Validation of mercury tip-switch and accelerometer activity sensors for identifying resting and active behavior in bears

• Published in: Ursus (International Association for Bear Research and Management) 2015-12, Vol.26 (2), p.86-96
• Main Authors: Ware, Jasmine V, Rode, Karyn D, Pagano, Anthony M, Bromaghin, Jeffrey, Robbins, Charles T, Erlenbach, Joy, Jensen, Shannon, Cutting, Amy, Nicassio-Hiskey, Nicole, Hash, Amy, Owen, Megan, Jansen, Heiko T
• Format: Article
• Language: English
• Subjects: activity sensors; biotelemetry; brown bear; collars; polar bear; satellite transmitters; Techniques; Ursus arctos horribilis; Ursus maritimus
• ISSN: 1537-6176; 1938-5439
• DOI: 10.2192/URSUS-D-14-00031.1

Activity sensors are often included in wildlife transmitters and can provide information on the behavior and activity patterns of animals remotely. However, interpreting activity-sensor data relative to animal behavior can be difficult if animals cannot be continuously observed. In this study, we examined the performance of a mercury tip-switch and a tri-axial accelerometer housed in collars to determine whether sensor data can be accurately classified as resting and active behaviors and whether data are comparable for the 2 sensor types. Five captive bears (3 polar [Ursus maritimus] and 2 brown [U. arctos horribilis]) were fitted with a collar specially designed to internally house the sensors. The bears’ behaviors were recorded, classified, and then compared with sensor readings. A separate tri-axial accelerometer that sampled continuously at a higher frequency and provided raw acceleration values from 3 axes was also mounted on the collar to compare with the lower resolution sensors. Both accelerometers more accurately identified resting and active behaviors at time intervals ranging from 1 minute to 1 hour (≥91.1% accuracy) compared with the mercury tip-switch (range = 75.5–86.3%). However, mercury tip-switch accuracy improved when sampled at longer intervals (e.g., 30–60 min). Data from the lower resolution accelerometer, but not the mercury tip-switch, accurately predicted the percentage of time spent resting during an hour. Although the number of bears available for this study was small, our results suggest that these activity sensors can remotely identify resting versus active behaviors across most time intervals. We recommend that investigators consider both study objectives and the variation in accuracy of classifying resting and active behaviors reported here when determining sampling interval.