HR: 1340h
AN: H23B-1313    [Abstracts]
TI: Prediction Uncertainty Analyses for the Combined Physically-Based and Data-Driven Models
AU: * Demissie, Y K
EM: ydemissi@uiuc.edu
AF: University of Illinois Urbana-Champaign, Department of Civil and Environmental Engineering, Urbana, IL 61801, United States
AU: Valocchi, A J
EM: valocchi@uiuc.edu
AF: University of Illinois Urbana-Champaign, Department of Civil and Environmental Engineering, Urbana, IL 61801, United States
AU: Minsker, B S
EM: minsker@uiuc.edu
AF: University of Illinois Urbana-Champaign, Department of Civil and Environmental Engineering, Urbana, IL 61801, United States
AU: Bailey, B A
EM: babailey@sciences.sdsu.edu
AF: San Diego State University, Department of Mathematics and Statistics, San Diego, CA 92182, United States
AB: The unavoidable simplification associated with physically-based mathematical models can result in biased parameter estimates and correlated model calibration errors, which in return affect the accuracy of model predictions and the corresponding uncertainty analyses. In this work, a physically-based groundwater model (MODFLOW) together with error-correcting artificial neural networks (ANN) are used in a complementary fashion to obtain an improved prediction (i.e. prediction with reduced bias and error correlation). The associated prediction uncertainty of the coupled MODFLOW-ANN model is then assessed using three alternative methods. The first method estimates the combined model confidence and prediction intervals using first-order least- squares regression approximation theory. The second method uses Monte Carlo and bootstrap techniques for MODFLOW and ANN, respectively, to construct the combined model confidence and prediction intervals. The third method relies on a Bayesian approach that uses analytical or Monte Carlo methods to derive the intervals. The performance of these approaches is compared with Generalized Likelihood Uncertainty Estimation (GLUE) and Calibration-Constrained Monte Carlo (CCMC) intervals of the MODFLOW predictions alone. The results are demonstrated for a hypothetical case study developed based on a phytoremediation site at the Argonne National Laboratory. This case study comprises structural, parameter, and measurement uncertainties. The preliminary results indicate that the proposed three approaches yield comparable confidence and prediction intervals, thus making the computationally efficient first-order least-squares regression approach attractive for estimating the coupled model uncertainty. These results will be compared with GLUE and CCMC results.
DE: 1829 Groundwater hydrology
DE: 1846 Model calibration (3333)
DE: 1873 Uncertainty assessment (3275)
SC: Hydrology [H]
MN: 2007 Fall Meeting