Time Series Clustering using the Total Variation Distance with Applications in Oceanography

TL;DR

This paper introduces a novel time series clustering algorithm using total variation distance on normalized spectra, effectively identifying stationary periods in ocean wave data and outperforming some existing methods.

Contribution

The paper presents a new clustering approach based on spectral dissimilarity, specifically designed for analyzing stationary segments in oceanographic time series.

Findings

Performance comparable to leading methods

In some tests, outperforms existing techniques

Effective in identifying stationary sea wave periods

Abstract

An algorithm for determining stationary periods for time series of random sea waves is proposed in this work. This is a problem in which changes between stationary sea states are usually slow and segmentation procedures based on change-point detection frequently give poor results. The method is based on a new procedure for time series clustering, built on the use of the total variation distance between normalized spectra as a measure of dissimilarity. The oscillatory behavior of the series is thus considered the central characteristic for classification purposes. The proposed algorithm is compared to several other methods which are also based on features extracted from the original series and the results show that its performance is comparable to the best methods available and in some tests it performs better. This clustering method may be of independent interest.