Phase‐dependent climate–predator interactions explain three decades of variation in neonatal caribou survival
Summary
-
Climate can have direct and indirect effects on population dynamics via changes in resource competition or predation risk, but this influence may be modulated by density‐ or phase‐dependent processes. We hypothesized that for ungulates, climatic conditions close to parturition have a greater influence on the predation risk of neonates during population declines, when females are already under nutritional stress triggered by food limitation.
-
We examined the presence of phase‐dependent climate–predator (PDCP) interactions on neonatal ungulate survival by comparing spatial and temporal fluctuations in climatic conditions, cause‐specific mortality and per capita resource limitation. We determined cause‐specific fates of 1384 caribou (Rangifer tarandus) from 10 herds in Newfoundland, spanning more than 30 years during periods of numerical increase and decline, while exposed to predation from black bears (Ursus americanus) and coyotes (Canis latrans).
-
We conducted Cox proportional hazards analysis for competing risks, fit as a function of weather metrics, to assess pre‐ and post‐partum climatic influences on survival on herds in population increase and decline phases. We used cumulative incidence functions to compare temporal changes in risk from predators.
-
Our results support our main hypothesis; when caribou populations increased, weather conditions preceding calving were the main determinants of cause‐specific mortality, but when populations declined, weather conditions during calving also influenced predator‐driven mortality. Cause‐specific analysis showed that weather conditions can differentially affect predation risk between black bears and coyotes with specific variables increasing the risk from one species and decreasing the risk from the other.
-
For caribou, nutritional stress appears to increase predation risk on neonates, an interaction which is exacerbated by susceptibility to climatic events. These findings support the PDCP interactions framework, where maternal body condition influences susceptibility to climate‐related events and, subsequently, risk from predation.
Introduction
Animal population dynamics are driven by a variety of factors, including abiotic conditions, per capita food limitation and predation, as well as interactions among these factors. For example, climate can have effects on population dynamics through its influence on plant productivity or inclement weather (Coulson, Milner‐Gulland & Clutton‐Brock 2000; Pettorelli et al. 2007), as well as via changes in resource competition or predation risk, in parallel with direct effects (Fig. 1, Melis et al. 2009). In these instances, when climate variability occurs in tandem with other risk factors, unexpected outcomes may arise (Tyler et al. 2008; Brook, Sodhi & Bradshaw 2008). Recent studies (e.g. Griffin et al. 2011; Brodie et al. 2013) have further shown that climate–predator interactions can drive population dynamics through variable recruitment and adult survival. Indeed, small fluctuations in temperature and precipitation can affect predator‐driven mortality across a variety of taxa (Yasué, Quinn & Cresswell 2003; Gilg, Sittler & Hanski 2009). Yet, for most systems, the mechanisms linking climate to predator impacts on prey populations are poorly elucidated (Lima, Stenseth & Jaksic 2002).

The influence of climate on population dynamics can also be modulated by phase‐dependent processes (Boonstra, Krebs & Stenseth 1998; Stenseth et al. 1998b; Barbraud & Weimerskirch 2003; Previtali et al. 2009). Many animal populations are limited by food resources (i.e. bottom‐up regulation); in these systems, individuals are exposed to food‐favourable periods followed by a period of nutritional restriction (Fryxell & Sinclair 1988; Forchhammer et al. 1998). These periods of nutritional stress can limit population abundance through reduced adult and juvenile survival and reduced reproductive output (Gaillard et al. 2000; Sinclair, Fryxell & Caughley 2006), and can be exacerbated by climatic variability, leaving some individuals to face increased starvation risk following harsh weather conditions (Coulson, Milner‐Gulland & Clutton‐Brock 2000; Patterson & Power 2002). Therefore, we expect that nutritionally stressed individuals will have more difficulty coping with stochastic variation in forage availability, such as that induced by climate (White 2008). When considered in the light of climate–predator interactions, we might expect the influence of climate on predation also to be modulated by per capita resource limitation through a complex set of interactions (Fig. 1; Lima, Stenseth & Jaksic 2002; Wang et al. 2009).
The influence of such interactions is likely to be realized first in juvenile mortality rates (Eberhardt 2002). For ungulate species, population dynamics are driven by the interplay between relatively high and stable adult survival and more variable juvenile survival (Gaillard et al. 2000). Griffin et al. (2011) formalized two mechanisms underlying how climate and predation can interact indirectly in neonates. The first refers to a ‘maternal condition’ mechanism − that pre‐partum climatic conditions over previous years affect physical condition of parturient females by influencing fat reserves and thus neonate survival during the subsequent calving season. Females in poorer condition should give birth to calves in poorer condition that may be more vulnerable to predation (pathway 4, Fig. 1), or to forage in habitats conferring higher predation risk to sustain the cost of lactation (pathway 3, Fig. 1; Patterson & Power 2002; Pettorelli et al. 2005b; Couturier et al. 2009b). The second is a ‘current condition’ mechanism − that negative and stochastic post‐partum climatic conditions negatively influence neonate survival (pathway 2, Fig. 1). Females and newborn calves may alter foraging behaviour (e.g. vigilance) or nutrient intake due to exposure to variable weather events, leading to variable predation risk (Pettorelli et al. 2005b, 2007; Gustine et al. 2006). Therefore, the maternal condition hypothesis reflects previous weather, whereas the current condition hypothesis is associated with current weather. An alternate pathway may involve the direct impact of climate on predator behaviour or abundance, with subsequent consequences on neonate survival (pathway 1, Fig. 1).
However, maternal condition (previous weather) and current condition interactions have rarely been examined (Wang et al. 2009), especially in the context of neonate survival, variable causes of death, and population context (i.e., phase). We suggest that phase‐dependent nutritional stress may be particularly important in influencing the strength and interrelations between these mechanisms in resource‐limited populations (Stenseth et al. 1998c). Nutritionally stressed individuals should have more difficulty compensating for stochastic variation in climatic conditions than those in better body condition, leading to increased predator‐driven mortality. Therefore, we surmise that climatic influences on juvenile survival close to parturition will be more pronounced under nutritional stress than during resource‐abundant periods. This would give rise to what we term phase‐dependent climate–predator (PDCP) interactions (Wang et al. 2009). We postulate that these interactions are likely to apply to a wide range of species.
We tested the effect of PDCP interactions on neonate survival by comparing spatial and temporal fluctuations in weather conditions and predation rates, across 10 caribou (Rangifer tarandus) herds in Newfoundland, Canada. These populations represent an ideal system for evaluating climate–predator interactions because unlike other long‐term studies investigating neonate survival (e.g. Clutton‐Brock et al. 1987), animals in our system experience predation risk and the predator guild has varied through time. Unlike most ungulates, caribou inhabiting Newfoundland display long‐term, periodic oscillations in population size that originate from resource limitation and climatic events (Festa‐Bianchet et al. 2011; Bastille‐Rousseau et al. 2013). Consistent with this pattern, studies on caribou morphology (Mahoney et al. 2011) and spatial ecology (Schaefer & Mahoney 2013) support the conclusion that population fluctuations are related to nutritional stress in adult female caribou, with nutritional stress first appearing as populations reach high densities and slow regrowth of their food sources imposing delays in body condition improvement and population recovery (Bastille‐Rousseau et al. 2013). Thus, phases of population increase and decline are markedly different in terms of caribou nutritional stress level. We evaluated how PDCP interactions shape patterns of neonate predation risk, and predicted that divergent weather influences during the increase and decline periods will play a stronger role on calf predation risk during the population increase, whereas the influence of climatic conditions during the calving period will be more strongly manifest during the population decline. We predicted that the direct impact of weather on predation risk (pathway 1, Fig. 1) should be of minimal importance but becomes evident when the detected impact of weather variable ran counter to our current understanding of herbivore ecology. Finally, we predicted that calves born during the increase period (in better condition) will experience reduced cumulative risk throughout the post‐calving season compared to calves during population decline.
Materials and methods
Study area
Newfoundland is a 108, 860 km2 island in eastern Canada (47°44′N, 52°38′W to 51°44′N, 59°28′W) dominated by a mixture of coniferous and mixed forest, bogs, lakes and barren rock. It has a maritime climate with ample year‐round precipitation and generally mild winters (Environment Canada 2013). During the last 50 years, caribou herds on Newfoundland have undergone drastic changes in abundance, starting with a population low prior to the 1950s that persisted until the 1970s, followed by rapid growth to a peak population size in the late 1990s (hereafter, ‘period of increase’), and then followed by a precipitous decline that persists today (hereafter, ‘period of decline’; Mahoney & Schaefer 2002; Mahoney et al. 2011). These fluctuations are largely synchronized among herds (Bastille‐Rousseau et al. 2013). During the period of decline, reductions in recruitment, parturition rate and adult body size were indicators of changes in determinants of demographic trends (Bastille‐Rousseau et al. 2013; Weir et al. 2014). Most importantly, neonate survival decreased markedly during the decline, with average yearly survival <35% (Mahoney et al. 2016). On average, around 3·8% of the Newfoundland caribou population has been harvested annually, and notably, rates of harvest reached 7·6% during the period of rapid population decline. Black bears (Ursus americanus) are a common predator of neonates throughout the increase and decline periods, whereas coyotes (Canis latrans) became an important predator only in the 2000s, after recently colonizing the island (Mahoney et al. 2016). Other, less frequent predators of calves are lynx (Lynx canadensis), bald eagles (Haliaeetus leucocephalus) and red foxes (Vulpes vulpes). Wolves (Canis lupus) were also an historical predator of caribou in Newfoundland, but went extinct around 1920.
Caribou data
We estimated survival time of caribou calves based on VHF telemetry, 1979–2013, for 10 herds (Fig. 2). During late May and early June of each year, neonates were located from helicopters and captured on foot, generally ≤5 days after birth. Calves were marked with ear tags, although information on mass and sex was missing for calves in the 1980s. When a mortality signal was detected, field investigations revealed calf remains followed by site assessment of cause of death. Recent assessments were complemented by necropsy or by DNA analysis to verify the predator identity (Mumma et al. 2014). Because our focus was on the two main predators of neonates, we considered three categories of cause of death: black bear, coyote and other causes. We censored animals when radio transmitters detached prematurely or were lost. All individuals were censored after 200 days, as monitoring became less frequent at that time (less than twice per month) and mortality assessment became less precise.

Weather and vegetation data
We identified three periods: spring (April–June) which reflected the pre‐calving and calving period, summer (May–October), which reflected the growing season, and winter (December–March), the period of continuous snow cover. We collected mean daily temperature, mean daily precipitation (rain or snow) and mean number of growing degree days (base temperature = 10 °C; GDD10) for each season from 6 different weather stations (Environment Canada 2013). We associated each herd to the closest weather station (Table S1, Supporting information). We also used the December–March North Atlantic Oscillation (NAO) index (http://www.cgd.ucar.edu/cas/jhurrell/naointro.html; Hurrell 1995). As population dynamics of caribou can reflect multiple‐year lags and additive effects through time (Bastille‐Rousseau et al. 2013), we took the average of each variable over the preceding 5 years to reflect winter and summer weather (hereafter, ‘Variable(t‐5)’). Such intervals can also account for the impact of cohort effects on caribou maternal condition. Five‐year average variables were correlated with their 3‐year (mean R = 0·92) and 2‐year (mean R = 0·83) counterparts, and our results were therefore not sensitive to this choice.
To estimate variables related to vegetation cover, we used the Normalized Difference Vegetation Index (NDVI). NDVI can be used to measure plant growth, vegetation cover or biomass production (Pettorelli et al. 2005a,b). NDVI 10‐day composites from the Advanced High Resolution Radiometer (AVHRR) for the period 1985–2013 at 1‐km spatial resolution were processed by the Canadian Centre for Remote Sensing following Latifovic & Trishchenko (2005). A correction for systematic bias between AVHRR sensors was applied (Latifovic, Pouliot & Dillabaugh 2012) to improve radiometric consistency over the period. The average cloud and shadow‐free NDVI value for each 10‐day composite was extracted for each herd's core area of occupation. The time series of average NDVI values was temporally smoothed to remove outliers using a robust Lowess filter where, in each iteration, data falling below the fit line were removed for the next iteration (Fernandes, Latifovic & Chilar 2005). Several time‐series metrics were extracted for each year and included: (i) maximum difference between 10‐day composites from May to July as a measure of the rate of spring green‐up, (ii) beginning of the growing season taken as the point where 50% of the maximum NDVI was observed, and (iii) average seasonal NDVI for each year as a measure of annual productivity. The annual NDVI‐based productivity for the birth year and the preceding year were used during analyses to account for the previous year's conditions.
The core area of occupation for each caribou herd was defined using a kernel density estimator from mortalities and final locations of censored individuals. The smoothing parameter was estimated using the ad hoc method of Worton (1989). We used a kernel probability isopleth of 90% for all herds, except when herds with few locations resulted in illogically large delineations, in which case we used a 50% isopleth (i.e., Pot Hill and Mount Peyton herds).
Statistical analyses
Our objective was to contrast periods of caribou population increase and decline, to infer differences that could be attributed to weather and phase‐dependent nutritional stress, based on cause of mortality. We measured the relative influence of these risk categories on neonate survival using cause‐specific mortality analyses. We used cumulative incidence functions (CIF) to estimate mortality rates from each category under a competing risk framework (Fine & Gray 1999; Heisey & Patterson 2006). This allowed us to test our prediction of whether a calf born during a resource‐favourable period experienced lower mortality.
To further investigate potential mechanisms between weather and predation, we used data augmentation in the competing risk framework to assess cause‐specific hazards (Lunn & McNeil 1995). Cause‐specific risk analyses are analogous to standard hazard‐based regression approaches except that the survival function in cause‐specific risk analyses considers both the cause of mortality and failure time. As direct causes of mortality are mutually exclusive in our study, cause‐specific mortality probabilities sum to the total mortality probability (Murray et al. 2010). The data augmentation approach takes advantage of the additive relationships of hazard functions; the data set is duplicated for each cause of mortality and a dummy variable assigns a risk to each cause. Within each risk set, death is identified only for the appropriate cause, while other entries are censored (Murray et al. 2010). We used a flexible, semi‐parametric Cox proportional hazards (CPH) model with herd as a random factor. We used a right‐censored design with time‐at‐risk based on the time (days) since the animal was live‐captured (Fieberg & Delgiudice 2009). Because we captured most calves when they were ≤5 days old, we are confident that survival timelines corresponded closely with age.
Because a mixed‐effects model with both periods violated the assumptions of the proportional hazards model, we analysed each period separately. For each herd, we used the estimated year of the population peak to distinguish the two periods (Bastille‐Rousseau et al. 2013; Table 1). For three herds (Corner Brook Lakes, Gros Morne, and Sandy Lake), we did not have herd‐specific estimates; instead, we relied on islandwide population estimates to approximate their peaks. Even though the strength of per capita resource limitation and associated nutritional stress likely vary prior to reaching carrying capacity (Sinclair, Fryxell & Caughley 2006), our population size estimates were interpolated (Bastille‐Rousseau et al. 2013). Therefore, the precise year of peak density was uncertain for some herds. Likewise, representative measures of nutritional stress in caribou are not available at the herd level and prevented us from using such variables directly in our modelling. Furthermore, owing to a low number of collared caribou during the late 1990s when most herds peaked, our distinction between increasing vs. decreasing phases for each herd were not unduly influenced by potential minor error in estimating the change in the direction of growth, or by a lag in density‐dependent effects. To confirm this point, we performed two additional analyses. First, we ran survival models on two smaller subsets of years that represented largely homogenous nutritional stress levels. These two subsets included the period 1980–1990 for the increase phase, and 2000–2010 for the decrease phase. We also ran analyses using a common, fixed year (1998) to reflect islandwide peak abundance. In both tests, we observed no qualitative difference in our results, confirming that our approach was robust (Tables S2 and S3).
| Herd | Years monitored | Year of population peak | Cause of mortality (%) | n | ||
|---|---|---|---|---|---|---|
| Black bear | Coyote | Other | ||||
| Corner Brook Lakes | 1994–1997 | 1998 | 10·87 | 0·00 | 4·35 | 46 |
| Gaff Topsails | 2003–2004 | 1996 | 2·08 | 18·75 | 43·75 | 48 |
| Grey River | 1979–1992 | 1991 | 7·31 | 0·00 | 9·59 | 219 |
| Gros Morne | 1993–1996 | 1998 | 19·12 | 0·00 | 13·25 | 68 |
| Lapoile | 1985–2012 | 1988 | 10·69 | 11·72 | 17·93 | 290 |
| Middle Ridge | 1983–2013 | 1995 | 23·35 | 16·15 | 16·54 | 514 |
| Mount Peyton | 1993–2003 | 1996 | 15·79 | 0·00 | 21·05 | 19 |
| Northern Peninsula | 2008–2012 | 1996 | 13·79 | 11·03 | 17·24 | 145 |
| Pot Hill | 1980–1982 | 1998 | 0·00 | 0·00 | 28·57 | 14 |
| Sandy Lake | 1982–1984 | 1998 | 4·76 | 0·00 | 14·29 | 21 |
To test our general prediction about the specific mechanisms for PDCP interactions, we used model selection approaches based on second‐order Akaike information criteria (AICc; Burnham & Anderson 2002). Our global model had a suite of variables representing current weather (post‐partum) and vegetation, and two suites of variables representing previous winter weather, and previous summer weather and vegetation (maternal condition; Table 2). To unravel the influence of weather on survival in a hierarchical manner, we also considered subsets of the global model containing different combinations of each suite – that is, one model including only the current variables, one model including all previous variables, and two models including only previous summer or previous winter variables (Table 2).
| Variable suite | Variables | |
|---|---|---|
| Current | Spring (Current weather) | Spring growing degree days (GDD) |
| Spring rain | ||
| Rate of spring green‐up * | ||
| Winter snow | ||
| Spring temperature | ||
| Beginning of growing seaso | ||
| Previous | Summer(t‐5) (Previous weather; Previous summer) | Summer(t‐5) growing degree days (GDD) |
| Summer(t‐5) rain | ||
| Average NDVI (t‐5) * | ||
| Summer(t‐5) temperature | ||
| Duration of growing season | ||
| Winter(t‐5) (Previous weather; Previous winter) | Winter(t‐5) temperature | |
| Winter(t‐5) snow | ||
| North Atlantic Oscillation (NAO) |
- *Not included in model for the period of increase.
To select which individual variables to include in each suite of variables in the final models (Table 2), we first assessed whether the proportional hazard assumption was met with model diagnostics, based on Schoenfeld residuals (Fox 2002). We also assessed collinearity by calculating Pearson correlation coefficients between each pair of predictor variables and keeping only those predictors that had the strongest influence when two predictors were highly correlated (¦r¦ > 0·70; Dormann et al. 2012; see also Fig. S1). Lastly, we also assessed multicollinearity using the variance inflation factor (VIF; Graham 2003) As NDVI variables were not available prior to 1985, we first performed the model selection exercise with the data set restricted to post‐1984. As none of the NDVI‐related variables were statistically significant in the top models (G. Bastille‐Rousseau, unpublished), we conducted the analysis using all years of data, but without NDVI variables for the period of increase subset, to maximize the temporal scope of our analysis. Time series of all variables included in our model are presented in Figs S2 and S3.
Calf mass can be a key predictor of calf survival, which can be affected by nutritional stress and prevailing weather conditions. However, these data were available only after 2003, during the decline phase. To understand the impact of weather conditions and phase‐dependent nutritional stress on mass, we first ran linear mixed‐effects models testing the impact of variables representing the previous summers and previous winters on mass. We performed multimodel averaging (Burnham & Anderson 2002) and took the residuals of this averaged linear model as representative of calf mass variation, independent of weather (Table S4). We then reran our top model for the period of population decline including these residuals and the subset of data with calf mass. We tested whether the addition of calf mass improved model fit. As this model did provide better support, we focused our interpretation on this model, although for full disclosure we also present the model without the mass variable (Table S5). Inferences from both models were nevertheless comparable for most variables.
We considered only two categories of mortality during the period of increase: black bear and other, because coyotes had yet to be confirmed in Newfoundland in the mid‐1980s (McGrath 2004); the importance of coyotes as predators of neonatal caribou became evident in the late 1990s. During the period of decline, we included all three categories. Ultimately, we think that our approach was more robust and our inferences regarding coyote and black bear predation more conservative through the inclusion of the ‘other causes’ category. We used black bear as our reference category in survival models, and reconstructed cause‐specific hazard ratios for coyotes and other causes of death by adding the single effect of a given variable for black bear, with the interaction between another predator and the variable. Standard errors were then calculated using the covariance matrix. As the cause of mortality did not conform to the proportional hazards assumptions based on Schoenfeld residuals, we stratified each model by the cause of mortality; each had its own baseline hazard (Kleinbaum & Klein 2012). We reported pseudo R2 based on Cox & Snell (1989) log‐likelihood derivation, but corrected for number of censored observations (O'Quigley, Xu & Stare 2005). All analyses were conducted using the statistical software R v3.0.2 with packages ‘survival’, ‘cmprsk’, ‘coxme’ and ‘adehabitatHR’.
Results
Neonate mortality
During 1979–2013, cause‐specific fates were determined for 1384 neonates (Table 1). During the increase period, 537 caribou calves were monitored, with 110 mortalities (mortality rate = 0·20). Cause of death was roughly comparable between black bears (n = 51) and other causes (n = 59); to an unknown degree, some in the latter category were also due to bear predation. During the decline period, 847 individuals were monitored, with 468 mortalities (mortality rate = 0·55), of which 159 being attributable to black bear, 142 to coyote and 167 to other causes.
Neonate mortality rate from either black bear predation or other causes roughly doubled from the increase to decline period. Predation risk quickly levelled off roughly 10 days after birth during the increase period − a pattern not evident during the decline (Fig. 3). This notable rise in risk during the decline suggests that the recent and pronounced reduction in recruitment was not simply due to recent arrival of a new predator (coyote) in Newfoundland, but also from increased risk from traditional predators, primarily bear. During the decline period, bear predation risk was slightly higher than risk from coyote, but overall probability of succumbing to any cause of death increased similarly (Fig. 3).

PDCP interactions
We predicted that the influence of weather during calving would play a stronger role during the decline period, when adult females and their calves were presumably under nutritional stress. During the increase period, the most parsimonious model for neonatal survival included a set of variables strictly related to previous weather conditions (AICc w = 0·90; pseudo R2 = 0·26; random effect variance <0·01, Tables 3 and 4), whereas the top model during the decline period was the global model (Table S5), which included variables related to both previous and current weather (AICc w = 0·95; pseudo R2 = 0·28; random effect variance = 1·02, Table 3). This relationship is in direct support of our prediction, as variables related to weather during calving were not solely present in the top model during the decline period. Adding calf mass to the model resulted in a modest increase in model fit compared to the model without this variable (AICc w > 0·99; pseudo R2 = 0·30; random effect variance <0·01, Table 4).
| Model | K | AICc | ΔAICc | w |
|---|---|---|---|---|
| Period of increase | ||||
| 1. Previous weather | 8 | 1318·3 | 0·901 | |
| 2. Global | 14 | 1323·4 | 5·104 | 0·071 |
| 3. Previous winter | 4 | 1326·8 | 8·537 | 0·013 |
| 4. Previous summer | 4 | 1327·8 | 9·549 | 0·008 |
| 5. Current weather | 6 | 1333·0 | 14·772 | 0·001 |
| Period of decline | ||||
| 1. Global | 27 | 5777·2 | 0·950 | |
| 2. Previous weather | 15 | 5783·1 | 5·876 | 0·050 |
| 3. Previous summer | 9 | 5804·9 | 27·688 | 0·000 |
| 4. Current weather | 12 | 5820·2 | 43·057 | 0·000 |
| 5. Previous winter | 6 | 5835·6 | 58·398 | 0·000 |
| Variable | Black bear | Coyote | Other causes |
|---|---|---|---|
| Period of increase | |||
| Winter(t‐5) temperature | 0·547 (0·402, 0·746) | 1·058 (0·831, 1·347) | |
| Winter(t‐5) snow | 1·450 (0·337, 6·246) | 5·736 (1·49, 22·071) | |
| Summer(t‐5) GDD10 | 0·170 (0·040, 0·721) | 1·234 (0·554, 2·745) | |
| Summer(t‐5) rain | 3·858 (1·221, 12·194) | 0·151 (0·046, 0·501) | |
| Random effect variance: 0·00039 | |||
| Period of decline | |||
| Spring GDD10 | 0·266 (0·117, 0·604) | 0·836 (0·424, 1·646) | 0·601 (0·311, 1·165) |
| Spring rain | 0·549 (0·405, 0·745) | 1·303 (0·951, 1·787) | 0·915 (0·712, 1·177) |
| Spring rate of green‐up | 4205·118 (0, >100) | 0·002 (0·000, >100) | 0·000 (0·000, 0·411) |
| Winter snow | 0·562 (0·393, 0·803) | 1·080 (0·766, 1·523) | 1·092 (0·792, 1·506) |
| Winter(t‐5) temperature | 2·419 (1·272, 4·601) | 0·745 (0·436, 1·273) | 0·717 (0·420, 1·222) |
| Winter(t‐5) snow | 0·382 (0·173, 0·848) | 1·083 (0·539, 2·175) | 0·271 (0·148, 0·496) |
| Summer(t‐5) GDD10 | 0·213 (0·073, 0·617) | 1·796 (0·651, 4·953) | 1·005 (0·367, 2·750) |
| Summer(t‐5) rain | 0·779 (0·294, 2·070) | 3·069 (1·111, 8·481) | 1·728 (0·715, 4·174) |
| AvgNDVI(t‐5) | 0·000 (0·000, 2·128) | 0·270 (0·000, >100) | 0·000 (0·000, 0·000) |
| Residual Mass | 0·812 (0·722, 0·914) | 0·894 (0·791, 1·011) | 0·789 (0·700, 0·889) |
| Random effect variance: 0·000015 | |||
Roles of weather variables on cause‐specific survival
Weather conditions had variable influence on the categories of mortality risk (Table 4). Temperature during previous winters and number of GDD were the only variables correlated with black bear risk over both increase and decline periods. A 1 °C increase in average daily temperature reduced black bear predation by 45% in winter during the increase period, but surprisingly, bear predation risk rose by 142% per 1 °C through the decline period (Table 4, Fig. S4). We acknowledge that this finding remains difficult to fully reconcile. Daily snowfall during the previous winter also reduced risk by 62% per cm of snow during the decline period. The model predicted that 1 mm average increase in previous daily summer's rain would increase risk by 286%. Previous summer's GDD decreased risk by 83% and by 79% per GDD during population increase and decline, respectively (Table 4). During the decline period, current weather also correlated with risk. The model predicted that 1 daily mm increase in rain and 1 GDD increase in temperature reduced bear predation risk by 74 and 45%, respectively. The model further predicted that during the decline period, each 1 cm increase in daily snowfall in the previous winter reduced risk by an average of 44% (Table 4).
We assessed coyote risk during the decline period and found that predator‐specific risk was differentially correlated with weather conditions (Table 4). Daily amount of precipitation in previous summers increased coyote predation risk by 207% per mm of rain (Table 4). Other causes of mortality also were related to weather conditions. During the increase period, a 1 cm increase in daily snowfall in the previous five winters increased risk by 473% whereas a 1 mm increase in rainfall in the previous five summers reduced risk by 85%. However, during the period of population decline, only snow was influential. Daily amount of snow in the previous five winters reduced risk of mortality due to other causes by 73%. A 1 kg increase in calf mass also reduced risk from either black bear or other causes by 19 and 21%, respectively, although the influence of calf mass on mortality risk was limited to the decline period.
Discussion
In this paper, we investigated how an intrinsic process (calf mortality risk) can interact with biotic (predation) and abiotic (climate) processes to influence the demography of an ungulate population. Our results supported the hypothesis that climatic conditions during the calving period are more important for predator‐driven neonate mortality during a population decline, when individuals are likely under nutritional stress, than during an increase when individuals are less likely to be nutritionally stressed. When caribou populations were in their increase period (and food was presumably abundant), predation was most strongly influenced by summer and winter conditions that preceded the calving period, which was related to maternal body condition. In contrast, when populations began to decline from nutritional stress, weather conditions during calving, which likely influence calf body condition, also influenced survival. Our cause‐specific survival analysis also showed important variation in temporal pattern of predation risk between population increase and decline periods (Fig. 3).
Role of weather on cause‐specific survival
Weather often influences survival of juvenile ungulates (Patterson & Power 2002; Pettorelli et al. 2007; Couturier et al. 2009b), but the importance of seasonal variability on risk is affected by prey species and location (Gaillard et al. 2000; Coulson, Milner‐Gulland & Clutton‐Brock 2000). In general, favourable weather during summer and winter prior to calving helps females sustain neonates by increasing access to forage and reducing energetic costs of movement and thermoregulation (Pettorelli et al. 2005b; Couturier et al. 2009a). Likewise, suitable spring conditions may enhance survival by facilitating foraging (Gustine et al. 2006; Pettorelli et al. 2007), but little is known about how these climatic influences affect risk from specific predators.
Interestingly, some of our findings run counter to the prevailing understanding of how weather patterns affect ungulate survival (Pettorelli et al. 2005b, 2007; Gustine et al. 2006; Couturier et al. 2009b), possibly indicating that predator‐mediated effects of climate can counteract positive, direct effects (pathway 1 vs. pathways 2, 3 & 4 in Fig. 1). Given the correlational nature of these results, it is sometime difficult to identify with certainty the mechanism at play. For juvenile caribou, warmer winters, often assumed to heighten survival of neonates, increased risk from bears during the period of population decline. This could reflect the effect of warmer winters on denning behaviour and foraging habits of bears during spring (McLoughlin, Cluff & Messier 2002; Manchi & Swenson 2005). Another sharp contrast between our results and general predictions regarding climate and neonate survival pertains to the role of rain in summers prior to calving. In general, wetter summers should favour improved body condition of parturient females and result in improved calf survival, through corresponding increases in forage availability (Taillon et al. 2012) and calf mass (Bergerud, Luttich & Lodewijk 2008). Our results, however, showed that heightened predation risk from coyotes follow summers with more precipitation, and that calf mass did not affect risk from coyotes. While the mechanisms for these patterns are still unclear, previous summer weather could have a positive impact on coyote populations by impacting food availability such as berries; numerical increase in coyote abundance could then increase predation risk. We caution that this potential linkage is speculative, but this point gives further support to another potential mechanism for climate–predator interactions (pathway 1, Fig. 1), whereby climate favours predators either by improving hunting success or by increasing predator density (Post et al. 1999; Yasué, Quinn & Cresswell 2003). This general mechanism might be important for omnivores such as coyotes or bears that rely on other food sources.
Phase‐dependent variation in temporal pattern of cause‐specific risk
Although the relative importance of climate on mortality risk differed according to cause of death, we found cumulative risk of neonates to be surprisingly consistent among predators, during both periods of population increase and decline (Fig. 3). We were surprised to observe similar shapes in CIFs between coyote and black bear; bears are regarded as effective neonatal ungulate predators only during the first 4–6 weeks of life (Zager & Beecham 2006). That the CIFs were qualitatively comparable between predator species begs additional investigation into predator‐specific hunting strategies on ungulate calves, and how these influence prey encounter rates, predator avoidance tactics, and predator–prey population dynamics.
The most striking difference in cumulative risk was manifest through variation in the shape of the CIF curves between periods of population increase and decline. Neonatal caribou are highly mobile within 2 weeks after birth (Gustine et al. 2006). However, mortality risks to neonates (<2 weeks old) are independent of birth mass, which suggests that neonates are essentially defenceless against predators (Jenkins & Barten 2005). This observation corresponds well with the cumulative risk of caribou calves in Newfoundland during both population phases for the first 10 days of life, yet prominent differences emerged during the ensuing post‐calving period. We found that during the population decline, the cumulative risk of neonate caribou continued to rise during the post‐calving period. This contrast in cumulative risk could be attributable to different, non‐exclusive causes. One possibility is that calves born to nutritionally stressed females are smaller and may experience a delay in establishment of full mobility, thereby extending the period of defencelessness, although this period should span at most the first month of life (Taillon et al. 2012). Secondly, nutritionally stressed females may need to forage in riskier habitats to sustain energetic demands of lactation (Brown 1999). Thirdly, predators may elicit contrasting prey behavioural responses. In Newfoundland, owing to recent coyote colonization and establishment, calving caribou may have lost predator‐free refugia and thereby experience increased predation from all sources (Holt & Lawton 1994; Schmidt 2004). This would explain the higher mortality rates from all causes in the more recent period.
There is support for all three above mechanisms for the decline of neonate survival during the post‐calving period in Newfoundland. The evidence presented here and elsewhere (Mahoney & Schaefer 2002; Mahoney et al. 2011; Bastille‐Rousseau et al. 2013) most strongly supports the first two mechanisms invoking the effect of nutrition on calf risk. Nevertheless, analysis of habitat selection during calving revealed a disparity between predators in terms of their space‐use (Bastille‐Rousseau et al. 2015b), supporting the hypothesis that coyotes occupy what otherwise would have been less risky habitat for neonatal caribou. Whether the decrease in calf recruitment was driven by coyote colonization or food limitation, it is evident that coyote predation is at least partly additive to other mortality risks. Differentiating among these mechanisms is critical from a conservation standpoint because reduction in calf recruitment is the main proximate cause of decline in Newfoundland caribou (Weir et al. 2014). Further analysis of fine‐scale space‐use patterns of parturient females and their predators is a logical next step towards a better understanding of caribou population regulation.
Concluding remarks
Similar to other survival studies across broad temporal and spatial scales (e.g. Murray et al. 2010; Griffin et al. 2011; Brodie et al. 2013), ours is constrained by its observational nature. Notably, we did not consider predator density as a covariate in our survival models because such information was not available over the whole study period. More generally, population density estimates for free‐ranging and elusive large carnivores are notoriously difficult to obtain. This is an important point, as predator density may have varied indirectly with caribou population density, as observed elsewhere (Dale, Adams & Bowyer 1994). However, it is unlikely that our results are driven substantially by variability in predator density. Black bear and coyote densities are not spatially homogenous across Newfoundland (Fifield & Lewis 2013). This enables us to consider variation in predator density among herds experiencing similar levels of nutritional stress. Our hazard models included caribou herd as a random factor; this should have largely addressed between‐herd variation independent of variables not included in the model, such as predator density. Marginal inferences from these mixed models therefore reveal an average response to weather variables among all 10 herds experiencing varying predator density. Indeed, our results indicated little variation among herds. Lastly, long‐term variation in the climatic variables included in our model (Figs S2 and S3) do not show evidence of temporal trends that span multiple continuous years. The absence of such trends indicates that fluctuations in weather variables are unlikely to be correlated with fluctuations in predation abundance, given that fluctuations in abundance for large mammals generally happen over multiple years (Post et al. 2002). Notwithstanding these shortcomings, our study establishes cause‐specific hazard analysis as a powerful, yet underused, approach in ecology. This approach provided us with insight into the impact of weather conditions independently of the cause of death.
Ecology is rife with interactions. Many studies have examined interactions between intrinsic and extrinsic factors and their effects on animal survival. We formalized a new survival mechanism linking nutritional stress, climate and predation; PDCP interactions, where maternal body condition influences susceptibility to climate‐related events and, subsequently, risk from predation. The magnitude of PDCP interactions is likely to vary among species (Stearns 1992) and across a species’ range (Bjørnstad, Falck & Stenseth 1995; Stenseth, Bjørnstad & Saitoh 1998a). Although it may be too soon to affirm the generality of the PDCP hypothesis, our proposed framework sets an important foundation for future research. Specifically, the clear set of predictions about how per capita resource limitation, climate and predation interact in a phase‐dependent fashion provides an important framework that to date has not garnered sufficient attention in ecology.
Acknowledgements
This project was a component of the Newfoundland & Labrador Caribou Strategy, 2008–2013. This study was funded by the Institute for Biodiversity, Ecosystem Science & Sustainability; the Sustainable Development & Strategic Science Branch of the Newfoundland & Labrador Department of Environment & Conservation; and the Safari Club International Foundation. G. Bastille‐Rousseau was supported by a scholarship from the Natural Sciences and Engineering Research Council of Canada. We thank C. Soulliere, A. Loison, M. Festa‐Bianchet and an anonymous reviewer for comments on an earlier version of this manuscript. We are indebted to the Newfoundland & Labrador Department of Environment & Conservation for providing these long‐term data as well as the many government personnel who conducted the caribou captures and monitoring over four decades.
Data accessibility
All data were collected by the Newfoundland & Labrador Department of Environment & Conservation. Survival data are available from the Dryad Digital repository: http://dx.doi.org/10.5061/dryad.5dj78 (Bastille‐Rousseau et al. 2015a).
Notes :
- *Not included in model for the period of increase.
Number of times cited: 8
- Guillaume Bastille‐Rousseau, Dennis L. Murray, James A. Schaefer, Mark A. Lewis, Shane P. Mahoney and Jonathan R. Potts, Spatial scales of habitat selection decisions: implications for telemetry‐based movement modelling, Ecography, 41, 3, (437-443), (2017).
- Keith P. Lewis, Stephen E. Gullage, David A. Fifield, David H. Jennings and Shane P. Mahoney, Manipulations of black bear and coyote affect caribou calf survival, The Journal of Wildlife Management, 81, 1, (122-132), (2016).
- Matthew A. Mumma, Joseph D. Holbrook, Nathaniel D. Rayl, Christopher J. Zieminski, Todd K. Fuller, John F. Organ, Shane P. Mahoney and Lisette P. Waits, Examining spatial patterns of selection and use for an altered predator guild, Oecologia, 185, 4, (725), (2017).
- Daniel Gibson, Erik J. Blomberg, Michael T. Atamian and James S. Sedinger, Weather, habitat composition, and female behavior interact to modify offspring survival in Greater Sage‐Grouse, Ecological Applications, 27, 1, (168-181), (2017).
- Guillaume Bastille-Rousseau, Nathaniel D. Rayl, E. Hance Ellington, James A. Schaefer, Michael J.L. Peers, Matthew A. Mumma, Shane P. Mahoney and Dennis L. Murray, Temporal variation in habitat use, co-occurrence, and risk among generalist predators and a shared prey, Canadian Journal of Zoology, 94, 3, (191), (2016).
- James A. Schaefer, Shane P. Mahoney, Jackie N. Weir, J. Glenn Luther and Colleen E. Soulliere, Decades of habitat use reveal food limitation of Newfoundland caribou, Journal of Mammalogy, 97, 2, (386), (2016).
- Alexander Singer and Karin Frank, Viability of cyclic populations, Ecology, 97, 11, (3143-3153), (2016).
- Guillaume Bastille-Rousseau, James A. Schaefer, Michael J. L. Peers, E. Hance Ellington, Matthew A. Mumma, Nathaniel D. Rayl, Shane P. Mahoney and Dennis L. Murray, Climate change can alter predator–prey dynamics and population viability of prey, Oecologia, 10.1007/s00442-017-4017-y, (2017).




