Hydrogeochemical spatialization and controls of the Serra Geral Aquifer System in southern Brazil: a regional approach by self-organizing maps and k-means clustering

Abstract The aim of this study is to investigate the regional hydrogeochemical spatialization and controls of the Serra Geral Aquifer System (SGAS), a transboundary fractured aquifer, across the southern region of Brazil. An extensive dataset of 1,564 groundwater wells represented by 16 attributes was analyzed to identify spatial patterns and groups with similar hydrogeochemical facies. An unsupervised machine learning approach, self-organizing maps (SOM), was used in combination with k-means clustering to carry out the analysis. SOM produces two-dimensional representations, allowing visual interpretation of nonlinear relationships between the attributes. The stochastic Davies-Bouldin index pointed out to an optimal number of four clusters, highlighting significant differences in geochemistry. Cluster 1 is the most abundant and widespread, corresponding to meteoric recharge; cluster 2 is influenced by the weathering of basaltic rocks, being widely distributed, but with a higher density at the southeastern region; cluster 3 is recognized as a mixing between all the other groups, with a sparse distribution, mostly in the extreme north of the area; cluster 4 is dominated by ascending flow from the underlying sedimentary aquifers, occurring in restricted areas. This study shows that SOM can identify large-scale spatial hydrogeochemical patterns of the SGAS driven by structural and stratigraphic elements.