Smart stitching: adding lateral priors to ensemble inversions as a post-processing step

Abstract

Summary The last decade has seen extensive development of Bayesian geophysical inversion methods which produce ensembles of models as outputs. Many of these are limited to producing 1D or very simple and narrow models. It is well established that tying such narrow inversions together using lateral priors can significantly improve inversion results. Such laterally constrained inversion can, however, be complicated to code and add computational overhead. For this reason, available Bayesian geophysical inversion codes often do not include lateral priors as an option. I introduce a simple and easy to use method that allows lateral priors to be added to Bayesian ensemble inversion results as a post-processing step. This method has the potential to extend the use of many existing inversion codes and results. It can significantly reduce computational costs when practitioners want to experiment with different lateral priors. The method is demonstrated using synthetic magnetotelluric data and VTEM data from Cloncurry in Queensland.