Damped Gauss-Newton Optimization Algorithm for Two-Dimensional Magnetotelluric Regularization Inversion

Abstract

The two-dimensional magnetotelluric inverse problem is ill-posed and the inverse results are unstable and non-unique. It means that different geo-electrical model could fit the observed data with the same accuracy. A stable solution of the ill-posed inverse problem can be obtained by utilizing the regularization methods in the objective function. Solving large scale linear equation of inverse problem, the damped Gauss-Newton algorithm was adopted, which can improve local convergence of Gauss-Newton method. Through the synthetic model simulation, the inversion results truly reflected the geo-electrical parameters of the model and accurately showed the depth and size of the abnormal body. On the one hand, inversion of TE mode was more sensitive for the low abnormal body and had poor resolution for the high abnormal body. On the other hand, inversion of TM mode had better resolution for the high abnormal body. So the damped Gauss-Newton algorithm can be used in two-dimensional magnetotelluric data analysis.