A projection pursuit application to predict extratropical transition of tropical cyclones

In this research, a multistage projection pursuit technique is presented that is used to predict the end results of the extratropical transition of the tropical cyclones. The technique extracts the spatial features relevant to extratropical transition that can be obtained from NOGAPS analyses and remotely sensed data. Various atmospheric variables at different levels of the atmosphere are used in the spatial and spatiotemporal techniques. The technique is tested in a forecasting environment on tropical cyclones from recent years, and results show prediction comparable to a global numerical weather prediction system with 75% prediction and 30% false alarm rate. The system is further developed to have utility for forecasting by examining ways to utilize forecast fields in various stages of the technique. This allows the prediction to occur ahead of time. Trials in this "forecast mode" indicate results that are slightly less accurate than those obtained using analysis fields during the development phase. However, the results are promising and further possible improvements are discussed.