Feature Extraction and Classification of Temporal Characteristics of Earthquake Motions

Abstract

A 98-dimensional feature vector has been proposed to represent temporal characteristics of a waveform of ground motion acceleration on the basis of its Husid plot. Dissimilarity of a pair of waveforms can be quantified in terms of Euclidian distance between corresponding feature vectors. Then a method has been proposed to cluster and classify temporal characteristics of waveforms by use of hierarchical cluster analysis. Illustrative examples using acceleration records observed in the 2011 Off the Pacific Coast of Tohoku Earthquake, Japan are shown. The distribution map of the clustering and classification show that the temporal characteristics affected by the distance from the source region and multiple-source process can be appropriately classified. Multidimensional scaling is also applied to characterize the result of clustering and classification. It is shown that hierarchical cluster analysis is capable of reflecting temporal characteristics better than significant duration.