Modelling reindeer rut activity using on-animal acoustic recorders and machine learning

Area of Science:

• Bioacoustics
• Wildlife Ecology
• Machine Learning in Biology

Background:

• Acoustic monitoring is crucial for wildlife research, but manual analysis of recordings is time-consuming.
• On-animal recorders offer unique insights into animal behavior and ecology.
• Integrating machine learning with on-animal acoustic data can overcome analytical bottlenecks.

Purpose of the Study:

• To evaluate the effectiveness of combining on-animal acoustic recorders with machine learning for wildlife behavior analysis.
• To investigate the vocalization patterns of rutting reindeer using advanced signal processing.
• To demonstrate the synergy between novel recording technology and AI in ecological studies.

Main Methods:

• Attached acoustic recorders to seven male reindeer during their 2019 and 2020 mating seasons.
• Utilized convolutional neural networks to analyze recorded reindeer vocalizations.
• Achieved 95% accuracy in distinguishing reindeer grunts using trained neural networks.

Main Results:

• Successfully trained machine learning models to identify reindeer grunts with high accuracy.
• Revealed distinct vocalization patterns in rutting reindeer.
• Identified that older, heavier male reindeer vocalized more frequently than younger, lighter males.

Conclusions:

• On-animal acoustic recorders coupled with machine learning are powerful tools for wildlife research.
• This approach significantly streamlines the analysis of animal vocalizations.
• The findings highlight the potential for broader applications in understanding animal behavior and ecology.