HR: 08:30h
AN: H21G-03    [Abstracts]
TI: Geostatistical Inversion of Conductivity and Dispersivities for Hydraulic Heads and Tracer Data from a Sandbox Experiment
AU: * Nowak, W
EM: wolfgang.nowak@iws.uni-stuttgart.de
AF: Institute for Hydraulic Engineering, University of Stuttgart, Pfaffenwaldring 61, Stuttgart, 70569 Germany
AU: Cirpka, O A
EM: olaf.cirpka@eawag.ch
AF: EAWAG (Swiss Federal Institute of Aquatic Science & Technology), šberlandstr. 133, Dübendorf, 8600 Switzerland
AB: In this study, the authors present a new method of geostatistical inverse modeling for simultaneous identification of spatially varying conductivity log K and dispersion coefficient log D fields. The method is applied to head and tracer data from a technical-scale experiment in a heterogeneously packed sandbox. As data from flow and transport processes are used to calibrate both and transport model parameters, the predictive capabilites of the method include both flow and transport processes. Up to presence, geostatistical inverse modeling has mainly focussed on identifying hydraulic conductivity fields of heterogeneous formations from data related to flow and/or transport processes. Since, in general, the input data are not sufficient to fully capture the heterogeneity of the system, the unresolved heterogeneity must either be simulated randomly in conditional realiations, or is averaged to obtain a conditional mean log K field. The lack of heterogeneity in the latter leads to an underprediction of the dispersion of solutes. For the sake of efficiency, our new method uses a conditional mean log K field, so that, compared to the unknown original formation, the conductivity field lacks small-scale heterogeneity. This lack of heterogeneity is made up for by simultaneously identifying a dispersion coefficient log D, which is included in the geostatistical inverse procedure as spacially distributed unknowns just like conductivity. Interpreted in the context of stochastic linear dispersion theory, this corresponds to an effective dispersion coefficient parameterizing the unresolved or conditional uncertainty of the log K field. Compared to methods that simulate the unresolved variability using large ensembles of conditional realizations, this new method is computationally highly efficient since it computes only the conditional mean field of the parameters. In order to compress the amount of data and focus on the most significant parts of information, only the first and second central temporal moments of the measured breakthrough curves are used. The greatest part of information for identifying the dispersion coefficient stems from the second central temporal moment. In the application featured here, the method performed well. The resulting parameter fields of log K and log D adequately represent the processes of flow and of advective-dispersive transport observed in the sandbox experiment. Although only lower order temporal moments of breakthrough were used, the system was sufficiently characterized to predict higher order moments to a reasonable extent. Further, the identified parameter values compare well to results from stochastic linear dispersion theory.
DE: 1829 Groundwater hydrology
DE: 1831 Groundwater quality
DE: 1869 Stochastic hydrology
DE: 1873 Uncertainty assessment (3275)
DE: 3260 Inverse theory
SC: Hydrology [H]
MN: Fall Meeting 2005