Three-dimensional magnetotelluric inversion: data-space method

A three-dimensional magnetotelluric (MT) minimum structure inversion algorithm has been developed based on a data-space variant of the Occam approach. Computational costs associated with construction and inversion of model-space matrices make a model-space Occam approach to 3D MT inversion impractical. These difficulties are overcome with a data-space approach, where matrix dimensions depend on the size of the data set, rather than the number of model parameters. With the transformation to data space it becomes feasible to invert modest 3D MT data sets on a PC. To reduce computational time, a relaxed convergence criterion is used for the iterative forward modeling code used to compute the sensitivity matrix. This allows reduction in computational time by more than 70%, without affecting the inversion results. Numerical experiments with synthetic data show that reasonable fits can be obtained within a small number of iterations, with a few additional iterations required to eliminate unnecessary structure and find the model with minimum norm.