Chemometrics to evaluate the quality of water sources for human consumption

The present study reveals the importance of chemometric data treatment in the interpretation of environmental monitoring data sets by supervised and unsupervised techniques such as two-way and three-way principal components analysis on the one hand and tree partitioning on the other. Environmental monitoring was performed in the region of Athens, Greece, with 17 sampling sites and 16 chemical and physicochemical water quality parameters monitored in bimonthly periods (total of 102 objects x 16 variables). It was found that the location of rural and urban sites is responsible for the spatial separation of the water sources since the data structure depends mainly on two latent factors, conditionally named "salinity" and "turbidity". Additionally, the three-way principal components analysis ensures separation into "low water" and "high water" latent factors, which provides information on seasonal data decomposition. A set of discriminant variables including turbidity, total hardness, conductivity and free acidity could be found by the CART (classification and regression trees) partitioning approach, which is able to explain the differences between water quality in natural potable water sources and purified potable water.