ADDING THE SPOTS: STOCHASTIC MODELING WITHIN A DETERMINISTIC HYDROSTRATIGRAPHIC FRAMEWORK TO ACCOUNT FOR SMALL-SCALE VARIABILITY AND UNCERTAINTY IN A COMPLEX GLACIAL AQUIFER SYSTEM

Glacial sediments contain variable and complex textures and sedimentary structures that can impede attempts to predict field scale subsurface groundwater flow or contaminant transport behavior. Refinement of deterministic hydrostratigraphic models to account for small-scale variability using stochastic modeling provides a means to assess flow and transport uncertainty. This study illustrates the application of a hybrid modeling approach integrating deterministic and stochastic components to assess uncertainty in selected contaminant transport metrics. The site is located in central Washtenaw County, Michigan, USA, where monitoring wells have been installed in more than 130 locations as part of an ongoing remediation effort.

Natural gamma radiation counts were recorded in monitoring well logs located throughout the site. Measurement intervals in each well were classified as aquifer or aquitard based on their position within a deterministic allohydrostratographic interpretation. Sequential Gaussian simulation was used to create an ensemble of realizations of gamma values conditioned to gamma well log measurements within a 14km² area of a regional groundwater model. Aquifer and aquitard simulations were generated separately and merged into a single 3D model honoring the original allohydrostratographic interpretation. Hydraulic conductivity was then assigned in each 30x30x3m MODFLOW model cell based on an experimentally determined exponential relationship between hydraulic conductivity and gamma count values. 100 stochastic realizations were ranked a priori using harmonic mean K values for flow paths along the primary migration direction between the source area and the Huron River, a potential groundwater discharge location at the site. Relevant transport metrics (e.g., first arrivals and breakthrough times at the river calculated using MODPATH and MT3D) were compared among realizations to evaluate the degree to which stochastic variability influences transport and whether a priori rankings can be used to identify realizations representing the range of transport behavior uncertainty predicted using the full ensemble.