Traditional spatial decompositions such as empirical orthogonal functions (EOFs) and principal components analysis (PCA) of meteorological field data have received wide attention in numerous forecasting and analysis applications. Typically, the EOFs are applied in the tropical cyclone (TC) extratropical transition (ET) problem by comparing the spatial distribution of variables (wind fields, imagery data, etc.) from many similar cases spatially referenced to the ET point. The resulting EOFs are then used to analyze the same fields for times preceding and following ET to understand the role of spatial patterns in predicting ET. There are a number of potential drawbacks to such an analysis, the most prominent being the highly nonlinear dynamics of TC field data that lead to bifurcation of solutions, i.e. ET in some cases and not in others that are nominally similar in their spatial patterns.

In this investigation, we explore the utility of using three-dimensional pattern recognition techniques that exploit both spatial and temporal data. Incorporating the temporal dimension allows us to track the temporal evolution of the EOFs for ET and non-ET cases in order to investigate the differences between them. In addition, we apply techniques such as independent components analysis (ICA) and markov field analysis, which allow us to derive decomposition functions that are not constrained to be orthogonal, and instead try to examine the underlying statistical distributions of the meteorological fields.

We are applying our analysis techniques to ET in storms in the western North Pacific between 1997 and 2003. We will compare the utility of using various fields to include height fields, winds, and remotely sensed optical and microwave imagery.