Locating sources of groundwater contamination is an inverse problem on steroids. It is, in fact, two inverse problems. Source locations must be back-calculated from field measurements. The hydraulic properties of aquifer material through which groundwater flows, and contaminants are transported, must also be inferred from these same measurements.
Another spice in the inverse problem soup is the nature of borehole contaminant measurements. These data are messy, to say the least.
Yet another problem is that numerical advection-dispersion models are slow. They are difficult to use in inverse problem settings. Solute concentrations which they calculate are blurred by numerical and hydrodynamic dispersion. The latter is a surrogate for stochastic detail that they cannot represent.
We demonstrate an innovative method to solve this two-tiered inverse problem. It is based on particle tracking. Simulation is fast and stable. The method makes use of the fact that, as well as carrying mass, particles carry information which is capable of being harvested. Outcomes of the analysis are probabilistic (as they should be). Meanwhile, uncertainties are reduced through assimilation of head and concentration data.
In addition to providing probabilistic maps of source location, the methodology can be used to optimise the locations of new monitoring wells.
The focus of our worked example report is a shallow aquifer that underlies Pavia, a city in northern Italy. Its story is typical of countless other stories that are told throughout the world. The aquifer is contaminated. Contamination must be contained. Responsible parties are sought.