Environ. Sci. Technol.
2009, 43, 743–749
Model for CO2 Leakage Including
Multiple Geological Layers and
Multiple Leaky Wells
JAN M. NORDBOTTEN,*
DMITRI KAVETSKI, MICHAEL A. CELIA,
AND STEFAN BACHU
Department of Mathematics, University of Bergen,
Bergen, Norway
Received May 13, 2008. Revised manuscript received
October 3, 2008. Accepted October 15, 2008.
Geological storage of carbon dioxide (CO2) is likely to be an
integral component of any realistic plan to reduce anthropogenic
greenhouse gas emissions. In conjunction with large-scale
deployment of carbon storage as a technology, there is an urgent
need for tools which provide reliable and quick assessments
of aquifer storage performance. Previously, abandoned wells from
over a century of oil and gas exploration and production
have been identified as critical potential leakage paths. The
practical importance of abandoned wells is emphasized by the
correlation of heavy CO2 emitters (typically associated with
industrialized areas) to oil and gas producing regions in North
America. Herein, we describe a novel framework for predicting
the leakage from large numbers of abandoned wells, forming
leakage paths connecting multiple subsurface permeable
formations. The framework is designed to exploit analytical
solutions to various components of the problem and, ultimately,
leads to a grid-free approximation to CO2 and brine leakage
rates, as well as fluid distributions. We apply our model in a
comparison to an established numerical solver for the underlying
governing equations. Thereafter, we demonstrate the
capabilities of the model on typical field data taken from the
vicinity of Edmonton, Alberta. This data set consists of over 500
wells and 7 permeable formations. Results show the flexibility
and utility of the solution methods, and highlight the role
that analytical and semianalytical solutions can play in this
important problem.
1. Introduction
Anthropogenic emissions have increased atmospheric con-
centrations of CO2 by about 35% during the past 200 years,
to a current value of about 385 ppm, well above the pre-
industrial level of 280 ppm (12). If the relentless increase is
to be reduced or reversed, technological solutions must be
implemented on a massive scale. While many options are
being considered, one attractive approach is carbon capture
and storage, or CCS (12, 13, 21). CCS involves the capture of
CO2 before it is emitted into the atmosphere and subsequent
storage away from the atmosphere. Within the storage options
for CCS, geological storage appears to be the most promising.
Geological storage involves injection of the captured CO2
into deep geological formations, where it would reside for
centennial to millennial time scales or longer. There is general
* Corresponding author phone: +47 55584869;
jan.nordbotten@math.uib.no.
10.1021/es801135v CCC: $40.75  2009 American Chemical Society
Published on Web 12/17/2008
agreement that the scale of the carbon problem will require
deep saline aquifers to be major subsurface repositories for
the CO2, in addition to depleted oil and gas reservoirs (with
or without enhanced oil recovery) and other subsurface
options (12).
Success in a geological CCS program will depend on the
ability to choose formations from which only minimal
amounts of CO2 will leak. This requires formations with
sufficient permeability to accept the CO2 injection rates, and
which are overlaid by one or more low-permeability units
referred to as caprock formations. One leakage pathway of
concern involves the large number of wells that may penetrate
otherwise excellent caprock formations (12). The century-
long legacy of oil and gas exploration, especially in North
America, has left many millions of exploration and production
wells, most of which perforate otherwise intact caprock
formations. In some locations, a typical plume of injected
CO2 might encounter tens to hundreds of existing wells (6).
If the large-scale injection of CO2 is viewed in this perspective,
then we may have a system similar to that shown schemati-
cally in Figure 1. Simulation of flow and leakage in this system
constitutes a severe computational challenge that cannot be
solved easily and efficiently by standard simulation methods
(1). Addressing CO2 storage from a risk assessment (3) or
system level (24) perspective will further increase the
computational effort required.
Herein we describe a semianalytical, grid-free approxi-
mation to the CO2 migration and leakage problem. Within
the context of a strong set of assumptions, the solution
captures the large-scale migration of both CO2 and brine
while still resolving the impacts of all of the potentially leaky
wells in the system. The model expands on earlier work of
the authors in several important ways. It builds from a simpler
version of this approach for the system of one injection well
and one leaky well, over two permeable layers (20) to include
arbitrary numbers of wells, and an arbitrary number of layers
of permeable formations. We have incorporated a new
formulation for interface upconing along a leaky well (18),
and we have developed robust solution methods for both
the pressure fields and saturation. Several example calcula-
tions are presented to demonstrate the flexibility and utility
of the model.
2. Fluid Migration and Leakage Model
Injection of CO2 into a deep saline aquifer involves many
complicated processes, including two-phase flow dynam-
ics, interphase mass transfer (dissolution), geochemical
reactions, and possible nonisothermal effects. While these
processes are all interesting and potentially important over
different length and time scales, we will focus on the most
FIGURE 1. Diagrammatic representation of pathways for CO2
migration and possible leakage through and along degraded
e-mail: wells. Reprinted with permission from the work of Bachu and
Celia (1) (modified from the work of Gasda et al. (6)) 2008 AGU.
VOL. 43, NO. 3, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 743
critical processes associated with possible leakage of
injected CO2 with an emphasis on leakage along existing
wells. Since the largest driving force for leakage occurs
during injection, when both pressure and strong buoyancy
govern the flow, we focus on the injection phase of a storage
operation. In addition, since dissolution and geochemical
reactions usually occur over longer time scales (12), we
will ignore these processes. Thermal effects are only
considered weakly, in that the fluid properties are allowed
to vary between aquifers.
2.1. Governing Equations and Model System. We con-
sider injection of CO2 into a deep saline aquifer and therefore
will model two fluid phases, namely, brine and a CO2-rich
phase that for simplicity we will refer to as CO2. We assume
the flow field can be described by the multiphase extension
of Darcy’s law,
KkR
vR ) - (∇pR + FRg ∇z)
µR
In eq 1, vR [L T-1] is the volumetric flux of phase R, pR [M L-1
T-2] is the fluid pressure, K and kR are the intrinsic [L2] and
relative [L0] permeabilities, FR and µR are the density [M L-3]
and dynamic viscosity [M L-1 T-1], respectively, g is the
gravitational constant [L T-2], and z is the vertical coordinate
[L], taken positive upward.
The equations of mass conservation can be stated for the
carbon (R ) c) and brine (R ) b) phases,
∂φFRSR
+ ∇ · (FRvR) ) QR, R ) c, b
∂t
The additional notation introduced represents the fluid
saturation SR [L0], time t [T], porosity φ [L0], and the source
(or sink) term QR [MT-1], which in our application will be
zero everywhere except at wells.
We close the system by requiring that the fluid saturations
fill the media, and by ignoring capillary pressure, the effect
of which is considered small at large spatial scales. This allows
us to use the relations
Sc + Sb ) 1, p c ) pb
As initial conditions, we take the system to be fully
saturated with brine (Sb ) 1) at hydrostatic pressure,
pb(x, 0) ) p0 - Fbgz
for some datum pressure p0, and coordinate vector x ) [x,
y, z]T.
We model the subsurface as a stack of alternating
permeable (aquifer) and low-permeable (aquitard) forma-
tions, which is typical of many sedimentary basins. In the
current work, we consider all aquitards to be impermeable;
however, we are working to relax this assumption in the
future. The system is considered to be (partially or fully)
penetrated by a number of wells. The aquitards will be
impermeable, thus we can omit them from the domain, and
take the domain Ω as the Nf aquifers and Nw wells. To enable
analytical treatment, we will limit the discussion to homo-
geneous and horizontal aquifers, and we will consider each
well segment (that is, each segment of a well passing through
an aquifer or aquitard) as having uniform properties. A
discussion of the assumption of homogeneous horizontal
aquifers, for this problem can be found in ref 8. Note that
each aquifer may have different properties; similarly, each
well segment (whether along the same well or different wells)
may also have different properties.
A CO2 plume arises due to injection into a given permeable
layer. The plume spreads into the formation and eventually
encounters some number of existing wells. Each well is
characterized by a set of effective permeability values, with
744 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 3, 2009
one value assigned to each segment of the well that crosses
a caprock formation. If there are Nf - 1 caprock (or aquitard)
formations, then each well in the system will have Nf - 1
values of effective permeability assigned to it. Leakage across
the caprock formations is restricted to flow along these wells,
and is therefore governed by these assigned values of effective
permeability. If CO2 leaks along one or more of the wells,
the CO2 will migrate upward, where it will encounter one or
more of the overlying permeable formations. Depending on
the flow conditions, some of the leaking CO2 may flow into
these intervening permeable formations, creating new CO2
plumes. These secondary plumes may in turn intersect some
number of existing wells, leading to additional leakage, and
so on. Assuming this basic model of flow and leakage, the
important parameters for the system include the permeability
and thickness of each of the permeable layers, the thickness
of each caprock formation, the CO2 injection rate, the fluid
properties (density and viscosity) of the CO2 and brine, the
(1)
size of the overall domain and the associated boundary
conditions along the boundaries, and the assignment of
effective well permeabilities along each segment of each well.
Note that in the solution for flow and leakage, the dynamics
of both the CO2 and the brine will be modeled. This allows
leakage dynamics of both fluids to be simulated and analyzed.
2.2. Approximation of Solution. Most standard mul-
tiphase flow simulators can solve eqs 1 and 2 subject to
the given boundary conditions (see, e.g, 4, 5, 10). However,
for the simulations of interest in the context of CCS, the
overall problem requires very large computational do-
(2) mains, while simultaneously requiring local grid refine-
ment around leaky wells. Because we may need to deal
with hundreds of wells within the domain of influence of
the injection (6) and on the order of 10 aquifers between
the injection formation and the surface, high-quality grids
conforming to this geometry easily require many millions
of grid cells, implying prohibitive computational cost even
for large-scale parallel computing.
In this section, we will outline a different approach, which
uses a much sparser spatial representation, where each node
corresponds to a leaky well, rather than to an arbitrary spatial
(3) location. This dramatically reduces the computational cost
of the simulation, essentially because the number of un-
knowns is the total number of well segments, e.g., 10 000 for
a system with 10 layers and 1000 wells, rather than the millions
(4) of unknowns if the flow equations are solved for closely
spaced spatial locations (as required by standard multiphase
simulators).
In deep terrestrial systems, the injected CO2 will be less
dense than the resident brine, with CO2 density between
25% and 75% of the brine density (19). The resulting strong
buoyant drive leads to gravity segregation, which makes the
assumption of a macroscopic sharp interface reasonable.
Under the associated assumption of essentially horizontal
flow, we can write the vertically integrated form of eq 2 for
an aquifer of thickness H as
∫ ∫ ∫
∂ H H H
φFRSR dz + ∇ · FRvR dz ) QR dz, R ) c, b
∂t 0 0 0
(5)
We assume a sharp interface separating CO2 (with residual
brine saturation) above the interface and brine below the
interface. Define the thickness of the CO2 layer as h(x, t).
With the sharp interface assumption, h(x, t) fully describes
the saturation distribution in the aquifer. Assuming
negligible vertical variation in porosity and densities, eq
5 simplifies to
( ∫ ) ∫
∂ H H
(φFc(1 - Sres)h) + ∇ · Fc vc dz ) Qc dz (6)
∂t 0 0
 ( ∫ ) ∫
∂ H H ∂p
(φFb(H - h) + φFbSresh) + ∇ · Fb vb dz ) Qb dz G(x, xi, h, t) ) (17)
∂t 0 0 ∂Q̃i
(7)
These quantities represent the sensitivity of the pressure field
Applying the vertical equilibrium assumption, which is to the sources and sinks. As such, they are effectively Green’s
consistent with the assumption of essentially horizontal flow, functions for a linearization of the pressure solution. Such
yields the vertically integrated flux from eq 1 as, a linearization is appropriate for a single well operating at
a constant rate of injection (see eq 20 later and ref 16).
Kkch
∫ Assuming that this linearization is appropriate for the full
H
vc dz ) - ∇ (pB + h(pb - pc)g) (8)
0 µc system, we can construct the solution by superposition,

∫ K(H - h)
H Nw

0
vb dz ) -
µb
∇ pB (9)
p(x, t) ) p0 + ∑ Q̃ G(x, x , h, t) + F(x, h, t)
i)1
i i (18)

Here kc is the relative permeability of CO2 at the residual

The offset F appears as a consequence of considering the
brine saturation, pB(x, y, t) ) p(x, y, zl ) 0, t), where zl is the
pressure at the bottom of the aquifer (recall that p ) pB),
local vertical coordinate in the aquifer. The pressures at the
which is thus dependent on the vertical fluid distribution at
top and bottom of the aquifer, pT and pB, respectively, are
that point. In terms of standard pressure-saturation for-
related through
mulations, pB - F is analogous to the so-called global pressure
pT ) pB - (H - h)Fbg - hFcg (10) for the system. We discuss the offset function F further in
section 2.3.1.
Summation over eqs 6 and 7 weighted by FR-1 leads to 2.3. Analytical Solutions. To complete our model de-
scription, we apply two additional analytical components.
h ( 1 - Sres ∂
Fc
φF +
Sres ∂
∂t c Fb ∂t b
φF + (H - h) )
1 ∂ 1
φF + ∇ ·
Fb ∂t b Fc
The choices of these components are not intrinsic to the
modeling concept and could be substituted for other models
depending on the accuracy needed (see, e.g., ref 9 for
( ∫ 1
( ∫ ) ∫
H H H
Fc
0 Fb )
vc dz + ∇ · Fb
0
vb dz )
0
Q̃ dz (11) preliminary work on integrating traditional numerical meth-
ods). As examples, the injection response described in section
We denote the total volumetric source (or sink) term as 2.3.1 might be generalized to include caprock flow following
the ideas of Hunt (11). Conversely, the upconing model given
Qc Q b in section 2.3.2 can be simplified by using e.g. Muskat’s
Q̃ ) + (12) approximation (15). And we could replace the assumed Darcy
Fc Fb
flow in leaky wells by more complex pipeflow models if that
To proceed to an analytically tractable pressure equation, were appropriate.
we make two assumptions. First, we approximate the Our first component is a similarity solution for two-phase
compressibility of CO2 and brine as equal and define the injection into porous media. This solution provides an
effective compressibility as expression from which estimates of the two primary variables,
pressure pB(x, t) and interface location h(x, t), can be made
ceff ≡ ( 1 - Sres ∂
Fc ∂p
φFc +
Sres ∂
Fb ∂p b
φF )
1 ∂
φF
Fb ∂p b ) (13)
for any individual plume. These estimates allow the functions
G and F to be calculated, thereby forming the basis for an
overall solution algorithm.
Second, let the spatial variability of density be negligible The second component is a generalization of traditional
within a given fluid phase, within a given aquifer. Then, eq upconing models for two-phase flow into a well. This solution
11, simplifies, provides us with the near-well flow field and saturation
distribution needed to more accurately compute flows along
ceff
∂p
∂t (
- ∇ · Kλeff ∇ p +
Kkch(Fb - Fc)g
µcH
∇h )
1
H ) ∫
H

0
Q̃ dz
any given well.
2.3.1. Fluid Injection into Confined Aquifers. In a previous
investigation (16), we considered the case of the constant-
(14)
rate injection into confined aquifers initially filled with brine.
The solution used the sharp interface assumption. It was
Here we have introduced the effective mobility given by (note
shown that the saturation distribution, given by h(x, t) has
that the relative permeability of brine kb ) 1)
a self-similar solution in time, such that the solution could
kch H - h be expressed in terms of a single space-time variable h(x, t)
λeff ) + (15) ) h(ri2/t), where ri ) ||x - xi||2 is the distance to the well. In
µcH µbH
self-similar coordinates, the equations governing the fluid
distribution and pressure are (see eqs 11 and 17 of ref 16)
Finally, within an aquifer, the source terms Qc and Qb appear
as point sources located at the wells, dh′ 2 d 1 - h′ dh′
Nw
) [
dχ χ dχ h′(λ - 1) + 1
2Γλhχ ′
dχ( +1 )] (19)

QR(x) ) ∑ δ(x - x )Q
i)1
i R,i (16) ∆p′ ) -
1
2Γ ∫ χ
ψ dχ
(h′(λ - 1) + 1)χ
+ F(h′) (20)

where δ(x - xi) is the Dirac delta function centered at the The dimensionless variables and parameters in these equa-
well location x ) xi. tions are defined as
Consider the formal solution p of eq 14 at some time t.
2πφH(1 - Sres)ri2
If the inverse of compressibility is large relative to the χ) (21)
perturbation from initial pressure in the system, which means Mc(t) ⁄ Fc
that pressure equilibrates fast relative to saturation changes,
2π(Fb - Fcw)gKλbH2 λc
we can approximate the pressure equations by superposition Γ) , λ) (22)
and introduce the quantities Q̃ λb

VOL. 43, NO. 3, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 745

 p - p0 h flow and significant thermal effects (22). These complex
∆p′ ) , h′ ) (23)
(Fb - Fc)gH H
where Mc(t) is the cumulative injected mass of CO2.
By comparison of eq 20 to eq 17 of the work of Nordbotten
and Celia (16), we see that the pressure offset F in eq 19 due
to the vertical fluid distribution given by
ln[(λ - 1)h′ + 1]
F(h′) ) -
λ
λ-1 [
h′(χ) -
λ-1 ] (24)
The compressibility of the system, ceff, enters through the
outer limit of integration in the pressure equation, and is
analogous to successive steady-state approximations (see
refs 18 and 20),
4.5πHφλbK(1 - Sres)
Ψ) (25)
ceffQ̃
The advantage of using the self-similar approximation
given in eq 19 is that it replaces a system of 3D nonlinear
partial differential equations (eqs 6 and 7) by a single
nonlinear ordinary differential equation. Even if the latter is
solved numerically, this strategy produces tremendous
computational savings. For the limit of Γ f 0, analytical
solutions to eqs 19 and 20 exist, which allows for simple
closed-form analytical solutions.
2.3.2. Fluid Upconing near Wells. Understanding the near-
well behavior of multiphase flow has a long history in both
groundwater and petroleum engineering. This is also an
important issue in the CO2 leakage problem, because near-
well flows may strongly influence the saturation in the wells
and, hence, larger-scale flow patterns.
Analytical solutions to near-well flow are typically ob-
tained by either (i) assuming vertical equilibrium or (ii)
applying perturbation equations around a known solution.
Since potentially large vertical flows near leaky wells may
cause significant local upconing effects, neither of these
traditional approaches gives satisfactory results in the current
application. Therefore we use a new solution approach, which
partially relaxes the vertical equilibrium assumption by
explicitly approximating vertical flow (17).
Single-phase flow in the well (Sc ) 1) occurs when the
upconing, near the well, is not sufficient to overcome the
fluid layering. This will be the case when the flow rates in
the well are small, or when the thickness of a CO2 plume at
a well is large. Using eqs 30 and 31 of ref 17, we see that this
situation occurs when the flow of the dense fluid (brine) in
the well is zero, Qd′ ) 0. Conditions for this are given by
k′ k′
(
ηouter2 < ηwell2 +
4Γλ)exp((h
0)2Γλ) -
4Γλ
(26)
Here η ) ri/Kh is a dimensionless radial distance scaled by
the horizontal permeability, and the dimensionless (inverse)
vertical permeability is given by k′) H2/Kz. The outer radius
of influence of the upconing is denoted by ηouter. For single-
phase flow, it is well-known (see, e.g., ref 2) that at steady
state
( )
Rη
Q̃
ηouter ) cη (27)
v̂H2
where v̂ is the background flow rate (neglecting the effect of
the well), and the exponent Rη ) 1. For two-phase flow, we
obtain good results using Rη ) 2.
When inequality 26 is violated, two-phase flow occurs in
the leaky well, as brine is pulled up into the well and flows
with the CO2. We describe this with the approach of ref 17,
referring to the online appendix for the relevant equations.
2.3.3. Well Flow Model. Flow in abandoned wells is in
general a complex process, which may include open well
746 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 3, 2009
effects may be important and may be incorporated into our
modeling framework. However, in the applications presented
herein, we will not focus on the complexities of the well flow,
and ignore nonisothermal effects in the flow. We assume
that flow along wells occurs mainly through fractured or
degraded porous media along the outside of the well casing,
including well cements and drilling fluids, and that the small-
scale flow paths can be represented at the larger scale by an
effective Darcy permeability (see right-hand part of Figure
1).
2.4. Overall Solution Algorithm. The major components
of our overall solution algorithm have been presented in the
previous sections, including solutions for the space-time
evolution of both the CO2 plume and the pressure field, and
the determination of leakage conditions along a well through
the use of an upconing model. These form the basis of our
solution algorithm, which we outline as follows.
Let the state of the system be well defined at time tn (e.g.,
at the initial condition t0 all CO2 masses are zero and the
initial pressure distribution is hydrostatic in all aquifers). To
calculate the solution at time tn+1 ) tn + ∆t, we use the known
CO2 plume masses from time tn to calculate a new pressure
field at time tn+1. This pressure field is defined for all space,
using the Green’s functions and the flow rates at the individual
wells. This should be contrasted to traditional multiphase
simulators, which discretize the pressure field on a much
denser computational grid.
Mass conservation along segment l of well i is imposed
using
d
dt R,i
l-
M ) QR,i (t) - ∑ l+
QR,j (t) (28)
l (t)
j∈SR,i
This equation is written for cumulative masses MR,i corre-
sponding to plumes of the CO2 (R ) c) and brine (R ) b)
phases associated with well segment i. It states that the mass
added to a plume associated with segment i is equal to the
sum of the fluxes into the plume, minus the sum of fluxes
out of the plume.
Vertical flows along any well segment are described by
the one-dimensional multiphase version of Darcy’s law,
written in discrete form as
( )
kR pBl+ - pTl-
QRl ) -πrw2K w
l
FR + FRg (29)
µR Hl
Equation 29 is used to describe the flux through a well
perforating caprock (or aquitard) layer l. We denote the well
by a subscript w, and the permeable formations immediately
above and below this layer are denoted by l+ and l-,
respectively. Note that we distinguish between pressures at
the top and bottom of permeable formation through the use
of subscripts T and B, respectively.
As the wells are treated as 1D objects intersecting the
aquifers, their internal volume is considered negligible.
Therefore, saturation of each phase within the well segment
becomes a key parameter in the Darcy equation (eq 29) which
governs the value of the relative permeability kR and hence
controls the flow rate. Whenever a neighbor plume intersects
the well in the formation immediately below caprock section
l (that is, formation l- in eq 29), the upconing solution of
section 2.3.2 is used to determine flow rates for both fluids
in the well segment. Given the flow rates, the fluid saturation
can be determined using the relative permeability function
(which is assumed to be knownshere we used linear
functions, but nonlinear forms can also be handled).
For wells that are not intersected by neighboring plumes
(so there is no upconing), we use Scl ) (Scl-1)ν where Scl-1 is the
saturation in the segment below, and ν ≈ 0.5. This specifica-
tion was determined by comparison with saturation profiles
observed in detailed, fully numerical simulations of the
system.
The overall solution algorithm involves the following steps:
(1) solve the linear system of pressure equations, which yields
pressures at each of well segments in each permeable layer
in the system, (2) use these pressures, in conjunction with
the upconing equations, to determine fluxes and saturations
in each well segment, (3) update all plumes and the associated
cumulative masses, and (4) move on to the next time step.
This numerical algorithm is analogous to the implicit-
pressure-explicit-saturation (IMPES) algorithm sometimes
used in petroleum reservoir engineering.
Several additional assumptions are needed when carrying
out the actual calculations. First, when calculating saturations
within the domain, we deal with overlapping plumes by
simply taking the maximum thickness, FIGURE 2. Comparison of leakage into successive formations
h ) max hi (30) between the new model (blue) and an established numerical
i simulator (green). Aquifer numbering is from the bottom up and
is indicated by the patterns shown with black color in the
Second, when calculating the pressure solutions, the as- legend.
sumption of radial symmetry inherent in the solutions of
are modeled as 20 m thick, with an areal extent of 250 km2.
section 2.3.1 requires that all parameters in the equation
Boundary conditions are constant hydrostatic pressure. The
have radial symmetry. However, since CO2 and brine have
densities of brine and CO2 are set to constant throughout the
different viscosities, and there are multiple (locally radial)
domain, at typical values of 1000 and 600 kg/m2, respectively,
plumes throughout the domain, the pressure field will fail
with viscosities of 0.5 and 0.05 mPa · s. The distance between
to have global radial symmetry. We construct an approximate
the central well, which we take as the injection well, and the
radially symmetric field using the procedures outlined in ref
four nearest (abandoned) wells is 1000 m. All abanonded
20 for the two-well problem.
wells have identical properties, with permeabilities of 100
Third, note that the solution of ref 16 was derived for a
mD and radii of 20 cm. This problem therefore has an 8-fold
constant injection rate. This is clearly not the case for leaky
symmetry. Such a symmetrical setup allows for sufficient
wells (which act as injection wells in upper layers), where
grid resolution in the numerical simulator that will be used
their flow rates vary in time, increasing from zero to a near-
to validate our new approximation method.
constant value. We account for this in two ways: In eq 21,
We simulated 2500 days of injection at 50 kg/s and
the denominator of the key dimensionless group χ involves
compare results obtained by two methods: (i) the semiana-
the cumulative mass in the plume, and not simply the
lytical model described herein and (ii) the industry-standard
(constant) flow rate multiplied by time. This corresponds to
simulator Eclipse (23). For the numerical Eclipse simulation,
using the current rate dMc/dt at time t to determine an
we discretized a quarter of the domain (exploiting symmetry),
effective time t* that would be required, at that flow rate, to
with 37 6320 grid cells. On comparable computational
reach a total mass of Mc. This effective time t* is then used
resources, our model reduces the run-time by more than 3
in the similarity solution. Additionally, when calculating the
orders of magnitude, with similar savings in memory usage.
outer limit for the pressure perturbation, denoted by Ψ in
In practical terms, this replaces days of computing with less
eq 25, the correction to account for the variable total flow
than a minute on a standard desktop machine.
rates is given by the modified equation
In Figure 2, we show accumulated mass from all wells in
4.5πHφλbK(1 - Sres) |Vtot,i| each layer. Note the logarithmic scale on the y-axis. The
Ψi ) (31) aquifer numbering is from bottom up, starting with the
ceffMc,i/Fc |Q̃i| injection aquifer. For early times, there is a substantial
deviation between the models, due to CO2 arriving at the
Here Vtot,i represents the total fluid volume which has entered leaky wells significantly earlier in the numerical simulation.
(or left) the formation through the given well segment i. The We attribute this, at least in part, to numerical diffusion in
absolute value signs appear because we use the same Eclipse. At late times, there is excellent agreement between
approach for both positive and negative flows. the numerical simulation and our model.
3.2. Full-Scale Field Example. To demonstrate some of
3. Representative Results the capabilities of our new model for CO2 injection and
The approximation and solution strategies detailed in the leakage, we have applied the model to a data set associated
previous sections provide the basis for an efficient numerical with a field location close to Edmonton, Alberta, Canada. At
simulator. This section illustrates some of the capabilities of this location, which is shown in Figure 3, four large power
our implementation. plants collectively emit about 35 million tons of CO2 per
We present two examples in this section. The first example year. As such, this is a potential location for carbon capture
serves as a validation of the model by comparing it to the and storage. Within a domain that is 30 km × 30 km, 508
results of a traditional numerical model. The second example existing wells were identified, and the local layered structure
demonstrates the applicability to large field-scale calcula- of the subsurface was characterized (see http://www.ags.
tions, which is far beyond the capabilities of traditional gov.ab.ca/activities/wabamun/Wabamun_base.html). These
methods. data form the basis for a model system with seven permeable
3.1. Eight Well, Five Aquifer Example. Our first example layers separated by six impermeable layers. These layers were
is a representative case from a suite of validation experiments assigned nominal permeability values of 10 milli-Darcy (mD),
we have conducted. This case has five aquifers, penetrated which is characteristic of the Alberta Basin. The existing wells
by nine wells which form a three by three square lattice when were assumed to penetrate all of the layers in the model.
seen from above. We consider an isotropic permeability of While the statistical properties of effective permeability
10 mD in all formations, with a porosity of 10%. All formations values along well segments are currently unknown, we use

VOL. 43, NO. 3, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 747

FIGURE 3. Characteristics of the Wabamun Lake area, Alberta,
Canada: (a) location and major stationary CO2 sources and (b)
distribution of deep wells (oil and gas and injection wells).
Reprinted with permission from the work of Bachu and Celia
(1) 2008 AGU.
FIGURE 4. Leakage into successive formations from a hypothetical
injection at a location in Alberta. Aquifer numbering is from the
bottom up.
values herein as illustrations of the variability we expect to
see. Experimental methodologies for measuring the perme-
ability of abandoned wells and ongoing field campains are
discussed elsewhere (7). Thus, each well segment was
assigned an effective permeability value sampled randomly
from a log-normal probability distribution. Here, we assumed
that log10 k ∼ N(µ ) log10(10 mD), σ2 ) 1), which corresponds
to a most likely permeability value of 10 mD and a variance
of about an order of magnitude. In this example, we show
a single realization of parameters from this distribution; a
detailed investigation of the model behavior in response to
the input parameter will be presented in a forthcoming
publication. Injection was modeled over a characteristic time
period of 30 years at 500 kg/s. Boundary conditions are
constant hydrostatic pressure. Note that while we only model
a single injection well in this run, this may also be considered
as a proxy for a cluster of injection wells.
The accumulation of CO2 in the six permeable layers
above the injection layer, formed from a series of secondary
plumes, is shown in Figure 4. Similarly to the previous
results, we see a clear reduction in the amount of leakage
that occurs as additional layers are encountered moving
upward in the stratigraphic sequence. This shows the
importance of intervening permeable layers above the first
impermeable layer and their ability to mitigate leakage
rates.
The only layer where this pattern is not followed is the
upper-most permeable layer, where a no-flow boundary
above it leads to a somewhat artificial accumulation of mass.
In this sense, the upper layer in our model represents the
748 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 3, 2009
FIGURE 5. Leakage statistics in terms of plume masses for
successive formations from a hypothetical injection at a
location in Alberta. Aquifer numbering is from the bottom up,
and the units on the x-axis refer to the logarithm of mass in
kilograms. The left y-axis is for the probability density curves,
and the right axis refers to number of well segments with
plumes of mass equal to 1 kg or less.
total mass that may leave the geological system and should
be thought of as a proxy for the shallow groundwater and
atmosphere.
These results show the importance of including all layers
in an analysis of an injection scenario, with the concomi-
tant implications for site characterization and regulatory
requirements. We note that the rate of brine flow is also
captured in the model and shows a similar trend to CO2
leakage. However, due to lack of buoyancy, in some cases
we observe negative brine fluxes, associated with coun-
tercurrent flow of brine. A discussion of brine leakage
patterns is included in the online Supporting Information
for this publication.
Additional insights given by our simulator are illustrated
in Figure 5, which shows the distribution of secondary plume
sizes, sorted by layer. The largest plumes occur in the layer
immediately above the injection layer, with a distribution
clearly skewed toward higher values. Plumes become pro-
gressively smaller as we move upward in the stratigraphic
succession, and the distributions become more symmetric.
Again, the highest permeable layer is somewhat anomalous
in its number of plumes, due to the boundary condition at
the top of the domain. The atom of probability at 1 kg (labeled
100 in the figure) represents all smaller plumes, almost all of
which are insignificant. This information about plume sizes
and their overall spatial distribution can provide practical
guidance for activities such as the design of subsurface
monitoring and leakage detection strategies.
Supporting Information Available
Upconing equations used and the flow of brine predicted for
the final example. This material is available free of charge via
the Internet at http://pubs.acs.org.
Literature Cited
(1) Bachu, S. and Celia., M. A. Assessing the Potential for CO2
Leakage, Particularly through Wells, from CO2 Storage Sites.
The Science and Technology of Carbon Sequestration, AGU
Monograph; American Geophysical Union: Washington, DC,
2008, in press.
(2) Bear, J. Dynamics of Fluids in Porous Media; Elsevier: New York,
1972.
(3) Celia, M. A.; Bachu, S.; Nordbotten, J. M.; Kavetski, D.; Gasda.,
S. Modeling critical leakage pathways in a risk assessment
framework: Representation of abandoned wells. In Proceedings
of the 4th Annual Conference on Carbon Capture and Seques-
tration, May 2-5, 2005, Alexandria, VA.
 (4) Class, H.; Bielinski, A.; Helmig, R.; Kopp, A.; Ebigbo, A. Numerical
simulation of CO2 storage in geological formations. Chem. Eng.
Tech. 2006, 78 (4), 445–452.
(5) Doughty, C.; Pruess, K. Modeling supercritical CO2 injection in
heterogeneous porous media. Vadose Zone 2004, 3 (3), 837–
847.
(6) Gasda, S. E.; Bachu, S.; Celia, M. A. Spatial characterization of
the location of potentially leaky wells penetrating a deep saline
aquifer in a mature sedimentary basin. Environ. Geol. 2004, 46
(67), 707–720.
(7) Gasda, S. E.; Nordbotten, J. M.; Celia, M. A. Determining the
effective wellbore permeability from a field pressure test:
Numerical analysis of detection limits. Environ. Geol. 2008, 54
(6), 1207–1215.
(8) Gasda, S. E.; Celia, M. A.; Nordbotten, J. M. Upslope Plume
Migration and Implications for Geological CO2 Sequestration
in Deep, Saline Aquifers. IES J. A: Civil Struct. Eng. 2008, 1 (1),
2–16.
(9) Gasda, S.; Nordbotten, J. M.; Celia., M. A. Vertical Equilibrium
with Sub-scale Analytical Methods for Geological CO2 Seques-
tration. Comput. Geosci., submitted for publication.
(10) Juanes, R.; Spiteri, E. J.; Orr, F. M.; Blunt, M. J. Impact of relative
permeability hysteresis on geological CO2 storage. Water Resour.
Res. 2006, 42 (12), W12418.
(11) Hunt, B. Flow to a well in a multiaquifer system. Water Resour.
Res. 1985, 21 (11), 1637–1641.
(12) Metz, B., Davidson, O., de Coninck, H. C., Loos, M., Meyer, L.
A., Eds. IPCC Special Report on Carbon Capture and Storage;
Cambridge University Press: New York, 2005.
(13) The Future of CoalsOptions for a Carbon-constrained World;
Massachusetts Institute of Technology: Cambridge, MA, 2007.
(14) U.S. Greenhouse Gas Abatement Mapping Initiative. Reducing
U.S. Greenhouse Emissions: How Much and at What Cost?;
McKinsey and Company, 2007.
(15) Muskat, M. Physical Principles of Oil Production; McGraw-Hill:
New York, 1949.
(16) Nordbotten, J. M.; Celia, M. A. Similarity solutions for fluid
injection into confined aquifers. J. Fluid Mech. 2006, 561, 307–
327.
(17) Nordbotten, J. M.; Celia, M. A. An improved analytical solution
for interface upconing around a well. Water Resour. Res. 2006,
42, W08433.
(18) Nordbotten, J. M.; Celia, M. A.; Bachu, S. Analytical solutions
for leakage rates through abandoned wells. Water Resour. Res.
2004, 40 (4), W04204.
(19) Nordbotten, J. M.; Celia, M. A.; Bachu, S. Injection and storage
of CO2 in deep saline aquifers: Analytical solution for CO2 plume
evolution during injection. Transport Porous Media 2005, 58
(3), 339–360.
(20) Nordbotten, J. M.; Celia, M. A.; Bachu, S.; Dahle, H. K. Semi-
analytical solution for CO2 leakage through an abandoned well.
Environ. Sci. Technol. 2005, 39 (2), 602–611.
(21) Pacala, S.; Socolow, R. Stabilization Wedges: Solving the Climate
Problem for the next 50 Years with Current Technologies. Science
2004, 305, 968.
(22) Pruess, K. On CO2 fluid flow and heat transfer behavior in the
subsurface, following leakage from a geologic storage reservoir.
Environ. Geol. 2007, 54 (8), 1677.
(23) Schlumberger Information Systems. Eclipse technical description;
Report, Houston, TX, 2004.
(24) Stauffer, P. H.; Viswanathan, H. S.; Pawar, R. J.; Guthrie, G. D.
A system model for geologic sequestration of carbon dioxide.
2009, 43, 565-570.
ES801135V
VOL. 43, NO. 3, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 749