Towards Quantitative Approach to Evaluating Greenhouse Gas Leakage from CO2 Enhanced Oil Recovery Fields

Abstract

AbstractGreenhouse gas (mainly CO2 and CH4) leakage from abandoned wells in CO2 enhanced oil recovery (EOR) sites is a long-standing environmental concern and health hazard. Although multiple CO2 capture, utilization, and storage programs (e.g., CarbonSAFE, Regional Carbon Storage Partnerships) have been developed in the U.S. to reach the net-zero emission target by 2050, one cannot neglect the significant amount of CO2 and CH4 leakage from abandoned wells. This study will investigate the potential of CO2 and oil components (e.g., CH4) leakages from the abandoned wellbore and develop the first-ever quantitative approach to evaluating CO2 and oil component leakage from a CO2-EOR field.We conducted wellbore leakage analysis for the CO2-EOR field. A numerical model which has aquifer, caprock, and reservoir components was developed. We used C1, C4, and C10 to represent the light, intermediate and heavy components of crude oil, respectively. All the required simulations were performed using Eclipse 300. We quantified the CO2/oil components leakage through the wellbore to the aquifer by varying abandoned reservoir pressure, effective wellbore permeability, caprock thickness, residual oil saturation, etc. Then, Monte Carlo simulations were performed to investigate the impact of uncertain characteristics (including reservoir depth, net-to-gross ratio, reservoir permeability, residual oil saturation, and mole fractions of oil components) on CO2 and oil components (e.g., CH4) leakages. After that, we developed a set of reduced-order models (ROMs) to predict CO2/oil components leakages through abandoned wellbore using a supervised machine learning technique.We observed that in addition to a large amount of CO2 leakage, a significant amount of light and intermediate oil components (i.e., C1 and C4) leaked through the wellbore. In contrast, a minimal amount of heavy oil component (C10) leaked. Oil components’ leakage is mainly through the gas phase rather than the liquid phase (relevant figures for mole fraction distributions of CO2/oil components in gas and oil phases are not shown). We observed that CO2 leakage is positively correlated to reservoir depth, wellbore pressure, and permeability through sensitivity analysis. In contrast, it is negatively related to net-to-gross ratio, residual oil saturation, and mole fraction of CH4. On the other hand, oil component leakages (C1 and C4) are positively correlated to all uncertain parameters, except the net-to-gross ratio. Lastly, the ROMs generated using the machine learning technique have a relatively high fidelity.