Classification of surface current maps

Abstract

Classification of ocean data maps is important for analysis of ocean data. Here, we compare self-organizing map (SOM) analysis with cluster methods such as the Ward method and K-means method. The HF (high-frequency) radar surface current data east of Okinawa Island, Japan were used for the comparison. There are two typical current patterns in the observation area: a strong southward current and a clockwise eddy-like current pattern. The classification results by the Ward method was similar to that by the SOM analysis. SOM analysis was insensitive to the cut-off empirical orthogonal function (EOF) mode number for reducing the data dimensions and noise, while the K-means method was the most sensitive to the EOF mode number.