A Translator That Hears the Ocean a Year Before the Event ✛
Imagine a translator sitting in on a conversation between two parties: the ocean and the atmosphere. Usually, the translator conveys the meaning with a six-month delay, and in spring it loses the thread entirely. The new model, called CTEFNet, translates up to two years ahead and, on top of that, explains which remark turned the tide.
The Pacific speaks nine languages: surface temperature, heat stored in the upper layer, the thickness of the wind-mixed layer, salinity, pressure, two components of currents, and two components of wind stress. Earlier translators listened only to surface temperature, so they often mistook the opening of the conversation. CTEFNet listens to all nine channels at once. First, convolutional layers, like attentive ears, pick up local whispers in each part of the ocean; then a transformer, like an experienced simultaneous interpreter, weaves these whispers into a long sentence that stretches over months.
The translator doesn't just translate; it also raises its hand when it hears something important. Gradient sensitivity analysis works like backpropagation: it asks how much the forecast would change if you tweaked one word in the conversation just a little. This revealed that 11 months before the peak of El Niño, warm water accumulates in the western Pacific, and the upper mixed layer becomes thinner. By 8 months, the signal spreads to the central and eastern ocean and to the eastern Indian Ocean. At the same time, cooling in the northern tropical Atlantic weakens the trade winds, and this atmospheric bridge nudges the conversation toward its climax.
The spring barrier, which used to sharply degrade forecasts made from March to May, is now softened: for forecasts from June to December, correlation above 0.5 holds for more than 20 months, and for spring forecasts up to 16 months. The translator still finds spring harder, but it no longer falls silent.
Now the same translator is ready to take on other climate conversations: distinguishing between El Niño types, the canonical and Modoki, or studying the asymmetry between El Niño and La Niña. And above all, it returns to the beginning: to how the ocean and atmosphere negotiate with each other, and to what we are finally beginning to understand of their language.
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.