Group and State-specific Estimation in Animal Movement Models Using a Bayesian Approach

This research study investigates statistical approaches for modeling the movement patterns of white-tailed deer in Louisiana using GPS tracking data. We start by classifying the latent behavioral states using a hidden Markov model (HMM) and then integrate those inferred states into a state-dependent step selection framework to evaluate the land cover preferences. Standard HMMs, however, assume the same movement patterns for all animals, overlooking the differences due to characteristics such as sex, age, breeding season, etc. To address this limitation, we extend the modeling framework to incorporate the group-level structure defined by similar characteristics or conditions to assess whether such unobserved attributes, combined with the behavioral states, affect the animal movement. Unlike other studies that incorporate such characteristics in transition probabilities and state-dependent distributions, our model provides a direct estimation of the state-dependent parameters governing the animal movement for the separate groups. We use a Bayesian inference framework with Markov chain Monte Carlo (MCMC) methods to estimate such group and state-specific parameters. To examine the performance of our model in parameter estimation, we conduct simulation studies to evaluate true parameter recovery and compare our Bayesian estimates to the traditional maximum likelihood estimates (MLEs). The results show that our model yields estimates with lower bias and smaller standard errors than MLEs. We then apply the group and state-specific model to the white-tailed deer data and carry out posterior predictive checks to evaluate model fit. The results show that our model effectively captures the central tendency of the observed data, although some refinement may be needed to allow greater flexibility in capturing the variability and distributional characteristics of step lengths, particularly at the lower end. Ultimately, our work provides a robust framework in ecology for studying the latent characteristics defined as groups, along with the behavioral states. We believe that this study not only deepens our understanding of animal movement but also holds potential applications across other domains beyond ecology. Advisor: Stephen Kachman