CMG Research: Multiscale data integration using facies based hierarchical Bayesian models

Texas A&M Research Foundation, College Station TX

Investigators

Abstract

This project focuses on uncertainty quantification for integrated geologic facies models. In many geologic environments, the distribution of subsurface properties is primarily controlled by the location and distribution of distinct geologic facies with sharp contrasts in properties across facies boundaries. Under such conditions, the orientation of the channels and channel geometry determine the flow behavior in the subsurface rather than the detailed variation in properties within the channels. Traditional geostatistical techniques for subsurface characterization have typically relied on variograms that are unable to reproduce the channel geometry and the facies architecture. Recently geostatistical models based on multiplepoint statistics have been proposed for reproduction of complex channel architecture. These methods rely on training images that can be difficult to obtain. In this project coherent Bayesian hierarchical models will be developed which will preserve the facies architecture and populate the petrophysically properties within the facies in a geologically consistent manner by incorporating available static and dynamic information. To maintain the contrast in facies properties, facies boundaries will be represented by level sets which represent variety of facies topology including splitting and merging of facies boundaries. The method relies on a Bayesian hierarchical approach to perturb the facies boundaries and properties to match the dynamic flow and transport data and multiphase production history at the wells. A novel aspect of the approach is the choice of a Langevian-type proposal perturbation of facies boundaries combined with multiscale simulations that allows us to implement efficient MCMC methods with higher acceptance rate without sacrificing the convergence characteristics. The facies based hierarchical formalism lends itself readily to efficient multiscale flow simulation with adaptivity that can provide significant speed up in the flow and transport calculations. The hierarchical approach will naturally integrate data from different scales and allow to condition on local hard and soft data. Proper exploitation of Bayesian simulation based algorithm will enable us to perform posterior inference to quantify uncertainty based on this model. The basic idea, novel models and algorithms developed by the project will significantly advance the current state-of-the-art in subsurface characterization by incorporating qualitative geological information into quantitative spatial modeling of properties. These, in turn, will improve the ability to model, scale-up and design problems related to environmental remediation, contaminant transport and CO2 sequestration in hydrocarbon reservoirs/aquifers. The focus of the application will be on CO2 sequestration in depleted hydrocarbon reservoirs. The sequestration of CO2 into geologic formations is a promising solution for reducing environmental hazards created by the release of green house gases in to the earth's atmosphere. In particular, existing and depleted oil and gas reservoirs are attractive candidates for CO2 sequestration for two principal reasons. First, the economic benefits associated with enhanced oil recovery through CO2 injection are commercially proven and widely practiced by the industry. Second, oil and gas reservoirs are likely to provide abundant data sources for subsurface characterization, design and performance assessment of any potential CO2 sequestration project.