Bayesian Inversion of 4D Seismic Data with a Machine Learning Prior: Application to the Catcher Fields

We present a workflow that integrates machine learning and Bayesian inversion methods to estimate reservoir pressure and saturation changes from 4D seismic data. The method uses a previously computed deep neural network inversion result as prior information for a stochastic Bayesian inversion. It is applied to data from the Catcher fields. We show that the Bayesian inversion benefits from the use of machine learning prior information, leading to improvements on the match to the observed 4D seismic signal as well as the total injected water volumes. In addition, stochastic sampling of the Bayesian posterior distribution also produces uncertainty estimations which are valuable in such non-unique and uncertain inversion problems. This allows us to extract multiple equiprobable realizations and define conditional bounds for the pressure and saturation changes, which can be useful for reservoir management and 4D seismic data assimilation into reservoir simulation models.