Establish stabilization of complex parameterizations by prognostic variables

Determine whether prognostic memory variables stabilize machine-learning parameterizations in configurations more complex than the online Lorenz-96 system.

Background

The study demonstrates online evaluation only for the Lorenz-96 system, while the precipitation experiment is conducted offline using coarse-grained, kilometer-scale atmospheric simulation data. The distilled linear ordinary differential equation is numerically stable and retains useful information during long rollouts in the tested settings.

The authors identify the absence of online tests in more complex atmospheric configurations as a limitation. It therefore remains unresolved whether adding prognostic variables will stabilize the parameterization itself when coupled to a more realistic dynamical core or Earth system model.

References

As a limiting factor, our online experiments are restricted to the L96 system, and it remains an open question whether these prognostic variables also stabilize the parameterization itself in more complex configurations.

— Learning Prognostic Variables for AI Convective Parameterizations via Symbolic Distillation  (2609.24882 - Schönfeld et al., 21 Sep 2026) in Section 6, Conclusion and Outlook