On the Choice of Training Data for Machine Learning of Geostrophic Mesoscale Turbulence

Aug 4, 2026

Upload by Eliana

Data choice is central for machine learning, yet it is often not the primary focus in studies applying neural models to Earth system modeling. This research examines eddy–mean interaction in rotating stratified turbulence with lateral boundaries, where the rotational component of eddy flux is known not to contribute directly to sub-grid forcing. Convolutional networks are trained using eddy fluxes with the rotational part filtered via an eddy force function, versus a divergence-based alternative. Results show comparable or improved accuracy alongside reduced sensitivity to small-scale features, supporting the value of curated training data for uncovering hidden physical processes.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

AccessFree to download

Key Points

Training-data choice and convolutional networks

Rotational-component filtering via eddy force function

Sensitivity to small-scale features

Abstract

Problem context: eddy–mean interaction

Competing training-data formulations

Main findings and implication for discovery

RESEARCH ARTICLE 10.1029/2023MS003915 Key Points: • Investigated the dependence of convolution neural networks on the choice of training data for geostrophic turbulence • Models are trained on eddy fluxes with rotational component filtered out by means of an eddy force function • Resulting models as accurate but less sensitive to small‐scale features than models trained on divergence of eddy fluxes Correspondence to: F. E. Yan and J. Mak, [feyan@connect.ust.hk](feyan@connect.ust.hk); [julian.c.l.mak@googlemail.com](julian.c.l.mak@googlemail.com) Citation: Yan, F. E., Mak, J., & Wang, Y. (2024) . On the choice of training data for machine learning of geostrophic mesoscale turbulence. Journal of Advances in Modeling Earth Systems, 16, e2023MS003915. [https://doi.org/10.1029/](https://doi.org/10.1029/)[ ](https://doi.org/10.1029/)2023MS003915 Received 2 JULY 2023 Accepted 20 DEC 2023 Author Contributions: Conceptualization: J. Mak Data curation: F. E. Yan, J. Mak Formal analysis: F. E. Yan, J. Mak Funding acquisition: J. Mak Investigation: F. E. Yan, J. Mak, Y. Wang Methodology: J. Mak, Y. Wang Project administration: J. Mak Resources: J. Mak Software: F. E. Yan, J. Mak Supervision: J. Mak Validation: F. E. Yan, J. Mak Visualization: F. E. Yan Writing – original draft: F. E. Yan, J. Mak, Y. Wang Writing – review & editing: F. E. Yan, J. Mak, Y. Wang © 2024 The Authors. Journal of Advancesin Modeling Earth Systems published by Wiley Periodicals LLC on behalf of American Geophysical Union. This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. On the Choice of Training Data for Machine Learning of Geostrophic Mesoscale Turbulence F. E. Yan1 , J. Mak1,2,3 , and Y. Wang1,2 1Department of Ocean Science, Hong Kong University of Science and Technology, Hong Kong, Hong Kong, 2Center for Ocean Research in Hong Kong and Macau, Hong Kong University of Science and Technology, Hong Kong, Hong Kong, 3National Oceanography Centre, Southampton, UK Abstract Data plays a central role in data‐driven methods, but is not often the subject of focus in investigations of machine learning algorithms as applied to Earth System Modeling related problems. Here we consider the problem of eddy‐mean interaction in rotating stratified turbulence in the presence of lateral boundaries, where it is known that rotational components of the eddy flux plays no direct role in the sub‐grid forcing onto the mean state variables, and its presence is expected to affect the performance of the trained machine learning models. While an often utilized choice in the literature is to train a model from the divergence of the eddy fluxes, here we provide theoretical arguments and numerical evidence that learning from the eddy fluxes with the rotational component appropriately filtered out, achieved in this work by means of an object called the eddy force function, results in models with comparable or better skill, but substantially reduced sensitivity to the presence of small‐scale features. We argue that while the choice of data choice and/or quality may not be critical if we simply want a model to have predictive skill, it is highly desirable and perhaps even necessary if we want to leverage data‐driven methods to aid in discovering unknown or hidden physical processes within the data itself. Plain Language Summary Data‐driven methods are increasingly being utilized in various problems relating to the numerical modeling of the Earth system. While there are many investigations focusing on the machine learning algorithms or the problems themselves, there have been relative few investigations into the impact of data choice or quality, given the central role of data. We consider here the impact of the choice of data for

FAQs

What is the main question of this study on machine learning for geostrophic mesoscale turbulence?

How the choice of training data affects convolutional neural network performance when modeling geostrophic mesoscale turbulence and eddy–mean interaction.

Why does filtering the rotational component of eddy flux matter?

The rotational component is known not to play a direct role in sub-grid forcing onto mean state variables, so including it in training can degrade performance.

How do the proposed models compare with models trained on divergence of eddy fluxes?

Models trained on eddy fluxes with rotational components appropriately filtered achieve comparable or better skill, but are substantially less sensitive to small-scale features.

Aug 4, 2026

Upload by Eliana