Influence of alternative models of spatial variability on solute transport, DNAPL entrapment, and DNAPL recovery in a homogeneous, nonuniform sand aquifer

Modeling decisions concerning the representation of spatially variable aquifer properties can exert a fundamental influence on numerical groundwater flow and contaminant transport model predictions. The purpose of this research was to evaluate the influence of alternative aquifer spatial variability models on such predictions for transport of both dissolved (aqueous phase) and immiscible (nonaqueous phase) contaminants within a relatively homogeneous but nonuniform aquifer. The approach involved the construction of deterministic and stochastic aquifer models of a glacial sand aquifer located in Oscoda, Michigan, USA, for use in numerical flow and transport simulations. Specific applications examined included a forced injection bromide tracer test, infiltration and entrapment of a tetrachloroethylene (PCE) spill, and dense nonaqueous phase liquid (DNAPL) recovery and effluent mass flux in response to surfactant-enhanced aquifer remediation.; Results from three-dimensional solute tracer transport modeling indicated that: (1) idealized, end-member deterministic models could constrain average transport behavior predicted by stochastic ensembles; (2) stochastic ensembles generated using sequential Gaussian simulation and sequential indicator simulation produced similar (overlapping) spaces of transport model prediction uncertainty; and (3) under certain conditions, advective particle track statistics could be used to identify stochastic ensemble realizations that displayed extremes in breakthrough metrics.; Results from two-dimensional DNAPL infiltration and dissolution simulations indicated that: (1) the degree of spatial correlation in capillary entry pressures exerts a controlling influence on predicted DNAPL entrapment; (2) complex DNAPL source zone architectures are not readily amenable to a priori definition based on correlation with permeability or entry pressure fields due to the dependence of entrapped organic saturation on infiltration pathways; (3) realizations sharing identical porosity and permeability distributions exhibited significant differences in predicted DNAPL saturations and distributions depending on whether entry pressures were assigned on the basis of Leverett scaling to permeability or calculated using grain size distributions associated with geostatistical indicator classes; (4) removal of 60% to 99% of entrapped PCE reduced dissolved contaminant concentration and mass flux by approximately two orders of magnitude; and (5) notable differences in predicted mass flux as a function of % DNAPL removed were observed for models incorporating Leverett-scaled versus unscaled entry pressure and permeability fields.