Numerical simulations are an essential tool for assessing the long-term storage of CO2 in saline aquifers, such as the proposed pilot project at the Wandoan site in Queensland’s Surat Basin.
However, the robustness of the numerical predictions depends on accurately modelling the important physical processes involved. Numerical predictions of the amount of dissolved CO2 during the early stages of a CO2 storage project, particularly during the injection period where the CO2 plume grows rapidly, overestimate the actual amount of dissolved CO2 due to the use of finite-sized grid blocks. A simple theoretical scaling has been used to demonstrate that this discretisation error can be accurately accounted for and effectively removed in numerical simulations.
In the long-term, the necessary use of coarse grid blocks in a computational model prohibits the accurate simulation of enhanced dissolution due to density-driven convective mixing. This type of mixing typically occurs at a spatial length scale that is smaller than the size of the grid blocks necessary in a field-scale simulation. In order to improve the long-term numerical predictions of CO2 dissolution in models that feature large grid blocks, a better understanding of the convective mixing process in heterogeneous reservoirs and the role of geochemical reactions is required.
A simple heterogeneity model consisting of a random distribution of impermeable horizontal barriers in an otherwise homogeneous porous media was used to demonstrate that an equivalent anisotropic model provided an adequate approximation of the long-term flux. The long-term flux for an anisotropic reservoir was shown to scale as (kv/kh) 1/2 times the isotropic estimate, a result that was confirmed by numerical simulations
A sub-grid-scale scheme, for reducing the error in the numerical predictions of long-term CO2 dissolution due to convective mixing, was developed. Several possibilities for implementing the scheme using grid-corrected properties were proposed and assessed using numerical simulations. This correction significantly reduced the difference between the fine-scale results and the results using much coarser models.
The results obtained in this project can be implemented in commercial simulation software to improve the modelling of the short-term and long-term behaviour of injected carbon dioxide.