CO2 is very mobile and when injected into a porous rock formation, the resulting plume of supercritical fluid can travel large distances. A typical saline aquifer considered for CO2 storage can be viewed as a thin, slightly inclined sheet that spans thousands of square kilometers. Inside this sheet, the flow of CO2 is usually confined to thin layers underneath the sealing caprock or other low-permeable vertical barriers. This gives a large disparity in lateral and vertical scales, which in combination with differences in density between the supercritical CO2 plume and the resident brine, means that the vertical fluid segregation will be almost instantaneous compared with the up-dip migration. The CO2 tendency to form a relative flat fluid interface is an effect of the pressure distribution, which in turn depends strongly on the flow in the vertical direction. To avoid introducing large errors in the forecast of the updip migration, the vertical fluid distribution must therefore be represented accurately. In practice, this means using a higher grid resolution than what is computationally tractable in standard 3D simulators, and as a result, such simulations tend to be severely under-resolved unless they are conducted using large-scale systems for high-performance computing.
In a vertical equilibrium model, the primary assumption is that the flow system is in vertical equilibrium so that the vertical distribution of fluid phases can be determined from analytical expressions. One can then integrate the flow equations in the vertical direction to obtain a reduced model. This is a common approach that is used in many other branches of physics, e.g., when describing water waves, creep flow, etc. Integration in the vertical direction not only reduces the number of spatial dimensions, and hence the required number of grid cells, but will also lessens the coupling between pressure and fluid transport and improves the characteristic time constants of the problem. As a results, so-called vertical-equilibrium simulations will typically be orders of magnitude faster and consume signficantly less memory than conventional 3D simulators.
Using a vertical equilibrium (VE) assumption, the flow of a thin CO2 plume in 3D can be approximated in terms of its thickness to obtain a 2D simulation model as illustrated in Figure 1. The vertical integration reduces the dimension of the model, while important information of the heterogeneities in the underlying 3D medium is preserved in an averaged sense through upscaled effective properties that depend on the particular assumption used to model the vertical fluid distribution. The integrated flow equations can be expressed on a form that resembles the standard flow equations know from the simulation of petroleum reservoirs. The only difference is that some of the new effective fluid parameters that originally only depended on saturation, now also depend on pressure.
Figure 1: Schematic illustration of how a vertical-equilibrium model is used to describe the migration and resulting trapping of CO2 under a sloping caprock. Here, we have made the simplifying assumption that fine-scale capillary forces are negligible so that CO2 and brine are separated by a sharp interface. The aquifer will then be filled by four different pseudo-phases, which from top to bottom are: pure CO2, brine with residually trapped CO2, brine with dissolved CO2, and pure brine. If fine-scale capillary forces are included, the sharp interfaces will instead consist of a smooth transition, commony referred to as the capillary fringe.
The primary numerical formulation used in MRST-co2lab is a fully-implicit discretization. This is a pragmatic choice, motivated by the fact that if VE models are formulated in the black-oil framework, one can easily include the VE models into existing commercial and academic simulators developed for this framework and potentially develop hybrid models that combine VE and 3D models in different parts of the domain, extend the VE models with the wide range of physical effects that are described in the literature using the black-oil framework, and utilize robust and reasonably efficient numerical methods developed over the past four decades. These methods are particularly suitable for accurately computing steady states that correspond to various kinds of trapping in the long-term transient migration of CO2. However, for simple incompressible, sharp-interface models, MRST-co2lab also offers sequential solvers.
When studying the long-term migration of a well-formed CO2 plume, the errors resulting from the VE assumption may be significantly smaller than the errors introduced by the overly coarse resolution needed to make the 3D simulation model computationally tractable. There are several reasons for this:
As an illustration, see the simple comparison of VE and 3D simulations, which illustrates how a 3D simulation gradually converges to the corresponding VE simulation as the number of grid points in the vertical direction is increased.
On the other hand, traditional 3D simulation will in most cases be more suitable than using a VE approximation when simulating the fluid movement close to the injection point during injection or when simulating systems in which the migration of the CO2 is not dominated by the topography of the caprock. One of the strenghts of MRST is that it offers both possibilities, and we even have developed example implementations that couple 3D flow equations in the near-well regions to VE models in the far-field region.
The vertical-equilibrium simulators in MRST-co2lab can either be called from a script written for the specific case you want to investigate -- examples of such scripts are given in the tutorials below -- or you can use either of the two graphical user-interfaces that are capable of setting up a simulation.
Figures 2 and 3 show two examples of sharp-interface VE simulations. The simulations report a detailed volume inventory: residual CO2 will not leak and can be found inside structural traps, in regions behind the tail of the plume, or within the movable plume; free CO2 is found in traps and in the movable plume and may possibly in the future leak through holes in the caprock or reach the perimeter of the model; and finally, some CO2 may already have exited across open boundaries. With the simplest type of sharp-interface models, a typical simulation takes from tens of seconds to a few minutes, which enables you to interactively study different injection scenarios.
Figure 2: Structural trapping and VE simulation for the Johansen formation using data from the CO2 Storage Atlas. Once an injection point is chosen, the user can interactively change parameters such as injection rate and reservoir pressure, injection period, migration time, and homogeneous petrophysical parameters (perm and poro).
Figure 3: Vertical equilibrium simulation of the Sleipner injection, assuming a sharp interface between CO2 and brine. The model from ieaghg.org provides injection rates for 11 years. The simulation describes 30 years of injection (using the rate of year 11 for the last 19 years), followed by 220 years of migration. See also: using VE simulation to match a benchmark model of the 9th Sleipner layer to observed plume outlines
The following articles gives more details about different ways of formulating VE models and performing simulations and how to construct top-surface grid that represents a 3D rock formation
In this example we show how to set up a standard format black-oil model that can be used to simulate a VE model. For the actual simulation, we use the fully-implicit solver in MRST from the 'ad-fi' module, which is based on automatic differentiation.
Since the MRST2015a, a new, class-based framework for setting up fully-implicit simulations in a simple way has been introduced. Here, we use the framework to setup and run multiple simulations of an injection scenario, using different assumptions on boundary conditions and capillary pressure.
This example shows a conceptual model of a multilayered system where there is leakage through the caprock that delineates each layer. The system is simulated using a standard fully-implicit simulator with a fluid model that mimics a vertical equilibrium formulation.
Read more about the VE formulations implemented in MRST and their applications in the following papers:
Published July 2, 2014