An interpreter in a large hall hears both a whisper right at their ear and a conversation across the room. That's how ResoNet works: a neural network that predicts El Niño.
Previously, forecasts a year or more ahead were unreliable. ResoNet tackles the problem differently. One part of the network, convolutional layers, looks at local patches of warm water in the ocean, like an ear catching a nearby whisper. Another part, the transformer, links those patches to distant regions, like hearing that picks up a conversation across the hall. Together, they produce forecasts 19–26 months out.
The network doesn't just predict; it also explains itself. The integrated gradients method reveals which parts of the temperature map it focused on most. It turns out that for long-range forecasts, the southern Pacific and the Indian Ocean are key, while for shorter ones, the equatorial zone matters. This matches what scientists already know: warm water accumulates off the equator before El Niño, and the Indian Ocean releases heat with a delay. For the first time, a neural network has captured the difference between El Niño and La Niña: the warm phase depends more on the southern Pacific, while the cold phase depends more on the eastern equator. So the interpreter not only repeats what it hears but also tells you which voices in the hall were decisive.
Bagging, used in ResoNet, comes from 1990s statistics: it was proposed by Leo Breiman, the same author of the famous random forests. And the integrated gradients method is so named because it accumulates the contribution of each pixel along a path from a zero image to the actual one.
In Kim Stanley Robinson's novel 'Forty Signs of Rain', climate scientists desperately try to predict abrupt weather shifts. ResoNet is a step toward making such predictions routine rather than a plot of catastrophe.