Exploring Noise Models in Approximate Bayesian Inversion for Facies

This paper demonstrates that a correlated noise model in a direct probabilistic inversion for facies significantly improves the resolution and detail as compared to an uncorrelated white noise model. The conservative LPF noise model shows potential to increase resolution for the inversion of real seismic data. However, real seismic noise is much more complex than a correlated Gaussian noise model and can be augmented by adding a weak white noise term (assuming that the correlated and the uncorrelated noise is independent).