NOAA’s GAIA project trains AI to spot endangered whales from space

Counting whales is one of the quiet, essential chores of ocean conservation — and one of the hardest. The animals range across enormous, remote stretches of sea, surfacing only briefly, and traditional aerial and ship surveys are expensive, weather-dependent and can cover only a sliver of habitat at a time. A NOAA initiative called GAIA — Geospatial Artificial Intelligence for Animals — is testing a way to widen that view dramatically, using satellites and machine learning to find whales from space.

Developed by NOAA’s Northeast and Alaska Fisheries Science Centers, GAIA integrates very-high-resolution satellite imagery with cloud computing, geospatial analysis and machine learning to build a scalable, automated detection system. The initiative focuses first on two endangered species chosen for their conservation urgency: the North Atlantic right whale, one of the most imperiled large whales on Earth, and the Cook Inlet beluga in Alaska. Because these populations are small and every individual matters, knowing where the animals are, and when, is directly tied to protecting them.

“The animals range across enormous, remote stretches of sea, surfacing only briefly, and traditional aerial and ship surveys are expensive, weather-dependent and can cover only a sliver of habitat at a time.”

The workflow blends automation with careful human oversight. Cloud-optimized imagery — including from commercial satellites such as Maxar’s WorldView-3 — is retrieved and preprocessed, then reviewed by multiple independent expert annotators who validate detections before they enter the dataset. That human-verified data both improves the AI models and ensures the results are trustworthy. The effort draws on an unusually broad set of partners, including the U.S. Geological Survey, Microsoft AI for Good, the Naval Research Laboratory and the British Antarctic Survey.

Better, faster whale detection has concrete conservation payoffs. Knowing where right whales are gathering can inform ship-speed rules and fishing measures that reduce deadly vessel strikes and gear entanglements — the two leading human causes of their deaths. A system that can scan large ocean areas repeatedly, rather than waiting for a survey plane, could give managers a far more current picture of where the animals actually are.

GAIA remains a research initiative under active development, and detecting a whale-sized object in a satellite image of open ocean is a genuinely hard technical problem that the team is still refining. But the direction is inspiring: pointing some of the same satellite and AI capabilities used for other purposes at the task of protecting endangered wildlife — quietly, at planetary scale, one validated detection at a time.