Eat or be eaten: Implications of potential exploitative competition between wolves and humans across predator- savvy and -naive deer populations

Study Area We conducted this study in Michigan’s UP and LP from September 21, 2018 to November 10, 2018. The UP study area was entirely within the Hiawatha National Forest and was interspersed with unpaved, US Forest Service roads. The bait sites in the LP study area were all located in Mackinaw State Forest of Michigan, and the area was also interspersed with private agriculture, forestland, and paved, secondary roads. The main vegetation cover, derived from the National Land Cover Database, at the UP study sites was evergreen (5 sites), woody wetland (3 sites), grassland (3 sites), deciduous (2 sites), and mixed forest (2 sites), while the LP study sites were deciduous (6 sites), scrub/shrub (4 sites), grassland (3 sites), and evergreen (Homer et al., 2015). Study Design We established 30 deer bait sites between the two study areas, 15 in the UP and 15 in the LP. To mimic recreational hunters, we selected sites based on deer hunting desirability such as habitat openness and/or proximity to a deer trail (Peterson, 2015). Sites were at least 1 km away from each other to avoid duplication and dependence (Winke, 2012). We baited each site with 3.8 liters of corn spread in a 3 by 3-meter- square plot, as per the Michigan deer baiting regulations (Michigan Department of Natural Resources, 2018). For the duration of the study, all sites were baited each week in order to maintain 3.8 liters of corn on the ground to replicate actual hunter behavior and to reduce the confounding effect of bait availability on deer behavior. We deployed one remote camera (Reconyx Hyperfire series or Cuddeback E) on a tree 4 meters from the center of the bait site. Each camera was programmed to take 5 burst images with no delay between triggers to record the number of deer present, their posture, sex (when possible), and presence/absence of fawn(s). We replaced batteries and memory cards weekly. BACI Experimental Setup We used a 10 week before-after-control-impact (BACI) design for the experimental treatment of both areas. This BACI design was conducted so that all sites were paired with themselves with all other factors (e.g., other predators, human visitation, environmental effects) remaining comparable. This allowed us to tease apart the effects of the treatment. Before experimental treatments, we collected data for one week preceding the six-week period to establish a baseline condition of deer visitation to the bait sites (Stewart-Oaten et al., 1986). For the first 3 weeks (September 28-October 20), we baited sites and recorded deer occurrence and behavior via remote cameras with no scent treatments applied. For the following three weeks, we continued to bait and record deer after treating the sites with three scents. In each area, UP and LP, 5 sites were treated each with wolf urine (predator treatment), lemon juice (novel scent; not a native plant), or distilled water (control). We assigned site treatment randomly (Atkins et al., 2016; Wikenros et al., 2015). When we treated sites in the field, we visited all ‘control’ sites in an area first followed by the ‘novel scent’ sites, and finally ‘experimental/wolf urine’ sites to avoid cross-site spread of scents. We also used different treatment application tools (i.e., pipets, travel containers) for each treatment type to avoid cross-contamination of treatments. At each site, we applied 10 mL of the given treatment (experimental, novel, or control) by dripping it from a pipet onto the bait site. This was done to mimic the amount and application of scent marking of a pair of wolves (Peters & Mech, 1975). Vegetation Cover We also recorded predator concealment or hiding potential (horizontal cover) at each site to account for site-specific differences in habitat openness - an index of perceived predation threat. We used a 2 m cover pole with 20 sections, each 10 cm long, to measure predator hiding potential at each site (Kuijper et al., 2014; Severud et al., 2019). From the center of each bait pile, we measured the horizontal cover in each cardinal direction. While one person held the cover pole 10-meters away from the center of the bait pile, another took a picture of the cover pole with a camera stationed on a 1-meter high pole at the center of the bait pile, mimicking deer visual height (Severud et al., 2019). We conducted this procedure twice during the study, once at the beginning and once when treatment started, to account for change in hiding potential with loss of leaves later in the year. For each of the 20 (10 cm) sections on the 2 m cover pole, we estimated obscurement to the nearest 25% (Severud et al., 2019). We calculated a single mean and standard error value for each site and each measurement time (beginning and middle of study). Analysis Photo Analysis Images were tagged in batches using the DigiKam photo editing software (Niedballa et al., 2016). Each batch was defined by any set of images taken within 5 minutes of the prior image. In each batch, for deer images only, we recorded if fawns were present, labelled all sexes when possible, and tallied a total count. This increased the likelihood that we accounted for all individuals that visited the bait pile and avoided miscounting any that were out of the camera frame. For individual pictures within batches, in addition to generic behaviours (e.g., fighting, nursing, foraging), we recorded the number of individuals with their heads above their shoulders, indicating vigilance/state of alertness (Flagel et al., 2016; Schuttler et al., 2017). We also labelled batches for different species that were captured at site visits including squirrels (Sciurus spp.), raccoons (Procyon lotor), wild turkeys (Meleagris gallopavo), and coyotes. We did not detect additional scent marking by other species. Treatment Impact on Deer Vigilance, Group Size, and Visitation Rate We explored the impact of different treatments in the two areas (UP and LP) on the number of deer at each site, the proportion of the group that was vigilant, and the number of visits made to the site. To account for the paired nature of our treatment design, we first averaged the variable values for each site individually and calculated a difference in values before and during treatment application. We then averaged across all sites within a given treatment. We also calculated a vigilance intensity metric using equation 1. This metric includes both group vigilance and event duration. (Eq 1) I=(∑v/g)/e Where I is the vigilance intensity metric for a given event, v is the number of deer vigilant in a single image, g is the group size in a single image, and e is the total time of the event. Hence, vigilance intensity simply standardizes individual vigilance across different group sizes and time spent in front of the camera. Similar to the previous analysis, we calculated a difference in vigilance intensity before and after treatment for each treatment type in each peninsula. Diel Activity To analyze possible temporal variability in the use of bait sites in the UP and LP before and after treatment, we used nonparametric kernel density estimation (Prugh et al., 2019; Wang et al., 2015). We converted times to radians and used a kernel density estimator to create a probability density distribution for each before or after period (Ridout & Linkie, 2009). We calculated the proportion of temporal overlap between the two treatment periods for each treatment in an area (Wang et al., 2015). We used a Δ̂4 with a smoothing parameter of 1 because our sample size for all analyses was greater than 50 (Ridout & Linkie, 2009). We conducted this analysis using the overlap package (Meredith & Ridout, 2018; Wang et al., 2015) in R (R Development Core Team 2013). We applied Watson’s U2 statistic with the CircStats package to test for homogeneity between the two samples of interest (i.e., test for/detect a statistically significant shift in the diel pattern before and after treatment) (Lashley et al., 2018; Lund & Agostinelli, 2012). If deer significantly shifted their temporal pattern between the two treatment periods, we expected Watson’s U2 statistic would be greater than the critical value (0.19 for an α value of 0.05) and P < 0.05. In the UP, we predicted that there would be no shift in temporal visitation by deer at the control and lemon treated sites (high Δ̂4, U2 ≤ 0.19), while a shift to more nocturnal activity at the wolf urine treated sites would be detected (lower Δ̂4, U2 > 0.19; Kohl et al. 2018) to potentially avoid wolves that are known to be typically less active at night (Kohl et al., 2018). In the LP, we expected that we would not see a significant shift in any treatment sites because deer in the LP are ostensibly naive to wolf predation. Effect on Vigilance Behaviour We fitted generalized linear models using pooled data from both the UP and LP to check for the effect on overall vigilance intensity by treatment effect (before/after) and type (predator/novel/control), area (UP or LP) and vegetation cover (i.e., predator hiding potential; S1; Fležar et al., 2019; Prugh et al., 2019) through additive and interactive effects between parameters of interest. Subsequently, we fitted models to examine the relationship between vigilance intensity and vegetation cover at each treatment type within both the UP and LP before and after treatment. Using generalized linear model, we further examined the effects of sex, vegetation cover, and presence of young on vigilance intensity, and included additive and interactive effects with area to detect any difference between the UP and LP. We ranked models based on AICc.