Support vector machine techniques to predict tropical cyclone re-intensification following extratropical transition

Intensity changes following the multi-stage process of extratropical transition have proven to be especially difficult to forecast, due to extremely similar storm evolutions prior to and during the first stages of transformation from a warm-cored axisymmetric tropical storm to a cold-cored asymmetrical extratropical low pressure system. In this study, differences in surrounding synoptic environments between dissipating and re-intensifying extratropical transitioning tropical cyclones are used to develop a prediction system for extratropical transition intensity change. Using a set of all historical transitioning storms between 2000 and 2008 in the northwest Pacific, we first identified common differences between 850mb potential temperature fields surrounding extratropical transition intensifiers and extratropical transition dissipators, respectively. These features were then used as inputs into a support vector machine classification system in hopes of creating a robust prediction system. Once the system was trained on a random subset of the data (80%), performance was tested on the remaining test set (20%). Overall, it was found that the prediction system was able to correctly predict extratropical transition intensity outcome in >75% of the test cases at 72 hours prior to extratropical transition. In the presentation, we will discuss the feature selection and classification system used, as well as the performance results, in detail.