Exploring the Typhoon Intensity Forecasting through Integrating AI Weather Forecasting with Regional Numerical Weather Model

Abstract Recent advancements in artificial intelligence (AI) have notably enhanced global weather forecasting, yet accurately predicting typhoon intensity remains challenging. This is largely due to constraints inherent in regression algorithm properties and inability of coarse resolution to capture the finer-scale weather processes. To address these insufficient in typhoon intensity forecasting, we propose an attractive approach by initiating regional Weather Research and Forecasting (WRF) model with Pangu-weather, a state-of-the-art AI weather forecasting system (AI-Driven WRF), whose forecasting power can be further augmented by the implementation of dynamic vortex initialization. The results revealed a significant gap in Pangu-weather's ability to predict typhoon intensity, while the AI-Driven WRF model, on the other hand, exhibited a remarkable improvement over Pangu-weather with dramatic reductions in the Root Mean Square Errors (RMSEs) for Typhoons Doksuri and Hato, demonstrating 87% and 63% improvements in forecast accuracy, respectively. The AI-Driven WRF model also showed potential advancements over traditional global numerical model-driven WRF models, performing more accurately typhoon intensity and wind details. Our analysis underscores the potential of AI weather forecasting models as a viable alternative for driving regional models, suggesting a promising avenue for future research in meteorology.