The Effects of Subsurface Heterogeneity on Detectability of CO2 Leakage to Shallow Groundwater Aquifers

Numerical simulations of CO₂ storage reservoir leakage can be used to assess risks of shallow groundwater aquifer contamination during monitoring network design. Improperly plugged and abandoned wells are well known to represent one of the greatest risks to successful containment at geologic carbon sequestration sites. Casing and cement seal failure of wells penetrating the confining layer may create fast-flow pathways for CO₂ and brine migration from the storage reservoir into the shallow subsurface. To protect drinking water aquifers from possible leaks, injection permits require identification of artificial penetrations and evaluation that wells are adequately plugged and abandoned. However, assumptions made during well evaluation may overlook the likelihood of well failure leading to a leak into an aquifer. We present a monitoring approach that provides quick and accurate detection in the event of a leak to an aquifer. Sand and shale facies are classified to simulate aquifer heterogeneity using representative borehole geophysical data from Texas, U.S.A. Gulf Coast Aquifer System wells. Numerical models simulate pressure perturbations in response to a leak to an aquifer overlying a storage reservoir. Candidate monitoring well locations for a possible leak of randomly selected location are chosen from a suite of possible wells based on the detectability of CO₂ leakage from the groundwater model. We first show that the locations and magnitudes of leakage can be identified for homogeneous aquifers by using an inversion procedure and pressure observations. We then consider the effects of conceptual model uncertainty, pressure measurement error, and background noise on detectability of leaky wells. While substantial previous work quantified pressure perturbations caused by leaky wells using analytical solutions or simple numerical model configurations, the effects of formation heterogeneity on pressure perturbation and other uncertain factors are not well examined. Therefore, the results of this work will provide valuable insight to the design of CO₂ leakage detection monitoring networks in the presence of uncertainty.