Multivariate wave climate using self-organizing maps

Abstract

ABSTRACT: The visual description of wave climate is usually limited to two-dimensional conditional histograms. In this work, self-organizing maps (SOMs), because of their visualization properties, are used to characterize multivariate wave climate. The SOMs are applied to time series of sea-state parameters at a particular location provided by ocean reanalysis databases. Trivariate (significant wave height, mean period, and mean direction), pentavariate (the previous wave parameters and wind velocity and direction), and hexavariate (three wave parameters of the sea and swell components; or the wave, wind, and storm surge) classifications are explored. This clustering technique is also applied to wave and wind data at several locations to analyze their spatial relationship. Several processes are established in order to improve the results, the most relevant being a preselection of data by means a maximum dissimilarity algorithm (MDA). Results show that the SOM identifies the relevant multivariate sea-state types at a particular location spanning the historical variability, and provides an outstanding analysis of the dependency between the different parameters by visual inspection. In the case of wave climate characterizations for several locations the SOM is able to extract the qualitative spatial sea-state patterns, allowing the analysis of the spatial variability and the relationship between different locations. Moreover, the distribution of sea states over the reanalysis period defines a probability density function on the lattice, providing a visual interpretation of the seasonality and interannuality of the multivariate wave climate.