Beyond a single rapid intensification threshold: a continuous intensification rate index for tropical cyclones using vortex-scale machine learning

Abstract

Abstract

The rapid intensification (RI) of tropical cyclones (TCs) remains most persistent challenges in operational forecasting, particularly in western North Pacific (WNP), where RI events are most frequent and intense globally. Although numerical weather prediction continues to advance, adaptive tools are needed to resolve the multi-scale processes driving sudden intensity changes. Existing studies often rely on static binary thresholds for RI occurrence. However, the physical mechanisms favoring RI emerge at varying stages during the intensification process across different cases, governed by concurrent environmental conditions and internal vortex dynamics rather than any fixed RI definition. To address these limitations, this study establishes a diagnostic system for TC intensification in the WNP by integrating vortex-scale reanalysis data for capturing structural drivers of RI, a new continuous intensification rate (IR) index that moves beyond traditional binary RI classifications, and machine learning techniques. Following systematic hyperparameter optimization, the Random Forest, Support Vector Regression, and Artificial Neural Network models all demonstrated consistent and reliable performance in linking higher IR index values to increased probabilities of TC intensification. By integrating physically meaningful vortex-scale features with a continuous IR metric, this framework offers a versatile approach to advance the understanding and forecasting of TC intensification in the WNP.