Introduction

Predators impact prey through two key processes, firstly by directly killing and removing individuals from the population and secondly through the indirect effects of predation risk that result in prey species modifying their behavior [1], [2]. These non-lethal effects have a strong influence on prey fitness with evidence suggesting substantial impacts at the population level possibly equal to or greater than the removal of individuals through direct predation [3]–[5]. Furthermore, the risk of predation varies across time and space with herbivores constantly balancing foraging effort against the need for safety from predators [6]–[9]. Prey species therefore inhabit ranges of shifting predation risk that has been termed the ‘ecology or landscape of fear’ [2], [10], [11]. The landscape of fear is specific to the prey species and will depend on the predators to which they are exposed, the encounter rate, predatory defense and the effectiveness of vigilance [1], [2], [12].

Recently, there has been considerable interest in the role of predation risk in regulating ecosystems through trophic cascades [3], [5], [13], including most notably the reintroduction of wolves (Canis lupus) to Yellowstone National Park in 1995 [10], [11], [14]. Interestingly, human disturbance in natural and protected areas has been shown to affect wildlife populations in a similar manner to predation risk, including greater rates of movement and vigilance, reduced foraging, shifting habitat use and increases in group size [15]–[19]. These behavioral responses result in potential fitness costs as a result of increased energy expenditure, loss of foraging opportunities and the direct impact of physiological stress on reproductive success and survival [15], [20]. However unlike predation risk, anthropogenic disturbance in protected areas exhibits greater spatial and temporal predictability, being largely a function of visitors using defined park infrastructure (e.g. roads and trail networks) during daylight hours.

Large-bodied carnivores such as wolves and bears often demonstrate the greatest sensitivity to human presence [17], [21], [22], whereas a number of large herbivore species have exhibited much greater tolerance of human activities and infrastructure [19], [23]–[25]. A primary reason why herbivores are predicted to exhibit greater tolerance to human presence in natural areas is that it provides a potential shelter from risk-sensitive carnivores. For example, female moose (Alces alces) in the Yellowstone Ecosystem showed a distinct preference for areas close to roads during their highly synchronous calving period because these zones functioned as refugia from human-averse brown bears (Ursus arctos) [26]. Meanwhile, two recent North American studies demonstrated that large herbivores were more likely to use areas in close proximity to roads and walking trails compared with carnivore species, and by doing so were experiencing a spatial refuge from potential predators [27], [28]. Disturbance driven by human activities might therefore alter community-level interactions between predators and prey [22], [29].

Although research has begun to indicate that ungulate species may select areas that have elevated levels of human use, much less attention has been focused on the specific behavioral outcomes associated with potential predator shelters. We now use a four-year behavioral dataset collected from the direct observation of elk (Cervus elephus) and pronghorn (Antilocapra americana) to explore the potential refuge effect generated by tourist infrastructure in Grand Teton National Park (GTNP). Based on the ‘predator shelter hypothesis’, we predicted that the Teton Park Road in GTNP, which is associated with significantly higher levels of traffic and human activity than our comparison site, the quieter River Road, would provide ungulates with the opportunity to trade off human disturbance against a shelter from predation. More specifically, we predicted that enhanced predation risk at the River Road would result in pronghorn and elk forming larger groups due to the benefits afforded by safety in numbers [30]–[32]. The specific behavioral responses mediated by the predator shelter were predicted to include an increase in the proportion of individuals feeding and a reduction in the proportion of animals engaged in alert behavior [2], [6], [7], concomitant with the lower perceived levels of predatory risk.

Materials and Methods

Results

Traffic levels differed significantly between the two sites (β-estimate  = 1.21, 95% CI  = 1.07/1.35) with an average of 2.6 (±0.11 SE) vehicles per observation at the Teton Park Road and 0.1 (±0.02 SE) vehicles per observation at the River Road. During the four-year study, 1040 behavioral scans and 6553 animal observations (Teton Park Road: 3707 elk, 1061 pronghorn; River Road: 1226 elk, 559 pronghorn) were conducted on 493 groups of elk and 547 groups of pronghorn. On average per road survey, we observed 19.7 (±SE 1.9) elk in 2.2 (±0.1 SE) groups and 4.3 (±0.2 SE) pronghorn in 1.7 (±0.1 SE) groups along the Teton Park Road, and we observed 37.2 (±6.3 SE) elk in 2.2 (±0.2 SE) groups and 10.3 (±1.1 SE) pronghorn in 2.4 (±0.2 SE) groups along the River Road. The average group size was smaller at the Teton Park Road than the River Road for both pronghorn (β-estimate  = −0.49, 95% CI  = −0.59/−0.39) and elk (β-estimate  = −0.65, 95% CI  = −0.93/−0.35; Figure 1a).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Comparisons of the average group size (a), and proportion of pronghorn and elk engaged in feeding (b), alert behavior (c) and movement (d) at the Teton Park Road and the River Road. Data are presented as means ±95% CI.

https://doi.org/10.1371/journal.pone.0094630.g001

The results of the behavioral scan analysis are presented in Table 1 for pronghorn and Table 2 for elk. We identified three top models predicting pronghorn feeding behavior, accounting for 0.95 of the AICc weight (Table 1). The β-estimates and confidence intervals extracted for each parameter in the best models revealed that pronghorn, as predicted, were more likely to feed at the Teton Park Road compared to the River Road (27% increase in proportion feeding, see Figure 1b). Pronghorn also fed more when in larger groups (Table 1). Moreover, the interaction between site and distance to road demonstrated that pronghorn were more likely to be observed feeding close to the Teton Park Road (<200 m) compared to distances further away (200–500 m and >500 m distance categories). At the River Road the opposite effect was observed, with greater levels of feeding further from the road (Table 1). The main effect of distance to road suggests an apparent overall increase in feeding >500 m from a road, however this appears to be an artifact of the interaction as demonstrated by the lack of an effect in a simple additive model that did not include an interaction term (β-estimate_200-500m  = −0.03, 95% CI  = −0.41/0.34; β-estimate_>500m  = −0.03, 95% CI  = −0.57/0.52).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Top models accounting for 0.95 of the AICc weight in the pronghorn behavioral analyses.

https://doi.org/10.1371/journal.pone.0094630.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Top models accounting for 0.95 of the AICc weight in the elk behavioral analyses.

https://doi.org/10.1371/journal.pone.0094630.t002

For pronghorn alert behavior, seven models accounted for ≥0.95 of the AICc weight and were within 3 AICc scores (Table 1). As predicted, pronghorn were less likely to be alert at the Teton Park Road compared to the River Road (18% reduction in proportion alert, see Figure 1c). The interaction between site and distance to road was driven by alert behavior increasing markedly >500 m from the Teton Park Road. In the pronghorn movement behavior analysis, four models within five AICc scores accounted for ≥0.95 of the AICc weight (Table 1). Only site had β-estimates with confidence intervals that did not overlap zero, indicating that pronghorn were more likely to be moving at the River Road compared with the Teton Park Road (30% increase in proportion moving, see Figure 1d).

Model selection for the analysis of elk feeding behavior generated four top models within seven AICc scores (Table 2). Consistent with predictions, elk were more likely to feed at the Teton Park Road in comparison to the River Road (21% increase in proportion feeding, see Figure 1b). For alert behavior, six models accounted for 0.95 of the AICc weight and were within four AICc scores (Table 2). As predicted, elk were less alert at the Teton Park Road compared to the River Road (Table 2), however the proportionate data demonstrate that this 9% reduction was comparatively weak (Figure 1c). Elk were more alert >500 m from the Teton Park Road compared to the closest distance category, while the converse relationship was evident at the river road (Table 2). Similar to the pronghorn feeding analysis, the inclusion of the interaction term is likely driving the main effect of alert behavior decreasing with distance, as demonstrated by the reduced strength of the effect in a simple additive model (β-estimate_200-500m  = −034, 95% CI  = −0.71/0.03; β-estimate_>500m  = −0.43, 95% CI  = −0.80/−0.06). The results for movement behavior of elk included six top models that accounted for 0.95 of the AICc weight and were within four AICc scores (Table 2). This analysis did not reveal any site-specific differences in movement (Table 2, Figure 1d), with the null model receiving the most support from the data.

Discussion

Our results suggest that human activity can alter ungulate behavior in accordance with the predator shelter hypothesis. More specifically, as predicted under the predator shelter hypothesis, elk and pronghorn formed smaller groups, were more likely to forage and less likely to be alert closer to the heavily used Teton Park Road compared with the nearby quieter River Road. Furthermore, pronghorn foraged less at distances greater than 200 m from the Teton Park Road, while both species exhibited increased alert behavior when greater than 500 m from the Teton Park Road. Interestingly, 500 m is commonly cited in the literature as a threshold distance at which large predators (e.g., wolves and bears) avoid roads [36]–[38]. Pronghorn also demonstrated lower levels of movement at the Teton Park Road study site. These behavioral differences suggest that in contrast to the River Road, the area immediately adjacent to the Teton Park Road is of considerably lower perceived risk, despite the increased exposure to human activity and disturbance. In addition to effects of site, we also detected effects of group size on behavior, as pronghorn, and to a much lesser degree elk, appeared to feed more when in larger groups, suggesting potential fitness benefits from aggregating with conspecifics [39], [40]. Group size, however, was not a key explanatory variable in the analyses of alert or movement behavior for either species.

Our finding that elk did not exhibit the same pronounced increase in alert behavior, or indeed greater movement at the River Road, as demonstrated by pronghorn, is consistent with known differences in anti-predator behavior between the species. Elk are predominantly susceptible to predation from wolves and to a lesser extent coyotes and bears, and compared with the smaller-bodied, highly-responsive pronghorn, are less likely to respond or take flight when faced with perceived threat [41]. Previous studies have in fact demonstrated that elk exhibit highly variable behavioral responses to wolves that have been dependent on site, sex and physiological condition [33], [39], [40], [42]–[44], which could explain the comparatively variable effect of site in predicting alert behavior among groups of elk.

Reduced anti-predator behavior, as evident by diminished alert behavior and greater feeding along the Teton Park Road, concurs with Brown et al. [24] who found that ungulates did not consistently respond to human disturbance as a form of predation risk in GTNP. Earlier research has demonstrated that ungulates have the capacity to habituate to the presence of humans - even in urban environments - if exposed repeatedly to predictable, low-risk human activities [19], [22], [23], [25]. However, if reduced anti-predator behavior along Teton Park Road was driven solely by habituation or tolerance to human disturbance, we would expect to find similarly reduced responsiveness across sites with and without humans, particularly as the roads are in relatively close proximity to one another and the study animals can likely move freely between them. The greater responsiveness of ungulates along the River Road is therefore not entirely consistent with the idea of habituation along the Teton Park Road. One further explanation could be that ungulates do in fact perceive human activity as a predatory risk, but they cannot maintain continuous levels of alert behavior associated with the chronic disturbance experienced at the Teton Park Road [24], as would be predicted if animals were making risk allocation decisions [45], [46]. Yet if this were so, then we would have expected that periods with greater vehicle traffic at the quieter River Road would have resulted in greater responsiveness and reduced foraging. No such relationship, however, existed for either pronghorn or elk.

Our results also concur with recent studies that demonstrated disparities in the spatial distribution of predator and prey species as a function of human presence [26]–[28]. Whilst the Teton valley supports a variety of large predator species, during at least two years of the study (2007 & 2008), wolves denned within approximately 1 km of the River Road (Sarah Dewey, Wildlife Biologist GTNP pers. comm.), likely elevating their activity and increasing predation risk in the vicinity. Large carnivores are likely to avoid habitats in close proximity (<500 m) to comparatively busy roads such as the Teton Park Road that experience consistent levels of vehicle traffic throughout the daylight hours [38], [47]–[50]. However, it is important to note that we were unable to address the proportion of each species susceptible to predation along the two sections of road. Meanwhile, ungulate species can adjust their behavioral responses to human disturbance according to the degree of risk that they experience across both time and space [24], [25]. A recent study demonstrated that the highest level of vigilance exhibited by elk was associated with the hunting season on private and public lands, whilst the lowest was found in a neighboring national park during the summer when human presence was elevated, but the associated activities were relatively benign [18]. Human impacts on wildlife in natural and protected areas is therefore highly context-dependent and can vary dramatically depending upon the type of activity, predictability in patterns of use, and the behavioral sensitivity of species that are likely to be affected.

We conclude that the most parsimonious explanation of our results is that ungulates in GTNP can use predictable human presence along the Teton Park Road as a potential refuge from predation risk and adjust their behavior accordingly, further demonstrating the important conservation and management implications of understanding behavioral responses of wildlife to anthropogenic disturbance [51], [52]. These refuges are potentially strong attractors for prey populations and thus have the potential to drive trophic-mediated effects similar to those proposed by the reintroduction or extirpation of large carnivores [10], [11], [22]. Human activity could tip the “space-race” balance in favor of prey over predators and reduce the area of habitat that is available to large carnivores such as bears and wolves that are typically human-averse [17], [21], [27], [28]. Meanwhile herbivore populations could potentially increase population growth due to reduced predation, resulting in elevated impacts on sensitive habitats [22]. With an ever-growing demand for access to natural and protected areas, the indirect effects of human presence on wildlife will become more prevalent and it is essential that these impacts be better understood in order that effective conservation management can be balanced with the visitor experience.

Acknowledgments

We would like to thank Grand Teton National Park, including S. Cain and S. Dewey, for permission to conduct this research and logistical support. e also thank R. Blaskovich, C. Brown, D. Ellsworth, D. Kinka, K. Knighton, K. Martinez, M. Scott, T. Sullivan, and K. Van Ort for assistance with data collection. We are grateful to A. Merenlender and three anonymous referees for their helpful suggestions for revisions, which greatly improved the paper.