Two musicians play a duet, but each hears only their own part. The atmosphere and ocean in a climate model are like such a duet: they exchange heat and moisture every few hours. The new model SamudrACE has, for the first time, made them play together for centuries without missing a beat.

Scientists took two ready-made neural networks: one mimics the atmosphere, the other the ocean. Each was trained separately on 155 years of climate model simulation. Then they connected them through a simple exchange: the atmosphere hands over heat, moisture, and wind stress fluxes to the ocean every 6 hours, while the ocean averages them over 5 days and returns surface temperature and ice concentration. After connecting, the model was fine-tuned so both parts get used to each other.

The key result: the model reproduces El Niño, the warm current in the Pacific that influences weather worldwide. Previously, no coupled neural network model had managed this phenomenon, because it arises only from the interaction between atmosphere and ocean. Now that the duet plays in harmony, scientists can quickly test climate hypotheses and prepare the model for scenarios with rising carbon dioxide.