AI Model Detects Lion Roars from Collar Movement Data

A new machine learning model accurately detects lion roars using only movement data from GPS collars, eliminating the need for power-intensive audio recorders.

Scientists have developed an artificial intelligence model that identifies lion roars exclusively from accelerometer data embedded in tracking collars. This method streamlines research into lion communication without relying on traditional audio recordings.

Traditional methods for studying lion roaring patterns present significant challenges. Lions use roars for long-distance communication within prides and for territorial marking. However, collecting comprehensive data on this behavior with traditional audio recording devices requires substantial energy and storage, often capturing irrelevant information. Accelerometer data, which tracks minute movements in three dimensions, offers an alternative.

Published in the journal *Ecological Informatics*, the U-Net model, developed by researchers from the GAIA Initiative at the Leibniz Institute for Zoo and Wildlife Research, analyzes these subtle movements. This machine learning approach achieves 90-96% accuracy in detecting lion roars from acceleration data. It also demonstrates approximately 81% precision, with false positive rates near 20%.

The model successfully identifies roars from both male and female lions, regardless of whether the animals are moving or stationary. During the study, researchers recorded 1,333 distinct lion roaring events. The U-Net model was trained using reference data from collared lions in Etosha National Park, which included both GPS accelerometers and audio loggers to create a comprehensive dataset.

This advancement means researchers can now analyze roaring behavior over extended periods without the logistical constraints of audio recording. By focusing solely on acceleration data, the new method significantly reduces battery consumption and data storage requirements on tracking collars. This allows for a deeper understanding of lion communication, including previously unstudied long-distance interactions among female lions and the spatial context of roaring.

Future research will likely apply this technology to unravel complex social dynamics and territorial strategies across diverse lion populations.