Pattern Classification of Typhoon Tracks Using the Fuzzy c-Means Clustering Method

Abstract

A fuzzy c-means clustering method (FCM) is applied to cluster tropical cyclone (TC) tracks. FCM is suitable for the data where cluster boundaries are ambiguous, such as a group of TC tracks. This study introduces the feasibility of a straightforward metric to incorporate the entire shapes of all tracks into the FCM, that is, the interpolation of all tracks into equal number of segments. Four validity measures (e.g., partition coefficient, partition index, separation index, and Dunn index) are used objectively to determine the optimum number of clusters. This results in seven clusters from 855 TCs over the western North Pacific (WNP) from June through October during 1965?2006. The seven clusters are characterized by 1) TCs striking the Korean Peninsula and Japan with north-oriented tracks, 2) TCs affecting Japan with long trajectories, 3) TCs hitting Taiwan and eastern China with west-oriented tracks, 4) TCs passing the east of Japan with early recurving tracks, 5) TCs traveling the easternmost region over the WNP, 6) TCs over the South China Sea, and 7) TCs moving straight across the Philippines. Each cluster shows distinctive characteristics in its lifetime, traveling distance, intensity, seasonal variation, landfall region, and distribution of TC-induced rainfall. The roles of large-scale environments (e.g., sea surface temperatures, low-level relative vorticity, and steering flows) on cluster-dependent genesis locations and tracks are also discussed.