Developing a Strategy for Parameter Estimation from CO2 Plume Migration During Geologic Carbon Sequestration in a Fluvial Depositional Setting

Abstract

Estimation of the CO2 plume from monitoring data is needed in order to assure safe carbon sequestration in geological formations. Synthetic field measurements and numerical simulations are used to estimate the plume position and obtain a better understanding of the characteristics of geological formations that govern the CO2 flow. The challenge is to be able to give an accurate prognosis of the plume location with relatively few monitoring observations while dealing with uncertainties and model error. We use the TOUGH2 program, which is a numerical simulator for multi-phase fluid and heat flow in porous and fractured media, along with the ECO2N module, specific for CO2 flow in brine. This model describes the coupling of flow and transport processes in heterogeneous geologic systems. The optimization program 'Stochastic RBF' is used to calibrate the model parameters. Stochastic RBF has proven to be computationally efficient for environmental models that are computationally expensive. It is a derivative-free method, which makes it easier to use in conjunction with a complex nonlinear simulation model. We use three-dimensional saline aquifers with different geological characteristics for the application. We show that estimating shale permeability is critical to determine the plume shape and position, while other facies higher permeabilities can be estimated with less accuracy with little effect on the estimate of the location of the CO2 plume. We also investigate how parameter lumping affects the calibration and the amount of measurements needed in order to accurately estimate plume position. In many cases, it has been found that pressure measurements suffice, while other types of measurements are needed in cases with more parameters to estimate. The plume position can be determined with a correlation coefficient equal to 87 percent with our method with a minimal amount of measurements and number of simulations. Using only pressure observations and no gas saturation samples, a slightly smaller correlation coefficient was obtained.