A Study on The Frequency Characteristics of Typhoon Landing in Guangdong, China, Based on Machine Learning Methods

Abstract

Located in the southern coastal region of China, Guangdong Province is perennially threatened by typhoons originating from the Northwest Pacific Ocean. Compared to studies focusing on typhoon intensity, paths, and catastrophic effects, research on typhoon landing frequency features remains relatively limited. This study systematically constructs a dataset based on the landing typhoon frequency of Guangdong over 71 years (as the target variable) and a set of 88 atmospheric circulation indices as well as 26 sea surface temperature indices (as the potential feature variables). Two machine learning models, including Random Forest Regression (RR) and Support Vector Regression (SVR), are used to predict the frequency of typhoons making landfall in Guangdong, and the results show that both models have good characterisation ability while the RR model has slightly better fitting performance on the training set. This study provides scientific support for understanding the characteristics of typhoons in Guangdong and better responding to typhoon disasters.