Habituation, den defence, and predatory opportunism appear to have contributed to coyote aggression in an urban park

Context
Between December 2020 and August 2021, there were an unprecedented 45 coyote attacks and 63 additional aggressive encounters involving humans in Stanley Park, Vancouver, BC, Canada.

Study Objectives
The objectives of this study were to (1) determine the spatial and temporal factors that distinguished aggressive human-coyote encounters from non-aggressive coyote sightings in Stanley Park and (2) identify the individual and group characteristics of human attack victims. Based on the complementary hypotheses that aggression typically stems from one or more of (1) habituation, (2) food conditioning, (3) defence of dens or resources, and (4) opportunistic predation, we predicted that coyote aggression would increase when there were more park users (causing habituation), closer to potential sources of human-associated food (causing food conditioning or as resource defence), in contexts with opportunities for predatory behaviour (e.g., times of the day or week when individuals were more likely to be alone), closer to den sites (den defence), and during the pup-rearing season (den or pup defence). We further predicted that attack victims would more often be exhibiting flight behaviour (e.g., running), vulnerability (e.g., alone, nighttime activity, or children), or threats (e.g., presence of dogs) at the time of the attack.

Description of Data
The file "sp_coyote_project_dataset" contains all variables used for objective 1. In the absence of daily human visitation data, we used COVID-19 pandemic-induced restrictions on indoor social gatherings as a proxy for long-term human park use, with the lockdown phase representing the phase of greatest human activity. Time of day was converted to radians and then cosine-transformed with a phase shift of -7π/6 to make 2pm the high and 2am the low, emulating hours of peak human activity in the park. Land cover variables are presented in pixels and in proportion of total area within a fixed radius corresponding to the distance included in the variable name. All distance measurements (i.e., land cover radii and variables starting with "d2") are in metres and all variables used for analyses were scaled to standard deviation=1 and mean=0, denoted as [variable]_scaled in the data.

Both human activity datasets contain data collected in 2024, which we used as the reference period for expected human activity and proportion of park users within each demographic and activity category in the park at various times of day.

The .lyrx file is an ArcGIS file that depicts the edited land cover classifications we used for this study.

Key Findings
We found that aggression increased with greater long-term human activity, times of day and week with lower human activity, proximity to dens, and areas of greater human activity in the park. We also found that human victims were more likely than expected by chance to be alone and running at the time of the attack, but that age, gender and dog presence had no significant effect.