In predator-prey systems, prey show anti-predator behaviors such as vigilance and altered foraging behavior to reduce the risk of being preyed upon (Ydenberg and Dill, 1986; Cooper and Frederick, 2007; Laundre et al., 2010). Detection is the first step in prey’s response to a predator (Bednekoff and Lima, 1998). Predation risk vary in time and space and vigilant behavior likely corresponds with the risk perception (Lima and Bednekoff, 1999; Gaynor et al., 2019). In low risk situations animals might spend less time on vigilant behavior and increase time feeding or resting (Lima and Bednekoff, 1999). In resting situation animals might select for protective cover, but potentially with an overview to detect any risks (Lazarus and Symonds, 1992).

Once a prey has detected a predator, it has four basic response options: flight, fight, hide, or freeze (Lima and Dill, 1990; Rupia et al., 2016; Roelofs, 2017). Different responses come with different energetic costs, which together with the perceived severity of risk (Cooper and Frederick, 2007) affects the response. Optimally, when the potential risk of staying exceeds the costs of fleeing, the animal should flee (Ydenberg and Dill, 1986; Lima and Dill, 1990; Cooper and Frederick, 2007). Additionally, the response in a given interaction can be affected by the animal’s personality and previous experience (Beale, 2007; Rupia et al., 2016; Found and Clair, 2018; Carricondo-Sanchez et al., 2020). Various disturbance intensities may affect the intensity and duration of the animal’s response (Beckmann et al., 2004; Ordiz et al., 2013; Petracca et al., 2019; Suraci et al., 2019).

Anti-predator behaviors are not limited to prey species. Even top predators might display similar behaviors to avoid intra-guild aggression (Holt and Polis, 1997; Swenson et al., 2001; Frid and Dill, 2002; Mech and Boitani, 2003; Wikenros et al., 2017) and as a response to human-induced disturbances (Gill et al., 1996; Frid and Dill, 2002; Moen et al., 2012). For wolves (Canis lupus), encountering humans is still not without risk. Lethal control, poaching, and traffic collisions are main sources of wolf mortality (Colino-Rabanal et al., 2011; Liberg et al., 2012, 2020; Recio et al., 2018), and human-related mortality currently limits the population growth of wolves in Europe (Kuijper et al., 2019; Liberg et al., 2020; Sunde et al., 2021). Hence, human-caused disturbances are expected to result in anti-predator behavior due to a potentially lethal risk for the wolf (Frid and Dill, 2002; Ordiz et al., 2011). In fact, even though wolves and other predators are known to make use of human-made structures (e.g., roads and bridges) (Blanco et al., 2005; Zimmermann et al., 2014; Dickie et al., 2017; Bojarska et al., 2020), they tend to avoid human activities, resulting in spatiotemporal segregation between wolf and human activities (Lesmerises et al., 2012; Milleret et al., 2019; Carricondo-Sanchez et al., 2020). The avoidance of human activity may have been shaped by the century-long history of wolf persecution by humans, as suggested for the brown bear (Ursus arctos) in Europe (Zedrosser et al., 2011).

Responses to approaching humans have been studied previously in wolves using information from VHF-collars (Karlsson et al., 2007; Wam et al., 2012, 2014), and more recently in brown bears using high-frequency GPS data (Moen et al., 2012, 2018; Ordiz et al., 2019). In wolves, flight initiation distance (FID, i.e., the distance at which an animal flees from an approaching threat) was affected by wind conditions, with shorter FID when the wind was blowing away from the wolf, but not by horizontal visibility at the wolf’s location (Karlsson et al., 2007). In contrast, reduced horizontal visibility at brown bear resting sites resulted in shorter FIDs when approached by humans (Moen et al., 2012). This indicates that the bear either made the decision to wait longer before fleeing (Beale, 2007; Cooper and Frederick, 2007), or it did not detect the observer (Moen et al., 2012), illustrating that the time of flight initiation does not necessarily equal the time of detection. After being disturbed by a human, VHF-collared wolves selected more concealed locations with lower horizontal visibility (Wam et al., 2012), and a recent study found that wolves select for more concealed resting sites during the day in response to increased human disturbances (Bojarska et al., 2021). Horizontal cover may benefit wolves more in terms of concealment more than it hampers their vigilance. Wolves have well developed auditory and olfactory systems, which may be more important in detection of threats over longer distances than visual detection (Mech, 1970; Harrington and Asa, 2003).

With this paper, we aimed to describe the flight response of seven wild GPS-collared wolves in Norway and Sweden by conducting experimental human approach trials. This is, to our knowledge, the first study assessing wolf flight responses toward humans with the use of high-resolution GPS data, which gives much more detailed and accurate information about flight intensity and flight patterns compared to VHF studies (Moen et al., 2012, 2018; Ordiz et al., 2019). High resolution GPS data makes it possible to re-construct details of the flight path without the need of snow tracking (Moen et al., 2012; Wam et al., 2012, 2014). With this advancement, fine-scaled studies of wolf flight responses no longer rely on snow cover, although supplementary snow-tracking may still have the potential to give additional behavioral information of importance.

Based on previous studies on wolves and bears (Karlsson et al., 2007; Moen et al., 2012) we hypothesized that wolves would avoid approaching humans (H1). Therefore, we predicted that during approach trials, wolves would show predominantly flight responses (P1) and flee before the observer would pass the wolf’s initial location (P2). If seen by the observer, we predicted that the wolves would retreat without signs of aggression (P3).

We hypothesized that wolves likelihood to flee would depend on the detectability of the approaching humans (H2). Previous studies have shown that FID during experimental human approaches can be affected by the number of observers for brown bears (Moen et al., 2012), and the wind condition for wolves (Karlsson et al., 2007). We predict larger FID with increasing number of observers (P4), observers walking though noisy vegetation (e.g., forest with dense regrowth of trees) (P5) and wind direction from observer to wolf (P6).

The wolf’s decision of whether and when to flee after detecting an observer may depend on the horizontal visibility at the individual’s location (i.e., wolf’s perception of risk) (H3). Previous study on bears found that individuals in more concealed resting site had shorter FIDs (Moen et al., 2012). We predicted that wolves resting at a more concealed location would wait for longer before fleeing (shorter FID) (P7) and have a higher occurrence of no flight (P8) compared to wolves at less concealed locations.

A higher response intensity is related to increased perceived risk (Ydenberg and Dill, 1986; Frid and Dill, 2002; Cooper and Frederick, 2007). Therefore, we hypothesized that an early retreat enabled by early detection would result in a lower perceived risk and thus a less intense flight (H4). From this hypothesis we predicted that larger FID would be associated with shorter (P9) and less straight (P10) flights at lower speed (P11).

Following the results from Wam et al. (2012), we hypothesized that after a flight, wolves will seek a more concealed resting location (H5). Therefore, we predicted that the wolf end position will have a shorter horizontal visibility than the wolf starting position (P12).

We performed 21 successful experimental wolf approaches over the course of 4 years (8 in 2018, 8 in 2019, 3 in 2020, and 2 in 2021). Individual wolves were approached on average 4 times (range: 2–7, Table 1). Wolves fled in 18 out of 21 interactions, and did not initiate a flight in the other three interactions (P1). In two thirds (N = 12) of the cases, the flight initiation occurred before the observer(s) passed the passing position, and in one third (N = 6) shortly after the observer(s) passed (Figure 1, P2). Observers did not see or hear the wolves during any of the approach trials (P3).

The flight initiation distance was significantly different between trials conducted with head wind and tail wind (Wilcoxon rank-sum test: W = 1, n = 11, p = 0.009, Figure 2, P6). However, the FID did not differ significantly between one or two observers present (Wilcoxon rank-sum test: W = 19, n = 18, p = 0.068, P4) or between noisy and not noisy conditions (Wilcoxon rank-sum test: W = 44, n = 17, p = 0.481, P5).

Flight initiation distance and horizontal visibility were positively correlated (r_s = 0.4), i.e., the FID increased when the wolf was less concealed (Figure 3, P7). Additionally, the horizontal visibility for three interactions where the wolf did not flee was low, with a mean visibility of 4, 6, and 14 m, respectively (P8), while the median horizontal visibility of the resting sites for fleeing wolves was 16 m (quartiles 7; 22 m).

We found that flight duration, total distance traveled, and flight displacement were positively correlated (r_s > 0.6, Figure 4 and Table 2). Therefore, we only analyzed the total distance traveled. Together with initial speed, initial straightness and overall straightness we considered those variables as a proxy representing flight intensity. We found no correlation between the FID and the total distance traveled (r_s = 0.1, P9), initial straightness (r_s = –0.12) and overall straightness (r_s = –0.24) (P10), and initial speed (r_s = 0.18, P11).

The horizontal visibility showed a significant increase between the wolf start position and the wolf end position (Wilcoxon signed-ranked test: V = 78, p = 0.025, N = 13, Figure 5, P12).

Most wolves that were approached in this pilot study showed what we interpret as an avoidance response, consistent with our first hypothesis (H1). In the majority of trials, the wolf left its initial resting site before the observer(s) passed the passing position (PP). During the remaining trials, the wolf fled shortly after the observer passed, or did not flee at all. No wolves were seen or heard during the approach trials, even when the observer(s) passed the wolf at less than 50 m. Kuijper et al. (2019) described getting closer than 100 m to wolves as “risk-enhancing human behavior”. However, similar to previous studies on wolves as well as brown bears (Karlsson et al., 2007; Moen et al., 2012; Wam et al., 2014; Ordiz et al., 2019), the wolves did not show any aggressive response to our approach trials.

A wolf’s decision to flee from an approaching human is contingent on the human being detected by the wolf. Despite their acute senses, Karlsson et al. (2007) demonstrated that it is possible to walk up to a resting wolf undetected. Hence, we hypothesized that FID would be larger under conditions that increase the detectability of humans (H2). Olfactory and auditory cues are important for a wolf’s ability for communication, tracking prey, and social interactions (Mech, 1970; Harrington and Asa, 2003). Therefore, it is likely that sound and smell might be important for detecting threats as well. Karlsson et al. (2007) found that a combination of wind direction and wind speed affected wolf FID during earlier approach trials and hypothesized that noise made by the observer likely affected the wolf’s FID. We found a significant difference between tail wind and head wind, where tail wind resulted in larger flight initiation distances. Unfortunately, we did not have sufficient observations of wind speed to make a similar comparison. The FID did not vary significantly with number of observers or the level of noise. Potentially, the noise made by the observer may have been masked by noise created by the wind. Exploring, this and other interactions between the explanatory variables will require a larger sample size and multiple regression analysis.

Once a wolf has detected an approaching human, the decision to flee should reflect the perception of risk (Ydenberg and Dill, 1986; Cooper and Frederick, 2007). We found a weak positive correlation between FID and the horizontal visibility at the wolf start position (H3). Furthermore, during three interactions without a flight, the horizontal visibility was low (<15 m). This is partly consistent with our hypothesis that wolves would perceive the risk as lower when they were more concealed, and allow the human to get closer before fleeing, or choose to not flee at all. However, GPS data alone does not allow us to identify the moment of detection, but rather shows the moment when the wolf responds spatially by dislocation. Hence, we cannot conclude whether the shorter FID or no flight at lower visibility was due to the wolf’s perception of being less detectable by the observer (i.e., lower perceived risk) or because the wolf did not detect the observer. This was also described by Moen et al. (2012) who found that the FID for brown bears increased as the horizontal vegetation structure became less dense. Potentially, the wolf’s social status, season, and habituation to human presence might affect the decision to flee, though the effect of habituation were found to be minor (Wam et al., 2014). Fine-scale accelerometer data or heart rate data from the wolves could provide additional information to distinguish between detection and risk perception by potentially measuring changes in heart rate, posture or fine-scale movement before the wolf flees (Græsli et al., 2020; Williams et al., 2020).

We expected that a shorter FID would be related with a more intense flight (larger distance traveled, higher speeds) due to a flush effect if the observer was not detected, or not perceived as a threat, until it was close by (H4). This could result in an energetically more costly flight (Ydenberg and Dill, 1986; Cooper and Frederick, 2007). However, we did not find a correlation between FID and the different flight variables. Other variables, such as landscape composition, road density and human population density might also have an effect on the initiation and path of the flight (Moen et al., 2018), but were not included in our study due to low sample size.

Wam et al. (2012) showed that wolves can adjust their strategy in choosing a resting site in a more concealed location after being disturbed. After centuries of persecution wolves might choose hiding and therefore sacrificing visual vigilance for better concealment to avoid human encounters (Wam et al., 2012). Similar strategies are found in lynx (Lynx lynx) and brown bear (Sunde et al., 1998; Ordiz et al., 2011). Therefore, we hypothesized that wolves would select for a more concealed (i.e., shorter horizontal visibility) resting site after being approached by a human (H5). Interestingly, we found the opposite effect where wolves chose resettling locations which were slightly more open and less concealed compared to their resting sites prior to the disturbance. It is likely that resting site selection after a disturbance might lead wolves to select places with a better overview on approaching humans, so they can initiate a new flight earlier. This would follow earlier results from a study on vigilance and protective cover on birds (Lazarus and Symonds, 1992). Additionally, concealed areas are available throughout the wolf’s territory, and there is no lack of available sites. An increased sample size would give the possibility to explore whether this pattern is consistent and to what degree other factors (e.g., elevation and vegetation types) affect the resettling site selection.

The sample size in this study was small and we could not account for individual differences between wolves, but we are aware that individual choices may exist (Beale, 2007). However, we did not perceive visually obvious differences in response variables between wolf territories. Such potential differences should be considered, and for future studies with larger sample size, we would advise considering the effects of individuals and territories, together with variables as e.g., social status and sex.

Moreover, the small sample size does not allow us to generalize the results across the Scandinavian wolf population, let alone to wolves in general. This first study rather serves to illustrate the application of standardized human approach trials with high-resolution GPS-data for describing wolf responses at a local scale. In continuation, the protocol (see text footnote 1) can be applied at larger spatial scales to compare wolf flight responses within and between populations and across anthropogenic gradients. Potential applications are to establish the range of responses by wolves in a given area, allowing the identification of individual wolves that show atypical response patterns, or to identify consistent differences between wolves inhabiting different habitat types in different parts of their range. This may lead a better understanding of how wolves can be expected to behave toward humans, and thus improve the potential for wolf-human coexistence across their range. If future research confirms what the current and previous studies suggest (Karlsson et al., 2007), i.e., that wolves generally avoid approaching humans, and that increased detectability increases the likelihood of an early flight by the wolf, people who wish to avoid a wolf encounter when walking in wolf areas should maximize their possibility of being detected.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The animal study was reviewed and approved by the Norwegian Food Safety Authority (FOTS ID 15370) and the Animal Welfare Ethics Committee of Uppsala, Sweden (ref. 5.8.18-13246/2019).

PW, BZ, AE, and EV conceived the study. AE, PW, BF, CW, HS, and BZ secured the funding. EV, BF, AE, BZ, and PW conducted field trials. EV conducted the analyses and drafted the manuscript. All authors contributed ideas and edits to the manuscript and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.