmelt.

| Survival analysis
Factors were assessed for influence on adult female deer weekly survival from 1 February to 31 May using Cox proportional hazards mixed-effects models in the package coxme (Therneau, 2016) for program R (R Core Team, 2016). Because deer captured using Clover traps and rocket nets can experience capture myopathy-related mortality (Beringer, Hansen, Wilding, Fisher, & Sheriff, 1996), deer were not included in survival models until 2 weeks postcapture. As yearling deer captured in this study had different patterns in body mass and pregnancy rates than older deer (Duquette, Belant, Beyer, & Jr, & Svoboda, N. J., 2012), survival analysis was limited to deer >2.5 years old.
Biological covariates of deer mortality risk included age (years) and body mass. Adult female survival was expected to follow a parabolic trend peaking at 5-6 years of age before declining (Delgiudice et al., 2006) and so it was modeled age as a polynomial quadratic term. Because deer body mass declined with capture date (see Results section for time-mass regression output), slope estimates from linear regressions of adult female body mass by capture date for each year were used to standardize body mass to 1 February (Festa-Bianchet & Jorgenson, 1998).
Time-varying covariates (Therneau, Crowson, & Atkinson, 2016) were included for cumulative winter severity index, cumulative snowmelt days, and snow depth, estimated for each week within years by averaging daily values over each weekly survival interval.
A staggered entry design was used to account for varying capture dates of deer (Pollock, Winterstein, Bunck, & Curtis, 1989).
Low-snowfall or mid-snowfall study area were included as a random effect (i.e., frailty term) in all models to account for variation in predator populations, land cover, and other factors that may influence deer mortality risk among study areas (Pankratz, de Andrade, & Therneau, 2005). Multicollinearity among covariates was tested for using Spearman's rank correlation tests and considered any covariates with |r| < 0.7 suitable for inclusion in the same model (Dormann et al., 2013). Although Cox proportional hazards models have fewer assumptions than parametric survival models, an important assumption of the Cox model is that the baseline hazard ratio for each covariate remains constant over time. Violations of this assumption were assessed by testing for a significant (p < 0.05) interaction between weekly time period and each predictor covariate (Bellera et al., 2010;Fox & Weisberg, 2011). The final candidate model set included 24 candidate models with noncollinear covariate combinations of 6 factors: age, body mass, body mass:time interaction, weekly average snow depth, cumulative winter severity index, and cumulative snow-free days. Because our goal was to explore the relative predictive value of model covariates on weekly deer survival, all candidate models were evaluated using backward stepwise model ranking based on Akaike's information criterion adjusted for small samples, where candidate models <2 ∆AIC c of the top-ranked model were considered plausible (Burnham & Anderson, 2002;Symonds & Moussalli, 2011).

| RE SULTS
One hundred and fifty pregnant adult female deer (>2.5 years old) were captured representing 1,784 deer-weeks of monitoring during February-May. This included 147 unique individuals and 3 individuals that were recaptured and included in 2 years of monitoring.
Median date of capture was 10 February (interquartile range = 21 January-25 February). Estimated age of captured deer ranged from 2.5 to 16.5 years (median = 6.5, interquartile range = 3.5-8.5). After standardization to 1 February, annual mean body mass of captured adult female deer ranged from 56.0 to 65.0 kg, and mean body mass within the mid-snowfall study area (64.0 kg, SD = 5.8) was greater than the low-snowfall study area (57.9 kg, SD = 7.2; t(116) = 5.57, p ≤ 0.001; Table 1). Among all capture years pooled, regression coefficient for body mass loss through the capture period was −0.88 kg/ week (n = 150, r 2 = 0.13, p < 0.001). Pooled across all years, mean weekly mortality rate during February-May (2.1%) was 3.5 times greater than mean weekly mortality during June-January (0.6%; obtained from the same study animals monitored year-round every 2 weeks, but not included in mortality risk models).
Coyote (n = 12) and wolf (n = 11) were the most common predators of adult female deer, followed by unidentified predator (n = 6) and bobcat (n = 2). Of the 8 malnutrition mortalities, 6 occurred during the winter with greatest winter severity (2014; WSI = 167). No deer were censored due to radio-collared failure or other reasons during the study interval.  (Table 3). The best-supported model included body mass, cumulative winter severity index, and cumulative snow-free days, but excluding body mass from this model also resulted in a plausible model (∆AICc = 0.994; Table 3).

| D ISCUSS I ON
The influence of winter weather on white-tailed deer winter mortality risk which we observed is consistent with other studies of deer survival in northern climates (e.g., Delgiudice et al., 2002;Dumont et al., 2000;Nelson & Mech, 1986), and generally supports the hypothesis that winter weather conditions are the primary fac- (the two years with latest snowmelt) remained high for 1-2 weeks following snowmelt (Figure 3). This could indicate a lag effect of snow conditions on deer mortality risk which our model did not account for. Such a lag effect may be expected because following snowmelt, deer physical condition is likely at an annual nadir, and nutritional recovery from winter likely takes several weeks to begin (Delgiudice et al., 1992 Note: SFD represents weekly cumulative snow-free days, and WSI represents weekly cumulative winter severity index. All models included study area as a random effect. a Cumulative snow-free days from 1 November to 31 May. b Cumulative winter severity index from 1 November to 31 May (Delgiudice et al., 2002).  , & Solberg, 1999). There are several plausible explanations for why body mass would be positively correlated with winter deer survival in our study. First, in northern white-tailed deer there are pronounced seasonal changes in body mass following nutritional gains and losses (Delgiudice et al., 1992). Within individuals, changes in mass are a good index of condition because ungulate body mass is correlated with condition because increasing fat stores adds body mass to individuals (Stephenson, Hundertmark, Schwartz, & Ballenberghe, 1998), and ungulates mobilize both fat and lean body tissue according to their condition at the onset of winter to meet the energetic demands of survival and pregnancy (Monteith et al., 2013). However, among individuals there are differences in skeletal size and lean body mass, unrelated to body fat reserves, which we did not account for by measuring body mass alone. Certainly, this adds unaccounted for variation to the effect of body mass on mortality risk in our study. Second, larger homeothermic vertebrates with greater lean body mass can carry more energy reserves relative to metabolic weight and endure longer periods of fasting (Boyce, 1979;Lindstedt & Boyce, 1985).

Langvatn
Hence, it is possible that a larger skeletal frame provides a survival advantage for northern ungulates. By only measuring deer body mass and not actual fat reserves, we cannot reach conclusions about which of these processes may have been important in our study. However, the considerably larger body mass of deer in the mid-snowfall study area does support the result of our mortality risk models, indicating a selective advantage for larger deer during longer winters.
Our results suggest that deer mortality in the Upper Peninsula may be similarly or more sensitive to the timing of spring snowmelt than temperatures or snow depth throughout early-mid-winter.
Previous research suggests that northern deer may lose body mass during winter at a similar rate during mild or severe winters (Giroux, Dussault, Tremblay, & Cote, 2016), which along with our results could indicate that the difference between a mild and severe winter from a deer energetic perspective is more determined by seasonal snow cover duration than by short-term (within-week) snow depth or temperature conditions. The number of snow-free days in February-May is likely correlated with WSI on an annual scale, but two winters can have similar WSIs with notable differences in spring snowmelt. For example, winter 2012-2013 (winter severity index = 108; 33 snow-free days) had a lesser total winter severity index but 15 more days of snow cover during February-May than winter 2014-2015 (winter severity index = 145; 48 snow-free days).
If the immediate physical effects of deep snow influenced deer mortality risk by impeding the ability to escape predators, snow depth should be positively correlated with deer mortality risk. In our study, weekly snow depth did not explain variation in weekly deer mortality risk, but other studies have found contrasting results.
In Minnesota, wolf predation rates on yearling and adult whitetailed deer were the greatest during months with the deepest snow (Nelson & Mech, 1986), and daily wolf kill rates of deer in a high snowfall area of Michigan were highly correlated with snow depth in a previous study . These studies suggest that the immediate effects of deep snow can increase predation on deer in some circumstances, but our results suggest that the gradual nutritional decline throughout winter was the primary mechanism influencing mortality risk during our study. A similar nutritional influ-  (Delgiudice et al., 1992). Finally, crusted snow conditions during spring facilitate deer predation by wolves and coyotes because of heavier foot-loading in deer (Telfer & Kelsall, 1984;Vucetich et al., 2012 forages. However, even with a positive energy budget, spring body mass gain by deer is a gradual process of weeks or months (Delgiudice et al., 1992 Overall, our study supports the nutritional integration model applied to northern white-tailed deer, which adds further support to the already substantial evidence that weather conditions, through nutritional effects, are the primary mediator of northern ungulate population dynamics (Parker et al., 2009;Post & Stenseth, 1999 Mautz (1978), and more broadly supports the nutritional integration model of northern ungulate ecology suggested by Parker et al. (2009). In future studies of northern ungulates with a negative energy budget during periods of snow cover, considering snow-free days during late winter or a similar measure of spring snowmelt timing may improve model accuracy.

AUTH O R CO NTR I B UTI O N S
TK, JB, DB, and BS conceived the ideas and designed methodology.
TK and JD collected the data. TK and JB analyzed the data. TK and JB led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.  (1996). Factors affecting capture myopathy in white-tailed deer.