Reducing Uncertainty Associated with Ground‐Water Flow and Transport Predictions

Abstract

Abstract Effective evaluation of ground‐water flow and transport problems requires consideration of the range of possible interpretations of the subsurface given the available, disparate types of data. Geostatistical simulation (using a modified version of ISIM3D) of hydrofacies units produces many realizations that honor the available geologic data and represent the range of subsurface interpretations of unit geometry. Hydraulic observations are utilized to accept or reject the geometric configurations of hydrofacies units and to estimate ground‐water flow parameters for the units (using MODFLOWP). These realizations are employed to evaluate the uncertainty of the resulting value of the response function (ground‐water flow velocity and contaminant concentration) using MT3D. The process is illustrated with a synthetic data set for which the “truth” is known, and produces a striking reduction in the distribution of predicted contaminant concentrations. The same system is evaluated three times: first with only hard data, then with both hard and soft data, and finally with only the realizations that honor the hydraulic data (i.e., those accepted after parameter estimation via inverse flow modeling). Using only hard data, the mean concentration predicted for all realizations at the point of interest is nearly two orders of magnitude lower than the true value and the standard deviation of the log of concentration is two. The addition of soft data brings the mean concentration within one order of magnitude of the true value and reduces the standard deviation of the log of concentration to one. After eliminating realizations using inverse flow modeling, the mean concentration is one‐third of the true value and the standard deviation of the log of concentration less than 0.5.