Evaluation of Concentration Data and Prior Information on Uncertainty in Simulations of Ground Water Flow in Four Hydrogeologic Settings

The U.S. Geological Survey’s National Water Quality Assessment topical study Transport of Natural and Anthropogenic Contaminants (TANC) to Supply Wells produced four detailed simulation models of ground-water flow. The purpose of the models was to estimate areas contributing recharge (ACR) to public-supply wells, which are essential to understanding the relation between land use and drinking-water quality. Simulation models commonly are used to estimate ACR, but the effects of uncertainty in the models often are not considered. The four models evaluated in this study represent very different hydrogeologic settings in California, Connecticut, Florida, and Nebraska.
Probabilistic ACR were predicted for each of these areas using Monte Carlo simulation.  Parameter realizations for the Monte Carlo simulation were derived from the parameter covariance matrix, which in turn was estimated using regression-based parameter estimation. All four models were calibrated using ‘parsimonious’ formulations, that is, hydraulic properties were simplified using the zonation approach. Even with this simplification, not all parameters could be estimated with available observations. Parameters that could not be estimated might affect predictions of the ACR; therefore, covariance matrices were calculated so as to include unestimated parameters. Often the covariances for unestimated parameters were so large that most Monte Carlo runs did not converge. Better (more frequent) convergence was obtained by including prior information on unestimated parameters. Including concentration data, such as tritium concentrations, in the model calibration and in the estimation of the covariance matrices also helped to obtain reasonable ACR. The simulations indicate that parameter uncertainty affects predicted travel times and that longer travel times lead to greater uncertainty in the estimated ACR. This result can be used to predict the probability of sampling various combinations of water from different sources to a well.