Xia et al. (2026) A Neural‐Network‐Based Scheme for Improving Wegener–Bergeron–Findeisen Process and Its Impact on Mixed Phase Cloud

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

Identification

• Journal: International Journal of Climatology
• Year: 2026
• Date: 2026-01-02
• Authors: Yang Xia, Siyu Yue, Zengyuan Guo
• DOI: 10.1002/joc.70254

Research Groups

Not explicitly mentioned in the abstract.

Short Summary

This study develops a neural network to parameterize the degree of heterogeneous distribution between liquid and ice clouds, which is then used to constrain the Wegener–Bergeron–Findeisen (WBF) process in General Circulation Models (GCMs), leading to improved simulations of mixed-phase clouds, cloud radiative forcing, and precipitation globally.

Objective

• To develop and apply a machine learning-based parameterization for the degree of heterogeneous distribution (DHD) between liquid and ice clouds to improve the simulation of mixed-phase clouds in General Circulation Models (GCMs) by constraining the unrealistically strong Wegener–Bergeron–Findeisen (WBF) process.

Study Configuration

• Spatial Scale: Global (GCM simulations), with specific improvements noted in mid-high latitudes.
• Temporal Scale: 2 years (for neural network training and evaluation using SPCAM6 simulations); two decadal experiments (for evaluation of the constrained scheme in CIESM).

Methodology and Data

• Models used: SPCAM6 (Superparameterized General Circulation Model) for training and evaluation of the neural network; CIESM (Community Integrated Earth System Model) for applying and evaluating the constrained microphysical scheme.
• Data sources: 2-year simulations from SPCAM6 were used to train and independently evaluate the neural network.

Main Results

• The neural network (NN) accurately captures the variability of the degree of heterogeneous distribution (DHD) between liquid and ice clouds, achieving a high correlation coefficient exceeding 0.9.
• The NN-based scheme, when applied to constrain the WBF process in CIESM, significantly improves the distribution of simulated mixed-phase clouds, particularly in mid-high latitudes.
• The constrained WBF process enhances the amount of liquid cloud and reduces ice cloud on mixed-phase levels (temperatures between -40 °C and 0 °C).
• Enhanced freezing processes lead to an increase in ice clouds on upper levels (temperatures below -40 °C), resulting in a reduction of the latent heating rate and an increase in atmospheric stability.
• Compared to constant-value schemes, the NN-based scheme provides stronger improvements for ice clouds in mid-high latitudes and avoids the overestimation near -20 °C at high latitudes.
• The scheme leads to a more reasonable modulation of liquid and ice clouds, resulting in more realistic global cloud radiative forcing and precipitation simulations.

Contributions

• Introduces a novel machine learning (neural network) approach to parameterize the degree of heterogeneous distribution between liquid and ice clouds, addressing a critical limitation in GCM microphysical schemes.
• Demonstrates significant improvements in the simulation of mixed-phase clouds, cloud radiative forcing, and precipitation by applying the NN-based scheme to constrain the Wegener–Bergeron–Findeisen process.
• Provides a more physically realistic representation of liquid and ice clouds compared to traditional constant-value schemes, particularly in mid-high latitudes and at specific temperature ranges.
• Offers a robust solution to a long-standing bias in GCMs regarding the underestimation of liquid cloud and overestimation of ice cloud on mixed-phase levels.

Funding

Not explicitly mentioned in the abstract.

Citation

@article{Xia2026NeuralNetworkBased,
  author = {Xia, Yang and Yue, Siyu and Guo, Zengyuan},
  title = {A Neural‐Network‐Based Scheme for Improving Wegener–Bergeron–Findeisen Process and Its Impact on Mixed Phase Cloud},
  journal = {International Journal of Climatology},
  year = {2026},
  doi = {10.1002/joc.70254},
  url = {https://doi.org/10.1002/joc.70254}
}