TL;DR
This study quantifies the uncertainties associated with CO2 plume expansion in shallow aquifers following wellbore leakage, a critical risk factor for geological carbon sequestration. By employing advanced uncertainty quantification (UQ) techniques, the research identifies key parameters that dominate plume behavior, providing crucial insights for risk assessment and site management.
What Problem Does This Paper Address?
Geological carbon sequestration (GCS) is a promising strategy for mitigating climate change by storing CO2 deep underground. However, the long-term security of these storage sites is paramount, and potential leakage pathways, particularly through abandoned wellbores, pose a significant risk. If CO2 leaks into shallow aquifers, it can alter groundwater chemistry, potentially impacting drinking water quality and the environment. Accurately predicting the extent and behavior of such CO2 plumes is challenging due to numerous uncertainties in geological, hydrological, and geochemical parameters. This paper addresses the critical need to understand and quantify these uncertainties to assess the risks associated with wellbore leakage into aquifers, ensuring the safety and public acceptance of GCS.
Methods
The researchers employed a comprehensive Uncertainty Quantification (UQ) framework to tackle this problem. Key methodological components include:
- Numerical Simulation: The STOMP-CO2-R simulator was used to model multiphase flow of CO2, water, and salt, coupled with geochemical reactions. This allowed for the simulation of CO2 plume migration and its impact on groundwater.
- Efficient Sampling: To manage the computational cost of running numerous simulations, a combination of quasi-Monte Carlo (QMC) and adaptive sampling (specifically, the Adaptive Response Modeling - ARM process) was utilized. QMC provides a more uniform exploration of the parameter space than traditional Monte Carlo methods, while ARM strategically identifies new sample points to refine the model and reduce the number of required simulations.
- Response Surface Modeling: Multivariate Adaptive Regression Splines (MARS) were used to build response surfaces that approximate the relationship between input parameters and output variables. This allows for a more efficient understanding of input-output relationships and the development of reduced-order models (ROMs).
- Statistical Analysis: Generalized Linear Model (GLM) analyses and statistical tests were performed to rank the significance of input parameters and identify their interactions. This included stepwise backward removal based on the Akaike Information Criterion (AIC) to develop parsimonious ROMs.
Datasets
The study utilized a well-characterized example: the carbonate Edwards Aquifer in south-central Texas.
- Geological and Hydrological Data: Information on aquifer properties such as porosity, hydraulic conductivity, and geological formations was sourced from existing U.S. Geological Survey (USGS) reports and other relevant literature (e.g., Musgrove et al., 2010; Lindgren, 2007; Hutchison and Hill, 2011).
- Groundwater Chemistry Data: Background groundwater chemistry data for the San Antonio area was also obtained from USGS reports (e.g., Lindgren et al., 2005; Musgrove et al., 2010).
- Geochemical Parameters: Kinetic and equilibrium reaction parameters for calcite and dolomite were taken from established literature (Palandri and Kharaka, 2004).
No new benchmark datasets were introduced; rather, existing data and literature values were used to parameterize the model.
Key Insights and Ideas
- Dominant Parameters Identified: The study successfully pinpointed the most influential parameters governing CO2 plume expansion. These are:
- Distance between leaky and injection wells (p1): This parameter primarily affects the timing of leakage, influencing the lag before CO2 reaches the aquifer.
- Maximum leakage rate (p2): This directly controls the volume of CO2 entering the aquifer, thus impacting plume size and atmospheric flux.
- Porosity (p3): A fundamental property influencing the storage capacity and movement of fluids within the aquifer.
- Hydraulic conductivity (p4): This dictates the rate of groundwater flow, which in turn influences how quickly the CO2 plume disperses.
- Time-Dependent Influence: The impact of some parameters, like hydraulic conductivity, increases over time as the plume has more opportunity to disperse.
- Geochemistry's Role: While geochemical parameters were considered, the study found that calcite equilibrium and dissolution played a more significant role than dolomite parameters in controlling pH changes. This suggests that in this specific aquifer context, the dominant geochemical processes are relatively straightforward.
- Adaptive Sampling Efficiency: The research demonstrated that adaptive sampling, when combined with QMC, can significantly reduce the number of required simulations while still producing reliable response surfaces and uncertainty estimates, a crucial advantage for computationally intensive UQ studies.
Key Results
- Significant Parameters: The analysis identified distance to the leaky well, maximum leakage rate, porosity, and hydraulic conductivity as the four most significant parameters controlling CO2 plume expansion.
- Parameter Interactions: The study revealed significant interactive effects between these dominant parameters, highlighting the complexity of plume behavior. For instance, the maximum leakage rate, porosity, and hydraulic conductivity collectively contributed over 85% of the output variability for pH < 6.5 plume volume.
- Reduced-Order Models (ROMs): The developed ROMs, based on the identified significant parameters, can predict CO2 plume behavior (volume, width, height, flux) without the need for full numerical simulations, facilitating rapid risk assessments.
- Output Distributions: Probability density functions for output variables like pH plume volume and CO2 flux were generated, showing that plume volume distributions tend towards Gaussian at later times, while CO2 flux distributions are more log-normal.
Limitations and Future Work
- Site Specificity: The developed ROMs are specific to the Edwards Aquifer and the considered parameters. The significance of parameters might change for different geological settings or if other leakage pathways (e.g., faults) or processes are included.
- Computational Demand: Despite efficient sampling methods, UQ studies can still be computationally intensive, especially for complex models or when a large number of parameters are considered.
- Future Work: The authors suggest that the UQ framework and developed ROMs can be extended to other sites and scenarios. Further research could incorporate a broader range of parameters, including more complex geological heterogeneities and dynamic caprock integrity, to refine risk assessments for GCS. The ability to predict CO2 plume behavior without extensive simulations is a valuable tool for informing regulatory decisions and site management strategies.
Summary updated