Classification of the quality of surface waters by means of pattern recognition

Pattern recognition is shown to provide a means of classifying various types of surface waters on the basis of physicochemical data. Based on eleven relevant features, five different clusters were found in an evaluation of about 150 000 observations made over a period of 8 years. Discrimination between these clusters pertaining to water quality required seven chemical and two physical features only. Combination of a selection of these features into one eigenvector, dominated by total ionic nitrogen and ortho-phosphate, and a second eigenvector dominated by nitrate and ammonia with different signs. defines a projection plane in a five-dimensional feature space that accounts for 91% of the information content of the data matrix; 65% is attributed to total ionogenic nitrogen and phosphate and 26% is associated with the nitrogen redox balance.