Convective dissolution of carbon dioxide in two- and three-dimensional porous media: The impact of hydrodynamic dispersion

Abstract

Subsurface storage of CO2 is widely regarded as the most promising measure to limit global warming of the Earth's atmosphere. Convective dissolution is the process by which CO2 injected in deep geological formations dissolves into the aqueous phase, which allows storing it perennially by gravity. The process can be modeled by buoyancy-coupled Darcy flow and solute transport. The transport equation should include a diffusive term accounting for hydrodynamics (or, mechanical) dispersion, with an effective diffusion coefficient that is proportional to the local interstitial velocity. A few two-dimensional (2D) numerical studies, and three-dimensional (3D) experimental investigations, have investigated the impact of hydrodynamic dispersion on convection dynamics, with contradictory conclusions drawn from the two approaches. Here, we investigate systematically the impact of the strength S of hydrodynamic dispersion (relative to molecular diffusion), and of the anisotropy α of its tensor, on convective dissolution in 2D and 3D geometries. We use a new numerical model and analyze the following quantities: the solute fingers' number density (FND), penetration depth and maximum velocity; the onset time of convection; the dissolution flux in the quasi-constant flux regime; the mean concentration of the dissolved CO2; and the scalar dissipation rate. We observe that for a given Rayleigh number (Ra) the efficiency of convective dissolution over long times is mostly controlled by the onset time of convection. For porous media with α = 0.1, commonly found in the subsurface, the onset time is found to increase as a function of dispersion strength, in agreement with previous experimental findings and in stark contrast to previous numerical findings. We show that the latter studies did not maintain a constant Ra when varying the dispersion strength in their non-dimensional model, which explains the discrepancy. When considering larger α values, the dependence of the onset time on S becomes more complex: non-monotonic at intermediate values of α, and monotonically decreasing at large values.