EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Data-driven subgrid-scale turbulence parameterization: methods for improving model efficiency

Xingyu Zhu, Yongquan Qu, Naigeng Wu, Hualong Zhang, Xiaoming Shi

Machine Learning: Earth · 01.12.2026 · 2 (2) · 025002

DOI: 10.1088/3049-4753/ae6dbe ↗

Kurz gefasst

Fachartikel in der Zeitschrift Machine Learning: Earth, online erschienen am 16.06.2026. Verfasst von Xingyu Zhu und 4 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Learning the Earth with Artificial Intelligence and Physics (LEAP) Science and Technology Center, China Meteorological Administration, Research Grants Council, University Grants Committee.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Data-driven parameterization schemes have gained much attention in atmospheric science in recent years. We designed an idealized testing case with the barotropic vorticity equation model, which employed a subgrid-scale turbulence parameterization based on a deep learning (DL) model. Its effective forecast lead time is substantially improved with the trained DL model parameterization. However, a challenge is the significant amount of computational time required to run the DL model, posing a hurdle for practical applications. Thus, this research aims at improving the DL parameterization’s efficiency while minimizing any negative impact on its efficacy. Here we propose three methods, reducing the amount of model parameters, focusing on critical sub-domain information only as the input, or combining both. In addition, we introduce new designs to the loss function and improve the baseline model’s effective forecast lead time. Through these methods and utilizing the solver-in-the-loop training approach, we reduced the model’s computing time while preserving its prediction skills. These acceleration approaches can contribute to the development of more efficient data-driven physical parameterizations and inspire further explorations.

Förderung

  • Learning the Earth with Artificial Intelligence and Physics (LEAP) Science and Technology Center
  • China Meteorological Administration
  • Research Grants Council, University Grants Committee

Zitieren

APA

Zhu, X., Qu, Y., Wu, N., Zhang, H., & Shi, X. (2026). Data-driven subgrid-scale turbulence parameterization: methods for improving model efficiency. Machine Learning: Earth, 2(2), 025002. https://doi.org/10.1088/3049-4753/ae6dbe

Vancouver

Zhu X, Qu Y, Wu N, Zhang H, Shi X. Data-driven subgrid-scale turbulence parameterization: methods for improving model efficiency. Machine Learning: Earth. 2026;2(2):025002. doi:10.1088/3049-4753/ae6dbe

BibTeX
@article{Zhu2026,
  author = {Zhu, Xingyu and Qu, Yongquan and Wu, Naigeng and Zhang, Hualong and Shi, Xiaoming},
  title = {Data-driven subgrid-scale turbulence parameterization: methods for improving model efficiency},
  journal = {Machine Learning: Earth},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {025002},
  doi = {10.1088/3049-4753/ae6dbe}
}
Kostenlos starten →