Zhang et al. (2025) Physics-Informed Deep Learning for 3D Wind Field Retrieval of Open-Ocean Typhoons

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-11-26
• Authors: Xingyu Zhang, Tian Zhang, Shitang Ke, Houtian He, Ruihan Zhang, Yan Miao, Teng Liang
• DOI: 10.3390/rs17233825

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study proposes a physics-informed deep learning framework for high-resolution three-dimensional (3D) typhoon wind field reconstruction over the open ocean using multi-channel Himawari-8/9 satellite data, achieving improved accuracy and physical consistency by embedding the continuity equation.

Objective

• To develop a physics-informed deep learning framework for accurate, high-resolution, three-dimensional (3D) typhoon wind field reconstruction over the open ocean, addressing observational gaps and physical inconsistencies in existing methods.

Study Configuration

• Spatial Scale: Regional to storm-scale, focusing on open-ocean typhoons and their fine-scale structures (e.g., eyewalls, spiral rainbands).
• Temporal Scale: Event-based reconstruction of typhoon wind fields at specific observation times.

Methodology and Data

• Models used: Physics-informed deep learning framework, convolutional neural network (CNN), continuity equation (embedded as a strong constraint in the loss function).
• Data sources: Multi-channel data from Himawari-8/9 geostationary satellites, specifically 16-channel satellite imagery including visible and infrared channels.

Main Results

• The full model, integrating both visible/infrared channels and the physical constraint, achieved the best performance.
• Mean absolute errors were 2.73 m/s for U-wind components and 2.54 m/s for V-wind components.
• This represents significant improvements over the baseline infrared-only model, with error reductions of 29.6% for U-wind and 21.6% for V-wind components.
• Notable error reductions were observed in high-wind regions (wind speeds greater than 20 m/s).
• The approach effectively captures fine-scale structures like eyewalls and spiral rainbands while maintaining vertical physical coherence.

Contributions

• Proposes a novel physics-informed deep learning framework for high-resolution 3D typhoon wind field reconstruction.
• Demonstrates the significant benefit of integrating multi-channel satellite data (visible/infrared) and physical constraints (continuity equation) for improved accuracy and physical consistency in wind field retrieval.
• Offers a robust foundation for enhanced typhoon monitoring and reanalysis by addressing observational gaps and physical inconsistencies in existing methods.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Zhang2025PhysicsInformed,
  author = {Zhang, Xingyu and Zhang, Tian and Ke, Shitang and He, Houtian and Zhang, Ruihan and Miao, Yan and Liang, Teng},
  title = {Physics-Informed Deep Learning for 3D Wind Field Retrieval of Open-Ocean Typhoons},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17233825},
  url = {https://doi.org/10.3390/rs17233825}
}