Coupling spectral analysis and hidden Markov models for the segmentation of behavioural patterns

Area of Science:

• Movement ecology
• Bio-logging
• Animal behavior

Background:

• Animal movement patterns reflect behavioral switches tied to environmental factors.
• Extracting behavioral signals from movement time series requires objective quantification tools.
• Hidden Markov Models (HMMs) are commonly used but lack direct cyclic pattern analysis.

Purpose of the Study:

• Develop a novel approach to extract new metrics of cyclic behaviors and activity levels from movement time series.
• Integrate spectral signatures of cyclic patterns into an HMM framework for latent behavioral state identification.
• Objectively quantify and classify animal behaviors from high-resolution movement data.

Main Methods:

• Applied time-frequency analysis to extract spectral signatures of cyclic behaviors and activity levels.
• Implemented these spectral signatures within a Hidden Markov Model (HMM) framework.
• Utilized 40 high-resolution European sea bass depth time series for illustration.

Main Results:

• Identified distinct activity regimes in European sea bass, associated with environmental cycles.
• Observed tidal rhythms correlating with reduced activity and shallower dives.
• Detected diurnal behavior linked to increased activity and deeper water column usage.

Conclusions:

• The method combines spectral analysis and HMMs for automated behavioral extraction from movement time series.
• The approach is suitable for large datasets and can incorporate environmental variables for enhanced analysis.
• This technique can advance understanding of habitat use, migration, and conservation strategies in movement ecology.