NeuralGCM: Hybrid ML + Physics Atmospheric General Circulation Model New

NeuralGCM (Kochkov et al., 2024; Google DeepMind; Nature) is a hybrid model of the Earth's atmosphere that couples a neural-network parameterization of sub-grid physical processes with a traditional, differentiable atmospheric dynamical core. By keeping the well-understood dynamics on a stable numerical solver and learning only the uncertain parameterization, it achieves both physical consistency and data-driven accuracy.

The project is distributed as a Python library (github.com/neuralgcm/neuralgcm). Licensing: the code is under the Apache License 2.0, and the trained model weights are released under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.

Why it matters for physics AI: NeuralGCM shows a pragmatic path for physics AI - augmenting, rather than replacing, classical numerical weather prediction with learned components. It delivers competitive skill at a fraction of the computational cost and demonstrates that hybrid models can be stable over long climate-scale integrations.