Exploring Noise Models in Approximate Bayesian Inversion for Facies

An inversion for geological parameters such as spatial distribution of facies with uncertainty is one of the fundamental goals of seismic inversion/interpretation, since facies mapping is an important step for well planning and field development. The residual noise in seismic data consists primarily of correlated noise, which tends to have the same frequency spectrum as the signal and therefore hard to distinguish from the true signal. The results from an approximate Bayesian inversion directly from AVO data to facies demonstrates that introducing a correlated noise model can significantly improve the resolution and lower the misclassification.