INFLUENCE OF DETERMINISTIC AND STOCHASTIC HYDRAULIC CONDUCTIVITY DISTRIBUTIONS ON PREFERENTIAL FLOW PATHWAYS IN A GLACIAL AQUIFER SYSTEM

Hybrid models incorporating stochastic variability within a deterministic hydrostratigraphic framework provide an effective way to assess uncertainty in flow and transport model predictions. This study evaluated the distribution of groundwater flow and contaminant transport pathways in two ensembles of spatially variable hydraulic conductivity (K) distributions. The models comprised an 80m thick sequence of Pleistocene glacial sediments in in an approximately 20 km² area across Washtenaw County, Michigan. Conditioned K fields were generated using Sequential Gaussian Simulation (SGS) and Sequential Indicator Simulation (SIS) constructed using indicator classes based on natural gamma ray logs from 77 monitoring wells. K fields were modeled independently for aquifer and aquitard materials and subsequently embedded within a 3D MODFLOW model constructed using a deterministic framework of eight aquifer and aquitard layers.

MODPATH was used to track the pathways of 100 particles released as line sources at five depth intervals with documented 1,4-dioxane concentrations along the boundary of the contaminant source area. Pathways for 100 realizations of each ensemble were combined to produce cross sections showing the frequency of particles passing through individual model cells along transects spaced at 1 km intervals downgradient of the line source and adjacent to the hypothesized groundwater discharge location along the Huron River approximately 8 km from the line source. Results revealed dispersion of preferential flow pathways related to stochastic K variability proximal to the line source, followed by increasing channelization into preferential flow pathways related to the deterministic hydrostratigraphic framework farther downgradient. Differences in the spatial distribution of particle pathways observed between the ensembles of SIS and SGS ensembles increased with increasing travel distances. These results suggest that the relative influence of deterministic versus stochastic components of hybrid models depends upon the scale of the flow and transport problem under investigation.