Sampling Network Design for Transport Parameter Identification

Abstract

An optimal experimental design algorithm is developed to facilitate the planning and the optimal configuration and scheduling of a ground‐water‐tracer test whose data are to be used to estimate model parameters. A maximal information criterion is used to select among competing designs. The proposed criterion is equivalent to a weighted sum of squared sensitivities, employing the observation that parameters are most accurately estimated at points with high sensitivity to the parameter but that the relative magnitudes of sensitivities to different parameters are different. The fundamental advantage of this criterion is that it is comparatively simple. The influence‐coefficient method is used to compute the sensitivity coefficient. A zero‐one integer hueristic is used to solve a simplified example for experiment configurations under a given experimental duration. The design considers the installation cost, which is a function of location and depth of the observation well and of the samples themselves. The resulting designs are intuitively reasonable. It was found that a dramatic increase in information can be obtained with an experimental budget increase in a heterogeneous example case.