Assessing uncertainty in groundwater flow directions by an iterative Ensemble Smoother technique

A sub-alpine catchment hosts a drinking water plant that collects groundwater through a series of drains. The catchment is crossed by a river that recharges the aquifer with potentially polluted water. The waterworks managers need a management strategy to maximize groundwater collection and minimize the probability to extract river water. This request was addressed by means of a groundwater model that simulates the mixing of river water and groundwater under a stochastic framework.

Samples of river water show the presence of Gadolinium, a rare earth element used as a contrast agent (GBCA) for magnetic resonance imaging. This element is also recurrently found in samples taken from some monitoring wells, and previous studies have determined its suitability as a tracer of solute dispersion. We used it as an indicator of partitioning between river water and original groundwater.

We built a simple, fast-running numerical groundwater model with the FEM code Feflow (DHI). We coupled it with PESTPP-IES, an optimization tool that implements the ensemble-smoother form of the Gauss-Levenberg-Marquardt algorithm. Through it, an ensemble of "realistic" parameter fields was generated, all of which support a good fit between model outputs and the calibration dataset. The latter included mixing ratios (calculated by measured Gadolinium concentrations) and groundwater levels. To simulate Gadolinium spread in groundwater, we used particle tracking instead of building an advective-dispersive transport model, because the latter is costlier to build and slower to run, therefore it does not allow the high number of runs required by PESTPP-IES. Although dispersion is not explicitly represented, its role is surrogated by uncertainty in hydraulic conductivity.

With this study, we built the engine of a decision support system that will optimize waterworks management. We also demonstrated that a lean, purpose-driven model is adequate in simulating solute transport in complex hydrogeological systems. Gadolinium concentrations were instrumental in identifying the partitioning between river water and groundwater.