Efficacy of GPS cluster analysis for predicting carnivory sites of a wide‐ranging omnivore: the American black bear

Abstract

AbstractThe capacity to describe and quantify predation by large carnivores expanded considerably with the advent ofGPStechnology. Analyzing clusters ofGPSlocations formed by carnivores facilitates the detection of predation events by identifying characteristics which distinguish predation sites. We present a performance assessment ofGPScluster analysis as applied to the predation and scavenging of an omnivore, the American black bear (Ursus americanus), on ungulate prey and carrion. Through field investigations of 6854GPSlocations from 24 individual bears, we identified 54 sites where black bears formed a cluster of locations while predating or scavenging elk (Cervus elaphus), mule deer (Odocoileus hemionus), or cattle (Bosspp.). We developed models for three data sets to predict whether aGPScluster was formed at a carnivory site vs. a non‐carnivory site (e.g., bed sites or non‐ungulate foraging sites). Two full‐season data sets containedGPSlocations logged at either 3‐h or 30‐min intervals from April to November, and a third data set contained 30‐min interval data from April through July corresponding to the calving period for elk. Longer fix intervals resulted in the detection of fewer carnivory sites. Clusters were more likely to be carnivory sites if they occurred in open or edge habitats, if they occurred in the early season, if the mean distance between all pairs ofGPSlocations within the cluster was less, and if the cluster endured for a longer period of time. Clusters were less likely to be carnivory sites if they were initiated in the morning or night compared to the day. The top models for each data set performed well and successfully predicted 71–96% of field‐verified carnivory events, 55–75% of non–carnivory events, and 58–76% of clusters overall. Refinement of this method will benefit from further application across species and ecological systems.