Editors’ Highlights are summaries of recent papers by AGU journal editors.
sauce: Journal of Advances in Modeling Earth Systems
Atmospheric models must represent processes over many orders of magnitude spatial scales. Small-scale processes such as thunderstorms and turbulence are important to the atmosphere, but are computationally expensive in most global models and cannot be addressed explicitly. In traditional models, heuristic estimates of the effects of these processes, called parameterizations, are designed by experts. A recent body of work used machine learning to create data-driven parameterizations directly from very high-resolution simulations, requiring fewer assumptions.
Yuval and O’Gorman [2023] We provide the first example of a neural network parameterization of the effect of subgrid processes on the vertical transport of momentum in the atmosphere. Considering the nuances of horizontal gridding in high-resolution models, a careful approach is taken to generate the training dataset. The new parameterization generally improves wind simulation on coarser resolution models, but overcorrects and biases heavily in one configuration. This work serves as a complete and clear example for researchers interested in applying machine learning for parameterization.
Quote: Yuval, J. & O’Gorman, PA (2023). Neural network parameterization of atmospheric subgrid momentum transport. Journal of Advances in Modeling Earth Systems15, e2023MS003606. https://doi.org/10.1029/2023MS003606
—Oliver Watt-Meyer, Associate Editor, James
Text © 2023. Author. CC BY-NC-ND 3.0
Images are subject to copyright unless otherwise noted. Reuse without the express permission of the copyright owner is prohibited.
Related
