Some thoughts on Darcy-type flow simulation for modelling underground CO 2 storage, based on the Sleipner CO 2 storage operation

摘要

We take three flow simulators, all based on Darcy’s Law but with different numerical solver implementations, to assess some of the issues surrounding their use to model underground CO2 storage. We focus on the Sleipner CO2 injection project, which, with its seismic monitoring datasets, provides unique insights into CO2 plume development during a large-scale injection operation. The case studies firstly compare simulator performance in terms of outputs and run-times on carefully matched model scenarios; then we compare numerical with analytical Darcy solutions to explore the potential for modelling simplification; finally we look at the effects of including conservation of energy in the simulations. The initial case-study used simplified axisymmetric model geometry to simulate the upward flux of CO2 through a heterogeneous reservoir, incorporating multiphase flow with coupled CO2 dissolution into formation brine. All three codes produced near-identical results with respect to CO2 migration velocity and total upward CO2 flux at the reservoir top. The second case-study involved 3D modelling of the growth of the topmost layer of CO2 trapped and migrating beneath topseal topography. Again the three codes showed excellent agreement. In the third case-study the simulators were tested against a simplified analytical solution for gravity currents to model the spreading of a single CO2 layer beneath a flat caprock. Neglecting capillary effects, the numerical models showed similar layer migration and geometry to the analytical model, but it was necessary to minimise the effects of numerical dispersion by adopting very fine cell thicknesses. The final case-study was designed to test the non-isothermal effects of injecting CO2 into a reservoir at non-ambient temperature. Only two of the simulators solve for conservation of energy, but both showed a near identical thermal anomaly, dominated by Joule-Thomson effects. These can be significant, particularly where reservoir conditions are close to the critical point for CO2 where property variations can significantly affect plume mobility and also seismic response. In conclusion, the three simulators show robust consistency, any differences far less than would result from geological parameter uncertainty and limitations of model resolution. In this respect the three implementations are significantly different in terms of computing resource requirement and it is clear that approaches with simplified physics will pay rich dividends in allowing more detailed reservoir heterogeneity to be included. Contrary to this, including conservation of energy is heavier on computing time but is likely to be required for storage scenarios where the injectant stream is significantly different in temperature to the reservoir and most critically for shallower storage reservoirs where CO2 is close to its critical point.