Classifying North Atlantic Tropical Cyclone Tracks by Mass Moments

Abstract

A new method for classifying tropical cyclones or similar features is introduced. The cyclone track is considered as an open spatial curve, with the wind speed or power information along the curve considered to be a mass attribute. The first and second moments of the resulting object are computed and then used to classify the historical tracks using standard clustering algorithms. Mass moments allow the whole track shape, length, and location to be incorporated into the clustering methodology. Tropical cyclones in the North Atlantic basin are clustered with K-means by mass moments, producing an optimum of six clusters with differing genesis locations, track shapes, intensities, life spans, landfalls, seasonal patterns, and trends. Even variables that are not directly clustered show distinct separation between clusters. A trend analysis confirms recent conclusions of increasing tropical cyclones in the basin over the past two decades. However, the trends vary across clusters.