Evaluation of a typhoon forecasting system based on AI weather model (AI-TRANS)

The application of artificial intelligence in scientific research (AI for Science) is transforming the traditional scientific research paradigm, and AI weather models have already been deployed in operational applications. Despite significant progress in typhoon track forecasting, AI weather models still exhibit limitations in capturing complex physical processes within typhoons and predicting changes in typhoon intensity, especially the rapid intensification (RI) of typhoons and the evolution of typhoon internal structures. This is mainly because AI weather models lack sufficient resolution support for determining typhoon intensity and fine structure. The use of AI weather model-driven regional model technology can overcome this bottleneck, enabling more accurate forecast of typhoon intensity and structural evolution. To this end, the Chinese Academy of Meteorological Sciences and the National Meteorological Center have jointly developed the AI-driven Typhoon Rapid Analysis and Forecasting System (AI-TRANS) based on the technical route mentioned above. The forecast results of AI-TRANS are compared with forecasts of six other operational models using typhoon forecast data from 2024. The results show that the performance of AI-TRANS is basically the same as that of the existing operational models in typhoon track forecasting. The prediction of typhoon intensity and structure shows obvious advantages. The system integrates the advantages of the AI weather model and numerical models, and significantly improves the accuracy of typhoon landfall forecast by improving the track forecast stability of the AI weather model. At the same time, the AI weather model-driven regional model technology has effectively improved the forecasting effect of key indicators such as typhoon RI, lifetime maximum intensity and fine structure. Taking the severe typhoons Gaemi and Yagi in 2024 as examples, the AI-TRANS system successfully predicted the track rotation, RI process and concentric eyewalls structure of Gaemi in the east side of Taiwan Island, and accurately captured the RI, concentric eyewalls formation and eye wall replacement process after Typhoon Yagi entered the South China Sea. This is the first time that a model can successfully predict the concentric eyewalls structure and the eyewall replacement process in real-time typhoon forecasting, which demonstrates the critical supporting role of the AI-TRANS system in improving the accuracy of typhoon intensity forecast and enhancing the capability of disaster mitigation.