Our study was conducted from November 2018 to November 2019 at 10 crossing structures in California, USA: eight culverts and two bridges under state highways (Figure 1). Three of these structures were purpose-built for wildlife crossing, and seven were built for water movement, human access and were opportunistically used by wildlife. Parameters for structural attributes and local conditions would be expected to control for these different uses. Crossing structures were located across Interstate 5 and U.S. Route 97 in northern California; Interstate 80, Interstate 680, and U.S. Route 50 in central California; and State Route 74 in Southern California. Crossing structures were located under highways consisting of one to three lanes of traffic in each direction, in areas classed as low- or medium-intensity development, evergreen forest, or shrub/scrub (Dewitz, 2019).

Behavior of coyotes and mule deer was assessed from videos obtained using camera traps. We placed six camera traps (Browning Dark Ops Pro) at each crossing structure and set cameras to record high-quality video and audio, triggered by motion and infrared (no-glow). Once triggered, cameras recorded videos for 20 s, with a 5-s delay between triggers. Cameras were equipped with the “smart IR setting,” i.e., videos during the daytime were recorded beyond 20 s if the camera continued to detect wildlife movement. We placed one camera at each of the two entrances of each crossing structure, ca. 5 m (range: 0–10 m) from the entrance, facing inward. We also placed two cameras at each of the two “approach zones” perpendicular to the crossing, the area ca. 50 m from each crossing structure entry from which wildlife approach the crossing (range: 11–84 m). Cameras remained operational at each crossing structure for an average of 45 days (range: 1–106 days). We captured the location of each camera station and the crossing structure entrance via a handheld GPS device (Garmin eTrex 20×) and estimated the Euclidean distance of each camera station to the entrance of the crossing structure via Google Earth Pro (V7.3.3.7786).

We extracted activity from each camera trap video that included mule deer and coyotes using the Behavioral Observation Research Interactive Software (BORIS; Friard and Gamba, 2016). To analyze flight and entry responses associated with acute noise, videos from cameras positioned at the crossing structure entrance (ca. 5 m from entrance) were checked for observations in which animals were facing the direction of the crossing structure entryway or walking toward the crossing structure entryway. Within these observations, we scored each time a movement-related decision-making event occurred: (1) Entry, defined as stepping across the concrete edge of the crossing structure entrance; and (2) Flight, defined as an alteration in body direction and movement away from the crossing structure (Figure 2).

To examine behaviors associated with chronic noise, we also scored activity of individual mule deer and coyotes for all videos from all cameras, including the videos analyzed for fight or entry responses. We focused on three categories of behavior associated with fear: (1) vigilance, defined as an individual standing or sitting upright, (2) running, defined as any movement faster than a walk, and (3) foraging, defined as feeding or drinking (Figure 2 and Supplementary Figure 1). For each of these behaviors, we assigned a start and stop time within each video and calculated time as a proportion of total time the individual was present in the video. The BORIS enabled classification of multiple behaviors recorded simultaneously. To distinguish between individuals within one video, we assigned each conspecific a unique identification based on order of appearance or distance from the camera. To avoid differential periods of exposure to a camera and to ensure observations were independent, video clips longer than 20 s were trimmed, and observations were filtered to include a minimum time interval of 5 min between observations. For each video, we documented whether the video was taken in day (color) or night (black and white) mode, the temperature listed for each video collected by the camera’s internal thermometer, and the maximum group size observed. We recorded the number of humans that passed through every video and quantified the frequency of daily visits by humans for each camera station (0–12/day). Research technicians who classified observations initially received extensive training in the recognition of behaviors across species and data capture using BORIS software.

We considered anthropogenic noise at multiple temporal and spatial scales. We quantified acute traffic-related noise (1 s) during each decision-making event occurring at the entrance of a crossing structure. We quantified chronic traffic-related noise for the duration of wildlife presence at a crossing structure across two spatial scales: (1) in each video (−20 s), representing variation within a crossing structure and (2) at each crossing structure (1 week), representing variation among crossing structures.

We analyzed the effects of acute noise on decision-making at the entrance of the crossing structure (entry or flight response), and chronic noise on behavior (proportion of time being vigilant, running, or foraging).

For mule deer, we observed 192 decision-making events at 16 cameras and 8 structures. A total of 35 (18%) of the events resulted in a flight response and 157 (82%) resulted in an entry (Supplementary Table 2). For coyotes, we observed 50 decision-making events at 11 cameras and 4 structures, nine (18%) of which resulted in a flight response and 41 (82%) resulted in an entry. For both mule deer and coyote, flight responses were more likely to occur as levels of acute noise increased (deer p ≤ 0.001, SE = 0.23; coyote p = 0.018, SE = 0.52; Table 1 and Figure 3). The best fit model also demonstrated that mule deer were more likely to enter underpasses with a smaller openness ratio (p = 0.001, SE = 0.58; range = 0.2–40.9), and coyotes were more likely to enter underpasses during warmer temperatures (p = 0.016, SE = 0.05). For model AIC and coefficient estimates, see Supplementary Table 3.

We collected a total of 3 h and 44 min of mule deer observations (n = 1,346) from 28 cameras at 10 structures. In the absence of a minimum 5-min time gap between observations, mule deer foraging increased during greater levels of within-structure noise. However, with a time gap (5 min) to increase independence of observations, we found that vigilance, foraging, and running behavior were not altered during greater levels of within-structure noise. For both sets of analyses, greater levels of among-structure noise demonstrated an increase in mule deer foraging (p < 0.001, SE = 0.02; Figure 4), coupled with a decrease in vigilance (p < 0.001, SE = 0.04; Figure 4).

We collected 20 min of coyote observations (n = 243) from 21 cameras at 8 structures. In the absence of a 5-min time gap between observations, vigilance increased at louder levels of within-structure noise, however, this relationship became non-significant with the addition of a 5-min gap between observations. In response to greater levels of among-structure noise, coyote demonstrated a different response to mule deer; vigilance decreased (p = 0.005, SE = 0.04; Figure 5), and running increased (p < 0.001, SE = 0.14; Figure 5). Other covariates and their associated effect on vigilance, running, and foraging are summarized in Table 1 and Supplementary Table 4.

We found support for our hypothesis that acute noise from vehicles invokes a flight response (Figure 3). Our results provide additional support to prior findings that a fear response can be induced by sudden increases in noise associated with transportation. We found that at crossing structures where coyotes and mule deer were exposed to higher levels of acute noise from vehicles, both species were more likely to initiate a flight response than they were to use the crossing. Hermit crabs (Pagurus bernhardus) (Walsh et al., 2017; Tidau and Briffa, 2019), fish (Voellmy et al., 2014), black-tailed prairie dogs (Cynomys ludovicianus) (Shannon et al., 2016a), mule deer and mountain sheep (Ovis canadensis) (Weisenberger et al., 1996) have similarly demonstrated startle responses from acute noise, leading to increased physiological stress. Through invoking a flight rather than a crossing-entry response, acute noise pollution resulted in suboptimal behavior (Sih, 2013) for coyote and mule deer; a result that impairs connectivity across roads with potential effects on genetic exchange and access to resources (van der Ree et al., 2009). If other wildlife species experienced similar responses, the consequence would be impaired effectiveness of wildlife crossing structures as mitigation for connectivity losses.

For coyotes, we detected a switch from vigilance to running at louder crossing structures (Table 1 and Figure 5). These behavioral changes are in accordance with the risk disturbance hypothesis, which posits that alterations to anti-predator behavior are due to wildlife perceiving the disturbance as a threat, and suggest that acute and chronic traffic noise can alter the “Landscape of Fear” in predators (Laundré et al., 2001). Our findings are supported by previous work that has demonstrated exposure to transportation noise can cause decreased foraging for a number of terrestrial mammals (Shannon et al., 2014b; Smith et al., 2017), marine mammals (Blair et al., 2016; Wisniewska et al., 2018), and other taxa (Wale et al., 2013; Castaneda et al., 2020). Though higher levels of chronic noise elicited a behavioral fear response in coyotes, for mule deer, chronic noise among structures elicited reduced anti-predator behavior, suggestive of a Human Shield effect (Table 1). Hence, wildlife responses to noise pollution can vary according to sound levels, temporal scale of sound, and according to species.

Our finding that increased foraging and lowered vigilance by mule deer coincided with higher levels of chronic noise among structures (Figure 4) suggests that zones of greater traffic noise may provide a refuge from predation, consistent with the Human Shield hypothesis (Berger, 2007). Ungulates have displayed similar increases in foraging in response to anthropogenic stressors within U.S. National Parks (Berger, 2007; Brown et al., 2012; Shannon et al., 2014b). Predators of mule deer have been shown to display strong fear responses to noise pollution (Smith et al., 2017). Large-bodied carnivores may also be particularly deterred from areas of high anthropogenic noise due to the low-frequency energy overlapping strongly with the frequencies in which carnivores communicate, or because of an inability to detect prey as a result of cue masking (Warren et al., 2006; Cardoso et al., 2018). Therefore, predation risk could be lower for ungulates during periods of higher noise pollution. However, in the case of roads and traffic, ungulates using the Human Shield to avoid predators may still suffer higher rates of mortality from collisions with vehicles.

Another explanation for mule deer behavior can be taken from the risk allocation hypothesis (Lima and Bednekoff, 1999), which states that wildlife exposed to long-term, continuous threats are unable to sustain anti-predatory behavior, and that anti-predator responses are used only when exposed to unpredictable, infrequent stimuli. Though we did document a fear response in mule deer to acute traffic noise, and no effect of noise when comparing behavioral events within a single crossing structure, under the risk allocation hypothesis, we would expect mule deer to exhibit no difference in anti-predator behavioral responses when exposed to chronic noise across structures. Rather, we detected a decrease in vigilance coupled with an increase in foraging of mule deer, rendering the risk allocation hypothesis a somewhat unlikely explanation for our findings.

Our results indicate a differential effect of noise pollution across temporal (acute vs. chronic noise) scales for deer, and a differential effect of noise pollution across spatial (within-structure vs. among-structure) scales for both species. Previous studies have recommended examining responses to fear across multiple spatiotemporal scales to capture variation in response (Creel and Christianson, 2008; Moll et al., 2017; Prugh et al., 2019). Here, in sampling vehicular noise across different spatial and temporal dimensions, we were able to identify the mechanisms driving behavioral responses that occur across the temporal spectrum of noise. Findings for mule deer demonstrate that fear and Human Shield effects are not mutually exclusive, and that prey species concurrently assess risk at various temporal scales. Employing a hierarchical modeling framework, we also identified that species-level behavioral responses differ across spatial scales when examining chronic noise. Much of our current knowledge surrounding the relation of risk perception and behavior to spatiotemporal scale is based upon herbivore-carnivore interactions. For example patch- and landscape-level fear occur in red deer (Cervus elaphus) when inhabiting wolf territories in Poland, and African ungulates increase vigilance during times of concurrent short-term and long-term risk from predation (Kuijper et al., 2015; Dröge et al., 2017). Here we add to the ecology of fear literature and demonstrate anthropogenic stressors also mediate fear responses of wildlife across multiple temporo-spatial scales.

Our primary goal was to assess the effect of anthropogenic noise on behaviors of mammals at highway crossing structures; however, we also examined several covariates which could alter the behavioral response (Table 1 and Supplementary Tables 3, 4). First, regardless of noise condition, for mule deer in larger groups we detected a lower likelihood of a flight response at the crossing structure. This trade-off between flight response and group size at crossing structures is likely due to individuals requiring less time to scan for predators (many eyes hypothesis) in larger groups, and individuals benefiting from a lowered predation risk through dilution effects (Roberts, 1996). As an example, vigilance of California ground squirrels (Otospermophilus beecheyi) only increased when small groups, but not large groups, were exposed to loud playbacks of natural sounds (e.g., rivers) (Le et al., 2019). The proportion of time mule deer spent running also increased when in larger groups, though this may be due to juveniles playfully interacting with one another.

Second, as expected, a greater number of humans present at the crossing structure caused a decrease in deer foraging and an increase in running, particularly during the daytime. This supports previous findings demonstrating that wildlife fear human presence generally, though differentially in scale (human shield effect) and will often shift to more nocturnal activity as a way to avoid humans (Stankowich, 2008; Ciuti et al., 2012; Gaynor et al., 2018). Coyotes also demonstrated less running activity during nocturnal hours, which could be associated with exposure to artificial light from headlights and streetlamps at crossing structures. For example, mule deer use brightly lit urban areas as a human shield (Ditmer et al., 2021). Future research should include experimental playbacks of anthropogenic sound to disentangle the relationship between noise and visual disturbances.

Third, coyote running decreased and mule deer vigilance increased at camera stations closer to the crossing structure, which is indicative of a switch to individuals assessing whether to enter the structure. At camera stations farther away from the crossing structure, mule deer foraging increased. In addition, during warmer summer months, mule deer foraging increased while vigilance decreased. These behavioral shifts are likely due to higher densities of vegetation during warmer seasons and in areas farther from the concrete crossing structure. In contrast, coyote vigilance increased during warmer temperatures, potentially due to greater activity of their predators. For example, black bears (Ursus americanus) are only a predation risk for coyotes during the spring and summer months when they are not hibernating (Rogers, 1992).

Fourth, mule deer increased vigilance and lowered foraging time at crossing structures associated with greater levels of traffic volume and evergreen forest (dominant NLCD category within 100 m buffer of the crossing). It is likely that deer perceive more forested habitats, often synonymous with dark edges, and highly trafficked areas, often synonymous with greater lanes of traffic and higher speeds, to be more risky (Altendorf et al., 2001; Blackwell et al., 2014; Knopff et al., 2014; Ditmer et al., 2021).

Finally, mule deer behavior was affected by the openness ratio of the crossing structure—for more open crossings, deer were less vigilant, but also less likely to forage, more likely to run and more likely to flee away from the entrance. Previous studies have typically observed a higher propensity for deer to use structures with adequate visibility and larger openness ratios (Clevenger and Waltho, 2005; Bissonette and Rosa, 2012). The structures in California (present study) potentially had greater associated traffic volumes than the studies in Clevenger and Waltho (2005) and Bissonette and Rosa (2012), which were in the Rockies. This could contribute to differences between the present study and previous published studies. Our finding could also be driven by certain unmeasured characteristics of the structure or surrounding landscape, and future studies should go beyond measuring the openness ratio and include measures such as brightness within the structure to tease this effect apart.

Though we filtered the dataset to include a minimum time interval of 5 min between observations, and identified unique individuals within each video, we did not monitor behavior at the individual level across the entire study. To determine the impact that behavioral alterations can have on individual fitness, and whether unsuccessful crossings have a population-level impact, future research should examine whether flight is the consistent response of certain individuals, or a proportion of decisions made by all individuals. Individuals could be monitored through GPS telemetry data or software that enables individual recognition from data collected by camera traps. Monitoring individual responses could also shed light on whether fear responses are characteristic of certain personalities. For instance, Eastcott et al. (2020) found among-individual behavioral differences in vigilance behavior of dwarf mongoose to playbacks of traffic noise.

Road networks are expanding globally, with the addition of 25 million km predicted by 2,050 (Dulac, 2013; Laurance et al., 2014). With this expansion comes increased interactions between wildlife and humans. Our work demonstrates that a fear response of wildlife to anthropogenic noise stressors at wildlife crossing structures is likely to differ depending on temporal scale, species, and characteristics of the stressor. This dynamic, behavioral response to noise pollution could alter predator-prey interactions, which can scale up to ecosystem-level consequences such as trophic cascades (Smith et al., 2017; Gaynor et al., 2019; Zanette and Clinchy, 2019). Furthermore, crossing structures are proposed as the primary conservation response to road impacts to wildlife. Ineffective crossing structures, as measured by wildlife use, could result in changes in wildlife species distributions and resource availability and use, which could degrade species interactions and ecosystem health. For example, differential responses of large, common ungulates (e.g., mule deer) and their predators (e.g., mountain lions, coyotes) could lead to spatial variation in grazing and browsing intensity based on prey and predators’ responses to traffic conditions, which could lead in turn to landscape-scale changes in plant communities and wildfires. To ensure a landscape of coexistence, additional research is needed to understand what anthropogenic stressors drive fear responses in wildlife and how spatiotemporal scale plays a role. Our work suggests that to ensure the viability of wildlife crossing structures, vehicular noise pollution should be considered during the design phase of these structures.