Modeling Complex Dependence Structures at Multiple Scales Using Hierarchical Hidden Markov Models

• Activity Number: 322 - Analyses in Ecology, Epidemiology, and Environmental Policy
• Type: Contributed
• Date/Time: Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Evan Sidrow* and Nancy Heckman and Sarah ME Fortune and Andrew W Trites and Ian Murphy and Marie Auger-Methe
• Companies: The University of British Columbia and The University of British Columbia and University of British Columbia and University of British Columbia and University of British Columbia and The University of British Columbia
• Keywords: accelerometer data; hierarchical modeling; animal movement; biologging; statistical ecology; time series

Abstract

Improved tagging technology has resulted in large, high-frequency data sets of marine animal diving behavior. While hidden Markov models (HMMs) are often used to model animal movement data, high-frequency diving data sets often exhibit complicated, multi-scale dependence structures that cannot be modeled using many modern HMMs. We detail a hierarchical approach that simultaneously labels dive types using a coarse-scale hidden Markov model and sub-dive behaviors using a more complicated fine-scale model. This fine-scale model may involve a variety of methods to deal with intricate dependence structures, including functional data analysis, Fourier analysis, and data transformations over moving windows. We use this approach to model the movement of a northern resident killer whale (Orcinus orca) off the coast of British Columbia, Canada. These results, together with our simulation study, show that our model produces more accurate parameter estimates and more interpretable state estimation compared to existing methods.