Causes and Consequences of Timing Errors Associated With Global Positioning System Collar Accelerometer Activity Monitors

Abstract

Abstract Direct behavioral observations of multiple free-ranging animals over long periods of time and large geographic areas is prohibitively difficult. However, recent improvements in technology, such as Global Positioning System (GPS) collars equipped with motion-sensitive activity monitors, create the potential to remotely monitor animal behavior. Accelerometer-equipped activity monitors quantify animal motion with different amounts of movement presumably corresponding to different animal activities. Variations in motion among species and differences in collar design necessitate calibration for each collar and species of interest. We paired activity monitor data collected using Lotek GPS_4400 collars worn by captive Rocky Mountain elk Cervus elaphus nelsoni with simultaneously collected behavior observations. During our initial data screening, we observed many sampling intervals of directly observed behavior that did not pair to activity monitor data in a logical fashion. For example, intervals containing behaviors associated with little or no motion sometimes aligned with relatively high activity monitor values. These misalignments, due to errors associated with collar timekeeping mechanisms, would likely result in inaccurate classification models. We corrected timing errors by using defined breaks in animal behavior to shift times given by collar output, improving the average correct classification rate 61.7 percentage points for specific behaviors. Furthermore, timing errors were significantly reduced by increasing the GPS fix rate, by using a sampling interval divisible by 8 seconds, and by accurately timing the initial collar activation. Awareness and management of collar timing error will enable users to obtain the best possible estimates of true behavior when calibrating these collars and interpreting data from free-ranging animals.