Human development has major implications for wildlife populations. Urban-exploiter species can benefit from human subsidized resources, whereas urban-avoider species can vanish from wildlife communities in highly developed areas. Therefore, understanding how the density of different species varies in response to landcover changes associated with human development can provide important insight into how wildlife communities are likely to change and provide a starting point for predicting the consequences of those changes. Here, we estimated the population density of five common mesocarnivore species (coyote (Canis latrans), bobcat (Lynx rufus), red fox (Vulpes vulpes), raccoon (Procyon lotor), and Virginia opossum (Didelphis virginiana)) along an urban to rural gradient in the greater Fayetteville Area, Northwest Arkansas, USA between November 2021, and March 2022. At each study site, we applied the Random Encounter Model (REM) to data from motion-triggered cameras to calculate the density of our five focal species. Coyote density ranged from 0–3.47 with a mean of 0.4 individuals/km2. Raccoon density ranged from 0–93.26 with a mean of 4.2 individuals/ km². Bobcat density ranged from 0–8.87 with a mean of 0.33 individuals/km². Opossum density ranged from 0–27.35 with a mean of 0.76 individuals/km². Red fox density ranged from 0–0.73, with a mean of 0.02 individuals/km². We used generalized linear models to evaluate the density of each species against environmental and anthropogenic variables. Coyotes and raccoons occurred in the greatest densities in areas with high anthropogenic noise levels, suggesting that both species are synanthropic and able to co-exist in areas of high human activity. Alternatively, Virginia opossum and red fox densities were greatest in open, developed areas (lawns, golf courses, cemeteries, and parks) and were absent (red fox) or rare (opossum) in natural areas. We found no evidence that bobcat density varied along the urban to rural gradient studied, but this lack of evidence may have been driven by the small spatial scale of many of our sites in relation to space needs of this wide-ranging species. The density estimates we report based on game camera data of unmarked animals were consistent with reports from the literature for these same species derived from traditional methods, providing additional support to the REM as a viable, non-invasive method to calculate density of unmarked species.

This data was collected through camera traps set between November 1, 2021, and March 14, 2022. Cameras were set at 12 study sites in the Ozark Mountain Ecoregion, Northwest Arkansas, USA. Sites were chosen to represent a continuum of human activity and ranged from 2km to 60km from downtown Fayetteville, Arkansas. Camera trap images were sorted using Timelapse 2.0 software. Detections were sorted into 5-minute "episodes", and each episode was treated as a single detection to avoid double counting individuals. 

To estimate the density (D) of our five focal species from game camera detections, we applied the Random Encounter Model (REM) equation, where y refers to the total detections of each animal per camera, and t is the total trap nights in hours (measure of trapping effort). V is the day range of each species, referring to how far an animal travels in a 24-hour period. We used published day range estimates for each species and used the median day range value for each species from all reported estimates to parameterize our models.  Values for the detection radius (r), and detection angle (θ) were collected for each camera in the field through walk tests. A walk test entailed walking directly towards each camera to calculate detection radius and from each side at 5m from the camera to calculate detection angle in degrees. Detection was determined by whether or not the detection light was triggered on the camera during each walk test. The detection angle was later converted to radians for density calculations.

To assess which landcover variables most influenced the density of each focal species, we used an iterative approach to assemble 31 Generalized Linear Mixed Models (GLMM) with additive effects using r programming and the “lme4” and “AICcmodavg” packages for five predictor variables: HUD, noise, distance to water, and developed open, including a global model (all variables with random effect) and a null model (only random effect). We used study site as a random effect in each model. The zero inflation in the data was accounted for by using a gamma distribution in all models. We then used AICc selection criteria with an a priori cutoff of two for the ∆AIC delta value. 

R programming and the “lme4” and “AICcmodavg” packages

Microsoft Excel

Timelapse 2.0

ArcGIS Pro (ArcGIS Pro 2.8.3, 2021; Esri Inc, Redlands, CA)