Animal movement modelling: Independent or dependent models?

Abstract

Hidden Markov models have become a popular time series method for the analysis of GPS tracked animals. Their advantage for identifying latent behavioural states compared with Independent Mixture models is that they take into account the time series dependency of successive displacement distances by the tracked animals. However, little is known about how the analysis results may differ depending on which of these approaches is used. We compared the results and interpretations obtained from fitting Hidden Markov and Independent Mixture models to simulated movement data as well as to field data recording the hourly movements of sable antelope and buffalo within the Kruger National Park, South Africa. Hidden Markov models consistently yielded narrower confidence intervals around parameters and smaller standard errors than simpler time independent mixture models, but for some data the improvement was marginal and the Independent Mixture model provided an adequate alternative for identifying the latent behavioural states of the animal. In general, it is expected Hidden Markov models will provide the better balance between model complexity and extensibility for animal movement modelling from a statistical perspective. However, in some cases, Independent Mixture models could provide an adequate alternative method and might be more faithful biologically. Keywords: Activity states, African buffalo, Hidden Markov models, Independent mixture models, Latent states