Inversion of magnetotelluric data for 2D structure with sharp resistivity contrasts

We have developed a linearized algorithm to invert noisy 2-D magnetotelluric data for subsurface conductivity structures represented by smooth boundaries defining sharp resistivity contrasts. We solve for both a fixed number of subsurface resistivities and for the boundary locations between adjacent units. The boundary depths are forced to be discrete values defined by the mesh used in the forward modeling code. The algorithm employs a Lagrange multiplier approach in a manner similar to the widely used Occam method. The main difference is that we penalize variations in the boundary depths, rather than in resistivity contrasts between a large number of adjacent blocks. To reduce instabilities resulting from the breakdown of the linear approximation, we allow an option to penalize contrasts in the resistivities of adjacent units. We compare this boundary inversion method to the smooth Occam inversion for two synthetic models, one that includes a conductive wedge between two resistors and another that includes a resistive wedge between two conductors. The two methods give good agreement for the conductive wedge, but the solutions differ for the more poorly resolved resistive wedge, with the boundary inversion method giving a more geologically realistic result. Application of the boundary inversion method to the resistive Gemini subsalt petroleum prospect in the Gulf of Mexico indicates that the shape of this salt feature is accurately imaged by this method, and that the method remains stable when applied to real data.