(756a) Impact of Model Complexity for Carbon Dioxide Migration in Structured Heterogeneous Domains

Carbon capture and storage (CCS) is seen as a near-term option to reduce anthropogenic carbon emissions from fossil fuel-based electricity generation and other stationary sources. In CCS, carbon dioxide (CO₂) is captured from a power plant’s flue gas and stored in the subsurface. While many types of storage reservoirs (e.g., depleted oil and gas reservoirs, unmineable coal seams) have been discussed, deep saline aquifers are the most promising, due to their large available storage volume and wide-spread distribution throughout the world. In order for a CCS operation to be successful, the injected CO₂ must remain in the subsurface over thousands of years. Even though CO₂ is injected in its denser supercritical state, the CO₂ is less dense than the resident brine; usually by about 500 kg/m³. This leads to a strong upward buoyant drive for the stored CO₂. Therefore, storage aquifers must be overlain by low-permeability formations (termed ‘caprocks’) that do not allow CO₂ to migrate vertically out of the storage aquifer. However, even for aquifers with competent caprocks, vertical CO₂ migration may occur along penetrations through the caprocks. These penetrations can be both artificial (e.g., abandoned wells) and natural (e.g., faults). As CO₂ is being injected it displaces the resident brine. While brine migration within the storage aquifer is typically not a problem, vertical brine migration may degrade overlying drinking water resources through the transport of high salt concentrations and other contaminants.

In order to determine if the injected CO₂ will remain in the subsurface, and that the displaced brine does not negatively impact drinking water sources, the migration of both CO₂ and brine needs to be predicted. The migration patterns tend to be complex due to the multi-phase flow behavior of the CO₂-brine system and the complex (and often uncertain) geometry of the storage and overlying aquifers. Mathematical models are useful tools to predict how CO₂ and brine migrate in the subsurface. Mathematical models applied to CCS systems range from fully-coupled three-dimensional multi-phase flow simulators to single-phase analytical solutions. The choice of model usually depends on the scale of the problem, the type of question being asked, and available computational resources.

The impact of model complexity on simulated migration of CO₂ is the focus of this study. Three modeling approaches are compared: fully-coupled three-dimensional two-phase flow, two-dimensional vertically-integrated two-phase flow, and macroscopic invasion percolation. The fully-coupled two-phase flow model solves the multi-phase flow governing equations in all three spatial dimensions and can thus take into account spatial variability in permeability in the horizontal and vertical directions. For the vertically-integrated approach the multi-phase governing equations are integrated in the vertical direction, leading to a two-dimensional system (horizontal) where the vertical profiles of pressure and saturation need to reconstructed. As this is a two-dimensional model, the vertical variability is not represented directly, but is taken into account as part of the two-dimensional permeability field. Lastly, the macroscopic invasion percolation model discretizes the domain into cells and uses threshold conditions to determine the invasion of CO₂ into brine filled cells. Just as for the fully-coupled model, the three-dimensional variability of aquifer parameters is directly represented; however, the solutions are based on a series of steady state calculations and no flow dynamics are represented.

In this study a hypothetical field site is used, with parameters based on the Ketzin CO₂ injection site in Germany. The spatial variability in permeability and porosity is chosen to represent a fluvial-deposition environment with stream channels embedded in a flood plain background. The domain is constructed by adding sinusoidal streams with varying characteristics (width, depth, wavelength, and amplitude) until approximately one third of the domain is filled by stream channels. The ranges of the stream characteristics are based on values found at the Ketzin site. CO₂ is injected at the center of the domain at a rate of approximately 0.1 million tonnes per year, and injection only occurs into stream channel grid cells. While the fully-coupled three-dimensional and macroscopic invasion percolation models directly take the vertical distribution of aquifers parameters into account, permeability and porosity are average over the thickness of the domain for the vertically-integrated model.

The results from the different models are compared based on the CO₂ plume shape and based on CO₂ arrival time at an offset monitoring well. The plume shape is chosen as a metric, because delineating the expected plume outline is an important piece of any CO₂ injection design. The arrival time was chosen as a second metric, because experience at several injection sites, including Ketzin, has shown that preferential flow paths have to be considered when estimating arrival times at offset wells. The results of this study will suggest to both practitioners and academics the appropriate level of model complexity for site-scale models of CO₂ migration in structured heterogeneous domains.