Magnetotelluric inversion for anisotropic conductivities in layered media

摘要

Electrical anisotropy in the Earth's crust and upper mantle has recently gained attention as a significant linking factor between electrical models and underlying structural and tectonic patterns. This interest has also motivated new methodological studies into the modelling and inversion for electrically anisotropic structures. We present an algorithm for the inversion of magnetotelluric data over layered anisotropic conductors which is a straightforward extension of the standard Occam 1-D inversion to anisotropic models. Owing to the essential limitation of magnetotellurics to resolve the complete conductivity tensor, we formulate the inversion for azimuthal anisotropy only. We treat the non-linear inverse problem as a multi-criterion minimization of the structure complexity, data misfit and anisotropy. To constrain the structure complexity, we employ the standard roughness penalty as well as non-quadratic penalties of the total variation and gradient support type that produce more focused model sections and thus conform better to the idea about sharp, non-diffuse boundaries of anisotropic structures in the Earth. Application of the anisotropy penalty is crucial for suppressing spurious anisotropy in the inverse models. We use a 2-D extension of the heuristic L-curve method to estimate the quasi-optimal penalty weights. With two non-linear iteration solvers, specifically the reweighted conjugate gradient method and the lagged diffusivity iteration, we can arrive at the minimum of the target functional, for one selected pair of regularization weights, typically after a few tens of iteration steps.