Spatial and spatiotemporal projection pursuit techniques to predict the extratropical transition of tropical cyclones

A multistage projection pursuit (PP) approach is applied to the classification of tropical cyclones (TCs) during extratropical transition (ET) using 500-hPa Navy Operational Global Atmospheric Prediction System (NOGAPS) geopotential height analyses. PP algorithms reduce the dimensionality of the high-dimensional data while minimizing the loss of information that discriminates among classes of ET types. In this paper, a prediction system is developed for ET TC classification and is applied to 85 western North Pacific storms during 1997-2004 using NOGAPS geopotential height analyses to group them as either intensifiers or dissipators. A classification is developed based on the 1997-2002 analyses, and the forecasting performance of the technique is tested on the storms from 2003 and 2004 in two different ways. In the first, the technique is applied at individual times utilizing just spatial information. In the second, the change in spatial patterns with time is taken into account with spatiotemporal algorithms. The preliminary classification results are slightly less accurate than those of NOGAPS but are found to be promising, and further possible improvements are discussed.