Prediction of pollutant remediation in a heterogeneous aquifer in Israel: Reducing uncertainty by incorporating lithological, head and concentration data

For large contaminant plumes in groundwater, Pump and Treat (PT) is usually applied, not only to remediate the aquifer, but also to control the spreading of the plume. The aquifer structure, as well as the spatial distribution of the contaminant plume, are typically subject to uncertainty. This uncertainty can highly effect the remediation predictions. Geostatistical tools combined with Monte-Carlo simulations can be used to quantify those uncertainties and to provide a more established basis for decision makers regarding the remediation strategy. Here, we focus on a case study of a polluted aquifer and inspect its response to remediation by a PT scheme using two pumping wells. The aquifer structure is combined of 4 lithological units and indicator realizations were generated using a modified sequential indicator simulation (SIS) approach. Then, the initial plume prior to remediation was estimated by a forward predictive model. As a result, each realization had a unique spatial distribution of the contaminant at the onset of the remediation. Spatio-temporal measurements of head and contaminant concentration were used to screen inappropriate realizations. Our results have shown that conditioning the realizations to the head alone has a minor effect upon the uncertainty regarding the remediation efficiency (RE). On the other hand, by conditioning the realizations to both head and concentrations, the uncertainty was reduced by more than 35% in terms of the standard deviation of RE. Risk analysis for the migration of the plume showed that the ability of the current configuration of the remediation wells to control the plume is limited to the top layers of the aquifer. Thus, with the present monitoring network, there is a significant risk that the plume will migrate downstream towards production wells, without being noticed.