Three-dimensional magnetotelluric inversion using a heterogeneous smoothness-constrained least-squares method

This paper presents a fast algorithm for electromagnetic data inversion to three-dimensional (3D) resistivity models. The algorithm is distinctive for the level of accuracy it attains while bypassing the sensitivity matrix update. A common sensitivity matrix for homogeneous half-space is used in all iterations. Instead of updating the sensitivity matrix, the smoothness filter coefficients at each model element are updated, based on the spatial variations in resistivity in the model derived from the latest iteration. This substitution is expected not only to reduce the computation time required for large-scale inversions, such as those for 3D surveys, but also to allow the resolution of sharp boundaries in resistivity structures. Our algorithm was applied to 3D magnetotelluric inversion in order to confirm its effectiveness. Using synthetic examples under several conditions, we demonstrated that the method can reduce the number of forward calculations required to reduce data misfits to noise level, and that the method is robust for constructing target models even with sharp boundaries without generating fatally false resistivity structures or boundaries under noisy conditions.