Performance Analysis of AI-based Global Models in Tropical Cyclone Forecasting

The integration of Artificial Intelligence (AI) into global circulation models has ushered in a new era of weather forecasting, particularly for high-impact events such as tropical cyclones. This study focuses on the performance evaluation of five state-of-the-art AI-driven global weather prediction models - FourCastNet v2, Pangu-Weather, GraphCast, FuXi, and FengWu - in forecasting tropical cyclones that have significantly affected the Korean Peninsula. The analysis centers on the models' ability to predict the tracks of notable recent typhoons, including Typhoon Maysak, Haishen, Hinnamnor, and Khanun, which have posed considerable threats to the region.

To ensure a consistent and rigorous evaluation framework, all models were initialized with identical conditions derived from initial analysis data. This approach allows for a direct comparison of model outputs, focusing on the accuracy of track predictions, which are critical for disaster preparedness and mitigation in affected regions.

Preliminary findings suggest that AI-driven models tend to outperform traditional numerical models in predicting the tracks of tropical cyclones. In particular, these models demonstrate superior predictive skill in cases where the typhoon tracks are recurving. While these models show promise, the study also acknowledges that further detailed analysis is required to fully understand their performance across various weather phenomena and lead times.

AI-based models, with their potential for higher accuracy in track predictions, could play a pivotal role in improving early warning systems and reducing the socio-economic impacts of tropical cyclones. The study underscores the importance of continued research and development in AI-driven weather prediction, advocating for a collaborative approach between AI researchers and meteorologists to enhance the reliability and accuracy of forecasts, particularly for extreme weather events like tropical cyclones.

Acknowledgments:
This research was supported by the Korea Institute of Marine Science & Technology Promotion (KIMST) funded by the Korea Coast Guard (RS-2023-00238652, Integrated Satellite-based Applications Development for Korea Coast Guard) and the Ministry of Oceans and Fisheries (RS-2022-KS221667), as well as by the National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science and ICT) (No. RS-2022-00144325).