A New Method of Estimating Soil Hydraulic Parameter Uncertainty and Heterogeneity Using Bayesian Updating and an Artifical Neural Network

Affiliation(s)PIProject periodFunded by
DHS Zhu, Jianting 01/01/2006 - 12/31/2009 DOE EPSCOR

Project Description

This proposed project will complete a comprehensive and quantitative study to predict heterogeneous soil hydraulic properties and the associated uncertainties. By applying the proposed innovative methods, we seek to thoroughly address a number of important issues related to the hydraulic property characterizations. We propose to extend a Bayesian updating method developed by our national laboratory collaborator to estimate soil hydraulic parameter probabilistic distributions, develop innovative artificial neural network (ANN) based pedo-transfer functions (PTFs) that incorporate the hydraulic parameter distributions and correlations and a new Bayesian geostatistical method to characterize soil heterogeneities. More specifically, the proposed extension will eliminate the potentially restrictive assumption that soil hydraulic parameters follow a normal distribution and will thus be more robust for practical applications. The proposed new neural network based PTFs for predicting hydraulic parameters will incorporate parameter distributions obtained from the Bayesian updating method and penalize any artificial correlations of hydraulic parameter estimates. The new Bayesian geostatistical inference method will be developed to estimate spatial correlation scale of input basic soil attributes to the developed neural network based PTFs. Heterogeneous input fields will be generated according to the statistics and the spatial correlation scales estimated by the new Bayesian geostatistical inference method and fed into the neural network based PTFs to create heterogeneous fields of the soil hydraulic parameters that can be used for numerical simulations and uncertainty analyses. The developed methods will be implemented using the existing data from the entire Hanford site and the robustness of the developed methods will be evaluated by applying them to numerical simulations of the vadose zone processes at the Hanford 300 Area, where soil parameter measurements are insufficient for comprehensive and accurate numerical simulations of uranium distributions. The method proposed in this research is aimed at facilitating numerical modeling to simulate physical and chemical processes in the vadose zone, where most of the contaminants occur and act as sources to the groundwater contaminations. We believe the results of proposed numerical simulation in this project can be used to provide insights to the DOE mission of appropriate contamination characterization and environmental remediation. The PIs have a very strong background in vadose zone hydrology. The national laboratory collaborator is conducting uncertainty analysis at the Hanford 300 Area and is well connected with other researchers actively involved in studies at this area. The project will leverage these previous and ongoing efforts. After being thoroughly evaluated, the developed method will be ready to be used for subsurface process simulations at other areas at the DOE Hanford Site. This project will initiate and strengthen a close and intense collaboration between researchers at the Desert Research Institute in the EPSCoR State of Nevada and their colleagues at the Pacific Northwest National Laboratory.