An experienced navigator steers the ship by the map, but the map is not always accurate. A cabin boy who has long watched the navigator err learns to point out where the map lies. Together they plot a better course than either could alone. That is how the new El Niño forecast model works. El Niño is a warm current in the Pacific that every few years shifts rainfall and drought patterns across the globe.

The conventional model is the navigator. It remembers how water temperature and sea level relate from month to month. This relationship is simple, so the model works even when data are scarce, and we have only about 150 years of them. But it fails to see that warm El Niño events are stronger than cold ones, so it makes mistakes.

The neural network is the cabin boy. It is trained not to predict El Niño but to correct the navigator's errors. First, the navigator makes sixteen slightly different forecasts. The cabin boy examines them along with the ocean's history and says where the navigator went wrong. In this way, the model captures the difference between warm and cold events.

Tests showed that after just a hundred years of training, the hybrid predicts a year ahead with skill above 0.5, while the neural network without the navigator stays below 0.4. Only after five hundred years of training does the pure neural network catch up with the hybrid. Previously, researchers thought thousands of years of data were needed, but now it is clear that a century suffices if one model does the heavy lifting and another corrects its mistakes. The navigator steers the ship, the cabin boy whispers where the map lies, and the course becomes truer.