Скачать или смотреть EMinar 3.7: Xushan Lu - Surface geometry inversion of geophysical electromagnetic data

Three-dimensional minimum-structure, Occam-style EM inversions are well-established and have been successfully applied to a wide range of exploration problems. Typically, constructed models have fuzzy, smeared-out characteristics, and it is difficult to find distinct boundaries between different geological units. Certain geological features, such as graphitic faults, have sharp, distinct interfaces with surrounding host. The recovered models from minimum-structure inversions provide limited information on these interfaces, making it difficult to target drill holes.

We present a new surface geometry inversion (SGI) algorithm which directly inverts for the interfaces between different geological units. Our SGI only focuses on localized anomalies and parameterizes them with coordinates of nodes used to form wireframe surfaces representing their shapes. Unstructured tetrahedral grids are used to discretize the anomaly and the background to calculate forward responses.

Our SGI can either treat the conductivities of various units as constants or as parameters to be inverted. An estimated background conductivity model (e.g., one obtained from 3D voxel inversions) is required and kept fixed during the inversion. When only inverting for geometrical parameters, the objective function only contains one sum-of-squares measure of data misfit. It can also include smoothness measures to obtain smooth anomalous conductivity models if conductivities are also inverted. A genetic algorithm (GA) is used to minimize the objective function. We parallelized our algorithm to calculate the forward response of each model in the GA population simultaneously.

We present the applications of the SGI algorithm to the synthetic and real-data inversions of marine CSEM, TEM, and DCR problems. We show that our SGI has the potential to effectively recover locations of thin, distinct, localized geological structures in complex realistic models.