Choosing Sampling Interval Durations for Remotely Classifying Rocky Mountain Elk Behavior

Abstract

Abstract Dual-axis accelerometer global positioning system collars can be used to remotely record the activity level and behavior of free-ranging animals, but inter- and intraspecific variations in motion among behaviors necessitate calibration for each species of interest. To date, little work has been done to determine the best duration for sampling intervals when using activity monitors that incorporate dual-axis accelerometers. However, we expected that the duration of behaviors relative to the duration of sampling intervals could affect the accuracy of calibration and behavior classification models. Furthermore, we considered the potential effect of winter diet supplementation (hay) on behavior classification. We used Lotek 4500 global positioning system collars featuring dual-axis accelerometer activity monitors to collect data for calibration and classification trials on Rocky Mountain elk Cervus elaphus nelsoni. We used discriminant function model structures to determine the number of accurately classifiable behaviors that could be derived from data sampled over three sampling interval durations (5 min, 152 s, and 64 s) while also considering the potential effect of hay supplementation on classification. Our results suggest that investigators should ascertain whether their focal elk herd accesses or might access supplemental hay before deployment and analysis of activity sensor data. Similarly, researchers must weigh priorities when choosing a sampling interval, because no optimal solution emerged from our investigation. For example, of our acceptable models, only those constructed using 64-s intervals were able to distinguish short bouts of running. However, only models constructed with 5-min intervals accurately classified browsing while also maximizing the number of behaviors identified.