Structure Adaptation in Stochastic Inverse Methods for Integrating Information

Abstract

The use of inverse modeling techniques has greatly increased during the past several years because the advances in numerical modeling and increased computing power. Most of these methods require an a priori definition of the stochastic structure of conductivity (K) fields that is inferred only from K measurements. Therefore, the additional conditioning data, that implicitly integrate information not captured by K data, might lead to changes in the a priori model. Different inverse methods allow different degrees of structure adaptation to the whole set of data during the conditioning procedure. This paper illustrates the application of a powerful stochastic inverse method, the Gradual Conditioning (GC) method, to two different sets of data, both non-multiGaussian. One is based on a 2D synthetic aquifer and another on a real-complex case study, the Macrodispersion Experiment (MADE-2), site on Columbus Air Force Base in Mississippi (USA). We have analyzed how additional data change the a priori model on account of the perturbations performed when constraining stochastic simulations to data. Results show how the GC method tends to honour the a priori model in the synthetic case, showing fluctuations around it for the different simulated fields. However, in the 3D real case study, it is shown how the a priori structure is slightly modified not obeying just to fluctuations but possibly to the effect of the additional information on K, implicit in piezometric and concentration data. We conclude that implementing inversion methods able to yield a posteriori structure that incorporate more data might be of great importance in real cases in order to reduce uncertainty and to deal with risk assessment projects.