| Description |
Carbon sequestration in oil fields is achieved as an outcome of injection of CO2 for enhanced oil recovery (CO2-EOR), albeit not necessarily by design. In addition to CO2 that is trapped during active oil recovery, CO2 may be sequestered in a subsequent intentional storage period via post-EOR CO2 injection. Once injected, CO2 may be trapped by dissolution in the oil or aqueous phase, or as a separate supercritical phase. Forecasting the nature of trapping and ultimate CO2 distribution in the subsurface is subject to many sources of uncertainty. The primary purpose of this dissertation research is to develop and apply a systematic uncertainty quantification approach to analyze CO2 trapping mechanisms in an active CO2-EOR field, the SACROC unit in western Texas, as well as forecasting of post-EOR injection (depleted oil field CO2 storage). This doctoral work is divided into three parts. The first part presents a probabilistic analysis of the primary CO2 trapping mechanisms, including oil solubility trapping, hydrostratigraphic trapping, and aqueous solubility trapping. Heterogeneity of model properties, particularly porosity and permeability, is considered as the most significant source of uncertainty in this part of the dissertation research. Such parameter uncertainty was analyzed by developing and calibrating reduced order models (ROMs) combined with Monte Carlo simulations. A polynomial chaos expansion (PCE) technique was used to develop the ROMs. The second part of the dissertation focuses on the impact of two important multiphase flow components on forecasting CO2 storage with the SACROC model. The two components are three-phase relative permeability and hysteresis effects, both of which are difficult to measure and as such are often represented by empirical models. Four commonly used three-phase relative permeability models and three hysteresis models were applied to the SACROC model, resulting in a total of 12 different permutations (or alternative models). The third part of the dissertation presents an integrated uncertainty quantification approach that combines both parameter uncertainty (analyzed in the first part) and model uncertainty (analyzed in the second part). Specifically, parameter uncertainty was first quantified using ROMs derived from PCE; model uncertainty was then integrated by using a Bayesian model averaging method. |
