Introduction

National parks, protected areas, and other natural areas are the cornerstone of global biodiversity conservation efforts, but are increasingly attracting visitors [1, 2]. These areas are particularly attractive for nature-based tourism and outdoor recreation such as hiking, mountain biking, or motorized activities (e.g., all-terrain vehicles, snowmobiles) [3]. While such recreation activities can generate economic and health benefits, they can also affect various components of the environment, including terrestrial wildlife [4, 5]. Recreation activities can impact wildlife in multiple ways, including positive or negative effects on animal use of space, such as facilitated movement along trails [6] or displacement from important habitats [7]. The considerable variation in observed effects of recreation on wildlife has been attributed to multiple factors that can influence animal responses to outdoor activities (hereafter “influencing factors”) [8]. Tablado & Jenni (2017) [8] classified these factors into three categories (i) recreation characteristics (ii) spatio-temporal context (iii) intrinsic characteristics of the animals.

Influencing factors such as the location, intensity, and type of recreation can affect animal responses. For example, elk (Cervus elaphus) in North America were more sensitive to off-road than on-road all-terrain vehicles [9]. Similarly, in the United States, carnivore species were more sensitive to recreation (e.g., hiking and camping) in protected areas with higher levels of landscape disturbance than they were to recreation in protected areas with lower levels of disturbance [10]. Moreover, depending on the species, different types of recreation have also been found to have different impacts on wildlife, such as stronger effects from motorized vehicles and mountain biking than hiking or horseback riding, potentially due to elevated levels of noise and speed [11, 12]. Mammals perceive recreationists visually, auditorily, and olfactorily [8] and this perception can be modified by such varied factors as recreationists’ color of clothing (but see [13]) or speed of movement [8].

Animal perceptions and responses to recreation may also be influenced by the spatio-temporal context within which interactions occur. For instance, high-density vegetation can reduce noise pollution and direct lines of sight, and thus habitats with lower visibility or sound may be preferred by wildlife in recreational landscapes [14]. Similarly, the same species may also show different responses to recreation in different areas or seasons [15, 16]. For example, Barja et al. 2007 [17] found that European pine martens (Martes martes) were most sensitive to human and other environmental disturbances during the reproductive season. Mammals can also habituate to human presence and infrastructure, and accordingly display fewer responses to recreation in busier or more developed areas. For instance, Prices et al. (2014) [18] found that responses of mule deer (Odocoileus heminonus) to humans were reduced near a biological field station in Colorado. Habituation to disturbances is a common result of outdoor recreationists-wildlife interactions [19, 20]. It depends on the animal’s risk perception and time exposition to the disturbance: the more the animal is exposed to human the more it is habituated [21]. Habituation and risk perception can be altered if hunting occurs simultaneously with outdoor recreation, leading to challenges in disentangling the various impacts of human disturbance [22–24].

Similarly, variation in the ecological and life-history traits of mammal species can also influence their perceptions and tolerances of risk from human disturbance [8]. For example, mammal responses to stress differ depending on their trophic level. Prey display vigilance behaviour, or aggregate in large groups, in response to stress [24, 25], while such behaviours are less often observed in predators. Specifically, the difference between predator and prey responses to recreation has been increasingly investigated in the context of the “human-shield hypothesis” [26–29]. This hypothesis suggests that predators are more sensitive to, and thus avoid, people, while prey find refuge from predators by associating with human activities. Underlying mechanisms of such hypotheses are still unclear; however, they make clear the importance of understanding variation in responses to recreation among species and guilds.

Previous studies of recreation impacts on wildlife have often focused on the relationship between one focal species and one type of recreation (e.g., elk and skiers [30]; red squirrels and mountain bikers [31]). While this approach can inform single-species conservation or the management of one activity, it may be inadequate to capture broader community dynamics and the complexity of the cumulative effects of multiple activities on wildlife [32]. For example, Procko et al. (2023) [33] found that human presence had different impact on wildlife nocturnality depending on the species. Similarly, human disturbance can lead to cascading effects that can alter community structure and ecosystem functioning such as inter-species interactions [34, 35]. Long-term exposure to human disturbance can impact animal fitness (i.e., survival and reproduction success) [36, 37]. However, different outcomes of human disturbance impacting the long-term can be found. Salvatori et al. (2023) [37] found that both community and species-level occurrences increased within a protected area in Europe, despite human activities causing strong temporal avoidance in the whole community. This highlights the complexity of human-wildlife interaction and the necessity of long-term studies. Lastly, alterations in herbivory and predation resulting from recreational activities can have broader environmental implications through functional ecosystem change, as demonstrated in Di Nicola et al. (2023), where human-mediated changes were found to impact local carbon stocks [38].

Just as wildlife responses to recreation vary across species and contexts, approaches to managing the impacts of recreation vary considerably. Different types of recreation management have been used to limit the intensity of recreation disturbance, such as by limiting visitor numbers [39], or restricting the timing or location of certain recreation activities [40]. However, assessments of the effectiveness of such management actions across wildlife communities are often lacking. Thus, there is a need to evaluate how multiple species interact with different types of activities in different contexts to inform land and wildlife managers about strategies to promote human-wildlife coexistence.

Recent advances in sampling and statistical methodologies have created an opportunity to pursue community-level studies of the impacts of diverse recreation activities on wildlife. In particular, the development and popularization of motion-triggered cameras, hereafter camera traps [41–43], has allowed studies that focus on wildlife responses to recreation to capitalize on the wealth of multi-species information that is collected non-invasively [44]. Similarly, the development of joint species distribution models [45, 46], such as the Hierarchical Modelling of Species Communities approach (HMSC), has opened new doors for community ecology by modelling the abiotic environment at the community level while accounting statistically for co-occurrence among species [47, 48].

In this study, we combined camera trapping and community modelling to understand how the distribution and intensity of recreation activities influences space use by multiple terrestrial mammal species and determine what are the factors influencing mammal responses. Specifically, we used multispecies modelling of camera trap detections of small-, medium- and large-bodied mammals in two areas increasingly used by recreationists in western Alberta, Canada. We hypothesized that (1) the impacts of recreation will differ among mammalian predators and prey, and be influenced by the following factors: (2) trail designation, (3) habitat structure, (4) human footprint, (5) type of management, and (6) season. We made several predictions related to these factors. First, under the human shield hypothesis, we predicted that predator species will more strongly avoid recreationists, while prey species will be attracted to areas with more recreationists and fewer predators. Second, we predicted that trails designated for motorized activities will be more strongly avoided by mammals than trails designated for non-motorized activities, due to the elevated noise and speed of motorized activities. Third, we predicted that habitat characteristics can influence interactions between recreation and mammals, such that greater vegetation density (measured as percentage of forest) will provide more security and thereby reduce mammal negative responses to recreation. Fourth, we predicted that the percentage of human footprint on the landscape, such as roads and other infrastructure, can lead to greater habituation to people by mammals, and thus reduced responses to recreation. Fifth, we predicted that impacts of recreation will be reduced in areas where there are greater restrictions on recreation activity (e.g., reduced access). Sixth, and finally, we predicted that responses may differ between summer and winter since recreation activities and mammal ecology vary seasonally.

Methods

We evaluated how mammal space use changed in response to recreation and influencing factors based on camera trap detections and characterizations of recreation intensity and landscape conditions around camera stations, using the HMSC multi-species modelling approach.

Study areas

Our study was carried out across two recreational landscapes in western Alberta, Canada: Bighorn Backcountry (5,000 km²) and Castle Provincial and Wildland Provincial (1,050 km²) parks (hereafter Bighorn and Castle, respectively; Fig 1). Both areas have experienced a high level of recreation pressure over recent years due to their proximity to some of the most popular national parks in Canada (e.g., Banff, Jasper, and Waterton). There are more than 900 and 500 km of designated (official) trails in Bighorn and Castle, respectively, with similar trail densities in both (see Table 1). Summer recreation activities include hiking, camping, off-highway vehicle (OHV) use, fishing, hunting, cycling, and horseback riding, while winter recreation includes backcountry skiing, snowshoeing, and snowmobiling (https://open.alberta.ca, https://www.albertaparks.ca). Visitor numbers to these areas are not available; however, based on indices of recreation from the apps Strava and AllTrails (mostly used by runners, hikers, and cyclists), the level of recreation was higher in the camera trap sampling area within Castle relative to Bighorn (section 2.2; Table 1).

Download:

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

Fig 1.

Camera traps deployed in Bighorn (left, yellow points) and Castle (right, purple points) areas in Alberta, Canada (inset). Shading represents different types of recreation management: PLREC: Public Land Recreation, PLUZ: Public Land Use Zone, PP: Provincial Park, PRA: Provincial Recreation Area, WA: Wilderness Area, and WPP: Wildland Provincial Park. Base Map: OpenStreetMap: https://www.openstreetmap.org/copyright].

https://doi.org/10.1371/journal.pone.0300870.g001

Download:

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

Table 1. Predictor variables used in our HMSC model to estimate effects of recreation and landscape on mammal space use.

For continuous variables, we show the mean and range (minimum and maximum) of values estimated at the camera trap sites. For categorical variables, we show the number of categories and which category was used as a reference level using an asterisk (*). BH referred to Bighorn and CA to Castle. Sources of each dataset can be found in S1 File.

https://doi.org/10.1371/journal.pone.0300870.t001

The prevailing vegetation composition is similar in the two areas and dominated by coniferous forest. Elevation is also similar and ranges from 1000 to 3000 m, with higher, rockier alpine terrain in the west transitioning through foothills to lower forest, grassland and shrubland in the east. The region has mild summers (20 to 25°C) and cold winters (-5 to -15°C). Annual precipitation in the foothills of the Rocky Mountains in Alberta can reach 600 mm/year.

Results

The total sampling effort was 90,467 camera trap days, with 52,188 days in Bighorn and 38,279 days in Castle. We obtained 16,833 independent detections of 34 mammal species (see S1 Table), from which 554 detections were removed due to either a) an inability to identify animal detections to the species level or b) a species having too few detections to achieve adequate model convergence (<60 independent detections; see S1 Table). This left 15 species with a sufficient number of detections to model (i.e., model effective sample size ≥ ~100 and PSRF ~ 1 (approximate convergence; see S1 Table).

Relative importance of groups to explain mammals’ space use

The HMSC model explained between 10–65% of the variation in patterns of space use across camera sites for the 15 species (Fig 2, top). Recreation on its own–as measured by the three different indices (distance, density, Strava; Table 1)–explained an average of only 9% of the variation in space use across species (range = 3–13%; Fig 2, bottom), and no species showed significant positive or negative responses to individual recreation variables (i.e., 95% CI overlapped 0; S2 Table). However, interactions between recreation and the hypothesized influencing factors (Table 1) explained on average 45% (19–61%) of mammal space use—considerably more than recreation main effects. Thus, in combination, the main and influencing factors of recreation explained considerable variation (Fig 2, bottom). The interacting factors on their own explained on average 14% (4–22%) of space use (Fig 2), with species more likely to be detected in areas with stricter management for some species in winter (S3 Fig—e.g., mule deer). The additive landscape variables explained an average of 6% (2–12%; Fig 2), including negative associations with elevation and grassland-shrubland habitat for several species (S3 Fig—e.g., gray wolf).

Download:

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

Fig 2.

Variance explained (pseudo R², top graph) and proportion of explained variance (bottom graph) for each species. “Random” represents the variance accounted for by the station level random effect; “Effort” represents the variance accounted for by the camera deployment length; “Landscape” includes: Elevation, Distance to water, Land Cover and NDVI; “Influencing Factors” includes the trail designation, % forest, % human footprint, Management Type, and Season; “Recreation” includes the main effects of distance, density and Strava; “Recreation interaction” includes the interaction effects between each factor of “Recreation” and “Influencing Factors”.

https://doi.org/10.1371/journal.pone.0300870.g002

Influencing factors of mammals’ responses to recreation

We first looked at the interactions between the group of recreation measures (distance, density and Strava) and hypothesized influencing factors (Table 1). The interaction with management type explained the most variation in species space use: an average of 19% (10–30%; Fig 3A). The trail designation (motorized or non-motorized) also had an important interaction with recreation activity in influencing mammal space use, accounting for on average 14% (5%-21%) of variation. The other variables (percentage forest, percentage human footprint, season) generally had weaker interactions with recreation, accounting for an average of less than 5% of the total variation across species (Fig 3A).

Download:

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

Fig 3. (above) Proportion of total variation in species detections.

The proportion of total variation was obtained by multiplying the variation partitioning by the corresponding pseudo R2 for each species (see Fig 2). In a) we combined the proportion of total variation of Strava, distance, and density with interaction factors. In b), c) and d) we showed, respectively, the results of the density, distance, and Strava measures in interaction.

https://doi.org/10.1371/journal.pone.0300870.g003

Trail density

At the finer model resolution of individual recreation variables, trail density explained the most variation in species space use. Trail density explained on average 11% (3–24%) of variation across species (Fig 3B) when considering both main and interactive effects. The strongest interactions were between trail density and a) management type and b) trail designation (Figs 3–5). For most mammal species (excepting mule deer and ground squirrel) there was a negative interaction between trail density and management type, such that mammals were less likely to use sites with higher trail density in wilder areas with more restricted recreation (i.e., provincial and wilderness park lands; Fig 5A). For most species, there was a positive interaction between trail density and trail designation, although the specific responses to motorized and non-motorized trails varied among species, resulting in a lack of clear signal (Figs 4 and 5B). Some species increased their use of higher trail density areas when the trail designation was motorized activity (e.g., wolf, cougar, snowshoe hare), some when the trail designation was with non-motorized activity (e.g., wolf, black bear, white-tailed deer), and some did not show this effect. For moose, use of areas with higher trail density was higher with increasing forest percentage, whereas for lynx and red squirrels, it was higher with increasing human footprint (Fig 4).

Download:

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

Fig 4. Subset of species responses to recreation with interacting factors.

We only show responses for interactions that were significant for at least one species. Effect coefficients with at least 95% posterior probability above zero are shown in red with a “+”, or below zero in blue with a “-“. Predators are in orange, ungulates in green and small mammals in pink. Vertical lines separate the different recreation measures. DT = Designation Trail, HF = Human Footprint. For full estimates and confidence intervals, see S3 Fig.

https://doi.org/10.1371/journal.pone.0300870.g004

Download:

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

Fig 5.

Conditional effects of a) density in interaction with the type of management. Values of management reflect the level of wilderness and recreation management in our study areas: public lands (value 1), Public Land Use Zones (PLUZs,) PLREC Public Land Recreation (PLREC), Provincial Recreation Areas (PRAs) (value 2), Provincial Park (PP) value 3 and Wildland Provincial Park (WPP) (value 4), b) density in interaction with the type of recreation, c) distance in interaction with the season, and d) Strava in interaction with non-motorized activities. Species represented with a response with at least 95% posterior probability (see Fig 3).

https://doi.org/10.1371/journal.pone.0300870.g005

Distance from trail.

The distance to the nearest trail explained on average 4% (1–9%) of variation in species detections, when considering both main and influencing factors (Fig 3C), with the strongest interactions with season. Space use for seven species showed a negative effect of interaction between distance and season (Fig 4): wolf, lynx, black bear, grizzly bear, moose, white-tailed deer and mule deer were found closer to trails during winter relative to summer (Fig 5C). Mule deer occurred further from trails where there was more forest, and closer to trails in areas with more restrictive management (e.g. park lands vs. public lands). Coyote occurred further from trails where there was more human footprint (Fig 4).

Recreation intensity (Strava).

The intensity or frequency of recreation activity around a camera site, as measured by our Strava variable, explained on average 3% (2–5%) of variation in species detections as a main effect and in interaction (Fig 3D). The interaction between Strava and the designation of trail was strongest. Specifically, lynx, black bear, grizzly bear, and mule deer, showed more negative responses to higher levels of Strava use in areas with trail designation for non-motorized activities (Figs 4 and 5D). Moose were more positively associated with Strava use in areas with higher human footprint. However, it is worth noting that Strava was only a measure of the intensity of non-motorized activities.

Discussion

There have been numerous studies examining the effects of outdoor recreation on wildlife, with many different types of responses being documented [70–73]. Our study aimed to elucidate not only the responses of multiple mammal species to recreation in two partially protected regions in western Alberta, but also the degree to which responses were influenced by factors such as local management restrictions, trail designations, and environmental conditions. Using a hierarchical Bayesian community-level modelling approach, we found that mammals did not have consistently strong negative or positive associations with measures of recreation distribution and intensity, but that recreation interacted with other influencing factors to explain considerable variation in patterns of space use across multiple species. More specifically, we did not find evidence that predator and prey species consistently differed in their responses to recreation (hypothesis 1). We also found that the trail designation (motorized vs non-motorized) influenced species responses to recreation pressure, but that there were not consistent differences in responses across species (hypothesis 2). We also found that most mammals were less likely to use sites with higher trail density when they occurred in areas with more restrictive management of recreation activities (e.g., parks and wilderness parks; hypothesis 5).

Our two study areas—the Bighorn and Castle regions—are both experiencing increasing amounts of visits by recreationists (while there is a control over the number of trails), raising concerns about potential negative effects on wildlife. We did not directly compare recreation impact on mammals’ space use in the two different areas, but instead used other landscape and influencing factors which are specific to these two areas (e.g., recreation management). It is encouraging that the camera trap surveys detected a diverse mammalian assemblage across these regions, and that we did not detect strong negative associations with recreation variables in regard to a possible avoidance of site with increased recreation pressure. It is also interesting that mammal responses to recreation appeared to be influenced by recreation management.

Our sampling covered a gradient of recreation management types across the two study areas, from public lands and recreation areas easily accessible to recreationists, to more remote park lands with a greater emphasis on protection of nature. We identified an interaction between the potential recreation activity at a sampling site—measured as trail density within a 500m circular buffer around a camera—and the recreation management, such that sites with a higher density of trails were less used by mammals in the wilder park lands. In such areas, mammals are generally less likely to encounter recreationists in the broader landscape, and may thus be more sensitive to, and accordingly avoid, localized sites where more recreation occurs [54]. By contrast, mammals in areas where recreation activity is more ubiquitous may become habituated to humans and less likely to show this avoidance response [15]. Habituation and avoidance behavior can be altered by hunting [24, 74], a factor not accounted for in our study due to a lack of information on spatial variation in hunting pressure. However, previous studies have demonstrated its influence on species responses to recreationists [22, 24].

We only detected an interaction between recreation and management for trail density, and not for distance to trail or our Strava indicator of recreation activity, with the exception of mule deer which were found closer to trails in park lands (vs public lands). The physical presence of multiple trails in these remote areas might represent a visual cue that leads to avoidance (or attraction) by wildlife [75]. We attempted to incorporate visibility into our models by using the percentage of forest within a 500m buffer around a site, but this did not influence most species responses to recreation and may have been too coarse a measure. Moose and mule deer were exceptions, but with two different responses, with moose using areas with higher trail density more often when there was more forest cover (consistent with the notion that cover provides security), while mule deer showed higher use of areas further from trails where there was more cover (potentially consistent with a preference for longer sightlines to avoid human disturbance). We recommend further research into understanding how recreation management interacts with recreation activity, and in turn with forest cover, to affect wildlife. This has important implications for management policies, including whether and where to prioritize actions that limit recreation infrastructure (e.g., trail density) vs. those that limit recreation activity (e.g., number of hikers allowed in an area per time period).

Interestingly, we did not find a strong effect of percentage human footprint on most focal mammal species, either on its own or as an interaction with recreation (hypothesis 4). There were some exceptions where significant interactions between percentage human footprint and recreation were identified in our model. Lynx and red squirrels used areas with higher trail density more often when there was more human footprint, and similarly moose were positively associated with recreation activity measured by Strava in areas with more human footprint. Both patterns are consistent with habituation to recreation in more developed areas. However, in the case of red squirrels, the lack of avoidance might indicate an inability to avoid disturbance rather than habituation to stress and recreation due to their small home range. By contrast coyotes occurred at sites further from trails in areas with higher footprint, suggesting they may have been warier of people in more developed areas. Other studies have concluded that the amount of human footprint on a landscape is a major driver of species space use [76]. The mixed results in our study highlight the complexity and variation in different species responses to the same landscape contexts [33], but the somewhat limited responses to footprint may also reflect the fact that our sampled landscapes had relatively low levels of development overall. We recommend that future research extend our efforts by applying standardized camera trapping sampling across a larger number and gradient of landscape contexts [77].

As expected, the type of recreation activities permitted in an area around sampling sites (i.e. by trail designation) influenced mammal responses to recreation. We identified an interaction between recreation activity level (measured by Strava) and non-motorized trail designation for four species (lynx, black and grizzly bear, mule deer), such that they used areas with higher Strava activity less when nearby trails were designated non-motorized. However, this result may be due to the fact that our Strava data primarily represents non-motorized activity. On trails that allow motorized activity, Strava data may not reflect recreation intensity as well, which may explain why there was not a significant interaction between Strava data and presence of trails designated as motorized. Thus, our finding may not actually indicate that non-motorized recreation has a stronger effect on these four species compared to motorized activity, but rather that we do not have an accurate representation of motorized recreation intensity in our dataset. Motorized activities are predicted to have a stronger impact on wildlife due to their higher levels of noise and speed [15], but recent reviews have reported that evidence of negative effects on wildlife from non-motorized activities is stronger than for motorized [70]. Motorized activities can have other impacts on the environment, such as soil erosion [78], and may cause impacts at larger spatial scales [70, 79]. They may also create wider travel corridors that are selected by some species (e.g., wolves, [80]). Thus, at local scales, non-motorized activities might affect behavioral responses but at landscape scales motorized activities are likely to have more impact on wildlife. In future research, it may be important to distinguish recreation activities at a finer resolution and obtain accurate estimates of motorized and non-motorized activity intensity, for example using Strava data it is possible to distinguish biking and running activities. Naidoo & Burton (2020) [12] found that mountain biking caused as much disturbance to wildlife as motorized activities, and more than other non-motorized activities like hiking and horseback riding. They also found that wildlife displacement could occur at finer temporal scales, and other studies have documented differences in wildlife responses to recreation during the day vs. at night [23, 81], such that areas with more recreation activity may be avoided during the day but used for movement under the cover of darkness [82, 83]. We recommend that future studies consider both spatial and temporal responses by wildlife and assess recreation activities based on the attributes that may underlie their impacts, such as speed, noise, and visibility.

Contrary to our prediction, we did not identify strong or consistent differences in responses to recreation between predator and prey species (hypothesis 1). The human shield hypothesis predicts that prey will be found close to human activity to protect themselves from predators which avoid human activities [26, 27]. We did not find evidence to support this hypothesis, as prey species were not consistently associated with our measures of recreation, nor did predators consistently avoid recreation. Predator-prey interactions are complex and human activities can lead to cascading effects on species interactions due to the impact on individual fitness and change in population demographics [29]. Similarly, predator-prey interactions are dynamic in time [84], which our study did not investigate as we only considered seasonal spatial patterns and not fine-scale temporal patterns. With more direct measurement of recreation activity—such as with camera traps placed on recreation trails, or using acoustic soundscape monitoring—finer-scale analytical methods such as avoidance-attraction ratios can be used to assess interactions between humans, predators, and prey [12, 73, 85]. Our results did suggest that smaller mammals were generally less responsive to recreation, despite indications from elsewhere that space use by small mammals may be affected by recreation, particularly due to snow compaction in alpine areas during winter [86]. Similarly, we did not conduct a fine-scale temporal analysis that would enable the identification of specific times when trails and areas with high recreation intensity were used by mammals. Mammals’ space use in high recreation intensity areas could result from a shift in their diel activity cycle, avoiding recreation during the day and utilizing the trail network at night [23, 87].

We were unable to directly measure recreation in our study areas (since camera traps were not placed on recreation trails), but rather used multiple indirect measures of potential recreation pressure on wildlife habitats. Previous studies have often only used one measure of recreation, such as the presence or number of recreationists, or the times of their peak activities [88]. We found interesting differences in species associations with the multiple measures of recreation. For instance, more variation in mammal detections was explained by an interaction between trail density and management type than by an interaction between distance to trail and management type. In contrast, an interaction between distance to trail and season explained more variation in space use for predators and ungulates, with these species being found closer to trails during winter (hypothesis 6). This result aligns with those of Gese et al. (2013) [6] and Bunnell et al. (2006) [89] that suggested recreation trails can facilitating animal movement in deep snow (e.g. trail compaction by skiing, snowmobiling [82]). It may also reflect more avoidance of trails by mammals in summer, when trails are busier, or more sensitivity due to reproduction statue (e.g., presence of young). Ultimately, the different patterns that we observed for different indirect measures of recreation could represent true variation in animal responses to different components of a “recreation landscape”, or they could reflect noise in the relationships between the measures and the true underlying aspects of recreation to which animals are responding. Eventually, more research is needed to test and compare different indicators of recreation pressure, including cost-effective indirect measures like those used in our study, and more labour intensive but direct monitoring of recreation activity, such as camera traps. Validation of indirect methods (e.g., Strava use) with direct ground-based monitoring is desirable to better understand their accuracy and limitations in different contexts, and critical before relying on them as a primary monitoring tool. Encouragingly, the more direct monitoring methods are being aided by the incorporation of machine learning tools, for example to process of camera trap images [85]. However, validation will require more purpose-designed sampling. For example, in our study we could not directly validate the Strava data with camera traps since cameras were placed off recreational trails. Likewise, we recommend that future research integrates various measures of recreation into a unified metric to comprehensively grasp how the spatial distribution (e.g., density) and the intensity of recreation (e.g., Strava) collectively influence mammals’ responses. Similarly, the proportion of trails used for a specific activity might affect the nature and strength of species responses to recreation.

There are additional opportunities to build on our study in future research. We used a community-level analytical approach to evaluating multispecies responses to recreation, in order to move beyond the inefficiencies of single-species case studies. However, the computational complexity of this modelling approach necessitated that we remove less common species from our analysis. These rare and elusive species, such as wolverine (Gulo gulo; [90, 91]), are usually of conservation interest but are difficult to assess in individual studies due to low detection rates. Continued monitoring over the long term will increase sample sizes for such species and confirm their continued presence—or conversely their extirpation from—recreation landscapes. Camera trap networks such as in National Parks and across ABMI sites can help meet this aim [92]. Similarly, these networks can aid in studying interactions between recreation and mammals at a large spatial scale. At an even larger spatial scale, comparing the impacts of recreation on mammals across continents that vary in the intensity and type of recreation activity (e.g. North America vs. western Europe) could potentially enhance our overall understanding of this subject. There are additional factors that may affect wildlife responses to recreation that we did not consider in this study. For example, species risk perception has been shown to depend on their long-term relationship with humans [93] and degree of habituation to human presence [25]. This phenomenon is difficult to quantify [44] but is a key factor in understanding the responses of species (and populations) to recreation. Other species traits such as relative brain size (as an index of cognitive ability [94]) and habitat or dietary breadth [7] could also influence species sensitivities to recreation. Such traits could be incorporated into modelling frameworks such as the one we used [69] in attempts to build on our work and explain more variation in species responses to variation in recreation activity.

Conclusion

Recreation activity is increasing globally [1, 2], and there is a need to develop management strategies that support the benefits of recreation while limiting any negative impacts on wildlife populations [95, 96]. Our results suggest that mammal communities are not transversally impacted by recreation activities (i.e., there were not strong or consistent responses across species), and that observed impacts were not easily decomposed into base ecological differences such as between predator and prey guilds. More research is needed to understand whether and how species traits influence sensitivity to recreation. Such knowledge could help guide multispecies management approaches, or identify where species-specific approaches are needed.

When studying the impacts of recreation on species, we must not only look at the sources of recreation disturbance, but also consider how their effects may be influenced by different anthropogenic and landscape contexts, some of which have direct linkages to management decisions. We observed that space use by a community of mammal species in the Rocky Mountains of western Alberta was influenced by an interaction between recreation intensity and the type of recreation management, such that mammals showed more avoidance of sites in the vicinity of higher trail density within areas managed for wilderness values. We suggest that limiting the overall density of trails in these protected areas, or aggregating trails within a small portion of protected areas, may help to reduce conflict between recreationists and mammals. We also recommend more direct monitoring of recreation and its potential mechanisms directly affecting wildlife, such as frequency, noise and visibility. While behavioural responses by wildlife can provide early warning of impacts, we recommend linking these to monitoring of population-level responses, such as changes in survival or reproduction, to understand long term effects of human disturbances [36, 37]. Only by linking rigorous scientific monitoring to recreation management will we more fully understand the prospects for human-wildlife coexistence within the world’s rapidly changing landscapes [97].

Supporting information

Acknowledgments

We thank Jason Fisher and Joanna Burgar for their important roles in the development of the research questions and methods. We thank members of the project steering committee for their support and feedback (Dallas Johnson, Dan Farr, Bonnie Drozdowski). Many staff at Innotech Alberta contributed to data collection and management (including Jason Fisher, Luke Nolan, Colin Twitchell, Greg Brooke). Members of the WildCo lab at UBC provided feedback on this research. Members of Jason Fisher’s ACME lab at the University of Victoria also contributed to field work. We thank Don Livingston (Alberta Environment and Protected Areas), Dragomir Vujnovic (Alberta Public Lands), Courtney Hughes (Alberta Environment and Parks), Graham Wylde (Alberta Environment and Protected Areas), Wonnita Andrus (Nature Conservancy of Canada), Brad Tucker (Alberta Environment and Parks), and Chad Willms (Alberta Environment and Parks) for sharing knowledge about the Bighorn and Castle areas. Chris Beirne and Patrick Thompson provided advice on statistical modelling.