SurfPerch: an AI that listens to coral reefs to track whether they are recovering

A healthy coral reef is surprisingly noisy, alive with the snaps, grunts and chirps of fish and other creatures, while a degraded reef falls quiet. Researchers have turned that insight into an AI tool called SurfPerch, built by Google Research and Google DeepMind with marine biologists at the University of Bristol and University College London, that analyzes underwater recordings to judge how healthy a reef is far faster than scientists could by listening manually.

The model learned to hear reefs with help from the public. Through a project called Calling in Our Corals, hosted on Google Arts & Culture, thousands of people listened to more than 400 hours of reef audio and tagged the fish sounds they heard. That crowd-labeled data trained SurfPerch to recognize reef sounds, and the system can now be quickly adapted to detect a new sound from only a handful of examples, making it practical for researchers working in very different waters.

“Researchers have turned that insight into an AI tool called SurfPerch, built by Google Research and Google DeepMind with marine biologists at the University of Bristol and University College London, that analyzes underwater recordings to judge how healthy a reef is far faster than scientists could by listening manually.”

In early trials the tool produced concrete conservation insights. It distinguished protected reefs from unprotected ones in the Philippines, tracked restoration outcomes in Indonesia and picked up relationships in fish communities on Australia's Great Barrier Reef. Because the analysis is automated, scientists can monitor reefs continuously, survey deeper waters and measure whether restoration efforts are actually working, rather than relying on slow, expensive manual review.

The limits are worth keeping in view. Acoustic monitoring is one signal among many, and a richer soundscape does not by itself save a reef from warming seas, pollution or bleaching, the underlying threats that conservation must still confront. The model also needs good recordings and careful validation in each new location. But giving reef scientists a fast, scalable way to listen, built partly by ordinary volunteers, turns passive ocean sound into a practical early-warning and progress-tracking tool, and a hopeful example of AI serving the natural world.