Abstract
A Bayesian inversion methodology for identification of large-scale subsurface structures (strata) from controlled source electromagnetic data is developed. The Bayesian inverse problem is solved by sampling from the posterior probability distribution, using the ensemble Kalman filter. Prior knowledge is incorporated in the methodology by utilizing structural prior information from, for example, interpreted seismic data. A model-based, composite parametrization of the electric conductivity distribution is applied to represent the subsurface structures. The model-based representation also allows for estimation of variation of electric conductivity within each stratum. To enhance stability and reduce model nonlinearity, a reduced representation of structure boundaries and conductivity variation within each stratum is applied. Numerical experiments on various test cases show that the methodology is able to identify fairly complex subsurface conductivity distributions reasonably well, with and without a hydrocarbon reservoir present. These experiments included strata with weak conductivity contrast and the application of various prior probability distributions. Furthermore, the methodology shows the ability to (almost completely) remove a reservoir present in the prior model that is not present in the true model (‘false positive’).