UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Geophysical survey decomposition and efficient 3D inversion of time-domain electromagnetic data Yang, Dikun


Rigorous three-dimensional (3D) forward and inverse modeling of geophysical electromagnetic (EM) data can be time-consuming and may require a large amount of memory on expensive computers. In this thesis, a novel framework, called survey decomposition, is proposed to make the 3D EM modeling more efficient. Recognizing the multi-scale nature of the EM modeling problems, the fundamental idea is to break down an EM survey, which consists of many transmitters, receivers and times/frequencies, into a number of subproblems, each of which is only concerned about data modeled by a localized source, receiver and time/frequency. The modeling is then carried out on the subproblems at different scales, instead of the original problem as a whole. Such a decomposition is able to speed up the numerical modeling, because: (1) A subproblem can have highly efficient discretizations in space and time customized to its localized source, receiver, time/frequency and the specific scale of investigation, for example, it uses a local mesh that is much smaller than the one used in the original global problem; (2) A subproblem is a self-contained EM modeling problem that does not depend on other subproblems, so it is suitable for massive parallelization; (3) Upon decomposition, no modeling is carried out on the global mesh and the amount of computation is proportional to the number of subproblems, so the scalability improves significantly. After decomposition, the large number of subproblems is further reduced by adaptive, random and dynamic subsampling of the data. The adaptive scheme matches the number of samples to the scale of investigation so that only the data necessary for the model reconstruction are selected. The framework of survey decomposition is applied to two types of time-domain EM (TEM) surveys: airborne TEM and ground large loop TEM. Both synthetic and field data are inverted using this new approach. I show that survey decomposition is capable of producing modeling and inversion results similar to those from the conventional methods with greatly reduced time and memory usage. Further speed-up by massive parallelization and generalization to other types of EM surveys is straightforward.

Item Media

Item Citations and Data


Attribution-NonCommercial-NoDerivs 2.5 Canada