AI-powered buoy listens for endangered killer whales
System could eventually alert ships in time to slow down and reduce underwater noise
An AI-powered acoustic buoy in the Salish Sea is being tested as a way to detect endangered southern resident killer whales in near-real time, potentially giving commercial ships enough warning to slow down and reduce underwater noise.
Only 76 southern resident killer whales remain. The population enters Washington's Puget Sound region to hunt Chinook salmon, but heavy vessel traffic can interfere with the whales' ability to hunt, communicate and echolocate. Ships also pose a collision risk. A 2024 study found that for every one-decibel increase in maximum noise, females were 58% less likely to pursue prey, while males and females were 12.5% less likely to successfully capture prey.
The project brings together Quiet Sound, SMRU Consulting and Microsoft. Their Communications Acoustics Buoy was deployed in July in the Haro Strait, which spans the U.S.-Canada border. It uses three underwater hydrophones along with Microsoft's Water SPARROW edge-computing technology and SMRU's acoustic analysis system to identify whale vocalizations.
Researchers say results have been encouraging, with false positives declining as the AI models are trained on more underwater recordings. The system, however, is not yet automatically sending detections to mariners because researchers want to further reduce the risk of false alerts.
The eventual goal is to connect detections with the Whale Report Alert System, operated by Ocean Wise. That system combines human sightings, thermal imaging and hydrophone data and can alert commercial marine pilots when they are within 10 nautical miles of a reported whale location. Large vessels can require about five miles to slow down, making early detection important.
The technology would complement Quiet Sound's existing voluntary vessel slowdown program. Quiet Sound says the program has reached 73% participation this season and has reduced underwater noise by as much as 50% since it was introduced in 2022.
Researchers ultimately hope the same approach can combine multiple detection methods and be expanded to monitor other marine species.