Work Plan
Multi-method inversions of geoelectric and electromagnetic data require optimized and parallelized algorithms for the forward simulation which have to be well adapted to the specific resolution and model regularization of a particular method. The efficiency and suitable parallelization strategies for forward operators depend on the particular electrical or electromagnetic method. Inversion strategies, in contrast, can be developed largely independently of the individual methods. Nevertheless, multi-scale, multi-method inversions depend strongly on the regularization and model resolution of the individual methods. We therefore divide our work packages into tasks that can be addressed separately before being merged in a second step.
Multi-scales are introduced by different spatial resolution capabilities of the methods under consideration. Borehole tomography, for instance, can image the direct vicinity of a well at very high resolution, whereas MT measurements conducted at the surface of the Earth cannot provide additional information at this scale. Conversely, well-logging cannot reveal structures of the subsurface at larger distances from the borehole. When combining both methods, the overlapping resolution domain is small and the multi-scale information content of the data is large. A common model that is consistent for both methods will be resolved in separate regions at dissimilar scales depending on the individual methods. The expression of the transition between resolved model domains and their inherent scales in the final model depends strongly on the regularization.
MT and CSMT are frequency domain methods using virtually identical sensor technology and similar survey geometry. Whereas passive MT sources are usually assumed to be plane waves, in processing CSMT data, the geometry of the source and the near-field characteristics of the excited EM fields have to be considered. CSMT and classical MT have overlapping resolution domains on similar spatial scales. However, their sensitivities with respect to conductivity structures differ and are partly complementary. Due to the excitation of different current systems in the subsurface in MT and CSMT, MT has a high resolution potential for lateral conductivity variations, whereas CSMT has a comparatively better vertical resolution power for horizontal layering and for poor conductors. A combination of both methods by joint inversion should therefore be well-suited to increase the overall resolution beyond that of the individual methods, independently of the applied regularizations (Commer and Newman, 2009).
In subproject I the GFZ Potsdam group concentrates on the above mentioned combination of MT und CSMT. First, a 3D joint inversion of "normal" surface MT and CSMT data shall be developed (WP 1.02 – 1.04). Subsequently, this inversion will be extended by introducing resolution- and scale-dependent regularization strategies (WP 1.05) to facilitate, e.g., an integration of borehole data.
The main focus of the Freiberg working group is the combination of transient electromagnetics (TEM) and the DC resistivity method in subproject II. In WP 2.1 and WP 2.2 the specific sensitivity patterns of the individual methods are combined to enhance the resolution power for a given target area (WP 2.6). Both, DC resistivity and TEM can be applied from the surface or within boreholes. Using sensitivity and resolution analyses (WP 2.4, WP 2.5), optimum transmitter/receiver configurations can be determined to optimize experimental designs. In the mathematical part of WP 2.3, procedures will be developed to increase the efficiency of the geophysical interpretation techniques. Particularly, we will address model reduction in the frequency domain (MRFD), the spectral Lanczos decomposition method including restarts, so called thick restarts, rational Lanczos methods, and multigrid techniques.
The enormous complexity of three-dimensional inversions calls for the use of massively parallel computing architectures. In addition to the parallelization of numerical algorithms (WP 1.03, 1.09), this requires the development of scheduling algorithms to coordinate the execution of computing jobs in a distributed computing environment (WP 1.10). Job scheduling is particularly important when grid architectures are used as a computing resource (Fernandez-Quiruelas et al., 2009).