Geostatistical Simulation of High-Transmissivity Zones at the Mirror Lake Site in New Hampshire: Conditioning to Hydraulic Information

By Frederick D. Day-Lewis, Paul A. Hsieh, Allen M. Shapiro, and Steven M. Gorelick

ABSTRACT

A new approach is presented to include hydraulic information as conditioning data in geostatistical simulation of high-transmissivity zones. A simulated-annealing algorithm is used to generate three-dimensional geostatistical realizations conditioned to borehole data and inferred hydraulic connections between packer-isolated borehole intervals. High-transmissivity zones were identified by Hsieh and Shapiro (1996) in the bedrock underlying the FSE well field at the U.S. Geological Survey Fractured Rock Research Site near Mirror Lake, Grafton County, New Hampshire. These zones are conceptualized to consist of connected, highly transmissive fractures that are embedded within a surrounding network of less transmissive fractures. During multiple-well hydraulic tests, well intervals connected by a high-transmissivity zone exhibit different responses. To analyze the test data, Hsieh and others (1999) constructed and calibrated a deterministic ground-water flow model. In this study, alternative spatial patterns of hydraulic properties are based on geostatistical realizations. The simulated-annealing algorithm is used to generate conditional realizations of high-transmissivity zones in the bedrock underlying the FSE well field, using an indicator-variogram model of spatial variability. Statistical analysis of the generated zones yields three-dimensional images of the probability that a high-transmissivity zone occurs and the likely spatial extents of specific zones. For selected realizations, ground-water flow is simulated with a finite-element model. Hydraulic conductivity and specific storage values of high-transmissivity zones and background rock are calibrated. Simulation results for realizations conditioned to borehole and hydraulic connection data compare favorably to the test data, and realizations exhibit more complex spatial variability than previous deterministic modeling by Hsieh and others (1999).