Record Details
Ungulate activity classification : calibrating activity monitor GPS collars for Rocky Mountain elk, mule deer, and cattle
ScholarsArchive at Oregon State University
| Field | Value |
|---|---|
| Title | Ungulate activity classification : calibrating activity monitor GPS collars for Rocky Mountain elk, mule deer, and cattle |
| Names |
Gaylord, Adam J.
(creator) Sanchez, Dana M. (advisor) |
| Date Issued | 2013-06-24 (iso8601) |
| Note | Graduation date: 2014 |
| Abstract | Ungulate behavior has been studied extensively but direct observation of free-ranging animals over long periods of time and large geographic areas is often prohibitively difficult. Improved technology, such as GPS collars fitted with motion-sensitive activity monitors, provides researchers with a potential tool to remotely collect fine scale activity and location data. Activity monitors record animal movement along one or more axes with different amounts of motion presumably corresponding to different animal behaviors. Inter- and intraspecific variations in motion among behaviors necessitate calibration for each focal species. Calibration generally consists of making detailed behavioral observations of captive collared animals and then pairing observed behaviors with collar activity data for the same sampling interval. This process results in a mathematical model that can be used to classify the activity level or behavior of novel free-ranging animals using remotely collected collar data. During the calibration process, we discovered that several factors associated with the time-keeping mechanisms of these collars can result in mismatches between collar activity monitor data and direct behavior observation. This results in inaccurate classification models. To correct for these timing errors, we used defined breaks in animal behavior to shift collar output times, improving the average correct classification rate up to 61.7 percentage points for specific behaviors. We also learned that timing errors can be minimized by activating a collar's GPS unit, increasing the GPS fix rate, and using a sampling interval divisible by 8 seconds. Awareness and management of collar timing issues will enable managers and researchers to best classify animal behavior when using these collars and interpreting data from free-ranging animals. No activity monitor calibration had been conducted for Lotek 4400 GPS collars featuring dual-axis activity monitors for Rocky Mountain elk (Cervus elaphus nelsoni), mule deer (Odocoileus hemionus), or cattle (Bos taurus). We used discriminant function models to determine what behaviors can be accurately classified using these collars. Additionally, we constructed models using only pure intervals (sampling intervals during which only one behavior occurred) and applied them to datasets containing only mixed intervals (sampling intervals during which >1 behavior occurred) to determine the effect of excluding the latter from the calibration process. Final full-dataset models accurately classified (correct classification rates > 70%) up to 4 behavior categories for elk, 3 for deer, and 2 for cattle. Our results showed that classification models constructed with only pure intervals can result in misclassification rates of up to 61% for mixed intervals of some behaviors. When remotely collecting data, researchers must balance sampling frequency with the battery life of the recording device. The duration of each behavior relative to sampling interval length might play an important role in activity monitor calibration. To date no efforts have been made to determine the optimal sampling interval duration to use with these sensors. Similarly, Lotek 4500 GPS collars featuring accelerometer activity monitors had not been calibrated for Rocky Mountain elk. We examined discriminant function model structures for 3 sampling interval durations (5-min, 152-sec, and 64sec) to determine what behaviors can be accurately classified for animals with and without access to supplemental feed in the form of hay. Models constructed using 5-min sampling intervals performed best, accurately classifying (≥ 70% classification rate) up to 5 behaviors for animals without access to supplemental feed and 4 behavior categories for those with access to supplemental feed. All of our calibration models will be made available on-line, allowing managers and researchers to interpret data from novel free-ranging animals for use in ongoing and future studies of ungulate ecology and management. |
| Genre | Thesis/Dissertation |
| Access Condition | http://creativecommons.org/licenses/by-nd/3.0/us/ |
| Topic | Ungulate |
| Identifier | http://hdl.handle.net/1957/40804 |
