A Posteriori Learning for Quasi‐Geostrophic Turbulence Parametrization

The use of machine learning to build subgrid parametrizations for climate models is receiving growing attention. State-of-the-art strategies address the problem as a supervised task and optimize algorithms that predict fluxes based on information from coarse resolution models. In practice, training data are generated higher numerical simulations transformed in order mimic simulations. By essence, these meet so-called priori criteria. But actual purpose parametrization obtain good performance terms posteriori metrics which imply computing entire model trajectories. this paper, we focus representation energy backscatter two-dimensional quasi-geostrophic turbulence compare obtained with different at fixed computational complexity. We show criteria yield tend be unstable direct describe how can alternatively trained end-to-end illustrate outperform known empirical data-driven schemes performance, stability, ability apply flow configurations. These results support relevance differentiable programming paradigms future.