Quantitative Risk Assessment of CO2 Sequestration in a commerical-scale EOR Site

Enhanced Oil Recovery with CO2 (CO2-EOR) is perhaps the most feasible option for geologic CO2 sequestration (GCS), if only due to existing infrastructure and economic opportunities of associated oil production. Probably the most significant source of uncertainty of CO2 storage forecasts is heterogeneity of reservoir properties. Quantification of storage forecast uncertainty is critical for accurate assessment of risks associated with GCS in EOR fields. This study employs a response surface methodology (RSM) to quantify uncertainties of CO2 storage associated with oil production in an active CO2-EOR field. Specifically, the Morrow formation, a clastic reservoir within the Farnsworth EOR Unit (FWU) in Texas, was selected as a case study. Four uncertain parameters (i.e., independent variables) are reservoir permeability, anisotropy ratio of permeability, water-alternating-gas (WAG) time ratio, and initial oil saturation. Cumulative oil production and net CO2 injection are the output dependent variables. A 3-D FWU reservoir model, including a representative 5-spot well pattern, was constructed for CO2-oil-water multiphase flow analysis. A total of 25 permutations of 3-D reservoir simulations were executed using Eclipse simulator. After performing stepwise regression analysis, a series of response surface models of the output variables at each step were constructed and verified using appropriate goodness-of-fit measures. The R2 values are larger than 0.9 and NRMSE values are less than 5% between the simulated and predicted oil production and net CO2 injection, suggesting that the response surface (or proxy) models are sufficient for predicting CO2-EOR system behavior for FWU case. Given the range of uncertainties in the independent variables, the cumulative distribution functions (CDFs) of dependent variables were estimated using the proxy models. The predicted cumulative oil production and net CO2 injection at 95th percentile after 5 years are about 3.65 times, and 1.7 times as large as the ones at 5th percentile, respectively. These results suggest that significant uncertainties of output variables are propagated from the parameter uncertainties. While we anticipated this result in general, quantitative confirmation is probably necessary for any and all specific cases (fields or operations).