Gradient Calculations for 3D Inversion of CSEM Data Using a Fast Finite-difference Time-domain Modelling Code

Abstract

Gradient-based 3D inversion of controlled-source electromagnetic (CSEM) data requires the gradient to a data-misfit functional to be computed within the volume where we seek to find a resistivity model that explains the observed data. Since 3D inversion typically requires more than 100 iterations in order to converge to a good model, fast modelling is important to make 3D inversion a feasible tool in advanced interpretation of CSEM data. We have previously presented a fast finite-difference time-domain (FDTD) modeling code which reduces the computational time of a typical modelling scenario by a factor of 40. In this presentation we show how we use this FDTD modeling to efficiently compute the misfit gradient. A benefit of this approach is that the gradient can be computed for several frequencies in one forward modelling thus saving valuable computational resources when inverting data from a large number of receivers. We present results showing that the gradients obtained are well-behaved. We also present results from inversion of synthetic data based on these gradient calculations showing that, for this specific example, a good result was obtained for the inverted resistivity model after relatively few iterations.