Use of neural networks for stable, accurate and physically consistent parameterization of subgrid atmospheric processes with good performance at reduced precision

TL;DR

This paper develops a neural network-based parameterization for subgrid atmospheric processes that is stable, physically consistent, and efficient, enabling accurate climate simulations at reduced computational cost.

Contribution

It introduces a physically constrained neural network approach that ensures stability and accuracy in climate model simulations, even with reduced numerical precision.

Findings

Neural network parameterization replicates high-resolution climate with similar accuracy to random forests.

The approach maintains stability across various architectures and resolutions.

Reduced precision neural networks can lower computational costs without sacrificing quality.

Abstract

A promising approach to improve climate-model simulations is to replace traditional subgrid parameterizations based on simplified physical models by machine learning algorithms that are data-driven. However, neural networks (NNs) often lead to instabilities and climate drift when coupled to an atmospheric model. Here we learn an NN parameterization from a high-resolution atmospheric simulation in an idealized domain by coarse graining the model equations and output. The NN parameterization has a structure that ensures physical constraints are respected, and it leads to stable simulations that replicate the climate of the high-resolution simulation with similar accuracy to a successful random-forest parameterization while needing far less memory. We find that the simulations are stable for a variety of NN architectures and horizontal resolutions, and that an NN with substantially reduced numerical precision could decrease computational costs without affecting the quality of simulations.