Unsupervised learning and similarity comparison of water mass characteristics with Gaussian mixture model for visualizing ocean data.

The temperature-salinity relationship is one of the most important characteristics for identifying water masses in marine research. Temperature-salinity characteristic changes dynamically with the geographic location and is sensitive to the depth at the same location. When depth is considered, however, it is not easy to compare the characteristics of the water masses efficiently for a wide range of areas of the ocean. In this paper, the Gaussian mixture model was proposed to analyze the temperature-salinity-depth characteristics of water masses, based on which the comparison between water masses may be conducted. First, the temperature-salinity-depth data for various locations are used to train a set of Gaussian mixture models individually. The distance between two Gaussian mixture models is then defined as the weighting sum of the pairwise Bhattacharyya distances among the Gaussian distributions. Accordingly, the distance between two water masses is fast measured, which makes possible the automatic and efficient comparison of the water masses for a large area. This approach not only approximates the temperature-salinity-depth distribution directly without prior knowledge of their relationship but controls the complexity by adjusting the number of mixtures when the samples are not evenly distributed. In addition, this approach facilitates the knowledge discovery in marine research, since it helps to represent, manage and share the characteristics of water masses flexibly. The proposed approach has been successfully applied to a real-time visualization system of ocean data, which allows the user to explore and compare water masses responsively through aggregating the data without degrading the resolution. The system provides an interface for querying the geographic locations with similar temperature-salinity-depth characteristics to track specific patterns of water masses, such as those in the Kuroshio near Taiwan or the South China Sea. [ABSTRACT FROM AUTHOR]