Zhang et al. (2026) Process‐Oriented Calibration of a Turbulence Scheme in the DOE's Global Storm‐Resolving Model Using Machine Learning

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

Identification

• Journal: Geophysical Research Letters
• Year: 2026
• Date: 2026-04-15
• Authors: Y Zhang, Tianning Su, Hewei Tang, Peter Bogenschutz, Stephen A. Klein, Charles Jackson, Peter Caldwell
• DOI: 10.1029/2025gl120241

Research Groups

• Department of Energy (DOE) / E3SM (Energy Exascale Earth System Model) project contributors

Short Summary

A machine learning-based calibration framework was developed for the SHOC turbulence scheme in the SCREAM model, significantly improving the representation of boundary-layer turbulence and shallow cumulus clouds.

Objective

• To efficiently calibrate the adjustable parameters of the Simplified Higher-Order Closure (SHOC) turbulence scheme across two distinct convective regimes: clear-sky dry convective boundary layers and fair-weather shallow cumulus clouds.

Study Configuration

• Spatial Scale: Local/Mesoscale (utilizing a doubly periodic version of the SCREAM model)
• Temporal Scale: Not specified (focused on convective boundary layer timescales)

Methodology and Data

• Models used: Simple Cloud Resolving E3SM Atmospheric Model (SCREAM), Simplified Higher-Order Closure (SHOC) turbulence scheme, and Large-Eddy Simulations (LES) for benchmarking.
• Data sources: ARM (Atmospheric Radiation Measurement) observations.
• Techniques: Machine learning surrogates, perturbed-parameter ensembles, and Markov Chain Monte Carlo (MCMC) sampling guided by cost functions.

Main Results

• Calibrated SHOC parameters led to substantial improvements in the modeling of boundary-layer turbulence and cloud boundaries.
• Modeled cloud fraction and radiative effects showed better alignment with observational data compared to the model's default settings.

Contributions

• Demonstrates that integrating process-specific convective regimes with machine-learning surrogates can effectively reduce parametric uncertainties and enhance the fidelity of cloud-turbulence interactions in atmospheric models.

Funding

• Not specified in the provided text.

Citation

@article{Zhang2026ProcessOriented,
  author = {Zhang, Y and Su, Tianning and Tang, Hewei and Bogenschutz, Peter and Klein, Stephen A. and Jackson, Charles and Caldwell, Peter},
  title = {Process‐Oriented Calibration of a Turbulence Scheme in the DOE's Global Storm‐Resolving Model Using Machine Learning},
  journal = {Geophysical Research Letters},
  year = {2026},
  doi = {10.1029/2025gl120241},
  url = {https://doi.org/10.1029/2025gl120241}
}