How to Hear the Ocean a Year Before the Event ✛
A translator listens to the conversation between the ocean and the atmosphere. Previously, it heard only one word out of nine, the sea surface temperature, so it translated late and got confused in spring. The new CTEFNet model listens to all nine words at once: temperature, heat content, upper-layer thickness, salinity, pressure, currents, and wind. First, convolutional layers pick up local rustles in each ocean region, then a transformer weaves them into a long sentence spanning months ahead.
The translator not only translates but also raises its hand when it hears an important word. Gradient analysis asks: if we tweak one word slightly, how much does the forecast change? This revealed that 11 months before the peak of El Niño, warm water accumulates in the western Pacific and the upper layer becomes thinner. By 8 months, the signal spreads eastward and into the Indian Ocean. Cooling in the Atlantic weakens the trade winds and nudges the event along. The spring barrier, which used to degrade forecasts in spring, is now softened: the translator no longer falls silent. The same method can be applied to other climate phenomena to better understand the language of the ocean and atmosphere.
The 1997–1998 El Niño was so strong that global sea surface temperature rose 0.2°C above the decadal average, and sea level in the eastern Pacific temporarily rose by 20 centimeters due to thermal expansion of water.
In Kim Stanley Robinson's novel 'Forty Signs of Rain', climatologists struggle with the consequences of a sudden change in ocean currents caused by warming; accurate long-term El Niño forecasting could be the tool the characters use to prevent disaster.