Abstract:
The tropical cyclone (TC) maximum sustained wind speed is an important indicator for assessing the impact of typhoon disasters. Due to limitations in observational technology and differences in data compilation methods, there is a noticeable overestimation in the maximum sustained wind speed data of TC records before the 1970s. This study employed an improved empirical formula method and the LightGBM machine learning model to correct the maximum sustained wind speeds. The results showed that the empirical formula method generally performs well in correcting wind speeds, but the correction error increases rapidly when wind speeds are less than 20 m/s or greater than 52 m/s. The wind speed correction results using the LightGBM model demonstrate higher accuracy and stability, especially in high/low wind speed scenarios. The combined method outperformed either the formula method or the LightGBM model alone, reducing root mean square errors by 24.2% and 7.1%, respectively. The combined method proposed in this study can significantly enhance the precision of early TC maximum sustained wind speed data, providing reliable support for disaster assessment, storm surge forecasting, and coastal disaster mitigation policies.