3D CSEM inversion using a source-type finite-element approach

Abstract

We propose a novel inversion algorithm for 3D controlled-source electromagnetic data which transforms the ill-posed nonlinear inverse problem into two coupled linear inverse problems. Similar to the so-called contrast source inversion (CSI) method, two sequences of variables, which are the contrast sources and the contrast itself, are iteratively reconstructed by alternately updating them. However, unlike the CSI method, the background reference medium as well as the associated Green functions are updated in our approach after each inversion iteration at the expense of performing several full forward simulations. The increasing availability of high-performance computers help us to address this high computational cost problem via the use of numerical techniques that are theoretically favorable, yet have been considered to be computationally demanding. Therefore, edge finite elements are used for the numerical treatment of our problem, and we also employ the highly tuned matrix inversion routines. Numerical experiments for a synthetic dataset from a simple model demonstrated that the proposed algorithm could provide reasonable inversion results. Our technique is comparable to the conventional local minimization methods, while it is very different from them in handling the nonlinearity of the inverse problem. Therefore our suggested approach can be an interesting and viable alternative in cases where the local optimization methods suffer from errors caused by linearization.