Skip to main content

Behavioral “bycatch” from camera trap surveys yields insights on prey responses to human-mediated predation risk

Cite this dataset

Burton, Cole et al. (2022). Behavioral “bycatch” from camera trap surveys yields insights on prey responses to human-mediated predation risk [Dataset]. Dryad.


Human disturbance directly affects animal populations but indirect effects of disturbance on species behaviors are less well understood. Camera traps provide an opportunity to investigate variation in animal behaviors across gradients of disturbance. We used camera trap data to test predictions about predator-sensitive behavior in three ungulate species (caribou Rangifer tarandus; white-tailed deer, Odocoileus virginianus; moose, Alces alces) across two boreal forest landscapes varying in disturbance. We quantified behavior as the number of camera trap photos per detection event and tested its relationship to predation risk between a landscape with greater industrial disturbance and predator abundance (Algar) and a “control” landscape with lower human and predator activity (Richardson). We also assessed the influence of predation risk and habitat on behavior across camera sites within the disturbed Algar landscape. We predicted that animals in areas with greater predation risk (more wolf activity, less cover) would travel faster and generate fewer photos per event, while animals in areas with less predation risk would linger (rest, forage), generating more photos per event. Consistent with predictions, caribou and moose had more photos per event in the landscape where predation risk was reduced. Within the disturbed landscape, no prey species showed a significant behavioral response to wolf activity, but the number of photos per event decreased for white-tailed deer with increasing line of sight (m) along seismic lines (i.e. decreasing visual cover), consistent with a predator-sensitive response. The presence of juveniles was associated with shorter behavioral events for caribou and moose, suggesting greater predator sensitivity for females with calves. Only moose demonstrated a positive association with vegetation productivity (NDVI), suggesting that for other species influences of forage availability were generally weaker than those from predation risk. Behavioral insights can be gleaned from camera trap surveys and provide information about animal responses to predation risk and the indirect impacts of human disturbances.


We deployed a single camera trap (CT) (Reconyx HyperFire PC900, Holman, WI) at each sampling site across the Algar and Richardson study areas in stratified random designs. The primary objectives of the surveys were to assess the distribution and relative abundance of medium- and large-bodied mammals in relation to landscape characteristics, particularly industrial disturbances like seismic lines. In the Algar study area, 73 CTs were deployed between November 2015 and November 2017, with year-round sampling continuing until November 2019. Sixty CT sites were on seismic lines and 13 were off of seismic lines. On-line sites were further stratified by restoration category (treated, regenerating, unrestored control, human use). In the Richardson study area, CTs were deployed in November 2017 and 2018 at 57 sites stratified by in (27) vs. out (30) of burned area and on (18) vs. off (39) of a seismic line, with year-round sampling continuing until November 2019. In both study areas, cameras were left in place and sampled continuously once deployed (mean sampling days per station was 1024 in Algar and 508 in Richardson).

At all sampling sites, CTs were placed on a tree 2 to 5 m from the edge of a seismic line or game trail, facing across the line (i.e., perpendicular to expected direction of animal travel), at a height of approximately 0.8 m above the ground (range = 0.7 – 1.1 m), targeting medium- to large-bodied mammals without bait or lure. One picture was taken per motion trigger with no delay between subsequent triggers and sensitivity was set to maximum with a fast shutter speed. CTs were active for 24 hours per day with no quiet period. One timelapse picture was taken each day at noon to ensure camera function (any camera-days with snow occluding the lens were excluded, but this rarely occurred).

We processed CT images by identifying the focal ungulate and predator species (woodland caribou, white-tailed deer, moose, wolf). For ungulate species, we counted the number of unique individuals (i.e., group size) and classified images by the sex (male, female) and age class (adult or juvenile, i.e. young of the year) of visible individuals. We characterized observed behavior for ungulate prey species in each image as either Secure (resting; foraging with food visible in its mouth or its head down with open mouth; inspecting camera with the face covering > 20% the image), or Travelling (animal seen walking past camera). Species identifications were made from images within Camelot software; any uncertain identifications were excluded from analysis. For each species, we defined a detection event at a given site as a sequence of images separated by no more than 15 minutes between consecutive images. This 15-minute threshold was based on identifying a consistent gap time between subsequent images across all focal species.

We evaluated three ways of distinguishing “at risk” from “secure” behaviors for each detection event of ungulate prey: classification, event duration, and number of photos per event. We first considered whether the majority of photos in an event demonstrated Secure or Travelling behaviors (classified following above definitions), but determined that behavioral classes could be difficult to discern over short event durations or when multiple behaviors were exhibited within a single event. We next inferred that risk-averse behaviors, such as travelling through an area without foraging, would result in detection events of shorter duration with fewer photos, whereas more secure behaviors (e.g., foraging, resting, inspecting camera) would have longer detection events and more photos, as the animal lingered in front of the camera. We found that large time intervals between detections could inflate event durations and skew the observed distribution (resulting in poor model fit). Therefore, while the three different response variables were correlated, we focused our analysis on the number of photos per detection event as the primary response of interest. 

Usage notes

Data files (csv) can be opened in Excel or text editor; scripts in text editor or R statistical software.


Natural Sciences and Engineering Research Council, Award: RGPIN-2018-03958

Canada Research Chairs

Alberta Upstream Petroleum Research Fund

Alberta Conservation Association