Analyzing the multidimensional wave climate with self organizing maps

The term ldquowave climaterdquo usually refers to the statistical distribution of several oceanographical geophysical variables. Components of the wave climate are variables such as wind velocity, wind direction, significant wave height - SWH, peak period, Tp, mean wave direction, swell SWH, sea SWH, etc. Usually, the classical analysis of the long-term distribution of wave climate is addressed using just one variable (f.i., long-term distribution of significant wave height) or at most bidimensionally (f.i., the bivariate distribution of SWH and Tp). It is clear that the joint probability distribution of the aforementioned variables is not easy to cope with. However, this problem is solved applying a non-linear clustering algorithm, namely the Self Organizing Maps (SOM), a neural network technique capable of classifying the high dimensional input data bases in a low number of centroids (clusters) in an ordered sheet shape representation, allowing an intuitive visualization of the results. The neurons are connected to adjacent elements by a neighbourhood relation. A multidimensional histogram of the sea state parameters is obtained allowing an easy further treatment of the classified sea states.