On the detection of leakage pathways in geological CO2 storage systems using pressure monitoring data: Impact of model parameter uncertainties

In this study, we examine the effect of model parameter uncertainties on the feasibility of detecting unknown leakage pathways from CO2 storage formations via inversion of pressure monitoring data, and discuss the strategies for enhancing detectability and reducing the impact of those uncertainties. We conduct a numerical study of leakage detection, using an idealized storage system consisting of a storage formation and an overlying aquifer separated by a caprock, with an injection well and a leaky well. Our uncertainty quantification analysis shows that (1) the anomalous leakage signals induced by the leaky well can be clearly detected in the overlying aquifer, with minimal impact of model parameter uncertainties, as long as the leaky well permeability is sufficiently large and the caprock permeability is small with the assumed aquifer and caprock thickness; and (2) the pressure monitoring data in the storage formation are not adequate for detecting leakage signals, because the model predictions can be significantly affected by the uncertainties of the model parameters (e.g., permeability and specific storativity of the storage formation and the overlying aquifer). Therefore, we propose an inverse-modeling methodology that combines leakage detection with model recalibration under conditions of model parameter uncertainties. Our results show that the combined leakage detection and model recalibration are most successful when pressure monitoring data from both the storage formation and the overlying aquifer are used, owing to the strong detectability in the overlying aquifer and the strong sensitivity of pressure in the storage formation to model parameters. The proposed methodology also shows that the effect of model uncertainties on leakage detection can be reduced by simultaneously estimating the leakage parameters and the uncertain model parameters, using long-term pressure data under various conditions of permeabilities and locations of the leaky well, and a wide range of uncertainties for the model parameters.