GROUNDWATER FLOW MODEL CALIBRATION DIFFICULTIES IN AREAS WITH GLACIALLY-DEPOSITED AQUITARDS: AN EXAMPLE FROM GLACIAL LAKE OSHKOSH

In Outagamie County, Wisconsin, glacio-lacustrine sediments form a regional aquitard comprised of very low hydraulic conductivity (K) lacustrine clays with embedded sand and gravel bodies. Several techniques were used to infer preferential flow paths occurring through connected high K zones at this study site. A combination of hard and soft data was used to create 300 three-dimensional hydrostratigraphic models of the site using multiple-point geostatistics. A representative set of six hydrostratigraphic models was selected, imported into groundwater flow models, and calibrated to head data. Results from particle tracking indicated preferential flow of individual particles moving through high K units in faster time than nearby particles moving through low K units.

During the calibration process, several problems arose related to boundary conditions that did not occur in previous models of glacial aquifers in Wisconsin. Specified head boundaries obtained using telescopic mesh refinement from a larger regional model were much too high for the glacial Lake Oshkosh site, likely due to a lack of calibration data in that area, less detailed information for the glacial deposits in the larger regional model relative to the telescoped model, the large volume of low K material, and, most importantly, the existence of strong vertical flow at this site. Additionally, a soil-water balance model and baseflow estimates at the basin scale gave recharge rates that were higher than calibrated recharge values. Comparison of particle tracking results and isotope values from previous studies suggests that a lower value for K of clay is more appropriate than was used in the calibration, which would allow higher values of recharge in the model. In general, when working in areas with vertical flow through thick glacial aquitards, local stream gaging data are essential for model calibration at the site scale and caution should be used in assigning boundaries from a larger regional model.