Cluster analysis for groundwater classification in multi-aquifer systems based on a novel correlation index.

Understanding the spatial variations in groundwater chemistry is fundamental to assess the groundwater pathways and identify the most advantageous strategies for a sustainable use of groundwater resources. In fact, such variations can be the result of the complex structure of flow systems or local pressures of natural or anthropic origin. Thus, a detailed analysis of spatial variations in hydrochemical data provides an insight into natural and anthropogenic effects on groundwater quality and into scale-dependent heterogeneity. Multivariate statistics and, in particular, clustering methods can effectively support those analyses, as remarked by a large number of studies in the literature. However, open issues still affect the reliability and general applicability of multivariate statistics and cluster analysis in hydrogeology, especially if a limited number of data and information about well-casing and screen characteristics are available. Such questions are related to the appropriate selection of end-members on one hand, and to the subjectivity of clustering results on the other hand. Starting from hydrogeological data of the Solofrana River basin in Southern Italy, the present paper illustrates an original approach to end-member selection in a pyroclastic-alluvial aquifer, and to the analysis of the geochemical evolution of groundwater in a given basin based on k-means clustering. The proposed approach has been applied to a real dataset collected in July 2010. Comparing the results against classical hydrogeological models and graphical methods typically used to classify water samples, a robust validation of the methodology has been achieved. • A new method to define groundwater end-members in complex aquifer systems is proposed. • An original vector index (EMAC), based on groundwater composition indices, is defined. • The vector index yields a significant data reduction in view of k-means clustering. • Physical insight on EMAC allows a clever selection of initial centroids. • Validation of the method is based on silhouette values and physical models. [ABSTRACT FROM AUTHOR]