Vianey Leos Barajas, PhD in Statistics, 2019 from Iowa State University
Abstract: Sensors play an important role in the automated collection of ecological and environmental data. For example, sensors can record animal movement and environmental data over time, possibly at high temporal resolutions, which in turn can provide insights into underlying latent ecological and environmental processes of interest. As such, the class of hidden Markov models (HMMs) and Markov-switching processes provide an intuitive framework to relate that the observed data stems from a set of different underlying processes that further exhibit persistence over time, e.g. an animal’s observed movements are drastically different when it is resting vs actively traveling. One of the primary benefits of the general HMM framework is the ease in which it can be extended and modified. This framework facilitates the identification of processes occurring at multiple temporal scales and incorporation of domain-specific knowledge to provide more realistic statistical models of biological and environmental processes. In this talk, I will discuss (i) multi-scale modeling of animal movement data via HMMs, (ii) modeling feedback between a Merino sheep’s physiological dynamics and its movement across the landscape and (iii) coupled hidden Markov models for spatiotemporal wind data.