Peng et al. (2025) Typhoon Prediction Analysis of Pangu Weather Model

Identification

• Journal: Lecture notes in civil engineering
• Year: 2025
• Date: 2025-10-08
• Authors: Yuehua Peng, Libo Yao, Jun Song, Yuchen Zhang
• DOI: 10.1007/978-981-95-1487-8_20

Research Groups

• Dalian Naval Academy, Dalian, China
• Naval Aviation University, Yantai, China
• Dalian Ocean University, Dalian, China
• Tsinghua University, Beijing, China

Short Summary

This paper evaluates the Pangu Weather Model for typhoon prediction, demonstrating its suitability and high accuracy with prediction errors not exceeding 1%, and proposes its potential for artificial typhoon modification through engineering cybernetics.

Objective

• To test and analyze the suitability and accuracy of the Pangu Weather Model for typhoon prediction.
• To explore the potential application of the Pangu Weather Model for artificial modification of typhoons, integrating engineering cybernetics principles.

Study Configuration

• Spatial Scale: Global (Pangu model's inherent scale), with specific analysis focused on the Western Pacific Ocean for typhoon prediction. Vertical resolution includes 13 pressure levels (50 hPa, 100 hPa, 150 hPa, 200 hPa, 250 hPa, 300 hPa, 400 hPa, 500 hPa, 600 hPa, 700 hPa, 850 hPa, 925 hPa, and 1000 hPa).
• Temporal Scale: Prediction range of 1 hour to 7 days. Analysis based on 88 tropical cyclones that emerged in 2018. Specific error analysis example for a 6-hour forecast from 00:00 on September 30, 2018.

Methodology and Data

• Models used: Pangu Weather Model (a 3D Earth-specific transformer neural network with approximately 64 million parameters). The integration of a PID (proportional-integral-differential) controller from engineering cybernetics is proposed for artificial typhoon modification.
• Data sources:
  • Global meteorological data (implicitly used for Pangu model training).
  • Interpolated correction data from the European Medium-term Weather Prediction Center (used by the system).
  • Raw data from the European Meteorological Center (for error comparison).
  • Data for 88 tropical cyclones that emerged in 2018.
  • Typhoon intensity forecast error verification index from the GB/T38308–2019 weather prediction standard.

Main Results

• The Pangu Weather Model is suitable for typhoon prediction.
• The typhoon prediction error of the model is not higher than 1% for key physical quantities (pressure, velocity).
• Pressure error is significantly low, often much lower than 0.1% in most areas, and estimated to be lower than 0.3% around the typhoon eye.
• Spatially averaged velocity error is approximately 0.3%.
• The model demonstrates very high accuracy for typhoon path forecasts in the Western Pacific Ocean, showing a strong fit with observed paths (e.g., for typhoons Connie and Jade Rabbit).
• The Pangu model's global and 3D prediction capabilities, coupled with its high accuracy and rapid inference time (approximately 1.4 seconds, which is over 10,000 times faster than traditional numerical methods), make it a viable tool for artificial typhoon modification and integration with engineering cybernetics.

Contributions

• Provides the first detailed analysis and validation of the Pangu Weather Model's performance specifically for typhoon prediction, confirming its high accuracy and suitability.
• Quantifies the prediction errors for critical typhoon parameters (pressure, velocity) using established meteorological standards, demonstrating errors consistently below 1%.
• Proposes a novel application of the Pangu Weather Model for artificial typhoon modification by integrating engineering cybernetics (e.g., PID controller) and building a simulation platform, leveraging the model's high accuracy and computational efficiency.
• Highlights the Pangu model's significant advantages over traditional numerical methods, particularly its 3D global prediction capability, speed, and potential for closed-loop control applications in meteorology.

Funding

No specific funding projects or grants for this research were explicitly mentioned in the paper. The Pangu Weather Model itself was recognized as one of the "Top 10 Scientific Advances in China" in 2023 by the National Natural Science Foundation of China.

Citation

@article{Peng2025Typhoon,
  author = {Peng, Yuehua and Yao, Libo and Song, Jun and Zhang, Yuchen},
  title = {Typhoon Prediction Analysis of Pangu Weather Model},
  journal = {Lecture notes in civil engineering},
  year = {2025},
  doi = {10.1007/978-981-95-1487-8_20},
  url = {https://doi.org/10.1007/978-981-95-1487-8_20}
}