Weather and climate predicted by hybrid model

On page 1060, Kochkov etai? report a hybrid global atmospheric model that overcomes these challenges, combining physics-based approaches with machine learning. The hybrid model could also run longer simulations (40-year projections) to estimate climate statistics, although some of these simulations (15 out of 37) were not stable throughout the 40-year period - a problematic feature that is not typical of conventional models. [...]when perturbed with a strong, uniform sea surface warming of 4 °C, the model developed by Kochkov and colleagues predicts unrealistic changes in atmospheric temperature - disagreeing with the expectations of fundamental physics and with conventional atmospheric models.