Predictability of Tropical Cyclone Rapid Intensification based on Statistical Approach

Abstract

This study investigated the spatial and temporal characteristics of rapid intensification (RI) in the Vietnam East Sea (VES) and evaluated the predictability of RI using statistical methods. For the purpose of the RI study, this work focused on a dataset of TCs that reach storm level higher, or having a maximum intensity of at least 34 knots (kn) during their existence. The results show that the annual TC activity in the VES is characterized by a dominance of strong TCs (Category 12 and above) and a significant occurrence of RI-TCs accounting for 73.7% and 23% of the total respectively. Remarkably, RI-TCs were consistently observed in 26 out of the 31 years studied, with a tendency to occur during the latter months of the year. Additionally, approximately 20% of these RI-TCs underwent RI near the Vietnam Coastal region. Given the increasing demand for accurate RI forecasts, four probability models namely Linear Discriminant Analysis (LDA), Logistic Regression (LogR), Naïve Bayes Classifier (Bayes) and Ensemble, using predictors from the SHIPS dataset, are developed to evaluate the predictability of the RI forecast. Among the predictors used, thermodynamic factors such as COHC, vertical wind shear (SHRD) and current TC states (PER) play crucial roles in constructing the RI probability models. Verification indices such as POD, FAR, CSI and BSS, indicate significant improvements in RI forecasting over the VES when utilizing the probability models, especially with the ensemble method.