Diagnosing Tropical Cyclone Rapid Intensification using Support Vector Machine Classification

Forecasting rapid intensification (RI) of tropical cyclones is a challenge due to limited understanding of the meteorological processes necessary to predict RI. Recent research identified large-scale synoptic controls that were relevant based on similar synoptic structure for RI in the North Atlantic basin via composite analysis. The compositing revealed variables that were good distinguishers of RI and non-RI cases including low-level specific humidity and potential temperature, mid-level vorticity, and pressure vertical velocity from 1000–100 mb. The current project used these variables, from the NASA MERRA dataset for all tropical cyclones from 1979-2009, to discriminate between RI and non-RI tropical cyclones.

The predictors for the statistical models consisted of RPC scores retained from a second RPCA of the MERRA fields, since these represent the variability structure of both RI and non-RI cyclones. In order to diagnose the generalization of the models, a bootstrap cross-validation method was implemented. The resulting analyses revealed model skill that is statistically similar to current classification capabilities of the SHIPS-RII model. Results from this study suggest a possible new approach for improving forecasting of RI using artificial intelligence techniques.