Determination of DNAPL entrapment architecture using experimentally validated numerical codes and inverse modeling

Abstract

The presence of dense non-aqueous phase liquid (DNAPL) in source zones in aquifers generates continuous mass flux long after the initial spill. It is our hypothesis that observed dissolved concentration in a monitoring well downstream of the source zone will provide inadequate and often misleading information on the entrapment architecture needed to design effective remediation schemes. An inverse modeling study was conducted to evaluate what additional information is needed to determine entrapment architecture. Three synthetic entrapment architectures generated by a multi-phase flow model UTCHEM in a spatially correlated random field were used in the analysis. The three selected from eighty realizations of the random field represent conditions that produce DNAPL pools, zones of residual DNAPL saturation and a combination of pools and residuals, respectively. For each of these three entrapment architectures, a data set of downstream concentration and solute mass flux was generated using a laboratory validated dissolution model based on MODFLOW and RT3D. An inverse modeling algorithm (PEST) was used to back-calculate the saturation distribution of DNAPL in the source zone. The inverse solution (i.e. DNAPL saturation distribution defining the architecture) was found to be non-unique when only concentration data was used. However when mass flux data that combines concentration and water flow, was added as an additional observation, the inverse problem converged rapidly to a unique solution. Further analysis to determine optimal monitoring strategies showed that the mass flux-matching technique could be used to determine entrapment architecture with some limitations. The technique becomes less accurate in terms of both total mass estimation and the ability to resolve the vertical distribution of DNAPL, when a source zone contains more pools than residuals. It was also found that with larger number of observations in the vertical direction (multi-level sampling points), the predictions become more accurate. The distance the monitoring well was placed downstream of the source zone affected the accuracy of prediction, but the estimate of the total mass entrapment was not affected. This is suspected to be an artifact of the two-dimensional test system that was used in this hypothetical analysis. More rigorous analysis using realistic three-dimensional systems is needed to make more definitive conclusions.