Prediction of wolf (Canis lupus) kill-sites using hidden Markov models

We explore hidden Markov models (HMMs) as a predictive modeling technique and assess the degree to which models encapsulate movement and kill-site behavior in three wolf packs that reside year-round in west central Alberta, Canada. Although HMMs have been used successfully to infer use of space by individual woodland caribou, their behavioral states remained hidden. In our study, global positioning satellite (GPS) radio-collars allowed us to derive data (distance-between-locations, turning angle and travel rate) meaningful to hidden Markov modeling techniques. Aerial relocation allowed confirmation of kill-site locations (uncover the hidden states). The primary objective of our study was to evaluate whether HMMs could predict observer-confirmed kill-sites solely from the GPS wolf relocation data (correctly discover the hidden states). The Markov structure inherent in the HMMs provided additional insight into wolf behavior, such as bedding and relocating. We discuss the potential of using HMMs to determine predation rates on populations of wild ungulates and compare model signatures between packs preying on different ungulate species.