3-D INVERSION OF FREQUENCY-DOMAIN CSEM DATA BASED ON GAUSS-NEWTON OPTIMIZATION

Quantitative interpretation of large-scale controlled-source electromagnetic (CSEM) data in frequency domain requires efficient and stable 3D forward modeling and inversion codes. In this work, we present an efficient approach to 3D inversion of CSEM data, which is based on Gauss-Newton (GN) optimization in combination with a direct solver for the forward modeling. In order to avoid computing and storing sensitivity matrix explicitly, a preconditioned conjugate gradient solver (PCG) is used to solve the system of the normal equations resulted from linearization at each GN iteration. This scheme only requires matrix-vector products of Jocabian and its transpose with vectors, which are equivalent to one forward and one adjoint problem. Therefore the matrix factorization obtained when solving forward problem can be used in subsequent PCG process, which dramatically speeds up PCG iterations and reduces overall computational cost. Numerical experiments on synthetic data from land and marine CSEM surveying configurations show that our inversion scheme exhibits excellent convergence rate and only ten-odd to tens of iterations are needed to reach desired data misfit, demonstrating its efficiency and stability.