Inversion of controlled‐source tensor magnetotelluric data for a layered earth with azimuthal anisotropy

Publisher Website
Google Scholar
Abstract
Fractures in the upper part of the crust are pervasive in crystalline areas. They often are aligned and nearly vertical. Magnetotelluric studies carried out in different parts of the world often indicate that the upper brittle part of the crust is electrically anisotropic. One model is that deep fluid‐filled fractures, when observed from the surface, mimic an anisotropic medium. The transfer functions of the controlled‐source tensor magnetotelluric (CSTMT) method contain information from current systems running both parallel and perpendicular to the principal horizontal axes of anisotropy, so that the CSTMT method is capable of detecting azimuthal anisotropy. For the CSTMT method, electric dipoles provide the source field, and hence fields are localized. Thus, the distortion generated by a major conductive anomaly lying outside the induction volume defined by the transmitter and receiver positions will be small. Therefore, one‐dimensional models often may be valid. We have developed a nonlinear least‐squares inversion approach to invert CSTMT data for azimuthal anisotropy in an one‐dimensional layered earth. Near‐surface distortion effects on the impedance tensor are parameterized as a real distortion matrix. The elements of the distortion matrix are incorporated into the model parameters. By using tipper functions which are less distorted by near‐surface structures, near‐surface distortion effects can be removed adequately. One striking feature of our inversion algorithm is that the partial derivatives of the response functions, with respect to model parameters, are given in analytical forms, which results in an efficient computation of the Jacobian matrix. Theoretical studies show good convergence and good resolution of the model parameters. We have applied the inversion scheme to a set of controlled‐source data from the Siljan impact structure in Sweden. The derived models give much better data fits than a corresponding isotropic model.