Niu et al. (2025) Machine‐Learning (ML)‐Physics Fusion Model Accelerates the Paradigm Shift in Typhoon Forecasting With a CNOP‐Based Assimilation Framework

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

Identification

• Journal: Geophysical Research Letters
• Year: 2025
• Date: 2025-08-04
• Authors: Zeyi Niu, Dongliang Wang, Mu Mu, Wei Huang, Xuliang Fan, Bo Qin
• DOI: 10.1029/2025gl115926

Research Groups

Not specified in the provided text.

Short Summary

The study develops a hybrid forecasting system integrating the FuXi machine-learning model with the physics-based Shanghai Typhoon Model (SHTM) to improve short-term predictions of typhoon track, intensity, and precipitation.

Objective

• To enhance the accuracy of short-term (0–120 h) typhoon forecasting by combining the large-scale capabilities of machine learning with the mesoscale strengths of physical models.
• To identify sensitive regions for data assimilation using the Conditional Nonlinear Optimal Perturbation (CNOP) method to further refine extreme typhoon forecasts.

Study Configuration

• Spatial Scale: Mesoscale to large-scale (regional typhoon domains).
• Temporal Scale: Short-term forecasting horizon of 0–120 h.

Methodology and Data

• Models used: FuXi (machine-learning model), Shanghai Typhoon Model (SHTM, physics-based model).
• Techniques: Spectral nudging (for model integration), Conditional Nonlinear Optimal Perturbation (CNOP) (for sensitivity analysis).
• Data sources: Satellite observations.

Main Results

• The hybrid FuXi-SHTM model significantly improved the prediction of track, intensity, and precipitation for Super Typhoons Yagi (2024) and Krathon (2024).
• The application of the CNOP method successfully identified sensitive regions where dense assimilation of satellite observations can further enhance forecast accuracy, despite the constraints of the large-scale ML forecast fields.

Contributions

• Proposes a novel hybrid framework that synergizes data-driven ML strategies with established physical modeling for typhoon prediction.
• First implementation of the Conditional Nonlinear Optimal Perturbation (CNOP) method to identify sensitive regions specifically for a hybrid ML-physics typhoon forecasting system.

Funding

Not specified in the provided text.

Citation

@article{Niu2025MachineLearning,
  author = {Niu, Zeyi and Wang, Dongliang and Mu, Mu and Huang, Wei and Fan, Xuliang and Yang, Mengqi and Qin, Bo},
  title = {Machine‐Learning (ML)‐Physics Fusion Model Accelerates the Paradigm Shift in Typhoon Forecasting With a CNOP‐Based Assimilation Framework},
  journal = {Geophysical Research Letters},
  year = {2025},
  doi = {10.1029/2025gl115926},
  url = {https://doi.org/10.1029/2025gl115926}
}