Multivariate exploration and classification applied to the chemical composition of spring waters in sanctuary forest areas

Abstract

This study was carried out in the Zimnik and Czyrna catchments (ZC and CC, respectively) located in the Silesian Beskid Mountains of southern Poland. The data matrix with 870 observations of major inorganic elements (Cl⁻, , , , Na⁺, K⁺, Ca²⁺ and Mg²⁺), pH and electrolytic conductivity of spring water samples was carried out using linear discriminant analysis, unsupervised Kohonen self-organising maps (SOM) classification and non-parametric tests. A multivariate analysis of the chemical composition of spring water was performed, taking into consideration the geological condition of the catchment area, the prevailing forest stands, slopes facing and different water levels due to the seasonal variations. For the ZC the first discriminant function (DF) reflected general geological conditions (absence of ) while the second DF reflected the nutrient biocycle (presence of ). In case of the CC, the first DF had a complex meaning and reflected both geological conditions and the nutrient biocycles. Based on specific chemical profiles, low water level related samples and high water level related samples were distinguished from each other in both the ZC and CC, respectively. The SOM-based classification indicated that forest types and springs location were the major factors affecting the spring water chemical profile. In general, in case of springs located above 1000 m, limited weathering connected with the effect of the highly acid spruce monoculture on leaching causes a decreasing concentration of Na⁺, K⁺, Mg²⁺, Cl⁻ and , while springs located between 700 and 800 m above sea level in mixed forests, showed the highest concentrations of K⁺ and Cl⁻ and the lowest concentrations of , values of pH and conductivity.

Keywords:

• mountain catchments
• Silesian Beskid
• spring water
• linear discriminant analysis
• self-organising maps

Acknowledgements

This research was supported financially in the framework of the project: ‘Optimisation of chemometrical techniques of exploration and modelling results originating from environmental constituent's pollution monitoring’ (1439/T02/2007/32). Moreover, the authors would like to kindly acknowledge the GIS and Remote Sensing Laboratory of Agricultural University of Cracow for maps delivering and to thank Prof Mike Kendall (Plymouth Marine Laboratory, UK) and Prof Bengt Nihlgård (Department of Plant Ecology and Systematics, Lund University, Sweden) for English correction and their invaluable suggestions and comments.