Quantifying kill rates remains a key limitation to addressing many predator-prey questions. To date the most common approach to identify lion kill sites is done by identifying clusters of locations obtained using Global Positioning System (GPS) collars on predators. However, if clusters were determined by different decision rules, comparison across studies were not possible. This study investigated Hidden Markov Models (HMMs) as a predictive modelling technique. The Hidden Markov Model (HMM) is a stochastic model in which the system is modelled by a Markov process with “hidden” states. The states of the Markov chain can be interpreted as providing rough classifications of the animal behavioural dynamics. HMMs have been previously used in the analysis of wolves (Canis lupus), where aerial surveys were conducted to locate wolf-killed ungulate carcasses. The objectives of this project were to evaluate whether HMMs can predict observer-confirmed kill sites from GPS lion relocation data and also provide additional insight into lion behaviour. GPS radio-collars have been used to obtain movement data of for six lions (Panthera leo) in the Kruger National Park, South Africa. HMMs were fitted for each lion and a set of states were predicted, from which the behaviours were inferred. This state sequence was used to identify potential kill sites, and the actual kill data were then used to confirm these sites. The observer-confirmed kill sites provide a unique opportunity to validate predicted kill sites from HMM modelling using GPS tracking data as it is seldom possible to validate the state predictions from an animal movement study.
Reference: Poongavanan, J. (2017). A novel approach at uncovering the latent movement states and predicting kill-sites of lions using Hidden Markov Models. Bachelor’s Thesis, Nelson Mandela Metropolitan University.