Cluster analysis of environmental data which is not interval scaled but categorical Evaluation of aerial photographs of groynefields for the determination of representative sampling sites

Cluster analysis is discussed for data which are categorical without ordering. The classification of groynefields from a dataset created by analyzing aerial photographs from a section of the river Elbe in order to determine representative sampling sites is the underlying environmental example. Problems of finding the appropriate distance metric and clustering method are discussed with respect to categorical data in general. For the groynefield data two similarity coefficients are compared and a satisfying result of four clusters is found with average linkage and the coefficient of Hyvärinen. Contingency tables are used to verify part of the resulting hypotheses.