Ensemble-based, Bayesian Inversion of CSEM Data Using Structural Prior Information
S. Tveit, S. Bakr, M. Lien and T. Mannseth
Event name: 76th EAGE Conference and Exhibition 2014
Session: EM Modelling and Inversion
Publication date: 16 June 2014
Info: Extended abstract, PDF ( 1.58Mb )
Price: € 20
The problem of identifying large-scale subsurface structures utilizing information from both controlled source electromagnetic (CSEM) data and seismics is considered. The proposed inversion methodology can use interpreted seismic data as structural prior information in a Bayesian inversion of CSEM data. The Bayesian method applied is the ensemble Kalman filter, where the conductivity model is updated without requiring sensitivity calculations. The ensemble Kalman filter also provides the ability to quantify uncertainty in the conductivity model. To be able to represent complex subsurface conductivity distributions, we utilize the recently proposed hierarchical level-set (LS) parameterization, with reduced representation of the LS function. The novel inversion methodology is applied on three numerical examples: one where there is no reservoir present; another where two reservoirs are present; and a third where the prior model includes a reservoir which is not present in the reference model (‘false positive’). In all examples, the methodology is able to identify the large-scale background structures (strata boundaries and a fault) with reasonably good accuracy. Furthermore, the methodology accurately identifies existing reservoirs, and removes the reservoir in the case containing the ‘false positive’.