A doctor who has followed the same patient for years learns to predict flare-ups from subtle signs: a slight pallor, labored breathing, restless sleep. The ocean is just such a complex patient, only instead of a thermometer it has thousands of buoys and satellites, and instead of a medical history, seventy years of daily measurements. The neural network created in Australia took that history and learned to look two years ahead, predicting where the water will become anomalously warm.
Previously, long-term ocean temperature forecasting hit a wall: physical models that honestly simulate air-sea interactions worked reliably only three to four seasons ahead. Regional anomalies such as marine heatwaves remained almost unpredictable. The neural network bypassed this limitation differently: it does not attempt to reproduce all the physics but learns from existing examples, like an experienced clinician who has seen thousands of patients and recognizes the disease from a combination of symptoms.
Training ran on four graphics processing units, and the model processed data from 1950 to 2021. It looked at sea surface and air temperature maps, compressed them into a compact representation, and then expanded them back into a forecast. The error stayed below 0.75 °C even at month 24, which is roughly the precision with which we can distinguish warm from cool water by touch.
The model predicted most of them many months before their peak. The Blob heatwave, which in 2014–2015 killed sea lions and crabs off the California coast, was seen by the neural network two years in advance. However, the Ningaloo Niño off the western coast of Australia proved a miss: the training set contained too few similar cases, and the network simply lacked examples.
A doctor who has followed a patient for years eventually begins to warn of attacks before they happen. A neural network fed on seventy years of ocean observations does the same: it sees how a warm patch off the coast of Peru portends drought in Australia, and a cold tongue in the Pacific means a rainy summer in India. And although the ocean's rare diseases are still beyond its reach, its overall diagnosis is becoming increasingly confident.
The model was able to predict the Blob marine heatwave 24 months before its peak, even though this phenomenon was considered elusive due to its irregular seasonal timing. However, with Ningaloo Niño off the western coast of Australia, the neural network did worse: there were too few similar cases in the training data.
The idea of predicting climate anomalies with machine learning echoes Kim Stanley Robinson's novel 'The Ministry for the Future', where algorithms help humanity respond to extreme heat and climate disasters. Here, the neural network acts as an early warning system that sees the onset of El Niño two years in advance.