Geostatistical approaches to characterizing the hydrogeology of glacial drift.

摘要

Subsurface correlation of aquifer and aquitard materials can be challenging, especially in glaciated terrain, because of complexity in the thickness and lateral extent of the depositional units and a general scarcity of data. Geostatistical methods provide a means for exploring that correlation structure, allowing comparisons of results from different geomorphological settings and approaches to populating the hydraulic conductivity arrays of numerical groundwater flow models.;In this study, indicator geostatistics and transition probability geostatistics are applied to nearly 300 km of drilling data derived mainly from over 11,000 private well drilling logs. Analyses are focused on a nearly 12,000-km 2 region in central Minnesota comprised of four counties plus a portion of a fifth, in a setting dominated by the Rainy, Superior, and Wadena lobes of late-Wisconsinan glaciation. The area of interest surrounds Camp Ripley, a 210-km2 Minnesota National Guard training facility. Data for Camp Ripley include 208 monitoring wells with detailed logging information obtained through split spoon sampling and nine even higher quality rotasonic boreholes. These data provide a basis for comparison of high-quality data to the abundant water well logs in the surrounding counties.;The drilling data are categorized according to the logged sediment descriptions. In one scheme, the materials are separated according to presumed high or low hydraulic conductivity (K) for use in binary indicator geostatistical and transition probability geostatistical analyses. In another, the number of categories is expanded to five on the basis of combined K and depositional setting information for use in a separate transition probability geostatistical analysis.;The drilling data are separated into different geomorphological settings associated with various depositional environments of several ice advances. Vertical variograms are very well supported in all geomorphological areas, while lateral variograms vary from well supported to indecipherable scatter. The ratio of vertical to lateral range varies but has an average value of ten. Results for the numerous geomorphological settings indicate overlapping geostatistical parameters in terms of ranges, sills, and vertical lens thicknesses. A lack of stationarity was observed, consistent with a fundamental complexity of glacial depositional and erosional processes. Correlation generally varied as much between geographically distinct zones of like geomorphology as it did between zones of different geomorphology. High-resolution data associated with monitoring well installation typically deviated from the private well data; this is attributed to site-specific geology and detailed logging of thin units.;Indicator variography and kriging, as well as binary and 5-category transition probability geostatistical simulation, were applied to a drilling data set in the Camp Ripley vicinity to evaluate their relative capability for 3-D numerical modeling of regional groundwater flow in the glacial drift. For the kriging approach, the flow model relied on the kriging results by grading the K values between high and low values on the basis of the indicator output. Calibration was performed by systematic testing of various plausible end members of the K range. The transition probability runs were calibrated by stochastic inverse modeling. A baseline single-K analysis was also calibrated through inverse modeling. Resulting K values from all four approaches were comparable, with the indicator kriging providing a slight edge in terms of the calculated errors at calibration targets. The modeling suggests that in complex glaciated terrain, in which lateral correlation of hydrostratigraphic units is small relative to grid spacing and drilling data spacing, a single-K model provides a suitable approach to determining regional groundwater flow, despite the large contrasts in K prevalent in glacial drift materials.