Source: Journal of Advances in Modeling Earth Systems
Atmospheric models must represent processes on spatial scales spanning many orders of magnitude. Although small-scale processes such as thunderstorms and turbulence are critical to the atmosphere, most global models cannot explicitly resolve them due to computational expense. In conventional models, heuristic estimates of the effect of these processes, known as parameterizations, are designed by experts. A recent line of research uses machine learning to create data-driven parameterizations directly from very high-resolution simulations that require fewer assumptions.
Yuval and O’Gorman  provide the first such example of a neural network parameterization of the effects of subgrid processes on the vertical transport of momentum in the atmosphere. A careful approach is taken to generate a training dataset, accounting for subtle issues in the horizontal grid of the high-resolution model. The new parameterization generally improves the simulation of winds in a coarse-resolution model, but also over-corrects and leads to larger biases in one configuration. The study serves as a complete and clear example for researchers interested in the application of machine learning for parameterization.
Citation: Yuval, J., & O’Gorman, P. A. (2023). Neural-network parameterization of subgrid momentum transport in the atmosphere. Journal of Advances in Modeling Earth Systems, 15, e2023MS003606. https://doi.org/10.1029/2023MS003606
—Oliver Watt-Meyer, Associate Editor, JAMES
Text © 2023. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.