Author(s): Young-Hoon Jin; Sung-Chun Park; Kyong-Bum Roh; Chang-Ryol Oh
Keywords: Water quality; Reservoir; Pattern classification; Self-organizing map
Abstract: Hydrological and environmental data measured from two stations (JSD1 and JSD2) in Jangseong reservoir, Korea are classified by application of Self-Organizing Map (SOM) in order to investigate the characteristics of multivariable data. Water quality parameters for environmental data include Dissolved Oxygen (DO), Chemical Oxygen Demand (COD), Total Nitrogen (TN) and Total Phosphorus (TP), while rainfall and storage amount are used to consider hydrological aspect of the reservoir. Recent monthly data for five years between 2006 and 2010 are classified by a methodologically systematic approach of SOM application including an appropriate data transformation, determination of total nodes and side lengths of SOM, initialization of weights, selection of a training mode and an optimized number of clusters, and a hierarchical cluster analysis for fine tuning after SOM training. The pattern classification results from SOM application are mainly visualized by radar charts for the respective clusters to efficiently show the characteristics of multivariate data. Detailed characteristics of each cluster are described using reference vectors which are the eventual purpose of SOM application. The classified patterns are interpreted for evaluating the hydrological and environmental conditions represented by the data used in the present study.