Objective classification of latent behavioral states in bio-logging data using multivariate-normal hidden Markov...

Area of Science:

• Ecology
• Quantitative Biology
• Bio-logging

Background:

• Ecological studies generate complex time-series data often affected by bias, autocorrelation, and noise.
• Objective classification of animal behavior from such data requires robust quantitative tools.

Purpose of the Study:

• To develop and validate a method for objective classification of complex ecological time-series data.
• To apply this method to high-resolution behavioral data from pelagic tuna.

Main Methods:

• Utilized multivariate-normal Hidden Markov Models (HMMs) with existing estimation techniques.
• Assessed the HMMs through simulation experiments and application to tuna bio-logger data.
• Incorporated covariate information, including diurnal effects, into the HMM analysis.

Main Results:

• HMMs accurately recovered known parameter values in simulations, achieving 90-97% correct classification rates.
• Analysis of tuna data revealed two distinct behavioral states: a shallow, warm state and a deeper, colder state.
• Predicted marked diurnal behavioral switching in tuna, consistent with previous studies.

Conclusions:

• Hidden Markov Models offer interpretable and objective classification for noisy, autocorrelated ecological data.
• The method provides a framework for analyzing bio-logging and other imperfect behavioral datasets.
• This approach facilitates hypothesis testing in diverse ecological systems.