Hybrid Model Enhances Typhoon Prediction Accuracy

Researchers have developed a new hybrid model to improve the accuracy of typhoon predictions by integrating data-driven machine learning (ML) models with physics-based models. The study, authored by Zeyi Niu, Wei Huang, Lei Zhang, Lin Deng, Haibo Wang, Yuhua Yang, Dongliang Wang, and Hong Li, focuses on addressing the limitations of current ML models, such as underestimating typhoon intensity and lacking interpretability.

The hybrid model, named Pangu_SP, uses forecast fields from the Pangu-Weather model to constrain large-scale forecasts of the Weather Research and Forecasting model through a method called spectral nudging. The results indicate that Pangu_SP significantly outperforms traditional models like the Global Forecast System (GFS_INIT) and the Integrated Forecasting System of the European Centre for Medium-Range Weather Forecasts (ECMWF IFS) in predicting the track and intensity of Typhoon Doksuri (2023).

Satellite observations show that the predicted typhoon cloud patterns from Pangu_SP are more consistent compared to other models. Additionally, the intensity forecasts from Pangu_SP are notably more accurate than those from the ECMWF IFS. This demonstrates the hybrid model's ability to leverage the strengths of both ML and physical models effectively.

The study also explores the importance of data assimilation in ML-driven hybrid dynamical systems. By assimilating water vapor channels from the Advanced Geostationary Radiation Imager onboard Fengyun-4B, the researchers were able to reduce errors in typhoon intensity forecasts.

These advancements in typhoon prediction models could lead to better preparedness and response strategies for communities affected by such natural disasters, potentially saving lives and reducing economic losses.

For more details, the full paper can be accessed at arXiv:2408.12630.