Untangling Effects of Human Disturbance and Natural Factors on Mortality Risk of Migratory Caribou

Plante, Sabrina; Dussault, Christian; Richard, Julien H.; Garel, Mathieu; Côté, Steeve D.

doi:10.3389/fevo.2020.00154

ORIGINAL RESEARCH article

Front. Ecol. Evol., 24 June 2020

Sec. Behavioral and Evolutionary Ecology

Volume 8 - 2020 | https://doi.org/10.3389/fevo.2020.00154

This article is part of the Research Topic Advances in Ungulate Ecology View all 22 articles

Untangling Effects of Human Disturbance and Natural Factors on Mortality Risk of Migratory Caribou

$\r\nSabrina Plante,*$ Sabrina Plante^1,2*

Christian Dussault^1,3

Julien H. Richard^1,2

Mathieu Garel⁴

Steeve D. Côté^1,2

¹Caribou Ungava, Département de Biologie, Université Laval, Québec, QC, Canada
²Département de Biologie, Centre d’Études Nordiques, Université Laval, Québec, QC, Canada
³Ministère des Forêts, de la Faune et des Parcs, Direction de l’Expertise sur la Faune Terrestre, l’Herpétofaune et l’Avifaune, Québec, QC, Canada
⁴ONCFS – Unité Ongulés Sauvages, Gières, France

Human disturbances are rapidly increasing in northern and Arctic regions, raising concerns about the recovery and persistence of declining caribou (Rangifer tarandus) populations. Yet, the consequences of behavioral responses toward human disturbances on vital rates rarely have been investigated. Herein, we assessed the cumulative and instantaneous effects of human disturbances (roads, human settlements, mines and mining exploration) at different temporal scales on the mortality risk of 254 GPS- collared migratory caribou monitored in two herds, the Rivière-aux-Feuilles (RFH) and Rivière-George (RGH) herds, in northern Québec and Labrador, Canada. We also assessed the relative importance of human disturbances on caribou mortality risk compared with non-anthropogenic factors, including habitat use by caribou, predation risk by gray wolves (Canis lupus), and local weather conditions. Human disturbances alone, exclusive of hunting, had a limited impact on mortality risk of caribou. Repeated exposure to disturbances did not have detectable effects on mortality risk during the early life period (1−7 years old), but more abundant precipitation (RFH) or the use of areas with a higher predation risk (RGH) did so. At the seasonal scale, non-anthropogenic factors, particularly the use of highly selected habitat by caribou and air temperature, had a greater effect than anthropogenic factors on the mortality risk in the RFH. Caribou of the RFH using more frequently higlhy selected habitats decreased their chance of mortality during winter, whereas individuals using warmer areas during summer faced a higher risk of mortality. At the daily scale, we observed that anthropogenic and non-anthropogenic factors generally had either no effect on the daily risk of mortality, or their effects were undistinguishable from the effect of latitude, with which they were highly correlated. The only exception was for the RFH in winter, for which the daily risk of mortality increased 10 folds for each 10-km increment closer to industrial disturbances. Although the impacts of human disturbances on caribou survival were limited to specific regions and areas, we nevertheless detected a negative effect on survival on the RFH, even at the currently low level of human development. Our study highlights the importance of assessing effects of human disturbances at various spatiotemporal scales, and of considering the relative influence of other non-anthropogenic factors to fully understand drives of wildlife populations.

Introduction

In the last decades, the human footprint has increased rapidly, including >70% of ecoregions around the world (Venter et al., 2016). The increased human pressure on ecosystems has been identified as the main cause of species extinction and biodiversity loss (Brooks et al., 2002; Davies et al., 2006; Fahrig and Rytwinski, 2009). Effects of human disturbances on wildlife vary from short-term modifications in behavior, activity budgets and nutrition (Dussault et al., 2007; Benhaiem et al., 2008; Marchand et al., 2014), to processes occurring at larger spatiotemporal scales, such as resources and range use (Moreau et al., 2012), distribution (Shackelford et al., 2018), population dynamics (Sorensen et al., 2008), and species interactions (Courbin et al., 2009).

Quantifying impacts of human disturbances on vital rates of wildlife is of particular interest in ecology because it can help determine if and how individual responses to disturbances translate into consequences at the population level. Human disturbances can directly effect survival, including mortality caused by collision with vehicles (Forman and Alexander, 1998; Lodé, 2000) or when species vulnerability to hunting increases because of human development (Lebel et al., 2012; Plante et al., 2017). Most effects of disturbance, however, are not directly lethal for wildlife, but include energetic and lost-opportunity costs (Frid and Dill, 2002). These costs can accumulate over time and space, and ultimately translate into long-term consequences on individual performance and vital rates (Bradshaw et al., 1998; Johnson and St-Laurent, 2011). Over time, the consequences of these indirect effects on population dynamics and trophic interactions can exceed those from direct sources of mortality (Creel and Christianson, 2008). Nevertheless, many studies still focus on single development projects and short-term behavioral responses toward disturbances, without further investigating the potential cumulative effects over broad spatiotemporal scales, and perhaps more importantly, without considering the consequences on vital rates (Weladji and Forbes, 2002; Tablado and Jenni, 2017).

Multiple natural factors may also act simultaneously with human disturbances to further effect vital rates in wildlife populations (Yamasaki et al., 2008; Tablado and Jenni, 2017; Ramey et al., 2018). Consequently, the study of relationships between habitat use and animal performance is an effective tool to assess the relative importance of multiple limiting factors acting simultaneously (Gaillard et al., 2010; DeCesare et al., 2014; Uboni et al., 2017). Patterns of habitat use often are assumed to represent an aggregated response to multiple factors acting at different spatiotemporal scales, and should reflect priorities or trade-offs among competing needs (Senft et al., 1987; Rettie and Messier, 2000). Linking habitat use to vital rates may thus help to reveal the relative importance of multiple natural and anthropogenic factors for wildlife populations (McLoughlin et al., 2010).

Animals may not be able to minimize the effect of one or multiple limiting factors without compromising their performance to face other factors. In risky environments for example, individuals can fail to reduce predation risk sufficiently through habitat use (DeCesare et al., 2014). Alteration of the environment also can create discrepancies between perceived and true habitat suitability (Schlaepfer et al., 2002). If poor or risky habitats are used or become attractive to individuals despite reduced quality, those habitats become ecological traps where individuals are attracted despite their reduced chance of survival (Battin, 2004). Such non-ideal or maladaptive behaviors have mostly been reported in rapidly changing environments (e.g., human-altered landscapes) where modifications of the habitat occur at a much faster rate than the potential adaptation rate of animals (Battin, 2004; Robertson and Hutto, 2006).

In past decades, northern and Arctic regions have witnessed a drastic increase in human disturbances, and serious concerns have been raised regarding the ability of individuals to adapt, or populations to persist in these newly modified landscapes (United Nation Environmental Programme [UNEP], 2001). Caribou and reindeer (Rangifer tarandus; hereafter caribou) are at the very heart of these considerations because they represent a key species both ecologically and culturally (Bergerud et al., 2008). Although caribou populations have fluctuated in the past, the current decline of most caribou populations worldwide indicates that large-scale modifications induced by the development of human activities may contribute to this situation (Vors and Boyce, 2009; Festa-Bianchet et al., 2011). Caribou could be particularly sensitive to human disturbances because they occupy broad ranges and have a limited intrinsic capacity of population growth because of their low productivity (Cardillo et al., 2005; Fahrig and Rytwinski, 2009). Numerous studies have also reported strong behavioral responses of caribou toward human disturbances (Vistnes and Nellemann, 2008), such as the avoidance of infrastructures over several kilometers (e.g., Boulanger et al., 2012; Johnson and Russell, 2014). Until recently, human development has been relatively limited in the ranges of the migratory tundra caribou compared with what has been experienced by boreal caribou populations. Yet, very few studies have assessed effects of human disturbance on vital rates of migratory caribou (i.e., reproductive rate only; Nellemann et al., 2003; Cameron et al., 2005). As human development continues in the north, it is crucial to assess the impacts of human disturbances on the vital rates and demography of migratory tundra caribou before levels of development occur that would impede population persistence and recovery (Festa-Bianchet et al., 2011).

To assess effects of human disturbances on the survival of wide-ranging animals such as migratory caribou, we need to consider natural limiting factors acting independently or simultaneously with human disturbances. Climate, predation, parasites, insect harassment, and diseases are natural factors known to reduce survival that may also ultimately contribute to the decline of caribou populations (Festa-Bianchet et al., 2011; Mallory and Boyce, 2018). Although warmer temperatures during summer have been positively associated with the growth of some caribou herds (Mallory and Boyce, 2018), those temperatures also can increase insect harassment, parasite load, and heat stress (Soppela et al., 1986; Weladji et al., 2003), all entailing physiological costs and energy expenditures that deteriorate body condition (Toupin et al., 1996; Pachkowski et al., 2013). Warmer temperatures can advance melt-down and delay freeze-up of water bodies, which could dramatically increase movement costs for caribou (Leblond et al., 2016), or lead to death when caribou try to cross over thin ice (Poole et al., 2010). Warmer temperatures during winter can also increase the frequency and intensity of rain-on-snow events, which limit access to food resources and may lead to major mortality events (Forbes et al., 2016; Berger et al., 2018). Climate warming is also increasing the frequency and intensity of fires in the tundra and boreal forest, thus shrinking the available habitat for caribou (Rupp et al., 2006; Joly et al., 2011; Skatter et al., 2017).

Although caribou have cohabited with natural predators for centuries and have evolved efficient antipredator tactics (e.g., long-distance migrations; Seip, 1991; Bergerud et al., 2008), human disturbances could modify predator-prey relationships and further increase the impact of predation (Latham et al., 2011; Dussault et al., 2012; Newton et al., 2017; DeMars and Boutin, 2018). For example, linear infrastructures can facilitate gray wolf (Canis lupus) movements and access to certain areas of caribou ranges, which further increases effects of predation on population dynamics (James and Stuart-Smith, 2000; DeMars and Boutin, 2018).

Our overarching hypothesis was that effects of human disturbances on the mortality risk of eastern migratory caribou as well as the relative importance of such disturbances compared with natural factors known to affect survival, including habitat use by caribou, predation risk by wolves and local weather conditions (i.e air temperature and precipitation) were driving caribou abundance. We focused on two caribou herds located in northern Québec and Labrador, Canada, the Rivière-aux-Feuilles (RFH) and the Rivière-George herds (RGH), between 2009 and 2016. Over the last decades, these two herds, which are part of the eastern migratory caribou designatable unit, have declined to a point where they have been recommended to be listed as Endangered in COSEWIC (2017). Simultaneously with these declines, the region has experienced a moderate increase in human activity, owing mainly to an expansion of the mining (MRNF, 2012). Previous research showed that caribou of these herds avoided human disturbances over distances up to 23 km, which can cause substantial habitat loss within their range (Plante et al., 2018). Despite this avoidance, caribou remain more vulnerable to harvest near human infrastructures (Plante et al., 2017). Thus, there is a need to better understand and quantify the interactions between human disturbances and the mortality risk of caribou.

We assessed the cumulative and instantaneous effects of human disturbances on mortality risk in the two caribou herds by considering three temporal scales: early life, seasonal, and daily scales. We first evaluated whether repeated exposure to disturbances increased mortality risk during the early period of caribou life (1−7 years old) and at the seasonal scale (winter and summer). We also evaluated whether daily exposure to disturbances increased daily mortality risk of individuals. Secondly, we used a conceptual framework inspired by DeCesare et al. (2014) to compare the relative importance of human disturbances and non-anthropogenic factors on mortality risk of caribou. We first considered a habitat use hypothesis, where mortality risk of caribou is mostly affected by the proportion of time spent in highly selected habitats (H1; Figure 1A). We expected that caribou spending less time in habitats that are strongly selected at the population level would face a higher risk of mortality because of the potentially lower quality of these habitats or to the marginality of their behavior (e.g., bolder personalities), which could increase predation risk (Lesmerises et al., 2019). Alternatively, the use of highly selected habitats could be a “non-ideal” behavior, being inefficient at reducing the mortality risk associated with one or more factors. We thus considered the two alternative hypotheses where, regardless of the use of highly selected habitats, most variation in the mortality risk would be explained by exposure to anthropogenic or other non-anthropogenic factors. Accordingly, we considered the disturbance risk and predation risk hypotheses, where individuals more exposed to disturbances (H2; Figure 1B), or predation risk (H3; Figure 1C), respectively, have a higher mortality risk. We also considered a weather hypothesis, where adverse temperature and precipitation conditions encountered by individuals during winter (e.g., heavy snow) and summer (e.g., warm temperature) increase mortality risk (H4; Figure 1D). We also tested the hypothesis of additive effects of habitat use and other anthropogenic or natural factors on survival (H5; Figure 1E). Finally, we considered the hypothesis of maladaptive habitat use (Battin, 2004; Robertson and Hutto, 2006), where spending more time in highly selected habitats could increase exposure to disturbances or predation risk and consequently reduce survival (H6; Figure 1F).

FIGURE 1

Figure 1. Hypotheses and predictions tested to explain variations in the risk of mortality of eastern migratory caribou of the Rivière-aux-Feuilles (RFH) and Rivière-George herds (RGH), northern Québec and Labrador, Canada (2009–2016). Competing hypotheses H1 through H6 (as given in the text) included (A) the use of highly selected habitats (% of individual locations in highly selected habitats), (B) exposure to disturbance risk, (C) exposure to predation risk, (D) exposure to poor weather conditions, (E) additive effects, and (F) maladaptive habitat use.

Materials and Methods

Study Area

The RFH and RGH have both experienced steep declines in the last decades. The RFH has peaked at >500,000 individuals around 2001 (Couturier et al., 2004), and decreased to an estimated population size of 199,000 (CI 90% = 183,080−214,920) in Taillon et al. (2016). The RGH has experienced a more drastic decline, with an estimated population peak at 823,375 (CI 90% = 721,375−925,375) individuals around 1993 (Couturier et al., 2004), 74,000 (CI 90%=60,680-87,320) in 2010, and an estimated population of <8,900 individuals (CI 90% = 8,232−9,568) in MFFP and NLFLR (2016, unpubl. data). Numerous factors are suspected to have contributed to these declines, including resource limitation and density-dependent effects, climate, hunting and predation (Côté et al., 2010).

The RFH and RGH range over more than 1,000,000 km² north of the 51st parallel in northern Québec and Labrador, mostly within the Nunavik and Nunatsiavut regions of Canada (Figure 2). The area is characterized by subarctic and arctic climates, with short and cool summers (1981−2010; 9.7°C on average for the warmest trimesters) followed by long and cold winters (−19.5°C on average for the coldest trimesters; Berteaux et al., 2018). Precipitation averages 898 mm⋅year^–1, mostly falling as snow between October and March. The RFH and RGH ranges are characterized by numerous lakes and rolling hills with more rugged terrain in the northeast portion of the RGH range.

FIGURE 2

Figure 2. Winter and summer ranges of eastern migratory caribou of the Rivière-aux-Feuilles (RFH) and Rivière-George herds (RGH) and distribution of human disturbances in northern Québec and Labrador, Canada (2009–2016).

In April, caribou of the RFH undertake a northward migration from their winter range, located in the boreal-taiga forests, to their calving ground and summer range, in the Arctic tundra (mean distance [2000−2011] = 615 km; Le Corre et al., 2017). Vegetation on the RFH summer range is mainly composed of shrubs (Betula sp. and Salix sp.), grasses, herbaceous plants and lichens (Latifovic et al., 2017). In autumn, the RFH migrates south and returns to its winter range in the taiga forest, which is dominated by black spruce (Picea mariana) and tamarack (Larix larcina) interspersed with lakes and open areas covered by shrubs and lichens. Caribou of the RGH used to undertake similar but shorter migrations between summer and winter ranges (mean distance from 2000 to2011 = 350 km; Le Corre et al., 2017). Migration paths have continued to shorten in recent years (Caribou Ungava, unpublished results), resulting in great overlap between summer and winter ranges for the 2009−2016 period. Summer range was composed mainly of grasses and shrubs, with conifer forest in the southern portion and in valleys. Winter range had a similar composition, but with a higher proportion of conifer forest because the range extended further south of the tree line.

Caribou of the RFH and RGH cohabit with two large predators: the gray wolf and the black bear (Ursus americanus). Wolves are known to predate both adults and calves (Bergerud et al., 2008), whereas black bears mainly predate calves opportunistically (Zager and Beecham, 2006; Dussault et al., 2012). Other large ungulates such as moose (Alces alces) and muskox (Ovibos moschatus) occur at low densities in the area. Caribou of the RFH and RGH are the main large prey available year-round for wolves on these ranges (M. Bonin, unpublished results).

Humans represent both a direct and an indirect risk of mortality for caribou of the RFH and RGH. Caribou are an important subsistence food for northern communities (Bergerud et al., 2008). On the RGH, sport hunting was discontinued in 2012, and ban on all hunting in Labrador was enacted in 2013. Despite this, subsistence harvesting continued through the remainder of the study period. Hunting is known to have been a predominant source of mortality for the RGH through the study period and to the present (Department of Fisheries and Land Resources, Newfoundland, unpubl. data). On the RFH, autumn and winter sport hunting and subsistence harvest occurred throughout the study period. Indirectly, human disturbances could affect caribou survival through habitat loss and cumulative costs associated with the behavioral and physiological responses of avoiding infrastructures. Human disturbances include human settlements, mostly located along the coast, mining operations and exploration (mostly drilling sites), three major roads stretching outside settlements, and hydroelectric infrastructures (power lines, reservoirs and dams; Figure 2). Winter ranges are more disturbed than summer ranges, but human disturbances remain at relatively low density across the area (Table 1).

TABLE 1

Table 1. Total area (km²) of caribou seasonal ranges and density of human disturbances (per 1000 km²) in the ranges of the Rivière-aux-Feuilles (RFH) and Rivière-George (RGH) herds in northern Québec and Labrador, Canada, 2009−2016.

Capture and Monitoring of Caribou

We focused on the mortality risk of adult female caribou because population dynamics are highly sensitive to small variation in the survival of adult females (Gaillard et al., 2000). Between 2009 and 2016, we captured 254 yearling and adult females (RFH = 119; RGH = 135) using a net gun fired from a helicopter, and we equipped them with GPS collars (GPS Iridium and Globalstar, Vectronic Aerospace GmbH) programmed to record locations every 1 to 13 h (MFFP, NLFLR and Caribou Ungava, unpubl. data). We determined environmental characteristics used by caribou and the survival status of individuals (alive or dead) using the information provided by GPS collars. We considered individuals as dead when we received a mortality signal or, in cases where mortality sensors failed, when caribou stopped moving permanently. In many instances (one-half of the individuals; Table 2), the cause of death was difficult to determine because we could visit carcasses only several months after death. We thus pooled all mortality causes in analyses.

TABLE 2

Table 2. Descriptive statistics of migratory caribou mortalities during summer and winter of 2009−2016 for the Rivière-aux-Feuilles (RFH) and Rivière-George (RGH) herds in northern Québec and Labrador, Canada.

Human Disturbance Effects on Mortality Risk

We first assessed the effects of human disturbances on caribou mortality risk at three temporal scales: early life (1−7 years old), seasonal, and daily scales. This allowed us to distinguish cumulative effects of disturbances (effects detectable over a season or a critical part of the lifetime) from instantaneous effects (effects observed within a day). We defined human disturbances as infrastructures and their potential associated human presence or activity. We had no information on the nature or level of activity associated with each infrastructure, but assumed that the effects of infrastructures on caribou behavior include the human activity occurring at, or around, infrastructures. For all scales of analysis, we quantified caribou exposure to a combination of all types of disturbances, which included mines, mining exploration, villages and roads. We excluded hydroelectric infrastructures because they were located at the southern limit of winter ranges of both herds. Caribou exposure to this type of disturbance was thus limited and previous research indicated a limited impact of this disturbance on caribou behavior (Plante et al., 2018). We also quantified exposure to industrial (mines and mining exploration) and non-industrial disturbances (villages and roads) separately. We expected non-industrial disturbances to have both instantaneous (e.g., higher risk of harvest near villages and roads) and cumulative effects on survival (e.g., stress and avoidance costs), whereas we mainly expected cumulative effects for industrial disturbances, because there is no harvest near exploration and mining sites. We restricted our assessment to disturbances located within 100 km of caribou locations. We considered caribou as exposed to disturbance within this radius because it included the largest zones of influence of disturbances reported for these herds (max = 23 km; Plante et al., 2018), and it encompassed the area where the risk of harvest by sport and subsistence hunters was the highest (90% of harvest occurred within <100 km of an infrastructure).

Early Life Risk of Mortality

We first quantified the effect of repeated exposure to human disturbances on the mortality risk during the early life of caribou (1−7 years old). We wanted to evaluate the possibility that the costs of repeated exposure to disturbance over the first 1−7 years-of- life could accumulate over time to a point where it would impact survival. We suspected that, if disturbance effects on survival were mostly observed through cumulative effects over time, smaller scales (one season or day) would not be enough to detect such effects. We modeled early life mortality risk of 112 female caribou (RFH = 62; RGH = 50) captured as yearling and monitored continuously until their death or the end of the study. We limited our analysis to caribou captured as yearling to avoid uncertainty in age estimation (aging for individuals captured at >1 year was based on tooth wear). Among this sub-sample, 40 caribou died during the study (RFH = 17; RGH = 23). The age at death was calculated as the difference in days between estimated birth date (set at June 15th the year before capture for all individuals; Taillon et al., 2016) and date of death. Because our study extended from 2009 to 2016, individuals that survived throughout the study period reached a maximum of 7 years old. This is near the upper limit of the prime-age age class for most large herbivores and explains why we considered our study to target the early life of migratory caribou (Gaillard et al., 2000). We quantified early life exposure to disturbances by computing the proportion of locations within 100 km of a disturbance during the complete monitoring period of each individual. We tested effects of human exposure on early life rate of mortality using Cox proportional hazard models (function coxph in Survival package, R 3.5.1 Software, R Core Team, 2018). We verified the assumption of proportional hazard over time (function cox.zph), and excluded models for which the assumption was violated. We assessed the mortality risk using the hazard ratio (HR; exponential of β), that is the ratio between hazard rates of one unit of the factors tested. As such, mortality risk increased when HR > 1, decreased when HR < 1 and was not significantly influenced by the tested variable when the 95% CI included 1. We quantified the proportion of variance in mortality risk explained by human disturbances using the Nagelkerke’s adjusted R² (function r.squaredLR in package MuMIn), and we verified model performance compared with the null model using a likelihood ratio test (in coxph output; Hosmer and Lemeshow, 2000).

Seasonal Mortality Risk

At the seasonal scale, we compared exposure to human disturbances for individuals that survived throughout our study, and individuals that died within a given season. We focused on summer and winter seasons because most mortalities occurred during these seasons and mortality events were too rare in other biological seasons to perform the survival analysis (Table 2). Summer and winter seasons were temporally delineated using First Passage Times that identify variations in caribou movements between seasons (Le Corre et al., 2014). We used the mean date of initiation and completion of migrations for all individuals to define each season on a yearly basis. Depending on the year, summer season started between mid-June and mid-July (RFH: 30 June – 10 July; RGH: 19 June – 8 July) and ended between the end of August and mid-October (RFH: 31 August– 20 October). Winter season started between mid-November and the end of December (RFH: 20 November – 11 December; RGH: 9 November – 31 December) and ended between the end of March and the beginning of May (RFH: 27 March – 27 April; RGH: 9 April – 5 May).

We used 465 individual-seasons to model summer mortality risk (RFH = 233 with 14 mortalities; RGH = 232 with 18 mortalities), and 329 individual-seasons to model winter mortality risk (RFH = 147 with 14 mortalities; RGH = 1 82 with 25 mortalities). We quantified repeated exposure of caribou to disturbances by calculating the proportion of seasonal locations within a 100-km radius of any disturbances, and types of industrial and non-industrial disturbances separately. We statistically tested the relationship between exposure to disturbances and mortality risk using mixed-effects Cox proportional hazard models (coxme, package Survival in R). We set the year as a random effect to account for the annual variations in exposure to disturbances, in the number of active industrial activities, and in the length of summer or winter seasons across years (Table 2). We determined the proportion of variance explained by disturbances and the performance of the model compared with a null model using the same approach as for the early life scale.

Daily Mortality Risk

As for the seasonal analysis, we focused the daily survival analysis on mortalities occurring during summer and winter. We used 858 individual-days during summer (RFH = 5 57 with 14 mortalities; RGH = 301 with 18 mortalities), and 636 individual-days during winter (RFH = 394 with 14 mortalities; RGH = 2 42 with 25 mortalities). We assessed the effect of human disturbances on daily mortality risk by comparing distance to disturbances of individuals that would eventually die and of individuals that survived during the 24 h before the death occurred. We created statistical clusters, which included the locations of an individual during the 24 h before it died and the locations of all caribou that were alive during the same period. For each caribou location, we calculated the Euclidian distance to the nearest disturbance of any type, and to the nearest industrial and non-industrial disturbances. We truncated distance values at 100 km, and averaged them over the 24-h period. To facilitate the interpretation of the coefficients associated with daily risk of mortality and make that value comparable to the analyses performed at other scales, we multiplied distance values by −1. By doing so, higher coefficients represented greater exposure to disturbances, as for the other scales. We used conditional logistic regressions (clogit, package Survival in R). We could not assess model performance compared to the null model using a likelihood ratio test because the calculation of R² was not possible for this type of model.

Relative Importance of Human Disturbances and Natural Factors

Use of Highly Selected Habitats by Caribou

We compared the relative impact of human disturbances on caribou mortality to that of natural factors at the three temporal scales of analysis. Following our hypotheses, we evaluated the role of habitat use by caribou on mortality risk (H1, Figure 1A). The observed patterns of habitat use were considered as aggregate responses of caribou toward multiple factors. These responses were described using Resource Selection Functions (RSFs; Manly et al., 2002) on 334 female caribou (RFH = 168; RGH = 166). The RSF was performed at the herd level, for each season separately (see Supplementary Appendix 1 for details). The RSF included 5 to 6 vegetation cover types, the Normalized Difference Vegetation Index (NDVI; 250 m 16-day composite VI from the MODIS MOD13Q1 product; Didan, 2015) as an index of vegetation productivity, the elevation and the proximity to water. We assessed the predictive performance of RSF models with a k-fold cross-validation (Boyce et al., 2002). When predictive performance was good (r_spearman > 0.70), we used the selection coefficients to predict the relative probability of occurrence of caribou on seasonal ranges.

We then described the individual patterns of use of highly selected habitats relative to the population RSF to assess how individual variations affected mortality. We first extracted the RSF score of the population model under each caribou location. We then defined highly selected habitats as the highest 25% of RSF scores among those for caribou locations. For the early life and seasonal scales, we quantified the use of highly selected habitats by calculating the proportion of caribou locations in these habitats. For the daily scale, we directly compared the RSF score at caribou locations for individuals that died and individuals that survived during the 24 h preceding death.

Predation Risk by Wolves

We determined whether caribou more exposed to predation risk by wolves faced an increased risk of mortality (H3; Figure 1C). The predation risk was defined based on RSF models of 42 wolves equipped with GPS collars in the caribou seasonal ranges (RFH = 28; RGH = 14; see Supplementary Appendix 1 for details). We included the same variables we used for caribou RSFs, i.e., vegetation cover type, NDVI, elevation and proximity to water in the wolf RSF models. We also used the same seasonal areas and periods for describing habitat selection by wolves that we used for caribou. We avoided including indices of caribou use in habitat selection-models for wolves because we aimed at identifying habitat characteristics associated with higher wolf occurrence, independently of caribou use. We extracted the relative probability of wolf occurrence at each caribou location and we defined risky habitats as those having a RSF score ≥ 75th percentiles among RSF scores occurring at caribou locations. We calculated the proportion of locations of each caribou in risky habitats when assessing the effect on early life and seasonal mortality risk, and used RSF scores of wolves to make predictions at caribou locations when assessing daily mortality risk.

Weather Conditions

For weather conditions (H4; Figure 1D), we extracted the mean daily temperature (°C) and total daily amount of precipitation (kg/m²) at each caribou location (National Oceanic and Atmospheric Administration [NOAA], 2017 climatic data; 35 km × 35 km-resolution). To ensure that weather conditions were comparable among individuals. even if they had different monitoring periods, we calculated the daily differences between temperatures and precipitations at locations of an individual and the mean values of these variables for all other individuals monitored during the same day. Therefore, we used weather conditions indices that contrasted the conditions faced by an individual compared to average conditions faced by all monitored individuals during a given day. We chose to model differences in the meteorological conditions encountered by individuals versus the mean value instead of actual local values because caribou in both herds spread over a large area within a given season, resulting in large differences in conditions encountered by individuals. This approach allowed us to test the effect of encountered conditions on survival, while keeping a certain amount of variability among individuals. It also limited the effect of extreme values on survival output, i.e., when individuals were far from the main groups.

Finally, we tested the hypotheses of additive effects of habitat use and human disturbances, predation or weather (H5; Figure 1E) by including different combinations of these factors in the models. We also tested the hypothesis of maladaptive habitat use in regards to human disturbances or predation risk (H6; Figure 1F) by including the interaction between the use of highly selected habitats and caribou exposure to human disturbances or predation risk by wolves in other candidate models.

Model Selection

We used the Akaike Information Criterion (AIC; Burnham and Anderson, 2002) to compare the relative fit of the models derived from our hypotheses. For nested models with ΔAIC ≤ 2, we retained the model with fewer variables because additional covariates were considered uninformative (Arnold, 2010). Otherwise, we interpreted all non-nested models with ΔAIC ≤ 2. We also compared our candidate models to a null model to verify that covariates included in models improved model fit (Mac Nally et al., 2018). We relied on the likelihood ratio test to assess the performance of best models compared with null models. At the daily scale, we could not test models including only the intercept with conditional logistic functions. We thus compared model performance to a model including only the latitude as a covariate (see next section for details). We verified multicollinearity among variables included in the same model with the variance inflation factor (VIF). We assumed no multicollinearity when VIF scores were <5 (Zuur et al., 2010).

Correlations and Confounding Effects

We also examined the correlations between all variables considered in the candidate models using Pearson’s paired tests of correlations (cor.test in R, Supplementary Appendix 2). This was done prior to the modeling step for two reasons. First, some factors could not be tested in the same model because of over-parametrization issues due to the low number of mortalities (King and Zeng, 2001). Thus, the collinearity of these variables could not be assessed with multicollinearity tests. Yet, their correlation would indicate some redundancy among factors used to explain variation in mortality risk of caribou, potentially affecting the interpretation of results. Variables that were weakly correlated (<0.50) were tested in the same set of candidate models. When correlation was >0.50, we retained the variable providing the best fit according to the model selection process. Secondly, we suspected that exposure of caribou to human disturbances and predation risk, as well as the effect of weather conditions on survival, could vary substantially with latitude. Caribou are highly mobile, even outside migration periods, which could make exposure to anthropogenic and non-anthropogenic factors highly variable across individuals, or over time, even within a single season. For example, during summer caribou of the RFH were more exposed to disturbances when they reached the northern tip of the Ungava peninsula because of the aggregation of infrastructures in this confined area. Exposure to disturbances, and their effects on the mortality risk of caribou could thus be correlated with latitude. We could not directly test the effect of latitude on early life and seasonal survival because latitudes used by individuals varied markedly during these periods. To ensure that effects could be attributed to the tested variable and not the confounding effect of latitude, we explored the potential correlation between the variables and latitude. When the correlation with latitude was >0.50, we tested the effect of the variable on mortality risk despite potential confounding effects of latitude, but we interpreted the results accordingly. At the daily scale, we could better test the effect of latitude on mortality risk by including a latitude model in the candidate set to assess whether other variables explained more variability in mortality risk than latitude alone. Model performance at this scale of analysis could not be assessed with a likelihood ratio test based on a null model (see above). Instead, we used latitude as the null model for comparison.

Results

Human Disturbance Effects

Early Life Mortality Risk

For the early life period (1−7 years old), all models except the one including all disturbance types for the RGH satisfied the assumption of proportional hazards. In all cases, models indicated no effect of human disturbances on the mortality risk of caribou over the early life period, for both the RFH and RGH (Figure 3A and Table 3).

TABLE 3

Table 3. Summary of the effects of human disturbances (based on Cox-proportional hazard analyses) on early-life, seasonal and daily mortality risk of migratory caribou of the Rivière-aux-Feuilles (RFH) and Rivière-George herds (RGH), northern Québec and Labrador, Canada, 2009−2016.

FIGURE 3

Figure 3. Human disturbance effects [hazard ratio (HR), represented on a log scale] on caribou mortality risk at three temporal scales, early-life 1–7 years old; (A), seasonal (B) and daily (C) mortality risk, and 95% confidence intervals, on the Rivière-aux-Feuilles (RFH) and Rivière-George herd (RGH), northern Québec, Canada, 2009–2016. A hazard ratio < 1 indicates that exposure to disturbances decreased mortality risk, whereas a hazard ratio >1 indicates that exposure to disturbances increased mortality risk (red line = hazard ratio of 1). Disturbances tested included all disturbance types (mines, mining exploration, roads, and human settlements), industrial disturbances only (indus.; mine and mining exploration) and non-industrial disturbances only (non-indus.; roads and human settlements).