Global classification of wave height, period, and direction multivariate distributions using principal component analysis

Abstract

Research on the classification of ocean surface wave patterns uses coarse spatial resolutions and approaches that are less usable for applications such as annual energy estimates for wave energy converters. Consequently, this study investigates the classification of 1–3-dimensional histograms of wave height, period, and direction at a 0.4° spatial resolution. Histograms of global grid-points have been decomposed into their main modes using principal component analysis. Each mode consists of a histogram and a map quantifying its contribution to each local wave pattern. The first mode displayed a meridional gradient of the wave height, a northeast-to-southeast gradient of the wave period, and a mix of both for the wave direction. Dominating wave systems consist of a) 1.9 m, b) 14.5 s in the basins’ southeast, c) 12.5 s north and south, and d) 9.5 s north-west. Generated remotely, c) travel and evolve over long distances to reach coasts, as b) pools, while d) is mainly generated locally or northwards. K-means was used to cluster wave patterns into 15 classes using the 10 dominant modes. Some clusters characterise regions in different basins. 1-parameter-based clusters provide more regions and alternations of clusters nearshore, while clustering the parameters together provides larger regions and less nearshore noise. The wave height and period combined classification showed good correlation and difference between the cluster's average histogram and those of the grid-points in that cluster, while the direction was lower in regions affected by strong currents or topographic obstacles. Consequently, combining all parameters greatly decreases these two metrics.