10a.7 Cluster Analysis of Western North Pacific Tropical Cyclone Tracks

Typhoons have a large socio-economic impact in many countries in Asia. Depending on the trajectory of the typhoon or tropical storm, landfall will occur or not. While it is well known that these trajectories vary strongly with season, and tend to be affected by ENSO, the probabilistic behavior of tropical cyclone trajectories needs to be better understood in order to isolate potentially predictable aspects of landfall. In this study, we apply a new clustering technique to the best track dataset of the Joint Typhoon Warning Center for the period 1950-2002. Only tropical cyclones (TCs) with at least tropical storm intensity were included, a total of 1393 TCs. The aim of this analysis is to identify different types of track, their seasonality, and their relationship to the large-scale circulation and ENSO. The clustering technique consists of building a mixture of polynomial regression models (i.e. curves), which are used to fit the geographical “shape” of the trajectories (Gaffney and Smyth, 1999). This technique is an extension of the standard multivariate finite mixture model to allow for the representations of mixture of underlying functions (in this case, quadratics), from which the observed TC tracks might have been generated. Finite mixture models enable highly non-Gaussian multimodal density functions to be expressed as a mixture of a few unimodal component PDFs. The model is fit to the data by maximizing the likelihood of the parameters, given the dataset. The mixture model framework allows the clustering problem to be posed in a rigorous probabilistic context, and to easily accommodate tropical cyclone tracks of different lengths, giving advantages over the K-means method used in previous studies (e.g. Harr and Elsberry (1985); Elsner and Liu (2003)).