The MAMBAT framework for acoustic tracking of multiple animals

Area of Science:

• Marine Biology
• Bioacoustics
• Acoustic Signal Processing

Background:

• Passive acoustic monitoring (PAM) is crucial for marine mammal research, but generates complex data with overlapping signals.
• Automated tools are needed to handle challenges like multiple sources, missed detections, and false alarms in PAM data.

Purpose of the Study:

• To introduce the Multiple-Animal Model-Based Acoustic Tracking (MAMBAT) framework for automated multi-source sound tracking.
• To demonstrate MAMBAT's capability in analyzing real-world marine mammal acoustic data.

Main Methods:

• Integration of model-based localization with Bayesian multi-target tracking.
• Implementation of a "Track-before-Localize" followed by a "Localize-then-Track" strategy.
• Elimination of the need for explicit detection, classification, or association steps.

Main Results:

• Successful automatic tracking of multiple sound sources from complex acoustic data.
• Demonstrated effectiveness on real-world datasets of sperm whales from two ocean basins.
• Validation of a novel approach that bypasses traditional detection and association steps.

Conclusions:

• MAMBAT provides an advanced automated solution for tracking multiple marine mammals using PAM data.
• The framework enhances the ability to monitor marine mammal distribution, abundance, and behavior.
• MAMBAT offers valuable insights for marine conservation and management strategies.