AI and Underwater Microphones Transform Right Whale Conservation

Scientists are using deep-learning models and underwater microphones to track North Atlantic Right whales with 86% accuracy—offering a breakthrough method for monitoring these critically endangered marine giants, even in challenging conditions.

Research: Acoustic abundance estimation for Critically Endangered North Atlantic right whales in Cape Cod Bay, Massachusetts, USA. Image Credit: lego 19861111 / Shutterstock

Using underwater microphones and machine learning (ML), Cornell University researchers have developed a new method to estimate North Atlantic Right whale numbers - offering a potentially safer and more cost-effective way to monitor this critically endangered species.

Their study, published in Endangered Species Research, demonstrates how microphones combined with ML and traditional aerial survey methods can help track right whale populations in Cape Cod Bay, a crucial feeding ground where the whales gather each spring.

Researchers rely on costly and dangerous airplane surveys or sound recordings to track this endangered species and identify their presence or absence.

The analysis pipeline consisted of a series of modular components. First, a trained analyst manually annotated upcalls in a subset of the recordings across 5 marine autonomous recording units (MARUs). The annotations were then used to evaluate the performance of an automated upcall detection and classification system. Next, the detections were associated across units through an automated time-difference-of-arrival (TDoA) procedure. Finally, the number of distinct upcalls was combined with visual observations from aerial surveys to fit a quasi-Poisson model and predict daily NARW abundance throughout the acoustic survey

"Using sound recordings to monitor whale populations isn't new," said lead author Marissa Garcia of the Cornell Lab of Ornithology's K. Lisa Yang Center for Conservation Bioacoustics. "What makes our study unique is that we were able to take those recordings and go beyond getting information on the presence or absence of whales to getting an approximate number of whales in an area."

The team set out an array of marine autonomous recording units (MARU) across Cape Cod Bay to capture right whale sounds.

Following the deployment of the MARUs, the team trained, validated, and applied a deep-learning model that could automatically detect right whale sounds with 86% precision.

"By analyzing their distinctive upcall vocalizations, we can detect their presence continuously, day and night," Garcia said. "This kind of round-the-clock monitoring that results from passive acoustic monitoring just isn't possible with traditional aerial surveys, which can only happen in daylight hours and in good weather."

Garcia says there's still some uncertainty in the counts that the team needs to address in future research. Still, the team is optimistic that monitoring whale vocalizations holds promise for estimating the abundance of right whales to aid in conservation and management efforts.

Having the ability to expand monitoring efforts across larger areas of the ocean will help scientists better assess the species' population numbers across the full extent of its range. Garcia said right whales have traditionally been considered a conservation challenge in New England, but right whales are found all along the East Coast.

"Using passive acoustic data and deep-learning tools, we can expand the area we can safely monitor and keep track of this critically endangered species," Garcia said.

The work comes at a critical time for North Atlantic right whales, whose population has declined to fewer than 370 individuals due to ship strikes, fishing gear entanglement, and changing ocean conditions affecting their food sources.