(389e) High-Resolution Numerical Simulation of CO2 Migration in Saline Aquifers

We study the migration of CO2 plumes in deep saline aquifers during the post-injection period using high-resolution numerical simulation. The interactions between buoyancy driven flow, dissolution of CO2 into the brine, capillary mixing, and diffusion/dispersion effects lead to complex behaviors in space and time. Here, a two-component, two-phase model is used, and the phase behavior is represented using an equation-of-state. In order to resolve the nonlinear coupling, the Fully Implicit Method (FIM) is used for time discretization. We find that the numerical solutions are quite sensitive to both grid resolution and time truncation errors. These challenges are compounded by the presence of miscible convective instability, which can dominate the overall evolution of the plume.
A lock-exchange configuration is set up, in which the total plume mass is emplaced near the bottom (inlet) of the aquifer. Following an initial equilibration period, the pressure gradient and the streamlines become nearly constant. In order to improve the overall computational efficiency of the simulations, a moving-outer-boundary (MOB) strategy was implemented as follows. The domain is split into two sub-domains: one that stretches from the inlet to just beyond the leading edge of the CO2 plume, followed by a region that extends all the way to the actual outer (updip) boundary of the aquifer. In the second sub-domain, the pressure and saturation distributions remain unchanged. Information at the end of a time step about the pressure, velocity, and saturation distributions is used to delineate the boundary between the two sub-domains. This is facilitated by the observation – made by many investigators including our group - that the leading-edge speed is nearly constant during the long post-injection period. Every time when the boundary between the sub-domains need to by moved, a few small time steps are performed on the global domain. This ensures that the evolution of the capillary transition zone and other mixing phenomenon is fully resolved before operating only on the “active” sub-domain.
This method shares similar idea with adaptive implicit method (AIM) when only the part of the domain where flow occurs is treated fully-implicitly. However, AIM approach has significant stability issues for this problem and can improve the overall performance by only a small margin. All simulations have been performed with the Automatic-Differentiation General Purpose Research Simulator (ADGPRS). The results demonstrate that compared with the standard approach of solving a global problem all the time, the MOB method improves the total computational time by 40-50% with 0.1% error (measured in terms of the updip speed). Simulation results are then shown for extremely large aquifer models using our MOB approach.