Song et al. (2025) Physically Interpretable Emulation of a Moist Convecting Atmosphere With a Recurrent Neural Network

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Geophysical Research Letters
• Year: 2025
• Date: 2025-09-09
• Authors: Qiyu Song, Zhiming Kuang
• DOI: 10.1029/2025gl114794

Research Groups

Information not available from the abstract.

Short Summary

This study develops a recurrent neural network (RNN) for data-driven convective parameterization, combining linear and nonlinear components to predict temperature, moisture, and precipitation time series. The model demonstrates stable and realistic long-term emulation performance, revealing physically interpretable properties of convectively coupled gravity waves.

Objective

• To develop a recurrent neural network to predict time series of temperature, moisture, and precipitation of a cumulus ensemble in response to large-scale forcings.

Study Configuration

• Spatial Scale: Limited-domain, cloud-resolving scale (specific dimensions not provided).
• Temporal Scale: Time series, long-term emulations (specific duration not provided).

Methodology and Data

• Models used: Recurrent Neural Network (RNN) with a linear component (pre-identified time-invariant state-space model) and a nonlinear component (multilayer neural network). The system being emulated is a cumulus ensemble, and the data is generated from a cloud-resolving model.
• Data sources: Ensembles of limited-domain cloud-resolving model simulation data.

Main Results

• The developed recurrent neural network exhibits stable and realistic performance in long-term emulations.
• This stable performance is observed both with prescribed large-scale forcings and when the model is coupled with two-dimensional gravity waves.
• Calculation of linear responses to perturbations for the coupled emulation reveals physically interpretable, state-dependent properties of the convectively coupled gravity wave system.

Contributions

• Development of a novel recurrent neural network architecture for data-driven convective parameterization, integrating a linear state-space model with a nonlinear multilayer neural network.
• Demonstration of stable and realistic long-term emulation capabilities for convective adjustments, addressing challenges in generalizability.
• The ability to extract physically interpretable, state-dependent properties (e.g., linear responses to perturbations) from the coupled emulation, enhancing model interpretability.

Funding

Information not available from the abstract.

Citation

@article{Song2025Physically,
  author = {Song, Qiyu and Kuang, Zhiming},
  title = {Physically Interpretable Emulation of a Moist Convecting Atmosphere With a Recurrent Neural Network},
  journal = {Geophysical Research Letters},
  year = {2025},
  doi = {10.1029/2025gl114794},
  url = {https://doi.org/10.1029/2025gl114794}
}