Terrence Larsen, Karen Graham, Gordon Stenhouse, Dario Fernández-Bellon

Understanding carnivore use of habitat resources and interactions with prey is key for the effective management and conservation of their populations. However, accurately quantifying such interactions for wide ranging or cryptic species requires resource-intensive field efforts. Here we combine GPS data, field investigations, and statistical modeling to identify predation and scavenging events by grizzly bears (Ursus arctos) in Alberta, Canada. We used GPS location data from 70 bears (37 females and 33 males) collared between 2001 and 2017 (n = 101 bear years when accounting for bears with multiple data years). We visited a selection of location clusters (n = 3,038) to determine bear activity and classified them as carcass (predation and scavenging of ungulate species; n = 352) or non-carcass clusters (bedding and other foraging; n = 2,686). At carcass clusters, we determined prey size based on ungulate species and age class. We then built random forest models to identify carcass clusters based on GPS point characteristics (including time of year, cluster diameter and duration, bear movement rate, distance travelled on departure from cluster, and distance to the next cluster). Models were built using 85% of the data and tested on the remaining 15% and validated over 50 iterations. The final models identified carcass clusters and prey size with 81.1% and 81.4% balanced accuracy, respectively. Our findings show how GPS data can be used to predict ungulate handling indicative of meat consumption events. We also applied these models across our entire GPS location dataset (n = 162 bears and 294 bear years during the same period) to identify 3,486 ungulate carcass clusters. We then explore the relationships between meat consumption event frequency, bear biology (sex, age, biometrics), and habitat characteristics (area protection, terrain, ungulate density), and discuss how this approach can be used to inform species management and conservation.