HR: 1330h
AN: H12E-1026    [PDF]
TI: Optimal Observation Network Design for Parameter Structure Identification in Groundwater Modeling
AU: * Chang, L
EM: lifang@ucla.edu
AF: UCLA, Department of Civil and Environmental Engineering, 5731 Boelter Hall, Los Angeles, CA 90095 United States
AU: Sun, N
EM: nezheng@seas.ucla.edu
AF: UCLA, Department of Civil and Environmental Engineering, 5731 Boelter Hall, Los Angeles, CA 90095 United States
AU: Yeh, W W
EM: williamy@seas.ucla.edu
AF: UCLA, Department of Civil and Environmental Engineering, 5731 Boelter Hall, Los Angeles, CA 90095 United States
AB: This research develops a methodology for optimal observation network design for parameter structure identification in groundwater modeling. The design is formulated as an integer nonlinear programming problem. The design objective is to minimize experimental cost subject to data sufficiency requirement. By incorporating the data sufficiency requirement as a constraint in the optimization model, the proposed methodology quantitatively unifies observation network design, model structure identification and model application reliability. Because the optimal design is predicated on parameter values, which themselves are to be estimated before performing the experiments, it is extremely important to utilize and integrate prior information into the design problem. In this study, we use a geostatistical simulation method to generate a large set of realizations for the real parameter field according to the available prior information. For each realization, we solve the integer nonlinear problem and find the minimum cost design that satisfies the data sufficiency requirement. After solving the design problems for each of the realizations, we analyze the overall results. We calculate the probability of a potential observation well location that has been chosen from all the realizations. Additionally, we analyze the reliability of providing sufficient data for each given cost. The criterion used to select the observation well location is to maximize a norm of the information matrix. We use the variational method (the adjoint state method) to compute the sensitivity coefficients which form the information matrix. When the parameter structure is unknown, the adjoint state method is highly efficient in computing the sensitivity coefficients for each of the computation nodes.
DE: 1829 Groundwater hydrology
DE: 3260 Inverse theory
SC: Hydrology [H]
MN: 2003 Fall Meeting