Satellite artificial intelligence can track shallow coral reef structures from space but cannot reliably follow routine annual changes. Many scientists assumed global satellite models would automatically monitor yearly ecological health across fragile ocean systems. Instead, spaceborne sensors gather sunlight bouncing off the submerged seabed over twelve months and compress that data into numerical vectors for each ten-meter square.
Water absorbs and scatters optical wavelengths, which alters the light patterns returning to orbiting detectors. The model functions like a long-exposure camera that blends seasonal water shifts into a single representative fingerprint for each patch of sea floor. Linear mathematical formulas then decode these numerical fingerprints to separate hard coral from sand. Damaged reefs with dead coral cover often reflect light similarly to living coral, confusing the decoder after disturbances.
Researchers tested the AlphaEarth model against 104668 underwater survey images collected across Great Barrier Reef locations. The system achieved a correlation score of 0.80 for mapping hard coral at fifty-meter resolution. During a severe 2024 heating event, the system detected ten-point coral loss blocks with an accuracy score of 0.83, but quiet years showed no detectable signal.
The researchers report that conservation teams can now deploy validated satellite channels to map stable seabed geography and flag severe bleaching disasters. However, tracking gradual reef recovery or subtle annual trends will still require local calibration and direct underwater field surveys.