Optimal observation network design for parameter structure identification in groundwater modeling

Abstract

Designing a pumping observation experiment for parameter identification in groundwater modeling is a very challenging problem because the parameter structure, such as the hydraulic conductivity of a heterogeneous aquifer, is usually unknown. Generally, to identify a more complex parameter structure requires more observation data. Traditional experimental design methods have two disadvantages: First, the sufficiency of information provided by a design is not considered in the design stage. Second, the decision variables often are dependent on the unknown parameter because of the nonlinearity of the model. As a result, a sequential design observation process should be used, but this often is impractical in groundwater modeling. This paper presents a new methodology for experimental design that can find a minimum cost design to provide sufficient information for identifying both the parameter structure and parameter values. Sequential Gaussian simulation is used to generate different realizations for the unknown parameter. For each possible realization the number of observation wells is increased gradually during the design process until the information provided by the design is sufficient. The selection criterion for locating a new observation well is the maximization of the information content for parameter identification. The overall sufficiency of a design is assessed by Monte Carlo simulation. The proposed methodology is explained in detail by a numerical example.