Geostatistical multimodel approach for the assessment of the spatial distribution of natural background concentrations in large-scale groundwater bodies

Quantification of the (spatially distributed) natural contributions to the chemical signature of groundwater resources is an emerging issue in the context of competitive groundwater uses as well as water regulation and management frameworks. Here, we illustrate a geostatistically-based approach for the characterization of spatially variable Natural Background Levels (NBLs) of target chemical species in large-scale groundwater bodies yielding evaluations of local probabilities of exceedance of a given threshold concentration. The approach is exemplified by considering three selected groundwater bodies and focusing on the evaluation of NBLs of ammonium and arsenic, as detected from extensive time series of concentrations collected at monitoring boreholes. Our study is motivated by the observation that reliance on a unique NBL value as representative of the natural geochemical signature of a reservoir can mask the occurrence of localized areas linked to diverse strengths of geogenic contributions to the groundwater status. We start from the application of the typical Pre-Selection (PS) methodology to the scale of each observation borehole to identify local estimates of NBL values. The latter are subsequently subject to geostatistical analysis to obtain estimates of their spatial distribution and the associated uncertainty. A multimodel framework is employed to interpret available data. The impact of alternative variogram models on the resulting spatial distributions of NBLs is assessed through probabilistic weights based on model identification criteria. Our findings highlight that assessing possible impacts of anthropogenic activities on groundwater environments with the aim of designing targeted solutions to restore a good groundwater quality status should consider a probabilistic description of the spatial distribution of NBLs. The latter is useful to provide enhanced information upon which one can then build decision-making protocols embedding the quantification of the associated uncertainty.