HR: 09:15h
AN: H11E-06    [Abstracts]
TI: Joint Assessment of Parameter and Conceptual Model Uncertainty at the Hanford Site 300 Area
AU: Meyer, P D
EM: philip.meyer@pnl.gov
AF: Pacific Northwest National Lab, 620 SW Fifth Ave Ste 810, Portland, OR 97204 United States
AU: * Ye, M
EM: ming.ye@dri.edu
AF: Desert Research Institute, 755 E. Flamingo Road, Las Vegas, NV 89119 United States
AU: Neuman, S P
EM: neuman@hwr.arizona.edu
AF: University of Arizona, 1133 E. North Campus Drive, Tucson, AZ 85721 United States
AB: Assessment of conceptual model and parameter uncertainties is important for science-based decision making, long-term stewardship and monitoring network design. We use the recently developed Maximum Likelihood Bayesian Model Averaging (MLBMA) method to assess jointly the uncertainty associated with alternative flow and transport models, and their parameters, at the 300 Area of the DOE Hanford Site in Washington. Eight groundwater flow and transport model alternatives are considered representing various levels of model complexity, e.g., the degree to which they resolve temporal variations in river stage and spatial variations in hydraulic and transport parameters. The eight models are implemented using MODFLOW and MT3DMS in conjunction with the statistical parameter estimation package PEST and its sensitivity routine SENSAN within the framework of GMS (Groundwater Modeling System). Each of the eight flow and transport models is calibrated jointly against site measurements of hydraulic head and concentration. Parameters found to be insensitive to these measurements and measurements found to have negligible effect on model parameters are excluded from the calibration. The relative plausibility of each calibrated model is expressed in terms of a posterior model probability computed on the basis of Kashyap's information criterion KIC. Models having relatively small posterior probabilities are discarded and those with relatively large posterior probabilities are retained to obtain MLBMA predictions of flow and transport at the site. This is done by (a) using the parameter estimates and their covariance matrices to generate numerous Monte Carlo realizations of model parameters; (b) generating corresponding realizations of predicted heads, concentrations and fluxes; (c) computing the mean and variance of these quantities over all realizations; and (d) averaging the latter over all models using their posterior probabilities as weights. We conclude by comparing the predictive performance of each individual model with that of MLBMA.
DE: 1829 Groundwater hydrology
DE: 1832 Groundwater transport
DE: 1846 Model calibration (3333)
DE: 1847 Modeling
SC: Hydrology [H]
MN: Fall Meeting 2005