Stochastic seismic inversion using greedy annealed importance sampling

Abstract

A global optimization method called very fast simulated annealing (VFSA) inversion has been applied to seismic inversion. Here we address some of the limitations of VFSA by developing a new stochastic inference method, named greedy annealed importance sampling (GAIS). GAIS combines VFSA and greedy importance sampling (GIS), which uses a greedy search in the important regions located by VFSA, in order to attain fast convergence and provide unbiased estimation. We demonstrate the performance of GAIS with application to seismic inversion of field post- and pre-stack datasets. The results indicate that GAIS can improve lateral continuity of the inverted impedance profiles and provide better estimation of uncertainties than using VFSA alone. Thus this new hybrid method combining global and local optimization methods can be applied in seismic reservoir characterization and reservoir monitoring for accurate estimation of reservoir models and their uncertainties.