Using Accelerometer Data to Remotely Assess Predation Activity of Arctic Wolves

• Main Author: Shipp, Heather
• Format: Text
• Language: unknown
• Published: Utah State University 2016
• Subjects: Arctic; Ellesmere Island; Fosheim Peninsula; Canis lupus
• Online Access: https://dx.doi.org/10.26076/b3f1-f640; https://digitalcommons.usu.edu/honors/545

Arctic wolves (Canis lupus arctos) play an important role in ecosystems located in the far northern regions of the world; however, unlike the gray wolves in Yellowstone National Park, little information is available about High Arctic wolves and their impacts on prey populations. This research uses data received from two GPS radio-collared Arctic wolves located in the Fosheim Peninsula on Ellesmere Island. Each radio-collar was programmed to record a position every 30-60 minutes, as well as the wolfs activity movement (forwards - backwards and left - right), which was generated by an accelerometer housed within the radio-collar. This research project focused on using location clusters and their associated activity data to remotely identify the locations and the frequency of wolf predation events. The activity data can be used to identify potential kill sites because it takes both time and energy for the Arctic wolves to take down and consume their prey, thus clusters of locations with high levels of activity are generated at these places. Over fifty of the cluster sites were visited and assessed for remains of a kill, such as bone remnants, teeth, or hair. A key objective of this study was to identify predictors and develop a statistical model that distinguishes kill sites from non-kill sites, including rendezvous sites, which I also analyzed. I used AIC model selection methods to compare different multinomial logistic regression models that measured the probability a cluster included a kill, a rendezvous, or neither as a function of several variables, including the sum of activity, total timespan of the cluster, average activity, and the initial slope in activity within the first few hours of each cluster, which is the rate at which activity decreased following the establishment of the cluster. The most predictive variable was number of points; other useful predictors included the average distance between each point and the cluster centroid, and the average value in sideways and rotary acceleration (Activity Y) across the cluster lifespan. These three variables comprise the best-fit multinomial model to distinguish kill and rendezvous clusters, as supported by the AIC results. When excluding the rendezvous clusters, the best-fit multinomial model included the three variables (number of points, average distance, and average in Activity Y) in addition to the slope in activity within the first two hours since cluster formation. Use of accelerometer data and multinomial logistic regression models may help differentiate clusters and enable scientists and wildlife managers to remotely monitor the predatory impact of Arctic wolves.