The ocean is a vast aquatic memory. Everything that happens in it lingers in currents and temperatures for years to come. To understand what lies ahead, you need to read that memory. In the past, this was done with complex programs that describe ocean physics through equations. But such programs take a long time to run and sometimes get things wrong. Now there is another way: a neural network that learns from examples, much like a child learns to speak by listening to adults.

This neural network is called ORCA-DL. It is shown what the ocean looked like in the past, including temperature, salinity, and currents, and what it became months or years later. After thousands of such examples, it begins to predict the future on its own. And it does not just predict the surface; it also captures the depths, and it does so better than traditional models. For instance, it more accurately forecasts El Niño, the warm current that affects weather across the entire planet.

The most important thing is that ORCA-DL can look ten years ahead without losing its way. It is as if, by looking at today's weather, you could tell what the climate would be like a decade from now. Such forecasts help us prepare for marine heatwaves that harm fish and corals, and plan the operations of ports and fishing fleets. Water remembers, and now a neural network helps us read that memory.