HR: 1340h
AN: H13A-1315    [Abstracts]
TI: Bayesian Model Averaging on Parameterization Non-uniqueness and Conditional Uncertainty Analysis
AU: * Tsai, F T
EM: ftsai@lsu.edu
AF: Department of Civil and Environmental Engineering, Louisiana State University, 3507 CEBA Building, Baton Rouge, LA 70803
AB: Understanding of subsurface heterogeneity, e.g., hydraulic conductivity, is inherently difficult in that natural heterogeneity and processes are extremely complex and the available data are limited. Although the parameter structure error in groundwater modeling has been assessed with one parameterization method (zonation or interpolation), with limited information many parameterization methods may interpret the same data satisfactorily. To cope with the non-uniqueness problem of parameterization, we introduce a Bayesian model averaging (BMA) method to integrate multiple parameterization methods in a Bayesian geostatistical framework. Moreover, a generalized parameterization (GP) method is adopted to estimate the highly complex spatial distribution of parameter heterogeneity. In this study, we combine BMA and GP as a Bayesian multi-parameterization (BMP) method to better represent the heterogeneity and reduce the model prediction uncertainty. The BMP avoids over-confidence in a single parameterization method. The proposed methodology is conducted in a numerical example where the spatially distributed hydraulic conductivity is estimated. The optimal weighting coefficients embedded in GP are identified through the maximum likelihood estimation (MLE) where the misfits between the observed and calculated groundwater heads are minimized. The conditional means and conditional covariances of the estimated hydraulic conductivity distribution are obtained to assess the estimation uncertainty.
DE: 1829 Groundwater hydrology
DE: 1869 Stochastic hydrology
DE: 3260 Inverse theory
SC: Hydrology [H]
MN: Fall Meeting 2005