HR: 08:00h
AN: H41B-01    [Abstracts]
TI: A new Joint Inversion Approach in Conjunction With Cluster Analysis to Improve the Reliability of Hydrogeophysical Models
AU: * Guenther, T
EM: Thomas.Guenther@gga-hannover.de
AF: Leibniz Institute for Applied Geosciences, Hannover, Stilleweg 2, Hannover, 30655, Germany
AU: Ruecker, C
EM: cruecker@uni-leipzig.de
AF: Institute of Geophysics and Geology, University of Leipzig, Talstr. 35, Leipzig, 04103, Germany
AB: Since the inversion of geophysical inversion usually suffers non-uniqueness we try to apply different physical fields to obtain a more reliable image of the subsurface. If the underlying parameters are connected by petrophysical laws or empirical relationships we are able construct a unique inversion scheme that considers all data. However, in many cases such a relationship does not exist. Nevertheless we expect structural similaries and want to allow for it without enforcement. That is basically the idea of our novel joint inversion: The structure of one parameter, i.e., the gradient of its distribution, is used to weight the other. For each boundary a weight is defined that determines one row of the derivative matrix. We use the techniques of robust modelling, the iteratively reweighted least squares method to determine the individual weights. The roughness vector of one parameter is the input for the weight of the other. Thus a large change in one parameter allows for easier change of the other and vice versa at the same position. Both inversion start and yield the first model iteration independently, then the coupling starts. By an example involving dc resistivity and refraction data we prove to yield more significant structures with structural coupling compared to inversion without coupling. Both images can be combined by means of cluster analysis determining a cluster value for each cell. By associating the cluster mean to the corresponding cells we obtain a very simple image of the subsurface. However the clustered model is not able to fit the data appropriately. Hence, we first try to optimize the values (post-iteration). As a second step we use a side-result of the fuzzy analysis, the cluster membership function, for model constraints in the next stage of inversion with the cluster model as reference. Basically the distance from the cluster center defines how much it may vary. Finally we end up in a very simple model combining two parameters that is able to explain the data to a high degree. The membership function serves as a measure for the reliability and helps to appraise resolution.
UR: http:www.resistivity.net/
DE: 1835 Hydrogeophysics
DE: 1894 Instruments and techniques: modeling
DE: 1899 General or miscellaneous
SC: Hydrology [H]
MN: 2007 Joint Assembly