A translator that learned the grammar of the wind, not just the vocabulary. The wind blows over the Pacific, and the water responds with waves, currents, and upwellings of cold depths. The KIST-Ocean neural network was taught not just to memorize how the ocean looked yesterday, but to understand why it will become what it is tomorrow.

Previously, it was unclear whether neural network ocean models could reproduce physically correct responses to atmospheric forcing, rather than just fitting statistics. KIST-Ocean was tested on idealized experiments: they applied wind stress and watched how the model reacted. It reproduced the eastward propagation of Kelvin waves and the westward propagation of Rossby waves, with Rossby wave speeds matching theoretical predictions for different latitudes. Cyclonic wind stress caused upwelling of cold water and surface cooling, while anticyclonic wind stress caused downwelling and warming, as predicted by Ekman theory.

The model outperformed the persistent forecast for 81.3% of targets in terms of anomaly correlation coefficient and for 86.5% in terms of RMSE. When forced with realistic boundary conditions, it maintained significant skill for up to 200 days for potential temperature, up to 85 days for zonal current at 145 m depth, up to 40 days for meridional current, and up to 100 days for salinity. Comparison with five NMME models showed that KIST-Ocean provides better sea surface temperature forecasts up to six months ahead under real forcing, and under climatological forcing it matches the best dynamical models for one to two months.

Now this translator can become part of a larger system where ocean and atmosphere communicate at every step. Then forecasts of El Niño and La Niña, which affect droughts, floods, and hurricanes worldwide, will become faster and more accessible. The wind keeps blowing, and the ocean now understands its grammar.