MATHEMATICAL MODELING AND MULTISCALE SIMULATION

OF CO2 STORAGE IN SALINE AQUIFERS

A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF ENERGY
RESOURCES ENGINEERING
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

Marc Andre Hesse

May 2008


c Copyright by Marc Andre Hesse 2008
All Rights Reserved

ii
Abstract

Carbon capture and storage (CCS) is a key technology to reduce CO2 emissions
from industrial processes, in particular from fossil-fuel based electricity generation.
One important aspect of CCS is the safe long-term storage of the captured CO2 in
geological formations, especially in deep regional saline aquifers. Predicting the long-
term evolution of the injected CO2 requires an understanding of the basic physical
mechanisms and the ability to capture them in field-scale numerical simulations.
Simple mathematical models of trapping processes are developed to allow the
identification of the dominant physical processes during CO2 storage and their as-
sociated length and time scales. First-order estimates of the duration of the active
storage period and the migration distance are obtained as a function of the average
properties of the aquifer. These estimates support the selection of storage sites, in
particular at the early stages when limited data is available. They also show that
the length scales associated with the physical processes in regional aquifers can span
several orders of magnitude.
Multiscale simulation techniques are necessary to resolve physical processes and
geological heterogeneity. In particular, robust multiscale methods of elliptic flow
problems, must be developed. The multiscale finite volume method is analyzed in
the context of multipoint flux approximations and shown to lose monotonicity for
anisotropic problems. Strong anisotropy arises in the simulation of CO2 storage,
because of the large aspect ratios of regional aquifers. A new compact coarse operator
and new local fine-scale problems are introduced to obtain monotone coarse pressure
solutions for anisotropic domains. This development presents a major step towards
multiscale simulation of CO2 storage in large regional saline aquifers.

iv
Acknowledgements

I would like to thank my PhD advisors Hamdi Tchelepi and Franklin Orr Jr. who
have provided the best possible intellectual environment for me. The foundation of
this dissertation is their enthusiasm for open enquiry, their appreciation of a good
argument, their willingness to question common assumptions, their patience when
results were not forthcoming, and their trust in my abilities that has allowed me
to explore and develop my own ideas. They have set an example that I hope to
emulate in the future. I am also grateful that they have always encouraged me to
seek collaborations with other scientists, which has lead to a rich and intellectually
diverse graduate experience. Amongst my collaborators I would like to acknowledge
Amir Riaz and Brad Mallison, who have shared their knowledge and experience with
me and who have made significant intellectual contributions to my dissertation. I
would also like to thank Prof. Brian Cantwell for his encouragement and interest in
my work, and Prof. Patrick Jenny who hosted me at ETH-Zürich in the summer of
2008 and who has allowed me to participate in the development of iterative multiscale
methods.
I want to thank Roland Horne, who carefully assigns students to each office, for
my office mates Rami Younis and Felix Kwok. Our office has been a stimulating place
to work that has been central to my graduate experience. I thank Felix and Rami
for their friendship, and I will miss their company. I hope our paths will cross in the
future. I would also like to thank my parents and my family for their support with all
my decisions and their love. Finally, I would like to thank Valentina for all her love
and the happiness she has brought to my life in the last two years of this dissertation.

v
Contents

Abstract iv

Acknowledgements v

1 Introduction 1
1.1 Carbon capture and storage . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Clean coal technology . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Geological CO2 storage . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Saline aquifer storage . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Storage security . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Leakage mechanisms . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.3 Trapping mechanisms . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Thesis overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.1 Trapping models . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.2 Multiscale simulation . . . . . . . . . . . . . . . . . . . . . . . 14

2 Dissolution Trapping & Convection in the Brine 16

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Onset of convection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Model problem . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.2 Previous work . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.3 Linear stability analysis . . . . . . . . . . . . . . . . . . . . . 21
2.2.4 Direct numerical simulation . . . . . . . . . . . . . . . . . . . 33
2.2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

vi
 2.2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3 Convective dissolution rates . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.1 Convection in open systems . . . . . . . . . . . . . . . . . . . 49
2.3.2 Convection in closed systems . . . . . . . . . . . . . . . . . . . 51
2.3.3 Convective time scales . . . . . . . . . . . . . . . . . . . . . . 53
2.3.4 Further work . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3 Residual Trapping & CO2 Gravity Currents 56

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.1.1 Residual saturations . . . . . . . . . . . . . . . . . . . . . . . 58
3.1.2 Gravity currents . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2 Governing equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2.2 Dimensionless form . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2.3 Discretisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.3 Regime transition: confined to unconfined . . . . . . . . . . . . . . . 66
3.3.1 Self-similar solution at early times . . . . . . . . . . . . . . . . 67
3.3.2 Self-similar solution at late times . . . . . . . . . . . . . . . . 75
3.3.3 Non-self-similar transition . . . . . . . . . . . . . . . . . . . . 78
3.3.4 Regime diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.4 Gravity currents with residual trapping . . . . . . . . . . . . . . . . . 86
3.4.1 Parabolic limit: horizontal aquifers . . . . . . . . . . . . . . . 87
3.4.2 Hyperbolic limit: sloping aquifers . . . . . . . . . . . . . . . . 92
3.4.3 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.5.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.5.2 Implications for CCS . . . . . . . . . . . . . . . . . . . . . . . 112
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

4 Multiscale Simulation 119

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.1.1 Numerical challenges . . . . . . . . . . . . . . . . . . . . . . . 119

vii
 4.1.2 Incompressible two-phase flow in porous media . . . . . . . . . 122
4.1.3 Multiscale methods . . . . . . . . . . . . . . . . . . . . . . . . 124
4.1.4 Fine-scale problem . . . . . . . . . . . . . . . . . . . . . . . . 125
4.1.5 Multipoint flux approximations . . . . . . . . . . . . . . . . . 127
4.2 Multiscale finite volume method . . . . . . . . . . . . . . . . . . . . . 129
4.2.1 Coarse operator . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.2.2 Local problems . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.2.3 Fine-scale reconstruction . . . . . . . . . . . . . . . . . . . . . 134
4.2.4 MSFV properties . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.3 Anisotropy and monotonicity . . . . . . . . . . . . . . . . . . . . . . 138
4.3.1 Effective tensor . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.4 Homogeneous coarse operator . . . . . . . . . . . . . . . . . . . . . . 141
4.4.1 Original operator . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.4.2 Compact operator . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.5 Heterogeneous compact operator . . . . . . . . . . . . . . . . . . . . 146
4.6 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.6.1 Transverse equilibrium . . . . . . . . . . . . . . . . . . . . . . 149
4.6.2 Bilinear local problems . . . . . . . . . . . . . . . . . . . . . . 150
4.6.3 Linear local problems . . . . . . . . . . . . . . . . . . . . . . . 152
4.7 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

5 Conclusions 167
5.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.1.1 Dissolution trapping & convection . . . . . . . . . . . . . . . . 168
5.1.2 Residual trapping & gravity currents . . . . . . . . . . . . . . 169
5.1.3 Multiscale methods for porous media . . . . . . . . . . . . . . 170
5.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

viii
List of Figures

1.1 (a) Predicted emission scenarios. The line labeled 500 ppm is the
emissions scenario necessary to stabilize the CO2 concentration at ap-
proximately twice the preindustrial average. (b) Schematic version of
figure a, showing the 7 stabilization wedges. Both graphs are modified
from Pacala & Socolow (2004). . . . . . . . . . . . . . . . . . . . . . 2
1.2 The different states of CO2 in the underground together with the phys-
ical and chemical processes responsible for the transfer between these
states. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 A schematic illustration showing the evolution of the fraction of CO2
in each state as a function of time, modified from Metz et al. (2006). 9
1.4 Sketches of the limiting cases investigated with simple process models:
(a) Dissolution trapping of CO2 ponded in structural trap. (b) Resid-
ual trapping of CO2 migrating in a horizontal aquifer. (c) Residual
trapping of CO2 migrating in a sloping aquifer. In all cases CO2 plume
is dark gray, residual CO2 light gray, and dissolved CO2 shaded. . . . 13

2.1 Supercritical CO2 accumulates along the impermeable top boundary.

It slowly dissolves into the underlying brine, forming a heavier bound-
ary layer. The resulting gravitational instability leads to the convective
transport of CO2 saturated brine plumes. . . . . . . . . . . . . . . . . 18

ix
2.2 Disturbance concentration profile for Ra = 500 and k = 0 at different
times, obtained from, (a) the IVP in the ξ-t coordinate system with
(2.21-2.23) and (b) the IVP in the z-t coordinate system with (2.14-
2.18). Random initial conditions are used for both cases. Plot (a)
shows that the concentration profile resolves into the correct solution
within a very short time, starting from any given set of white noise
initial conditions. For the z-t coordinate system on the other hand,
plot (b), convergence is much slower and is strongly dependent on
initial conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 (a) Growth rate vs. wavenumber curves for Ra = 500 computed by the
dominant mode method and the IVP. The flow is stable for small times.
The growth rate increases with time to become positive at the critical
time tc and the critical wavenumber kc . Flow instability increases with
time beyond tc with both shortwave and longwave cutoffs. Compar-
ison with the IVP shows exact agreement for small times and small
wavenumbers. (b) Comparison of the dominant mode method and the
QSSA with the initial value problem for two perturbation wavenum-
bers at Ra = 500. The dominant mode solution gives exact results for
small times but becomes inaccurate for later times particularly for large
wavenumbers. The QSSA on the other hand is reasonably accurate for
all times. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 (a) Critical time vs. Ra. (b) Critical wavenumber vs. Ra, given by the
dominant mode solution. The critical time varies as the 1/Ra, while
the critical wavenumber scales linearly with Ra. . . . . . . . . . . . . 31
2.5 (a) Maximum growth rate σmax as a function of time, and (b) The
most dangerous wavenumber kmax as a function of time, for different
Rayleigh numbers. All the results are obtained using the QSSA. . . . 32
2.6 (a) The longwave cutoff wavenumber kL as a function of time for dif-
ferent Rayleigh numbers, computed by the dominant mode method.
kL behaves similarly for all Ra and varies approximately linearly with
time. (b) The shortwave cutoff kS computed by the QSSA. . . . . . . 33

x
2.7 Comparison of the growth rates obtained from nonlinear direct numer-
ical simulation and the linear initial value problem, for two different
initial perturbations with wavenumber, k. . . . . . . . . . . . . . . . 35
2.8 Concentration contours at different times for Ra = 4000. (a) t = 1,
(b) t = 1.8, (c) t = 2.3 and (d ) t = 3.8 A large number of fingers
consistent with the linear stability analysis develop initially. Nonlinear
interactions rapidly reduce the number of fingers and give rise to large
scale structures at later times. The fingers at later times are connected
to the top boundary at discrete locations. These connections act as
feeding sites of the high density fluid to the convecting fingers below. 37
2.9 Vorticity contours in the background with overlapping streamlines for
Ra = 4000, t = 2.3. Corresponding concentration contours are shown
in figure 2.8(c). The vorticity field has a dipole structure that drives the
high density fluid through the fingers. The stream lines show how the
fluid travels laterally and then descends through the isolated feeding
sites for the fingers. The direction of fluid circulation is shown by the
arrows. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.10 The dominant mode of the nonlinear simulations as a function of time
for different values of the Rayleigh number. Also shown are the most
dangerous modes given by the linear stability analysis. Good agree-
ment is observed between the two for early times. At later times the
onset of nonlinear behavior leads to a significant deviation from linear
results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.11 Concentration contours for (a) Ra = 1000, t = 2.2 and (b) Ra =
8000, t = 1.6. The larger Rayleigh number case is marked by intense
nonlinear interaction between competing fingers. The small Ra case
shows larger fingers with a strong diffusive spreading. . . . . . . . . . 41

xi
2.12 Concentration contours at time tb , when the concentration front reaches
the bottom. (a) Ra = 1000, tb = 8.9. (b) Ra = 2000, tb = 7. (c)
Ra = 4000, tb = 8.0. (d ) Ra = 8000, tb = 8.5. The late time dy-
namics is governed by large scale fingers. Large Rayleigh number cases
continue to display vigorous fingering interactions. . . . . . . . . . . . 42
2.13 The position of the most advanced portion of the front as a function
of time for different values of the Rayleigh number. The tip moves
faster for larger Rayleigh numbers for early times. Fingering interac-
tions significantly influence the rate of front propagation at later times,
particularly for large Ra cases. . . . . . . . . . . . . . . . . . . . . . . 43
2.14 (a) Variation with permeability of the critical time, tc , the critical
wavelength, λc and the penetration depth at the critical time, δc , for
∆ρ = 5 kg/m3 , φ = 0.3, µ = 0.5 cP and D = 10−9 m2 /s. For this range
of permeability variation tc varies between 2000 yrs and 10 days while
λc goes from about 100 m to less than a meter. The penetration depth
δc gives information regarding the applicability of our analysis with
respect to the layer thickness H, such that δc  H. (b) The advance
of the fastest finger tip is shown until the bottom of the simulation
domain is reached. A diffusive, t0.5 , behavior is observed at early times,
while the fingers advance proportional to t at later times. . . . . . . . 45
2.15 Evolution of the concentration in an open aquifer for Ra = 4000. . . . 49
2.16 (a) Evolution of the dissolution rate at Ra = 4000. Average dissolution
rate over time is shown as dashed line. (b) Time averaged of the
dissolution rate for different Ra. . . . . . . . . . . . . . . . . . . . . . 50
2.17 Evolution of the concentration in a closed aquifer, for Ra = 1000. (a)
Plumes of dissolved have not reached the base of the aquifer. Convec-
tion similar to an open aquifer. (b-d ) Convection in a closed aquifer. . 52
2.18 (a) Evolution of the dissolution rate, dC/dτ , for various Ra in a closed
aquifer. (b) Variation of the time scales discussed in the text as a
function of Ra. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

xii
3.1 (a) Microtomogram of non-wetting CO2 bubbles (black) and water
(white) in the pore space between grains (grey), from Benson et al.
(2006). (b) Comparison between non-wetting phase saturation profile
after imbibition, inferred from a wire-line log (Koperna & Kuuskraa,
2006), and the simplified step function saturation profile corresponding
to the sharp-interface approximation. (c) The geometry of the porous
layer and the variables used in the derivation are shown. . . . . . . . 60
3.2 The geometry of the initial condition and the three associated length
scales H, Ld , and Lf . A particular initial condition h0 [x] and the
corresponding idealized step function initial condition h∗0 [x] are shown.
The arrows indicate that the fluid has been injected over the entire
thickness of the reservoir. . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.3 The evolution of buoyant CO2 -vapor (gray) released into a horizontal
porous layer saturated by brine (white). The outward propagating
tip of the interface is marked by xo . All figures are exaggerated in
the vertical direction, to make the late solution (d ) visible. In many
situations of interest the width of the invaded region in figure (a) is
several times larger than the aquifer thickness. . . . . . . . . . . . . . 68
3.4 Similarity solution is shown for different values of M. The variable θ
is the non-dimensional thickness of the CO2 , it is measured from the
top of the aquifer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.5 The numerical values for the outer (a) and inner (b) tip positions are
shown as a function of M (solid lines). The scaling laws (3.17) and
(3.18) are shown as dashed lines. . . . . . . . . . . . . . . . . . . . . 71
3.6 (a) The numerical solution to (3.2) with initial condition (3.22) and
M = 10 is plotted at various times, t, and compared to the self-similar
solution obtained from (3.15) (solid grey line). (b) Numerical solutions
to (3.2) with M = 20 and initial condition (3.25) is shown for several
values of the parameter a in the initial condition and compared to the
solution of (3.15) with a = 0 (solid grey line). . . . . . . . . . . . . . 73

xiii
3.7 The transition of the numerical solution (light lines) from the early to
the late similarity solution (both heavy lines) is shown in scaled coordi-
nates. (a) The numerical solution is shown at ϑ = ϑb +{1, 4, 10, 30, 100},
where ϑb = 0.61; (b) The numerical solution is shown at ϑ = ϑb +
{101 , 102 , 103 , 104 , 105 }, where ϑb = 7.1. . . . . . . . . . . . . . . . . . 79
3.8 The numerical results for the non-dimensional tip position ξo are shown
as a function of non-dimensional time ϑ, for different mobility ratios
M. In all figures the numerical solution is given by dots (· · · ), the tip
scaling from the early similarity solution by a dashed line (- - -), and
the tip scaling from the late similarity solution as a full line (—). (a)
ϑt = 2.0, ϑb = 0.1; (b) ϑt = 2.5, ϑb = 0.24; (c) ϑt = 29.3, ϑb = 1; (d )
ϑt = 811.3, ϑb = 1; (d ) ϑt = 811.3, ϑb = 7.1. . . . . . . . . . . . . . . 81
3.9 Regime diagram for a finite release of fluid into a horizontal porous
slab, showing the non-dimensional time scales obtained in this study,
and the shapes of the gravity current as a function of the mobility
ratio M. The shaded region indicates the transition period between
the similarity solutions. The characteristic time to dimensionalize all
results is td = L2d κ−1 H −1 . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.10 Current evolution in an aquifer with 30 mD permeability, for details of
the physical properties see § 3.5.2. CO2 currents (dark grey) and the
zone containing residual saturation in their wake. Figures a to d show
the evolution in a horizontal aquifer. These figures are equivalent to
figure 3.3, but they include the effect of residual trapping. Figures e
to h show the evolution in an equivalent aquifer sloping with 5◦ . . . . 86
3.11 (a) Interface shapes for M = 1 and  increasing from 0 to 1 in 0.2
increments. (b) Interface shapes for M = 10 and  increasing from 0
to 1 in 0.2 increments. (c) Contours of the outward propagating tip
position, ζo . (d ) Contours of the inward propagating tip position, -ζi . 88

xiv
3.12 Change of the solution from the early to the late scaling law (M = 1).
(a) The scaling laws for ξo ∝ ϑβ at early and late times as a function
of the residual, . The symbols at  of 0, 1/4, 1/2, and 3/4 show the
exponents obtained by fitting the numerical data shown in figure b. For
the early scaling law the data was fit for ϑ between 0.1 and 10 and for
the late scaling law the data was fit for ϑ > 1000. (b) The shifted tip
position, ξˆ = ξo − ξ,
˜ is shown as function of ϑ, for increasing residuals,
where ξ˜ = 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.13 (a) The flux function f is shown for M = 10 and increasing trapping .
(b) The corresponding solution profiles given by (3.56) in self-similar
coordinates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.14 Shock construction for M = 10. (a) In the absence of trapping,  = 0,
the construction for advancing and receding shocks is the same. (b) In
the presence of trapping,  = 1/2, the shock construction for advancing
shocks is modified. (c) Shock speed as function of ηL , for ηL > ηc the
shock speed depends on the amount of trapping. . . . . . . . . . . . . 97
3.15 Characteristic portraits of the evolution of initial data (3.71), in the
ξτ -plane. The current is bounded by the shock, A to D, in the back
and the leading characteristic of the rarefaction, F to D, in the front.
(a) Analytic solution for M = 1 and  = 0.4. (b) Semi-analytic solution
for M = 5 and  = 0.4. (c) Evolution of the current interface (black)
and the residual surface (grey). . . . . . . . . . . . . . . . . . . . . . 100
3.16 (a) Portrait of the shock path and the leading characteristic of the rar-
efaction, for increasing residual. The locus of the termination points,
τf = τf [ξ↑ ], is shown as a dotted line. (b) The corresponding tem-
poral evolution of the shock strength, ηs . (c) The corresponding final
residual surfaces, ηr = ηr [τf ]. . . . . . . . . . . . . . . . . . . . . . . 102
3.17 Contour plots of the up-dip migration distance, log(ξ↑ ), on the left and
of the total migration time, log(τf ), on the right. The data for several
potential storage aquifers in Alberta, Canada, is shown as circles to
indicate typical parameter values (Bachu & Bennion, 2007). . . . . . 103

xv
3.18 The effect of the governing parameters on the evolution of the current
volume. The passing and final migration times, are shown as circles
and triangles. (a) Increasing the residual, , from 0.1 to 0.9 at constant
M = 1. (b) Increasing the mobility ratio, M, from 1 to 40 at constant
 = 0.4. Circles denote . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.19 This figure shows several comparisons between the numerical solution
(∆ξ = 0.01) to (3.7a) and the hyperbolic solution for Pe = 2 at τ = 50.
(a) Increasing M at constant  = 0. (b) Increasing  at constant M = 10.106
3.20 Evolution of the current volume, V, as a function of diffusive dimen-
sionless time, ϑ, for M = 5 and increasing Pe. The time axis at the
top gives dimensional values corresponding to the parameter choice
discussed in § 3.5.2, assuming cos θ ≈ 1. . . . . . . . . . . . . . . . . . 107
3.21 Time and length scales obtained from the numerical solution of (3.7)
are shown as symbols (∆ξ = 0.01). All results are for a residual of
 = 0.4. The limiting hyperbolic and self-similar solutions are indicated
by full and dashed lines. (a) Down-dip migration distance, ξ↓ . (b)
Transition time, τt . (c) Up-dip migration time, ξ↑ . (d ) Final migration
time, τf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.22 Contour plots of the critical angle, θc , assuming d0 = 2 and the sweep,
S, as function of M and . The data for several potential storage
aquifers in Alberta, Canada, is shown as circles to indicate typical
parameter values (Bachu & Bennion, 2007). . . . . . . . . . . . . . . 115

4.1 An example of a typical regional aquifer under consideration for CO2

storage, the Carrizo Formation in east Texas is shown in grey. The
figure is vertically exaggerated 16:1, so that the true aspect ratio is
much larger than it appears in the figure. The sketch is simplified
from Nicot (2008). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

xvi
4.2 (a) Nine cells of the finite volume grid are shown by dashed lines. The
four dual interaction regions, Ω̃(1) to Ω̃(4) , which connect the shaded
finite volume cell in the center to its eight nearest neighbors are shown
by solid lines. (b) A dual interaction region is shown with the four
coarse cell boundary segments labeled s1 to s4 and the corresponding
fluxes f1 to f4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.3 The top row shows contour plots of the four standard basis functions of
the MSFV method, calulated numerically for a heterogeneous, isotropic
fine permeability distribution. The second row shows the underlying
dual interaction region with the fine grid and the fine fluxes t∗m,k along
the coarse boundaries (small arrows) and the total coarse fluxes tm,k
integrated over each face (big arrows). . . . . . . . . . . . . . . . . . 133
4.4 The fine-scale solution as well as the reconstructed conservative and
the interpolated multiscale solutions are shown for the same problem.
The log of the permeability is shown at the base of each figure. A
unit pressure gradient is applied in the x-direction, and homogeneous
Neumann boundary conditions are applied in the y-directions. (a)
The fine-scale solution, ph . (b) The conservative solution from the
flux reconstruction on the fine grid, p̄h . (c) The coarse solution, pH ,
interpolated onto the fine grid using the MSFV basis functions, p̂h . In
(b) and (c) the coarse solution, pH , is shown as large black dots. . . . 136
4.5 Three solution profiles in the x-direction located at y = 1/2 are shown
for increasing values of the ratio a/b with c = 0. (a) anisotropy ratio,
a/b = 1 > 1/3, corresponding to A = ∆x/∆y = 1 . (b) anisotropy
ratio, a/b = 1/100 < 1/3, corresponding to A = 10. (c) anisotropy
ratio, a/b = 1/2500 < 1/3, corresponding to A = 50. . . . . . . . . . . 139
4.6 Numerical results for the stencil components m1,0 and m1,1 of the
MSFV method are compared to the analytic expressions of the MPFA
O(1) method. The components are shown as a function of the orien-
tation of the tensor, θ, for several ratios of the principal components,
λ1 /λ2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

xvii
4.7 The monotonicity regions of the original and compact coarse operator
are shown in the parameter space a/b vs. |c|/b. A gray grid shows the
corresponding angle, θ, and the ratio of principal components, λ1 /λ2
in the effective permeability tensor given by (4.35). All methods are
monotone above the labeled line. (a) The monotonicity regions of both
the MPFA O(1) method and the coarse operator of the MSFV method
are compared. (b) The monotonicity region of the CMSFV method
equals that of the 7-point stencil. . . . . . . . . . . . . . . . . . . . . 143
4.8 The triangular domains of dependence of the fluxes in a dual interaction
region are shown for both signs of c. The corresponding orientation of
the permeability tensor is indicated by the ellipses. . . . . . . . . . . 145
4.9 The approach of the global fine-scale solution to transverse equilibrium,
as the effective anisotropy a/b increases. The boundary conditions for
this problem are p(x = 0, y) = 1, p(x = 1, y) = 0, and py (x, y = 0, 1) =
0 and the effective permeability on the fine-scale is diagonal, c = 0.
The log of the permeability field is shown at the base. . . . . . . . . . 148
4.10 Comparison between the fine-scale solution ph and the interpolated
pressure field p̂h = φh1 + φh4 in a dual crossing the high permeabil-
ity channel at a/b = 1/1000. (a) Fine-scale solution, ph , in the dual,
rescaled to a unit pressure gradient, is in transverse equilibrium. (b)
Interpolated fine pressure p̂h with reduced boundary conditions. (c)
Interpolated fine pressure p̂h with linear boundary conditions. (d) In-
terpolated fine pressure p̂h with hybrid boundary conditions. . . . . . 151
4.11 (a) The unit square domain Ω, with a 20 × 20 coarse grid. The back
squares indicate coarse blocks with sources fH = ±1. In these blocks
the fine-scale permeability is constant at the mean to give a uniform
distribution of the fine-scale flux. (b) A realization of the permeability
field with small correlation length. Figures c and d show realizations of
the layered permeability fields at angles of 15◦ and −30◦ respectively.
The logarithm of the permeability is shown in all cases. . . . . . . . . 154

xviii
4.12 Comparison of time of flight maps between multiscale methods and the
fine-scale solution for three isotropic but heterogeneous fine-scale per-
meability fields. All solutions are shown at the same injected volume,
corresponding to the breakthrough of the fine-scale solution. The fine-
scale solution, fs, is shown in the first row, a to c, the original MSFV -
red method in the second row, d to f, the CMSFV -lin method in the
third row, g to i, and the CMSFV -hyb method in the fourth row, j
to l. The first column shows solutions for a realization of the perme-
ability field shown in figure 4.11b, columns 2 to 3 show solutions for
realizations of layered permeability fields at 15◦ and -30◦ (figure 4.11c
and d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.13 Comparison of time of flight maps between multiscale methods and the
fine-scale solution for three heterogeneous fine-scale permeability fields
with grid aligned fine-scale anisotropy, a/b = 0.01. All solutions are
shown at the same injected volume, corresponding to the breakthrough
of the fine-scale solution. The figure layout is the same as in figure 4.12.156
4.14 Figures a and b show the mean L2 -errors of the interpolated solution p̂h
and the reconstructed flux field ūh = −k∇p̄h as function of the orien-
tation, θ, of the fine-scale layering for isotropic fine-scale permeability,
a/b = 1. Figures c and d show the same for grid aligned fine-scale
anisotropy a/b = 0.04 (∆x/∆y = 5). All values shown are means over
20 realizations and the standard deviations are shown as vertical bars,
if they are larger than the symbols. . . . . . . . . . . . . . . . . . . . 158
4.15 Figures a and b show the mean L2 -errors of the interpolated solution
p̂h and the reconstructed flux field as function of increasing fine-scale
anisotropy a/b. All values shown are means over 20 realizations of the
permeability field shown in figure 4.11a. . . . . . . . . . . . . . . . . 160

xix
4.16 Results for the quarter 5-spot shown in figure 4.11a with homogeneous
Dirichlet boundary conditions. The first column shows contour maps
of p̂h , the second column shows a horizontal view of p̂h along the y-axis,
and the third column shows streamlines obtained from the conservative
reconstruction p̄h . Figures a to c show the results for MSFV-lin, figures
d to f show results for MSFV-hyb, and figures g to i show results for
CMSFV-hyb. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.17 The L∞ and L2 -norms of M (LH ) are shown as a function of the angle
of the layering. Figures a and b show results for a isotropic fine-scale
permeability field (a/b = 1), while figures c and d are for a fine-scale
anisotropy a/b = 1/100. . . . . . . . . . . . . . . . . . . . . . . . . . 163

xx
Chapter 1

Introduction

1.1 Carbon capture and storage

Carbon Capture and Storage (CCS) involves capturing the CO2 arising from indus-
trial processes and transportation to a suitable storage site, where most of it will
be sequestered from the atmosphere permanently. CO2 capture is more easily ap-
plied to large point sources, such as fossil fuel power plants, fuel processing plants
and plants for the manufacture of iron, steel, cement and bulk chemicals, than to
dispersed emission sources. The transport of large quantities of carbon dioxide from
the source to the storage site will require a large network of pipelines. Currently
carbon dioxide pipelines extend over more than 2500 km in the western US, where
they carry 50 MtCO2 /yr from natural sources to enhanced oil recovery projects in
West Texas and elsewhere (Metz et al., 2006). Several geological storage repositories
have been suggested, including depleted oil and gas reservoirs, unmineable coal beds,
saline aquifers, and deep sea sediments. Gigatonnes of CO2 per year must be stored in
the subsurface to make a significant contribution to the mitigation of climate change
(Metz et al., 2006). The capture and storage of carbon dioxide can be implemented
at a large scale by applying known technology developed for other purposes, that are
is technically feasible today (Metz et al., 2006).

1
CHAPTER 1. INTRODUCTION 2

fossil fuel emissions GtC/yr

fossil fuel emissions GtC/yr

16 (a) 16 (b)
14 14
ual
12 ual 12
s a s us
s us nes
10
nes s a 10 busi CCS
8 busi 500 ppm 8
6 6 500 ppm
4 4
2 2
0 0
2000 2010 2020 2030 2040 2050 2060 2000 2010 2020 2030 2040 2050 2060
year year

Figure 1.1: (a) Predicted emission scenarios. The line labeled 500 ppm is the emis-
sions scenario necessary to stabilize the CO2 concentration at approximately twice the
preindustrial average. (b) Schematic version of figure a, showing the 7 stabilization
wedges. Both graphs are modified from Pacala & Socolow (2004).

1.1.1 Clean coal technology

According to the U.S. Department of Energy, the term clean coal technology describes
a new generation of energy processes that sharply reduce air emissions and other
pollutants from coal-burning power plants.
Coal is a cheap mainstay of electricity generation in both the developed and
developing world, and its use is projected to increase. Coal has the highest carbon
content of all fossil fuels, and its increased use will worsen the climate change problem
(Katzer, 2007). The large-scale implementation of CCS is therefore an important part
of clean coal technology. CCS allows significant reduction in CO2 emissions while
allowing coal to meet future energy needs. However, critics warn that CCS should
not be seen as a silver bullet to reducing emissions, nor should it be researched and
developed at the expense of other environmentally sound, technologically feasible, and
economically affordable solutions to climate change (Union of Concerned Scientists,
2001).
It is hoped that limiting atmospheric CO2 concentrations to less than double
the preindustrial concentration of 280 ppm will prevent the most damaging climate
change. This requires that emissions are held constant at 7 billion tons of carbon per
year (GtC/yr) for the next 50 years, while the current trend will lead to emissions of
CHAPTER 1. INTRODUCTION 3

approximately 14 GtC/yr in 2060 (figure 1.1a). Pacala & Socolow (2004) have intro-
duced the concept of a stabilization wedge (figure 1.1b), as an activity that reduces
emissions to the atmosphere that starts at zero today and increases linearly until it
accounts for 1 GtC/yr of reduced carbon emissions in 50 years. A few examples of
such activities are: efficient vehicles and buildings, substitution of gas baseload power
for coal baseload power, substitution of nuclear, wind or photovoltaic power for coal
power, biomass fuel for fossil fuel, and CO2 capture at baseload power plants, H2
plants or coal-to-synfuel plants. Seven stabilization wedges are necessary to avoid a
doubling of preindustrial atmospheric CO2 concentrations. Currently, no single avail-
able technology can provide even half of the necessary emission reductions. However,
Pacala & Socolow (2004) argue that the necessary reductions in carbon emissions are
already possible with a portfolio of options available today. They have identified 15
activities, including those listed above, which can contribute one stabilization wedge
each.
Assuming that CCS prevents 90% of the fossil carbon from reaching the at-
mosphere, a stabilization wedge could be achieved by the installation of 800 GW
of baseload coal power plants with carbon capture. The associated geological storage
would require that current CO2 injection, for enhanced oil recovery, be scaled up by
a factor of 100 over the next 50 years, or the equivalent of 3500 storage projects
comparable to the current CO2 storage operations at the Sleipner-Vest natural gas
field, operated by Statoil (Torp & Gale, 2004).

1.1.2 Geological CO2 storage

Geological storage of anthropogenic CO2 as a mitigation option for climate change
was first proposed by Marchetti (1977). No further published research was done
until the Norwegian Parliament decided to stabilize Norway’s CO2 emissions at the
1989 levels by the year 2000 and introduced a carbon tax, approximately equal to
50 $/ton CO2 , on CO2 emissions from burning natural gas offshore. In response to
this, Statoil initiated small-scale CO2 -storage tests in 1989 (Kaarstad, 1992). The
options for geological storage where first fully explored during the First International
CHAPTER 1. INTRODUCTION 4

Conference on Carbon Dioxide Removal in Amsterdam in 1992. Storage in saline

aquifers (Kaarstad, 1992; Koide et al., 1992), depleted oil and gas reservoirs (van der
Burgt et al., 1992), as well as a combination of enhanced oil recovery with CO2 storage
(Bondor, 1992; Legg, 1992; van der Meer, 1992) were proposed. The storage of CO2
in deep sea sediments is distinct from aquifer storage on land or on the continental
shelf, due to the neutral buoyancy of the injected CO2 and the potential of a self-
sealing CO2 hydrate cap (Koide et al., 1995, 1997; House et al., 2006). Enhanced
coal bed methane recovery from unminable coal seams also provides a possibility for
geological CO2 storage (Stevens et al., 2001). Some have argued that CO2 can be
stored in basaltic rocks in regions where large sedimentary basins are not common,
such as the Indian subcontinent (McGrail et al., 2006). However, questions remain
about the potential for leakage through fractures that are very common in basalts.
The worldwide storage capacity in depleted oil and gas fields is estimated between
675 and 900 GtCO2 , the capacity of unminable coal seams is estimated between 3 and
200 GtCO2 , and the capacity of saline aquifers is at least 1000 GtCO2 and possibly
up to 10000 GtCO2 (Metz et al., 2006). House et al. (2006) initially estimated that
the economic zone of the US alone has a storage capacity of more than 10000 GtCO2 .
More recently, Levine et al. (2007) showed that the average formation permeabil-
ity and geomechanical considerations may limit the storage capacity substantially.
McGrail et al. (2006) estimate that the Columbia River Flood Basalt Province alone
has a CO2 storage potential greater than 100 GtCO2 . The Deccan Volcanic Province
in India has roughly three times the volume of the Columbia River Flood Basalt
Province, and it may be an important storage option.
CO2 storage combined with enhanced oil recovery or enhanced coal bed methane
recovery is favorable, because hydrocarbon recovery will offset a part or all of the
costs associated with CCS. Unfortunately, the storage capacity of these systems is
limited, and CO2 may have to be transported over long distances from the CO2
sources in urban areas to the storage locations in hydrocarbon provinces. Carbon
storage on a scale large enough to make a significant contribution to the reduction
of CO2 emissions, appears to be viable in saline aquifers. The source–sink matching
is easiest for saline aquifers, because many urban areas are located on sedimentary
CHAPTER 1. INTRODUCTION 5

basins. As of today, there are no economic benefits associated with storage in saline
aquifers that can help offset the storage cost. Therefore, other incentives, such as the
carbon-emissions tax in Norway, must be in place for the adoption of aquifer storage.
Since CO2 emission regulations, taxes, or both, are expected in the near future, saline
aquifer storage is currently the most promising large-scale geological storage option.
Therefore, this thesis focuses on accurate mathematical modeling of the complex flow
behaviors associated with CO2 storage in saline aquifers.

1.2 Saline aquifer storage

The term saline aquifer or saline formation is used to identify deep sedimentary
rocks saturated with formation waters or brines, which contain high concentrations
of dissolved salts. Due to their high salinity, these brines are not used as sources
of drinking water and cannot be used for irrigation. On the contrary, unsustainable
exploitation of shallow groundwater resources may lead to upconing of underlying
brines and to a subsequent deterioration of the water quality (Zhou et al., 2003). In
some locations deep brines are used for low-temperature geothermal power generation
(Lund et al., 2005). In other areas, deep aquifers are used for the injection of drilling
slurries (Reed et al., 2002), hazardous chemical waste (Saripalli et al., 2000), and
liquid radioactive waste (Rumynin et al., 2005). In comparison with the enormous
volumes of saline aquifers available, only a negligible fraction is currently used.
Saline aquifers occur in sedimentary basins throughout the world and are not
restricted to coal, oil, or gas provinces. Holloway & Savage (1993) have argued that
CO2 should be injected at depths greater than 800 m. For a geothermal gradient of
25 ◦ C/km, this is the depth required for CO2 to be in a dense supercritical state,
although this depth may vary between 650 and 1200 m in a particular sedimentary
basin (Bachu & Stewart, 2002).
Carbon dioxide is injected in the supercritical state to decrease the storage volume,
but the long-term storage capacity of an aquifer is determined by the amount of CO2
that can be dissolved into the brine (Bachu & Adams, 2003). Supercritical CO2 is
less dense than the brine under all continental and shallow marine storage conditions.
CHAPTER 1. INTRODUCTION 6

An impermeable seal, such as a shale or clay layer, overlying the storage formation is
therefore always necessary to prevent upward migration of the injected CO2 towards
the surface. To prevent upward migration of the CO2 along an inclined seal towards
the surface a structural trap, (figure 1.4a), is thought to be necessary for long-term
storage security. Based on these and other criteria, Bachu (2003) have proposed a
screening and ranking methodology to identify suitable storage aquifers.
CCS projects targeting saline aquifers are under discussion in many countries,
including Norway, the US, Japan, Canada, Australia, the UK, France, Germany,
Italy, and Poland (Metz et al., 2006). The Sleipner Project, a gas field operated by
Statoil on the Norwegian continental shelf, is a good example of CO2 storage in a
saline aquifer. Approximately 1 MtCO2 /yr is removed from the natural gas stream
produced at the Sleipner-Vest platform and injected into the Utsira formation. The
operation started in October 1996, and over the lifetime of the project a total of 20
MtCO2 is expected to be stored. The evolution of the Sleipner CO2 plume has been
successfully monitored by seismic time-lapse surveys. The surveys also show that the
caprock prevents migration out of the storage formation (Torp & Gale, 2004).

1.2.1 Storage security and CO2 migration

We expect that some fraction of the injected CO2 will remain mobile and buoyant
in the subsurface for several hundreds or even thousands of years. Therefore, the
security of the storage depends strongly on the quality of the geological seal preventing
upward migration. The geological seal in oil and gas fields is intact, because they have
retained buoyant fluids over millions of years. As long as the integrity of the seal has
not been compromised by hydrocarbon extraction, CO2 stored in depleted oil and gas
reservoirs is unlikely leak to the surface. In saline aquifers no such test of the seal
integrity exists, and therefore leakage of CO2 into the shallow subsurface and back
into the atmosphere is a major concern that must be evaluated carefully. On the
other hand, storage security is increased by physical and chemical processes in the
subsurface that either make the CO2 negatively buoyant or immobile. Such trapping
processes progressively reduce the amount of CO2 available for leakage, and increase
CHAPTER 1. INTRODUCTION 7

atmospheric CO2

leakage
geological seal

injection CO2 plume

on re
uti g tra sidual
sol ppin ppi
tra ng
primary
trapping
dissolved CO2 residual CO2
trapping
mineral

solution
trapping
secondary
trapping
mineralized CO2

Figure 1.2: The different states of CO2 in the underground together with the physical
and chemical processes responsible for the transfer between these states.
CHAPTER 1. INTRODUCTION 8

the storage security over time (Metz et al., 2006).

The redistribution of the injected CO2 between different physical and chemical
states is illustrated schematically in figure 1.2. The competition between leakage
and trapping processes determines what fraction of CO2 remains permanently in the
subsurface.

1.2.2 Leakage mechanisms

Leakage of CO2 stored in saline aquifers is a major concern, and the research in this
field is growing rapidly. Given a laterally continuous seal and a structural trap, Metz
et al. (2006) identify the following leakage pathways:

1. Through the pore system in low-permeability caprocks, such as shales, if the

capillary entry pressure at which CO2 may enter the caprock is exceeded.

2. Through openings in the caprock, such as lateral discontinuities or fractures

and faults.

3. Through anthropogenic pathways, such as poorly completed and/or abandoned

pre-existing wells.

Gasda et al. (2004) have identified abandoned wells as one of the most probable leak-
age pathways for CO2 storage projects, due to their high density in many sedimentary
basins, i.e., 350000 wells in the Alberta basin. Nordbotten et al. (2004) have used
semi-analytic solutions to show that leakage through multiple passive wells is not a
simple sum of single well leakage rates, due to leakage induced draw-down around
passive wells. They have also identified that multiple aquifers and aquitards mitigate
leakage into shallow zones, because of loss of CO2 into the intervening aquifers. Nu-
merical comparisons under less restrictive assumptions have shown that the inclusion
of additional processes leads to reductions in leakage rates (Ebigbo et al., 2006).
Pruess (2005) showed that non-isothermal effects are important during leakage
along faults and lead to a self-limiting behavior of the leakage rate. In case CO2
does leak into the shallow subsurface, CO2 will displace the lighter resident soil gas.
Therefore, CO2 concentrations may approach 100% in some parts of the vadose zone,
CHAPTER 1. INTRODUCTION 9

100 leaked CO2

CO2 Reservoir %
natural
trapping

CO2 plume residual CO2 dissolved CO2

engineered
trapping
mineralized CO2
0
1 10 102 103 104 105 106

injection period active storage passive storage

Figure 1.3: A schematic illustration showing the evolution of the fraction of CO2 in
each state as a function of time, modified from Metz et al. (2006).

even for small leakage fluxes (Oldenburg & Unger, 2005). Leakage of CO2 plumes is
similar to migration of plumes of volatile organic compounds, and similar remediation
approaches may be applicable (Zhang et al., 2005).

1.2.3 Trapping mechanisms

Carbon dioxide can leak from the geological storage formation back into the at-
mosphere, because it is mobile and positively buoyant with respect to the brine.
Any process that either immobilizes CO2 , makes it negatively buoyant, or both, is
considered a trapping mechanism. Three dynamic trapping mechanisms have been
recognized in saline aquifers:

1. Residual trapping, which refers to the formation of residual CO2 through capil-
lary snap-off in the wake of a migrating immiscible CO2 plume. Residual CO2 is
still buoyant, but it is immobilized because it is disconnected from the flowing
saturation. Holtz (2002) was among the first to point out the potentially large
CHAPTER 1. INTRODUCTION 10

residual saturations in CO2 –brine systems. Kumar et al. (2005) showed that
the formation of residual saturation may be the dominant trapping mechanism
in saline aquifer storage, and they highlighted the importance of the magnitude
and variation of the residual saturation in the formation as a key petro-physical
property.

2. Dissolution trapping, which refers to the dissolution of CO2 into the brine. Dis-
solved CO2 is still mobile, through the migration of the CO2 -rich brine. How-
ever, the brine density increases with increasing CO2 saturation (Yang & Gu,
2006), and the negative buoyancy prevents leakage of dissolved CO2 (Lindeberg
& Wessel-Berg, 1997). Convective motion induced by dense plumes of CO2 -rich
brine may have the potential to increase the overall dissolution rate of the CO2
significantly (Ennis-King et al., 2005).

3. Mineral trapping, which refers to precipitation of dissolved CO2 as carbonate

or calc-silicate minerals. Gunter et al. (1997) have shown that the CO2 trap-
ping reactions are expected to take hundreds of years to complete. Xu et al.
(2003) showed that the mineral trapping capability of arenaceous formations
after 10,000 years is comparable to CO2 dissolution in pore waters. Johnson
& Nitao (2003) and others have argued that the feedback between the dissolu-
tion/precipitation and the formation permeability may be important.

Figure 1.2 illustrates the relationship between the different CO2 trapping processes
and the states of CO2 in the subsurface. Mineralization offers the hope for perma-
nent storage on geological timescales, but CO2 is effectively trapped by dissolution
and residual trapping much earlier. Therefore, we refer to direct dissolution of the
mobile CO2 plume and residual trapping as primary trapping processes, and to the
dissolution of residual CO2 and the mineralization of dissolved CO2 as secondary
trapping processes. Figure 1.3 is a conceptual sketch of the evolution of the different
states of CO2 over time.
During the injection period, the CO2 plume contacts fresh brine as it is advancing
into the aquifer, leading to some CO2 dissolution. Residual trapping is expected
to be very effective in the period immediately following the end of injection, as the
CHAPTER 1. INTRODUCTION 11

CO2 migrates toward the top of the aquifer (Kumar et al., 2005). It is possible to
enhance trapping during the injection period and for some period after the end of
injection, leading to engineered trapping. Examples of engineered trapping include
water-alternating-gas injection (Kovscek & Cakici, 2005; Juanes et al., 2006; Ide et al.,
2007), and the circulation of brine to increase dissolution (Leonenko & Keith, 2008).
However, engineered trapping at large scales will increase the storage cost and is
therefore unlikely in aquifers, where economic benefits do not offset the increased
cost.
In most situations we have to rely on the trapping of CO2 by natural processes,
in the post-injection period. This reliance on natural trapping processes makes the
proper selection of the storage site very important. This thesis focusses on natural
trapping processes in the post-injection period. In situations where the immiscible
CO2 plume has ponded in a structural trap (figure 1.4a) and has come to rest, no
further residual saturation forms, and dissolution of CO2 into the underlying brine
is the main process reducing the volume of the stationary CO2 plume. As long as
the immiscible CO2 plume continues to migrate, CO2 dissolves into fresh brine and
residual CO2 forms in the wake of the moving CO2 plume (figures 1.4b-c).
We may also distinguish an active storage period, during which mobile buoyant
CO2 is present and leakage is possible, from a passive storage period, where CO2 is
only redistributed between trapped phases (figure 1.3). During the active period,
leakage and trapping processes compete to reduce the mobile buoyant CO2 volume.
The active period ends when all injected CO2 has either leaked or been trapped. The
time when a CO2 storage site switches from the active to the passive period is an
important time scale that provides an indication of the storage security of a particular
site.

1.3 Thesis overview

The concern about leakage of CO2 from a storage site is a dominant theme in the
technical literature and also the main concern of the broader public. An important
part of the portfolio of measures to avoid leakage is to choose the storage site and
CHAPTER 1. INTRODUCTION 12

the injection scheme to accelerate trapping processes. Effective trapping reduces

the time available for leakage, the active storage period, and is therefore crucial to
storage security. Effective trapping also reduces the migration distance of the injected
CO2 , which reduces the likelihood of encountering a leakage pathway. To estimate
the trapping rates, we need to understand the underlying physics of the primary
trapping processes, namely dissolution and residual trapping. We also need to ensure
that numerical models used to predict storage performance are able to resolve these
dynamic flow and transport processes adequately. Therefore, this thesis consists of
two parts:

1. Chapters 2 and 3 present simple mathematical models that aim to capture the
fundamental dynamics of dissolution and residual trapping.

2. Chapter 4 presents multiscale numerical algorithms for the numerical simula-

tions of CO2 storage in large-scale heterogeneous formations.

We develop simple mathematical models of trapping processes to construct a con-

ceptual framework to elucidate the relations among the driving forces, flow charac-
teristics, and the rates of trapping. These models allow us to identify the dominant
processes during CO2 storage and their associated length and time scales. This gives
us first-order estimates of the duration of the active storage period and the migration
distance, and how they depend on the average properties of a storage site. These
simple estimates support the selection of storage sites, in particular at the begin-
ning when limited data is available. Such a conceptual framework also allows us to
understand and evaluate the results of complex numerical models.
The models of CO2 plume migration and trapping show the limitations of current
numerical simulation tools for large-scale geological systems, and have lead us to
the development of new multiscale methods for flow in heterogeneous porous media.
In § 4 we present a robust framework for multiscale discretizations of heterogeneous
elliptic equations, which lie at the heart of any reservoir simulator. This framework
provides the core of a multiscale numerical method for CO2 storage; in the near
future we hope to be able to simulate CO2 storage at the field scale, while resolving
important trapping processes appropriately. We believe that such an approach has to
CHAPTER 1. INTRODUCTION 13

(a) (b) (c)

Figure 1.4: Sketches of the limiting cases investigated with simple process models:
(a) Dissolution trapping of CO2 ponded in structural trap. (b) Residual trapping of
CO2 migrating in a horizontal aquifer. (c) Residual trapping of CO2 migrating in a
sloping aquifer. In all cases CO2 plume is dark gray, residual CO2 light gray, and
dissolved CO2 shaded.

be adopted to obtain reliable predictions of the long-term evolution of CO2 plumes

and the duration of the active storage period.

1.3.1 Process models for CO2 trapping

To understand the two primary trapping processes we consider the three limiting cases
shown in figure 1.4. In § 2 we assume that the injected CO2 has ponded in an anticline.
In this scenario, no residual saturation forms and the volume of the plume decreases
only by dissolution into the underlying brine (figure 1.4a). If the mass transfer of the
dissolved CO2 is diffusive, then the dissolved CO2 will increase proportional to t1/2 . In
this case, the active storage period is very long (>100,000 yrs), and significant leakage
may occur even if the leakage rate is small. However, the unstable density gradient
in the brine may allow for convective mass transport of dissolved CO2 . Assuming the
aquifer is large enough, the dissolved CO2 increases linearly. Rapid dissolution short-
ens the active storage period, and the cumulative leakage will be reduced significantly.
A criterion that determines the dominant mode of mass transport is therefore very
important. In § 2.2 we analyze the linear stability of the diffusive boundary layer,
forming below the interface separating supercritical CO2 from the underlying brine,
and determine the onset of convection as a function of average reservoir parameters.
In § 2.3 we present preliminary results on two-dimensional dissolution rates in the
fully non-linear regime using direct numerical simulations.
CHAPTER 1. INTRODUCTION 14

In § 3 we consider the spreading of CO2 in horizontal and inclined aquifers (fig-

ure 1.4b-c). In both cases, the continued migration of the CO2 plume leads to sig-
nificant residual trapping. We have used a vertical-equilibrium sharp-interface model
to study the migration of CO2 plumes in the post-injection period. In § 3.3 we use
similarity solutions to the governing equations to study the effect of vertical confine-
ment on the migration of CO2 plumes. We show that accounting for the confinement
is important to obtain correct scaling laws for the CO2 plume. In § 3.4 we study
the migration of a CO2 plume in a sloping confined aquifer. We show that the CO2
migration can be approximated by a hyperbolic conservation law, if the aquifer is con-
fined. This has allowed us to model residual trapping semi-analytically, and we have
obtained estimates for the duration of the active storage period and the maximum mi-
gration distance. Comparison of our results for the sloping case with the self-similar
solutions for horizontal aquifers shows that residual trapping is much more efficient
in sloping aquifers.

1.3.2 Multiscale simulation of CO2 storage

In § 3 we show that the CO2 plume migrates like a traveling wave for tens and up
to a few hundred kilometers. Accurate description of the interface separating the
gravity current from the brine is necessary for reliable predictions of the transport
behavior of CO2 plumes, and this poses significant challenges for numerical models
of the large-scale flow systems of interest.
Gravity currents form a narrow tongue that requires resolution at the sub-meter
scale. In addition, we show in § 2 that the convective instability in high-permeability
aquifers may require numerical resolution on the decimeter scale. The length scales
associated with the physical processes during CO2 storage may therefore span as
much as six orders of magnitude in a sloping high-permeability aquifer. The fraction
of the simulation domain that requires these high resolutions is generally small and
may be associated with several features, such as multiple supercritical CO2 plumes
and interfaces between brine and potable ground water.
However, adaptive methods for multiphase flow and transport in heterogeneous
CHAPTER 1. INTRODUCTION 15

porous media are not well developed. This is partly because of upscaling and down-
scaling problems when the mesh size is changed. Recently, multiscale methods that
work on a single coarse grid and incorporate finescale effects through numerical ba-
sis functions have been developed. Multiscale methods offer adaptivity through the
selective updating of basis functions. In § 4 we discuss the multiscale finite volume
(MSFV) method introduced by Jenny et al. (2003). The MSFV method is currently
the only multiscale method for porous media transport that has been extended to in-
clude wells (Lee et al., 2008), compressibility (Lunati & Jenny, 2006; Zhou & Tchelepi,
2008), and gravity (Lunati & Jenny, 2008). However, the original MSFV method is
only robust for uniform grids, which prevents the application to saline aquifers that
typically have very large aspect ratios. In chapter 4 we analyze the monotonicity of
the original MSFV method, and we develop a new Compact Multiscale Finite Volume
(CMSFV) discretization that is robust for large aspect ratios. Once these different
results have been combined into a single multiscale simulator, the MSFV method
should be able to simulate CO2 storage in saline aquifers at dramatically reduced
computational cost.
Chapter 2

Dissolution Trapping &

Convection in the Brine

2.1 Introduction
Dissolution of the supercritical CO2 into the brine is an important primary trapping
process that reduces the volume of the CO2 plume, and it therefore determines the
duration of the active storage period (figure 1.3). Lindeberg & Wessel-Berg (1997)
pointed out that the density of the brine increases with increasing CO2 saturation,
and Yang & Gu (2006) verified this for a CO2 –brine system at reservoir conditions.
Dissolved CO2 is considered trapped, because the CO2 -rich brine migrates downward
and leakage is not likely. The associated reduction in the total volume of the system
also allows the pressure in the aquifer to relax.
At an interface where both the supercritical CO2 and the brine are at rest the
dissolution of CO2 will be slow. In this case, the dissolution rate of the CO2 across
the interface is controlled by molecular diffusion of dissolved CO2 in the brine, and
we expect the dissolution rate of CO2 to decay as t−1/2 . Much more rapid dissolution
of CO2 occurs across an interface where the supercritical CO2 continues to contact
fresh brine, due to advection in either phase. Both the buoyancy driven migration
of the supercritical CO2 as well as convective motion in the brine can lead to rapid
dissolution.

16
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 17

Once the CO2 plume has ponded in an anticline, only convective motion in the
underlying brine can increase the dissolution rate of the CO2 (figure 1.4a). We
assume that the supercritical CO2 is separated from the brine below by a horizontal
interface. Across this horizontal interface, CO2 dissolves into the brine to form a
diffusive boundary layer that grows with time (Lindeberg & Wessel-Berg, 1997). The
CO2 -rich brine in this boundary layer is heavier than the underlying fresh brine. Once
the boundary layer has grown to a sufficient thickness, it will become unstable, so
that fingers of dense CO2 -rich brine propagate downward. In this case the dissolution
rate is determined by convective mass transport in the brine.
We use both a linear stability analysis and direct numerical simulation to study
the onset of this instability. Linear stability analysis is a standard method used to
determine the stability of a particular base-state. In 2.2.3 we study the stability of
the concentration profile of dissolved CO2 in the brine below the interface. If the
concentration profile is unstable to small perturbations, convective motion in the
brine will develop. The advantage of linear analysis is that it can detect very small
unstable wavelengths, because it is not limited by numerical resolution. However,
the linear analysis is valid only at early times when the perturbations are small. In
§ 2.2.4 we validate the direct numerical simulations against the linear theory, and
we use simulations to investigate the nonlinear evolution of the instability at larger
times. These direct numerical simulations are used in § 2.3 to study the long-term
dissolution rate of CO2 in saline aquifers.

2.2 Onset of convection in the brine

2.2.1 Model problem

Numerical simulations of the full two-phase problem by Lindeberg & Bergmo (2003)
have shown that the interface between the supercritical CO2 and brine, remains rel-
atively sharp and is not deflected by the density fingering in the underlying brine,
as shown schematically in figure 2.1. Both Lindeberg & Bergmo (2003) and Ennis-
King & Paterson (2003) showed that local equilibrium between the supercritical CO2
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 18

CO2 gas phase

z=0 g
CO2 dissolved in brine
z

Brine

Figure 2.1: Supercritical CO2 accumulates along the impermeable top boundary. It
slowly dissolves into the underlying brine, forming a heavier boundary layer. The
resulting gravitational instability leads to the convective transport of CO2 saturated
brine plumes.

and the brine can be assumed across the interface. We assume, therefore, that the
supercritical CO2 layer acts as a horizontal upper boundary, with a constant CO2
concentration. These authors propose that the gravitational instability can be stud-
ied in the context of a finite domain, bounded at the top by a constant concentration
boundary, using the Boussinesq approximation. This problem is analogous to ther-
mal convection in a porous medium with insulated boundaries, that is rapidly heated
from below, or cooled from the top, and initially at a fixed temperature. Elder (1967)
showed that the diffusive layer at the boundary becomes unstable only after an ini-
tial period of decaying perturbations. After this critical time, plumes or fingers of
hot fluid rise into the porous medium. This onset time for convection is an impor-
tant timescale that determines the efficiency of dissolution trapping in saline aquifer
storage.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 19

2.2.2 Previous work

The stability of a concentration base profile to small perturbations depends on the
nature of the base-state. For steady base-states, linear stability theory leads to an
algebraic eigenvalue problem (Drazin & Reid, 1981; Nield & Bejan, 1999). The dif-
fusive boundary layer, however, is a time dependent base-state, and linear stability
theory leads to a non-autonomous linear system of ordinary differential equations. A
common approach to such stability problems is the use of frozen time coefficients, the
so-called quasi-steady-state approximation (QSSA). This assumption reduces the sys-
tem of differential equations to a standard algebraic eigenvalue problem (Lick, 1964;
Robinson, 1976). The QSSA is valid if the growth rate of perturbations is large com-
pared to the growth rate of the diffusive boundary layer. Initially, when the boundary
layer grows rapidly, the QSSA is not justified, but it becomes valid for large times
when the base-state changes relatively slowly (Tan & Homsy, 1987; Riaz & Meiburg,
2003a,b). Gresho & Sani (1971) have shown that the QSSA is not valid for the onset
of the instability of a thermal boundary layer in a viscous, incompressible fluid.
In order to avoid using the QSSA for small times, several authors have solved
the initial value problem (IVP) numerically for a large number of random initial
conditions. However, the solution of the initial value problem is sensitive to the
particular initial conditions and is therefore not unique (Foster, 1965). Foster (1965,
1968), Gresho & Sani (1971) and Jhavery & Homsy (1982) solved the IVP for a
thermal boundary layer in a viscous, incompressible fluid. Caltagirone (1980) and
Kaviany (1984) investigated the IVP for a thermal boundary layer in a porous medium
due to a step change in temperature. Ennis-King & Paterson (2003) extended the
analysis of Caltagirone (1980) to anisotropic porous media, and they applied it to
CO2 storage. All of these investigations give a wide range of critical times, depending
on the particular method of measurement of the growth rate.
The stability of time dependent base-states can be investigated by energy methods
(Homsy, 1973; Caltagirone, 1980). Energy methods give a lower bound for the onset
of the instability, but give no information about the growth rate and wavenumber of
the most dangerous disturbance.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 20

For unbounded domains Pritchard (2004) and Ben et al. (2002) argue that a fun-
damental limitation on the accuracy of the solution at short times is the assumption
of decoupled normal modes in the streamwise direction. Since the disturbances are
confined within a narrow diffusive zone, the global Fourier eigenfunctions of the diffu-
sion operator do not provide an optimal basis for streamwise perturbations (Pego &
Weinstein, 1994). Ben et al. (2002) note that the Hermite polynomial based discrete
eigenspectrum of the self-similar diffusion operator is the natural basis for streamwise
perturbations in an unbounded domain. This eigenspectrum is the solution of the
stability problem for the zero wavenumber. The first mode is neutral, it has zero
growth rate, while the rest of the modes decay with time. Therefore, the first mode
becomes dominant after a relatively short time. For small wavenumbers, the per-
turbation dynamics can then be projected onto this dominant mode to increase the
growth rate from zero to positive values.
A similar approach is adopted here for the stability analysis in the semi-infinite
domain. We use the self-similar transform of the diffusion operator in the inhomoge-
neous direction with localized eigenfunctions. The zero wavenumber solution shows
that the first mode of the self-similar diffusion operator decays with time, while the
rest of the spectrum decays more rapidly. We use this dominant first mode to capture
the perturbation dynamics for larger wavenumbers, which shift the growth rate from
negative to positive values at later times. We show that in contrast to the longwave
instability in an unbounded domain, a critical time, critical wavenumber, and a long
wavelength cutoff characterize the stability problem in the semi-infinite domain.
It is important to note that the dominant mode solution becomes inaccurate for
large times and large wavenumbers. For these cases, we use the QSSA in self-similar
coordinates, and we show by comparion with solutions to the IVP that the QSSA
gives good results. Compared to the poor accuracy in the original coordinates, the
success of both QSSA and IVP in self-similar coordinates is due to the localized basis
functions in the streamwise direction.
In order to analyze the long-term evolution of the unstable modes predicted by
the linear stability theory, we carry out high-accuracy direct numerical simulations
(DNS) using a vorticity based formulation (Tan & Homsy, 1988). Voss & Sousa
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 21

(1987) have also carried out numerical simulations of density-driven flows. We use
the methodology developed by Ruith & Meiburg (2000) using Fourier-Galerkin de-
composition in the lateral direction and compact finite differences in the streamwise
direction to solve for the velocity field. A 4th order Runga-Kutta method is used
for time integration. The nonlinear simulations are validated by comparison to the
linear stability analysis. Excellent agreement is observed at early times for growth
rates associated with individual wavenumbers. The long time behavior, however, is
characterized by complex nonlinear interactions that require DNS.

2.2.3 Linear stability analysis

Governing equations

The equations describing the Boussinesq-flow in a horizontal porous layer, where

gravity points in the positive z direction, are:

K
u = − (∇P − ρgẑ) , (2.1)
µ
∂C
φ = −u · ∇C + φD∇2 C , (2.2)
∂t
∇·u = 0 , (2.3)
ρ = ρ0 + ∆ρC . (2.4)

The normalized concentration C of the heavier fluid, u = (u, w) is the Darcy velocity.
K is the permeability, D is the diffusion coefficient, φ is the porosity, and µ is the
viscosity. The unit vector in the direction of the gravitational acceleration is ẑ.
Density ρ is specified as a linear function of concentration, and ρ0 is the density of
the lighter fluid. The initial conditions are u(x, z, t = 0) = 0 and C(x, z, t = 0) = 0,
and the boundary conditions are given by
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 22

w(x, z = 0, t) = 0 , w(x, z = H, t) = 0 ,


∂C


C(x, z = 0, t) = 1 , =0.
∂z
x,z=H,t

We define the following characteristic scales:

K∆ρg
U = , (2.5)
µ
ρ∗ = ρ1 − ρ0 = ∆ρ , (2.6)
µU H
P∗ = = ∆ρgH , (2.7)
K
φH φµH
t∗ = = . (2.8)
U K∆ρg

where H is the domain thickness and U is the buoyancy velocity. The corresponding
dimensionless equations are

∇·u = 0 , (2.9)
u = − (∇P  − Cẑ) , (2.10)
∂C 1 2
= −u · ∇C + ∇C. (2.11)
∂t Ra

A modified pressure P  = ∇(P − ρ0 gẑ/∆ρgH) appears in the above equations. The

Rayleigh number, Ra, is the only dimensionless parameter in the problem, which is
defined as
UH K∆ρgH
Ra = = . (2.12)
φD φDµ
The boundary conditions are given by

w(x, z = 0, t) = 0 w(x, z = 1, t) = 0 ,


∂C


C(x, z = 0, t) = 1 =0.
∂z
x,z=1,t
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 23

We eliminate pressure by taking the curl of (2.10) and substitute the transverse
velocity from the continuity equation. The streamwise velocity, w, and concentration
are decomposed into the base-state and perturbation components

(w, c)(x, z, t) = (wo , Co )(z) + (ŵ, ĉ)(z, t)eikx . (2.13)

The base velocity, wo , is zero, and the base concentration, Co , is assumed constant
in the x-direction and given by the solution of (2.11), with u = 0. The perturbation
variables are decomposed into eigenfunctions, which depend on time and the stream-
wise coordinate, and normal modes in the transverse x-direction with wavenumber k.
The linearized perturbation equations can then be expressed as
 
∂2
2
− k ŵ = −k 2 ĉ ,
2
(2.14)
∂z
 2 
∂ĉ 1 ∂ 2 ∂Co
− 2
− k ĉ = − ŵ , (2.15)
∂t Ra ∂z ∂z

with boundary conditions

ĉ(z = 0, t) = ŵ(z = 0, t) = 0 , (2.16)

∂ĉ


= ŵ(z = 1, t) = 0 . (2.17)
∂z
z=1,t

Equation 2.11 admits a streamwise 1-D, base-state solution, Co , which is

4 1
∞
2 2
Co (t, z) = 1 − sin ((n − 1/2)πz) e−(n−1/2) π t/Ra .
π n=1 2n − 1

The length δ(t) over which C0 is significantly different from zero is the so-called

“penetration depth” of the diffusive boundary layer. For δ ∝ 4t/Ra  1, the
domain can be considered semi-infinite in the positive z-direction, and the base-state
is given by
  
Ra
Co (z, t) = 1 − erf z on z ∈ (0, ∞) (2.18)
4t
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 24

with the boundary conditions

Co (z = 0, t) = 1 , (2.19)
Co (z → ∞, t) → 0 . (2.20)

The problem has now been redefined as the stability of a diffusive boundary layer
in a semi-infinite domain. The parameter range over which the results are valid,

for the original layer geometry in a finite domain, is given by δ ∝ 4t/Ra  1.
Note that the semi-infinite domain does not impose any external length scale on the
problem. An internal length scale in the problem is the time dependent penetration
depth δ(t). A Rayleigh number based on δ(t) is itself time dependent, such that a
critical Rayleigh number, Rac = Ra(δ(tc )), is merely a function of the critical time,
tc , at which the boundary layer becomes unstable. On the other hand, the Rayleigh
number can also be scaled out of the equations by specifying the length scale as a
ratio of diffusion to buoyancy velocity, H = D/U . In this case, the critical time, tc ,
is the only criterion for the onset of instability. For our analysis, we choose to work
with the latter. However, we retain the Rayleigh number, with an arbitrary length
scale H, for convenience in comparing the linear stability analysis with the nonlinear
results discussed in § 2.2.4. In the discussion of the dimensional results (§ 2.2.5), we
show that the imposed length scale cancels out, so that the choice of length scale is
arbitrary.
The perturbation equations (2.14-2.17) can be solved in a straight forward manner
using the quasi-steady-state approximation (QSSA). Earlier investigations have shown
that the QSSA gives accurate results for relatively long times. A fundamental problem
with such an approach is that the concentration eigenfunctions are localized in the
boundary layer, while the eigenfunctions of the operator ∂ 2 /∂z 2 , z ∈ (0, ∞) have
global support. Hence, they do not provide an appropriate basis for streamwise
perturbations (Chang et al., 1998). Therefore, a finite time is required, in the initial-
value-problem (IVP), before the global eigenfunctions can accurately represent the
localized structure of the streamwise perturbations. Since the critical time of this
problem can relatively short, the QSSA in the original coordinate system may not
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 25

able to resolve the early time behavior appropriately.

Localized eigenmodes of the diffusion operator

In a semi-infinite domain, we transform the perturbation equations such that the

eigenfunctions associated with the streamwise diffusion operator are localized around
the base-concentration front. The objective is to achieve considerable improvement in
accuracy at small times, even with the QSSA. Following a coordinate transformation

to the similarity variable of the base-state, ξ = z Ra/4t, the base-state and the
perturbation equations can be expressed as

Co (ξ) = 1 − erf(ξ) , (2.21)

 2

Ra ∂
− k ŵ = −k 2 ĉ ,
2
(2.22)
4t ∂ξ 2
  
∂ĉ 1 1 ∂ 2 ξ ∂ k2t Ra −ξ2
− 2
+ − ĉ = e ŵ , (2.23)
∂t t 4 ∂ξ 2 ∂ξ Ra πt

with boundary conditions

ĉ(ξ = 0, t) = ŵ(ξ = 0, t) = 0 , (2.24)

ĉ(ξ = ∞, t) = ŵ(ξ = ∞, t) = 0 . (2.25)

Note that the self-similarity applies only to the base concentration. The amplitude
and the spatial structure of perturbations are time dependent. The streamwise oper-
ator of the concentration perturbation in the transformed coordinate, ξ, is

1 ∂2 ξ ∂
L= 2
+ , ξ ∈ (0, ∞) . (2.26)
4 ∂ξ 2 ∂ξ

We expand the perturbation concentration as


∞
ĉ(ξ, t) = An (t)φn (ξ) , (2.27)
n=1
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 26

(a) (b)
1
t=10-5 t=0.1
t=10-4 t=0.5
1 t=10-3 t=1.0
0.5

c 0.5 c0
〈

〈
0 -0.5

-1
0 1 2 3 4 5 0 0.2 0.4 0.6 0.8 1
ξ z

Figure 2.2: Disturbance concentration profile for Ra = 500 and k = 0 at different

times, obtained from, (a) the IVP in the ξ-t coordinate system with (2.21-2.23) and
(b) the IVP in the z-t coordinate system with (2.14-2.18). Random initial conditions
are used for both cases. Plot (a) shows that the concentration profile resolves into
the correct solution within a very short time, starting from any given set of white
noise initial conditions. For the z-t coordinate system on the other hand, plot (b),
convergence is much slower and is strongly dependent on initial conditions.

with
2
Lφn = λn φn (ξ) = λn e−ξ Hn (ξ) ; n = 1, 2, 3, ..... (2.28)

The eigenfunctions φn of L are the Hermite polynomials Hn (ξ) in a semi-infinite

2
domain with weight function e−ξ (Robinson, 1976). The associated eigenvalues are
λn = −n for n = 1, 2... , and the dominant mode of the perturbation concentration is
2
φ1 = ξ e−ξ . These eigenfunctions, which are localized around the base-state, provide
an optimal basis for streamwise perturbations in the semi-infinite domain (Pego &
Weinstein, 1994). We note from (2.22) and (2.23) that for k = 0 and using (2.27) the
perturbation amplitude can be expressed as

dAn
= λn An . (2.29)
dt

All modes decay as t−n for k = 0, so that the flow is stable in the longwave limit.
2
The perturbation eigenfunction related to the largest eigenvalue, φ1 = ξ e−ξ , decays
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 27

as 1/t.
We solve the initial-value problem given by (2.21-2.23) to show that the preferred
mode of the streamwise perturbations for k = 0 is φ1 . Figure 2.2(a) shows that the
2
concentration eigenfunction resolves into the dominant mode of L, given by ξ e−ξ , in
a very short time (t = O(1/Ra)) for any given set of white-noise initial conditions.
We show in § 2.2.3 that the time required for convergence to the first eigenfunction is
several orders of magnitude smaller than the critical time for the onset of instability.
Therefore, the IVP in ξ-t coordinates can yield accurate results for short times.
In order to highlight this phenomenon, figure 2.2(b) plots the perturbation eigen-
function in the z-t coordinates, obtained by solving the IVP, given by (2.14-2.18).
2 Ra/4t
Although the white noise initial conditions converge to ĉ ∼ z e−z , the time re-
quired for convergence is several orders of magnitude larger than in the self-similar
coordinates, as shown in figure 2.2(a). Hence, the small-time dynamics in the z-t
coordinate system are obscured during the period it takes for the random initial
perturbations to resolve into the dominant mode. This explains why earlier investi-
gations using the IVP in the z-t coordinate system did not produce accurate results.
For the ξ-t coordinate system, on the other hand, the localized eigenfunctions of L
converge rapidly to the exact solution, thereby giving an accurate growth rate of the
disturbance at small times.

Dominant mode solution

Robinson (1976) used a one-term approximation to the solutions in his QSSA analysis
in z-t coordinates, and he found that the error is small when compared to expansions
using many terms. In the ξ-t coordinates such a one-term approximation is expected
to be even better, because we use an eigenfunctions expansion. We use the leading
order approximation for ĉ from (2.27), substitute it into the IVP, given by (2.23), and
integrate across the domain to obtain

dA1 A1 A1 k 2 Ra −ξ2
=− − +
e ŵ , (2.30)
dt t Ra πt
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 28

where ∞
2
e−ξ ŵ dξ
2

e−ξ ŵ = 0 ∞ . (2.31)
2
ξ e−ξ dξ
0

Again using only the leading order estimate of ĉ to solve for ŵ, we obtain
 
∂2 4tk 2 4tk 2 2
− ŵ = − A1 ξ e−ξ . (2.32)
∂ξ 2 Ra Ra

This can be solved analytically for ŵ to obtain


√  ξ √ 
t 2k t/Ra −2kx t/Ra −x2
ŵ = −A1 k e ξe dx − B1
Ra 0
√  ξ √ 
−2k t/Ra 2kx t/Ra −x2
−e ξe dx − B2 , (2.33)
0

where

   
ξ √ 1 √ k πt k2 t
t/Ra −x2 2
ξ e−2kx dx = − e−2kξ t/Ra − ξ − e erf ξ + k , (2.34)
0 2 2 Ra Ra

and

   
ξ √ 1 √ k πt k2 t
t/Ra −x2 2
ξ e2kx dx = − e2kξ t/Ra − ξ + e erf ξ − k . (2.35)
0 2 2 Ra Ra

The constants B1 and B2 are obtained by satisfying the boundary conditions ŵ = 0

at ξ = 0, ∞. Using this value of ŵ(ξ, t), we can then solve the integral in (2.31)
numerically to obtain the amplitude equation for the dominant mode

dA1
= σ(t; k) A1 , (2.36)
dt

with growth rate σ. Equation 2.36 shows that the perturbations grow exponentially.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 29

2.5
2 t=1
2
t=tc=0.34 k=8π
0
1.5

growth rate
-2
σ t=0.05 1 k=4π
-4
0.5
-6 IVP
IVP 0 QSSA
dominant mode dominant mode
-8
-0.5
0 20 40 60 80 0 1 2 3 4
k t

Figure 2.3: (a) Growth rate vs. wavenumber curves for Ra = 500 computed by
the dominant mode method and the IVP. The flow is stable for small times. The
growth rate increases with time to become positive at the critical time tc and the
critical wavenumber kc . Flow instability increases with time beyond tc with both
shortwave and longwave cutoffs. Comparison with the IVP shows exact agreement for
small times and small wavenumbers. (b) Comparison of the dominant mode method
and the QSSA with the initial value problem for two perturbation wavenumbers at
Ra = 500. The dominant mode solution gives exact results for small times but
becomes inaccurate for later times particularly for large wavenumbers. The QSSA on
the other hand is reasonably accurate for all times.

The growth rate σ(t; k) is obtained without the QSSA and is expected to yield accu-
rate results.
It is interesting to compare the above development for the semi-infinite case with
that of the infinite case analyzed by Ben et al. (2002). The eigenfunctions of the
self-similar operator in the latter case are the full range of Hermite polynomials, The
associated eigenvalues are λn = −n/2. for n = 0, 1, 2... resulting in a neutral mode for
the zeroth eigenvalue. In a semi-infinite domain, on the other hand, the self-similar
operator has eigenfunctions based upon only those Hermite polynomials which satisfy
the boundary condition ĉ(ξ = 0, t) = 0. The associated eigenvalues are −1, −2, ... ;
therefore, a neutral mode is not present.
The presence of a zero eigenvalue in the unbounded case implies a longwave insta-
bility (σ = 0) for k = 0, such that the flow is always unstable for small wavenumbers.
For the semi-infinite domain, on the other hand, the dominant mode decays as t−1 ,
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 30

hence σ = −1/t for k = 0, so that the flow is always stable for small wavenumbers. A
longwave cutoff therefore exists along with a critical time at which the flow becomes
unstable.
The existence of a critical time as a function of the Rayleigh number has been
noted previously (Ennis-King & Paterson, 2003; Ennis-King et al., 2005). However,
its exact value as well as the fundamental mechanism have not been given explicitly.
Consider the growth rate, given by

1 k2 k
σ(t; k) = − − + √ F (t; k) , (2.37)
t Ra π
where F (t; k) is computed numerically from (2.31). Since F (t; k) > 0 the stabilizing
effects come from the first two negative terms on the left hand side. The first term,
−1/t, which is due to the nonzero eigenvalue of the dominant mode of L, insures that
σ < 0 for very small times. In physical terms, the flow can become unstable only
when the perturbations grow at a rate faster than the decay rate of the first mode of
L.
It is important to note that selecting only the first mode to capture the pertur-
bation dynamics cannot be accurate for the entire range of length and time scales
of interest, in view of the absence of a neutral mode of L. We show in § 2.2.3 that
the dominant mode solution gives exact results in comparison with the IVP only for

small values of k t/Ra. Therefore, we use the QSSA for (2.21-2.25), to solve for

the growth rates for larger values of k t/Ra. We also show that when the QSSA is
used with the governing equations in the self-similar coordinates, accurate results are
obtained.

Results

The growth rate vs. wavenumber curves given by both the dominant mode method
and the numerical solution of the IVP are shown in figure 2.3(a) for different times.
High accuracy numerical simulations of the IVP were carried out for each wavenumber
k on a fine computational grid, by using standard methods for 1-D problems. The
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 31

(a) (b)
101 103

critical wavenumber, kc
102
critical time, tc
100

101
-1
10

100

-2
10
101 102 103 104 101 102 103 104
Ra Ra

Figure 2.4: (a) Critical time vs. Ra. (b) Critical wavenumber vs. Ra, given by
the dominant mode solution. The critical time varies as the 1/Ra, while the critical
wavenumber scales linearly with Ra.

evolution of the maximum value of either the concentration or the velocity eigen-
function forms the basis of the growth rate plotted at a particular time for each
wavenumber. Completely stable behavior is indicated by σ < 0 for all wavenumbers
at early time. A critical time, tc , is also shown when the growth rate just becomes
positive at a critical wavenumber kc . At larger times, the stability curve displays a
maximum growth rate at a corresponding most dangerous wavenumber, and a long-
wave and a shortwave cutoff. Comparison of the dominant mode solution with the

IVP results shows exact agreement for all wavenumbers at small times when k t/Ra
is small. For longer times, the growth rate begins to deviate from the IVP result at
larger wavenumbers. However, the critical time and the longwave cutoff are computed
exactly by the dominant mode solution.
Since the dominant mode solution does not give accurate results for large values of

k t/Ra, we use the QSSA in self-similar coordinates to compute the growth rates.
Figure 2.3(b) compares the growth rate as a function of time obtained from the
IVP with the results computed from the dominant mode method and the QSSA, for
two wavenumbers. The dominant mode method again gives exact results for small
times, but deviates from the IVP solution for large times, particularly for the larger
wavenumber. The QSSA, on the other hand, gives reasonably accurate results for
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 32

(a)
(b)
(b) (a)
(b)
1 1
10
250
150 200
10
250
200
125 200
150
100
150 150

σkmaxcutoff, kS

cutoff, kS
100
75
100 100

max
max

σkmax
50 0 500
10 10
shortwave

shortwave
50 50

Ra=500 Ra=500
25 Ra=1000 Ra=500 Ra=1000 Ra=500
Ra=500 Ra=1000 Ra=500 Ra=1000
Ra=2000
Ra=1000 Ra=2000
Ra=1000
Ra=2000 Ra=2000
Ra=2000 Ra=2000
1020
-1
-1
10-1-1
0
20
-1
1020
10
10 10010
10 0
1011
10 10-1-1-1
10
10 10000
10
10 10111
10
10
ttt ttt

Figure 2.5: (a) Maximum growth rate σmax as a function of time, and (b) The most
dangerous wavenumber kmax as a function of time, for different Rayleigh numbers.
All the results are obtained using the QSSA.

all times. This clearly shows that the QSSA in the self-similar coordinates can be
employed to obtain reliable results even for short times.
Figures 2.4(a) and (b) respectively show the critical time and the critical wavenum-
ber as a function of the Rayleigh number computed by the dominant mode method.
The critical time varies as Ra−1 , while the critical wavenumber scales linearly with
Ra. Similar scalings have been obtained by Caltagirone (1980), but our analysis elim-
inates the effect of the initial condition. We obtain tc ≈ 146/Ra and kc ≈ 0.07 Ra for

the critical time and wavenumber. These relationships apply only when 4t/Ra  1.
The maximum growth rate as a function of time and the corresponding most dan-
gerous mode for various Ra, computed by the QSSA, are shown in figures 2.5(a) and
(b) respectively. The maximum growth rate, σmax , increases rapidly at early time
beyond tc , reaches a maximum value and decays approximately as t1/4 at late times.
The most dangerous wavenumber kmax also displays approximately t1/4 scaling.
The evolution of the longwave, kl , and the shortwave, ks , cutoff is plotted in
figure 2.6. The former decays approximately as t1/5 , while the latter decays much
faster as approximately t4/5 at long times.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 33

(a) (b)
(a)
(b)
103 250
10
150
8
200
125
Ra=500 6
100
150

kmaxcutoff, kS
longwave cutoff, kL
Ra=1000
Ra=2000
102 4
75
100

max
50

shortwave
2

σ
50
101
1 Ra=500
25 Ra=1000 Ra=500
Ra=500 Ra=1000
Ra=2000
Ra=1000 Ra=2000
Ra=2000
100 -2 0.5
20
10 10-1 100 101 10-1-1
10
10
0 0
100
1010 10
1
t ttt

Figure 2.6: (a) The longwave cutoff wavenumber kL as a function of time for different
Rayleigh numbers, computed by the dominant mode method. kL behaves similarly
for all Ra and varies approximately linearly with time. (b) The shortwave cutoff kS
computed by the QSSA.

2.2.4 Direct numerical simulation

Numerical method and validation of the numerical solution

Nonlinear fingering dynamics govern the long term flow behavior. We solve the non-
linear problem with a high accuracy, vorticity based method proposed by Ruith &
Meiburg (2000). This method has been employed successfully to obtain highly accu-
rate results for various miscible flow problems in porous media. (Camhi et al., 2000;
Riaz & Meiburg, 2003b, 2004b). High resolution finite-difference simulations have
been reported recently by Otero et al. (2004), for high Rayleigh number convection
in a porous layer of finite depth. They compare the numerical results with analytical
heat flow estimates. We compare our numerical simulations with the results from the
linear stability analysis.
The governing equations used for direct numerical simulations are (2.10) and
(2.11). We use a vorticity formulation for (2.10) to eliminate pressure. By taking
the curl of (2.10) we obtain,

∂C
ω=− = −∇2 ψ , (2.38)
∂x
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 34

where ω is the vorticity and ψ is the streamfunction which is related to velocity as

∂ψ
w = , (2.39)
∂x
∂ψ
u = − . (2.40)
∂z

We assume the vorticity and streamfunction to be periodic in the transverse x-

direction, and for concentration we impose symmetry conditions. The boundary
conditions are therefore given by

∂C
C(z = 0, x, t) = 1 , (z = 1, x, t) = 0 , (2.41)
∂z
∂C ∂C
(z, x = 0, t) = 0 , (z, x = A, t) = 0 , (2.42)
∂x ∂x
w(z = 0, x, t) = 0 , w(z = 1, x, t) = 0 , (2.43)
u(z, x = 0, t) = 0 , u(z, x = A, t) = 0 . (2.44)

The aspect ratio A = L/H, where L is the lateral extent of the computational domain
and H is the layer thickness (figure 2.1). The streamfunction ψ = 0 on all boundaries
while ω = 0 at x = 0 and x = A. Boundary conditions for vorticity at z = 0, 1 are
obtained from (2.38).
We solve the Poisson equation (2.38) by expanding ω and ψ in Fourier modes in
the x-direction. We then solve the resulting ODE for the decoupled z-direction eigen-
functions with 6th order compact finite differences (for details see Riaz & Meiburg,
2003b). The velocities are then computed from (2.39) and (2.40), where the deriva-
tives of ψ are evaluated with 6th order compact finite differences. Time integration
of (2.11) is carried out using a standard 4th order Runga-Kutta method, where all
the spatial derivatives are again evaluated with 6th order compact finite differences.
The resulting numerical scheme resolves all relevant length and time scales accurately.
The initial condition for the concentration is given by (2.18) with a starting time of
t = 0.2. The initial condition for velocity is w = u = 0.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 35

2.5

k=8π
2

1.5

growth rate
1 k=4π

0.5

0 initial value problem

DNS

-0.5
0 1 2 3 4
t

Figure 2.7: Comparison of the growth rates obtained from nonlinear direct numerical
simulation and the linear initial value problem, for two different initial perturbations
with wavenumber, k.

The explicit nature of time integration imposes a strict limit on the time steps.
Although stable time steps are given by the CFL condition, we use even smaller time
steps, on the order of 10−5 , to insure accuracy. Spatial resolution of the computational
grid ranges from 512 × 512 grid points for small Rayleigh number cases to 2048 × 2048
grid points for larger Rayleigh number cases. These fine spatial and temporal reso-
lutions produce converged results. Appropriate grid spacing for different parameter
combinations is obtained by consideration of the cutoff mode provided by the linear
stability analysis. The grid spacing is chosen such that it is smaller than the cutoff
wavelength. Additionally, the divergence of the velocity field is checked throughout
to ensure exact mass conservation for a given grid spacing.
Validation of the numerical simulations was performed by comparing the growth
rates with those obtained from the linear stability analysis. The numerical simulations
are perturbed with pure sinusoidal modes in the transverse direction, superimposed
on the initial concentration profile. The growth rate is measured using the norm of
vorticity, defined as
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 36

1 A
ω(t) = ω(z, x, t) dx dz . (2.45)
0 0

The growth rate for the numerical simulations is then defined as

 
1 ω(t)
σDNS = ln . (2.46)
∆t ω(t − ∆t)
Figure 2.7 shows that the growth rates for the two cases as a function of time for
two wavenumbers, although not exactly equal, are in good agreement. The reason
for this small discrepancy may be related to the 2-D nature of perturbations in the
numerical simulations, which show a slightly higher rate of growth than those in the
1-D initial value problem.

The flow structure for Ra = 4000

We begin our discussion of the nonlinear dynamics by analyzing the concentration

contours for Ra = 4000. An aspect ratio, A = 1 will be used throughout unless
noted otherwise. In order to observe the nonlinear behavior, we perturb the initial
concentration with white noise. The initial disturbances are localized within the
diffusive zone to avoid unphysical conditions of C > 1 or C < 0. The wavelength
selection mechanism is largely independent of the amplitude and the particular values
of the random initial perturbations.
Figure 2.8 presents the concentration contours at different times. This simulation
was carried out with 1024 × 1024 grid points. At an early time of t = 1, figure 2.8(a)
shows a multitude of competing fingers. The number of fingers is about 24, which
is close to 22, predicted by the linear stability analysis. In § 2.2.4 we will precisely
quantify the dominant wavenumber for nonlinear simulations and compare them with
the linear stability results.
Figure 2.8(b) shows the concentration contours at a later time of t = 1.8. The
number of fingers is much smaller than that at t = 1. This increase in the finger
wavelength is primarily due to vigorous nonlinear interactions such as merging and
shielding. Fingering instability in this case is reminiscent of viscous and gravitational
instability of displacement type flows in porous media (Zimmerman & Homsy, 1992;
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 37

Figure 2.8: Concentration contours at different times for Ra = 4000. (a) t = 1, (b)
t = 1.8, (c) t = 2.3 and (d ) t = 3.8 A large number of fingers consistent with the linear
stability analysis develop initially. Nonlinear interactions rapidly reduce the number
of fingers and give rise to large scale structures at later times. The fingers at later
times are connected to the top boundary at discrete locations. These connections act
as feeding sites of the high density fluid to the convecting fingers below.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 38

Figure 2.9: Vorticity contours in the background with overlapping streamlines for
Ra = 4000, t = 2.3. Corresponding concentration contours are shown in figure 2.8(c).
The vorticity field has a dipole structure that drives the high density fluid through
the fingers. The stream lines show how the fluid travels laterally and then descends
through the isolated feeding sites for the fingers. The direction of fluid circulation is
shown by the arrows.

Tchelepi & Orr , 1994; Manickam & Homsy, 1995). The driving force for instability,
i.e. the density gradient, is weakened progressively as the fingers move away from
the top boundary. In addition to the diffusive spreading, nonlinear finger interactions
tend to further smooth out the concentration gradients of a large number of competing
fingers. At a later time of t = 2.3 shown in figure 2.8(c), some of the smaller fingers
disappear due to diffusive smearing, while others merge to form large-scale structures
that develop relatively independently. This trend continues for later times, as shown
in figure 2.8(d ) for t = 3.8. It is interesting to note that even at late times, large-
scale fingers are connected to the diffusive boundary layer at discrete narrow locations,
which serve as feeding sites of high density fluid.
Fingering dynamics are revealed by analysis of the vorticity profile that generates
the rotational flow responsible for driving the fingers. Figure 2.9 plots the vorticity
field contours for Ra = 4000 at t = 2.3. The corresponding concentration field
is shown in Fig, 2.8(c). Superimposed on the vorticity field in figure 2.9 are the
streamlines showing the negative (dashed lines) and positive (solid lines) circulation
paths. The vorticity field displays a dipole structure of negative (light) and positive
(dark) vorticity pairs. Very high vorticity pairs are concentrated at the root of the
fingers, which act as feeding sites for the finger portions away from the top boundary.
The corresponding streamlines show how the fluid is drawn from the sides to flow
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 39

laterally along the top boundary layer and down through the high-vorticity regions
at the finger roots. Although the highest vorticity magnitude occurs at the finger
roots, the tips of the fingers also display a moderate accumulation of vorticity.

Dominant wavenumber of nonlinear flows

As noted above in § 2.2.4, the early time wavelength developed by the nonlinear
simulations is in close agreement with that predicted by the linear stability analysis.
In order to compare the time evolution of the preferred mode of the nonlinear flow,
we define a dominant mode as
K
kE(k) dk
0
n̂ = K
, (2.47)
E(k)dk
0

where n̂ is the dominant wavenumber, k is the Fourier mode and E(k) is the energy
spectrum associated with the Fourier transform of the vorticity field. We compute
E(k) as
 
 x 1
E(k; t) = ω(z, x, t)dz e−ikx dx . (2.48)
0 0

The dominant mode n̂ is an integral measure of the transverse perturbation spec-

trum. We have observed the spectrum to be highly localized around the high energy
modes. The magnitude of n̂ is a reasonable approximation of the high energy modes
in the spectrum. We have also observed that the energy spectrum is only weakly
dependent on the random disturbances introduced as initial conditions. Figure 2.10
shows the evolution of n̂ for various Rayleigh numbers. The most dangerous mode
kmax from the linear stability analysis is also shown for comparison. We observe a
good agreement of n̂ with kmax for short times. The onset of the nonlinear regime leads
to the deviation of the dominant mode away from the most dangerous mode. This
deviation from kmax is stronger and occurs earlier for higher values of Ra, indicating
a stronger influence of nonlinearity for larger Rayleigh numbers.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 40

40

〈
30 DNS (n)
linear stability (kmax)
20

wavenumber
10

Ra=500
1000
2 2000
4000
0.4 2 4 6 8 10
t

Figure 2.10: The dominant mode of the nonlinear simulations as a function of time for
different values of the Rayleigh number. Also shown are the most dangerous modes
given by the linear stability analysis. Good agreement is observed between the two
for early times. At later times the onset of nonlinear behavior leads to a significant
deviation from linear results.

Influence of the Rayleigh number

Figure 2.10 shows that the nonlinear behavior is strongly dependent on the Rayleigh
number. In order to estimate the influence of Ra on the flow dynamics, we com-
pare the relatively early-time behavior at two Rayleigh numbers. Figure 2.11 plots
the concentration contours for Ra = 1000 at t = 2.2 and Ra = 8000 at t = 1.6.
These simulations employ 512 × 512 and 2048 × 2048 grid points, respectively. The
concentration front moves much faster for the latter case. The Ra = 1000 case de-
velops a few large fingers and also shows a substantial amount of diffusive smearing
in other regions of the concentration front. The large-scale fingers appear to move
independently, without interacting with neighboring fingers. The Ra = 8000 case,
on the other hand, develops a vigorous instability that results in complex fingering
structures. Compared to the Ra = 4000 case shown in figure 2.8(b) at t = 0.18, the
Ra = 8000 case already shows well developed discrete feeding sites. Note that new
feeding sites develop as the old ones are abandoned. In other regions, multiple fingers
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 41

Figure 2.11: Concentration contours for (a) Ra = 1000, t = 2.2 and (b) Ra = 8000,
t = 1.6. The larger Rayleigh number case is marked by intense nonlinear interaction
between competing fingers. The small Ra case shows larger fingers with a strong
diffusive spreading.

can attach to a single feeding site, nonlinear interactions then select finger one over
the others as the preferential flow path.
The complexity of the fingering structures suggest that the long term fate of the
nonlinear competition cannot be predicted from the concentration profiles at earlier
times, especially for large Rayleigh numbers. In order to analyze the fingering dynam-
ics at long times, we plot the concentration profiles in figure 2.12 for various Rayleigh
numbers at times when the concentration front reaches the bottom boundary. We
designate this time as tb . The small Rayleigh case, Ra = 1000 at time tb = 8.9,
shows that two large fingers survive to reach the bottom boundary. Comparison with
figure 2.11(a) shows that the fingering configuration is completely different from that
at an earlier time of t = 2.2. The Ra = 2000 case at tb = 7 displays slightly narrower
fingers, with one isolated finger in the middle making it to the bottom boundary,
while two other fingers, attached to a single feeding site, are still in competition. For
larger Rayleigh number cases, Ra = 4000, tb = 8 and Ra = 8000, tb = 8.5, more
fingers reach the bottom boundary. Many of these fingers can be observed to undergo
strong interactions while others are in the process of fading out. Figures 2.12(c) and
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 42

Figure 2.12: Concentration contours at time tb , when the concentration front reaches
the bottom. (a) Ra = 1000, tb = 8.9. (b) Ra = 2000, tb = 7. (c) Ra = 4000, tb = 8.0.
(d ) Ra = 8000, tb = 8.5. The late time dynamics is governed by large scale fingers.
Large Rayleigh number cases continue to display vigorous fingering interactions.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 43

0
10

tip position
10-1

Ra=500
Ra=2000
Ra=8000
10-2 -1
10 100 101
t

Figure 2.13: The position of the most advanced portion of the front as a function
of time for different values of the Rayleigh number. The tip moves faster for larger
Rayleigh numbers for early times. Fingering interactions significantly influence the
rate of front propagation at later times, particularly for large Ra cases.

2.12(d) clearly show the fading of the fingers is due to shifts in the feeding sites as well
as due to lateral pinch-off, where one diagonally moving finger cuts the fluid supply
of a neighboring finger. An interesting consequence of these fingering interactions is
an increase in tb for larger values of the Rayleigh number, as compared to the small
Ra cases. Note that tb = 8.9 for Ra = 1000, it decreases to 7 for Ra = 2000 but then
again increases to 8 for Ra = 4000 and then to 8.5 for Ra = 8000.
The position of the most advanced section of the concentration front, the tip
position, as a function of time is shown in figure 2.13 for various Rayleigh numbers.
For the small Ra = 500 case, the front initially propagates as t1/2 and then switches to
a linear growth for larger times. Higher Rayleigh number cases display the diffusive
t1/2 behavior for relatively shorter times. The Ra = 2000 case displays a faster
than linear growth of the tip position at later times. The tip travels quickly for the
Ra = 8000 case, but for a very small initial time. It is clear that by t = 5 the tip is
moving faster for the Ra = 2000 case as compared the Ra = 8000 case. Consequently
the time for a finger to reach the bottom boundary is shorter for the Ra = 2000 case.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 44

As shown by the concentration contours in figure 2.12 this behavior is due to more
intense finger interaction at higher Rayleigh number cases, such that the interacting
fingers do not allow any one finger to clearly breakaway ahead of the front. For the
smaller Ra cases, on the other hand, isolated fingers travel faster due to the absence
of interference from neighboring fingers.

2.2.5 Discussion
We analyze the stability of a diffusive boundary layer in a semi-infinite domain. This
analysis is applicable when the penetration depth of the diffusive boundary layer, δ,
is small relative to the domain thickness, H. The penetration depth at the onset
of instability is given in dimensional form as δc ≈ 24µD/(K∆ρg). We must have
δc  H for the assumption of the semi-infinite domain to be valid. The dimensional
critical time, which is then independent of the length scale, is given by

φµ2 D
tc = 146 . (2.49)
(K∆ρg)2

Similarly, we define the dimensional critical wavelength λc as

2πµD
λc = . (2.50)
0.07K∆ρg

To validate both the linear analysis and the numerical simulations, the predictions
of tc and λc should be compared with experimental observations. Unfortunately
Elder (1968) only states the Rayleigh number of his experiments and does not give
all the data necessary to calculate tc and λc . Green & Foster (1975) have reported
experiments of a salt solution diffusing into the top of a Hele Shaw cell. The Rayleigh
number in their experiment is Ra = KρgH/(µD) = 90500. They do not report the
onset time, but they give the wavelength of the first observed fingers as λ = 1.8 mm
and note that this may be an over estimate. For the same parameters, our linear
stability analysis predicts a critical time of tc = 4 s and a critical wavelength of
λc = 0.6 mm. The critical wavelength predicted by linear theory is generally smaller
than the wavelength first observed, because the fingers are initially not visible in
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 45

103 tc

0
λc

1.
tc (yrs), λc (m), δc (m)
2
10 102

≈
pe
slo
depth (m)
101
δc
0 .5
10 ≈0
101 pe
slo

10-1

10-2
1 2 3 4 100 1 2 3 4
10 10 10 10 10 10 10 10
k (mD) time (yrs)

Figure 2.14: (a) Variation with permeability of the critical time, tc , the critical wave-
length, λc and the penetration depth at the critical time, δc , for ∆ρ = 5 kg/m3 ,
φ = 0.3, µ = 0.5 cP and D = 10−9 m2 /s. For this range of permeability variation
tc varies between 2000 yrs and 10 days while λc goes from about 100 m to less than
a meter. The penetration depth δc gives information regarding the applicability of
our analysis with respect to the layer thickness H, such that δc  H. (b) The ad-
vance of the fastest finger tip is shown until the bottom of the simulation domain
is reached. A diffusive, t0.5 , behavior is observed at early times, while the fingers
advance proportional to t at later times.

the experiments. By the time they become observable, they have already coarsened.
Given this experimental limitation on the detection of the critical parameters, high
resolution numerical studies as reported above are valuable because the growth of
perturbations can be detected before they become visible.
The critical time and the critical wavelength can vary by orders of magnitude
depending on the properties of the geological formation. While viscosity, diffusion and
density difference have more or less similar values for typical aquifers, permeability
can vary over a large range of values, and therefore introduces the largest variations
in tc and λc . Figure 2.14(a) shows that δc decreases from 55 m to about 0.07 m when
the permeability increases from 1 mD to 3 D. Hence our results for the semi-infinite
domain will apply to high permeability layers with a thickness of few tens of meters,
but for low permeability formations the thickness has to be several hundred meters
for our results to be applicable. The critical wavelength decreases from λc ≈ 200 m
to λc ≈ 0.3 m and the critical time for the onset of the instability decreases from
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 46

tc ≈ 2000 yrs to tc < 10 days as the permeability decreases from 3 D to 1 mD.

The critical wavelength, λc , is an indication of the length scales that have to be re-
solved by numerical simulations to capture the convective transport and the resulting
solution trapping. In high-permeability aquifers the initial unstable perturbations
are much smaller than the size of grid blocks typically used in field-scale reservoir
simulations. A numerical study of the high permeability Sleipner injection site by
Lindeberg & Bergmo (2003) shows that a grid block size of 3 × 4 cm is necessary to
capture the initial fingering with a standard reservoir simulator. For field scale simu-
lation the grid size has to be increased to 100 × 100 m. Lindeberg & Bergmo (2003)
show that simulations using such coarse grids delay the time of onset to 100 yrs, so
that the onset of convection is delayed by 2 orders of magnitude, compared to the
fine grid. Such increases in the onset time may, or may not, be acceptable depending
on the time scale of storage. More importantly, the failure to resolve the unstable
length scales at small times, may lead to significant errors in the nonlinear regime,
even if the dominant length scales at long times are coarse enough to be resolved on
the chosen coarse grid. Diersch & Kolditz (2002) have investigated the convergence
of the Elder problem. They show that even at late times, when most of the structures
are large enough to be resolved on a coarse grid, the solution is sensitive to the grid
size.
Following the onset of instability, the strength of convective mixing can be de-
termined qualitatively from the speed of the most advanced finger tip, shown in
figure 2.14(b) for several values of the permeability. All other properties are held
fixed at the same values as in figure 2.14(a). For all permeability values, the position
of the finger tips is initially proportional to t1/2 . Eventually, however, they acceler-
ate and the penetration is approximately proportional to t. Note that the late time
rate of advance is similar for all values of permeability. Larger permeability values
simply lead to an earlier switch from t1/2 to the linear regime. In high permeability
formations, solution trapping is therefore strongly enhanced by convection, and it
may be the dominant mechanism reducing the mass of the CO2 plume. To assess the
safety and effectiveness of CO2 storage in high-permeability formations, it is therefore
necessary to simulate this convective process accurately.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 47

Heterogeneity is important for the miscible, viscous fingering instability (Tchelepi

& Orr , 1994; Riaz & Meiburg, 2004b; Prassad & Simmons, 2003). Some attempts
have been made to model the effect of heterogeneity on the mechanism of this in-
stability (Gounot & Caltagirone, 1989), but this remains one of the directions of
future research. The fine stratigraphic layering of sedimentary rocks causes strong
anisotropy of the permeability. In general, the vertical permeability may be several
orders of magnitude smaller than that in the horizontal direction. Ennis-King & Pa-
terson (2003) have shown that anisotropy has an equally strong effect on the critical
time and wavelength as the magnitude of the permeability. The analysis developed
here can easily be extended to include the effects of anisotropy.
Our analysis does not include the full dispersion tensor, which accounts for the
influence of velocity induced dispersion (Yortsos & Zeybek, 1988; Tchelepi et al.,
1993). Recent work by Riaz & Meiburg (2004a) suggests that for homogeneous porous
media, velocity induced dispersion is generally equivalent to a slight increase in the
level of molecular diffusion.

2.2.6 Conclusions
Accurate numerical simulation of density fingering over long times at the field scale
is one of the main challenges in predicting the movement of CO2 underground. Our
theoretical and numerical results are valuable for understanding density driven con-
vection during CO2 storage in saline aquifers. Our analysis is directly applicable to
the thermal instability problem and resolves the issue of the critical time and the
critical wavenumber. We also highlight the physical mechanisms of instability that
give rise to both the long wavenumber cutoff and the critical time for the onset, in a
semi-infinite domain.
The disturbances only become experimentally observable a finite period after the
critical time, and it is therefore difficult to measure the critical time with reasonable
accuracy. For typical aquifers with moderate permeability, the onset time can be as
large as hundreds of years. The prediction of the critical wavelength is crucial in
choosing the grid resolution in the numerical simulation. Although the large scale
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 48

fingers at later times can be resolved with fewer grid blocks, the errors introduced by
ignoring the small-scale dynamics at early times, may have a significant influence on
the late time behavior.
The simplifications necessary to treat the problem theoretically may introduce
significant errors when such results are applied to real aquifers. The important as-
sumptions are the homogeneity and isotropy of the porous medium, as well as the
assumption of single-phase flow and the absence of velocity induced dispersion. The
assumption of single-phase flow will break down if capillary forces are significant.
Additionally, physical mechanisms related to dissolution, precipitation and geochem-
ical reactions, which are not accounted for in our analysis, can be expected to play
a role. Some of these processes may be incorporated in the linear stability analysis,
while others can only be investigated with high-resolution nonlinear simulation. The
results for the basic gravitational fingering instability presented here, form the basis
to investigate the effect and importance of these additional processes.

2.3 Convective dissolution rates

Once the onset time for convection in an aquifer has been determined, the evolution
of the convective dissolution rate determines the time scale necessary to dissolve the
injected CO2 . To obtain estimates of the convective dissolution rate, we use direct
numerical simulations to study the long-term behavior of the convective mass trans-
port in open and closed systems. In a closed system, the background concentration of
dissolved CO2 in the domain increases and convection decays as the domain is satu-
rated. In an open system, the domain is infinite so that the background concentration
of dissolved CO2 remains effectively zero, and convection continues as long as CO2
is supplied. Regional saline aquifers are often very large and are likely to behave as
open systems, so that dissolution may be a very effective trapping process.
All simulations reported below are solutions to the two-dimensional model prob-
lem introduced in § 2.2.1. All results are non-dimensionalized using the characteristic
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 49

(a) ¿ = 7.3 (b) ¿ = 14.1 (c) ¿ = 22.6 (d) ¿ = 29 C

0 0 0 0 1.0

0.9

0.8
1 1 1 1
0.7

0.6

z 2 2 2 2 0.5

0.4

0.3
3 3 3 3
0.2

0.1

4 4 4 4 0.0
0 1 0 1 0 1 0 1
x x x x

Figure 2.15: Evolution of the concentration in an open aquifer for Ra = 4000.

scales (2.5) to (2.8), and the concentration of CO2 in the brine, C = c/ceq , is normal-
ized by the equilibrium concentration at the CO2 -brine interface, ceq . The concentra-
tion is given by c = xc ρc , where xc is the mass-fraction of CO2 in the brine and ρc is
the density of the dissolved CO2 . As before, the governing parameter is the Rayleigh
number, Ra = k∆ρgl/φµD, where l is a suitable length scale specified below.

2.3.1 Convection in open systems

We consider an open aquifer of width, L, and infinite depth. In this case the appro-
priate length scale, l, is the width of the domain, L. For the numerical simulations
we choose H L, and we report results only for the time during which the plumes
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 50

10-2
3 2.5

dC dC 1.5
dτ 2 dτ
1

0.5

1 0
1 10 20 30 40 50 0 1000 2000 3000 4000
τ Ra

Figure 2.16: (a) Evolution of the dissolution rate at Ra = 4000. Average dissolution
rate over time is shown as dashed line. (b) Time averaged of the dissolution rate for
different Ra.

of dissolved CO2 have not reached the bottom of the domain. Figure 2.15 shows the
evolution of gravity driven fingers, or plumes, of dense CO2 -rich brine in a typical
simulation. The nonlinear interactions of the fingers have already been described in
§ 2.2.4-2.2.4, and here we focus on the evolution of the dissolution rate, dC/dτ . We
define the dissolution rate as the rate of change of the total dissolved CO2 in the
domain, given by
L H
dC d
≡ Cdx dz. (2.51)
dτ dτ 0 0

Figure 2.16(a) shows the evolution of dC/dτ , for a typical simulation at Ra = 4000.
This evolution can be divided into three stages. For τ < 1 the diffusive boundary
layer is stable and the dissolution rate decays as, dC/dτ ∝ τ −1/2 . After the onset of
convection the dissolution rate fluctuates, and the peaks correspond to the growth of
large fingers. After an initial transient, 1 < τ < 10, the dissolution rate fluctuates
around a constant value of dC/dτ ≈ 0.017. We refer to this average value as the
open-system dissolution rate,
dC/dτ . Figure 2.16(b) shows that
dC/dτ  is nearly
independent of Ra, because Ra can be scaled out of the problem in a domain of
infinite depth, see also the discussion in § 2.2.3.
This open-system dissolution rate allows simple estimates of the time necessary
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 51

to dissolve a given mass of CO2 injected into a saline aquifer based on its average
petrophysical properties. The mass dissolution rate of CO2 per unit width in the
open convection regime is given by

dmc dC k∆ρgceq
≈ , (2.52)
dt dτ φµ

where mc is the total mass of dissolved CO2 . The onset time, τo , of the convective
mass transport is given by the first minimum in the evolution of the dissolution rate
(figure 2.16a, see also figure 2.18a). For the range of Ra investigated (see figure 2.18b),
it is given by
φµ11/5 D6/5
to = 6215 , (2.53)
(k∆ρg)11/5 H 1/5
as computed from DNS. This onset time, to , decreases with Ra and it is larger than
the theoretically predicted critical time, given by (2.49), obtained from linear theory
in § 2.2.3.

2.3.2 Convection in closed systems

If the aquifer is of finite thickness, the suitable length scale, l, is the aquifer thickness,
H. Figure 2.17 shows how the nature of the concentration field changes after the
fingers have started to interact with the bottom boundary. After the fingers reach
the bottom the convection cells coarsen rapidly until the size of a convective cell is
approximately H. Figure 2.18(a) shows the evolution of the CO2 dissolution rate
for various values of Ra. The early phase of the evolution is similar to the semi-
infinite aquifer. An initial phase of diffusive mass transport is followed by open-
system convective mass transport after the onset time. However, after the fingers
have reached the bottom of the aquifer, the dissolution rate decays very rapidly. We
refer to this late stage as closed-system convective mass transfer, to distinguish it
from the earlier vigorous regime. The time when the plumes start to interact with
the bottom boundary, τi , is independent of Ra (figure 2.18a). The dimensional form
is given by
φµH
ti ≈ 15 . (2.54)
k∆ρg
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 52

(a) ¿ = 7.4 C
0 1.0
0.8

z 0.6
0.4
0.2
1 0.0
0 1 2 3 4
(b) ¿ = 22.1 C 1.0
0
0.8

z 0.6
0.4
0.2
1 0.0
0 1 2 3 4
(c) ¿ = 36.7 C
0 1.0
0.8
0.6
z
0.4
0.2
1 0.0
0 1 2 3 4
(d) ¿ = 124.7 C
0 1.0
0.8

z 0.6
0.4
0.2
1 0.0
0 1 2 3 4
x

Figure 2.17: Evolution of the concentration in a closed aquifer, for Ra = 1000. (a)
Plumes of dissolved have not reached the base of the aquifer. Convection similar to
an open aquifer. (b-d ) Convection in a closed aquifer.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 53

10-1 102
τi (a) (b)

101
10-2
dC τ0 τ
dτ 100
DNS:
10-3 τc
10 -1
τ0
τi
τc (theory)
10-4 10-2
10-1 100 101 102 102 103 104
τ Ra

Figure 2.18: (a) Evolution of the dissolution rate, dC/dτ , for various Ra in a closed
aquifer. (b) Variation of the time scales discussed in the text as a function of Ra.

At t ≈ 15ti the aquifer is saturated to 95%, and convective motion is negligible. If

this stage is reached in a CO2 storage project, the size of the chosen aquifer is too
small to dissolve all the injected CO2 into the formation brine. In general, we expect
the storage aquifer to be large enough to dissolve several times the volume of the
injected CO2 .

2.3.3 Convective time scales

Using linear theory and high accuracy numerical simulation, we identify three con-
vective time scales:

1. The critical time, tc ∝ k −2 , is the time after which small perturbations grow, it
gives an analytic lower bound for the onset of convection.

2. The onset time to ∝ k −2.2 , marks the transition from diffusive to convective
mass transfer, and corresponds to the appearance of the first fingers in the
concentration profile. For large Ra, we expect to to approach tc from above.

3. The interaction time, ti ∝ H/k, marks the transition to a regime of decaying

convection, and corresponds to the first fingers interacting with the bottom of
the domain.
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 54

The first two time scales are independent of the thickness of the aquifer, because
they are determined by the growth of the boundary layer, which we assume to be
much smaller than H. The interaction time, ti , is of course directly proportional to
H. These time scales identify three dynamic regimes of mass transport in the brine,
shown in figure 2.18(b). These three regimes may be characterized as:

1. In the diffusive regime (t < to ) the concentration of dissolved CO2 in the aquifer
increases proportional to t1/2 , and the dissolution rate of CO2 decreases rapidly
as t1/2 . In the absence of the gravitational instability this would be the only
regime.

2. In the open-system regime (to < t < ti ) the dynamics are independent of the
domain size. In this regime the total concentration increases linearly with t,
and the open-system dissolution rate is given by
dC/dτ  ≈ 0.017. This is the
most effective regime of mass transfer (gray triangle in figure 2.18b), and is
characterized by rapidly growing fingers with strong nonlinear interactions.

3. The closed-system regime (t > ti ), where the finite size of the aquifer influences
the dynamics of convection. This regime is characterized by large, slow convec-
tive cells. The dissolution rate decays rapidly and dissolved CO2 approaches its
equilibrium concentration throughout the aquifer.

Consider an aquifer with the following properties: H = 200 m, µ = 5 · 10−4 Pa s,

∆ρ = 5 kg/m3 , and D = 10−9 m2 s−1 . We illustrate the effect of permeability variation
by considering a high-permeability aquifer with, porosity φ = 0.3 and permeability
k = 3400 mD, and a low-permeability aquifer, with φ = 0.1 and k = 10 mD. In
the high-permeability aquifer, the critical time is 4 days, the onset of convection
enhanced dissolution is after 20 days, the plumes reach the base of the aquifer after
approximately 80 yrs, and the aquifer would be saturated after approximately 1200
yrs. In the low-permeability aquifer, the critical time is 450 yrs, the onset of convective
dissolution is not until 680 yrs, the plumes reach the base of the aquifer after 9510 yrs,
and the aquifer would be saturated after 150000 yrs. This simple example illustrates
that the time scales of dissolution trapping are strongly dependent on the permeability
CHAPTER 2. DISSOLUTION TRAPPING & CONVECTION IN THE BRINE 55

of the aquifer. In general, the dissolution of CO2 into the brine will be greatly
enhanced by density driven convection in large high-permeability aquifers, because
the onset time is short, the dissolution rate in the open-system regime is high, and
the transition to finite-acting convection is late.

2.3.4 Further work

The results on the dissolution rate presented above are preliminary; in the future we
plan to extend this work to three-dimensions and to heterogeneous domains. These
results would allow simple order-of-magnitude estimates of the time necessary to
dissolve a given mass of CO2 injected, through a given interfacial area. Unless the
supercritical CO2 has ponded in a structural trap, the area of the interface changes as
the supercritical CO2 migrates due to buoyancy. In this case, it would be instructive
to combine the dissolution rates obtained from high accuracy simulations with the
simple models of gravity currents discussed in § 3.
Chapter 3

Residual Trapping &

CO2 Gravity Currents

3.1 Introduction
The formation of residual saturations is the second primary trapping process that
reduces the volume of mobile CO2 , and therefore determines the duration of the active
storage period (figure 1.3). When the saturation of the non-wetting CO2 decreases,
a process referred to as imbibition, an interplay of viscous and capillary forces at
the pore-scale leads to the formation of disconnected essentially-immobile bubbles
of residual CO2 (3.1a). Imbibition takes place in the wake of the migrating CO2
plume, and residual trapping continues as long as the CO2 migrates. To understand
residual trapping it is therefore essential to understand the buoyancy driven migration
of the CO2 plume. In the two aquifers shown in figure 1.4(b) and (c) the CO2
plume migrates as a gravity current and residual trapping is likely to be an important
trapping process. In § 3.1.1 we give a brief introduction to the pore-scale phenomenon,
and we introduce a nomenclature suitable for the simple model we adopt. A discussion
of the current research on the formation of residual saturations is beyond the scope
of this thesis. Instead we adopt the simplest trapping model, and we focus on the
interplay between sweep and residual trapping, which has so far not received much
attention in CO2 storage.

56
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 57

After the end of injection the advective forces are small compared to gravitational
and capillary forces, and we study the migration of the CO2 using a simplified model
assuming gravity-capillary equilibrium in the vertical direction and a sharp interface
between the CO2 plume and the ambient brine. Vertical-equilibrium sharp-interface
models are commonly used in hydrology, petroleum engineering, and applied math-
ematics, because they allow analytic and semi-analytic solutions that give insight
into the underlying dynamics. CO2 storage in saline aquifers raises new questions
that have not been addressed by the literature on vertical-equilibrium sharp-interface
models, reviewed in § 3.1.2. We focus on the effects of confinement, residual trap-
ping and slope on the migration of CO2 plumes in saline aquifers. In § 3.2 we derive
the vertical-equilibrium sharp-interface equations for a confined sloping aquifer with
constant residual saturations.
In § 3.3 we study the effect of confinement on the spreading of the CO2 plume,
in absence of residual trapping. We use the scaling laws, associated with similarity
solutions for confined and unconfined aquifers, to detect the influence of confinement
in more general numerical solutions. Using this novel approach, we are able to show
that confinement has a strong effect on the migration of the CO2 plume, due to the
less mobile ambient brine. In other words the mobility ratio between the CO2 and
the brine is an important parameter governing the migration of the CO2 .
In § 3.4 we turn our attention to the relation between the sweep of the gravity
current and residual trapping. We extend the work on horizontal confined aquifers,
presented in § 3.3, to include residual trapping, and show that the solutions remain
self-similar. This implies that the volume of the current decreases as a power-law
in horizontal aquifers. Using the horizontal aquifer as a base case we show that
residual trapping is much more effective in sloping aquifers. The hyperbolic limit of
the equations allows us to study analytically how far the CO2 migrates up-dip and
how long it remains mobile.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 58

3.1.1 Immiscible displacements and residual saturations

In many sedimentary rocks supercritical CO2 is typically the non-wetting phase rela-
tive to the ambient brine. At the front of the CO2 plume the CO2 saturation increases,
and the brine is drained from the pore space. The capillary entry pressure prevents
the drainage of the brine from the smallest pores, resulting in an incomplete displace-
ment (Lenormand et al., 1983). We refer to the brine left behind the advancing CO2
front as the residual brine, Sbr . Bachu & Bennion (2007) showed that Sbr can range
from 0.2 to 0.68 at storage conditions in saline aquifers. The high end of these values
is surprising and may in part be due to heterogeneity and gravity segregation in the
experiments. They also show that the presence of residual water reduces the apparent
permeability of the CO2 to approximately 1/5 of the single phase permeability. We
refer to this value as the relative permeability of CO2 , denoted krc .
If the CO2 plume is migrating laterally as a gravity current the CO2 saturation
decreases at the trailing edge of the plume (figure 3.10), and the ambient brine imbibes
into the pore space previously occupied by CO2 . Preferential imbibition of the brine
into the smaller pores and interfacial instabilities leave CO2 behind as disconnected
bubbles and ganglia of CO2 (figure 3.1a), which are effectively immobile (Lenormand
et al., 1983). We refer to this immobile CO2 saturation as the residual CO2 saturation,
Scr , and to the process as residual trapping. Bachu & Bennion (2007) report values
of Scr from 0.1 to 0.35 for saline aquifers in the Alberta basin, indicating that they
will trap CO2 efficiently.

3.1.2 Gravity currents in porous media

While the residual saturation is an upscaled parameter that is determined by viscous
and capillary forces on the pore scale, the vertical sweep is a macroscopic quantity
that can be studied using fluid mechanical models of gravity currents in porous media.
Simple models to describe the fundamental behavior of such flows have been developed
independently in several disciplines, including petroleum engineering (Lake, 1989) and
hydrology (Bear, 1972). They are based on two common assumptions: hydrostatic
distribution of the pressure and complete segregation of the fluids. In this case the
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 59

dimensionality of the problem is reduced, and an equation governing the evolution

of the interface is obtained. We will refer to such models as vertical-equilibrium
sharp-interface models.
Sedimentary rocks are layered, and therefore aquifers may have very large aspect
ratios (length:height 1). In this case the pressure distribution is close to hydro-
static throughout the aquifer. A hydrostatic pressure distribution indicates that the
saturation be given by gravity-capillary equilibrium in the vertical direction. In hy-
drology this is also called Dupuit’s approximation, and in petroleum engineering it
is referred to as vertical equilibrium and was first introduced by Dietz (1953). Yort-
sos (1995) presented a rigorous derivation of the vertical-equilibrium sharp interface
model from the full governing equations of incompressible two-phase flow in porous
media.
If the thickness of the released fluid is large compared to the capillary transition
zone, the saturation in each fluid is approximately constant (figure 3.1b). As a first
approximation we replace the transition zone by a sharp interface and assume that
the saturations above and below the interface are constant.
An additional assumption that is commonly made is that the flow is unconfined.
In this case the ambient fluid is assumed to be inviscid, so that the pressure on
the interface is constant. This approximation results in the so-called porous medium
equation, which has received a lot of attention in the applied mathematics community
(Aronson, 1985). Woods (2002) reviewed the large variety of similarity solutions for
gravity currents obtained from porous medium type equations. We show in 3.3.2
that this approximation is also valid if the gravity current is much thinner than the
aquifer.
The groundwater table is often well described by an unconfined vertical-equilibrium
sharp-interface model. This set of approximations gives rise to the Boussinesq’s equa-
tion (Bear, 1972). Henry (1959) used a confined vertical-equilibrium sharp-interface
model the describe a salt-water intrusion into fresh-water aquifers. In petroleum en-
gineering confined vertical-equilibrium sharp-interface models have been used water
or gas injection (Dietz, 1953; Fayers & Muggeridge, 1990). Both disciplines have
used unconfined models to describe the upconing of gas-liquid interfaces near wells
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 60

(a) 1 mm (b) 1-Sbr Scr 0 (c)

Sc 0
¢x x,t)
Q c(x+¢
h
Q c(x,t) H
h
x,t)
0 Q b(x+¢
Q b(x,t)
hr z
x
saturation H µ
approx. sat.

Figure 3.1: (a) Microtomogram of non-wetting CO2 bubbles (black) and water (white)
in the pore space between grains (grey), from Benson et al. (2006). (b) Comparison
between non-wetting phase saturation profile after imbibition, inferred from a wire-
line log (Koperna & Kuuskraa, 2006), and the simplified step function saturation
profile corresponding to the sharp-interface approximation. (c) The geometry of the
porous layer and the variables used in the derivation are shown.

(Bear, 1972; Lake, 1989). Coats et al. (1971) used a vertical-equilibrium model, which
accounts for presence of a capillary transition zone, to formulate efficient quasi-three-
dimensional reservoir simulators. Parker & Lenhard (1989) introduced a similar model
for the migration of non-aqueous phases along the groundwater table.
Nordbotten et al. (2005), Lyle et al. (2005), Vella & Huppert (2006), and Bickle
et al. (2007) used sharp-interface models to study the development of CO2 plumes
during the injection period in confined and unconfined aquifers. During injection
the CO2 saturations increase, and no residual CO2 forms. After several decades the
injection of CO2 ends, and the CO2 plume migrates as a gravity current for several
hundreds or thousands of years, so that the long term evolution of the CO2 plume
may be modeled as an instantaneous release of finite volume. In the post-injection
period the CO2 saturation decreases in the wake of the gravity current, and residual
trapping reduces the current volume (figure 3.10). Kochina et al. (1983) extended
the sharp-interface model to allow for a constant residual saturation, and studied the
decay of the plume volume after a finite release. We extend their work to confined
sloping aquifers and consider the limiting cases.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 61

3.2 Governing equations

3.2.1 Derivation
We consider the flow of supercritical CO2 with density, ρc = ρ, and viscosity, µc , and
of brine with density, ρb = ρ + ∆ρ, and viscosity, µb , in a sloping porous layer of
constant thickness, H, with dip angle, θ ≥ 0, and infinite lateral extent (figure 3.1c).
Both fluids are considered incompressible, and we assume that the porous medium
is homogeneous and isotropic with permeability, k, and porosity, φ, and that the
top and bottom boundaries are impermeable. We also assume that gravity–capillary
equilibrium is maintained in any vertical cross section of the current (figure 3.1b), and
we replace the transition zone by sharp-interface and assume that the saturations
above and below the interface are constant. We denote the thickness of the CO2
plume by hc = h so that the depth of the brine is given by hb = H − h. In this case
the pressure in the aquifer is given by

pI − gρ(z − hb ) + Pc , for z > hb ,
p= (3.1)
pI − g(ρ + ∆ρ)(z − hb ), for z ≤ hb ,

where pI is the unknown pressure at the interface, Pc is the constant capillary pressure,
and g is the gravitational acceleration. The volume flux per unit width, qp , of phase
p ∈ {c, b} is given by the multiphase extension of Darcy’s law qp = −kλp ∂φp /∂x,
where φp = p − gρp (x sin θ + z cos θ) is the potential of phase p, and λp = krp /µp is the
mobility of phase p. The flow rate per unit width of phase p is given by Qp = hp qp .
In the absence of a source term, and with the assumption of incompressibility the
global conservation of volume is given by Qc + Qb = 0. Using this constraint we can
eliminate ∂pI /∂x from the expressions for the flow rates, and we obtain
 
hλc (H − h)λb ∂h
Qc = −Qb = kg∆ρ sin θ − cos θ .
hλc + (H − h)λb ∂x
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 62

To obtain an equation for the evolution of the interface, we consider the conser-
vation of the CO2 volume Vc over region ∆x and time ∆t as shown in figure 3.1(c).

 
∆Vc = φ (1 − Sbr ) ∆h∆x = Qc |x − Qc |x+∆x ∆t + Rc .

The source term, Rc , accounts for the volume of CO2 that is lost as residual saturation
Scr in the wake of the plume. Following Kochina et al. (1983) we assume a constant
residual saturation, Scr , so that

−∆h∆xScr , for ∂h/∂t < 0,
Rc =
0, for ∂h/∂t ≥ 0.

Taking limits for small ∆x and ∆t the equation for the evolution of the interface is
given by   
∂h ∂ h(H − h) ∂h
=κ − sin θ + cos θ , (3.2)
∂t ∂x h(M − 1) + H ∂x
where we have introduced two parameters. The conductivity, κ, of the CO2 is given
by

kλc ∆ρg
κ1 = φ(1−Sbr −Scr )
, for ∂h/∂t < 0,
κ= kλc ∆ρg
(3.3)
κ0 = φ(1−Sbr )
, for ∂h/∂t > 0,

where κ1 ≥ κ0 , and the mobility ratio that is given by M = λc /λb . The mobility ratio
M measures the change in the mobilities across the advancing part of the interface,
because we neglect the effect of the residual CO2 on the mobility of the brine. For
geological CO2 storage the ambient brine is less mobile, M > 1 (Adams & Bachu,
2002), and we restrict the discussion to this range. For Scr = 0 the coefficient 3.3
becomes continuous and equation 3.2 reduces to the form given by Bear (1972).
Note that the parameters are not independent; λc occurs in both. This choice of
parameters allows a simple reduction of (3.2) in the limits max(h)  H or M  1,
as discussed in § 3.3.2. In these limits the equation loses its dependence on M, but
retains the parameter κ. To allow this reduction κ must be defined in terms of the
mobility of the CO2 , and must be independent of the mobility of the ambient brine.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 63

Ld xo,0 x
0

plane of symmetry
CO2 brine
h0 h0
?

H xi;0

h Lf

Figure 3.2: The geometry of the initial condition and the three associated length
scales H, Ld , and Lf . A particular initial condition h0 [x] and the corresponding
idealized step function initial condition h∗0 [x] are shown. The arrows indicate that
the fluid has been injected over the entire thickness of the reservoir.

Equation (3.2) is invariant under the substitution

hb = H − hc , M = λb /λc , κ = κM = κλb /λc , θ = −θ. (3.4)

This transformation can be used to obtain a similar expression for the evolution of
the thickness of the ambient brine hb from (3.2).
The thickness of the zone saturated with residual saturation is given by hr [x, t1 ] =
hmax [x, t1 ] − h[x, t1 ], where hmax is the largest thickness the CO2 plume has achieved
at a particular location up to the time of consideration

hmax [x, t1 ] = max(h[x, t]), ∀x.

t≤t1

We also refer to hr as the residual surface that identifies the fraction of the aquifer
swept by the CO2 plume. In the absence of residual trapping this surface has no
significance.

Initial and boundary conditions

We consider the evolution of the CO2 plume after injection has stopped (t ≥ t0 ).
Figure 3.2 illustrates the plume shape at the end of the CO2 injection, which serves
as the initial condition for the post-injection period under consideration here. We
assume that CO2 has been injected along the whole depth of the aquifer, and that
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 64

it has formed a gravity tongue along the upper boundary of the porous layer (Riaz
& Tchelepi, 2006). Near the injection site the CO2 has displaced the water over an
average distance of Ld . The initial saturations in the CO2 plume and in the brine are
Sc = 1 − Sbr and Sc = 0, respectively. This corresponds to a porous medium that is
invaded by the released fluid for the first time, which is appropriate for CO2 storage
in saline aquifers. The lateral extent of the fluid invasion is determined by the viscous
to gravity ratio Rvg = (u/Ld )/(ug /H) = (uµg H)/(k∆ρgLd ), where u is the average
horizontal flow velocity, and ug = k∆ρg/µ is a gravitational velocity (for detailed
discussion see Tchelepi, 1994). When Rvg is small, gravitational forces dominate the
flow, and a thin gravity tongue forms at the top of the aquifer. When Rvg is large
the interface advances over the entire thickness of the aquifer. Ld increases with
time during the injection period, and Rvg generally decreases over time. During CO2
storage large quantities of fluid are injected, and the horizontal velocity, u, is high. As
a result Rvg is initially large, and the interface advances over across the full thickness
of the aquifer. Over time Rvg decreases, and a gravity tongue will form, leading to
the initial condition shown in figure 3.2.
The CO2 -brine interface transitions from h0 [xi,0 ] = H to h0 [xo,0 ] = 0 over a
frontal region of width Lf = xo,0 − xi,0 . The volume of CO2 is given by the integral
over the initial distribution
xo,0
V = 2φ(1 − Sbr ) H − h0 dx. (3.5)
0

The length scale Ld is chosen so that an idealized step function initial profile located
at x = Ld has the same CO2 volume as the particular initial condition (V = 2φLd H).
The idealized initial condition is

H, for |x| ≤ Ld ,
h∗0 =
0, for |x| > Ld .

This initial configuration imposes three length scales: the layer height H, the average
displacement distance Ld , and the width of the front at the end of injection Lf
(figure 3.2). The two boundary conditions for (3.2) require that h[x, t] → 0 for
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 65

|x| → ∞.

3.2.2 Dimensionless form

We chose the following dimensionless variables

η = h/H, ξ = x/L, σ = κ/κ1 ≤ 1, t/t∗ ,

In § 3.3 the natural length scale is given by L = Ld because the evolution in a

horizontal aquifer is symmetric. In § 3.4 we consider sloping aquifers and the natural
length scale is L = 2Ld , so that the initial aspect ratio of the current is therefore
given by A0 = L/H. We may choose either diffusive or an advective characteristic
time scale, t∗ , given by

ta = L(κ1 sin θ)−1 , or td = L2 (κ1 H cos θ)−1 , (3.6)

so that ϑ = t/td and τ = t/ta . Substituting these definitions into equation 3.2 we
obtain two dimensionless equations

σ −1 ητ + fξ = Pe−1 (f ηξ )ξ , or σ −1 ηϑ + Pefξ = (f ηξ )ξ , (3.7)

depending on the choice of the characteristic time scale. The flux function is defined
as
η(1 − η)
f= , (3.8)
η(M − 1) + 1
and the dimensionless discontinuous coefficient is given by

1, ητ < 0, or ηϑ < 0,
σ= (3.9)
1 − , ητ > 0, or ηϑ > 0,

depending on the choice of the characteristic time. Equations 3.7a and b have the
same dimensionless governing parameters

krc µb Scr
M= , Pe = A0 tan θ, = .
µc krb 1 − Sbr
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 66

The mobility ratio, M ≥ 0, the Peclet-number, Pe ≥ 0, and the residual, 0 ≤  < 1.

The advective and diffusive time scales are related by τ = Pe ϑ. Next, we study the
evolution of CO2 plumes with respect to these three parameters.

3.2.3 Discretisation
High accuracy numerical solutions of equation 3.7 were obtained as follows. The spa-
 
tial domain was divided into N grid cells of width ∆ξ centered at ξi = i − 12 ∆ξ,
where i ∈ [1, N ]. The temporal domain has been divided into T constant time steps
of size, ∆τ , so that the solution is obtained at times τn = n∆τ , for n ∈ [1 T ]. The nu-
 ξi+1/2
merical approximation of the cell average in the i-th cell is given by ξi−1/2 η(ξ, τn )dx =
ηin +O(∆ξ 2 ). The right hand side of (3.7) was discretised in divergence form to ensure
discrete conservation (Leveque, 2002), and central differences were used for all spa-
tial derivatives. The time derivative was discretised using the explicit forward Euler
method with a constant time step ∆τ . The update formula is given by

∆τ  n 
ηin+1 = ηin − Fi+1/2 − Fi−1/2
n
. (3.10)
∆ξ
 
n
The numerical flux function Fi+1/2 = F ηin , ηi+1
n
is given by
 
 
n
ηi+1/2 1− n
ηi+1/2
n
ηi+1 − ηin
F ηin , ηi+1
n
=− , (3.11)
n
ηi+1/2 (M − 1) + 1 ∆ξ

n
 n 
where ηi+1/2 = ηi+1 + ηin /2. The numerical results were validated against the early
similarity solution derived in § 3.3.1 (see figure 3.6a).

3.3 Regime transition: confined to unconfined

First we study the effect of the mobility ratio, M, on a finite release in a two-
dimensional horizontal and confined aquifer without residual trapping (Pe =  = 0 or
Scr = Sbr = θ = 0). We choose this simple model problem, illustrated in figure 3.3, to
study how the transition from confined to unconfined flow depends on the mobility
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 67

ratio. The current is initially given by two tilting interfaces and later by a spreading
mound. The tilting interfaces correspond to self-similar solutions of (3.2) for a con-
fined aquifer that predict that the tips of the interface propagate as x ∝ t1/2 . At late
time the governing equation simplifies to the porous medium equation for an uncon-
fined aquifer. The spreading mound is a similarity solution to the porous-medium
equation and predicts tip propagation as x ∝ t1/3 . We use these scaling laws, which
are predicted by the similarity solutions, to study when this transition occurs and
how the transition from confined to unconfined flow depends on the mobility ratio.

3.3.1 Self-similar solution at early times

For Lf < Ld the fronts are initially separated, and the finite propagation speeds of
the tips of the two interfaces ensure that the fronts will evolve independently until
their inward propagating tips collide (figure 3.3b). As a result, during the early period
each front can be analysed in isolation, and it is convenient to shift it to the origin
(x̂ = x − Ld ), so that the initial condition becomes
⎧
⎪
⎪ for x̂ ≤ x̂i,0 ,
⎨ H,
ĥ0 = h0 [x̂ + Ld ] , for x̂i,0 < x̂ < x̂o,0 ,
⎪
⎪
⎩ 0, for x̂ ≥ x̂o,0 .

The new boundary conditions are

ĥ [x̂ → −∞] = H, ĥ [x̂ → ∞] = 0. (3.12)

The early evolution of the interface is independent of the front separation 2Ld , but
the duration of this early period depends on Ld .

Dimensional analysis

We follow the general procedure for dimensional analysis given by Barenblatt (1996).
The problem defined above has three dimensions: length L, height H ∗ , and time T .
The dimensions of the variables and parameters appearing in (3.2) and the initial
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 68

plane of
symmetry
(a) xo xo∝ t1/2

early similarity solution early time

(b) xo xo

interface detaches back-propagation time

(c) xo xo

non-self-similar transition intermediate time

(d) xo xo ∝ t1/3

late similarity solution late time

Figure 3.3: The evolution of buoyant CO2 -vapor (gray) released into a horizontal
porous layer saturated by brine (white). The outward propagating tip of the interface
is marked by xo . All figures are exaggerated in the vertical direction, to make the late
solution (d ) visible. In many situations of interest the width of the invaded region in
figure (a) is several times larger than the aquifer thickness.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 69

1.0

0.8

0.6

M=1
0.4
M = 10
M = 102
M = 103
0.2
M = 104

0.0
-1.0 -0.5 0.0 0.5 1.0

Figure 3.4: Similarity solution is shown for different values of M. The variable θ is
the non-dimensional thickness of the CO2 , it is measured from the top of the aquifer.

condition are

−1
[ĥ] = [H] = H ∗ , [x̂] = [Lf ] = L, [t] = T, [κ] = L2 H ∗ T −1 , [M] = 1.

The dimensions of the parameters κ, H, and t are independent, and the parameters
1
give the length scale l = (κHt) 2 . We obtain the dimensionless parameters

ĥ x̂ Lf
Π= , Π1 = ζ = 1 , Π2 = ζf = 1 , Π3 = M. (3.13)
H (κHt) 2 (κHt) 2

The non-dimensional interface height Π can be written as a dimensionless function,

ψ, of the dimensionless variables Π = ψ [Π1 , Π2 , Π3 ] . We seek a similarity solution
for times after the details of the particular initial condition have disappeared. As
time increases Π2 → 0, while Π3 remains finite, and x̂ can always be chosen so that
Π1 is finite. Following the procedure given by Barenblatt (1996) we assume complete
similarity in the parameter Π2 , and we seek a solution of the form Π = ψ [Π1 , 0, Π3 ] =
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 70

θ [Π1 , Π3 ]. The expressions for ĥ and x̂ in these variables are

1
ĥ = Hθ [ζ, M] , x̂ = ζ [M] (κHt) 2 . (3.14)

Dimensional analysis shows that the tip propagation is proportional to t1/2 when
this scaling analysis is valid. The inner tip position is given by x̂i = ζi [M] (κHt)1/2 ,
and the outer tip position by x̂o = ζo [M] (κHt)1/2 , where ζi and ζo are dimensionless
quantities that depend only on the mobility ratio M. Substituting relationships (3.14)
into (3.2), we obtain a nonlinear ordinary differential equation for θ:
 
ζ dθ d θ(1 − θ) dθ
− = . (3.15)
2 dζ dζ θ(M − 1) + 1 dζ

The mobility ratio M is the only parameter determining the shape of the similarity
solution at early times. The inner and outer boundaries of integration ζi and ζo are
unknown, and must be determined as part of the solution. The boundary conditions
are:



dθ

ζi M dθ

ζo
θ (ζi ) = 1,
= , θ (ζo ) = 0,
=− .
dζ ζi 2 dζ ζo 2

The boundary conditions on θ are the non-dimensional form of (3.12), and the condi-
tions on dθ/dζ come from inserting the conditions on θ into (3.15). Equation (3.15)
and the boundary conditions are invariant under reflection in ζ, so that if θ1 (ζ) is
a particular solution θ1 (−ζ) is also a solution. The physical interpretation of this
reflection is exchanging the position of the fluids on either side of the initially ver-
tical interface. The evolution of the interface at early times has been reduced to a
nonlinear eigenvalue problem for a second order ordinary differential equation, with
two unknown eigenvalues and four boundary conditions. The two additional bound-
ary conditions allow the unique determination of the eigenvalues as a function of the
mobility ratio M. For unit mobility ratio (3.15) reduces to a simpler equation
 
ζ dθ d dθ
− = θ(1 − θ) (3.16)
2 dζ dζ dζ
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 71

0 0
10 10

−1 −1

i
−ζ
ζo

10 10

(a) (b)
−2 −2
10 10
0 1 2 3 4 0 1 2 3 4
10 10 10 10 10 10 10 10 10 10
M M

Figure 3.5: The numerical values for the outer (a) and inner (b) tip positions are
shown as a function of M (solid lines). The scaling laws (3.17) and (3.18) are shown
as dashed lines.

that has been obtained by Huppert & Woods (1995). The solution is symmetric with
respect to the origin, and the eigenvalues become ζi = −ζo = 1. Huppert & Woods
1
(1995) have obtained the solution θ = 2
(1 + ζ). We note that ζi = −ζo = −1 and
1
θ = 2
(1 − ζ) is also a solution. In the numerical solutions for the case M = 1 we
have chosen ζi < 0 and ζo > 0 (see figure 3.4). This choice places the CO2 on the
left side and the brine on the right side of the tilting interface and is consistent with
geometry shown in figure 3.2.

Numerical solution of eigenvalue problem

The nonlinear eigenvalue problem is solved numerically for the shape of the interface
and the tip positions as a function of M. We only need to obtain numerical solutions
for M > 1, the corresponding solutions for M < 1 can be obtained from the transfor-
mation (3.4). A shooting method is used to integrate inward from both boundaries of
the domain. The mismatch of θ and dθ/dζ at the origin was minimized to determine
the eigenvalues for a given value of M. The analytical solution for M = 1 was used as
an initial guess for ζi and ζo , and M was increased incrementally to obtain solutions
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 72

for M > 1. The resulting interface shapes are shown in figure 3.4.
As the mobility ratio increases a gravity tongue develops along one of the horizon-
tal boundaries. In the limit of M → ∞ the interface appears to approach a vertical
line. The CO2 viscosity, µc , is kept constant, because the similarity variable ζ de-
pends on µc through κ. Assuming krc and krb are constant, increasing the mobility,
M = µb /µc · krc /krb , we increase µb , and in the limit µb → ∞ the brine becomes
immobile and the interface remains vertical.
Figure 3.5(a, b) shows the position of the inner and outer tips as a function of
increasing mobility ratio. For large values of M the positions of the tips follow scaling
laws given by

1.24
ζi = −e0.2210 M−0.4997 ≈ − √ , for M > 10, (3.17)
M
2.37
ζo = e0.8645 M−0.4163 ≈ 0.42 , for M > 200, (3.18)
M

which are shown as dashed lines in figure 3.5. The outward propagating non-dimensional
tip position for M < 1 is obtained from the following argument
  12
1 κp Ht
x̂o [Mp < 1] = −x̂i [Mq > 1] = −ζi [Mq ] (κq Ht) = 2 −ζi [M−1
p ] .
Mp

The inward propagating tip xi [Mp < 1] can be obtained by an analogous argument.
√
The self-similar tip positions for M < 1 are given by ζo [M < 1] = −ζi [M−1 ]/ M
√
and ζi [M < 1] = −ζo [M−1 ]/ M. The position of the outward propagating tip of the
interface at early times is given by
 1
Ld + ζo [M] (κHt) 2 , M ≥ 1,
xeo =   1 (3.19)
Ld − ζi [M−1 ] κHt
M
2
, M < 1,

where the superscript e is used to indicate the scaling for the early similarity solution.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 73

1.0 1.0
t = 10-2
a = 0.1
0.8 t = 10-1 0.8 a = 0.2
t = 100 a = 0.3
t = 101 a = 0.4
0.6 0.6
θ(ζ) a = 0.5

θ(ζ)
0.4 0.4

0.2 0.2
(a) (b)
0.0 0.0
-5 -2.5 0 2.5 5 -0.5 -0.25 0 0.25 0.5 0.75
ζ ζ

Figure 3.6: (a) The numerical solution to (3.2) with initial condition (3.22) and
M = 10 is plotted at various times, t, and compared to the self-similar solution
obtained from (3.15) (solid grey line). (b) Numerical solutions to (3.2) with M = 20
and initial condition (3.25) is shown for several values of the parameter a in the initial
condition and compared to the solution of (3.15) with a = 0 (solid grey line).

The position of the inward propagation tip is given by

 1
Ld + ζi [M] (κHt) 2 , M ≥ 1,
xi =   1 (3.20)
Ld − ζo [M−1 ] κHt
M
2
, M < 1.

The numerical values of ζo and ζi can be obtained from figure 3.5 or from (3.17) and
(3.18) in the appropriate limits.

Range of validity of the early similarity solution

The similarity solutions described above were obtained under the assumption of com-
plete similarity in Π2 , which corresponds to a step function initial profile (Lf = 0).
Barenblatt & Zeldovich (1972) have shown that similarity solutions are intermedi-
ate asymptotic solutions for a much larger class of initial conditions. Therefore,
the analysis presented above also applies to initial profiles with a finite front width
(Lf = 0), for which Π2 = 0. For this larger class of initial conditions the similarity
solution will be valid after the details of the initial conditions have dissipated, because
Π2 = Lf / (κHt)1/2 approaches zero for Lf  (κHt)1/2 . Hence every particular initial
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 74

condition will be asymptotic to the similarity solution for

L2f
t te = . (3.21)
κH

This is illustrated for a particular initial condition in figure 3.6(a). The initial condi-
tion is a ramp defined by
⎧
⎪
⎪ x̂ < −0.5,
⎨ 0,
ĥ0 = x̂ + 0.5, −0.5 ≤ x̂ ≤ 0.5, (3.22)
⎪
⎪
⎩ 1, x̂ > 0.5.

In this case Lf = κ = H = 1, and we have chosen a mobility ratio of M = 10.

The partial differential equation (3.2) was solved numerically (see § 3.2.3), and the
solutions at various times are plotted as dashed lines. The similarity solution obtained
from the eigenvalue problem is shown for comparison. The particular solution is
essentially identical to the similarity solution at t = 10, which is only an order of
magnitude larger than the lower bound te = 1.
For t te the similarity solution is valid until the inward propagating tip reaches
the origin xi (tb ) = 0, where tb is the back-propagation time (figure 3.3b). Solving
(3.20) for tb we obtain ⎧
⎨ L2d
κHζi [M]2
, M ≥ 1,
tb = 2 (3.23)
⎩ Ld M
, M < 1.
κHζo [M−1 ]2

Hence the early self-similar solution is valid for te  t  tb . We can also define
a new length scale Lb = 2xo (tb ), the width of the current at the back-propagation
time . For small M, Lb provides a suitable initial length scale for the late similarity
solution in § 3.3.2. Lb is given by
⎧  
−1 √
⎪
⎨ 2Ld 1 + ζi [M −1] M , M < 1,
ζo [M ]
Lb =   (3.24)
⎪
⎩ 2Ld 1 + ζo [M] ,
ζi [M]
M ≥ 1.

In some situations the CO2 may not fill the entire depth of the domain, so that the
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 75

idealized initial condition is given by


H − a, for |x| ≤ Ld ,
h∗0 = (3.25)
0, for |x| > Ld ,

where a < H is the thickness of the brine in the injection zone. Figure 3.6(b) compares
numerical solutions to (3.2) with M = 20 for various values of a, with the solution to
(3.15), which corresponds to a = 0. Solutions for a = 0 are also self-similar in the
early similarity variable, so that the outward propagating tip propagates as t1/2 for
a = 0. The position of the tip ζo in self-similar coordinates decreases with increasing
a, but for a/H < 0.2 the difference is less than 10% compared to the values of ζo
given in figure 3.5(a). The evolution of currents that do not occupy the full thickness
of the layer is similar to those investigated here and follows the same early scaling
law. These results indicate clearly that the viscosity of the brine cannot be neglected,
even if the CO2 plume does not occupy the full thickness of the aquifer. A full
investigation of this larger family of similarity solutions for a = 0 is beyond the scope
of this investigation.

3.3.2 Self-similar solution at late times

Reduction to the porous medium equation

At the back-propagation time tb the interface detaches from one of the horizontal
boundaries, and the thickness, h, of the CO2 decreases monotonically as a function
of time (figure 3.3c, d ). At late times h  H, and we expect the solution for a finite
layer to be similar to the solution in a half-space. The equation for the half-space
can be obtained from (3.2) by taking the limit for H → ∞, for finite h and M, or
equivalently taking the limit h → 0 for finite H and M. Consider the limit of the
nonlinear diffusion coefficient in (3.2) for small h, keeping M and H constant,

h (H − h)
lim = h. (3.26)
h→0 h(M − 1) + H
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 76

In this limit (3.2) reduces to the porous medium equation

 
∂h ∂ ∂h
=κ h , (3.27)
∂x ∂x ∂x

which has been studied intensively, and the similarity solution for a finite release of
fluid into a two-dimensional porous half-space was found by Barenblatt (1952).
We expect that the limit (3.26) becomes a good approximation even if h  H
is finite, and (3.27) becomes a good approximation for (3.2) after some time. The
parameter M can vary over several orders of magnitude, and we need to consider its
effect on the validity of approximation (3.26). Consider the approximation in the
denominator of (3.26) for finite but small values of h

h(M − 1) + H ≈ H.

Large values of M require even smaller values of h to allow this approximation. Since
h is a monotonically decreasing function of time, the half-space approximation will
become valid for all M eventually. In § 3.3.3 we develop an expression for the onset of
half-space behaviour as a function of M. For small M the half-space approximation
becomes valid very quickly. In the limit of small mobility ratios we obtain

h (H − h)
lim = h, (3.28)
M →0 h(M − 1) + H

and (3.2) reduces to (3.27) at all times and for all values of h and H. As mentioned
in § 3.1.2, the simplification in this limit is responsible for the success of (3.27) in
problems of unconfined flow, where the ambient fluid is a gas (M  1).
Equation (3.27) depends only on the CO2 mobility, λc , not on the mobility ratio
M. From the global conservation of mass Qc + Qb = 0, we can obtain an expression
for the Darcy velocity qb in the ambient fluid

hc qc
qb = − . (3.29)
(H − hc )

For finite h and qc the flux in the brine qb becomes negligible as H → ∞. In contrast
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 77

to the early evolution, where M is the governing parameter, the problem becomes
independent of the mobility ratio at late times, because the brine is stationary.

Barenblatt’s solution

In either limit the initial condition for the porous medium equation is a particular gas
distribution h̃[x], with finite width Lb , in a half-space otherwise saturated by water.
The volume of current is given by

V = h̃[x]dx = 2Ld H. (3.30)

A similarity solution in the parameters

  13
V2 1
h= ϕ[ς] and x = ς(κV t) 3 (3.31)
κt

has been found by Barenblatt (1952) and is given by

⎧   13  
⎪
⎨ 1 V2 x2
6 κt
ςo2 − 2 , for |x| ≤ xlo ,
h[x, t] = (κV t) 3 (3.32)
⎪
⎩ 0, for |x| > xlo .

From the definition of the self-similar coordinate, ς, the tip propagation at late times
is proportional to t1/3 , and the front position at late time is given by

1
xlo = (9κLd Ht) 3 . (3.33)

The superscript l identifies the tip scaling for the late similarity solution. The late
similarity solution depends on the CO2 volume V = 2φLd H, but it is independent of
the local length scale Lf of the initial front, and the mobility ratio M. In contrast
the early tip scaling (3.19) is independent of the global length scale Ld , but depends
on Lf and M.
The similarity solution obtained for the idealized initial condition is an intermedi-
ate asymptotic solution for a larger range of initial conditions with Lb = 0, for times
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 78

larger than
L3b
t tl = t̃ + , (3.34)
2κg Ld H
where t̃ is the time at which (3.26) becomes valid. This lower bound becomes impor-
tant when M  1, and the porous medium equation becomes valid very quickly. In
this case t̃ = tb and (3.24) is a suitable initial length scale Lb , so that tl becomes
⎛  √ 3 ⎞
L2d −1
⎝8 1 + Mζi [M ] M ⎠.
tl = − (3.35)
κH −1
ζo [M ] ζo [M−1 ]2

3.3.3 Non-self-similar transition

We obtain a description of the front propagation speed at early times from the sim-
ilarity solution describing a tilting interface (§ 3.3.1). At late times the governing
equations simplify to (3.27), and the similarity solution of Barenblatt (1952) gives
the propagation speed at late times. The transition from the early to the late simi-
larity solution will not be self-similar, and must be investigated numerically.
We study this transition as a function of the dimensionless mobility ratio, M. We
consider initial distributions that are symmetric with respect to the origin, so that
we only need to consider the spatial domain [0, a], where a > 0 is chosen larger than
the maximum propagation distance estimated from (3.33). The initial condition in
all simulations is the following step function

1, ξ ≤ 1,
η [ξ, ϑ = 0] = (3.36)
0, ξ > 1.

The problem is symmetric with respect to the origin, so that the boundary condition
at the origin is ∂η(0, ϑ)/∂ξ = 0, and the outer boundary condition is η(a, ϑ) = 0.

Transition time

The two examples in figures 3.7(a) and 3.7(b) show the numerical transition from
the early to the late similarity solution. The initial condition is the early similarity
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 79

1 1
la M = 1/2 lat M = 10
t e
0.8 0.8 s

e
sim

im
ea

il a
rl

ila

rit
ea
´ ´

rit
sm
y
0.6 0.6

y
r
y

ly
i

so
max(´) max(´)

so
ila

sim

lut
lut
rit

ion
ila
ion
y
0.4 0.4

rit
so
lu

y
ol
tio

s
u tio
n
0.2 0.2 n
(a) (b)
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
»/»o »/»o

Figure 3.7: The transition of the numerical solution (light lines) from the early to
the late similarity solution (both heavy lines) is shown in scaled coordinates. (a) The
numerical solution is shown at ϑ = ϑb + {1, 4, 10, 30, 100}, where ϑb = 0.61; (b) The
numerical solution is shown at ϑ = ϑb + {101 , 102 , 103 , 104 , 105 }, where ϑb = 7.1.

solution at the non-dimensional back-propagation time ϑb given by

⎧
⎨ 1
, M ≥ 1,
ζi [M]2
ϑb = (3.37)
⎩ M
, M < 1.
ζo [M−1 ]2

In figure 3.7(a) the evolution is shown for M = 1/2. In this case the curvature of
the early and the late similarity solution is of the same sign. The main difference
between them is the slope at the origin, where the early similarity solution has a
finite slope, but the late similarity solution has zero slope. In this case the transition
period is relatively short, and the late similarity solution is a good approximation to
the solution for ϑ > 100. Figure 3.7(b) shows the transition for the case M = 10. In
this case the curvature of the early and the late similarity solution is of opposite sign.
The numerical solution adjusts very slowly and the transition period is very long.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 80

Transition of the scaling for the tip position

Figure 3.8(a-d ) shows the numerical results for the non-dimensional position of the
outward propagating tip ξo of the CO2 as a function of non-dimensional time ϑ. The
figure shows the effect of increasing the mobility ratio M on the tip propagation and
the timing of the transition. The scaling laws for the tip position obtained from the
early and late similarity solutions are also shown. In non-dimensional coordinates
these scaling laws (3.19, 3.33) simplify to
 1
1 + ζo [M] ϑ 2 , M ≥ 1,
ξoe = −1 1
− 12
(3.38)
1 − ζi [M ] ϑ M 2 , M < 1,
1
ξol = (9ϑ) 3 , (3.39)

respectively. The shifted tip position ξo −1 is plotted as a function of time in logarith-

mic axes, so that the early scaling law (3.38) plots as a straight line with slope 1/2.
In these variables the late scaling law (3.39) does not plot as a straight line, but it
approaches a straight line with slope 1/3 for large times, where it becomes valid. The
late scaling law is independent of M, and therefore represented by the same curve in
all four figures, while the straight line corresponding to the early similarity solution
is shifted downward as M increases.
Comparison of the numerical results with the scaling laws from the early and late
similarity solutions leads to the following four observations:

1. The numerical tip position initially follows the early scaling law ξo ∝ ϑ1/2 , and
then the scaling law for late times ξo ∝ ϑ1/3 .

2. The transition time ϑt increases monotonically with increasing M. Comparison

of figures 3.8(a) and 3.8(b) shows that this increase is very small for M < 10−1 .
Figures 3.8(c) and 3.8(d ) show a rapid increase of the transition time for M >
10−1 .

3. The transition from early to late scaling is short for M ≈ 10−1 (figure 3.8b),
and increases rapidly for M > 10−1 (figure 3.8d ).
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 81

2
10
1
(a) M =
100
1
10 ϑb
ξo − 1
0
10

ϑt
−1
10
−2 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10

2
10
1
(b) M =
10
1
10 ϑb
ξo − 1
0
10

ϑt
−1
10
−2 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10

2
10
(c) M = 1
1
10 ϑb
ξo − 1
0
10

ϑt
−1
10
−2 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10

2
10
(d) M = 10
1
10 ϑb
ξo − 1
0
10

ϑt
−1
10
−2 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10
ϑ

Figure 3.8: The numerical results for the non-dimensional tip position ξo are shown
as a function of non-dimensional time ϑ, for different mobility ratios M. In all figures
the numerical solution is given by dots (· · · ), the tip scaling from the early similarity
solution by a dashed line (- - -), and the tip scaling from the late similarity solution
as a full line (—). (a) ϑt = 2.0, ϑb = 0.1; (b) ϑt = 2.5, ϑb = 0.24; (c) ϑt = 29.3,
ϑb = 1; (d ) ϑt = 811.3, ϑb = 1; (d ) ϑt = 811.3, ϑb = 7.1.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 82

4. The tip position follows the early scaling law ξo ∝ ϑ1/2 even after the early
similarity solution has become invalid at ϑb (3.37). In figure 3.8(a-c) the early
scaling law continues to be valid almost up to ϑt . This shows that the finite
depth of the layer continues to have a strong effect on the solution, even after
the interface has detached from one of the boundaries.

Although the early and late scaling behaviours are separated by a transition period,
it is useful to define a dimensionless transition time ϑt that falls within this transition
period. This transition time defines a lower bound for the validity of the late similarity
solution. The difference between the early and the late scaling laws is given by

 
f [ϑ; M] = ξol − 1 − ξoe , (3.40)

where ξol is given by (3.39), and ξoe by (3.38). We use the substitution ϑ = y 6 to
eliminate ϑ1/2 and ϑ1/3 and obtain a cubic in y. Let Mt denote the value of M, where
the early scaling law is tangent to the late scaling law such that f (ϑ; Mt ) = 0. For
M ≥ Mt the non-dimensional transition time ϑt can be defined as the intersection
of the early and late time scaling laws (figure 3.8c, d ), and is therefore given by
the largest real root of f (ϑ; M ≥ 1) = 0. For M < Mt the two scaling laws do
not intersect, but the transition time can be defined as the point of minimal vertical
distance between the two scaling laws (figures 3.8c and 3.8d ) given by the local
minimum of (3.40). Solving for the appropriate root and the minimum we obtain the
following expression for the non-dimensional transition time
⎧   π θ 6
1
⎪
⎪ 1 + 2 cos −3 , M ≥ 1,
⎨ 9ζo [M] 6

3
 
M3 6
ϑt = 9ζi [M−1 ]6
1 + 2 cos π3 − 3θ , Mt ≤ M ≤ 1, (3.41)
⎪
⎪
⎩ 64M3
, M ≤ Mt ,
9ζi [M−1 ]6

where θ is the principle argument of the following complex numbers

⎧  !
⎪ 2
⎨ Arg −2 + 3ζo [M] + iζo [M] 12 − 9ζo [M] , M ≥ 1,
    
θ= √  1 2 1  1 2
⎪ 3
⎩ Arg −2M 2 + 3 Mζi M − iζi M 3M 4M − 3ζi M , Mt ≤ M ≤ 1.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 83

Due to the change in the definition of the transition time at Mt the graph is not smooth
at this point (figure 3.9). For M < Mt the transition time increases very slowly with
M, while it increases strongly for M ≥ Mt (figure 3.9). Mt can be obtained by finding
the value of M for which the local minimum of (3.40) is zero.

3  2
Mt − ζi M−1 t = 0. (3.42)
4
 
This equation must be solved numerically, because ζi M−1
t is not known analytically,
and we obtain Mt = 0.1839. For large values of M, ξo (ϑt ) 1, and (3.40) simplifies
to f ≈ fˆ = ξol − ξoe and gives a scaling law for the transition time ϑt = 0.45M5/2 .
For small values of M, (3.20) can be used to simplify (3.41) to obtain a constant
ϑt = 1.96. Equation 3.41 is complicated to evaluate, and we therefore introduce
a simple expression based on the two limits discussed above and a simple fit for
intermediate values. ⎧
⎪
⎪
5
2
⎨ 0.45M 2 , 10 < M,
3
ϑt ≈ 36.6M 2 , Mt ≥ M ≤ 102 , (3.43)
⎪
⎪
⎩ 2, M < Mt .

3.3.4 Regime diagram

We summarize our results on the evolution of a finite release of CO2 into a horizontal
aquifer saturated with brine, where the density difference between the immiscible
fluids is the only driving force. Figure 3.9 combines all time scales into a M-ϑ regime
diagram that describes the evolution of a finite CO2 plume. The only parameter in
this problem is the mobility ratio, M = λc /λb , between the CO2 and the brine. The
magnitude of a particular dimensional time scale is given by the characteristic time
td = L2d κ−1 H −1 formed using the displacement distance Ld , the conductivity of the
CO2 given by κ, and the height of the layer H. The CO2 mobility, λc , appears in
κ, and κ enters the characteristic time td used for the non-dimensionalisation, hence
λc is constant and changes in M are due to changes in λb . For all finite values of M
the evolution can be divided into three dynamic stages: an early self-similar regime,
a transition period, and a late self-similar regime.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 84

1010
? ϑt ∝ M 5/2

?
105
e
im
iont
it
ns
tra ϑb ∝ M
# ?
ime
on t
8 ←ϑl lower bound late soln. p agati
ϑt -pro
1.96 ← back
100
ϑb ∝ M1/6

10-5
10-4 10-2 100 102 104
M

Figure 3.9: Regime diagram for a finite release of fluid into a horizontal porous slab,
showing the non-dimensional time scales obtained in this study, and the shapes of the
gravity current as a function of the mobility ratio M. The shaded region indicates
the transition period between the similarity solutions. The characteristic time to
dimensionalize all results is td = L2d κ−1 H −1

After the details of the initial condition are lost, the interface shape and dynamics
are asymptotic to an early similarity solution that corresponds to a tilting interface.
The early similarity variable is ζ = x(κHt)−1/2 , so that the non-dimensional tip posi-
tion is given by ξo ∝ ϑ1/2 . During this period the left and the right interfaces evolve
independently, and the length scale of their separation, 2Ld , does not appear in the
similarity variable. In this phase, both fluids move with non-zero velocities, and there-
fore the mobility ratio M determines the shape of the interface. We have not plotted
the lower bound for the onset of the early similarity solution (3.21), because this time
scale depends on the initial width of the front Lf , which is given by ϑe = (Lf /Ld )2 .
The validity of the early similarity solution ends at the back-propagation time, ϑb ,
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 85

because the inward propagating tips of the two initially separated fronts begin to
interact at the origin (figure 3.3b). Figure 3.9 shows that ϑb increases monotonically
with time, and follows simple scaling laws for small and large M. Physically we can
explain the increase of ϑb with increasing M by the increasing viscosity of the brine,
which slows down the inward propagating tip of the tilting interface. The period
during which the early similarity solution is valid increases with increasing M, be-
cause the ϑe = ϑe [M]. As long as Lf = 0 there are always values M ≤ Mc such that
ϑe ≥ ϑb , and hence the early similarity solution is not realized. Mc is determined by
√
the equation Lf /Ld = Mc /ζo [M−1 c ] for Mc < 1, a similar equation can be found for
Mc > 1. Even in the case Lf = 0 the early similarity solution will not be realized
in the limit M → 0, because ϑb → 0. Figure 3.8 shows that the scaling law for the
non-dimensional tip position ξo ∝ ϑ1/2 is valid for a significant time even after the
early similarity solution itself has become invalid at ϑb .
The initial similarity solution is followed by a period where the solution is not
self-similar and must be obtained numerically (figure 3.7). For M  1 we can define
the transition period as ϑb < ϑ < ϑl , where ϑl is given by (3.35). The duration of
the transition period increases as M → 0, because the upper boundary is constant
ϑl = 8, while the lower boundary is proportional to ϑb ∝ M1/6 . For M > 1 we have
no estimate of the upper boundary of the transition period. The numerical results in
figure 3.8(b-d ) show that the transition period increases with increasing M. From the
transition of the scaling laws for the tip position we have defined a transition time ϑt ,
that provides a lower bound on the onset of the late similarity solution. Equation 3.41
or (3.43) shows a rapid increase of ϑt with increasing M for M > Mt .
After the transition period the late similarity solution becomes valid, because
the CO2 occupies only a small fraction of the thickness of the aquifer, and (3.2)
reduces to (3.27). Equation 3.27 admits a similarity transformation in the variable
ς = x/(κV t)−1/3 , and the analytical solution was obtained by Barenblatt (1952). In
contrast to the early similarity solution, this late similarity solution is independent
of M and depends on the volume of the gravity current, given by V = 2Ld H. In
this limit the velocity of the brine is negligible, which explains why the problem
is now independent of the mobility of the brine, and hence the mobility ratio M.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 86

(a) (e)
-15 10 -5 0 5 10 15 km -5 0 10 20 30 40 50 km
trapped CO2: 4% # = 0.3 trapped CO2: 6% # = 0.3
t = 112 yrs t = 112 yrs

(b) (f)
20% # = 10 19% # = 1.5
t = 3730 yrs t = 560 yrs

(c) (g)
32% # = 100 44% #=4
t = 37300 yrs t = 1500 yrs

(d) (h)
41% # = 1_000 69% #=8
t = 373000 yrs t = 3000 yrs

Figure 3.10: Current evolution in an aquifer with 30 mD permeability, for details of

the physical properties see § 3.5.2. CO2 currents (dark grey) and the zone containing
residual saturation in their wake. Figures a to d show the evolution in a horizontal
aquifer. These figures are equivalent to figure 3.3, but they include the effect of
residual trapping. Figures e to h show the evolution in an equivalent aquifer sloping
with 5◦ .

When the late similarity solution is valid, the non-dimensional tip position is given
by ξo ∝ ϑ1/3 . Again the scaling for the tip position becomes valid before the solution
is fully self-similar.

3.4 Gravity currents with residual trapping

The regime diagram of gravity currents in horizontal aquifers (figure 3.9) shows clearly
that the confined nature of the aquifer has to be taken into account if gravity currents
with M > 1 are considered. The mobility ratio for the migration of the CO2 plume
is M ≈ 5, so that we have to study the effect of residual trapping on a CO2 plume
migrating in a confined aquifer. In particular the early similarity solution (figure 3.4)
shows that a gravity tongue forms for M > 1, which reduces the vertical sweep of the
CO2 plume. Most research on residual trapping during CO2 storage has focused on
the effect of hysteresis on the magnitude of Scr , and the design of injection strategies
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 87

that maximize residual trapping during or shortly after the injection period (Mo et al.,
2005; Juanes et al., 2006; Ide et al., 2007). We focus on the effect of the vertical sweep
efficiency, S, which is defined as the fraction of the aquifer contacted by the CO2 plume
during its buoyancy-driven migration in the post-injection period. A volume balance
argument shows that the up-dip migration distance, x↑ , is approximately given by

x↑ ≈ Vc / (φHSScr ) , (3.44)

where Vc is the volume of CO2 , H the thickness of the aquifer, and φ the porosity.
Ennis-King & Paterson (2002) used this relationship to highlight the sensitivity of the
migration distance to the magnitude of Scr . Equation (3.44) shows that the effect of S
on the migration distance of the CO2 plume is comparable to that of Scr . The sweep
is expected to be less than unity, because gravity segregation and viscous instabilities
lead to the formation of a gravity tongue along the top of the aquifer.

3.4.1 Parabolic limit: horizontal aquifers

Here we extend the analysis presented in § 3.4 to include the effect of residual trapping.
In the presence of residual trapping an analogous transition from an early to a late
scaling law is observed. However, we show below that the exponents of the early
scaling laws are independent of the residual, while Bear & Ryzhik (1998) showed that
the exponents of the late scaling laws are a function of the residual. In the terminology
of Barenblatt (1996), the evolution of the current transitions from a similarity solution
first kind to a similarity solution of the second kind.

Early similarity solutions with loss

Initially the two interfaces of the current evolve independently and symmetrically
(figure 3.10a). Until they interact they evolve as tilting interfaces described by a self
similar solution given by a step function initial condition

˜
a, ξ < ξ,
η= (3.45)
˜
b, ξ ≥ ξ,
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 88

1 1
ζi ζi
ζo ζo

η η

=
=
0.5 0 0.5

=

0

=
1

1
(a) (b)
0 0
-1.5 -1 -0.5 0 0.5 1 -0.5 -0.25 0 0.25 0.5 0.75
ζ ζ

20 20
(c) ζo 0.1 (d) ζi
0.2 -0.3
15 0.3 15
0.4
M M
10 0.5 10 -0.4
0.6
-0.5
5 0.7 5
-0.6
0.8 -0.7 -0.8
1 1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Figure 3.11: (a) Interface shapes for M = 1 and  increasing from 0 to 1 in 0.2
increments. (b) Interface shapes for M = 10 and  increasing from 0 to 1 in 0.2
increments. (c) Contours of the outward propagating tip position, ζo . (d ) Contours
of the inward propagating tip position, -ζi .
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 89

where ξ˜ is the initial position of the interface. Here we consider only the cases a = 1
and b = 0, but self-similar solutions are not limited to this choice. The outward and
inward propagating tips of the interface are denoted by ξo and ξi respectively. For
ξ˜ < ξ ≤ ξo the interface is advancing, and the thickness of the current is increasing,
ηϑ > 0, so that σ = 1 − . For ξi ≤ ξ < ξ˜ the interface is receding, and the thickness
of the current is decreasing, ηϑ < 0, so that σ = 1. The location ξ˜ of the discontinuity
in the coefficient is constant at early times.
Equation 3.7b and the initial condition (3.45) are self-similar in the variable ζ =
ξ/ϑ1/2 . Therefore, the propagation of the plume tip at early time is given by the
scaling law ξo = a[, M]ϑ1/2 , where a is a constant that decreases with increasing
. Inserting Pe = 0 and the similarity variable into (3.2b), we obtain the ordinary
differential equation
 
ζ dη d η(1 − η) dη
− = σ̃ , (3.46)
2 dζ dζ η(M − 1) + 1 dζ

where σ[ηζ ζϑ ] = σ̃[ηζ ]. The mobility ratio, M, and the residual, , determine the
shape of the similarity solution at early times. The inner and outer boundaries of
integration ζi and ζo are unknown, and they are generally functions of both M and .
These eigenvalues must be determined as part of the solution, requiring two additional
constraints. The boundary conditions on the inward and outward propagating tips
are given by


dη

ζi M
η [ζi ] = 1,
= , (3.47)
dζ ζi 2


dη

−ζo
η [ζo ] = 0,
= . (3.48)
dζ ζo 2(1 − )

respectively. The conditions on η specify the vertical extent of the interface. The
conditions on ηζ have no direct physical interpretation, but they are required to satisfy
(3.15) at the boundaries ζi and ζ0 , and they provide the two additional constraints
necessary to determine the eigenvalues. Together (3.46) and (3.47) form a nonlinear
eigenvalue problem for the shape of the interface and the two unknown tip positions,
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 90

ζi and ζo . This eigenvalue problem is solved numerically as a function of M and 

using a shooting method.
Figures 3.11(a-b) show that increasing M leads to the formation of a gravity
tongue, while increasing  tends to shorten the gravity tongue and may lead to an
inflection in the interface. For M ≥ 1 the outward propagating tip position, ζo , is
mostly a function of , while the inward propagating tip position, ζi , is more strongly
affected by M (figure 3.11c-d ). In the limit where all fluid is trapped,  = 1, the
˜ but the interface does recede, ξi = ξ.
interface does not propagate, ξo = ξ, ˜ However,
the distinction between the fluids across the receding interface is lost, and the interface
looses it physical significance. The evolution of this imaginary interface is still given
by (3.46), but it does not correspond to any fluid movement.
The early similarity solution shows that residual trapping has only a weak influence
on the evolution of the receding part of the interface. Instead residual trapping slows
down the advancing interface, because less fluid is supplied from the receding part.
A similar behavior is observed in the hyperbolic model discussed in § 3.4.2.

Regime transition and late solution

In § 3.3.2 we show that f → η at late times when the current is much thinner than
the thickness of the aquifer, so that (3.7b) reduces to

ηϑ = σ (ηηξ )ξ , (3.49)

for all M. The only parameter in this equation is the residual, , in the discontinuous
coefficient, σ. Semi-analytical solutions for the interface shape are available for the
radial (Kochina et al., 1983; Barenblatt, 1996), and linear cases (Bear & Ryzhik,
1998).
For a compact initial condition, the solution of a parabolic equation with a moving
discontinuous coefficient is a similarity solution of the second kind (Barenblatt, 1996).
These solutions are characterized by an anomalous exponent in the scaling laws that
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 91

102
1/2 (b)
early
late 101
1/3
β ξ̂o
1/4
100 =0
1/8  = 0.25
 = 0.5
(a)  = 0.75
0 10-1
0 0.25 0.5 0.75 1 10-2 10-1 100 101 102 103 104 105
 ϑ

Figure 3.12: Change of the solution from the early to the late scaling law (M = 1).
(a) The scaling laws for ξo ∝ ϑβ at early and late times as a function of the residual,
. The symbols at  of 0, 1/4, 1/2, and 3/4 show the exponents obtained by fitting
the numerical data shown in figure b. For the early scaling law the data was fit for
ϑ between 0.1 and 10 and for the late scaling law the data was fit for ϑ > 1000. (b)
˜ is shown as function of ϑ, for increasing residuals,
The shifted tip position, ξˆ = ξo − ξ,
where ξ˜ = 1.

govern the solution. In this case we obtain

ξo = c1 [, M, η0 ] ϑβ() , (3.50)
ηmax = c2 [, M, η0 ] ϑ2β()−1 , (3.51)
V = c3 [, M, η0 ] ϑ3β()−1 , (3.52)

for the tip of the plume, the height of the plume, and the volume of the plume.
Figure 3.12(a) shows the decrease of the anomalous exponent β with increasing .
For (3.49) the exponent must be determined numerically from a nonlinear eigenvalue
problem (see Bear & Ryzhik, 1998). As expected the increasing residual trapping
leads to a faster decrease of the plume volume and slower tip propagation. Another
characteristic of this type of solution is that the constants of proportionality (c1 -c3 )
in these scaling laws depend on the initial condition.
Figure 3.12(b) shows transition in the scaling law for the tip position from the
early scaling law, with the exponent 1/2, to the late scaling law with the anomalous
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 92

exponent β ≤ 1/3. Initially only the constant in the scaling law is affected by ,
but at late times both the coefficient and the exponent depend on . The sharp-
interface model presented here predicts that the volume of the current always decays
as a power-law in horizontal aquifers. In this case the current is never exhausted and
continues to spread. Capillary forces, which have been neglected in our model, will
arrest the movement of the current once it becomes too thin.

3.4.2 Hyperbolic limit: sloping aquifers

For confined sloping aquifers the current evolution is not self-similar, and no general
solutions to (3.2) are known. For large Pe, equation 3.7a reduces to a simpler equation
that allows an analysis of the current evolution. For Pe → ∞ we obtain the following
quasi-linear hyperbolic equation

ητ + σληξ = 0, (3.53)

where the derivative of the flux function is denoted λ ≡ fη .

Riemann problems

First we consider the two Riemann problems to understand the effect of trapping,
i.e. the discontinuous coefficient σ, on rarefactions and shocks. Then we consider a
hat function initial condition, representing a finite release of fluid, and the resulting
wave interaction. For non-zero values of loss, the wave interaction will lead to an
extinction of the current in finite time.

Rarefaction Consider the piecewise constant initial data


˜
1, ξ ≤ ξ,
η[ξ, τ = 0] = (3.54)
˜
0, ξ > ξ.

The weak solution, a spreading wave or rarefaction, is obtained by the method of

characteristics. This rarefaction is a tilting interface analogous to the self-similar
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 93

0.06 1
(a) ²=0 (b)

0.25 0.75
0.04
¾f 0.5 ´
0.5
0.02
0.75
0.25

²
 = 1
0 1 ²=0
0
0 0.25 0.5 0.75 1 -0.5 0 0.5 1
´ »=¿

Figure 3.13: (a) The flux function f is shown for M = 10 and increasing trapping .
(b) The corresponding solution profiles given by (3.56) in self-similar coordinates.

solutions presented in § 3.4.1. Following the notation introduced there, we refer to

the tip propagating to the left as ξi and to the tip propagating to the right as ξo . The
characteristics of the solution originating at the discontinuity are given by

˜
ξ = σ [ητ ] λ[η]τ + ξ, (3.55)

therefore all solutions are self-similar in the coordinate ξ/τ . The discontinuity in
˜ but the interface itself
σ leads to a discontinuity of the slope of the interface at ξ,
˜ The equation for the evolution of η[ξ, τ ],
remains continuous, because λ = 0 at ξ.
given by inverting (3.55) is
⎧
⎪
⎪ 1, ξ ≤ ξi ,
⎪
⎪

⎪
⎪
⎪
⎪
−τ +M̂(ξ0 −ξ)+ Mτ (τ +M̂(ξ−ξ0 ))
˜
⎪
⎪ , ξi < ξ < ξ,
⎨ M̂(τ +M̂(ξ−ξ0 ))
1 ˜
η [ξ, τ ;  < 1, M = 1] = √ , ξ = ξ, (3.56a)
⎪
⎪
(1+ M)

⎪
⎪ −τ ˆ+M̂(ξ0 −ξ)+ Mτ ˆ(τ ˆ+M̂(ξ−ξ0 ))
⎪
⎪ , ξ˜ < ξ ≤ ξo ,
⎪
⎪ M̂(τ ˆ+M̂(ξ−ξ0 ))
⎪
⎪
⎩ 0, ξo < ξ,
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 94

where M̂ = M − 1, ˆ = 1 − . For M = 1 the solution becomes piecewise linear and is

given by ⎧
⎪
⎪ 1, ξ ≤ ξi ,
⎪
⎪  
⎪
⎪ 1 ξ0 −ξ ˜
⎪
⎨ 2 2τ + 1 , ξi < ξ < ξ,
1 ˜
η [ξ, τ ;  < 1, M = 1] = 2
, ξ = ξ, (3.56b)
⎪
⎪ 
⎪
⎪ 1 ξ0 −ξ
+ 1 , ξ˜ < ξ ≤ ξo ,
⎪
⎪ 2 2τ (1−)
⎪
⎩
0, ξo < ξ,
where the positions of the outward and inward propagating tips of the rarefaction are
given by

˜ τ ˜
ξo = (1 − )τ + ξ, and ξi = − + ξ. (3.57)
M

For tilting interfaces in the parabolic limit we have shown that ξi is a stronger func-
tion of M than . In the hyperbolic limit the entire receding interface has lost its
dependence on . Where the interface is advancing, f is scaled by σ = 1 −  (fig-
ure 3.13a), and the residual reduces the propagation speed of the advancing interface
(figure 3.13b). This behaviour is analogous to the self-similar solutions in the par-
abolic limit. The solution for  = 1 is still given by (3.56), but the advancing interface
is replaced by a discontinuity at ξ˜ with a jump from 0 to ηmax . A comparison of fig-
ures 3.11b and figure 3.13b shows that M and  have a similar effect on the shape of
the interface in both the parabolic and hyperbolic limits.
For M = 0, equation 3.53 reduces to the linear advection equation

ητ + σηξ = 0, (3.58)

with a moving discontinuity in the velocity, σ. In this limit an ambiguity arises

in σ(ητ ), because ητ is singular at the discontinuity. The solution to (3.58) can
be obtained taking the limit of (3.56) as M → 0 and requiring that the solution
changes continuously with changes in the parameter M. In this limit the rarefaction
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 95

degenerates to a contact discontinuity


˜
1, ξ ≤ Λc τ + ξ,
η [ξ, τ ; M = 0] = (3.59)
0, Λc τ + ξ˜ < ξ,

where Λc = 1 −  is the speed of the discontinuity.

Shock Consider the piecewise constant initial data


˜
ηL , ξ ≤ ξ,
η[ξ, τ = 0] = (3.60)
˜
ηR , ξ > ξ,

where ηL < ηR . The weak solution of (3.53) with (3.60) is a shock of strength,
ηs = ηR − ηL , given by

˜
ηR , ξ ≤ Λτ + ξ,
η [ξ, τ ; M] = (3.61)
ηL , Λτ + ξ˜ < ξ,

where Λ is the shock speed. Similar to (3.58) an ambiguity arises, because ητ is singu-
lar at the shock. Therefore we define the shock speed considering (3.53) regularized
by a diffusion term
ητ + σ [ητ ] fη ηξ = νηξξ , (3.62)

where ν > 0. For initial data (3.60) the entropy-satisfying solution to (3.53) is defined
by the solution of (3.62) in the limit of small ν, the so-called vanishing viscosity solu-
tion. We assume the solution of (3.62) with (3.60) takes the form of a traveling wave
with constant speed Λν , so that η [ξ, τ ] = η̂ [ζ], where the traveling wave coordinate
is, ζ = ξ − Λν τ . For a steady traveling wave the time derivative that determines the
value of σ is given by
ητ = ζτ η̂ζ = −Λν η̂ζ . (3.63)

For initial data (3.60) we expect the solution to be monotonically increasing, η̂ζ > 0,
so that the sign of ητ is determined by the direction of wave propagation, given by the
sign of Λν . The constant wave speed implies that ητ does not change sign, and hence
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 96

σ is continuous, but its magnitude depends on the sign of the wave speed, σ = σ̂ [Λν ].
In the moving coordinate system (3.62) reduces to the ordinary differential equa-
tion with the boundary conditions η̂ → ηR for ζ → +∞, and η̂ → ηL for ζ → −∞.
After integrating once and eliminating the constant of integration with either bound-
ary condition, the following expression is obtained

σM (η̂ − ηL ) (ηR − η̂)

η̂ζ =    . (3.64)
ν M̂η̂ + 1 M̂ηL + 1 M̂ηR + 1

This confirms our assumption that η̂ζ > 0 for ηL < η̂ < ηR . Equation (3.64) does
not depend on the wave speed, because consistency with both boundary conditions
requires that the wave speed be given by

1 − M̂ηR ηL − ηL − ηR
Λν = σ   , (3.65)
M̂ηL + 1 M̂ηR + 1

which has already been substituted into (3.64). For M = 1 the solution to (3.64) is
  
1 (ζ − Cν)σ (ηR − ηL )
η̂ = ηL + ηR + (ηR − ηL ) tanh , (3.66)
2 2ν

where C is a constant of integration that can be determined by the initial position of

˜ and σ is given by (3.9). Equations (3.63) and (3.64) allow us to
the discontinuity, ξ,
write an equivalent definition of σ in terms of the sign of the wave speed

1, Λν > 0,
σ [ητ ] = σ̂ [Λν ] = (3.67)
1 − , Λν < 0.

This definition is not ambiguous at discontinuities in the limit ν → 0. With this

definition of σ the Rankine-Hugoniot jump condition for the shock is

f [ηL ] − f [ηR ]
Λ = σ̂ = Λν , (3.68)
ηL − ηR

so that the shock speed of the limiting solution is identical to the wave speed of the
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 97

10 -2 10 -2 10 -1
6 6 1
(a) (b) (c)
0. 5
4 4
σf σf Λ rec.
0
adv.
2 2
adv. −0. 5
rec.
0 0 −1
0 0.25 0. 5 0.75 1 0 0.25 0. 5 0.75 1 0 0.25 0. 5 0.75 1
η η ηL

Figure 3.14: Shock construction for M = 10. (a) In the absence of trapping,  = 0,
the construction for advancing and receding shocks is the same. (b) In the presence
of trapping,  = 1/2, the shock construction for advancing shocks is modified. (c)
Shock speed as function of ηL , for ηL > ηc the shock speed depends on the amount
of trapping.

viscous solution.
The geometric interpretation of the jump condition (3.68), is a cord joining f [ηL ]
and f [ηR ], as shown in figure 3.14a and b. For a given ηR and M the direction of
wave propagation as function of ηL is given by
⎧
⎪
⎪
⎨ < 0, ηc < ηL < ηR ,
Λν = 0, ηL = ηc , (3.69)
⎪
⎪
⎩ > 0, 0 < η < η ,
L c

where
1 − ηR
ηc = , (3.70)
(M − 1)ηR + 1
which corresponds to a horizontal tangent and a stationary shock.
The speed of shocks moving to the right is not affected by trapping, while the
speed of waves traveling to the left is reduced due to trapping (figure 3.14c). This
behavior of the shock is analogous to the behavior of the rarefaction. In both cases
the speed of the interface is reduced, by a factor 1 −  due to trapping, when it is
advancing, but it is not affected if the interface is receding. However, the parameter
that identifies whether the interface recedes or not is different for the rarefaction and
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 98

the shock. For an interface with a finite slope, i.e. a rarefaction, the change of the
height of the interface with time, ητ , determines which parts of the interface recede.
For a vertical interface, i.e. a shock, the sign of shock speed, Λ, determines whether
the interface recedes across the shock or not.

Wave interaction

The evolution of a finite release of fluid is governed by the interaction of the shock
at the back of the current with the rarefaction at the front of the current. In the
presence of trapping this interaction leads to a rapid reduction of current volume,
and give rise to a finite migration distance and time. Consider the piecewise constant
initial data 
1, |ξ| ≤ 1/2,
ηH = (3.71)
0, |ξ| > 1/2.

Initially the solution of the front is a rarefaction given by 3.56 with ξ˜ = 12 , for all
M > 0, and the evolution of the back is a stationary shock (A to B in figure 3.15)
with constant shock strength (figure 3.16b), because Λ = 0 for ηL = 0 and ηR = 1
for all M > 0. When the back end of the rarefaction reaches the shock, the shock
strength decreases, and it begins to move. The solution, η, will then be given by
(3.56), but η [ξ < ξs ] = 0, where the ξs is the shock position.

Shock evolution The waves begin to interact at the transition time, τt = M (B in

figure 3.15), and the shock speed begins to increase as the shock strength decreases.
Considering the shock strength, ηs , as a parameter along the shock path (B to D),
so that τ = τ [ηs ], and ξ = ξ [ηs ], we differentiate (3.55) and (3.68) to obtain two
differential equations

ξηs = ληs [ηs ] τ + λ [ηs ] τηs and ηs ξηs = f [ηs ] τηs . (3.72)

Eliminating ξηs we obtain the differential equation

dηs f [ηs ] − ηs λ [ηs ]

= , (3.73)
dτ ηs λη [ηs ] τ
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 99

for the evolution of the shock strength. Together with the initial condition ηs [τt ] = 1,
it determines the evolution of the shock strength (figure 3.16b). Once ηs is known,
the shock position is given by ξs = λ [ηs ] τ + 12 . We solve for ηs instead of ξs , because
(3.73) is always separable.
If trapping occurs,  = 0, then the evolution of ηs is divided into an early period,
τt ≤ τ ≤ τp , during which the shock interacts with the receding portion of the
rarefaction (B to C in figure 3.15), and a later period τp < τ ≤ τf , where the shock
interacts with the advancing portion of the rarefaction (C to D). Hence, the problem
has three time scales: the transition time, τt , when interaction between waves begins
(B ), the passing time, τp , when the shock passes the stationary point of the rarefaction
and starts to interact with the advancing section of the rarefaction (C ), and the final
migration time, τf , when the volume of the current goes to zero (D).
In the early period the shock strength and position are independent of  and are
given by

1 1 √ √ 2
ηs = √ and ξs = − + M− τ . (3.74)
1 − M + Mτ 2

We have shown in (§3.4.2) that the receding section of the interface is not affected by
trapping. Therefore, the shock strength and position are independent of the amount
of trapping. The early period extends until the shock migrates past the initial extent
of the current, ξs < 12 . The passing time and the shock strength at the passing time
are given by
 √ 2 1
τp = 1 + M and ηs [τp ] = √ . (3.75)
1+ M

In the absence of trapping (3.74) remain valid for all τ > τt (dashed line in figure 3.16).
After τp an explicit expression for ηs and ξs is only obtained for M = 1. We give the
full details for the case M = 1 to illustrate the structure of the solution, and for M > 1
we integrate (3.73) numerically, using (3.75) as an initial condition. For τp ≤ τ ≤ τf
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 100

16 50
(a) (b)
D D
40
12

30
τ 8 τ
20
C
4 C
10
B
B
A
0 A 0
-1.5 0F 2 4 6 8 10 -5 0F 10 20 30
ξ ξ

(c)

η 16
12
8
τ
0 4
-1 0 2 4 6 8 10 0
ξ

Figure 3.15: Characteristic portraits of the evolution of initial data (3.71), in the
ξτ -plane. The current is bounded by the shock, A to D, in the back and the leading
characteristic of the rarefaction, F to D, in the front. (a) Analytic solution for M = 1
and  = 0.4. (b) Semi-analytic solution for M = 5 and  = 0.4. (c) Evolution of the
current interface (black) and the residual surface (grey).
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 101

the shock strength for M = 1 is given by

⎧ 1 1
⎨ −2
1+
−1 −1−
t 2(
−1)
,  = 12 ,
ηs =  
2−1
(3.76)
⎩ log 2
√ + 12 ,  = 12 .
t

1
For  = 2
the right hand side of (3.73) loses the explicit dependence on ηs , leading to
a simplified expression. The expression for the shock path is given by
⎧  
⎨ (−1)
t−2
1
2+
−1
t
1
2−2
+ 12 ,  = 12 ,
2−1
ξs = (3.77)
⎩ 1 (log( t )t + 1),  = 12 .
2 4

In the second phase the shock speed and position depend strongly on the amount
of trapping. Figure 3.16a illustrates the effect of increasing trapping on the current
evolution in the ξτ -plane for M = 1, and similar behavior is observed numerically for
other M (figure 3.15b). From (3.77), (3.76), and (3.56) the aspect ratio of the current
is given by ⎧
⎪
ξo − ξs ⎨ 1 + (1 − )τ, τ < 1,
⎪
A= = 2τ − τ 3/2 , 1 < τ ≤ 4, (3.78)
ηs ⎪
⎪
⎩ 2(1 − )τ, 4 < τ < τf ,
for M = 1. Even in the presence of trapping A is always increasing for M ≥ 1. The
final residual surface, ηr , is the interface that divides the part of the aquifer that has
been invaded by the current, and now contains trapped fluid, from the region of the
1
aquifer that has not been invaded. The final residual surface for ξ < 2
is the initial
1
condition. For ξ > 2
the final residual surface is defined as ηr = ηs [ξs ] and can be
calculated by eliminating τ between (3.76) and (3.77) to obtain ξ = ξ [ηr ]. However,
it is not possible to invert this relationship to obtain an explicit expression for ηr
⎧  
⎨ 1
+ (−1) 1 1
ϕ(ηr ) − 22+
−1 ϕ(ηr ) 2−2
,  = 12 .
2 2−1
ξ= (3.79)
⎩ e1−2ηr (2 − 4ηr ) + 1 ,  = 12 .
2

  2
−1
1
−1
−
−1


were ϕ [ηr ] = 2 (−2ηr + ηr + ) . The shape of the final residual surface
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 102

16
(b) (a)

14

12

10

τ 8 4
0.

=
0
5

=
0.

6

=
.6

τp
0
0.7

=

4 C

=

=1

2
τt B
A F
0
1 0.5 0
1 (c)
ηs
0.5
ηr
0
-1 0 2 4 6 8 10
ξ

Figure 3.16: (a) Portrait of the shock path and the leading characteristic of the
rarefaction, for increasing residual. The locus of the termination points, τf = τf [ξ↑ ],
is shown as a dotted line. (b) The corresponding temporal evolution of the shock
strength, ηs . (c) The corresponding final residual surfaces, ηr = ηr [τf ].
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 103

20 20
(a) log(ξ↑) (b) log(τf)

15 8 15
8
6 6
M M
10 4 10
4
2
5 5
0 2
1 1
0 0.2 0.4 0.6 0. 8 1 0 0. 2 0.4 0.6 0.8 1
 

Figure 3.17: Contour plots of the up-dip migration distance, log(ξ↑ ), on the left and of
the total migration time, log(τf ), on the right. The data for several potential storage
aquifers in Alberta, Canada, is shown as circles to indicate typical parameter values
(Bachu & Bennion, 2007).

is shown in figure 3.16c for increasing residual saturations.

Final migration time and distance The finite migration distance is an impor-
tant characteristic of the hyperbolic limit, in contrast to the self-similar solutions for
the horizontal case. The maximum migration distance and the corresponding total
migration time are amongst the most important time and length scales for CO2 stor-
age, and the analytic and semi-analytic results (figure 3.17) presented here show how
these scales depend on the governing parameters.
The total migration distance, ξ↑ , is given by
 1 1
1
2
− 21+ 2
−1 ( − 1) 2
−1 −1 ,  = 12 ,
ξ↑ = 1
(3.80)
2
+ 2e,  = 12 .

Increasing mobility increases ξ↑ , because a gravity tongue forms at the top of the
aquifer and only a small fraction of the aquifer is swept and contributes to residual
trapping.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 104

The final migration time for M = 1 is given by

⎧  − 2(
−1)+1
2(
−1)
⎨ 1
−1−
−1
2  ,  = 12 ,
τf = (3.81)
⎩
4e,  = 12 .

The dimensionless final migration time increases strongly with increasing M. The
dimensionless time is only based on the properties of the released fluid. As M increases
the released fluid becomes increasingly less mobile, which retards the movement of
the released fluid.

Volume evolution The final migration time, τf , bounds the time available for
leakage from the storage reservoir. For smaller times the volume of mobile CO2 is a
basic quantity necessary to estimate maximum potential leakage over time. Analytic
expressions for the evolution of the current volume are given by integrating (3.56)
over the extent of the plume. The normalized current volume V = V (τ )/V (τ0 ) is
given by ξo √ −2
V= η dξ = 1 − τ  1 + M (3.82a)
ξs

for τ < τp . In this early period both the shock and the receding part of the rarefaction
are independent of , and therefore the area containing residual saturation is also
independent of . Hence, the current volume decreases proportional to . The decay
of the current volume slows down with increasing M, because the receding interface
slows down. This behavior that is also observed in semi-analytic solutions for τ > τp .
At later times the lower limit of the integral is only known for M = 1, and in this
case the current volume is given by
⎧  2
⎨ 1−
1
2− 1−
t 2(1−
) − t ,  = 12 ,
V= t(1−2)2 (3.82b)
⎩  t 2
t
8
log 4 − 1 ,  = 12 ,

for τ < τf = 4e. Figure 3.18(a) shows how strongly the current volume is affected by
the magnitude of . This evolution is distinctly different from the power-law decay
obtained from the self-similar solutions for horizontal aquifers. It suggests that sloping
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 105

1 1
(a) (b)

0.8 0.8

0.6 0.6
V V
0.4 0.1 0.4
1
0.2 5
0.2 0.3 0.2 10
15 20 40
0.4 25
0 0
0 25 50 0 100 200 300
τ τ

Figure 3.18: The effect of the governing parameters on the evolution of the current
volume. The passing and final migration times, are shown as circles and triangles.
(a) Increasing the residual, , from 0.1 to 0.9 at constant M = 1. (b) Increasing the
mobility ratio, M, from 1 to 40 at constant  = 0.4. Circles denote

aquifers reduce the current volume particularly effectively. An evolution similar to

(3.82) is observed for a wide range of M (figure 3.18b).
The residual volume is given by Vr = 1 − V, and from (3.82) we see that volume
of trapped CO2 initially increases linearly as Vr ∝ τ , and it increases more slowly
for τ > τp . The efficiency of residual trapping decreases with time, because the size
of the receding interface, ηs , decreases rapidly, once the shock is interacting with the
advancing section of the rarefaction.

3.4.3 Numerical results

In this section we present numerical solutions to (3.7) and show that the current evo-
lution for intermediate values of Pe can be understood in terms of an early essentially
parabolic period followed by a late near hyperbolic regime. We also discuss important
length and time scales of the solution as a function of increasing Pe. The numerical
solutions were computed using standard finite volume techniques and explicit time
integration (Leveque, 2002).
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 106

0.2 hyp.
(a) 1 0 num.
sim.
´ 0.1 M = 20 15 10 5

0
0 10 20 30 40 50 60
»
0.1
(b) hyp.
num.
´ ²
 = 0
0.1
0.2
0.3
0
10 20 30 40 50 60
»

Figure 3.19: This figure shows several comparisons between the numerical solution
(∆ξ = 0.01) to (3.7a) and the hyperbolic solution for Pe = 2 at τ = 50. (a) Increasing
M at constant  = 0. (b) Increasing  at constant M = 10.

Near hyperbolic behavior for large M

In CO2 storage Pe is not necessarily large, but the mobility ratio is generally larger
than unity, M ≈ 5. Here we show that the particular shape of f leads to near
hyperbolic behavior of (3.7a) for small Pe and large M. Consider the differentiated
form of (3.7a) given by

 
σ −1 ητ + ληξ = Pe−1 λ (ηξ )2 + f ηξξ . (3.83)

In the absence of source terms the thickness η of the current decreases monoton-
ically with time and for M 1 the current takes the shape of a triangle composed
of a very elongate, flat, advancing tongue in the front and a short, steep, receding
tongue in the back (figure 3.10g-h). This asymmetry increases with Pe, M, and τ .
The hyperbolic approximation introduced in § 3.4.2 replaces the short, steep, receding
interface of the current by a shock.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 107

t [yrs]
37 373 3’730 37’300 373’000

100 100

10-1 10-1

V
full equation:
Pe = 0 (θ = 0 ◦ )
◦
Pe = 0 .44 (θ = 12 )
◦
10-2 Pe = 0 .87 (θ = 1 ) 10-2
Pe = 4 .37 (θ = 5 ◦ )
Pe = 8 .82 (θ = 1 0 ◦ )
Pe = 1 3 .4 (θ = 1 5 ◦ )
limiting cases:
similarity soln.
hyperbolic soln.
10-3 10-3
10-1 100 101 102 103
ϑ = t/td = tκ1 H cos θ/L2

Figure 3.20: Evolution of the current volume, V, as a function of diffusive dimension-

less time, ϑ, for M = 5 and increasing Pe. The time axis at the top gives dimensional
values corresponding to the parameter choice discussed in § 3.5.2, assuming cos θ ≈ 1.

In the advancing tongue ηξξ  ηξ  1, and hence both the diffusive, f ηξξ , and the
second order term, ληξ2 are small. The diffusive term is reduced further over the entire
√
current, because max(f ) = 1/(1 + M)2 decreases with increasing M. The second
order term ηξ2 is not small in the steep receding back of the current, and figure 3.19
shows that the error in the hyperbolic approximation is localized there.
Figure 3.19(a) illustrates how the error introduced by the hyperbolic approxima-
tion decreases with increasing M. For M > 5 and τ > 2τp the only significant error in
the hyperbolic approximation is due to small off-set in the shock position. In the limit
M = 0 a similarity solution to the full equation is available (Huppert & Woods, 1995),
and it is shown to provide an indication of the accuracy of the numerical method used.
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 108

Once loss is introduced,  > 0, the small differences in the sweep of the full solution
and the hyperbolic solution lead to differences in plume volume at any given time.
This volume error accumulates over time and leads to differences between the shock
location in the hyperbolic model and the back of the plume in the numerical solution
(figure 3.19b).

Transition from parabolic to hyperbolic behavior

In gently sloping media the current transitions from initial parabolic behavior to
hyperbolic behavior at late times (figure 3.20). Due to the distinctly different behavior
of these limiting cases for  > 0 the transition is apparent in the evolution of the
current volume. An initial power-law decay of the current volume is followed by a
much more rapid decay characteristic of the hyperbolic limit.
For very large Pe the plume volume is given 3.82 and its semi-analytic extension
to M = 1. This limiting solution is plotted in figure 3.20 as a function of the diffusive
dimensionless time, ϑ = τ /Pe, so that changes in Pe correspond to a stretching of
the time axis. We observe that the limiting solution describes the rapid decay of the
current volume at late times, even for Pe < 1.

Length & time scales for gravity currents with residual

The length and time scales that are important for gravity currents in sloping layers are
the propagation distance against gravity, ξ↓ , and the time of the associated reversal
in the direction of the interface movement, τt . For a buoyant current the down-dip
propagation of the interface is due to the parabolic part of the equation. The reversal
of the interface movement indicates the transition from early near-parabolic to late
near-hyperbolic behavior. The continued volume loss and the resulting extinction
of the current give rise to additional length and time scales, the up-dip migration
distance, ξ↑ , and the final migration time, τf . Numerical results for all four scales as
a function of increasing Pe are shown in figure 3.21. To determine the scales associated
with the end of the current the numerical simulation was terminated at V = 10−3 .
This value was chosen to allow a reasonable detection of the gravity current tips,
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 109

2
20
(a) numerical: (b)
M=0
1.5 M=1 15
M=5
-»↓ M = 10
M = 15 ¿t
1 10
M = 20

0.5 5
similarity soln.
hyperbolic soln. 0
0
0 5 10 15 20 0 5 10 15 20
Pe Pe

100 160
(c) (d)

75 120

»↑ ¿f
50 80

25 40

0 0
0 5 10 15 20 0 5 10 15 20
Pe Pe

Figure 3.21: Time and length scales obtained from the numerical solution of (3.7)
are shown as symbols (∆ξ = 0.01). All results are for a residual of  = 0.4. The
limiting hyperbolic and self-similar solutions are indicated by full and dashed lines.
(a) Down-dip migration distance, ξ↓ . (b) Transition time, τt . (c) Up-dip migration
time, ξ↑ . (d ) Final migration time, τf .
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 110

which become strongly affected by numerical diffusion for smaller cut-off volumes.
The rapid decay of the plume volume at late times (figure 3.20) suggests that the
results are not very sensitive to the cut-off, and the reasonable agreement between
numerical and limiting analytic results in figure 3.17 confirms this.
The down-dip migration distance, ξ↓ , is not very sensitive to M or . For small Pe
√
it follows the relation τt = 1/ 6Pe obtained for the similarity solution for M =  = 0.
As the τt increases with decreasing Pe the current thickness decreases as well, and
the similarity solution for the unconfined aquifer becomes a better approximation.
As Pe increases ξ↓ approaches the position of the initial condition, which is also the
down-dip migration distance in the hyperbolic limit. The transition time, τt , appears
to approach the limiting hyperbolic solution much slower than the other parameters.
The up-dip migration distance, ξ↑ , and the final migration time, τf , show a more
complicated behavior as Pe increases. As Pe goes to zero both become very large,
and this is consistent with the late similarity solution for the limiting case Pe = 0,
introduced in section § 3.4.2. As Pe increases the hyperbolic limit is approached
quickly, except for M = 0. However, for Pe ≈ 0.2 − 0.5 a minimum in both quantities
is observed for M > 1. Hesse et al. (2006) showed that both ξ↑ and τf decrease with
increasing Pe for M = 0. Figures 3.17(c-d ) show a similar behavior of ξ↑ and τf for
M < 1, but we observe an increase of ξ↑ and τf for M > 1 for Pe > 1.

3.5 Discussion

3.5.1 Assumptions
The most limiting assumptions in the derivation of the governing equations are the
homogeneity of the porous medium, the uniform saturation in the CO2 plume, and the
incompressibility of the CO2 . While simple forms of heterogeneity in the across slope
direction can be included (Huppert & Woods, 1995) general forms of heterogeneity
require a full numerical solution. Ide et al. (2007) investigated gravitational spreading
of a CO2 plume in a horizontal, two-dimensional, heterogeneous aquifers with a reser-
voir simulator. They showed that the effect of capillary forces on residual trapping
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 111

is larger than the effect of moderate permeability heterogeneity. In the absence of

capillary forces, their numerical simulations resemble figure 3.10(a-d ). Simulations
including capillary forces lead to a thicker gravity current with a non-uniform vertical
saturation profile and an increase in capillary trapping (Mo & Akervoll, 2005).
It may therefore be interesting, to extend the model presented here to non-uniform
saturation distributions. The infinitely spreading solution obtained for horizontal
aquifers is a direct result of neglecting capillary forces. The current will stop prop-
agating once the hydrostatic potential at the tip of the plume decreases below the
entry pressure necessary for the CO2 to invade the largest pores. Even in very gently
dipping aquifers, the CO2 may rise vertically by several hundred meters, if the up-dip
migration distance is large. In these cases, the expansion of the CO2 will lead to an
increase in plume volume that may be significant, in particular if the CO2 becomes
gaseous.
In the hyperbolic model for sloping aquifers we neglect the second order term,
λ(ηξ )2 . This simplification leads to an unphysical vertical interface at the trailing
end of the current. Despite this simplification, we find that the resulting quasi-
linear hyperbolic model describes many characteristics of the current evolution very
well. The minimum in the migration distance and time observed in figure 3.21 is not
explained by the limiting cases we study. The most likely explanation is the down-dip
migration of the current against gravity. This down-dip migration is represented by
the second order term, λ(ηξ )2 we have neglected. Inclusion of this term leads to a
fully nonlinear hyperbolic model

σ −1 ητ + ληξ = Pe−1 λ (ηξ )2 , (3.84)

which is likely to exhibit this behavior. This model could still be solved by the method
of characteristics allowing fast and accurate solutions to be obtained by numerical
integration.
In terms of geological CO2 storage we have neglected the dissolution of CO2 into
the brine. Two types of dissolution must be distinguished, direct dissolution of CO2
into the residual brine and the enhanced dissolution of CO2 at the interface between
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 112

CO2 and brine due to convective currents in the underlying brine, see § 2. The direct
dissolution into the residual brine can be modeled in this framework, but the effect is
likely to be small compared with residual trapping. The enhanced dissolution due to
convective transport may become important at later times, because the gravity tongue
provides a large interfacial area. This process could be modeled conceptually through
an effective loss term. Detailed information regarding the convective dissolution rates,
similar to those presented in § 2.3, will allow us to estimate the magnitude of this loss
term. An extension of the model presented here would then allow for a comparison
of the relative contributions of dissolution and residual trapping.
The leading edge of the CO2 plume is both by gravitationally and viscously un-
stable and this leads to the formation of a gravity tongue in our solutions. Riaz &
Tchelepi (2006) have shown that this gravity tongue is also the dominant feature
in high resolution numerical simulations of two-dimensional vertical displacements.
The displacement may also be viscously unstable in the transverse direction, similar
to the Saffman-Taylor instability in a Hele-Shaw cell (Saffman & Taylor, 1958), or
the instability of viscous currents on inclined planes discussed by Lister (1992). The
linear instability of an immiscible displacement in a porous medium has been studied
by Yortsos & Hickernell (1989) and Riaz & Meiburg (2004a). However, these studies
consider advection dominated one dimensional base flows and not the two-dimensional
base flow in gravity-capillary equilibrium that is of interest here. The study of the
stability of the two-dimensional flows presented here is an interesting direction for
future research, but beyond the scope of this contribution.

3.5.2 Implications for CO2 storage in saline aquifers

We model the migration and the associated residual trapping of CO2 plumes in saline
aquifers using vertical-equilibrium sharp-interface models. The analysis of these sim-
plified models shows us how the most important dynamical parameters affect the
evolution of the footprint of the CO2 plume over time and the efficiency of residual
trapping. First, we compare the evolution of CO2 plumes in horizontal and sloping
aquifers. Then we consider a particular field case and show how our results have first
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 113

order implications on the selection of storage sites.

Residual trapping in horizontal and sloping aquifers

Let us consider a deep saline aquifer with the following constant properties: H ≈
20 m, k ≈ 30 mD, and φ ≈ 0.1. In addition, we assume the following constant
fluid properties: ∆ρ ≈ 300 kg/m3 , Scr = Sbr = 0.2, µc ≈ 0.06 · 10−3 , µb /µc ≈ 10,
krc ≈ 0.2, and at the advancing interface krb = 1. In this case, the mobility ratio is
M ≈ 5, so that we expect the formation of a gravity tongue of CO2 along the top
of the aquifer. Finally, we assume that the aquifer has been invaded by CO2 over a
distance of Ld ≈ 500 m, so that the aspect ratio at the end of injection is given by
A = 2 · Ld /H ≈ 50.

Horizontal aquifers The evolution of the CO2 plume in a horizontal aquifer with
these properties is shown in figure 3.10(a-d ), and the evolution of the volume of the
mobile CO2 plume is given in figure 3.20. For gravity currents with M ≈ 5 we expect
an extended early period where the tip position is given by x ∝ t1/2 (figure 3.9).
The regime diagram in figure 3.9 does not account for the effect of residual trapping,
but figure 3.12(b) shows that the transition from the early to the late scaling law is
similar for  = 0.25 and  = 0. We can therefore use the theory developed in § 3.3 to
estimate the duration of the early period, given by tb , and the time, tt , after which
we can expect the late scaling law to hold.
For M = 5, the dimensionless time scales are given by ϑb ≈ 4 and ϑt ≈ 265,
and the characteristic time is td = L2d (κ1 H)−1 ≈ 250 yrs. The interface will detach
from the aquifer after approximately tb ≈ 1000 yrs. At this time the footprint of the
CO2 plume is 4.6 km. In the period right after the detachment of the CO2 plume
residual trapping is the most effective, and the fraction of CO2 trapped as residual
saturation increases to 20% (figure 3.10b). The period that corresponds to rapid
residual trapping in a horizontal aquifer is approximately given by tb to 3tb .
The tip of the CO2 plume will continue to propagate proportional to t1/2 until
tt ≈ 66000 yrs, when the footprint of the plume is 20 km. Therefore, although the
CO2 plume shown in figure 3.10(c) is already very thin compared to the thickness of
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 114

the aquifer, the plume tips are is still propagating according to the early scaling law,
x ∝ t1/2 . This illustrates that the effects of confinement, expressed in the mobility
ratio, must be taken into account to understand the evolution of the CO2 plume. In
particular, confinement has an effect on the propagation of the gravity current long
after the CO2 plume has detached from the base of the aquifer at tb .
After tt the spreading of the CO2 plume slows down significantly, and in fig-
ure 3.10(d ) the tips of the plume propagate according to the late scaling law. In this
late stage, residual trapping decreases the propagation velocity in a manner similar
to figure 3.12. As the plume slows down, residual trapping decreases rapidly. From
figure 3.10(a) to (d ) we can see that it takes 3,000 yrs to trap the first 20%, about
30,000 yrs to trap the next 10%, and approximately 300,000 yrs to trap an additional
10%. This drastic decay of the rate of residual trapping in horizontal aquifers is a
consequence of the self-similar nature of the solution at late times, which requires
that the volume of the mobile CO2 plume decays as a power-law.

Sloping aquifers Figure 3.20 shows that the effectiveness of residual trapping in-
creases dramatically even for small dip angles. The evolution of the interface shape
for θ = 5◦ is shown in figure 3.10(e-h). We see that the migration time of the cur-
rent is reduced from approximately 40,000 yrs for a slope of 1/2◦ to approximately
2,000 yrs for a slope of 15◦ . The up-dip migration distance decreases from 85.6 km
at 1/2◦ to a minimum of 82.8 km at 1◦ and slowly increases again to 84.5 km at 15◦ .
As the slope of the aquifer increases the distance the plume can migrate before
it reaches the surface decreases. For an injection depth of 2 km, only angles of 1/2◦
and 1◦ prevent the CO2 from reaching the surface, and even for θ = 1/2◦ the CO2
rises 750 m vertically. Therefore, the depth of the current at the end of migration
is the important criterion for storage security. Given the dimensionless initial depth
d0 = D0 /L the dimensionless final depth is d = −d0 + ξ↑ sin θ. We can define the
critical angle at which the injected CO2 reaches the surface as
 
d0
θc = arcsin (3.85)
ξ↑
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 115

20 20
(a) θc (b) S
0.1° 0.01%
15 15
0.5°
1°
2° 0.1%
M 4° M
10 6° 10 1%
10°
20° 10%
5 35° 5
90° 50%
90%
1 1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
 

Figure 3.22: Contour plots of the critical angle, θc , assuming d0 = 2 and the sweep,
S, as function of M and . The data for several potential storage aquifers in Alberta,
Canada, is shown as circles to indicate typical parameter values (Bachu & Bennion,
2007).

Figure 3.22(a) shows that the critical angle, θc , increases with  and decreases with
M. For a given M and  the CO2 will not reach the surface for θ < θc . The optimal
storage aquifer therefore has a few degrees of slope, just enough to cause lateral up-
dip migration and the corresponding efficient residual trapping, but without reducing
the distance to the surface too much.
Despite rapid trapping compared to the horizontal case, CO2 storage in saline
aquifers is limited by the poor vertical sweep, S, and the resulting long migration
distances. The vertical sweep can be defined as S =
hr  /H =
ηr , where
hr  is the
average residual surface. The initial volume of mobile CO2 is given by φ(1 − Sbr )HL,
and the residual volume of CO2 at the end of the migration is given by φScr
hr  (x↑ −
x↓ ). An expression for the vertical sweep, S, can be obtained from the conservation
of volume and is given by

S= , (3.86)
ξ↑ − ξ↓
where ξ↓ ≈ 0 in the hyperbolic limit. Figure 3.22(b) shows that the sweep is a strong
function of , increasing monotonically from zero at  = 0 to unity for  = 1. For the
data from Bachu & Bennion (2007), the sweep, S, can vary from less than 10−3 to
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 116

more than 10−1 , roughly three orders of magnitude, while the residual, , itself only
varies between 0.1 and 0.7. Therefore, the residual CO2 saturation, Scr , influences
the up-dip migration distance given by (3.44) more strongly through its influence on
the sweep than through its direct volumetric effect.

Large scale CO2 storage: Carrizo-Wilcox aquifer in Texas

Nicot (2008) considered the storage of roughly a fifth of the CO2 emissions from
coal power stations in the state of Texas over the next 50 years. This would re-
quire the injection of approximately 370·106 m3 /yr of supercritical CO2 , equivalent
to 50 MtCO2 /yr, into the central section of the Carrizo-Wilcox aquifer. They envi-
sioned a line of 50 wells aligned perpendicular to the dip of the aquifer with a 1.6 km
spacing. These wells act as a 80 km long line source and our one-dimensional solution
is a reasonable approximation of the displacement away from the edges.
To obtain first order estimates of the migration distance of the injected CO2 we
assume the following constant aquifer properties estimated from the data given in
Nicot (2008): H ≈ 200 m, k ≈ 500 mD, φ ≈ 0.15, the average depth of injection
2.7 km, the average distance to the outcrop 100 km, and therefore and average dip
angle, θ ≈ 1.5◦ . We assume the following fluid properties: ∆ρ ≈ 300 kg/m3 , Scr =
Sbr = 0.2, µc ≈ 0.06·10−3 , µb /µc ≈ 10, krc ≈ 0.2, krb = 1 (at the advancing interface).
Given this data the appropriate length scale at the end of 50 years of injection is
L ≈ 10 km, so that the initial aspect ratio is A = L/H ≈ 50. The three governing
parameters are therefore M ≈ 5,  ≈ 0.25, and Pe = A tan θ ≈ 1.4. Figure 3.21 shows
that the migration distance and time for M = 5 are already close to the hyperbolic
limit for Pe = 1.4, so that we obtain ξ↑ ≈ 80 and τf ≈ 110 from figure 3.17.
The along slope migration distance would be approximately ξ↑ L = 800 km, so
that the migration distance exceeds the up-dip extent of the aquifer significantly.
Under the simplifying assumptions made here, this particular injection scenario does
not ensure that all CO2 would be trapped by residual saturation alone. The initial
length scale L would have to be reduced to 100 km/ξ↑ = 1.25 km to achieve residual
trapping of all injected CO2 . This could be achieved by increasing the width of the
injection zone from 80 to approximately 620 km. Another option would be to reduce
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 117

the volume injected into this aquifer from 50 to roughly 6.5 MtCO2 /yr, assuming the
original injection width of 80 km. In this case the CO2 would remain mobile for 1550
years after the end of injection.

3.6 Conclusion
We derive a vertical-equilibrium sharp-interface model governing the migration of an
immiscible gravity current in a two-dimensional confined aquifer. A simple model of
residual trapping allows the analysis of the relationship between the vertical sweep of
CO2 plume and residual trapping.
In the parabolic limit of the governing equations, corresponding to a horizontal
aquifer, the evolution of the CO2 plume is divided into three regimes. The mobility
ratio M is the parameter that determines the time scales separating these regimes.
We have obtained new similarity solutions in the variable ζ = x(κHt)−1/2 that are
valid at early times when the interface is tilting, due to a horizontal exchange flow.
These similarity solutions are a strong function of the mobility ratio M, and of the
residual, . In this regime the position of the tip of the interface is given by x ∝ t1/2 ,
so that the exponent is independent of . The numerical solution continues to follow
the early scaling law long after the interface has detached from one boundary. This
indicates that the finite thickness of the aquifer is important for gravity currents with
M > 1, and solutions that assume infinite depth are not valid until the current has
become very thin.
In the limits h → 0 and M → 0, the governing equation simplifies to the porous
medium equation. For  = 0, this equation admits a similarity solution in the variable
ς = x/(κV t)−1/3 , where the position of the tip of the interface is given by x ∝ t1/3 .
For  = 0, the exponent in is a function of , x ∝ tβ() . This dependence is given by a
nonlinear eigenvalue problem. We obtain an expression for the transition time tt from
the t1/2 to the tβ() scaling. The transition time, tt , increases monotonically with M,
but it is a weak function of M for M < 0.18, and it increases rapidly for M > 0.18. The
two self-similar regimes are separated by a transition period that is roughly centered
on tt . Numerical solutions of the governing partial differential equation are used to
CHAPTER 3. RESIDUAL TRAPPING & CO2 GRAVITY CURRENTS 118

describe the evolution during the non-self-similar transition period. Numerical results
show good agreement with the early and the late similarity solutions.
During CO2 storage in horizontal saline aquifers M ≈ 5, and we expect a prolonged
period where the current spreads according to the early scaling law. Residual trapping
will be most effective in the time just after the current detaches from the base of the
aquifer. In the long term limit, the the current spreads very slowly and residual
trapping becomes less important. Despite continued loss of volume, the current is
never reduced to zero and continues to spread. This behavior is an artefact of the
model that neglects capillary pressure.
In the hyperbolic limit of the governing equations, appropriate for dipping aquifers,
the current is described by a travelling wave with hysteresis. The leading edge of the
interface is a rarefaction, and the trailing edge is a shock. The current volume is
reduced continuously by the residual, and the volume is reduced to zero in finite
time. We have obtained expressions for the up-dip migration distance and the final
migration time of the current.
In both the parabolic self-similar solutions and the hyperbolic rarefaction and
shock we observe that residual trapping affects the advancing section of the interface
more strongly than the receding section of the interface. Although trapping occurs
at the receding interface, the advancing interface is affected more strongly because
less fluid is supplied to it from the receding section.
In gently sloping aquifers, an initial near-parabolic regime with power-law decay
of the volume is followed by a near-hyperbolic regime with very rapid volume decay.
Increasing the slope of the aquifer or the initial aspect ratio of the current reduces
the duration of initial parabolic period.
Our results suggest that lateral migration of the injected CO2 along the seal will
trap the CO2 relatively quickly as residual saturation. Residual trapping is optimized
in sloping aquifers with small mobility ratios and high residual CO2 saturations.
However, the long migration distances of CO2 , due to the formation of a gravity
tongue, may limit the volume of CO2 that can be stored in sloping regional aquifers.
Chapter 4

Multiscale simulation of
CO2 storage in saline aquifers

4.1 Introduction

4.1.1 Challenges in the numerical simulation

of CO2 storage in saline aquifers
The simulation of CO2 storage in saline aquifers is challenging because: the domi-
nant physical processes change as the plume evolves, the domain may be very large,
several key physical processes as well as geological heterogeneity require high spatial
resolution, and the uncertainty in the geological parameters is large.
During the relatively short injection phase, advection and gravitational forces are
expected to be important, while gravitational and capillary forces dominate the long
post-injection period. Finally, geochemical reactions that bind the CO2 in minerals
will become important. This change in the dominant processes over time limits the use
of highly specialized and efficient numerical methods, such as streamline simulations
(Jessen et al., 2005; Kovscek & Wang, 2005), or vertical equilibrium models discussed
in § 3, which exploit the dominance of particular terms in the governing equations.
The size of the simulation domain is another main challenge in the numerical
simulation of geological CO2 storage. In comparison to reservoir simulation for oil

119
CHAPTER 4. MULTISCALE SIMULATION 120

recovery, much larger domains have to be simulated and over much longer time pe-
riods. The CO2 plume may migrate laterally over 10’s to 100’s of kilometers in
regional aquifers, and toward the surface along wells and fractures. Nordbotten et al.
(2004) and Silin et al. (2006) showed that the interaction of the CO2 with shallower
aquifers and aquitards along the leakage pathway can reduce the leakage rate to the
atmosphere significantly. It may therefore be necessary to simulate fluid movement
in the overburden of the storage site, leading to even larger simulation domains. In
addition, large quantities of CO2 have to be injected to lead to a meaningful reduction
of CO2 emissions, and this may lead to displacements of the interface between brine
and potable groundwater on a basin scale (Nicot, 2008).
On the other hand, geological porous media, including saline aquifers, are hetero-
geneous on all length scales (Journel, 1986), and numerical simulation with very high
spatial resolution is necessary to capture the complex flow patterns that develop in
unstable displacements (Tchelepi & Orr , 1994). The simulation of CO2 storage also
requires high numerical resolution to capture the small length-scales associated with
important physical processes, such as diffusion and dispersion. In § 2 we show that
buoyancy driven fingers in the brine determine the dissolution rate of CO2 , and may
require even higher decimeter-scale resolution. In § 3 we show that the adverse mobil-
ity ratio leads to dynamical instabilities such as long thin gravity tongues and viscous
fingers that require sub-meter-scale resolution in numerical simulation. The length
and time scales associated with the physical mechanisms during CO2 storage may
span six orders of magnitude in a sloping high-permeability aquifer, and this range
of scales considers only the continuum description where Darcy’s law is thought to
apply.
Finally, the uncertainty in the description of the geological storage site requires
many simulations to assess the uncertainty in model predictions (Kovscek & Wang,
2005). Similarly, the updating of the geological model through monitoring data re-
quires a large number of simulations (Doughty et al., 2007). Current reservoir simula-
tion technology may not be able to provide adequate numerical resolution to capture
physical processes and at the same time allow for the reasonable exploration of the
CHAPTER 4. MULTISCALE SIMULATION 121

0 km

Carrizo Fm.
1 km

10 km

2 km

Figure 4.1: An example of a typical regional aquifer under consideration for CO2
storage, the Carrizo Formation in east Texas is shown in grey. The figure is vertically
exaggerated 16:1, so that the true aspect ratio is much larger than it appears in the
figure. The sketch is simplified from Nicot (2008).

uncertainty through the investigation of many equally probable geostatistical real-

izations of the storage site. We therefore aim to develop computationally efficient
solution algorithms that can provide high resolution locally to capture the fine-scale
details of the flow field in very large simulation domains.
The fraction of the simulation domain that requires these high resolutions is gener-
ally small and may be associated with several features, such as multiple CO2 plumes
and interfaces between brine and potable groundwater. Recently multiscale meth-
ods have been developed that can resolve the large-scale flow on a coarse grid, using
numerical basis functions and have the ability to reconstruct fine-scale flow features
locally. Multiscale methods offer the desired adaptivity in highly heterogeneous do-
mains through the selective reconstruction and updating of basis functions (Jenny
et al., 2005). We focus on the multiscale finite volume (MSFV) method introduced
by Jenny et al. (2003). The MSFV method is currently the only multiscale method
for modeling flow in porous media that has been extended to most of the physical
processes necessary for CO2 storage in saline aquifers: compressibility (Lunati &
Jenny, 2006; Zhou & Tchelepi, 2008), gravity (Lunati & Jenny, 2008), and wells (Lee
CHAPTER 4. MULTISCALE SIMULATION 122

et al., 2008).
Grids with high aspect ratios are common in reservoir simulation, because oil
reservoirs are kilometers in areal extent and tens of meters thick, leading to simula-
tion domains with high aspect ratios (100:1 in 2D). In addition, geological layering
and fractures can introduce severe anisotropy, which is often miss-aligned with the
computational grid. Grid aspect ratios and permeability anisotropy have similar ef-
fects on the numerical discretization (see § 4.3.1) and can lead to significant challenges
in numerical modeling.
Large regional aquifers extend for several tens or even a few hundred kilometers,
but they are at most a few hundred meters thick (figure 4.1), leading to even larger
aspect ratios than in reservoir simulation of oil and gas fields (1000:1 in 2D). There-
fore, the problems with accurate and robust numerical discretization of anisotropic
domains with high aspect ratios are even more relevant in CO2 storage than in con-
ventional reservoir simulation. The original MSFV method, however, is only robust
for uniform grids and isotropic fine-scale permeability (Kippe et al., 2007; Lunati &
Jenny, 2007). The development of a multiscale finite volume discretization that is
robust for large aspect ratios is therefore an important step towards the simulation
of CO2 storage at the field scale.

4.1.2 Incompressible two-phase flow in porous media

Using a simple model problem, we illustrate the structure of the governing equations
that have to be solved for field-scale simulation of CO2 storage. We show how an
elliptic equation for the pressure arises. The repeated solution of this equation is
numerically expensive and multiscale solution algorithms are necessary to solve the
very large problems arising in CO2 storage in regional aquifers.
We consider the case of immiscible two-phase (supercritical CO2 and brine) flow in
a heterogeneous porous medium. For incompressible fluids and rock, the conservation
CHAPTER 4. MULTISCALE SIMULATION 123

equation for each phase can be written as

∂Sc
φ + ∇ · uc = −fp , (4.1)
∂t
∂Sb
φ + ∇ · ub = −fp , (4.2)
∂t

where Si , ui , and fi denote the saturation, Darcy velocity and source term (well) of
fluid phase i, and φ is the porosity. The subscripts c and b for the phase denote CO2
and brine. The saturation, Si , is the volume fraction of phase i in the pore space, so
that Sc + Sb = 1. For our problem, the viscosities of the CO2 and brine phases can be
different, but the viscosity of each phase is assumed to be constant. We also assume
that capillarity and gravity effects are negligible, so that Darcy’s law in terms of the
phase mobility, λi , is given by

kri (Si )
ui = −λi ∇p = −K(x) ∇p, (4.3)
µi

where K(x) is the absolute permeability tensor field, p is the pressure, and kri and
µi are the relative permeability and viscosity of phase i. Summing the conservation
equations of both phases we obtain

∇ · u = f, (4.4)

where f = −fc − fb , u = uc + ub = −λ∇p is the total velocity, and the total mobility,
λ, is defined as  
krc (Sc ) krb (Sb )
λ(x, Si ) = K(x) + . (4.5)
µc µb
This system of equations is supplemented with appropriate initial and boundary con-
ditions. In reservoir simulation, the system is often assumed to be closed, so that
no-flow boundary conditions are imposed at the periphery.
The total velocity, u, defined above allows us to rewrite (4.1) in terms of the
fractional flow, fi = ui /u, of phase i. Incompressible two-phase flow in porous media
CHAPTER 4. MULTISCALE SIMULATION 124

is therefore governed by the following coupled elliptic-hyperbolic system of equations

−∇ · [λ(x, Si )∇p ] = f, (4.6)

∂Si
φ − ∇ · [fi (Si )λ(x, Si )∇p ] = −fi . (4.7)
∂t

Equations (4.6) and (4.7) are usually referred to as the flow and transport problems,
respectively. The flow problem is an elliptic equation for p, and the transport problem
is a hyperbolic conservation law for Si . They are coupled through the dependence of
λ on the saturations Si , which are a function of space and time. These equations are
representative of the type of system that must be solved accurately and efficiently by
a simulator for CO2 storage.
At every time step, tn , in the solution process (4.6) has to be solved for the pressure
field, and in this context the total mobility is only a function of space. Therefore, we
drop the dependence of λ on µp and krp in (4.5) and set λ(x, tn ) = K(x) to emphasize
its tensorial nature.

4.1.3 Multiscale methods for porous media

Natural porous media are heterogeneous at all length scales, and current advances
in data integration and subsurface description provide increasingly detailed descrip-
tions of the subsurface. Numerical methods for incompressible flow in porous media
therefore lead to discrete elliptic problems with highly oscillatory coefficients. Full
resolution of the fine-scale features of the coefficient in realistic problems is often too
expensive. In the last few years, several multiscale methods have been developed to
reduce the computational complexity by incorporating fine-scale features into a set of
coarse grid equations. A characteristic of multiscale methods for flow in porous me-
dia is that they allow a reconstruction of an approximate fine-scale solution from the
coarse solution. The fine-scale reconstruction of the flow field (p and u) is necessary
for the subsequent transport calculations, which are very difficult to upscale. Meth-
ods designed for flow in porous media include the multiscale finite element method
(Hou & Wu, 1997), the mixed multiscale finite element method (Chen & Hou, 2002;
CHAPTER 4. MULTISCALE SIMULATION 125

Aarnes, 2004), the two-scale conservative subgrid approach (Arbogast, 2002), the fi-
nite difference heterogeneous multiscale method (Abdulle & E, 2003), the variational
multiscale method (Juanes & Dub, 2008), and many others.
We discuss the multiscale finite volume (MSFV) method for elliptic problems,
which was introduced by Jenny et al. (2003, 2005, 2006). The method described in
§4.2 is identical to the original MSFV method, but we describe it from a different
perspective that motivated the new results and developments presented in §4.4 & §4.6.
For simplicity, we restrict the discussion to a two-dimensional (2D) orthogonal grid.
In § 4.1.2 we show that the pressure field in incompressible porous media flows is given
by the scalar elliptic boundary value problem (4.6). Introducing operator notation
the complete definition of the flow problem is

Lp(x) = f (x) (x ∈ Ω),

(4.8)
LΓ p(x) = g(x) (x ∈ Γ := ∂Ω).

Here x = (x, y), and Ω ⊂ R2 is a rectangular open domain with boundary ∂Ω. L
is the linear elliptic operator on Ω and LΓ represents the linear boundary operators.
The elliptic operator L is defined by

L[•] = −∇ · (K(x)∇•) , (4.9)

where the tensor K is symmetric positive definite. In problems of flow and transport
in porous media the pressure equation, (4.8), is coupled to one or more transport
equations, given by (4.7).

4.1.4 Fine-scale problem

The hyperbolic nature of the transport equations requires that any numerical ap-
proximation of L be conservative, so that the corresponding flux field matches source
terms exactly. The finite volume approximation of (4.8) is given by

Lh ph (xi , yj ) = fh (xi , yj ) ((xi , yj ) ∈ Ωh ),

(4.10)
LΓh ph (xi , yj ) = gh (xi , yj ) ((xi , yj ) ∈ ∂Ωh ).
CHAPTER 4. MULTISCALE SIMULATION 126

Here h is a discretization parameter defining the centers of the finite control volumes
or cells Gh := {(x, y) : xi = (i + 12 )h, yj = (j + 12 )h; i ∈ (0, ..., n − 1), j ∈ (0, ..., m − 1)},
where m and n are the number of cells in the x- and y-directions. We only consider
square grids, because simple forms of grid stretching and skewness can be absorbed
into an effective permeability tensor, see § 4.3.1. The control volume centered on
(xi , yj ) is denoted Ω̄hij , and the four interaction regions that connect it to the sur-
(d)
rounding eight neighbors are labeled Ω̃hij , where d ∈ (1, 2, 3, 4). When we discuss
a control volume and its associated interaction regions, we drop the i, j subscript
(figure 4.2). The discrete operator for the ij-th cell is given by


1
Lh ph (xi , yj ) = ma,b ph (xi + ah, yj + bh) = [ma,b ]h ph , (4.11)
a,b=−1

where we have dropped the i, j subscript on the discrete operator, Lh , and the coef-
ficients, ma,b , for clarity. For homogeneous problems it is convenient to express the
discrete operator using the stencil notation (Trottenberg et al., 2001)
⎡ ⎤
m−1,1 m0,1 m1,1
⎢ ⎥
Lh ph = [ma,b ]h ph = ⎢
⎣ m −1,0 m 0,0 m 1,0
⎥ ph .
⎦ (4.12)
m−1,−1 m0,−1 m1,−1
h

To represent a constant solution exactly the coefficients ma,b must satisfy


1
ma,b = 0. (4.13)
a,b=−1

A stencil with n non-zero coefficients will be referred to as a n-point stencil. The

discrete problem 4.10 gives rise to the linear system, Lh ph = fh , and the stencil,
[ma,b ]h , corresponds to the non-zero entries in a row of Lh for a cell in the interior of
Ωh .
CHAPTER 4. MULTISCALE SIMULATION 127

Pi-1,j+1 Pi,j+1 Pi+1,j+1

P4 P3
~ (2)
f2 f4
(1) ~ s3 f3
(1)
f4 f2
(2)
f1 f1
Pi-1,j s4 s2
Pi,j Pi+1,j
f3(3) (4)
f2

~ s1 f1
f2(3) f4(4) ~

Pi-1,j-1 Pi,j-1 Pi+1,j-1

P1 P2

a) b)

Figure 4.2: (a) Nine cells of the finite volume grid are shown by dashed lines. The
four dual interaction regions, Ω̃(1) to Ω̃(4) , which connect the shaded finite volume
cell in the center to its eight nearest neighbors are shown by solid lines. (b) A dual
interaction region is shown with the four coarse cell boundary segments labeled s1 to
s4 and the corresponding fluxes f1 to f4 .

4.1.5 Multipoint flux approximations

On rectangular grids with aligned anisotropy finite volume methods use central dif-
ferences to approximate the fluxes across cell boundaries. Such two-point approxi-
mations for the fluxes give rise to a 5-point stencil that gives a second-order approx-
imation to L. On general non-orthogonal grids, as well as for non-diagonal tensors,
two-point flux approximations are not a consistent approximation of (4.8), and lead
to errors that do not diminish with grid refinement. Multipoint flux approximations
(MPFA) have been introduced to give a more accurate approximation of the fluxes
in these cases (for an introduction, see Aavatsmark, 2002). The properties of porous
media can exhibit very strong anisotropies and robust discretizations, Lh , for these
cases have been discussed by several authors (Aavatsmark et al., 1996; Lee et al.,
1998; Edwards & Rogers, 1998; Aavatsmark, 2002; Lee et al., 2002; Nordbotten &
Aavatsmark, 2005; Nordbotten et al., 2007). Although other MPFA methods are
possible, we use the term for methods that approximate the fluxes using a subset of
the six nodes surrounding the face through which the flux is computed (figure 4.2).
This gives rise to a 9-point stencil that approximates the integral form of (4.10) in
CHAPTER 4. MULTISCALE SIMULATION 128

each cell Ω̄hij in the interior of Ωh

Lh ph = [ma,b ]h ph ≈ Lp dΩ. (4.14)
Ω̄h

With (4.13) the stencil [ma,b ]h has eight degrees of freedom. These eight coefficients
are determined by considering the fluxes induced across the cell boundaries by unit
sources in the eight neighboring cells (figure 4.2a). Every cell is linked to its eight
neighbors by four interaction regions, Ω̃(1) to Ω̃(4) . Each interaction region contains
four cell boundary segments s1 to s4 (figure 4.2b). When considering a single inter-
action region the pressures in the corners are labeled p1 to p4 as shown in figure 4.2b.
In this contribution we only consider MPFA schemes based on the interaction
regions shown in figure 4.2a. Other choices of the interaction regions are possible
and have been discussed by Lambers et al. (2008). The MPFA schemes considered
here assume that the fluxes f1 to f4 across cell boundaries in each interaction region
can be expressed as a linear combination of the four coarse pressures p1 to p4 in the
corners of the dual
⎛ ⎞ ⎛ ⎞⎛ ⎞
f1 t1,1 t1,2 t1,3 t1,4 p1
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ f2 ⎟ ⎜ t2,1 t2,2 t2,3 t2,4 ⎟ ⎜ p2 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ f ⎟=⎜ t ⎟⎜ p ⎟. (4.15)
⎝ 3 ⎠ ⎝ 3,1 t3,2 t3,3 t3,4 ⎠⎝ 3 ⎠
f4 t4,1 t4,2 t4,3 t4,4 p4

Therefore, we need to determine the sixteen coefficients of the transmissibility matrix

T = {tm,k } that represent fluxes induced across the four boundary segments sm by
the four pressures pk . Once the four matrices T(1) to T(4) are known the discrete
CHAPTER 4. MULTISCALE SIMULATION 129

operator Lh is given by Lee et al. (1998) as

(1) (1) (4) (4)

m1,0 = t1,2 + t4,2 − t4,3 + t3,3 (4.16)
(2) (2) (3) (3)
m−1,0 = t2,1 − t1,1 − t3,4 − t2,4 (4.17)
(1) (1) (2) (2)
m0,1 = t1,4 + t4,4 − t1,3 + t2,3 (4.18)
(3) (3) (4) (4)
m0,−1 = −t2,2 − t3,2 − t4,1 + t3,1 (4.19)
(1) (1)
m1,1 = t1,3 + t4,3 (4.20)
(2) (2)
m−1,1 = −t1,4 + t2,4 (4.21)
(3) (3)
m−1,−1 = −t2,1 − t3,1 (4.22)
(4) (4)
m1,−1 = t3,2 − t4,2 , (4.23)

where [ma,b ]h differs from Lee et al. (1998) by a minus sign due to the definition of
Lh . The form of these equations can be understood from figure 4.2a.
In this framework, any method that determines T will lead to a conservative dis-
cretization, Lh . Originally MPFA methods were developed for a full tensor permeabil-
ity field that is constant in every control volume, Ω̄h , and analytical considerations
have been used to obtain the entries in T (Aavatsmark, 2002). In reservoir engi-
neering the tensor permeability field is usually computed using upscaling techniques
(Durlofsky, 1991).

4.2 Multiscale finite volume method

The spatial structure of K is often very complex, and it may contain discontinuities
of several orders of magnitude as well as patterns on many different length scales.
The grid size, h, for (4.10) must be chosen to resolve all features of the tensor K that
are considered important for the subsequent solution of the transport equations. In
the case of flow and transport in porous media this is often the geostatistical grid
(Journel, 1986) and in the context of the multiscale method it is referred to as the
fine grid. In applications of petroleum engineering and groundwater hydrology the
fine grid is generally too large to allow direct numerical simulation at reasonable cost.
CHAPTER 4. MULTISCALE SIMULATION 130

The MSFV method addresses this problem by a three step procedure (Jenny et al.,
2003): 1) Local numerical fine-scale solutions in each interaction region, are used to
construct a coarse operator that accounts for fine-scale information on a coarser grid.
2) Problem 4.8 is solved on the coarse grid with the coarse operator. 3) A locally
conservative fine-scale flux field is reconstructed solving local fine-scale problems in
each coarse cell. The MSFV flux reconstruction on the fine-scale is consistent with
the fluxes of the coarse pressure field. The MSFV method is distinguished from
other upscaling methods by the construction of the coarse operator and the locally
conservative flux reconstruction that is consistent with the coarse operator.
The geostatistical or fine-scale permeability field is commonly isotropic Journel
(1986), although skewed and stretched grids can introduce an effective fine-scale per-
meability anisotropy (see § 4.3.1). All multiscale methods introduced in § 4.1.3 con-
sider isotropic fine-scale permeability fields on rectangular grids, corresponding to
a diagonal effective fine-scale permeability tensor field. The multiscale method dis-
cussed here also assumes a diagonal fine-scale permeability tensor, except in § 4.4
where fine-scale anisotropy is introduced to allow an analysis of the MSFV method
for the homogeneous case.

4.2.1 The coarse operator

The coarse elliptic problem is defined by

LH pH (xi , yj ) = fH (xi , yj ) ((xi , yj ) ∈ ΩH ),

(4.24)
LΓH pH (xi , yj ) = gH (xi , yj ) ((xi , yj ) ∈ ∂ΩH ).

The coarse grid has a grid size H = nc h, where the integer nc is the coarsening ratio,
assumed to be the same in both directions. In porous media applications values of nc
between 3 and 15 are a common choice, so that the coarse linear system can be up
two orders of magnitude smaller than the fine system in two dimensional problems
(Jenny et al., 2003). The centers of the coarse finite control volumes are defined by
GH := {(x, y) : xi = (i + 12 )H, yj = (j + 12 )H; i ∈ (0, ..., N − 1), j ∈ (0, ..., M − 1)}.
The coarse control volume centered on node (xi , yj ) is denoted by Ω̄Hij . The coarse
CHAPTER 4. MULTISCALE SIMULATION 131

grid operator of the MSFV method in stencil notation is given by

LH pH = [ma,b ]H pH , (4.25)

and the corresponding coarse linear system is denoted by LH pH = fH , so that [ma,b ]H

contains the non-zero entries in a row of LH in the interior of ΩH . The coarse operator
LH of the MSFV method incorporates the fine-scale information of K through a series
of local elliptic solutions on portions of the fine grid that allow the determination of
the coefficients in the coarse stencil [ma,b ]H . Even if K is diagonal on the fine-scale, the
heterogeneity may give rise to an effective anisotropy on the coarse scale (Durlofsky,
1991; Wu et al., 2002; Jenny et al., 2003; Chen et al., 2007). The coarse operator LH
of the MSFV method is designed to capture this effective anisotropy on the coarse
scale. This contribution focuses on the construction of LH from the local elliptic
problems and the resulting properties of the coarse operator LH .
The control volumes shown in figure 4.2a correspond to the coarse cells Ω̄Hij in
(i)
the MSFV method and the interaction regions form the dual cells Ω̃Hij . These coarse
grids are underlain by a fine grid containing the detailed structure of K. Numerical
solutions of local elliptic problems in the dual interaction regions are used in the
MSFV method to obtain the entries of T(i) . In the MSFV method the entire set
of fine fluxes along each boundary segment, sm , is necessary for the reconstruction
(§4.2.3), therefore the vectors of fine transmissibilities are stored instead of the total
transmissibilities (figure 4.3). In analogy to (4.15) the matrix of fine transmissibilities
* +
T∗ = t∗m,k is defined as
⎛ ⎞
t∗1,1 t∗1,2 t∗1,3 t∗1,4
⎜ ∗ ⎟
⎜ t2,1 t∗2,2 t∗2,3 t∗2,4 ⎟
T =⎜
∗
⎜ t∗ t∗ t∗ t∗
⎟.
⎟ (4.26)
⎝ 3,1 3,2 3,3 3,4 ⎠
t∗4,1 t∗4,2 t∗4,3 t∗4,4

The t∗m,k are column vectors and the matrix is of size (4 · ns ) × 4, where ns is the
number of fine cells on the coarse boundary segment. The transmissibilities on the
coarse-scale, T, can be obtained by summing the column vectors in T∗ . Therefore, the
CHAPTER 4. MULTISCALE SIMULATION 132

coarse operator of the MSFV method, LH , is defined analogous to MPFA methods,

and the coefficients of the coarse stencil [ma,b ]H are given by (4.16) to (4.23). The
LH is therefore similar to discretizations Lh for non-diagonal tensors on the fine grid.
There is a large body of work in this field, and we use these results to understand the
properties of LH in the MSFV method.

4.2.2 Local elliptic problems — basis functions

In the general heterogeneous case T(i) will be different in every dual interaction re-
gion, but the determination of tm,k in the MSFV method is identical in all four dual
interaction regions. Therefore, we restrict the discussion to a single dual, and we drop
the superscript indicating the dual. The sixteen degrees of freedom tm,k in each dual
are determined by any four linearly independent elliptic solutions, φhl , in Ω̃H . The
l-th local solution in the dual interaction region is given by

L̃h φhl = 0 ((xi , yj ) ∈ Ω̃H ),

(4.27)
L̃Γhl φhl = g̃hl (pk ) ((xi , yj ) ∈ Γ̃ := ∂ Ω̃H ).

Here the linear operator L̃h is the restriction of Lh to the open domain Ω̃H ⊂ Ωh , and
L̃Γh is the linear boundary operator. The local support of the L̃h requires that artificial
boundary conditions are imposed on Γ̃. The choice of L̃Γhl is an important design choice
in the MSFV method. The two most common choices for L̃Γhl , are the linear boundary
condition and the oscillatory boundary condition, both were introduced by Hou &
Wu (1997) in the context of the multiscale finite element method.
For any vector pk that contains the solution in the corners of Ω̃H , the boundary
condition g̃hl (pk ) interpolates the solution values onto the sides of Ω̃H . The simplest
choice is linear interpolation, in this case L̃Γhl := I, where I is the identity and the
boundary condition is given by

φhl = g̃hl (pk ) = (1 − xb /H) pn + xb /Hpn+1 (4.28)

CHAPTER 4. MULTISCALE SIMULATION 133

0 0 0 0 0 1 1 0

1 0 0 1 0 0 0 0

t4,1 t2,2
t3,1 t3,2 t3,3 t3,4
t2,1 t4,2

t1,1 t1,2 t4,3 t2,4

t1,3 t1,4
t2,3 t4,4

Figure 4.3: The top row shows contour plots of the four standard basis functions of the
MSFV method, calulated numerically for a heterogeneous, isotropic fine permeability
distribution. The second row shows the underlying dual interaction region with the
fine grid and the fine fluxes t∗m,k along the coarse boundaries (small arrows) and the
total coarse fluxes tm,k integrated over each face (big arrows).

where n ∈ (1, 2, 3, 4) is circular, and xb is the appropriate coordinate along the bound-
ary. The oscillatory boundary condition is obtained by solving a reduced elliptic
problem along the boundary of Ω̃H . The reduced boundary operator is defined by

L̃Γhl φhl = − ∂x∂ b kb ∂xbj = 0,

∂φ
(xb ∈ Γ̃n )
(4.29)
φhl (xb = 0) = pn , φhl (xb = H) = pn+1

where n ∈ (1, 2, 3, 4) is circular, kb is the component of the permeability tensor along

the boundary, and Γ̃n is the boundary segment between pn and pn+1 . In the context of
the MSFV method oscillatory boundary conditions are referred to as reduced bound-
ary conditions, and they are used to improve the accuracy of the MSFV solution in
problems with strong heterogeneity (Jenny et al., 2003).
While the reduced boundary condition is optimal for space-separable problems,
either choice introduces an error localized in a boundary layer, δ, near Γ̃. It has been
CHAPTER 4. MULTISCALE SIMULATION 134

shown that local oversampling can reduce this error at higher computational cost
(Hou & Wu, 1997; Hou et al., 1999). In all results presented in this paper the domain
has not been oversampled, but all developments presented below can be used with
oversampling.
In the MSFV method these four local solutions are interpreted as basis functions,
and the vectors of corner pressures are chosen as pk = [δ1k , δ2k , δ3k , δ4k ]T , where δik is
the Kronecker delta. In the homogeneous case the solution to (4.27) reduces to the
bilinear basis functions known from finite element methods. We also use the term
bilinear local problem or basis function for heterogeneous problems with the same pk .
Similarly we use the term linear basis function or local problem for cases where the
homogeneous limit admits a linear solution. Figure 4.3 shows these solutions for a
heterogeneous, isotropic permeability field. From each of the four local solutions we
obtain a vector of fine transmissibilities, t∗m,k , across each of the coarse cell boundary
segments. These vectors contain the fine transmissibilities ordered by increasing x or
y-coordinate. At the boundaries the dual interaction regions overlap by a fine cell so
that the corresponding entries in t∗m,k are half of the associated fine-scale flux. The
, s ∗
coarse transmissibilities are given by tm,k = ni=1 (tm,k )i , where ns is the number of
fine cells along the coarse boundary segment. The four local problems give a total
of sixteen constraints that determine T uniquely in a given dual interaction region
Ω̃H . The MSFV basis functions are a particularly convenient way of determining the
entries of T, because the entries can be obtained directly by integrating the fluxes in
the local problems along the coarse cell boundaries (figure 4.3). In general any set of
four linearly independent local pressure solutions can be used to determine T.
(d)
Once the transmissibility matrices are known for all dual interaction regions Ω̃Hij ,
the coarse operator can be assembled. The coarse solution pH is obtained from the
coarse linear system, LH pH = fH .

4.2.3 Reconstruction of the fine-scale flux field

A conservative fine-scale flux field can be reconstructed from pH , solving Neumann
problems in each coarse cell (Gautier et al., 1999; Jenny et al., 2005). The Neumann
CHAPTER 4. MULTISCALE SIMULATION 135

problem for a particular coarse cell Ω̄H is defined as

L̄h p̄h = 0 ((xi , yj ) ∈ Ω̄H ),

(4.30)
L̄Γh p̄h = ḡ(pH ) ((xi , yj ) ∈ Γ̄ := ∂ Ω̄H ).

Here the linear elliptic operator L̄h is the restriction of Lh onto the open domain
Ω̄H ⊂ Ωh , and L̄Γh is the linear boundary operator. To obtain a fine reconstruction
that is locally conservative on Ωh the linear boundary operator L̄Γh must impose the
fluxes on Γ̄, and is hence defined by

∂ p̄h
L̄Γh p̄h := −kn = ḡ(pH ), (4.31)
∂xn

where xn is the direction normal to the boundary, and kn is the permeability in

direction xn . The fluxes ḡ(pH ) are informed by the coarse solution pH . Jenny et al.
(2003) use the superpositon of a second set of basis functions to obtain ḡ(pH ). This
method becomes inefficient, if the basis functions have to be updated frequently.
Lunati & Jenny (2006) avoid the second set of basis functions and extract ḡ(pH )
from the interpolation of the coarse solution onto the fine grid, using the MSFV basis
functions.
If the MSFV method is formulated in terms of T∗ as above, the fluxes on the
boundary ḡ(pH ) can be obtained directly without explicitly interpolating the coarse
solution. In each of the four dual interaction regions, Ω̃, surrounding the coarse cell,
Ω̄, the fine fluxes on the coarse boundary segments s1 to s4 can be evaluated from
⎛ ⎞ ⎛ ⎞⎛ ⎞
f1∗ t∗1,1 t∗1,2 t∗1,3 t∗1,4 p1
⎜ ∗ ⎟ ⎜ ∗ ⎟⎜ ⎟
⎜ f2 ⎟ ⎜ t2,1 t∗2,2 t∗2,3 t∗2,4 ⎟ ⎜ p2 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ f ∗ ⎟ = ⎜ t∗ t∗ t∗ t∗ ⎟⎜ p ⎟, (4.32)
⎝ 3 ⎠ ⎝ 3,1 3,2 3,3 3,4 ⎠⎝ 3 ⎠
f4∗ t∗4,1 t∗4,2 t∗4,3 t∗4,4 p4
- ./ 0 - ./ 0- ./ 0
(4·ns )×1 (4·ns )×4 4×1

where ns is the length of the boundary segments and (p1 , p2 , p3 , p4 )T are the values
of the coarse solution pH in the corners of the dual interaction region. Once the fine
CHAPTER 4. MULTISCALE SIMULATION 136

local local
maximum maximum

a) finsecale solution b) conservative solution c) interpolated solution

Figure 4.4: The fine-scale solution as well as the reconstructed conservative and the
interpolated multiscale solutions are shown for the same problem. The log of the
permeability is shown at the base of each figure. A unit pressure gradient is applied
in the x-direction, and homogeneous Neumann boundary conditions are applied in
the y-directions. (a) The fine-scale solution, ph . (b) The conservative solution from
the flux reconstruction on the fine grid, p̄h . (c) The coarse solution, pH , interpolated
onto the fine grid using the MSFV basis functions, p̂h . In (b) and (c) the coarse
solution, pH , is shown as large black dots.

fluxes in all dual interaction regions are known, the fluxes on the entire boundary
of the coarse cell ḡ(pH ) can be assembled. Again, care should be taken to properly
treat the fine-scale fluxes at the boundaries of dual interaction regions; these fluxes
receive contributions from two interaction regions. To specify the unknown constant
in (4.30) the fine solution is set to the coarse solution at the center of the coarse cell.
The reconstructed solution is conservative on the fine grid, and the corresponding
fine flux field can be used in transport calculations.
In §4.4.2 the coarse operator, LH , is modified by changing the entries of T∗ ,
and it is important to maintain consistency between the coarse operator and the
reconstruction. The reconstruction of the flux boundary condition given in (4.32) is
always consistent with LH .
CHAPTER 4. MULTISCALE SIMULATION 137

4.2.4 Properties of the MSFV solution

Figure 4.4a shows the reference fine-scale solution for a heterogeneous permeability
field. Figure 4.4b shows that the solution obtained from the reconstruction of fine
fluxes, p̄h , is discontinuous at the coarse cell boundaries. These discontinuities are
not merely a function of the arbitrary constants in problem (4.30). There is no set of
pressure constants that eliminates the discontinuities in the reconstructed solutions.
The presence of strong heterogeneity leads to peaks at ∂ Ω̄H , if the boundary condition
forces a high flux into a fine cell with low permeability. The magnitude of these peaks
in the fine solution is not bounded by the surrounding coarse solution. This leads
to the loss of monotonicity in the reconstructed fine-scale solution even if the coarse
solution is monotone. The variational multiscale method exhibits a similar loss of
monotonicity in the fine-scale pressure reconstruction Juanes & Dub (2008). Despite
these undesirable features of the reconstructed fine solution, the associated flux field
is generally a good approximation, and it has been used successfully in transport
calculations (Jenny et al., 2003; Juanes & Dub, 2008).
Figure 4.4c shows the interpolation of the coarse solution onto the fine grid using
the bilinear basis functions. This solution is denoted p̂h and defined in each dual
interaction region, Ω̃H , by

4
p̂h = φhl pHl , (4.33)
l=1

where pHl are the coarse pressures in the corners of Ω̃H .

This interpolated solution is not necessary for the MSFV method described above,
but has some attractive properties. Given a monotone coarse solution, pH , the in-
terpolation, p̂h , is monotone, because each of the numerical basis functions used for
interpolation is monotone. The interpolation p̂h is not conservative along the bound-
aries of the dual interaction regions on the fine grid. The fine fluxes derived from the
interpolation p̂h are therefore not locally conservative and cannot be used for trans-
port calculations. The interpolation p̂h is only continuous for pure Dirichlet local
boundary conditions in (4.27) without oversampling.
CHAPTER 4. MULTISCALE SIMULATION 138

4.3 Anisotropy and monotonicity

Kippe et al. (2007) and Lunati & Jenny (2007) have shown that the MSFV method
leads to unphysical oscillations in the coarse solution and unphysical recirculation in
the reconstructed fine-scale velocity field, if the aspect ratio of the grid is not close
to unity. Figure 4.4b & c show a local maximum in the coarse solution of the MSFV
method inside the domain even for an isotropic, heterogeneous problem.

4.3.1 Effective permeability tensor

In § 4.1.4 we introduced the discrete fine-scale operator, Lh for a uniform square grid.
A uniform rectangular grid with permeability tensor Kr is equivalent to a uniform
square grid with an effective permeability tensor, K, given by
     
a c 1 ∆y 0 ∆y 0
K= = Kr , (4.34)
c b ∆x∆y 0 ∆x 0 ∆x

where ∆x and ∆y are the length of the finite control volumes on the rectangular
grid. The relation between the aspect ratio, A = ∆x/∆y, used by Kippe et al.
(2007) and Lunati & Jenny (2007), and the effective permeability tensor used here is
therefore a/b ∝ A−2 . In all heterogeneous numerical examples Kr is isotropic and K
is diagonal, in this case a/b = A−2 .
To understand the loss of monotonicity of the MSFV method we need to under-
stand the effect of coarse-scale anisotropy introduced by heterogeneity on the fine
scale. However, it is difficult to analyze the properties of multiscale methods for
heterogeneous cases with long correlation length. Durlofsky (1991) shows that the
effect of the fine-scale heterogeneity on the coarse scale is equivalent to a homoge-
neous anisotropic fine-scale permeability. This equivalent fine-scale permeability is
generally not aligned with the grid, c = 0. In § 4.4 we use this analogy to study the
properties of the coarse operator of the MSFV method on a homogeneous anisotropic
fine-scale permeability field. Numerical examples in § 4.7 show that the behavior of
MSFV for these problems gives a good indication of its accuracy and robustness for
CHAPTER 4. MULTISCALE SIMULATION 139

−3 −5
x 10 x 10
0.04 pH 1 4
a) b) c)
ph
0.03
pH , ph
2
0.02 0.5

0.01 0
0 0
0 0.5 1 0 0.5 1 0 0.5 1
x x x

Figure 4.5: Three solution profiles in the x-direction located at y = 1/2 are shown
for increasing values of the ratio a/b with c = 0. (a) anisotropy ratio, a/b = 1 > 1/3,
corresponding to A = ∆x/∆y = 1 . (b) anisotropy ratio, a/b = 1/100 < 1/3,
corresponding to A = 10. (c) anisotropy ratio, a/b = 1/2500 < 1/3, corresponding
to A = 50.

problems with anisotropy resulting from fine-scale heterogeneity.

Therefore, we introduce a rotated fine-scale permeability tensor only for the analy-
sis of homogeneous problem in § 4.4, while we assume K to be diagonal otherwise.
Since K is symmetric positive definite, the components of the effective tensor are
required to satisfy a > 0, b > 0, and c2 < ab. So that a, b, and c can also be expressed
as a rotation of a diagonal tensor
     
a c cos θ sin θ λ1 0 cos θ − sin θ
= , (4.35)
c b − sin θ cos θ 0 λ2 sin θ cos θ

where λ1 and λ2 are the principle components of the effective tensor (λ1 ≥ λ2 ) and θ
is the angle between the x-axis and the eigenvector associated with λ1 .

Loss of monotonicity

Incompressible flow in porous media leads to elliptic problems with strongly heteroge-
neous anisotropic coefficients. Many numerical discretizations of the elliptic operators
lead to unphysical oscillatory solutions, and the design of robust schemes has received
a lot of attention. Generally it is not possible to construct a 9-point discretization
that is locally conservative, exact for linear pressure fields, and monotonic for all
CHAPTER 4. MULTISCALE SIMULATION 140

admissible a, b and c (Nordbotten et al., 2007). Diagonal dominance and the M-

matrix property together ensure the monotonicity of the discrete scheme, where an
M-matrix is defined as a nonsingular matrix A = {aij } whose off-diagonal elements
are non-positive (aij ≤ 0 for i = j) and whose inverse is monotone, A−1 > 0 (Bunse
& Bunse-Gerstner, 1985). Nordbotten et al. (2007) have obtained less restrictive con-
ditions sufficient for monotonicity. For homogenous uniform grids considered in this
section their results reduce to

m0,0 > 0, (4.36)

max(m1,0 , m0,1 ) < 0, (4.37)
m0,0 + 2 max(m1,0 , m0,1 ) > 0, (4.38)
m1,0 m0,1 − max(m1,1 , m−1,1 )m0,0 > 0. (4.39)

Kippe et al. (2007) and Lunati & Jenny (2007) have reported unphysical oscillatory
solutions that occur when the MSFV method is applied to heterogeneous anisotropic
problems. In heterogeneous domains it is not clear whether the loss of monotonicity is
caused by the assumptions on the boundary condition of the local problems 4.27, or by
the properties of LH itself. The following example will show that pH has oscillations
even in homogeneous domains, indicating loss of monotonicity of the coarse operator
LH .
Consider (4.10), with Ωh = (0, 1) × (0, 1), a square grid of 135 × 135 fine cells,
homogeneous Dirichlet boundary conditions, so that LΓh = I and gh = 0, the source
term 
1 1/3 < x, y < 2/3,
fh = (4.40)
0 otherwise,

and aligned anisotropy (c = 0). The source term is non-negative, fh ≥ 0, in Ωh

and therefore ph should be non-negative and have no local minimum in Ωh by Hopf’s
maximum principle (Strauss, 1992). For grid aligned anisotropy the discrete operator
on the fine grid Lh is the standard 5-point stencil. Figures 4.5a-c show that ph obeys
Hopf’s maximum principle, because the 5-point stencil satisfies conditions 4.36 to 4.39
for all ratios a/b.
CHAPTER 4. MULTISCALE SIMULATION 141

5 0
a) b)
-1

0
-2 λ /λ = 2.5: O(1)
1 2
MSFV
m1,0

m1,1
λ /λ = 5.0: O(1)
1 2
-3 MSFV
-5
λ /λ = 7.5: O(1)
1 2
-4 MSFV
λ1/λ2 = 10 : O(1)
MSFV
-10 -5
0 45 90 135 180 0 45 90 135 180
θ θ

Figure 4.6: Numerical results for the stencil components m1,0 and m1,1 of the MSFV
method are compared to the analytic expressions of the MPFA O(1) method. The
components are shown as a function of the orientation of the tensor, θ, for several
ratios of the principal components, λ1 /λ2 .

Consider the same problem solved with the MSFV method with a coarsening ratio
nc = 5, giving 27 × 27 coarse cells. Figures 4.5a & b show that the coarse solution pH
is monotonic for a/b close to unity, but figure 4.5c shows that oscillations and local
minima occur as a/b decreases further.

4.4 Coarse operator for homogeneous problems

The properties of the MSFV coarse operator are analyzed with respect to anisotropy
and monotonicity. This section focusses on homogenous anisotropic permeability
fields. In this case the oscillatory local boundary conditions (4.29) reduce to linear
interpolations (4.28).

4.4.1 Original coarse operator

In §4.1.5 we show that the MSFV coarse operator is completely determined by T
and (4.16) to (4.23). In the homogenous case the transmissibilities tm,k of the MSFV
method are determined by the four bilinear basis functions in the dual interaction
CHAPTER 4. MULTISCALE SIMULATION 142

region. For grid aligned anisotropy, c = 0, the basis functions are given by

φ1 = (1 − x)(1 − y), φ2 = x(1 − y), φ3 = xy, and φ4 = (1 − x)y.

The analytical solutions for the bases functions are not known for anisotropy not
aligned with the grid (c = 0), but these basis functions can be calculated numerically.
The discussion is therefore, initially restricted to problems with aligned anisotropy,
and in this case the matrix T is given by
⎛ ⎞
3a −3a −a a
⎜ ⎟
1⎜ b 3b −3b −b ⎟
T= ⎜
⎜
⎟. (4.41)
8⎝ a −a −3a 3a ⎟
⎠
3b b −b −3b

From (4.16) to (4.23) we obtain the coarse operator of the MSFV method as
⎡ ⎤
− (a + b) , 2 (a − 3b) , − (a + b)
1⎢ ⎥
⎢
LH = [ma,b ]H = ⎣ 2 (b − 3a) , 12(a + b), 2 (b − 3a) ⎥⎦ , (4.42)
8
− (a + b) , 2 (a − 3b) , − (a + b)
H

for homogeneous problems with aligned anisotropy, c = 0. The coefficients m±1,0 and
m0,±1 in the stencil are only non-positive for an anisotropy ratio, a/b, between 1/3 and
3, outside this narrow interval LH violates (4.37). This is the cause of the oscillations
in homogeneous problems with aligned anisotropy observed in §4.3.1. The violation
of (4.36) to (4.39) does not immediately lead to a coarse solution pH that violates
Hopf’s maximum principle. In figure 4.5b the coarse solution pH is still oscillation
free although LH violates (4.37), but oscillations occur eventually if the anisotropy
increases further.
The coarse operator of the MSFV method is similar to the MPFA O(η) method,
for η = 1. This family of schemes was first introduced by Edwards & Rogers (1998),
however we follow the parametrization of Nordbotten et al. (2007) in terms of the
parameter η ∈ [0, 1). Although the general construction of the O(η) method requires
CHAPTER 4. MULTISCALE SIMULATION 143

λ1/λ2 = 1.5 2 3 5 10 100 λ1/λ2 = 1.5 2 3 5 10 100

1 1

θ= θ=
40 º 40 º
0.8 0.8
r
d rato
θ=

r
o

θ=

to
h
et p e i a
30

30

se c l
0.6 0.6

er
-m se o

θ=2

º
θ=2

e p

a n
)

θ = 10
θ = 10

1
O( coar

o
a/b
b

co st
a/b

b
0º
t
0º

=a
r

=a
V in
FV po

º
º

c2
0.4

c2
0.4
MS 7

SF -
it:

it:
lim

lim
M

C
0.2 0.2
ic
pt

ic
i i pt
ell ell
a) b)
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
|c|/b |c|/b

Figure 4.7: The monotonicity regions of the original and compact coarse operator are
shown in the parameter space a/b vs. |c|/b. A gray grid shows the corresponding
angle, θ, and the ratio of principal components, λ1 /λ2 in the effective permeability
tensor given by (4.35). All methods are monotone above the labeled line. (a) The
monotonicity regions of both the MPFA O(1) method and the coarse operator of the
MSFV method are compared. (b) The monotonicity region of the CMSFV method
equals that of the 7-point stencil.

η ∈ [0, 1), for the homogeneous case the O(1) method is well defined Nordbotten et al.
(2007). Previously the O(0) method has been emphasized by Aavatsmark (2002), the
O(1/2) method by Edwards & Rogers (1998), and the O(0.9) method by Pal et al.
(2007).
For homogeneous problems with aligned anisotropy (c = 0) the MSFV coarse
operator is identical to the MPFA O(1) method. For c = 0 we have used numerical
solutions on the fine grid to compute the MSFV coarse operator. The MPFA O(0)
method was used to compute the basis functions on the fine grid. Figure 4.6 compares
the numerical results for the MSFV method to the analytic expression for the O(1)
method. Both methods have a similar behavior as the tensor is rotated. The difference
between the methods increases as the ratio of the principal components of K increases.
CHAPTER 4. MULTISCALE SIMULATION 144

The region where a stencil satisfies (4.36) to (4.39) will be referred to as the
monotonicity region of the stencil. The monotonicity region of the MSFV method is
shown in figure 4.7a and has the same characteristics as that of the O(1) method.
The monotonicity of both methods is very limited when anisotropy is aligned with
the grid and largest, if the tensor is rotated by 45◦ . This unusual property of the
original MSFV method is an unanticipated consequence of using bilinear dual basis
functions within the finite volume framework of § 4.1.5.

4.4.2 Compact coarse operator

The limited monotonicity region of the MSFV coarse operator, when the anisotropy is
aligned with the grid, prevents the application of the MSFV method to many realistic
problems in subsurface flow, where grids with high aspect ratios are commonly used.
In §4.2 the MSFV method was formulated in terms of a MPFA method. It has been
shown that it is possible to construct MPFA methods with different monotonicity
properties while maintaining accuracy and convergence (Edwards & Rogers, 1998;
Nordbotten et al., 2007). For homogeneous anisotropic problems the following 7-
point stencil
⎡ ⎤
(c − |c|)/2 −b + |c| −(c + |c|)/2
⎢ ⎥
[ma,b ]h = ⎢
⎣ −a + |c| 2(a + b − |c|) −a + |c| ⎥ ,
⎦ (4.43)
−(c + |c|)/2 −b + |c| (c − |c|)/2
h

is second order accurate. Two of the diagonal neighbors are always zero depending
on the sign of c, and therefore the stencil has 7 non-zero entries for c = 0. All four
diagonal neighbors are zero for c = 0, and the method reduces to the standard 5-
point stencil. Nordbotten et al. (2007) have shown that the 7-point stencil has an
optimal monotonicity region, in the sense that no other 9-point stencil satisfies con-
ditions 4.37 to 4.39 over a larger range of parameters (figure 4.7b). For homogeneous
problems and inside its monotonicity region this 7-point stencil also leads to a linear
system, where Lh is a M-matrix. The 7-point stencil can be characterized using a
3-point flux approximation (figure 4.8), leading to one zero entry in every row of the
CHAPTER 4. MULTISCALE SIMULATION 145

P4 P3 P4 P3

f3 f3
f4 f2 f4 f2

f1 f1

P1 P2 P1 P2
a) c > 0, 0° < θ < 90° b) c < 0, 0° > θ > −90°

Figure 4.8: The triangular domains of dependence of the fluxes in a dual interaction
region are shown for both signs of c. The corresponding orientation of the permeability
tensor is indicated by the ellipses.

transmissibility matrix
⎧ ⎛ ⎞
⎪
⎪ a, c − a, −c, 0
⎪
⎪ ⎜ ⎟
⎪
⎪ ⎜ c, b − c, −b, ⎟
⎪
⎪ 1 ⎜ 0 ⎟
⎪
⎪ 2 ⎜ c, ⎟ c ≥ 0,
⎪
⎪ ⎝ 0, −a, a − c ⎠
⎪
⎪
⎪
⎪
⎪
⎨ b, 0, −c, c − b
T= (4.44)
⎪
⎪ ⎛ ⎞
⎪
⎪ a − c, −a,
⎪
⎪
0, c
⎪
⎪ ⎜ ⎟
⎪
⎪ ⎜ 0, c − b, −c ⎟
b,
⎪
⎪ 1⎜ ⎟ c < 0,
⎪ 2⎜
⎪
⎪
⎪ ⎝ 0, −c, c − a, a ⎟⎠
⎪
⎩
b − c, c, 0, −b

whose position depends on the sign of the off-diagonal component, c. Figure 4.8 shows
that the domain of dependence of each flux is aligned with the direction of maximum
connectivity. We refer to a stencil with three non-zero entries in each row of T as
a compact operator. The definition is based on T rather than the resulting stencil,
because the latter may lose its zero entries in the heterogeneous case, while the zero
entries in T always remain.
CHAPTER 4. MULTISCALE SIMULATION 146

4.5 Compact coarse operator for

heterogeneous problems
To obtain a multiscale finite volume method with better monotonicity properties,
a compact coarse operator is introduced that reduces to the 7-point stencil in the
homogeneous limit. The resulting compact multiscale finite volume (CMSFV) method
is otherwise identical to the original MSFV method proposed by Jenny et al. (2003).
The framework presented in §4.2 allows the necessary modification of T without
introducing inconsistencies in the reconstruction.
Four degrees of freedom are necessary to modify T to obtain a compact operator,
the corresponding compact transmissibility matrix is denoted T̄. These degrees of
freedom are obtained by reducing the number of local solutions from four to three,
providing twelve constraints on the sixteen unknown transmissibilities tm,k . Instead
of the four basis functions of the original MSFV method, defined in §4.2.2 the basis


4
Φ1 = φi , Φ2 = φ1 + φ4 , Φ3 = φ1 + φ2 , and Φ4 = φ4 ,
i=1

can be used to determine the transmissibilities tm,k . This change in the basis does
not affect the coarse operator, but it separates the linear and the bilinear parts of the
solution. Linear solutions can be represented exactly by the first three bases, and we
eliminate Φ4 to obtain the four degrees of freedom necessary to modify T. The first
three basis functions Φ1 to Φ3 provide the following twelve constraints

0 = T̄( 1 1 1 1 )T , (4.45)
f x = T̄( 1 0 0 1 )T , (4.46)
f y = T̄( 1 1 0 0 )T , (4.47)

on T̄. The definition of T̄ is completed by placing a zero entry in each row. As in the
homogeneous case the zeros are placed to align the stencil with the principle direction
of anisotropy, which can be determined from the local flow information in (4.46) and
CHAPTER 4. MULTISCALE SIMULATION 147

(4.47). The compact transmissibility matrix, T̄, has the following rows,
⎧  
  ⎨ f x , f y − f x , −f y , 0 , f1y ≥ 0,
1 1
t̄1,1 , t̄1,2 , t̄1,3 , t̄1,4 =  1 1
 (4.48)
⎩ f x − f y , −f x , 0, f y , f1y < 0,
1 1 1 1

⎧  
  ⎨ f x , f y − f x , −f y , 0 , f2x ≥ 0,
2 2
t̄2,1 , t̄2,2 , t̄2,3 , t̄2,4 =  2 2
 (4.49)
⎩ 0, f y , f x − f y , −f x , f2x < 0,
2 2 2 2
⎧  
  ⎨ f y , 0, −f x , f x − f y , f3y ≥ 0,
t̄3,1 , t̄3,2 , t̄3,3 , t̄3,4 =  3 3 3 3
 (4.50)
⎩ 0, −f y , f y − f x , f x , f3y < 0,
3 3 3 3
⎧  
  ⎨ f y , 0, −f x , f x − f y , f4x ≥ 0,
t̄4,1 , t̄4,2 , t̄4,3 , t̄4,4 =  4 4 4 4
 (4.51)
⎩ f y − f x , f x , 0, −f y , f4x < 0,
4 4 4 4

expressed in terms of the transmissibilities f x and f y of the new bases functions Φ2

and Φ3 . This definition of the compact transmissibility matrix reduces to (4.44)
in the homogeneous, anisotropic case. Since the MSFV method does not compute
an explicit upscaled effective permeability tensor, the orientation of the domain of
dependence is based on the numerically computed fluxes rather than the off-diagonal
component of the effective tensor.
The corresponding compact coarse operator is given by (4.16) to (4.23), where tm,k
has been replaced by t̄m,k in all equations. For the homogeneous case the compact
coarse operator is identical to the 7-point stencil, and hence the monotonicity of the
CMSFV method in the homogeneous parameter space is identical to that of the 7-
point stencil (figure 4.7b). In the general heterogeneous case the entries of T̄ are
calculated numerically from the basis functions Φ2 and Φ3 . The choice of boundary
conditions for these local problems is discussed in §4.6.3.
CHAPTER 4. MULTISCALE SIMULATION 148

a) a/b = 1 b) a/b = 1/10 c) a/b = 1/100 d) a/b = 1/1000

Figure 4.9: The approach of the global fine-scale solution to transverse equilibrium,
as the effective anisotropy a/b increases. The boundary conditions for this problem
are p(x = 0, y) = 1, p(x = 1, y) = 0, and py (x, y = 0, 1) = 0 and the effective
permeability on the fine-scale is diagonal, c = 0. The log of the permeability field is
shown at the base.

4.6 Localization of fine-scale problems

The boundary conditions that are used to localize the elliptic problems in the dual
interaction regions, L̃Γh , have a strong influence on the properties of the coarse opera-
tor for heterogeneous problems. Two common choices, linear and reduced boundary
conditions have been introduced in §4.2.2. Either choice introduces an error into the
basis functions that is localized in a boundary layer. In many anisotropic porous
media problems the boundary layer quickly grows to the size of the local domain,
and local oversampling cannot eliminate the error introduced by the local bound-
ary conditions. However, in these cases the solution has a simple form, and this is
exploited here to construct local solutions that lead to a robust and accurate coarse
operator. A different approach to this problem is to introduce global information into
the basis functions (Chen & Durlofsky, 2006; Efendiev et al., 2006; Durlofsky et al.,
2007), which leads to improved accuracy but at increased computational cost.
CHAPTER 4. MULTISCALE SIMULATION 149

4.6.1 Transverse equilibrium

Problems of flow in porous media are often characterized by domains with high ef-
fective anisotropy and homogeneous Neumann boundary conditions. This allows the
solution of the elliptic problem to reach transverse equilibrium (TVE). Transverse
equilibrium is characterized by negligible gradients in the direction of high effective
permeability. This property has been exploited in the derivation of simplified analyti-
cal models (Yortsos, 1995), and efficient pseudo three-dimensional numerical methods
(Coats et al., 1971).
Consider the elliptic problem given by (4.8) with a diagonal permeability ten-
sor, where the components of the tensor a(x) and b(x) are related by a constant
a(x) = A−2 b(x). This is the typical effective permeability tensor arising from isotropic
problems in stretched domains, where A = ∆x/∆y. The pressure equation is then
given by    
1 ∂ ∂p(x) ∂ ∂p(x)
b(x) + b(x) = 0. (4.52)
A2 ∂x ∂x ∂y ∂y
Consider the case where homogenous Neumann boundary conditions are applied in the
y-direction, and constant Dirichlet conditions on the remaining boundaries, so that
an overall pressure gradient in the positive x-direction is imposed. The numerical
solutions in figure 4.9 show that the solution becomes one-dimensional in the x-
direction in the limit of large A. We integrate (4.52) in the y-direction to obtain an
expression for the effective one-dimensional pressure equation

1
1 ∂ 1
∂p(x) ∂p(x)


b(x) dy + b(x) = 0. (4.53)
A2 ∂x 0 ∂x ∂y
0

The second term is zero due to the homogeneous Neumann boundary conditions. In
a boundary layer, δy ∝ A, near the Neumann boundaries the transverse gradients in
the solution vanish, ∂p/∂y ≈ 0. In the limit of large A the boundary layers from the
opposite sides of the domain merge so that p is in TVE and no longer a function of
y, and (4.53) is approximately equal to
 1   
1 ∂ ∂p(x) ∂ ∂p(x)
b(x)dy =
ay (x) ≈ 0, (4.54)
A2 ∂x 0 ∂x ∂x ∂y
CHAPTER 4. MULTISCALE SIMULATION 150

where
ay (x) is the average of a(x) in the y-direction.

4.6.2 Bilinear local problems with

Dirichlet boundary conditions
The local elliptic problems, introduced in § 4.2.2, are constructed with Dirichlet
boundary conditions, and their behavior for large effective anisotropy is different
from the limit of the global solution with homogeneous Neumann boundary condi-
tions. In particular the second term of (4.53) is not zero and the limit of (4.52) for
large A is  
∂ ∂p(y; x)
b(y; x) ≈ 0. (4.55)
∂y ∂x
The solution consists of a family of decoupled one dimensional problems in the y-
direction, which are parameterized by x. Unless the pressures specified at the y-
boundaries are equal there will be significant fluxes in the y-direction.
When basis functions are constructed for problems with high effective anisotropy,
the boundary layer containing the error introduced by the localization spreads through
the entire dual interaction region and dominates the solution. Since it is not possible
to eliminate the localization error by oversampling, it appears that the MSFV method
cannot be accurate for high aspect ratios. However, it has been shown above that
the solution has a simple form as it approaches transverse equilibrium with increasing
anisotropy. If the local boundary conditions are chosen to allow transverse equilibrium
in the local solution, then the MSFV method can produce good results at high aspect
ratios.
We have shown in §4.4.1 that very large transverse fluxes occur in homogeneous
problems at high effective anisotropy and lead to the loss of monotonicity of the coarse
operator. The compact coarse operator is designed to minimize these fluxes and give
the largest possible monotone domain. To retain a monotone coarse operator for het-
erogeneous problems, it is therefore important that the local elliptic problems do not
introduce artificial gradients in the transverse direction. Given a coarse pressure field
in transverse equilibrium, ∂pH /∂y = 0, the local solutions must be able to interpolate
the coarse pressure onto the fine grid without introducing artificial gradients in the
CHAPTER 4. MULTISCALE SIMULATION 151

a) b) c) d)
1 1 1 1

p̂h

p̂h
p̂h
ph

0.75 0.75 0.75 0.75

0 0 0 0
0.25 y 0.25 y 0.25 y 0.25 y
x 0.25 x 0.25 x 0.25 x 0.25
0.75 0.75 0.75 0.75

Figure 4.10: Comparison between the fine-scale solution ph and the interpolated
pressure field p̂h = φh1 +φh4 in a dual crossing the high permeability channel at a/b =
1/1000. (a) Fine-scale solution, ph , in the dual, rescaled to a unit pressure gradient,
is in transverse equilibrium. (b) Interpolated fine pressure p̂h with reduced boundary
conditions. (c) Interpolated fine pressure p̂h with linear boundary conditions. (d)
Interpolated fine pressure p̂h with hybrid boundary conditions.

y-direction.
Consider a domain with a high permeability channel in the y-direction, small
amplitude random noise, an effective anisotropy a/b = 1/1000, and the boundary
conditions p(x = 0, y) = 1, p(x = 1, y) = 0, and py (x, y = 0, 1) = 0. Figure 4.10a
shows that the fine-scale solution is in transverse equilibrium, and the reduced one
dimensional problem in the x-direction reflects the high permeability channel.
The advantage of the reduced boundary conditions is that they are also solutions
to one dimensional problems, and therefore recognize the high permeability channel.
Figure 4.10b shows that the channel is reflected in the superposition of the local
solutions. The reduced boundary conditions sample different realizations of the per-
meability and are therefore not identical. This introduces finite transverse gradients,
∂ p̂h /∂y = 0, even if the coarse pressure is in transverse equilibrium. Even small
transverse pressure gradients lead to very large fluxes as the effective anisotropy in-
creases, and this leads to the loss of monotonicity of the coarse operator even for
the compact stencil. Eliminating these gradients is therefore the key to a monotone
coarse operator and a robust MSFV method for heterogeneous anisotropic problems.
Linear boundary conditions are identical by construction and do not introduce
artificial transverse gradients. Therefore, linear boundary conditions give rise to a
CHAPTER 4. MULTISCALE SIMULATION 152

more robust coarse scale operator, but figure 4.10c shows that they give poor accu-
racy. In this case the linear boundary conditions have a large error. For isotropic
problems this error would be localized in a narrow boundary layer, but for high effec-
tive anisotropy the error spreads through the entire dual interaction region. Linear
boundary conditions therefore give a robust operator, but they may lead to poor
accuracy.

4.6.3 Linear local problems with hybrid boundary conditions

For highly anisotropic problems with homogeneous Neumann boundaries on the global
domain, the boundary conditions for the local elliptic problems must have the follow-
ing properties: To ensure robustness it is essential that the local elliptic problems can
reach transverse equilibrium. To ensure accuracy in the isotropic case the boundary
condition should be close to the reduced boundary condition. In the limit of high ef-
fective anisotropy the local elliptic problem should reduce to a close approximation of
the one dimensional problem (4.54). To allow a simple implementation the boundary
condition should adjust automatically to the magnitude of the effective anisotropy.
We propose hybrid boundary conditions for the local elliptic problems that ensure
transverse equilibrium, give good accuracy for all aspect ratios, and are parameter
free. To allow the local problems to reach transverse equilibrium homogeneous Neu-
mann boundary conditions need to be imposed in the direction of TVE instead of
Dirichlet boundary conditions. For the linear basis functions already introduced in
§4.4.2, homogeneous Neumann boundary conditions can be imposed naturally on the
faces perpendicular to the imposed gradient. The linear local problems are defined
by 4.27, with the boundary operator given as

L̃Γh2 = I, g̃h2 = 1, (xi = 0, yi ), (4.56)

L̃Γh2 = I, g̃h2 = 0, (xi = 1, yi ), (4.57)
∂
L̃Γh2 = , g̃h2 = 0, (xi , yi = 0) and (xi , yi = 1) (4.58)
∂y
CHAPTER 4. MULTISCALE SIMULATION 153

for linear gradient, Φh2 , in the x-direction, and

L̃Γh3 = I, g̃h3 = 1, (xi , yi = 0), (4.59)

L̃Γh3 = I, g̃h3 = 0, (xi , yi = 1), (4.60)
∂
L̃Γh3 = , g̃hl = 0, (xi = 0, yi ) and (xi = 1, yi ), (4.61)
∂x

for linear gradient, Φh3 , in the y-direction. We can use these two solutions to define
the boundary conditions for a bilinear local problem Φh4

L̃Γh4 = I, g̃h4 = Φh2 , (xi , yi = 1), (4.62)

L̃Γh4 = I, g̃h4 = 1 − Φh3 , (xi = 0, yi ), (4.63)
L̃Γh4 = I, g̃h4 = 0, (xi = 1, yi ) and (xi , yi = 0). (4.64)

Together with the constraint, 0 = T1, these three problems determine the coarse
operator uniquely. The standard set of bilinear basis functions can be obtained from
Φh2 to Φh4 through linear superposition. These bilinear basis functions have hybrid
boundary conditions and hence the transmissibility matrix T and the coarse operator
LH are modified. This modification is independent from the implementation of the
compact coarse operator.
The local problems defined above attain transverse equilibrium at high anisotropy
by construction (figure 4.10d ). Transverse gradients in the local problems are re-
duced faster than in the global fine-scale solution, because the domain is smaller and
therefore boundary layers merge earlier. Therefore, hybrid boundary conditions have
robustness similar to linear boundary conditions. Section 4.11 shows that the accu-
racy of the hybrid boundary conditions is similar to that of the reduced boundary
conditions at low anisotropy. At high anisotropy the accuracy of the hybrid boundary
conditions is good if
ay in the dual interaction region is a good representation of

ay in the global domain.

CHAPTER 4. MULTISCALE SIMULATION 154

Γ fH = -1
1

a) b) c) d)
0 fH = 1 1

Figure 4.11: (a) The unit square domain Ω, with a 20 × 20 coarse grid. The back
squares indicate coarse blocks with sources fH = ±1. In these blocks the fine-scale
permeability is constant at the mean to give a uniform distribution of the fine-scale
flux. (b) A realization of the permeability field with small correlation length. Figures
c and d show realizations of the layered permeability fields at angles of 15◦ and −30◦
respectively. The logarithm of the permeability is shown in all cases.

4.7 Numerical results for heterogeneous problems

Two coarse scale operators and three local boundary conditions have been introduced
above, allowing six different combinations in addition to the fine-scale solution. To
distinguish the resulting combinations we denote the original operator by, MSFV,
the compact operator by, CMSFV, and the reduced, linear and hybrid local boundary
conditions by the post fix -red, -lin, and -hyb, respectively. This section focuses mostly
on the comparison between MSFV-red that is most commonly used and CMSFV-hyb
presented above.
In all tests a unit square domain has been used, and a diagonal pressure gradient
has been imposed by introducing coarse source terms fH in opposing corners. This is
similar to the quarter five spot configuration commonly used in reservoir simulation
(figure 4.11a). The fine grid has a dimension of 100 × 100, and it has been coarsened
uniformly by a factor of five to produce a 20 × 20 coarse grid. Two types of per-
meability fields were used; fields with short correlation length resulting in a patchy
pattern, and fields with a short and a long correlation length resulting in a layered
pattern. Twenty realizations of the patchy permeability field with a log-normal dis-
tribution, a variance of log k of 2.0, a mean of 3.0, a dimensionless correlation length
CHAPTER 4. MULTISCALE SIMULATION 155

a) fs b) fs c) fs

d) MSFV-red e) MSFV-red f) MSFV-red

g) CMSFV-lin h) CMSFV-lin i) CMSFV-lin

j) CMSFV-hyb k) CMSFV-hyb l) CMSFV-hyb

Figure 4.12: Comparison of time of flight maps between multiscale methods and
the fine-scale solution for three isotropic but heterogeneous fine-scale permeability
fields. All solutions are shown at the same injected volume, corresponding to the
breakthrough of the fine-scale solution. The fine-scale solution, fs, is shown in the first
row, a to c, the original MSFV -red method in the second row, d to f, the CMSFV -lin
method in the third row, g to i, and the CMSFV -hyb method in the fourth row, j to
l. The first column shows solutions for a realization of the permeability field shown
in figure 4.11b, columns 2 to 3 show solutions for realizations of layered permeability
fields at 15◦ and -30◦ (figure 4.11c and d).
CHAPTER 4. MULTISCALE SIMULATION 156

a) fs b) fs c) fs

d) MSFV-red e) MSFV-red f) MSFV-red

g) CMSFV-lin h) CMSFV-lin i) CMSFV-lin

j) CMSFV-hyb k) CMSFV-hyb l) CMSFV-hyb

Figure 4.13: Comparison of time of flight maps between multiscale methods and
the fine-scale solution for three heterogeneous fine-scale permeability fields with grid
aligned fine-scale anisotropy, a/b = 0.01. All solutions are shown at the same injected
volume, corresponding to the breakthrough of the fine-scale solution. The figure
layout is the same as in figure 4.12.
CHAPTER 4. MULTISCALE SIMULATION 157

λ1 = λ2 = 0.05, and a spherical variogram have been generated with sequential

gaussian simulation (figure 4.11b). Layered permeability fields with a log-normal dis-
tribution, a variance of log k of 2.0, a mean of 3.0, a dimensionless correlation length
λ1 = 0.5, and λ2 = 0.02, and a spherical variogram have been generated with sequen-
tial gaussian simulation (Chen et al., 2007). The angle θ between the direction of
long correlation and the horizontal has been varied between θ = −45◦ and 45◦ in 15◦
intervals (figure 4.11c&d ). For each angle, 20 realizations have been created and all
values reported are averages and standard deviations over all realizations.
For transport in porous media a major objective of multiscale methods is to obtain
accurate transport on the fine-scale. Comparing error norms of the reconstructed fine-
scale pressure does not give a good indication of the quality of the fine-scale transport.
Therefore, it is common to assess the quality of multiscale methods designed for
transport in porous media by comparing the concentration fields of a passive tracer
advected on the fine scale (Jenny et al., 2003; Aarnes, 2004). To eliminate numerical
diffusion introduced by advection schemes, the tracer distribution is calculated using
particles.
First we consider the domain shown in figure 4.11a with homogeneous Neumann
boundary conditions, so that the solution approaches transverse equilibrium at high
anisotropy. Figure 4.12 shows the tracer particles introduced at the source in the bot-
tom left corner. The particles are colored according to the time since they have been
introduced into the domain, their time of flight (TOF). For isotropic fine-scale per-
meability fields both the MSFV-red method and the CMSFV-hyb method are able to
reproduce the fine-scale structure in the tracer field. In these cases where the MSFV-
red method performs well the quality of the CMSFV-hyb solution is comparable to
the MSFV-red solution.
Figure 4.13 shows the tracer maps for the same permeability fields that are shown
in figure 4.12, but with uniform, grid aligned, fine-scale anisotropy of a/b = 0.01,
corresponding to an orthogonal grid with aspect ratio 1 : 10. The MSFV-red solu-
tion shows unphysical recirculation cells. The CMSFV-hyb solution shows no such
recirculation cells, has a monotonic coarse scale solution, and recovers the fine-scale
CHAPTER 4. MULTISCALE SIMULATION 158

0.25 MSFV-red 0.4

a) MSFV-lin b)
0.2 CMSFV-lin
0.3
CMSFV-hyb

||ūh −uh ||2

||p̂h −ph ||2

0.15

||uh ||2
||ph ||2

0.2
0.1

0.1
0.05

0 0
-45 -30 -15 0 15 30 45 -45 -30 -15 0 15 30 45
θ θ
0.2 3
c)
2.5
d)
0.15
2
||ūh −uh ||2
||p̂h −ph ||2

||uh ||2
||ph ||2

0.1 1.5

1
0.05
0.5

0 0
-45 -30 -15 0 15 30 45 -45 -30 -15 0 15 30 45
θ θ

Figure 4.14: Figures a and b show the mean L2 -errors of the interpolated solution
p̂h and the reconstructed flux field ūh = −k∇p̄h as function of the orientation, θ, of
the fine-scale layering for isotropic fine-scale permeability, a/b = 1. Figures c and d
show the same for grid aligned fine-scale anisotropy a/b = 0.04 (∆x/∆y = 5). All
values shown are means over 20 realizations and the standard deviations are shown
as vertical bars, if they are larger than the symbols.
CHAPTER 4. MULTISCALE SIMULATION 159

detail of the tracer distribution. In the presence of fine-scale anisotropy the CMSFV-
hyb solution is a good approximation to the fine-scale solution, while the MSFV-red
solution has large errors and unphysical behavior.
The linear boundary conditions give comparable results for isotropic problems,
but lose accuracy as the aspect ratio increases. For flow across layers the errors are
largest (figure 4.13i ), as expected from the discussion in §4.6.2.
Error norms for the pressure and the reconstructed flux obtained by both methods
are shown in figure 4.14. The interpolated pressure p̂h is proportional to the error
in the coarse solution, while the reconstructed, fine-scale flux ūh is proportional to
the error in the reconstructed pressure p̄h . Figure 4.14a shows that linear boundary
conditions lead to larger errors than reduced or hybrid boundary conditions for both
coarse operators. For clarity, the results for linear boundary conditions are omitted
from the other graphs. Figures 4.14a and b show that for isotropic fine-scale perme-
ability fields the errors in the MSFV-red and the CMSFV-hyb solutions are of the
same order of magnitude for flow along layers (0◦ < θ < 45◦ ). For flow across layers
(−45◦ < θ < 0◦ ) the error in the MSFV-red solution is significantly smaller, because
the reduced boundary condition is optimal for this special case. However, for any sig-
nificant anisotropy on the fine-scale the CMSFV-hyb solution becomes more accurate
than the MSFV-red for all angles (figure 4.14c and d ).
Figures 4.15a and b show that for both the MSFV-red and the CMSFV-red
solutions the errors in the flux, ūh , grow exponentially with increasing fine-scale
anisotropy, while the errors for both the MSFV-hyb and the CMSFV-hyb remain
bounded. Therefore, in problems where the fine solution approaches transverse equi-
librium, the hybrid local boundary condition rather than the compact operator lead
to a monotone multiscale solution.
Consider the domain shown in figure 4.11a with homogeneous Dirichlet bound-
ary conditions, so that the solution cannot approach transverse equilibrium with
increasing anisotropy. Figure 4.16 shows that both the MSFV-lin and the MSFV-hyb
solution have lost monotonicity and the streamlines show unphysical recirculation
cells. The CMSFV-hyb solution has remained monotone and captures the fine-scale
transport. Similar behavior has been observed in §4.3.1 for homogeneous problems,
CHAPTER 4. MULTISCALE SIMULATION 160

2
0.4 10
a) MSFV-red b)
MSFV-hyb
0.3 CMSFV-red
1
10

||ūh −uh ||2

CMSFV-hyb
||p̂h −ph ||2

||uh ||2
||ph ||2

0.2
0
10
0.1

-1
0 10
-4 -3 -2 -1 0 -4 -3 -2 -1 0
10 10 10 10 10 10 10 10 10 10
a a
b b

Figure 4.15: Figures a and b show the mean L2 -errors of the interpolated solution
p̂h and the reconstructed flux field as function of increasing fine-scale anisotropy a/b.
All values shown are means over 20 realizations of the permeability field shown in
figure 4.11a.

where all local boundary conditions become linear. Therefore, the compact operator
is necessary to obtain a monotone solution for strongly constrained pressure fields.
A similar situation is shown in figure 4.4c where a local maximum occurs in the
coarse solution computed by the MSFV-red method even for an isotropic problem
with Dirichlet constraints on only two boundaries. The compact operator gives a
monotone coarse solution in this case, demonstrating that it improves the monotonic-
ity even for isotropic problems with few constraints. Such situations arise commonly
in reservoir engineering in areas with horizontal wells that constrain the pressure over
many gridblocks.
In cases with or without transverse equilibrium the compact operator leads to a
linear system, LH , much closer to an M -matrix. A simple measure M (LH )i given by
by the ratio of the largest and smallest off-diagonal entries,
 
max( LHi,j
=i
M (LHi,j )i = −  , (4.65)
min( LHi,j
=i

is introduced to quantify the magnitude of the violation of the M -matrix condition

CHAPTER 4. MULTISCALE SIMULATION 161

−4
x 10
1 3 1
(a) (b) (c)
1.5

p̂h
0
y

y
−1.5

0 −3 0
−4
x 10
1 3 1
(d) (e) (f)
1.5
p̂h

0
y

y
−1.5

0 −3 0
−4
x 10
1 3 1
(g) (h) (i)
1.5
p̂h

0
y

−1.5

0 −3 0
0 1 0 1 0 1
x x x

Figure 4.16: Results for the quarter 5-spot shown in figure 4.11a with homogeneous
Dirichlet boundary conditions. The first column shows contour maps of p̂h , the second
column shows a horizontal view of p̂h along the y-axis, and the third column shows
streamlines obtained from the conservative reconstruction p̄h . Figures a to c show
the results for MSFV-lin, figures d to f show results for MSFV-hyb, and figures g to
i show results for CMSFV-hyb.
CHAPTER 4. MULTISCALE SIMULATION 162

in the i-th row of the system matrix LH . If M (LH )i is non-positive, then the i-th row
of LH satisfies the M -matrix condition. As the magnitude of M (LH )i increases the
violation of the M -matrix condition becomes stronger. The norm ||M (LH )|| allows
us to quantify the quality of the coarse linear system resulting from a multiscale
discretization.
For heterogeneous problems LH does generally not satisfy the M-matrix property,
for any choice of operator and local boundary conditions, but the compact operator
always gives a system much closer to an M -matrix. The violations of the M -matrix
property are larger in magnitude (figure 4.17a and c) and much more common (fig-
ure 4.17b and d ) in the coarse linear system arising from the original coarse operator.
For isotropic fine-scale permeability fields the dependence of ||M (LH )|| on the ori-
entation of the layering, shown in figures 4.14a&b, is analogous to the homogeneous
analysis in §4.4. For the original operator the violation of the M -matrix property
is largest for grid aligned layering and smallest for layering oriented at 45◦ , corre-
sponding to the region of monotonicity shown in figure 4.7a, which is most limited
for anisotropy at 0◦ and unlimited for anisotropy at 45◦ to the grid. The compact
operator on the other hand produces linear systems with a maximum in ||M (LH )|| for
layering oriented between 15◦ and 30◦ , corresponding to the region of monotonicity
shown in figure 4.7b, which is unlimited for anisotropy at 0◦ and 45◦ . The improve-
ment in the quality of LH relative to the original operator is largest for layering that
is aligned, or close to aligned, with the grid. Comparing the first and second row of
figure 4.17, we see the that the violations of the M -matrix criterion produced by the
MSFV operator increase rapidly with increasing fine-scale anisotropy, a/b → 0, while
the CMSFV-hyb remains close to a M -matrix.

4.8 Discussion
In the formulation presented in §4.2 the MSFV method is defined by the properties
of the transmissibility matrix, T∗ . The integrated form of T∗ directly defines the
coarse operator through (4.16) to (4.23) and the reconstruction of the conservative
fine-scale velocity is defined by T∗ through (4.32). The properties of T∗ are therefore
CHAPTER 4. MULTISCALE SIMULATION 163

MSFV-red
2 MSFV-lin 0.02
a) MSFV-hyb b)
CMSFV-red
0.015
CMSFV-lin
||M(LH)||∞

||M(LH)||2
CMSFV-hyb
1 0.01

0.005

0 0
-45 -30 -15 0 15 30 45 -45 -30 -15 0 15 30 45
θ θ
15 0.06

0.05
c) d)
||M(LH)||∞

10 0.04
||M(LH)||2

0.03

5 0.02

0.01

0 0
-45 -30 -15 0 15 30 45 -45 -30 -15 0 15 30 45
θ θ

Figure 4.17: The L∞ and L2 -norms of M (LH ) are shown as a function of the angle
of the layering. Figures a and b show results for a isotropic fine-scale permeability
field (a/b = 1), while figures c and d are for a fine-scale anisotropy a/b = 1/100.
CHAPTER 4. MULTISCALE SIMULATION 164

the starting point for the analysis of MSFV formulations and T∗ provides a direct
link to MPFA methods. This link has made it possible to take advantage of the
existing analysis of MPFA methods and to formulate the compact coarse operator
for the MSFV method. The construction of many other coarse operators is possible
in this framework, and as long as the modification is done at the level of T∗ , the
resulting MSFV method will be: 1) conservative on the coarse level; 2) the fine scale
reconstruction of the fluxes will be locally conservative; 3) the fine-scale fluxes will
be consistent with the coarse fluxes.
The CMSFV method described here has been designed to give a coarse operator
with optimal monotonicity, but other design criteria, e.g. rotational invariance of the
discretization, may lead to other preferred coarse operators. The formulation given
here provides a consistent framework for the design of coarse operators by modifying
T∗ . In a parallel Aavatsmark et al. (2008) have developed a compact MPFA method
with improved monotonicity for non-orthogonal quadrilateral grids with piecewise
constant, full-tensor permeability fields.
Our framework emphasizes that local elliptic solutions determine the numerical
entries of T∗ , rather than their role as basis functions used for the interpolation of
the coarse solution. The concept of the basis function has been important in the
development of the MSFV method, but the formation of the coarse operator and the
fine-scale reconstruction of the conservative velocity field do not require explicit basis
functions. In the framework presented here coarse operators, which are not naturally
associated with a set of basis functions, can be considered, e.g. the compact operator.
In these cases the interpolation of the coarse solution with the standard basis functions
still gives a good approximation of the fine-scale solution, but the corresponding fine
fluxes are not consistent with the coarse fluxes. This is generally not a problem since
the fluxes derived from the interpolated solution are not used for transport.
The original bilinear basis functions require Dirichlet boundary conditions, which
are too restrictive for many anisotropic problems. Once the local elliptic problems
are not interpreted as basis functions, a larger range of local problems and local
boundary conditions can be considered. This allows the design of local problems
that take advantage of simplifications in the fine-scale solution, such as the linear
CHAPTER 4. MULTISCALE SIMULATION 165

flow problems with hybrid boundary conditions that are designed to capture TVE.
Lunati & Jenny (2007) modify the boundary condition for the flux reconstruction,
L̃Γh , to eliminate recirculation cells in the velocity field. Here we modify the boundary
condition of the local elliptic problems, L̄Γh , and hence the coarse operator to achieve
a monotone coarse pressure, which eliminates the unphysical recirculation cells.
The CMSFV-hyb method introduced here is only based on local information, be-
cause local methods are computationally efficient. It has been shown that the accu-
racy of purely local methods may be low if the permeability field has structures with
very long correlation lengths. Several authors have suggested the incorporation of
global data to increase accuracy at increased cost (Chen & Durlofsky, 2006; Efendiev
et al., 2006; Durlofsky et al., 2007). The coarse operator introduced here may also
increase the robustness of these applications.
Most recent studies of multiscale methods for flow in porous media have been
restricted either to tracer flow or to immiscible two phase flow (Chen & Hou, 2002;
Jenny et al., 2003, 2005; Aarnes, 2004; Juanes & Dub, 2008). In these problems only
the flux field is necessary for the solution of the transport problem, while the quality
of the solution itself - the pressure field - is not important. In many applications,
such as the three-phase black-oil formulation recently presented by Lee et al. (2008),
mass transfer between phases is an important physical mechanism. In particular
strongly non-linear phenomena such as phase appearance and disappearance are a
strong function of the pressure. For these applications the increased robustness of the
CMSFV-hyb method for the pressure equation will increase the stability of the coupled
problem. In these cases the pressure interpolation, p̂h , may be more appropriate for
the evaluation of the phase properties.
In the general heterogeneous anisotropic case neither coarse operator leads to a
linear system where LH is an M-matrix. However, compared to the original coarse
operator the compact operator leads to a LH with fewer positive off-diagonal entries
and these entries have a smaller magnitude. The performance of algebraic multigrid
solvers will not deteriorate seriously, unless the off-diagonal entries are substantial
(Ruge & Stüben, 1987). Therefore, we expect that the CMSFV method will avoid
problems with the performance of the linear solver.
CHAPTER 4. MULTISCALE SIMULATION 166

4.9 Conclusion
The construction of the MSFV coarse operator is identical to the construction of a
certain type of MPFA methods. The MSFV method is a natural extension of these
MPFA methods to numerical evaluation of the transmissibility matrix. Therefore,
the rich literature on the analysis of MPFA methods is directly relevant to the design
and analysis of MSFV methods.
The original MSFV coarse operator has a limited region of monotonicity for
aligned anisotropy. This leads to loss of monotonicity of the coarse pressure in prob-
lems with Dirichlet boundary conditions with increasing anisotropy.
A compact coarse operator for the MSFV method that reduces to a 7-point stencil
in the homogeneous case has been introduced. This compact operator has the largest
possible monotonicity region. In particular it is always monotone for grid aligned
anisotropy in homogeneous problems. In heterogeneous problems the compact opera-
tor leads to coarse linear systems that are close to a M -matrix. The compact operator
eliminates oscillations in anisotropic problems with Dirichlet boundary conditions.
For anisotropic problems with homogeneous Neumann boundary conditions the
fine-scale solution approaches transverse equilibrium as the anisotropy increases. To
obtain a monotone, accurate, multiscale solution the local boundary conditions must
be able to reach transverse equilibrium as well. A hybrid boundary condition has been
introduced that naturally allows transverse equilibrium and gives accurate solutions
for anisotropic problems.
The compact coarse operator in combination with the hybrid boundary condition
is the most robust multiscale finite volume formulation for large anisotropy and similar
in accuracy to the original MSFV method for isotropic problems.
Chapter 5

Conclusions

Saline aquifer storage of CO2 has the potential to reduce the emissions from fossil
energy significantly. The leakage of CO2 from the storage aquifer back into the
atmosphere is a main concern in developing this technology. Carbon dioxide can leak
from the geological storage formation back into the atmosphere, because it is mobile
and buoyant relative to the resident brine. Therefore, any process that immobilizes
the injected CO2 , makes it negatively buoyant, or both, is considered a trapping
mechanism. The competition between leakage and trapping processes will determine
what fraction of the injected CO2 will remain in the subsurface permanently. While
mineral trapping offers permanent storage, leakage is unlikely after the CO2 has
dissolved into the brine or has become disconnected at the pore-scale, as a residual
saturation. Consequently dissolution and residual trapping determine the length of
the active storage period, during which leakage is possible.

5.1 Results
We studied dissolution and residual trapping using two different approaches. In order
to resolve the small length and time scales associated with buoyancy driven miscible
convection, we employed an innovative linear stability analysis in self-similar coordi-
nates, and we performed high accuracy nonlinear simulations to study the nonlinear
evolution of the instability.

167
CHAPTER 5. CONCLUSIONS 168

To study the migration and residual trapping of immiscible CO2 plumes in the
post-injection period, we employed vertical-equilibrium sharp-interface models, in
which the dimensionality of the problem is reduced in favor of the ability to focus on
the first-order behavior of the gravity current.
The dominant physical and chemical processes during CO2 storage in saline aquifers
change with time, and natural geologic formations are strongly heterogeneous. Pre-
dictions of the flow and transport behaviors for a particular storage site must account
for the presence of heterogeneity and be able to model the wide range of length and
time scales associated with the physical and chemical processes. To make numerical
modeling of such systems possible, we extend the multiscale finite volume (MSFV)
method to highly anisotropic problems. This will allow the robust numerical modeling
of CO2 storage in saline aquifers that commonly have large aspect ratios.

5.1.1 Dissolution trapping & convection

Dissolution is the dominant trapping process, if the CO2 has ponded in a structural
trap. Dissolution trapping will be effective, if the increased density of the saturated
brine induces convective currents. The permeability, k, of a formation is the most im-
portant physical parameter controlling convective dissolution. Linear stability theory
has shown that the critical time for the onset of the convective motion is proportional
to 1/k 2 and the associated critical wavelength is proportional to 1/k. Direct numer-
ical simulations confirm the predictions of the linear theory at early times, and we
use them to study the nonlinear evolution of the dissolution rate of CO2 after the
onset of convection. In very large aquifers, the long-term dissolution rate is essen-
tially constant over time, independent of the Ra number, and proportional to the
permeaability. In closed aquifers, the dissolution rate evolves in a manner similar to
open aquifers until the plumes of dissolved CO2 reach the base of the aquifer. Once
these plumes interact with the base of the aquifer the dissolution rate decays very
rapidly.
We conclude that dissolution trapping is favored in large high-permeability aquifers,
because the onset of convection is early, the dissolution rate is high, and it remains
CHAPTER 5. CONCLUSIONS 169

nearly constant for long times.

5.1.2 Residual trapping & gravity currents

In situations where the buoyant supercritical CO2 migrates as a gravity current along
the top of the aquifer, residual saturation forms continuously in its wake. In this case,
residual trapping is likely to be the dominant trapping process. The effectiveness of
residual trapping is determined by the magnitude of the residual saturation and the
sweep efficiency of the CO2 plume. We have studied the interplay between the vertical
sweep of the CO2 plume and residual trapping analytically, using models of gravity
currents that assume vertical equilibrium and a sharp interface. We have identified the
mobility ratio between the CO2 and the brine as a key physical parameter governing
the spreading of the current and the sweep during the migration. In sloping aquifers,
the volume of the mobile CO2 plume is reduced to zero in finite time, while it decays
as a power-law in horizontal aquifers. Residual trapping is more efficient in sloping
aquifers, because the CO2 plume migrates along-slope at constant speed, while the
speed of migration decays in horizontal aquifers. For sloping aquifers, we obtain semi-
analytic estimates of the migration distance of the CO2 plume and the duration of
the active storage period.
Our results show that the poor vertical sweep of the CO2 plume gives rise to
long migration distances that may limit the storage capacity of saline aquifers. In-
creasing the slope of the aquifer leads to faster residual trapping, without increasing
the maximum migration distance of the CO2 plume significantly. However, for any
given injection depth, the along-slope distance to the surface decreases rapidly with
increasing dip of the aquifer. Leakage of CO2 at an outcrop of the aquifer will occur, if
the maximum migration distance of the CO2 plume exceeds the available along-slope
distance between the injection site and the outcrop. To reduce the duration of the
active storage period, potential storage aquifers should have enough dip to induce
lateral migration of the CO2 plume. However, the reduced along-slope distance to
the outcrop for large dip angles limits the volume of CO2 that can be stored. There-
fore, high-permeability, gently-dipping, and deep saline aquifers are optimal targets
CHAPTER 5. CONCLUSIONS 170

for CO2 storage, because they have relatively short active storage periods and large
storage capacities.

5.1.3 Multiscale methods for porous media

We developed a robust compact multiscale finite volume (CMSFV) method for het-
erogeneous anisotropic elliptic equations, which describe the flow (pressure) field in
porous media. The original multiscale finite volume (MSFV) method loses monotonic-
ity in the presence of moderate grid aligned anisotropy or equivalently for stretched
grids. By comparison with multipoint flux approximations, we show that the coarse
operator of the original MSFV method has a limited region of monotonicity, and
we introduce a new compact coarse operator with an optimal stability region. We
also show that the construction of the boundary conditions of the local elliptic prob-
lems must allow for transverse equilibrium to increase the monotonicity of the coarse
operator in heterogeneous problems.
We introduced new local elliptic problems with hybrid boundary conditions that
allow for transverse equilibrium and lead to a robust multiscale discretization in
the presence of strong heterogeneity. This compact discretization will allow the ap-
plication of the (C)MSFV method to CO2 storage in strongly heterogeneous and
anisotropic saline aquifers with large aspect ratios.

5.2 Future work

Here we briefly outline some of the research directions that follow directly from the
work presented in this thesis.

Convective dissolution The direct numerical simulations should be extended to

three-dimensions and the inclusion of permeability heterogeneity in order to obtain,
more realistic estimates of the long-term dissolution rate. These results will allow
simple estimates of the time necessary to dissolve a volume of CO2 through a given
interfacial area, an important quantity currently not available.
CHAPTER 5. CONCLUSIONS 171

Gravity currents The most interesting continuation of the work presented in § 3 is

the extension of the model to include the effects of CO2 dissolution into the brine. The
resulting relatively simple model would capture some of the main aspects of residual
and dissolution trapping and allow a comparison between trapping mechanisms. If a
simple model of leakage is included, the competition between leakage and trapping
processes can be investigated. Such a model may allow the computation of semi-
analytic versions of figure 1.3, which show the contribution of each trapping process.
The three scenarios shown in figure 1.4 are likely to produce qualitatively different
features in figure 1.3, which may form the basis of a classification of storage sites.

Multiscale simulation of CO2 storage The MSFV method has been extended to
include all the necessary physics: gravity and capillary forces, compressibility, wells,
and three-phase problems. We have extended it to problems with severe anisotropy,
expected in regional aquifers. Adaptivity criteria have been developed for the pressure
solver as well as for the solution of the transport equation promising large speed-
up compared to conventional simulations. We therefore plan to demonstrate the
application of the MSFV method to saline aquifer storage in the near future.
BIBLIOGRAPHY 172

Bibliography
Aarnes, J. E. 2004 On the use of a mixed multiscale finite element method for
greater flexibility and increased speed or improved accuracy in reservoir simulation.
Multiscale Model. Simul. 2 (3), 421–439.

Aavatsmark, I. 2002 An introduction to multipoint flux approximations for quadri-

lateral grids. Comput. Geosci. 6, 405–432.

Aavatsmark, I., Barkve, T., Bøe, Ø. & Mannseth, T. 1996 Discretization

on non-orthogonal, quadrilateral grids for inhomogeneous, anisotropic media. J.
Comput. Phys. 127, 2–14.

Aavatsmark, I., Eigestad, G. T., Mallison, B. T. & Nordbotten, J. M.

2008 A compact multipoint flux approximation method with improved robustness.
Num. Meth. for Part. Diff. Eqns., DOI 10.1002/num.20320 (In Press) .

Abdulle, A. & E, W. 2003 Finite difference heterogeneous multi-scale method for

homogenization problems. J. Comput. Phys. 191 (1), 18–39.

Adams, J. J. & Bachu, S. 2002 Equations of state for basin geofluids: algorithm
review and intercomparison for brines. Geofluids 2, 257271.

Arbogast, T. 2002 Implementation of a locally conservative numerical subgrid

upscaling scheme for two-phase darcy flow. Comput. Geosci. 6 (3-4), 453481.

Aronson, D. G. 1985 The Porous Medium Equation, Lecture Notes in Mathematics,

vol. 1224. Springer-Verlag.

Bachu, S. 2003 Screening and ranking of sedimentary basins for sequestration of

CO2 in geological media in response to climate change. Environmental Geology 44,
277–289.

Bachu, S. & Adams, J. J. 2003 Sequestration of CO2 in geological media in re-

sponse to climate change: capacity of deep saline aquifers to sequester CO2 in
solution. Energy Convers. Manage. 44, 31513175.
BIBLIOGRAPHY 173

Bachu, S. & Bennion, B. 2007 Effects of in-situ conditions on relative permeability

characteristics of CO2 -brine systems. Environ. Geol. (doi 10.1007/s00254-007-0946-
9).

Bachu, S. & Stewart, S. 2002 Geological sequestration of anthropogenic carbon

dioxide in the western canada sedimentary basin: Suitability analysis. Can. J.
Petrol. Tech. 41 (2), 32–40.

Barenblatt, G. I. 1952 On some unsteady motions of fluids and gases in a porous

medium. Appl. Math. Mech.-Engl. (PMM) 16, 67–78.

Barenblatt, G. I. 1996 Scaling, self-similarity, and intermediate asymptotics.

Cambridge University Press.

Barenblatt, G. I. & Zeldovich, Ya. B. 1972 Self-similar solutions as interme-

diate asymptotics. Annual Reviews in Fluid Mechanics 4, 285–312.

Bear, J. 1972 Dynamics of Fluids in Porous Media. American Elsevier Pub. Co.

Bear, J. & Ryzhik, V. 1998 On displacement of NAPL lenses and plumes in a

phreatic aquifer. Transport Porous Med. 33, 227–255.

Ben, Y., Demekhin, E. A. & Chang, H.-C. 2002 A spectral theory for small-
amplitude miscible fingering. Phys. Fluids 14 (3), 999.

Benson, S. M., Tomutsa, L., Silin, D. & Kneafsey, T. 2006 Core scale and
pore scale studies of carbon dioxide migration in saline formations. In Proceedings
GHGT-8 . Trondheim, Norway.

Bickle, M., Chadwick, A., Huppert, H. E., Hallworth, M. & Lyle, S. 2007
Modelling carbon dioxide accumulation at Sleipner: Implications for underground
carbon storage. Earth Planet. Sc. Lett. 255, 164–176.

Bondor, P. L. 1992 Applications of carbon dioxide in enhanced oil recovery. Energy

Convers. Manage. 33 (5-8), 579–586.
BIBLIOGRAPHY 174

Bunse, W. & Bunse-Gerstner, A. 1985 Numerische lineare algebra. Stuttgart:

Teubner.

van der Burgt, M. J., Cantle, J. & Boutkan, V. K. 1992 Carbon dioxide
disposal from coal-based IGCC’s in depleted gas fields. Energy Convers. Manage.
33 (5-8), 603–610.

Caltagirone, Jean-Paul 1980 Stability of a saturated porous layer subject to a

sudden rise in surface temperature: comparison between linear and energy methods.
Q. J. Mech. Appl. Math. 33, 47–58.

Camhi, E., Ruith, M. & Meiburg, E. 2000 Miscible rectilinear displacements

with gravity override. Part 2. Heterogeneous porous media. J. Fluid Mech. 420,
259.

Chang, H.-C., Demekhin, E. A. & Kalaidin, E. 1998 Generation and suppres-

sion of radiation by solitary pulses. SIAM J. App. Math. 58, 1246.

Chen, Y. & Durlofsky, L. J. 2006 Efficient incorporation of global effects in up-

scaled models of two-phase flow and transport in heterogeneous formations. SIAM
Multiscale Model. Simul. 5, 445–475.

Chen, Y., Mallison, B. T. & Durlofsky, L. J. 2007 Nonlinear two-point

flux approximations for modelling full-tensor effects in subsurface flow simulations.
Comput. Geosci., Special Issue on Multiscale Methods for Flow and Transport in
Heterogeneous Porous Media, DOI 10.1007/s10596-007-9067-5 (In Press) .

Chen, Z. & Hou, T. Y. 2002 A mixed multiscale finite element method for elliptic
problems with oscillating coefficients. Math. Comp. 72 (242), 541–576.

Coats, K. H., Dempsey, J. R. & Henderson, J. H. 1971 The use of vertical equi-
librium in two-dimensional simulation of three-dimensional reservoir performance.
Soc. Petrol. Eng. J. 251 (March), 63–71.
BIBLIOGRAPHY 175

Diersch, H. J. G. & Kolditz, O. 2002 Variable-density flow and transport in

porous media: approaches and challenges. Advances in Water Resources 25, 899–
944.

Dietz, D. N. 1953 A theoretical approach to the problem of enchroaching and by–

passing edge water. Akad. van Wetenschappen, Proc. V.56 B, 83–92.

Doughty, C., Freifeld, B. M. & Trautz, R. C. 2007 Site characterization for

CO2 geologic storage and vice versa: the Frio brine pilot, Texas, USA as a case
study. Environ. Geol. (doi 10.1007/s00254-007-0946-9).

Drazin, P. G. & Reid, W. H. 1981 Hydrodynamic Stability. Cambridge University

Press.

Durlofsky, L. J. 1991 Numerical calculation of equivalent grid block permeability

tensors for heterogeneous media. Water Resour. Res. 27, 699–708.

Durlofsky, L. J., Efendiev, Y. & Ginting, V. 2007 An adaptive local-global

multiscale finite volume element method for two-phase flow. Adv. Water Resour.
30, 576–588.

Ebigbo, A., Class, H. & Helmig, R. 2006 CO2 leakage through an abandoned
well: problem-oriented benchmarks. Computat. Geosci. 11 (2), 103–115.

Edwards, M. G. & Rogers, C. F. 1998 Finite volume discretization with imposed

flux continuity for the general tensor pressure equation. Comput. Geosci. 2, 259–
290.

Efendiev, Y., Ginting, V., Hou, T. & Ewing, R. 2006 Accurate multiscale
finite element methods for two-phase flow simulations. J. Comput. Phys. 220,
155174.

Elder, J. W. 1967 Transient convection in a porous medium. J. Fluid Mech. 27,

609–623.

Elder, J. W. 1968 The unstable thermal interface. J. Fluid Mech. 32, 69–96.
BIBLIOGRAPHY 176

Ennis-King, J. & Paterson, L. 2002 Engineering aspects of geological seques-

tration of carbon dioxide. In Asia Pacific Oil and Gas Conference and Exhibition.
Melbourne, Australia.

Ennis-King, J. & Paterson, L. 2003 Role of convective mixing in the long-term

storage of carbon dioxide in deep saline formations. SPE 84344 pp. 1–12.

Ennis-King, J., Preston, I. & Paterson, L. 2005 Onset of convection in

anisotropic porous media subject to a rapid change in boundary conditions. Phys.
Fluids 17 (8).

Fayers, F. J. & Muggeridge, A. H. 1990 Extensions to Dietz theory and behavior

of gravity tongues in slightly tilted reservoirs. Soc. Petrol. Eng. Res. Eng. 5, 487–
494.

Foster, T. 1965 Stability of a homogeneous fluid cooled uniformly from above.

Phys. Fluids 8, 1249–1257.

Foster, T. 1968 Effect of boundary conditions on the onset of convection. Phys.

Fluids 11, 1257–1262.

Gasda, S. E., Bachu, S. & Celia, M. A. 2004 Spatial characterization of the

location of potentially leaky wells penetrating a deep saline aquifer in a mature
sedimentary basin. Environ. Geol. 46 (6-7), 707–720.

Gautier, Y., Blunt, M. J. & Christie, M. A. 1999 Nested gridding and

streamline-based simulation for fast reservoir performance prediction. Comput.
Geosci. 3 (3-4), 295–320.

Gounot, J. & Caltagirone, Jean-Paul 1989 Stabilité et convection naturelle au

sein d’une couche poreuse non homogène. International Journal of Heat and Mass
Transfer 32, 1131–1140.

Green, L. L. & Foster, T. 1975 Secondary convection in a Hele Shaw cell. J.

Fluid Mech. 71, 675–687.
BIBLIOGRAPHY 177

Gresho, P. M. & Sani, R. L. 1971 The stability of a fluid layer subjected to a

step change in temperature: Transient vs. frozen time analysis. Int. J. Heat Mass
Transfer 14, 207.

Gunter, W. D., Wiwchar, B. & Perkins, E. H. 1997 Aquifer disposal of

CO2 -rich greenhouse gases: extension of the time scale of experiment for CO2 -
sequestering reactions by geochemical modelling. Miner Petrol 59, 121–140.

Henry, H. R. 1959 Salt intrusion into fresh-water aquifers. J. Geophys. Res. 64,
1911–1919.

Hesse, M. A., Tchelepi, H. A. & F. M. Orr Jr. 2006 Scaling analysis of

the migration of CO2 in saline aquifers. In SPE Annual Technical Conference and
Exhibition. San Antonio, TX.

Holloway, S & Savage, D. 1993 The potential for aquifer disposal of carbon-
dioxide in the UK. Energy Convers. Manage. 34 (9-11), 925–932.

Holtz, M. H. 2002 Residual gas saturation to aquifer influx: A calculation method

for 3-D computer reservoir model construction. In SPE Gas Technologies Sympo-
sium. Calgary, Alberta, Canada.

Homsy, G. M. 1973 Global stability of time-dependent flows: impulsively heated or

cooled fluid layers. J. Fluid Mech. 60, 129–139.

Hou, T. Y. & Wu, X.-H. 1997 A multiscale finite element method for elliptic
problems in composite materials and porous media. J. Comput. Phys. 134, 169–
189.

Hou, T. Y., Wu, X.-H. & Cai, Z. 1999 Convergence of a multiscale finite ele-
ment method for elliptic problems with rapidly oscillating coefficients. Math. Comp.
68 (227), 913–93.

House, K. Z., D. P. Schrag, C. F. Harvey & Lackner, K. S. 2006 Permanent

carbon dioxide storage in deep-sea sediments. P. Natl. Acad. Sci. USA 103, 12291–
12295.
BIBLIOGRAPHY 178

Huppert, H. E. & Woods, A. W. 1995 Gravity-driven flows in porous media. J.

Fluid Mech. 292, 55–69.

Ide, S. T., Jessen, K. & F. M. Orr Jr. 2007 Storage of CO2 in saline aquifers:
effects of gravity, viscous, and capillary forces on amount and timing of trapping.
Int. J. Greenh. Gas Control 1 (4), 481–491.

Jenny, P., Lee, S. H. & Tchelepi, H. A. 2003 Multi-scale finite-volume method

for elliptic problems in subsurface flow simulation. J. Comput. Phys. 187, 47–67.

Jenny, P., Lee, S. H. & Tchelepi, H. A. 2005 Adaptive multiscale finite volume
method for multi-phase flow and transport. SIAM Multiscale Model. Simul. 3 (1),
50–64.

Jenny, P., Lee, S. H. & Tchelepi, H. A. 2006 Adaptive fully implicit multi-scale
finite-volume method for multi-phase flow and transport in heterogeneous porous
media. J. Comput. Phys. 217, 627–641.

Jessen, K., Kovscek, A. R. & F. M. Orr Jr., 2005 Increasing CO2 storage in
oil recovery. Energy Conv. Manag. 46 (2), 293–311.

Jhavery, B. S. & Homsy, G. M. 1982 The onset of convection in fluid layer heated
rapidly in a time-dependent manner. J. Fluid Mech. 114, 251.

Johnson, J. W. & Nitao, J. J. 2003 Reactive transport modeling of geologic

CO2 sequestration at Sleipner. In Proceedings of the 6th International Conference
on Greeenhouse Gas Technologies (ed. J. Gale & Y. Kaya), , vol. 1, pp. 327–333.
Elsevier Science Ltd.

Journel, A. G. 1986 Geostatistics - models and tools for the Earth-sciences. Math-
ematical Geology 18 (1), 119–140.

Juanes, R. & Dub, F.-X. 2008 A variational multiscale method with locally con-
servative treatment of multiscale source terms. Comput. Geosci., Special Issue on
Multiscale Methods for Flow and Transport in Heterogeneous Porous Media, doi
10.1007/s10596-007-9070-x (In Press) .
BIBLIOGRAPHY 179

Juanes, R., Spiteri, E. J., F. M. Orr Jr. & Blunt, M. J. 2006 Impact
of relative permeability hysteresis on geological CO2 storage. Water Resour. Res.
42 (W12418), 1–13.

Kaarstad, O. 1992 Emission-free fossil energy from norway. Energy Convers. Man-
age. 33 (5-8), 781786.

Katzer, J. 2007 The future of coal, options for a carbon constrained world. Tech.
Rep.. Massachusetts Institute of Technology.

Kaviany, M. 1984 Onset of thermal convection in a saturated porous medium:

experiment and analysis. International Journal of Heat and Mass Transfer 27,
2101–2110.

Kippe, V., Aarnes, J. E. & Lie, K.-A. 2007 A comparison of multiscale meth-
ods for elliptic problems in porous media flow. Comput. Geosci., Special Issue on
Multiscale Methods for Flow and Transport in Heterogeneous Porous Media, DOI
10.1007/s10596-007-9074-6 (In Press) .

Kochina, I. N., Mikhailov, N. N. & Filinov, M. V. 1983 Groundwater mound

damping. Int. J. Eng. Sci. 21, 413–421.

Koide, H., Shindo, Y., Tazaki, Y., Iijima, M., Ito, K., Kimura, N. &
Omata, K. 1997 Deep sub-seabed disposal of CO2 - the most protective storage.
Energy Convers. Manage. 38, S253–S258.

Koide, H., Takahashi, M., Tsukamoto, H. & Shindo, Y. 1995 Self-trapping

mechanisms of carbon-dioxide in the aquifer disposal. Energy Convers. Manage.
36 (6-9), 505–508.

Koide, H., Tazaki, Y., Noguchi, Y., Nakayama, S., Iijima, M., Ito, K.
& Shindo, Y. 1992 Subterranean containment and long-term storage of carbon
dioxide in unused aquifers and in depleted natural gas reservoirs. Energy Convers.
Manage. 33 (5-8), 619626.
BIBLIOGRAPHY 180

Koperna, G. J. & Kuuskraa, V. A. 2006 Assessing technical and economic

recovery of oil resources in residual oil zones. Tech. Rep.. Advanced Resources
International, 4501 Fairfax Drive, Suite 910, Arlington, VA 22203 USA.

Kovscek, A. R. & Cakici, C. M. 2005 Geological storage of carbon dioxide and

enhanced oil recovery I. Cooptimization of storage and recovery. Energy Conv.
Manag. 46 (11-12), 1941–1956.

Kovscek, A. R. & Wang, Y. 2005 Geological storage of carbon dioxide and en-
hanced oil recovery I. Uncertainty quantification employing a streamline based
proxy for reservoir simulation. Energy Conv. Manag. 46 (11-12), 1920–1940.

Kumar, A., Ozah, R., Noh, M., Pope, G. A., Bryant, S., Sepehrnoori, K.
& Lake, L. W. 2005 Reservoir simulation of CO2 storage in deep saline aquifers.
Soc. Petrol. Eng. J. pp. 336–348.

Lake, L. L. 1989 Enhanced Oil Recovery. Prentice-Hall, Inc.

Lambers, J. V., Gerritsen, M. G. & Mallison, B. T. 2008 Accurate local

upscaling with variable compact multi-point transmissibility calculations. Comput.
Geosci., Special Issue on Multiscale Methods for Flow and Transport in Heteroge-
neous Porous Media, doi 10.1007/s10596-007-9068-4 (In Press) .

Lee, S. H., Durlofsky, L. J., Lough, M. F. & Chen, W. H. 1998 Finite

difference simulation of geologically complex reservoirs with tensor permeabilities.
Soc. Petrol. Eng. Res. Eval. & Eng. pp. 567–574.

Lee, S. H., Tchelepi, H. A., Jenny, J. & DeChant, L. J. 2002 Implementation

of a flux-continuous finite-difference method for stratigraphic hexahedron grids.
Soc. Petrol. Eng. J pp. 267–277.

Lee, S. H., Wolfsteiner, C. & Tchelepi, H. A. 2008 Multiscale finite-volume

method for multiphase flow in porous media: Black oil formulation of compressible,
three phase flow with gravity and capillary force. Comput. Geosci., Special Issue
BIBLIOGRAPHY 181

on Multiscale Methods for Flow and Transport in Heterogeneous Porous Media, doi
10.1007/s10596-007-9069-3 (In Press) .

Legg, J. F. 1992 Overview of carbon dioxide removal and disposal in canada. Energy
Convers. Manage. 33 (5-8), 787–794.

Lenormand, R., Zarcone, C. & Sarr, A. 1983 Mechanisms of the displacement

of one fluid by another in a network of capillary ducts. J. Fluid Mech. 135, 123132.

Leonenko, Y. & Keith, D. W. 2008 Reservoir engineering to accelerate the dis-

solution of CO2 stored in aquifers. Environ. Sci. Technol. 42 (8), 2742–2747.

Leveque, R. J. 2002 Finite Volume Methods for Hyperbolic Problems. Cambridge

University Press.

Levine, J. S., Matter, J. M., Goldberg, D., Cook, A. & Lackner, K. S.

2007 Gravitational trapping of carbon dioxide in deep sea sediments: permeability,
buoyancy, and geomechanical analysis. Geophys Res. Lett. 34 (L24703).

Lick, W. 1964 The instability of a fluid layer with time-dependent heating. J. Fluid
Mech. 21, 565–576.

Lindeberg, E. & Bergmo, P. 2003 The long-term fate of CO2 injected into an
aquifer. In Proceedings of the 6th International Conference on Greeenhouse Gas
Technologies (ed. J. Gale & Y. Kaya), , vol. 1, pp. 489 – 494. Elsevier Science Ltd.

Lindeberg, E. & Wessel-Berg, D. 1997 Vertical convection in an aquifer column

under a gas cap of CO2 . Energy Conservation and Management 38, SS229–SS234.

Lister, J. R. 1992 Viscous flows down an inclined plane from point and line sources.
J. Fluid Mech. 242, 631–653.

Lunati, I. & Jenny, P. 2006 Multiscale finite-volume method for compressible

multiphase flow in porous media. J. Comput. Phys. 216, 616–636.

Lunati, I. & Jenny, P. 2007 Treating highly anisotropic subsurface flow with the
multiscale finite-volume method. SIAM Multiscale Model. Simul. 6 (1), 308–318.
BIBLIOGRAPHY 182

Lunati, I. & Jenny, P. 2008 Multiscale finite-volume method for density-driven

flow in porous media. Comput. Geosci., Special Issue on Multiscale Methods for
Flow and Transport in Heterogeneous Porous Media, doi 10.1007/s10596-007-9071-
9 (In Press) .

Lund, J. W., Freeston, D. H. & Boyd, T. L. 2005 Direct application of geother-

mal energy: 2005 worldwide review. Geothermics 34 (6), 691–727.

Lyle, S., Huppert, H. E., Hallworth, M., Bickle, M. & Chadwick, A. 2005
Axisymmetric gravity currents in a porous medium. J. Fluid Mech. 543, 293–302.

Manickam, O & Homsy, G. M. 1995 Fingering instability in vertical miscible

displacement flows in porous media. J. Fluid Mech. 288, 75.

Marchetti, S. 1977 On geoengineering and the CO2 problem. Climatic Change 1,

59–68.

McGrail, B. P., Schaef, H. T., Ho, A. M., Chien, Y.-J., Dooley, J. J. &
Davidson, C. L. 2006 Potential for carbon dioxide sequestration in flood basalts.
J. Geophys. Res. 111 (B12), B12201.

van der Meer, L. G. H. 1992 Investigations regarding the storage of cabon dioxide
in aquifers in the Netherlands. Energy Convers. Manage. 33 (5-8), 611–618.

Metz, B., Davidson, O., de Coninck, H., Loos, M. & Meyer, L., ed.
2006 Special Report on Carbon Dioxide Capture and Storage. Cambridge University
Press.

Mo, S. & Akervoll, I. 2005 Modeling long-term CO2 storage in aquifer with a
black-oil reservoir simulator. In SPE/EPA/DOE Exploration and Production En-
vironmental Conference. Galveston, TX.

Mo, S., Zweigel, P., Lindeberg, E. & Akervoll, I. 2005 Effect of geologic
parameters on CO2 storage in deep saline aquifers. In SPE Eurospec/EAGE Annual
Conference. Madrid, Spain.
BIBLIOGRAPHY 183

Nicot, J.-P. 2008 Evaluation of large-scale CO2 storage on fresh-water sections of

aquifers: An example from the Texas Gulf Coast Basin. (In Press) Int. J. Greenh.
Gas Control .

Nield, D. A. & Bejan, A. 1999 Convection in porous media, 2nd edn. Springer
Verlag.

Nordbotten, J. M. & Aavatsmark, I. 2005 Monotonicity conditions for control

volume methods on uniform parallelogram grids in homogeneous media. Comput.
Geosci. 9, 61–72.

Nordbotten, J. M., Aavatsmark, I. & Eigestad, G. T. 2007 Monotonicity of

control volume methods. Numer. Math. 106, 255–288.

Nordbotten, J. M., Celia, M. A. & Bachu, S. 2005 Injection and storage

of CO2 in deep saline aquifers: Analytical solution for the CO2 plume evolution
during plume injection. Transport Porous Med. 58, 339–360.

Nordbotten, J. N., Celia, M. A. & Bachu, S. 2004 Analytical solutions for

leakage rates through abandoned wells. Water Resour. Res. 40 (4), W04204.

Oldenburg, C. M. & Unger, A. J. A> 2005 Coupled vadose zone and at-
mospheric surface-layer transport of carbon dioxide from geologic carbon seques-
tration sites. Vadose Zone J. 3 (3), 848–857.

Otero, J., Dontcheva, L. A., Johnston, H., Worthing, R. A., Kurganov,

A., Petrova, G. & Doering, C. R. 2004 High-rayleigh-number convection in
a fluid-saturated porous layer. J. Fluid Mech. 500, 263–281.

Pacala, S. W. & Socolow, R. 2004 Stabilization wedges: solving the climate

problem for the next 50 years with current technologies. Science 305, 968–972.

Pal, M., Edwards, M. G. & Lamb, A. R. 2007 Convergence study of a family

of flux-continuous finite-volume schemes for the general tensor pressure equation.
(submitted) Int. J. Num. Meth. in Fluids .
BIBLIOGRAPHY 184

Parker, J. C. & Lenhard, R. J. 1989 Vertical integration of three-phase flow

equations for analysis of light hydrocarbon plume movement. Transport Porous
Med. 5, 187–206.

Pego, R. L. & Weinstein, M. I. 1994 Asymptotic stability of solitary waves.

Commun. Math. Phys. 104, 305.

Prassad, A. & Simmons, C. T. 2003 Unstable density driven flow in heteroheneous

porous media: A stochastic study of the Elder [1967b] “short heater” problem.
Water Resour. Res. 39, 1–21.

Pritchard, D. 2004 The instability of thermal and fluid fronts during radial injec-
tion in porous media. J. Fluid Mech. 508, 133.

Pruess, K. 2005 Numerical studies of fluid leakage from a geologic disposal reservoir
for CO2 show self-limiting feedback between fluid flow and heat transfer. Geophys-
ical Research Letters 32 (L14404), 1–4.

Reed, A. C., Mathews, J. L., Bruno, M. S. & Olmstead, S. E. 2002 Safe dis-
posal of one million barrels of NORM in Louisiana through slurry fracture injection.
SPE Drill. Completion 17 (2), 72–81.

Riaz, A. & Meiburg, E. 2003a Radial source flows in porous media: Linear sta-
bility analysis of axial and helical perturbations in miscible displacements. Phys.
Fluids 15 (4), 938.

Riaz, A. & Meiburg, E. 2003b Three-dimensional miscible displacement simu-

lations in homogeneous porous media with gravity override. J. Fluid Mech. 494,
95.

Riaz, A. & Meiburg, E. 2004a Linear stability of radial displacements in porous

media: influence of velocity-induced dispersion and concentration-dependent diffu-
sion. Phys. Fluids 16 (10), 3592.
BIBLIOGRAPHY 185

Riaz, A. & Meiburg, E. 2004b Vorticity interaction mechanisms in variable-

viscosity heterogeneous miscible displacements with and without density contrast.
J. Fluid Mech. 517, 1.

Riaz, A. & Tchelepi, H. A. 2006 Numerical simulation of immiscible two-phase

flow in porous media. Physics of Fluids 18 (014104).

Robinson, J. L. 1976 Theoretical analysis of convective instability of a growing

horizontal thermal boundary layer. Physics of Fluids 19, 778–791.

Ruge, J. W. & Stüben, K. 1987 Multigrid Methods, chap. 4, pp. 73–130. SIAM.

Ruith, M. & Meiburg, E. 2000 Miscible rectilinear displacements with gravity

override. Part 1. Homogeneous porous medium. J. Fluid Mech. 420, 225.

Rumynin, V. G., Konosavsky, P K & Hoehn, E 2005 Experimental and mod-

eling study of adsorption-desorption processes with application to a deep-well in-
jection radioactive waste disposal site. J. Contam. Hydrol. 76 (1-2), 19–46.

Saffman, P. G. & Taylor, G. I. 1958 The penetration of a fluid into a porous

medium or hele-shaw cell containing a more viscous fluid. Proc. R. Soc. London
Ser. A 245, 312–329.

Saripalli, K. P., Sharma, M. M. & Bryant, S. L. 2000 Modeling injection

well performance during deep-well injection of liquid wastes. J. Hydrol. 227 (1-4),
41–55.

Silin, D., Patzeck, T. W. & Benson, S. M. 2006 Exact solutions in a model of

vertical gas migration. SPE 103145, San Antonio, TX.

Stevens, S. H., Kuuskra, V. A., Gale, J. & Beecy, D. 2001 CO2 injection
and sequestration in depleted oil and gas fields and deep coal seams: worldwide
potential and costs. Environmental Geosciences 8 (3), 200–209.

Strauss, W. A. 1992 Partial Differential Equations, an introduction. John Wiley

& Sons.
BIBLIOGRAPHY 186

Tan, C. T. & Homsy, G. M. 1987 Stability of miscible displacements in porous

media: radial source flow. Phys. Fluids 30 (5), 1239.

Tan, C. T. & Homsy, G. M. 1988 Simulation of nonlinear viscous fingering in

miscible displacement. Phys. Fluids 31 (6), 1330.

Tchelepi, H. A. 1994 Viscous fingering, gravity segregation and permeability het-

erogeneity in two dimensional and three dimensional flows. PhD thesis, Department
of Petroleum Engineering, School of Earth Sciences, Stanford University.

Tchelepi, H. A. & F. M. Orr Jr. 1994 Interaction of viscous fingering, per-

meability inhomogeneity and gravity segregation in three dimensions. SPE Res.
Engng. p. 266.

Tchelepi, H. A., F. M. Orr Jr., Rakotomalala, N., Salin, D. & Woumeni,

R. 1993 Dispersion, permeability heterogeneity and viscous fingering: acoustic ex-
perimental observations and particle tracking simulations. Phys. Fluids 5 (7), 1558.

Torp, T. A. & Gale, J. 2004 Demonstrating storage of CO2 in geological reservoirs:

the Sleipner and SACS projects. Energy 29, 13611369.

Trottenberg, U., Oosterlee, C. W. & Schüller, A. 2001 Multigrid . Acad-

emic Press.

Union of Concerned Scientists 2001 Policy context of geologic carbon seques-

tration. Tech. Rep..

Vella, D. & Huppert, H. E. 2006 Gravity currents in a porous medium at an

inclined plane. J. Fluid Mech. 555, 353–362.

Voss, C. I. & Sousa, W. R. 1987 Variable density flow and solute transport
simulation of regional aquifers containing a narrow freshwater-saltwater transition
zone. Water Resour. Res. 28, 1851.

Woods, A. W. 2002 Gravity Driven Flows in Porous Rocks: Effects of layering,

reaction, boiling and double diffusion, Transport phenomena in porous media, vol. 2.
Pergamon.
BIBLIOGRAPHY 187

Wu, X. H., Efendiev, Y. & Hou, T. Y. 2002 Analysis of upscaling absolute

permeability. Discrete Contin. Dyn. Syst. 2 (2), 185–204.

Xu, T., Apps, J. A. & Pruess, K. 2003 Reactive geochemical transport simulation
to study mineral trapping for CO2 disposal in deep arenaceous formations. Journal
of Geophysical Research 108 (doi:10.1029/2002JB001979,).

Yang, C. & Gu, Y. 2006 Accelerated mass transfer of CO2 in reservoir brine due
to density-driven natural convection at high pressures and elevated temperatures.
Ind. Eng. Chem. Res. 45, 2430–2436.

Yortsos, Y. C. 1995 A theoretical analysis of vertical flow equilibrium. Transport

Porous Med. 18, 107–129.

Yortsos, Y. C. & Hickernell, F. J. 1989 Linear stability of immiscible displace-

ment in porous media. SIAM J. Appl. Math. 49 (3), 730–748.

Yortsos, Y. C. & Zeybek, M. 1988 Dispersion driven instability in miscible

displacement in porous media. Phys. Fluids 31 (12), 3511.

Zhang, Y. Q., Oldenburg, C. M. & Benson, S. M. 2005 Vadose zone reme-

diation of carbon dioxide leakage from geologic carbon dioxide sequestration sites.
Vadose Zone J. 3 (3), 858–866.

Zhou, H. & Tchelepi, H. A. 2008 Operator based multiscale method for com-
pressible flow. Soc. Petrol. Eng. J. (June), pp. 1–7.

Zhou, Q. L., Bear, J. & Bensabat, J. 2003 Saltwater upconing and decay
beneath a well pumping above an interface zone. Transport Porous Med 61 (3),
337–363.

Zimmerman, W. B. & Homsy, G. M. 1992 Viscous fingering in miscible dis-

placements: Unification of effects of viscosity contrast, anisotropic dispersion and
velocity dependence of dispersion on nonlinear finger propagation. Phys. Fluids
4 (11), 2348.