NUMERICAL SIMULATIONS OF TWIN LAKE NATURAL-GRADIENT TRACER TESTS - A COMPARISON OF METHODS

The primary objectives of the Twin Lake experiments have been to study geologic heterogencities and field-scale dispersion and to provide da ta for developing and evaluating groundwater flow and transport models . The experiment reported here was undertaken in 1987-1988 to expand t he studies of heterogeneity and dispersion from a 40-m scale investiga ted in 1983 to a full 270-m scale from the point of tracer injection t o the groundwater discharge zone. The hydrogeologic setting, the exper imental and analytical methods, and the interpretation of experimental data are the topics of discussion in this paper. The 1983 and 1987-19 88 tracer tests were used to evaluate the predictive abilities of vari ous numerical groundwater flow and transport models. The modeling is d one along the mean flow direction in the vertical plane. The two-dimen sional finite-element modeling was successful in simulating the hydrau lic head distribution but not the groundwater motion or the tracer tra nsport. In addition, owing to the advection-dominated nature of the gr oundwater flow, the transport simulation suffered from numerical dispe rsion. Significant improvements in simulations were obtained using met hod-of-characteristics- and random-walk-based computer models. The inf luence of dimensionality on the transport simulation was evaluated by comparing the two- and three-dimensional random-walk simulations. It w as found that over the first 40 m along the mean flow path, the three- dimensional simulation does not reproduce the plume migration any bett er than the two-dimensional simulation does. Random-walk simulations o f the transport observed in the 1983 and 1987-1988 tests are also done on the basis of the advection-dispersion equation and tracer test-der ived velocities without invoking the concept of hydraulic conductivity and Darcy's law. These simulations were found to be superior to simul ations based on the coupled flow and transport models. It was also fou nd that the simulated hydraulic head distribution is insensitive to sp atial variations in hydraulic conductivity and the simulated groundwat er velocity distribution is inappropriate for transport modeling. It w as concluded that the advection-dispersion models of the local-scale t ransport perform better in predicting the tracer migration at the Twin Lake aquifer using measured velocities rather than velocities simulat ed by a flow model.