THE SPATIAL DISTRIBUTION OF THE TROPICAL CYCLONE PRESSURE USING MACHINE LEARNING TECHNIQUE

Accurate modeling of storm surges induced by typhoons is crucial for effective disaster preparedness and mitigation. To simulate storm surges caused by typhoons, external forces include 10-meter winds and sea-level pressure fields are essential. Typically, such data is obtained using either parametric models, which utilizing simple mathematical parameters for modeling, or numerical weather prediction (NWP) models, interpreting physical processes numerically for predictions. Parametric models, despite being less accurate than NWP's full physics models, are often preferred in situations where rapid predictions are crucial due to their computational efficiency. However, parametric models, relying on statistical equations formulated from past observed data, may fall short in accurately describing the complex and nonlinear relationships between prediction variables related to typhoons using current simplistic statistical relationships. Additionally, parametric models, with their typical radial distribution, may inadequately represent the terrain impacts in regions distant from the center of a typhoon.Therefore, this study employed a Convolutional Neural Network (CNN) model, beneficial for spatial information in image pattern analysis, to derive pressure fields based on wind fields alone.