Multivariate statistical analysis of hydrochemical and radiological data of Serbian spa waters

Abstract

Principal component analysis (PCA) and hierarchical cluster analysis (HCA) have been applied in order to recognize and classify spa waters collected at 30 sites on Serbian territory. The data set comprised 30 observations and 18 measured variables; natural radionuclides (K-40, (238)u, Ra-226, Ra-228), gross alpha (GA) and gross beta (GB) activities, and hydrochemical parameters (temperature, pH, electrical conductivity, total solids, HCO3-, Ca2+. Mg2+. Na+,K+, Cl-, SO42-, SiO2). A Box-Cox transformation was used as a data pretreatment before the statistical methods applied. Analyzed waters were qualified in 14 categories strongly predominated by hydrogen carbonate. The exploration of the correlation matrix allowed to uncover strong associations between some variables (the alkaline and alkaline-earth elements, the radium isotopes, GA, GB, and chloride) as well as a lack of association between the others (pH, T, Mg2+ and SiO2). PCA has revealed four latent factors which are responsible for the data structure covering 74.2% of the observed variations among the variables studied. Two of them can be initially assigned to mineralization of the components of the host rock whereas the other PCs are built from variables indicative of natural radioactivity. A reliable grouping of given data set of spa water samples with respect to their geotectonic units was found. The total correct classification of 83.3% was achieved for predefined geotectonic units. The resulting dendogram of HCA was interpreted to have classified the 30 spa water samples into four major groups and eleven subgroups using 18 variables. (C) 2011 Elsevier B.V. All rights reserved.