HR: 1340h
AN: H13C-1346    [Abstracts]
TI: Time-lapse Inversion of Electrical Resistivity Data
AU: * Nguyen, F
EM: f.nguyen@fz-juelich.de
AF: Research Center Juelich Agrosphere Institute, Forschungszentrum Juelich ICG IV, Juelich, 52425 Germany
AU: Kemna, A
EM: a.kemna@fz-juelich.de
AF: Research Center Juelich Agrosphere Institute, Forschungszentrum Juelich ICG IV, Juelich, 52425 Germany
AB: Time-lapse geophysical measurements (also known as monitoring, repeat or multi-frame survey) now play a critical role for monitoring -non-destructively- changes induced by human, as reservoir compaction, or to study natural processes, as flow and transport in porous media. To invert such data sets into time-varying subsurface properties, several strategies are found in different engineering or scientific fields (e.g., in biomedical, process tomography, or geophysical applications). Indeed, for time-lapse surveys, the data sets and the models at each time frame have the particularity to be closely related to their "neighbors", if the process does not induce chaotic or very high variations. Therefore, the information contained in the different frames can be used for constraining the inversion in the others. A first strategy consists in imposing constraints to the model based on prior estimation, a priori spatiotemporal or temporal behavior (arbitrary or based on a law describing the monitored process), restriction of changes in certain areas, or data changes reproducibility. A second strategy aims to invert directly the model changes, where the objective function penalizes those models whose spatial, temporal, or spatiotemporal behavior differs from a prior assumption or from a computed a priori. Clearly, the incorporation of time-lapse a priori information, determined from data sets or assumed, in the inversion process has been proven to improve significantly the resolving capability, mainly by removing artifacts. However, there is a lack of comparison of these methods. In this paper, we focus on Tikhonov-like inversion approaches for electrical tomography imaging to evaluate the capability of the different existing strategies, and to propose new ones. To evaluate the bias inevitably introduced by time-lapse regularization, we quantified the relative contribution of the different approaches to the resolving power of the method. Furthermore, we incorporated different noise levels and types (random and/or systematic) to determine the strategies' ability to cope with real data. Introducing additional regularization terms yields also more regularization parameters to compute. Since this is a difficult and computationally costly task, we propose that it should be proportional to the velocity of the process. To achieve these objectives, we tested the different methods using synthetic models, and experimental data, taking noise and error propagation into account. Our study shows that the choice of the inversion strategy highly depends on the nature and magnitude of noise, whereas the choice of the regularization term strongly influences the resulting image according to the a priori assumption. This study was developed under the scope of the European project ALERT (GOCE-CT-2004-505329).
UR: http://www.fz-juelich.de/icg/icg-iv/index.php?index=47
DE: 1899 General or miscellaneous
DE: 3260 Inverse theory
DE: 7260 Theory
SC: Hydrology [H]
MN: Fall Meeting 2005