Are bottled mineral waters and groundwater for human supply different?

Abstract

Increasingly, bottled natural mineral water (NMW) is proposed as a healthy and safe alternative to supply water. However, tap supply water often comes from aquifers (TGW), even from the same aquifers as NMW, sharing the exact formation mechanisms and mineralization processes. Therefore, it is hypothesized that NMW and TGW cannot be distinguished. The chemical composition of TGW and NMW samples in Spain has been compared using five criteria: expert judgment, hydrochemistry, legal regulations, statistical analysis, and machine learning (ML). Hydrochemical criteria included all the NMW samples in the TGW group, as did the legal criterion, whereas classical statistical analysis could not find significant differences between the two groups. Although experts could correctly differentiate a small subsample of both types of water with an accuracy of 0.67, ML-based classification with Extreme Gradient Boosting yielded a balanced accuracy of 0.92 on an extremely imbalanced data set. Shapley Additive Explanations (SHAP) analysis identified pH, SiO2, E, K+, Ca2+, K+/Na+ and NO3− as the most relevant variables for water type discrimination. The overall consistency and generalization ability of the ML classifier has been proven by the spatial distribution of hits and misses, where the few cases of indistinguishable waters seem to be related to proximity to nature reserves (i.e., land use) more than to geological characteristics. Therefore, it can be concluded that NMW and TGW are indeed different and that only ML could find the hidden structure in the chemical data that determines the differences. This structure originates in how the market and consumers decide which water is ultimately bottled. The results can help on future choices of TGW and NMW in a context of water scarcity.