Analysis of clustering and selection algorithms for the study of multivariate wave climate

Abstract

Recent wave reanalysis databases require the application of techniques capable of managing huge amounts of information. In this paper, several clustering and selection algorithms: K-Means (KMA), self-organizing maps (SOM) and Maximum Dissimilarity (MDA) have been applied to analyze trivariate hourly time series of met-ocean parameters (significant wave height, mean period, and mean wave direction). A methodology has been developed to apply the aforementioned techniques to wave climate analysis, which implies data pre-processing and slight modifications in the algorithms. Results show that: a) the SOM classifies the wave climate in the relevant "wave types" projected in a bidimensional lattice, providing an easy visualization and probabilistic multidimensional analysis; b) the KMA technique correctly represents the average wave climate and can be used in several coastal applications such as longshore drift or harbor agitation; c) the MDA algorithm allows selecting a representative subset of the wave climate diversity quite suitable to be implemented in a nearshore propagation methodology.