Han et al. (2025) A Decadal Hybrid GCM Simulation Using Deep‐Learning‐Based Cloud and Convection Parameterization Generalized to a Warm Climate

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

Identification

• Journal: Journal of Advances in Modeling Earth Systems
• Year: 2025
• Date: 2025-12-01
• Authors: Yilun Han, Guang J. Zhang, Yong Wang, Hui Wan
• DOI: 10.1029/2025ms005231

Research Groups

Not specified in the provided abstract.

Short Summary

This study demonstrates that a global climate model (GCM) with neural-network-based cloud and convection parameterization, trained solely on present-day climate data, can successfully perform a stable, decade-long simulation of a warm climate with a +4 K sea surface temperature anomaly, matching the performance of conventional and superparameterized models.

Objective

• To achieve stable, accurate simulations using machine-learning (ML) parameterization in global climate models (GCMs) under climates not seen during training, specifically a warm climate scenario.

Study Configuration

• Spatial Scale: Global
• Temporal Scale: Decadal (decade-long simulation)

Methodology and Data

• Models used:
  • Global Climate Model (GCM) with neural-network (NN) based cloud and convection parameterization (based on Han et al., 2023, with additional inputs).
  • Superparameterized CAM (SPCAM) for comparison.
  • Conventional CAM5 for comparison.
• Data sources:
  • Present-day climate data (used exclusively for training the neural network).
  • Warm climate with +4 K sea surface temperature (SST) for simulation and evaluation.

Main Results

• The GCM with NN parameterization successfully performed a stable, decade-long simulation of a warm climate with a +4 K SST anomaly, despite being trained only on present-day data.
• The simulation accurately captured global precipitation distribution, surface temperatures, vertical atmospheric structures, and extreme precipitation, closely matching results from SPCAM and CAM5 in the warm climate.
• The NN-GCM produced a climate response to +4 K SST in atmospheric thermodynamic states and circulations similar to those from SPCAM and CAM5.
• Prognostic ablation tests revealed that:
  • The NN without convective memory as an input suffered from numerical instability.
  • The NN without considering radiative variables and land fraction as inputs, or with reduced training samples, produced less accurate results.

Contributions

• This is the first reported instance of an ML parameterization successfully achieving online extrapolation to a warm climate without requiring additional warm-climate data for training.
• It demonstrates the significant potential of ML-driven parameterizations for credible and stable long-term climate projections under future climate scenarios.

Funding

Not specified in the provided abstract.

Citation

@article{Han2025Decadal,
  author = {Han, Yilun and Zhang, Guang J. and Wang, Yong and Wan, Hui},
  title = {A Decadal Hybrid GCM Simulation Using Deep‐Learning‐Based Cloud and Convection Parameterization Generalized to a Warm Climate},
  journal = {Journal of Advances in Modeling Earth Systems},
  year = {2025},
  doi = {10.1029/2025ms005231},
  url = {https://doi.org/10.1029/2025ms005231}
}