A Methodology for Risk Assessment of CO2 Sequestration Based On Surrogate Models of Detailed Simulations

Large scale implementation of CO2 geologic sequestration requires detailed uncertainty analysis and risk assessment for leakage potential. Probabilistic risk assessment based on Monte Carlo simulation is one option for quantifying the risk. It involves repeated random realizations of a simulator to generate probability distributions of interesting variables such as leakage amount. However, in the context of modeling CO2 plume migration underground after injection, numerical solution of non-linear equation systems for heat and mass balances is too time-intensive for Monte Carlo simulation.  To cope with this computational expense, investigators are often forced to make simplifying assumptions and resort to approximate models.  To address this challenge, instead of solving a large system of non-linear equations, we first develop a surrogate model for the multi-component multi-phase flow processes involved in the injection of CO2 into deep saline aquifers.  In particular, we use TOUGH2 in conjunction with adaptive sampling and cross validation to obtain a surrogate algebraic model.  Then, probabilistic risk assessment for CO2 leakage potential is performed based on Monte Carlo simulation using the surrogate model.  We demonstrate that our approach not only reduces significantly the computational requirements of risk assessment but also provides accurate results.  The accuracy of the model was validated by comparisons to TOUGH2 itself.