An analysis of pilot whale vocalization activity using hidden Markov models

Area of Science:

• Marine biology
• Bioacoustics
• Statistical modeling

Background:

• Cetacean vocalizations are crucial for social interactions.
• Whale behavior states are not directly observable, necessitating advanced modeling techniques.
• Latent-state models offer a promising approach for analyzing animal communication data.

Purpose of the Study:

• To apply hidden Markov models (HMMs) to analyze vocalization activity in long-finned pilot whales (Globicephala melas).
• To identify and characterize distinct behavioral states influencing whale calling patterns.
• To investigate the impact of covariates on state-switching dynamics in pilot whale communication.

Main Methods:

• Utilized hidden Markov models (HMMs) to analyze three years of vocalization data from pilot whales in Norway.
• Developed baseline HMMs to establish the presence of distinct behavioral states.
• Fitted more complex HMMs incorporating covariates to explore state-switching influences.

Main Results:

• Identified three distinct latent behavioral states driving pilot whale vocalization patterns.
• Demonstrated that HMMs can accurately describe complex stochastic patterns in animal communication.
• Showcased the influence of covariates on the dynamics of state transitions in whale behavior.

Conclusions:

• Hidden Markov models provide a powerful and concise framework for analyzing cetacean vocalizations.
• This approach facilitates meaningful biological inference regarding whale social interactions and behavior.
• The study highlights the utility of HMMs for understanding complex animal communication systems.