Improving Typhoon Predictions by Integrating Data-Driven Machine Learning Model With Physics Model Based on the Spectral Nudging and Data Assimilation

Abstract

The rapid advancement of data-driven machine learning (ML) models has improved typhoon track forecasts, but challenges remain, such as underestimating typhoon intensity and lacking interpretability. This study introduces an ML-driven hybrid typhoon model, where Pangu forecasts constrain the Weather Research and Forecasting (WRF) model using spectral nudging. The results indicate that track forecasts from the WRF simulation nudged by Pangu forecasts significantly outperform those from the WRF simulation using the NCEP GFS initial field and those from the ECMWF IFS for Typhoon Doksuri (2023). Besides, the typhoon intensity forecasts from Pangu-nudging are notably stronger than those from the ECMWF IFS, demonstrating that the hybrid model effectively leverages the strengths of both ML and physical models. Furthermore, this study is the first to explore the significance of data assimilation in ML-driven hybrid typhoon model. The findings reveal that after assimilating water vapor channels from the FY-4B AGRI, the errors in typhoon intensity forecasts are significantly reduced.