The ocean's green is mostly hidden by clouds — a self-taught AI just learned to fill the gaps

Satellites read the colour of the sea to follow phytoplankton — the tiny drifting plants that feed ocean life and soak up carbon. But clouds and patchy coverage hide most of the picture: in the records this team studied, about 85% of observations were missing. Researchers at the Institute of Oceanology, part of the Chinese Academy of Sciences, have built GDCM-EOF, an AI that teaches itself from the incomplete records themselves, with no need for a ready-made complete map to learn from. In tests where real readings were deliberately hidden and then checked, its reconstruction error was about 82% lower than the standard method in the Caribbean Sea and about 40% lower worldwide — sharp enough to keep features like river plumes and swirling eddies visible instead of smoothing them away. More continuous records could help scientists track how the ocean's living surface shifts as the climate changes.

SDG 14 · Life below water — Mechanism: tech innovation