2.1 Writing capillary pressure in terms of $C_{1}$

As CO₂ is injected into the saline formation, H₂O evaporates from the brine leaving NaCl behind as a precipitate in a dry-out zone that develops around the injection well. Following the commencement of CO₂ injection, there are therefore three distinct zones within the saline formation that should be considered (see figure 1): (i) The dry-out zone, which surrounds the injection well and contains only precipitated salt and CO₂ in the non-wetting fluid phase. (ii) The full mixture zone, which surrounds the dry-out zone and contains CO₂, H₂O and NaCl, distributed between the wetting and non-wetting fluid phases. (iii) The initial saline formation fluid zone, which surrounds the full mixture zone and contains only H₂O and NaCl in a wetting fluid phase.

Figure 1.

A schematic diagram illustrating the distribution of CO₂, water and salt around a CO₂ injection well in a saline formation.

Inspection of (2.13) and (2.14) reveals that the problem is hyperbolic for $C_{1}\notin (c_{12}(1-S_{3}),c_{11}(1-S_{3}))$ and not hyperbolic for $C_{1}\in (c_{12}(1-S_{3}),c_{11}(1-S_{3}))$ , because of the $\unicode[STIX]{x2202}\unicode[STIX]{x1D713}/\unicode[STIX]{x2202}\unicode[STIX]{x1D709}$ term. For the CO₂ injection scenario described above, both Zones 1 and 3 are hyperbolic. In contrast, Zone 2 is not hyperbolic. The discontinuities that separate the three zones are shock waves, which must satisfy the Rankine–Hugoniot condition (e.g. Orr Reference Orr2007).

Within Zone 2, the displacement of a wetting phase by a non-wetting phase represents a continuous drainage cycle such that $\unicode[STIX]{x1D713}$ can be treated as a unique function of  $S_{2}$ . Furthermore, because $S_{3}=0$ and $S_{2}=1-S_{1}$ , it follows, from (2.4), that

(2.19) $$\begin{eqnarray}S_{2}=\left\{\begin{array}{@{}ll@{}}1, & C_{1}\leqslant c_{12}\\ \displaystyle \frac{c_{11}-C_{1}}{c_{11}-c_{12}}, & c_{12}<C_{1}<c_{11}\\ 0, & C_{1}\geqslant c_{11}\end{array}\right.\end{eqnarray}$$

and

(2.20) $$\begin{eqnarray}\frac{\unicode[STIX]{x2202}S_{2}}{\unicode[STIX]{x2202}C_{1}}=\frac{1}{(c_{12}-c_{11})},\quad C_{1}\in (c_{12},c_{11})\end{eqnarray}$$

such that it can be said that

(2.21) $$\begin{eqnarray}\frac{\unicode[STIX]{x2202}\unicode[STIX]{x1D713}}{\unicode[STIX]{x2202}\unicode[STIX]{x1D709}}=\frac{1}{(c_{12}-c_{11})}\frac{\unicode[STIX]{x2202}\unicode[STIX]{x1D713}}{\unicode[STIX]{x2202}S_{2}}\frac{\unicode[STIX]{x2202}C_{1}}{\unicode[STIX]{x2202}\unicode[STIX]{x1D709}}.\end{eqnarray}$$

In this way, equation (2.14) can be substantially simplified to get

(2.22) $$\begin{eqnarray}F_{i}=\unicode[STIX]{x1D6FC}_{i}-\unicode[STIX]{x1D6FD}_{i}\unicode[STIX]{x1D709}\frac{\unicode[STIX]{x2202}C_{1}}{\unicode[STIX]{x2202}\unicode[STIX]{x1D709}},\end{eqnarray}$$

where

(2.23) $$\begin{eqnarray}\displaystyle & \displaystyle \unicode[STIX]{x1D6FC}_{i}=\left\{\begin{array}{@{}lll@{}}C_{i}, & C_{1}\notin (c_{12},c_{11}), & i\in \{1,2\}\\ \displaystyle \mathop{\sum }_{j=1}^{2}f_{j}c_{ij}, & C_{1}\in (c_{12},c_{11}), & i\in \{1,2\}\\ f_{2}\unicode[STIX]{x1D70E}_{32}, & C_{1}\in [0,1], & i=3\end{array}\right. & \displaystyle\end{eqnarray}$$

(2.24) $$\begin{eqnarray}\displaystyle & \displaystyle \unicode[STIX]{x1D6FD}_{i}=\left\{\begin{array}{@{}lll@{}}0, & C_{1}\notin (c_{12},c_{11}), & i\in \{1,2,3\}\\ \displaystyle G\mathop{\sum }_{j=1}^{2}(-1)^{j}c_{ij}, & C_{1}\in (c_{12},c_{11}), & i\in \{1,2\}\\ \displaystyle G\unicode[STIX]{x1D70E}_{32}, & C_{1}\in (c_{12},c_{11}), & i=3\end{array}\right. & \displaystyle\end{eqnarray}$$

and

(2.25) $$\begin{eqnarray}G=\frac{f_{2}k_{r1}}{Ca(c_{11}-c_{12})}\frac{\unicode[STIX]{x2202}\unicode[STIX]{x1D713}}{\unicode[STIX]{x2202}S_{2}}.\end{eqnarray}$$

When $Ca\rightarrow \infty$ and $\unicode[STIX]{x1D70E}_{32}=0$ , the above problem reduces to the hyperbolic problem solved by Orr (Reference Orr2007) using the MOC. When $c_{11}=1$ , $c_{12}=0$ and $\unicode[STIX]{x1D70E}_{32}=0$ , the above problem reduces to the immiscible two-phase flow problem with capillary pressure, previously solved by McWhorter & Sunada (Reference McWhorter and Sunada1990) and Bjornara & Mathias (Reference Bjornara and Mathias2013). The $G$ term in (2.25) is analogous to the $G$ term in (16) of Bjornara & Mathias (Reference Bjornara and Mathias2013).

2.2 Relative permeability and capillary pressure functions

Relative permeability is calculated from Corey curves but with relative permeability assumed to linearly increase with saturation to one beyond residual saturations:

(2.26) $$\begin{eqnarray}k_{rj}=\left\{\begin{array}{@{}ll@{}}0, & S_{j}\leqslant S_{jc}\\ \displaystyle k_{rj0}\left(\frac{S_{j}-S_{jc}}{1-S_{1c}-S_{2c}}\right)^{n_{j}}, & S_{jc}<S_{j}<1-S_{ic},\\ \displaystyle k_{rj0}+(1-k_{rj0})\left(\frac{S_{j}-1+S_{ic}}{S_{ic}}\right), & S_{j}\geqslant 1-S_{ic}\end{array}\right.\quad i\neq j.\end{eqnarray}$$

Dimensionless capillary pressure, $\unicode[STIX]{x1D713}$ , is calculated using the empirical equation of van Genuchten (Reference van Genuchten1980) in conjunction with, following Oostrom et al. (Reference Oostrom, White, Porse, Krevor and Mathias2016) and Zhang, Oostrom & White (Reference Zhang, Oostrom and White2016), the dry-region extension of Webb (Reference Webb2000):

(2.27) $$\begin{eqnarray}\unicode[STIX]{x1D713}=\left\{\begin{array}{@{}ll@{}}\displaystyle (S_{e}^{-1/m}-1)^{1/n}, & S_{2}>S_{2m}\\ \displaystyle \unicode[STIX]{x1D713}_{d}\exp \left[\ln \!\left(\frac{\unicode[STIX]{x1D713}_{m}}{\unicode[STIX]{x1D713}_{d}}\right)\frac{S_{2}}{S_{2m}}\right], & S_{2}\leqslant S_{2m},\end{array}\right.\end{eqnarray}$$

where $S_{e}$ [-] is an effective saturation found from

(2.28) $$\begin{eqnarray}S_{e}=\frac{S_{2}-S_{2c}}{1-S_{2c}}\end{eqnarray}$$

and $k_{rj0}$ [-], $S_{jc}$ [-] and $n_{j}$ [-] are the endpoint relative permeability, residual saturation and relative permeability exponent for phase  $j$ , respectively, $m$ [-] and $n$ [-] are empirical exponents associated with van Genuchten’s function, $\unicode[STIX]{x1D713}_{d}=P_{cd}/P_{c0}$ [-] where $P_{cd}$ [ML^-1T^-2] is the capillary pressure at which ‘oven-dry’ conditions are said to have occurred (according to Webb (Reference Webb2000), this is taken to be $10^{9}$  Pa) with

(2.29) $$\begin{eqnarray}S_{2m}=(1-S_{2c})S_{em}+S_{2c}\end{eqnarray}$$

and

(2.30) $$\begin{eqnarray}\unicode[STIX]{x1D713}_{m}=(S_{em}^{-1/m}-1)^{1/n},\end{eqnarray}$$

where $S_{em}$ [-] is a critical effective saturation at which the switch over between van Genuchten’s function and Webb’s extension take place, defined in the subsequent sub-section.

Differentiation of (2.27) with respect to $S_{2}$ leads to

(2.31) $$\begin{eqnarray}\frac{\unicode[STIX]{x2202}\unicode[STIX]{x1D713}}{\unicode[STIX]{x2202}S_{2}}=\left\{\begin{array}{@{}ll@{}}\displaystyle \frac{\unicode[STIX]{x1D713}}{(1-S_{2c})mnS_{e}(S_{e}^{1/m}-1)}, & S_{2}>S_{2m}\\ \displaystyle \frac{\unicode[STIX]{x1D713}}{S_{2m}}\ln \!\left(\frac{\unicode[STIX]{x1D713}_{m}}{\unicode[STIX]{x1D713}_{d}}\right), & S_{2}\leqslant S_{2m}.\end{array}\right.\end{eqnarray}$$

The van Genuchten capillary pressure function has been widely used in many previous CO₂ injection studies (e.g. Pruess & Muller Reference Pruess and Muller2009; Kim et al. Reference Kim, Han, Oh, Kim and Kim2012; Mathias et al. Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013; Oostrom et al. Reference Oostrom, White, Porse, Krevor and Mathias2016; Zhang et al. Reference Zhang, Oostrom and White2016). The Corey relative permeability functions have previously been useful in describing CO₂-brine relative permeability data from at least 25 different experiments from the international literature (Mathias et al. Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013).

2.3 Determination of $S_{em}$

Considering (2.31), Webb (Reference Webb2000) defines $S_{em}$ as the effective saturation at which

(2.32) $$\begin{eqnarray}\frac{\unicode[STIX]{x1D713}_{m}}{(1-S_{2c})mnS_{em}(S_{em}^{1/m}-1)}=\frac{\unicode[STIX]{x1D713}_{m}}{S_{2m}}\ln \!\left(\frac{\unicode[STIX]{x1D713}_{m}}{\unicode[STIX]{x1D713}_{d}}\right).\end{eqnarray}$$

Substituting (2.30) and (2.29) into (2.32) and rearranging leads to

(2.33) $$\begin{eqnarray}S_{em}=\frac{S_{em}+S_{2c}(1-S_{2c})^{-1}}{mn(S_{em}^{1/m}-1)\ln [(S_{em}^{-1/m}-1)^{1/n}\unicode[STIX]{x1D713}_{d}^{-1}]},\end{eqnarray}$$

which must be solved iteratively. Webb (Reference Webb2000) suggests that four to five iterations are sufficient. However, this will be strongly dependent on the initial estimate of $S_{em0}$ applied.

For $S_{2c}>0$ , a good initial estimate of $S_{em}$ , $S_{em0}$ , can be obtained by assuming $S_{em0}\ll 1$ such that (2.33) reduces to

(2.34) $$\begin{eqnarray}S_{em0}=\frac{S_{2c}(1-S_{2c})^{-1}}{\ln [S_{em0}\unicode[STIX]{x1D713}_{d}^{nm}]},\end{eqnarray}$$

which can be rearranged to get

(2.35) $$\begin{eqnarray}W\exp (W)=z,\end{eqnarray}$$

where

(2.36) $$\begin{eqnarray}z=\frac{S_{2c}\unicode[STIX]{x1D713}_{d}^{nm}}{(1-S_{2c})}\end{eqnarray}$$

and

(2.37) $$\begin{eqnarray}W=\frac{S_{2c}}{(1-S_{2c})S_{em0}}.\end{eqnarray}$$

Note that the functional inverse of $z(W)$ in (2.35), $W(z)$ , is given by the Lambert $W$ function. Furthermore, because $z$ is always positive and real, $W(z)=W_{0}(z)$ , otherwise referred to as the zero branch, which has the following asymptotic expansion (Corless et al. Reference Corless, Gonnet, Hare, Jeffrey and Knuth1996)

(2.38) $$\begin{eqnarray}W_{0}(z)=L_{1}-L_{2}+\frac{L_{2}}{L_{1}}+O\left(\left[\frac{L_{2}}{L_{1}}\right]^{2}\right),\end{eqnarray}$$

where $L_{2}=\ln L_{1}$ and $L_{1}=\ln z$ .

In this way, it can be said that

(2.39) $$\begin{eqnarray}S_{em0}=\frac{S_{2c}}{(1-S_{2c})W_{0}(z)},\end{eqnarray}$$

where $z$ is found from (2.36).

Examples of the iterative calculation of $S_{em}$ from initial guesses obtained from (2.39) are presented in table 1. When $S_{2c}\leqslant 0.3$ , it can be seen that convergence is achieved after just two iterations. When $S_{2c}=0.5$ , three iterations are required. When $S_{2c}=0.7$ , six iterations are required. The increase in the number of iterations required with increasing $S_{2c}$ is due to reducing validity of the $S_{em}\ll 1$ assumption.

Table 1.

Examples of the iterative calculation of $S_{em}$ for different values of $S_{2c}$ (as indicated in the top row) using (2.33) with $m=0.5$ , $P_{c0}=19.6$  kPa and $P_{cd}=10^{9}$  Pa. The initial guess, $S_{em0}$ , is calculated using (2.39).

2.4 Application of a similarity transform

The partial differential equation in (2.13) can be reduced to an ordinary differential equation by application of the following similarity transform

(2.40) $$\begin{eqnarray}\unicode[STIX]{x1D706}=\frac{\unicode[STIX]{x1D709}}{\unicode[STIX]{x1D70F}}.\end{eqnarray}$$

Substituting (2.40) into (2.13) and (2.22) leads to

(2.41) $$\begin{eqnarray}\frac{\text{d}F_{i}}{\text{d}C_{i}}=\unicode[STIX]{x1D706}\end{eqnarray}$$

and

(2.42) $$\begin{eqnarray}F_{i}=\unicode[STIX]{x1D6FC}_{i}-\unicode[STIX]{x1D6FD}_{i}\unicode[STIX]{x1D706}\frac{\text{d}C_{1}}{\text{d}\unicode[STIX]{x1D706}}.\end{eqnarray}$$

Differentiating both sides of (2.41) with respect to $C_{i}$ yields

(2.43) $$\begin{eqnarray}\frac{\text{d}^{2}F_{i}}{\text{d}C_{i}^{2}}=\frac{\text{d}\unicode[STIX]{x1D706}}{\text{d}C_{i}},\end{eqnarray}$$

which on substitution into (2.42), along with (2.41), and rearranging leads to

(2.44) $$\begin{eqnarray}\frac{\text{d}^{2}F_{1}}{\text{d}C_{1}^{2}}+\frac{\unicode[STIX]{x1D6FD}_{1}}{(F_{1}-\unicode[STIX]{x1D6FC}_{1})}\frac{\text{d}F_{1}}{\text{d}C_{1}}=0.\end{eqnarray}$$

In the event that the boundary and initial values of $C_{1}$ , $C_{10}$ and $C_{1I}$ , respectively, are $\notin (c_{12},c_{11})$ , the boundary conditions for (2.44) must satisfy the Rankine–Hugoniot conditions (similar to Orr Reference Orr2007, p. 75):

(2.45) $$\begin{eqnarray}\displaystyle & \displaystyle \frac{\text{d}F_{1}}{\text{d}C_{1}}=\frac{\unicode[STIX]{x1D6FC}_{10}-F_{1}}{C_{10}-C_{1}},\quad C_{1}\geqslant c_{11}, & \displaystyle\end{eqnarray}$$

(2.46) $$\begin{eqnarray}\displaystyle & \displaystyle \frac{\text{d}F_{1}}{\text{d}C_{1}}=\frac{\unicode[STIX]{x1D6FC}_{1I}-F_{1}}{C_{1I}-C_{1}},\quad C_{1}\leqslant c_{12}, & \displaystyle\end{eqnarray}$$

where $\unicode[STIX]{x1D6FC}_{10}$ and $\unicode[STIX]{x1D6FC}_{1I}$ represent the boundary and initial values of $\unicode[STIX]{x1D6FC}_{1}$ associated with $C_{10}$ and $C_{1I}$ , respectively. Alternatively, when $C_{10}$ and $C_{1I}$ are $\in (c_{12},c_{11})$

(2.47) $$\begin{eqnarray}\left.\begin{array}{@{}ll@{}}F_{1}=\unicode[STIX]{x1D6FC}_{10}, & C_{1}=C_{10},\\ F_{1}=\unicode[STIX]{x1D6FC}_{1I}, & C_{1}=C_{1I}.\end{array}\right\}\end{eqnarray}$$

An efficient way of expressing both (2.46) and (2.47) simultaneously is to state instead:

(2.48) $$\begin{eqnarray}\left.\begin{array}{@{}ll@{}}\displaystyle (C_{10}-C_{1})\frac{\text{d}F_{1}}{\text{d}C_{1}}+F_{1}=\unicode[STIX]{x1D6FC}_{10}, & C_{1}=\tilde{C}_{10},\\ \displaystyle (C_{1I}-C_{1})\frac{\text{d}F_{1}}{\text{d}C_{1}}+F_{1}=\unicode[STIX]{x1D6FC}_{1I}, & C_{1}=\tilde{C}_{1I},\end{array}\right\}\end{eqnarray}$$

where

(2.49) $$\begin{eqnarray}\displaystyle & \displaystyle \tilde{C}_{10}=H(C_{10}-c_{11})c_{11}+H(c_{11}-C_{10})C_{10}, & \displaystyle\end{eqnarray}$$

(2.50) $$\begin{eqnarray}\displaystyle & \displaystyle \tilde{C}_{1I}=H(c_{12}-C_{1I})c_{12}+H(C_{1I}-c_{12})C_{1I} & \displaystyle\end{eqnarray}$$

and $H(x)$ is a Heaviside function.

2.5 Pseudospectral solution

Following Bjornara & Mathias (Reference Bjornara and Mathias2013), the boundary value problem described in the previous section is solved using a Chebyshev polynomial differentiation matrix, $\unicode[STIX]{x1D63F}$ (Weideman & Reddy Reference Weideman and Reddy2000).

The coordinate space for the Chebyshev nodes is $x\in [-1,1]$ . However, the solution space for $F_{1}$ is $C_{1}\in [\tilde{C}_{1I},\tilde{C}_{10}]$ . Therefore the Chebyshev nodes, $\boldsymbol{x}$ , need to be mapped to the $C_{1}$ space by the following transform

(2.51) $$\begin{eqnarray}C_{1}=\frac{\tilde{C}_{10}+\tilde{C}_{1I}}{2}+\left(\frac{\tilde{C}_{10}-\tilde{C}_{1I}}{2}\right)x.\end{eqnarray}$$

Consequently, it is necessary to introduce an appropriately transformed differentiation matrix, $\unicode[STIX]{x1D640}$ , where

(2.52) $$\begin{eqnarray}\unicode[STIX]{x1D640}=\frac{\text{d}x}{\text{d}C_{1}}\unicode[STIX]{x1D63F}\end{eqnarray}$$

and from (2.51)

(2.53) $$\begin{eqnarray}\frac{\text{d}x}{\text{d}C_{1}}=\frac{2}{\tilde{C}_{10}-\tilde{C}_{1I}}.\end{eqnarray}$$

By applying the Chebyshev polynomial on the internal nodes and the Robin boundary conditions in (2.48) on the end nodes, equation (2.44) can be written in matrix form (similar to Piche & Kanniainen (Reference Piche and Kanniainen2009) and Bjornara & Mathias (Reference Bjornara and Mathias2013))

(2.54) $$\begin{eqnarray}\boldsymbol{R}(\boldsymbol{F})=\left[\begin{array}{@{}c@{}}\displaystyle \unicode[STIX]{x1D640}_{2:N-1,:}^{(2)}\boldsymbol{F}+\unicode[STIX]{x1D644}_{2:N-1,:}\text{diag}\left[\frac{\unicode[STIX]{x1D6FD}_{1}}{F_{1}-\unicode[STIX]{x1D6FC}_{1}}\right]\unicode[STIX]{x1D640}^{(1)}\boldsymbol{F}\\ (\boldsymbol{C}_{N}-C_{1I})\unicode[STIX]{x1D640}_{N,:}^{(1)}\boldsymbol{F}-\unicode[STIX]{x1D644}_{N,:}\boldsymbol{F}+\unicode[STIX]{x1D6FC}_{1I}\\ (\boldsymbol{C}_{1}-C_{10})\unicode[STIX]{x1D640}_{1,:}^{(1)}\boldsymbol{F}-\unicode[STIX]{x1D644}_{1,:}\boldsymbol{F}+\unicode[STIX]{x1D6FC}_{10}\end{array}\right],\end{eqnarray}$$

where $\boldsymbol{R}$ is the residual vector, $\boldsymbol{F}$ is the solution vector for the dependent variable $F_{1}$ , $\unicode[STIX]{x1D644}$ is an identity matrix, $\boldsymbol{C}$ is the vector containing the corresponding values of $C_{1}$ and $N$ denotes the number of Chebyshev nodes to be solved for. The two last rows on the right-hand side of (2.54) impose the Robin boundary conditions. Also note that $\unicode[STIX]{x1D640}^{(n)}$ can be obtained from  $\unicode[STIX]{x1D640}^{n}$ .

The solution vector, $\boldsymbol{F}$ , can be obtained by Newton iteration, whereby new iterations, $\boldsymbol{F}_{(i+1)}$ , are obtained from

(2.55) $$\begin{eqnarray}\boldsymbol{F}_{(i+1)}=\boldsymbol{F}_{(i)}-(\unicode[STIX]{x2202}\boldsymbol{R}/\unicode[STIX]{x2202}\boldsymbol{F}_{(i)})^{-1}\boldsymbol{R}(\boldsymbol{F}_{(i)}),\end{eqnarray}$$

where $\unicode[STIX]{x2202}\boldsymbol{R}/\unicode[STIX]{x2202}\boldsymbol{F}$ is the Jacobian matrix defined as

(2.56) $$\begin{eqnarray}\frac{\unicode[STIX]{x2202}\boldsymbol{R}}{\unicode[STIX]{x2202}\boldsymbol{F}}=\left[\begin{array}{@{}c@{}}\displaystyle \unicode[STIX]{x1D640}_{2:N-1,:}^{(2)}+\unicode[STIX]{x1D644}_{2:N-1,:}\text{diag}\left[\frac{\unicode[STIX]{x1D6FD}_{1}}{F_{1}-\unicode[STIX]{x1D6FC}_{1}}\right]\unicode[STIX]{x1D640}^{(1)}-\unicode[STIX]{x1D644}_{2:N-1,:}\text{diag}\!\left[\text{diag}\!\left[\frac{\unicode[STIX]{x1D6FD}_{1}}{(F_{1}-\unicode[STIX]{x1D6FC}_{1})^{2}}\right]\unicode[STIX]{x1D640}^{(1)}\boldsymbol{F}\right]\\ (\boldsymbol{C}_{N}-C_{1I})\unicode[STIX]{x1D640}_{N,:}^{(1)}-\unicode[STIX]{x1D644}_{N,:}\\ (\boldsymbol{C}_{1}-C_{10})\unicode[STIX]{x1D640}_{1,:}^{(1)}-\unicode[STIX]{x1D644}_{1,:}\end{array}\right]\!.\end{eqnarray}$$

Note that $F_{1}$ is bounded by $\unicode[STIX]{x1D6FC}_{1}$ and $\unicode[STIX]{x1D6FC}_{10}$ . Therefore, a good initial guess is to set $F_{1}=\unicode[STIX]{x1D6FC}_{10}$ . Following Bjornara & Mathias (Reference Bjornara and Mathias2013), an additional correction step should be applied in the Newton iteration to force the solution, $F_{1}$ , to be less than $\unicode[STIX]{x1D6FC}_{1}$ . The Newton iteration loop is assumed to have converged when the mean absolute value of $\boldsymbol{R}\leqslant 10^{-9}$ . With 100 Chebyshev nodes (i.e. $N=100$ ), convergence is typically achieved with less than 200 iterations.

2.6 Dealing with salt precipitation in the dry-out zone

Now consider the case where pure CO₂ is injected into a porous medium (i.e. $\unicode[STIX]{x1D6FC}_{10}=1$ ) initially fully saturated with brine (i.e. $\unicode[STIX]{x1D6FC}_{1I}=0$ ). Let $\unicode[STIX]{x1D70E}_{32}$ be the volume fraction of NaCl in phase 2 throughout the system. In this way, the volume fraction of H₂O in phase 2 prior to CO₂ injection is $(1-\unicode[STIX]{x1D70E}_{32})$ .

Let $r_{0}$ [L] and $r_{I}$ [L] be the radial extents of the dry-out zone and injected CO₂ plume respectively. At any given time, the volume of H₂O evaporated by the CO₂, $V_{e}$ [L³], can be found from

(2.57) $$\begin{eqnarray}V_{e}=2\unicode[STIX]{x03C0}\unicode[STIX]{x1D719}H(1-c_{11})\int _{r_{0}}^{r_{I}}rS_{1}\,\text{d}r.\end{eqnarray}$$

The volume of salt precipitated in the dry-out zone, $V_{s}$ [L³], is found from

(2.58) $$\begin{eqnarray}V_{s}=\frac{\unicode[STIX]{x1D70E}_{32}V_{e}}{1-\unicode[STIX]{x1D70E}_{32}}.\end{eqnarray}$$

The volume of the dry-out zone where the salt is precipitated, $V_{d}$ [L³], is found from

(2.59) $$\begin{eqnarray}V_{d}=\unicode[STIX]{x03C0}\unicode[STIX]{x1D719}Hr_{0}^{2}.\end{eqnarray}$$

Another quantity of interest is the volume of CO₂ dissolved in the brine, $V_{c}$ [L³], which can be found from

(2.60) $$\begin{eqnarray}V_{c}=2\unicode[STIX]{x03C0}\unicode[STIX]{x1D719}Hc_{12}\int _{r_{0}}^{r_{I}}r(1-S_{1})\,\text{d}r.\end{eqnarray}$$

Considering the definition of $\unicode[STIX]{x1D706}$ in (2.40) in conjunction with (2.15) and (2.16)

(2.61) $$\begin{eqnarray}r_{0}^{2}=\frac{Q_{0}t\unicode[STIX]{x1D706}_{0}}{\unicode[STIX]{x03C0}\unicode[STIX]{x1D719}H}\quad \text{and}\quad r_{I}^{2}=\frac{Q_{0}t\unicode[STIX]{x1D706}_{I}}{\unicode[STIX]{x03C0}\unicode[STIX]{x1D719}H},\end{eqnarray}$$

where, recall (2.41) and (2.48), $\unicode[STIX]{x1D706}_{0}$ and $\unicode[STIX]{x1D706}_{I}$ can be found from

(2.62) $$\begin{eqnarray}\unicode[STIX]{x1D706}_{0}=\left.\frac{\text{d}F_{1}}{\text{d}C_{1}}\right|_{C_{1}=c_{11}}\quad \text{and}\quad \unicode[STIX]{x1D706}_{I}=\left.\frac{\text{d}F_{1}}{\text{d}C_{1}}\right|_{C_{1}=c_{12}}.\end{eqnarray}$$

In this way it can be understood that:

(2.63) $$\begin{eqnarray}\displaystyle & \displaystyle V_{e}=(1-c_{11})Q_{0}t\int _{\unicode[STIX]{x1D706}_{0}}^{\unicode[STIX]{x1D706}_{I}}S_{1}\,\text{d}\unicode[STIX]{x1D706}, & \displaystyle\end{eqnarray}$$

(2.64) $$\begin{eqnarray}\displaystyle & V_{d}=Q_{0}t\unicode[STIX]{x1D706}_{0}, & \displaystyle\end{eqnarray}$$

(2.65) $$\begin{eqnarray}\displaystyle & \displaystyle V_{c}=c_{12}Q_{0}t\int _{\unicode[STIX]{x1D706}_{0}}^{\unicode[STIX]{x1D706}_{I}}(1-S_{1})\,\text{d}\unicode[STIX]{x1D706}. & \displaystyle\end{eqnarray}$$

Noting that the rates at which $V_{s}$ and $V_{d}$ grow with time are constant it can also be understood that the volume fraction of precipitated salt, $C_{3}$ , will be both uniform within the dry-out zone and constant with time. The value of $C_{3}$ within the dry-out zone, hereafter denoted as $C_{30}$ , can be found from

(2.66) $$\begin{eqnarray}C_{30}=\frac{(1-c_{11})\unicode[STIX]{x1D70E}_{32}}{(1-\unicode[STIX]{x1D70E}_{32})\unicode[STIX]{x1D706}_{0}}\int _{\unicode[STIX]{x1D706}_{0}}^{\unicode[STIX]{x1D706}_{I}}S_{1}\,\text{d}\unicode[STIX]{x1D706}.\end{eqnarray}$$

Given that $C_{10}=1-C_{30}$ , $C_{1I}=0$ , $\unicode[STIX]{x1D6FC}_{10}=1$ and $\unicode[STIX]{x1D6FC}_{1I}=0$ , the boundary conditions in (2.48) reduce to

(2.67) $$\begin{eqnarray}\left.\begin{array}{@{}ll@{}}\displaystyle \frac{\text{d}F_{1}}{\text{d}C_{1}}=\frac{1-F_{1}}{1-C_{30}-c_{11}}, & C_{1}=c_{11},\\ \displaystyle \frac{\text{d}F_{1}}{\text{d}C_{1}}=\frac{F_{1}}{c_{12}}, & C_{1}=c_{12}.\end{array}\right\}\end{eqnarray}$$

Values of $C_{30}$ can be obtained iteratively by repeating the procedures outlined in § 2.5 with successive estimates of $C_{30}$ obtained from (2.66). Using an initial guess of $C_{30}=0$ , this process is found to typically converge after less than 60 iterations. The integrals in (2.65) and (2.63) can be found by trapezoidal integration.

3.1 Gas displacing oil

As a first example, the gas-displacing-oil scenario previously presented in figures 4.13 and 4.15 of Orr (Reference Orr2007) is adopted. The parameters describing the scenario include $c_{11}=0.95$ , $c_{12}=0.20$ , $\unicode[STIX]{x1D70E}_{32}=0$ , $\unicode[STIX]{x1D707}_{2}/\unicode[STIX]{x1D707}_{1}=2$ , $S_{1c}=0.05$ , $S_{2c}=0.1$ , $k_{r10}=k_{r20}=1$ and $n_{1}=n_{2}=2$ . For the pseudospectral solution, a value for the van Genuchten (Reference van Genuchten1980) parameter, $m$ , is set to 0.5.

Plots of $C_{1}$ against $\text{d}F_{1}/\text{d}C_{1}$ (which, recall, is equal to $\unicode[STIX]{x1D709}/\unicode[STIX]{x1D70F}$ ) for this scenario are shown in figure 2. The different subplots show the effect of varying the boundary volume fraction, $C_{10}$ , and the initial volume fraction, $C_{1I}$ . The different colours relate to different assumed values of  $Ca$ . Increasing $Ca$ can be thought of as analogous to an increased injection rate. The $Ca\rightarrow \infty$ curves were obtained from the MOC solutions previously presented in figures 4.13 and 4.15 of Orr (Reference Orr2007). The finite $Ca$ value solutions were obtained using the pseudospectral solution described above, with 100 Chebyshev nodes.

Figure 2.

Sensitivity analysis based on gas-displacing-oil examples. The infinite $Ca$ value curves were obtained from the method of characteristics solutions presented in figures 4.13 and 4.15 of Orr (Reference Orr2007). The finite $Ca$ value curves were obtained using the pseudospectral solution documented in the current article.

When $Ca=100$ , the pseudospectral solution is virtually identical to the infinite- $Ca$ -MOC solutions. As $Ca$ is decreased, the solution becomes more diffused. In figure 2(a,b,d and f), the infinity $Ca$ results exhibit a trailing shock, which represents a dry-out zone where all the liquid oil has been evaporated by the gas. Of particular interest is that decreasing $Ca$ leads to a reduction in the thickness of the dry-out zone, ultimately leading to its complete elimination.

3.2 CO₂ injection in a saline formation

Here the CO₂-injection-in-a-saline-formation scenario, previously presented by Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013), is revisited. The example involves injecting pure CO₂ at a constant rate via a fully penetrating injection well at the centre of a cylindrical, homogenous and confined saline formation, initially fully saturated with brine. Relevant model parameters are presented in table 2. In this case, components 1, 2 and 3 are CO₂, H₂O and NaCl, respectively, and phases 1, 2 and 3 represent a CO₂-rich phase, an H₂O rich phase and precipitated salt, respectively.

Table 2.

Relevant model parameters used for the CO₂ injection in saline formation scenario, previously presented by Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013).

The relevant fluid properties are obtained using equations of state (EOS) and empirical equations provided by Batzle & Wang (Reference Batzle and Wang1992), Fenghour, Wakeham & Vesovic (Reference Fenghour, Wakeham and Vesovic1998), Spycher et al. (Reference Spycher, Pruess and Ennis-King2003) and Spycher & Pruess (Reference Spycher and Pruess2005). Mathias et al. (Reference Mathias, Gonzalez Martinez de Miguel, Thatcher and Zimmerman2011a ) found that when using analytical solutions in this context, to account for the relatively high compressibility of CO₂, it is important to use an estimate of the final pressure rather than the initial pressure for calculating the fluid properties relating to CO₂. Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013) found that, for the scenario described in table 2, the well pressure increased by just over 5 MPa after 10 years. Therefore, for the current study, fluid properties are calculated using 15 MPa as opposed to 10 MPa.

The EOS of Spycher et al. (Reference Spycher, Pruess and Ennis-King2003) and Spycher & Pruess (Reference Spycher and Pruess2005) provide equilibrium mole fractions as opposed to volume fractions. Pruess & Spycher (Reference Pruess and Spycher2007) show how mole fractions can be converted to mass fractions, $x_{ij}$ [-], which can be converted to volume fractions, $\unicode[STIX]{x1D70E}_{ij}$ [-], using (similar to Orr Reference Orr2007, p. 19)

(3.1) $$\begin{eqnarray}\unicode[STIX]{x1D70E}_{ij}=\frac{\unicode[STIX]{x1D70C}_{j}x_{ij}}{\unicode[STIX]{x1D70C}_{ij}},\end{eqnarray}$$

where $\unicode[STIX]{x1D70C}_{ij}$ [ML^-3] is the density of component $i$ in phase $j$ and $\unicode[STIX]{x1D70C}_{j}$ [ML^-3] is the composite phase density, which can be found from

(3.2) $$\begin{eqnarray}\unicode[STIX]{x1D70C}_{j}=\left(\mathop{\sum }_{i=1}^{N_{c}}\frac{x_{ij}}{\unicode[STIX]{x1D70C}_{ij}}\right)^{-1},\end{eqnarray}$$

where $N_{c}$ [-] is the number of components present. Because the pseudospectral solution above assumes component densities remain constant throughout, a decision is made that $\unicode[STIX]{x1D70C}_{12}=\unicode[STIX]{x1D70C}_{11}$ , $\unicode[STIX]{x1D70C}_{21}=\unicode[STIX]{x1D70C}_{22}$ and $\unicode[STIX]{x1D70C}_{32}=\unicode[STIX]{x1D70C}_{33}$ .

Table 3 shows how the various fluid properties vary with depth below sea level in this context. Depth is related to pressure by assuming hydrostatic conditions and then adding 5 MPa to allow for pressure induced by CO₂ injection. Depth is related to temperature by assuming a geothermal gradient of $40\,^{\circ }\text{C}~\text{km}^{-1}$ . It can be seen that the volume fractions are largely unaffected by depth. However, the variation in brine viscosity and CO₂ density are more noticeable.

Table 3.

Relevant model parameters used for the CO₂ injection in a saline formation scenario with a brine salinity of 150 ppt.

A comparison of results from the pseudospectral solution with those from the TOUGH2 simulation reported by Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013) is shown in figure 3, alongside results for when $Ca\rightarrow \infty$ , obtained using a MOC solution similar to that previously presented by Zeidouni et al. (Reference Zeidouni, Pooladi-Darvish and Keith2009) and Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel and Hosseini2011b ). Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013) assumed $P_{c0}=19.6$  kPa. Considering the other parameters in tables 2 and 3, this leads to a $Ca$ value of 1.7. There is excellent correspondence between the MOC solution, the TOUGH2 results and the pseudospectral solution when $Ca=1.7$ .

Figure 3.

Plots of CO₂ saturation against radial distance after injecting 4.73 Mt of CO₂ whilst assuming a range of different capillary numbers, $Ca$ . The TOUGH2 results are from the simulations previously presented by Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013). Other associated model parameters are presented in table 2. The results for $Ca\rightarrow \infty$ were obtained using a method of characteristics solution, also presented by Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013). The results for finite $Ca$ values were obtained using the pseudospectral solution.

A value of $P_{c0}=19.6$  kPa is often used to describe saline formations in a CO₂ storage context (Rutqvist et al. Reference Rutqvist, Birkholzer, Cappa and Tsang2007; Zhou et al. Reference Zhou, Birkholzer, Tsang and Rutqvist2008; Mathias et al. Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013; Zhu et al. Reference Zhu, Zuo, Zhang, Zhang, Wang and Wang2015, e.g.). Experimental analysis looking at four different sandstone reservoirs revealed a range of $P_{c0}$ values from 1.3–7.1 kPa (Oostrom et al. Reference Oostrom, White, Porse, Krevor and Mathias2016). Smaller values of $P_{c0}$ imply larger pore diameters.

A hallmark of hyperbolic theory is that the problem can be reduced to a fundamental wave structure which constitutes the solution. In figure 3, it can be seen that such a wave structure is largely preserved, despite the inclusion of capillary diffusion. Furthermore, the wave velocity of the leading shock is virtually independent of $Ca$ for the range of $Ca$ values studied. However, decreasing $Ca$ leads to a more diffused spreading wave caused by the increase in capillary diffusion, which in turn leads to a reduction in the wave velocity of the trailing shock, as also seen in figure 2(a). The decrease in steady-state CO₂ saturation in the dry-out zone is caused by an increase in the volume fraction of precipitated salt (recall that $C_{10}=1-C_{30}$ ).

For the scenarios depicted in figure 3, $C_{30}$ is found to be insensitive to $Ca$ for $Ca$ values greater than or equal to 1.7. However for $Ca$ values less than 1.7, the volume of the dry-out zone is significantly reduced and the volume fraction of precipitated salt is significantly increased. The value of $C_{30}$ for $Ca=0.2$ is almost double the value for $Ca=1.7$ . The value of $C_{30}$ for $Ca=0.1$ is around 10 times that of when $Ca=1.7$ . The $Ca=1.7$ scenario described in table 2 assumes an injection rate of $15~\text{kg}~\text{s}^{-1}$ . The results shown in figure 3 therefore suggest that reducing the injection rate down to $1.8~\text{kg}~\text{s}^{-1}$ would lead to a doubling of the volume fraction of precipitated salt around the injection well. Furthermore, reducing the injection rate from $15~\text{kg}~\text{s}^{-1}$ down to $0.9~\text{kg}~\text{s}^{-1}$ would lead to an almost 10 times larger volume fraction of precipitated salt around the injection well.

Figure 4.

Plots of $F_{1}$ , $\unicode[STIX]{x1D6FC}_{1}$ and $\unicode[STIX]{x1D6FD}_{1}$ against $C_{1}$ for the simulation results presented in figure 3.

For the hyperbolic case when $Ca\rightarrow \infty$ , it is common to study plots of $F_{1}$ and $C_{1}$ (Orr Reference Orr2007). Figure 4(a) shows plots of $F_{1}$ against $C_{1}$ for all the values of $Ca$ presented in figure 3 along with a plot of $\unicode[STIX]{x1D6FC}_{1}$ . The MOC solution (i.e. with $Ca\rightarrow \infty$ ), which sits almost exactly underneath the $Ca=1.7$ line, intersects the $\unicode[STIX]{x1D6FC}_{1}$ line at tangents, which is symptomatic of satisfying the shock waves satisfying the Rankine–Hugoniot condition. To better visualize the results for finite $Ca$ values, $(1-F_{1})$ is shown on a log scale in figure 4(b). Here it can be seen that the models approach a value of $F_{1}=1$ at different $C_{1}$ values depending on the volume fraction of precipitated salt. The volume fraction of precipitated salt increases with decreasing $Ca$ . Figure 4(c) shows a close-up view of the trailing shocks on linear axes for further reference. For finite $Ca$ values, the  $F_{1}$ lines never actually intersect the $\unicode[STIX]{x1D6FC}_{1}$ line except at where $C_{1}=0$ . The reason for this is due to $\unicode[STIX]{x1D6FD}_{1}$ , which is plotted in figure 4(d). The highest values of $\unicode[STIX]{x1D6FD}_{1}$ are at the centre of the two-phase region, $C_{1}\in (c_{12},c_{11})$ . $\unicode[STIX]{x1D6FD}_{1}$ smoothly grades down to zero as it reaches the single-phase regions, $C_{1}\notin (c_{12},c_{11})$ .

A further sensitivity analysis is presented in figure 5. The three depth scenarios presented in table 3 are applied with three different brine salinities. Figure 5(a) shows how the volume of the dry-out zone decreases with decreasing $Ca$ . The size of the dry-out zone increases with increasing depth. In contrast, brine salinity has very little impact on dry-out zone volume.

Figure 5.

Sensitivity analysis based around the scenario presented in figure 3. The different colours relate to different brine salinities, as indicated in the legend. The solid lines, dashed lines and dash-dotted lines represent results obtained using fluid properties calculated assuming the saline formation exists at a depth of 1000 m, 1500 m and 2000 m, respectively (based on hydrostatic pressure conditions and a geothermal gradient of $40\,^{\circ }\text{C}~\text{km}^{-1}$ as in table 3). (a) Shows plots of the ratio of dry-out zone volume ( $V_{d}$ ) to injected CO₂ volume ( $Q_{0}t$ ) against capillary number ( $Ca$ ). (b) Shows plots of the ratio of volume of evaporated water ( $V_{e}$ ) to $Q_{0}t$ against $Ca$ . (c) Shows plots of the ratio of volume of dissolved CO₂ ( $V_{c}$ ) to $Q_{0}t$ against $Ca$ . (d) Shows plots of precipitated salt volume fraction in the dry-out zone ( $C_{30}$ ) against $Ca$ .

Figure 5(b) shows the volume of the evaporated water also reduces with decreasing $Ca$ . At first this seems surprising given that capillary pressure effects should bring more water into the dry-out zone. However, the effect of the capillary pressure is also to spread the CO₂ out further (see leading edge of CO₂ plumes in figure 3). As a consequence, more CO₂ is dissolved (see figure 5 c). Consequently, less of the CO₂-rich phase is available for water from the brine to evaporate into. The volume of evaporated water increases with depth because the equilibrium volume fraction of water in the CO₂-rich phase increases with depth (see table 3). The volume of dissolved CO₂ is insensitive to depth but decreases with increasing brine salinity. The latter is because the solubility limit of CO₂ in brine decreases substantially with increasing salinity (Spycher & Pruess Reference Spycher and Pruess2005).

Figure 5(d) shows how volume fraction of precipitated salt in the dry-out zone, $C_{30}$ , superlinearly increases with decreasing $Ca$ . For $Ca>0.25$ , the quantity of precipitated salt is mostly controlled by brine salinity. However, for $Ca<0.25$ , depth plays an increasingly important role, with higher levels of salt precipitation in shallower formations. This is because the dry-out zone increases with depth, despite increasing water evaporation with depth. Figure 6 shows the same results as figure 5(d) but with $C_{30}$ normalized by dividing by the salinity of the brine, $X_{32}$ . Here it can be seen that $C_{30}$ almost linearly scales with  $X_{32}$ .

Figure 6.

The same as figure 5(d) except that salt volume fraction, $C_{30}$ , is divided by the salinity of the brine,  $X_{32}$ .

The volume fraction of precipitated salt is also strongly controlled by the relative permeability parameters, $k_{rj0}$ , $S_{jc}$ and $n_{j}$ (Zhang et al. Reference Zhang, Oostrom and White2016). The analysis performed to provide figure 6 was repeated for the 1000 m depth scenario for each of the six groups of relative permeability parameters presented in table 4. These six parameter sets were selected from a database of 25 core experiments previously compiled by Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013). The six cores were selected to provide a representative range of possible outcomes, given the wide variability generally observed in such data sets.

From figure 7 it can be seen that the high $Ca$ values of $C_{30}$ range from 0.019 to 0.044. Furthermore, the critical $Ca$ value below which $C_{30}$ superlinearly increases ranges from 0.025 to 10. Comparing these results with the parameter sets in table 4 it can be seen that when the relative permeability for brine is more linear, the value of $C_{30}$ at high values of $Ca$ tends to be lower. However, this linearity also leads to the superlinearly increasing of $C_{30}$ with decreasing $Ca$ to occur at a relatively low value of $C_{30}$ (see for example Cardium no. 1 and Basal Cambrian). Exactly the opposite happens when the relative permeability for brine is highly nonlinear (see for example Paaratte and Tuscaloosa). This is probably due to counter-current flow of water being less efficient when relative permeability is highly nonlinear.

Figure 7.

Plot of normalized precipitated salt volume fraction, $C_{30}$ , against capillary number, $Ca$ , using the 1000 m depth model scenario described in tables 2 and 3 in conjunction with the different relative permeability parameters given in table 4.

Table 4.

Relative permeability parameters for six different sandstone cores (after Mathias et al. Reference Mathias, Gluyas, Gonzlez Martnez de Miguel, Bryant and Wilson2013). Note that for each core $k_{r20}=1$ and $S_{1c}=0$ . Data for Cardium no. 1, Basal Cambrian and Viking no. 1 were originally obtained by Bennion & Bachu (Reference Bennion and Bachu2008). Data for Otway were originally obtained by Perrin & Benson (Reference Perrin and Benson2010). Data for Paaratte and Tuscaloosa were originally obtained by Krevor et al. (Reference Krevor, Pini, Zuo and Benson2012).

4.1 Incompressible fluids

Fluid densities are assumed to be independent of pressure. The compressibilities of H₂O and NaCl are commonly ignored. For pressures and temperatures associated with depleted gas reservoirs, the compressibility of CO₂ is very high and has a significant impact on fluid movement (Mathias et al. Reference Mathias, McElwaine and Gluyas2014). However, for CO₂ injection in saline formations, fluid pressures are expected to be hydrostatic or above. Under these conditions, providing a sensible reference pressure is used to determine the fluid properties of CO₂ (i.e. an estimate of pressure towards the end of the injection cycle), the compressibility of CO₂ has been found to have a negligible effect in this context (Mathias et al. Reference Mathias, Gonzalez Martinez de Miguel, Thatcher and Zimmerman2011a ,Reference Mathias, Gluyas, Gonzlez Martnez de Miguel and Hosseini b ).

4.2 No volume change on mixing

Component densities are assumed to be uniform across phases. In fact, the densities of CO₂ and H₂O are both higher in the aqueous phase as compared to in the CO₂-rich phase. For a wide range of different CO₂ injection scenarios, this volume change on mixing is found to lead to an increase in volumetric flow rate of around 0.05 % in Zone 2 and a decrease in volumetric flow rate of around 5 % in Zone 3 (see table 2 of Mathias et al. Reference Mathias, Gluyas, Gonzlez Martnez de Miguel and Hosseini2011b ). See § 2.1 above for an explanation of the zone numbers.

With regards to NaCl, the density of precipitated NaCl, $\unicode[STIX]{x1D70C}_{33}$ , is $2160~\text{kg}~\text{m}^{-3}$ . Using (3.2) in conjunction with the EOS for brine given by Batzle & Wang (Reference Batzle and Wang1992), it can be shown that the density of NaCl dissolved in brine, $\unicode[STIX]{x1D70C}_{32}$ , is around $2800~\text{kg}~\text{m}^{-3}$ . In the above analysis we have set $\unicode[STIX]{x1D70C}_{32}=\unicode[STIX]{x1D70C}_{33}$ such that the model precipitates the correct volume of salt in the dry-out zone. The consequence is that the volume fractions of water and CO₂ in the brine are underestimated by around 2 %.

Figure 3 compares model results from TOUGH2 with those from the similarity solution. TOUGH2 properly incorporates fluid compressibility and volume change on mixing and there is negligible difference between the two models.

4.3 Ignoring gravity effects

As stated earlier, another important assumption is that the vertical permeability of the formation is sufficiently low that gravity effects can be ignored. Extreme changes in density and/or viscosity can lead to instabilities and fingering phenomena, which cannot be represented using one-dimensional models. Indeed, Kim et al. (Reference Kim, Han, Oh, Kim and Kim2012) found that buoyancy driven flow, associated with the different densities of brine and CO₂, played an important part in controlling the spatial distribution of precipitated salt around an injection well. However, this was mostly after the cessation of injection. During the injection phase, gravity segregation within the dry-out zone was much less significant and no viscous fingering was observed.

Mathias et al. (Reference Mathias, Gluyas, Gonzlez Martnez de Miguel and Hosseini2011b ) presented a comparison of simulation results where gravity was accounted for and ignored using TOUGH2 and the MOC solution of Zeidouni et al. (Reference Zeidouni, Pooladi-Darvish and Keith2009), respectively. For a 100 m thick isotropic saline formation, gravity was found to have a strong effect on the leading edge of the CO₂ plume. However, gravity effects were found to be negligible on the dry-out zone development and the associated volume fraction of the precipitated salt. For a 50 m thick isotropic saline formation, gravity effects were found to be negligible throughout.

The dry-out zone is generally unaffected by gravity segregation due to the larger velocities situated close around the injection well, which are mostly horizontal due to the horizontal driving force provided by the injection well boundary (Mathias et al. Reference Mathias, Gluyas, Gonzlez Martnez de Miguel and Hosseini2011b ). From the discussion above it is expected that gravity effects are unlikely to significantly affect the dry-out zone in the 30 m thick saline formations studied in this current article, at least for the lower capillary numbers studied. However, as the capillary numbers are increased, the horizontal injection velocities will become less significant and gravity will play a more important role. However, our analysis has shown that excessive salt precipitation can also develop in the absence of gravity effects due to the counter-current imbibition associated with capillary pressure.