Data Space Occam?s Inversion: Three-Dimensional Magnetotelluric Case

Abstract

A major challenge in 3-D Magnetotelluric (MT) data inversion is the considerable computational resources required. Most efforts at 3-D MT inversion have been formulated directly in the model space, so that the size of all computations depends strongly on the value of M, the number of model space parameters. For geologically realistic 3-D models, M can be quite large, and in particular is typically much larger than the total number of MT impedance data N. By transforming the inversion into the data space, the computational demands can be significantly decreased. In this paper, we present an Occam style 3-D inversion algorithm based on a data space formulation. Inversion results for synthetic data display good convergence. The data space method can also be used with other types of inversion algorithm such as data sub-space techniques. These refinements will allow more efficient and robust 3-D inversion algorithms to be developed.