Cluster Analysis with Application to Geochemical Data

The goal of this presentation is to give a deeper insight into the large data sets of the C-horizon and the O-horizon soil samples with the help of cluster analysis and fuzzy cluster analysis. These mineral soil samples were taken from a big area in the course of a regional study in the European Arctic. An overview of the Kola data project can be found on the web site http://www.ngu.no/Kola. The aim of cluster analysis is to find groups in data, in which objects of the same group should be as similar to each other as possible, whereas objects of different groups should be as dissimilar as possible. First we want to explain the steps necessary for doing cluster analysis. In the literature many methods of clustering algorithm are described. There are some characteristics which should be the basis for the choice of the clustering algorithm. The most widely used clustering techniques will be explained in more detail. The procedure of evaluating the results of a clustering algorithm is known under the term cluster validity. We distinguish between external, internal and relative criteria. With validity measures one can find the optimal number of clusters. Moreover, they are helpful for making statements about the structure of the existing data and for the selection of an appropriate clustering algorithm. A good graphical presentation of cluster results is important for the interpretation. Whereas objects can easily be shown in maps using different colors or grey-scale, the presentation of the variables that are influential for each cluster is not trivial. We propose a new plot based on Reimann et al. (2000), which shows the degree of average concentrations of the elements in the clusters. Finally, we will present some cluster results. The clusters of the C-horizon reflect the various groups of rock types in this area. For the O-horizon we can localize environmental pollution from heavy industry and recognize the influence of weather and sea.