Every few years, a warm patch appears in the Pacific Ocean off the coast of South America. It is so large that it alters weather worldwide. Scientists call it El Niño. Predicting it is challenging: simple computer models accurately estimate how often it occurs but miss the details. Complex models capture more but frequently make mistakes and require extensive computation time.

The authors of the new study created a hybrid of two models. The first, a simple one, acts like a soloist in an orchestra: it sets the rhythm, indicating when to expect a warm event. The second, a complex model, behaves like the rest of the musicians: it follows the soloist and completes the full picture of the ocean. Inside the complex model runs a neural network, a computer program that learns from data. It was trained on a long simulation from another model and then fine-tuned on real observations.

The model was run for 1000 years, with the last 960 analyzed. The hybrid reproduced El Niño's correct rhythms, with activity peaks every 3–4 years. It also correctly positioned the warm patches in the ocean, correcting the complex model's errors. Previously, it was unclear how to reconcile simplicity and accuracy, but now it is evident: combining two models makes them work better together than separately. This approach will be useful for training forecasting programs and studying rare events.