Numerical modeling for DNAPL source characterization

DNAPL (dense non-aqueous phase liquid) contamination in the subsurface is a persistent environmental problem that threats groundwater resources. In order to clean up DNAPL contaminated sites, it is important to characterize the distribution of DNAPL sources in the field. However, DNAPL source delineation is a difficult task due to both the complexity of DNAPL distribution and the inadequate capability of current DNAPL detection technologies. It is recognized that a better understanding of DNAPL migration in site formations can assist the process of DNAPL source delineation. Numerical modeling is an efficient way to achieve this.; This research used numerical modeling of two-phase flow to predict the features of DNAPL migration in water saturated porous medium. It was applied to three studies: (1) assisting the design of a controlled release experiment by identifying the relationship between release conditions and DNAPL distribution features, (2) evaluating the influence of the sampling density of permeability on the uncertainty in model predicted DNAPL migration features through Monte Carlo simulations of a two-dimensional flow, and (3) exploring spatial sampling design under uncertainty for identifying the horizontal presence of DNAPL, developed on the basis of Monte Carlo simulations of two-dimensional DNAPL migration.; Numerical simulations of DNAPL infiltration in controlled release experiments were conducted at a wide range of release heads. It was found that the release head could not significantly influence the distribution of DNAPL in the test cell. Based on simulation results, it was suggested that a constant-flux injection design might be a better strategy than a constant-head injection design, since it controls the infiltration rate and is less sensitive to changes in soil properties.; The hypothesis that permeability sampling may reduce the uncertainty in model predicted DNAPL migration features was tested through Monte Carlo simulations of DNAPL-water flow that incorporated the variability in permeability and fluid-medium constitutive relationships. The results suggested that denser permeability sampling can significantly reduce the uncertainty in the space between DNAPL fingers, and moderately reduce the uncertainty in the center and size of the DNAPL-contaminated area, while achieving relatively small reduction for the uncertainty in the volume-aggregated features-average saturation and DNAPL-invaded area.; Monte Carlo simulations of DNAPL spill into heterogeneous field were conducted to observe the variability of DNAPL distribution at two aquifers with different degrees of permeability contrast (variance). Based on simulation results, spatial sampling design under uncertainty was conducted with the objective of minimizing the expectation of the relative classification error for delineating the horizontal extent of DNAPL presence. It was found that the optimized designs exhibited similar patterns for both aquifers. However the spaces between sampling probes were significantly larger at high permeability contrast. It was also found that deep sampling is needed for aquifer formations with high permeability contrast, while a shallow sampling might be sufficient for aquifer formations with low permeability contrast.