Comparison of Kohonen's Self-Organizing Map algorithm and principal component analysis in the exploratory data analysis of a groundwater quality dataset

Regional monitoring of groundwater chemistry yields large, multivariate data sets. Summarizing available data, extracting useful information and formulating hypotheses for further research are the key aspects in the exploratory data analysis of these data sets. Traditionally multivariate statistical techniques such as principal component analysis (PCA) are applied for this purpose. In PCA a linear dimensionality reduction of the original, high dimensional data set is carried out in order to identify orthogonal directions (principal components) of maximum variance in the dataset based on linear combinations of correlated variables [3]. In this study, PCA is compared to the Self-Organizing Map (SOM) algorithm. The SOM-algorithm is a neural network designed to carry out a non-parametric regression process in order to represent high-dimensional, nonlinearly related data items in a topology-preserving, often two-dimensional display, and to perform unsupervised classification and clustering [11]. PCA and SOM are applied to a groundwater chemistry data set from a regional monitoring network in two sandy, phreatic aquifers in Central Belgium. The 47 monitoring wells are each equipped with three well screens at different depths, in which 14 variables are measured. Both techniques succeed in distinguishing between both aquifers and reveal the apparent relationships between variables. The main advantage of PCA is the expression of each variable in terms of the principal components and the quantification of the amount of variance explained by each component. The visualization of the SOM-analysis has the advantage of allowing a straightforward interpretation of the structure of the data set in which even non-linear relationships can be identified. Additionally, the SOM-algorithm can handle a limited amount of missing values in the data set, contrary to PCA.