Matthew D. Fortney (md-fortney@northwestern.edu)

and

Howard W. Reeves, Assistant Professor (h-reeves@northwestern.edu)

Department of Civil Engineering

Northwestern University

A116 Technological Instute

2145 Sheridan Road

Evanston, IL 60208-3109

The ability to reliably predict the behavior of groundwater systems to additional stresses has become increasingly important for both groundwater supply and groundwater remediation. Uncertainty in the predictions of groundwater models used in design arises from the inherent uncertainty in the inputs to the models. Accounting for these uncertainties has become ever more important as the sophistication and complexity of groundwater models increases. This research seeks to provide an objective measure of reliability through use of a reliability index.

Overall, the uncertainty in system performance may be expressed as a reliability index. This index relates the predicted ability of the design to meet the design objectives to the uncertainty in the predicted performance. The uncertainty in predicted performance is estimated by combining model sensitivity with geologic parameter uncertainty. The impact on model prediction of the spatially-correlated uncertainty of the input depends on the sensitivity of the numerical model to changes in the input values.

This research capitalizes on features of the USGS groundwater flow model MODFLOW-2000. MODFLOW-2000 can be used to calculate the sensitivities of the groundwater model to changes in selected input parameters. In the program, these sensitivities are used in an inverse procedure to calibrate the numerical model to field observations. In this research, these sensitivities are combined with geologic uncertainties in a Taylor series to yield a first-order second-moment approach (FOSM) that produces an estimate of performance uncertainty. This estimate can be used to assess the reliability of predicted impacts and can be used to compare different designs or management strategies. The spatial uncertainties generated through this method also can be used to direct site exploration to points that will have the greatest impact on the model. Through a Bayesian approach, new data collected at the site can be incorporated into the model, and new estimates of reliability can be made.