제목: Physics-augmented Machine Learning Model to Estimate Tropical Cyclone Intensity​​​​​​​​​​

​​

​​​​​​​​​​​​​​​​​​연사: 김대휘 교수 (공주대학교)​​​​​

​

장소: 과학관 B102호​​

Abstract:

​Tropical cyclones (TCs) are low-pressure systems accompanied by heavy rainfall and severe winds that cause socioeconomic damage. Accurate real-time estimation of TC intensity is essential for effective decision-making. This study introduces a machine learning (ML) model for TC intensity estimation based on geostationary satellite observations. The ML model, comprised of light gradient boosting machines and the multilayer perceptron, is developed for the North Atlantic basin using TCs from 2017–2023. Rather than using full satellite imagery as inputs to complex neural networks, physics-based TC features are extracted from the satellite imagery and used as inputs to the ML model. This approach achieves estimation errors of 6–7 knots for the 2020–2023 seasons. Analysis of the inputs suggests that TC features representing the eye development and the symmetry of the cloud structure are important for intensity estimation. This physics-augmented approach offers a promising direction for improving TC intensity estimation while maintaining interpretability and low computational cost, with potential for extension to intensity forecasting.​