Multiscale modeling of flow and transport in fractured porous

media via the lattice Boltzmann method

Dongxiao Zhang
Mewbourne School of Petroleum and Geological Engineering, University of Oklahoma, Norman, USA

ABSTRACT: Our physical observations and theoretical treatments of flow and transport in fractured porous
media are usually associated with various length scales: pore- (microscopic), local- (macroscopic), fracture-,
and field (megascopic)-scales. Dominant processes and governing equations may vary with scales. Extending
from one scale to another requires upscaling (or downscaling). This allows the essence of physical processes at
one level to be attributed at a coarser (or finer) level. In this conference proceedings paper, we review some of
our recent work in modeling flow and transport in fractured porous media at coupled scales by the method of
lattice Boltzmann and in upscaling from the microscopic to the macroscopic scale.

1 INTRODUCTION In this study, we quantify such interactions with

Flow and solute transport in naturally fractured porous
media have been studied extensively due to their
importance in water resources management, contam-
inant migration in aquifers, oil/gas production, and
radioactive waste repositories [e.g., Bear et al., 1993].
Owing to statistically complex distribution of geolog-
ical heterogeneity and the multiple length and time
scales involved, three approaches are commonly used
in describing fluid flow and solute transport in natural
fractured porous formations: (1) discrete fracture mod-
els, (2) continuum models, including double porosity
models, using effective properties to represent the
effects of fractures, and (3) hybrid models that com-
bine the above two. In the discrete fracture approach,
the effect of porous matrix on flow and transport is
usually neglected (e.g., Smith and Schwartz, 1993;
Berkowitz and Scher, 1997). A recent work of Kang
et al. (2002a) on flow in a fractured porous system
reveals that this may not be a good assumption when
the matrix permeability is finite. In this case, the matrix
flow has significant effects on the fractured system and
the assumption that the matrix is impermeable does
not hold, and hence the use of the cubic law to calcu-
late the fracture permeability may cause a significant
error. When solute transport is considered in such a
fractured porous system, it is more important to take
into account the contribution of the porous matrix, and
it is essential to represent the interactions between the
factures and the matrix blocks in the continuum and
hybrid approaches (e.g., Warren and Root, 1963;Tsang
et al., 1996; Lee et al., 2001).
detailed, pore scale simulations of flow and solute
transport in porous media with a single fracture of
varying apertures with the lattice Boltzmann (LB)
method. The LB simulation can provide us with
detailed flow and solute transport quantities at vari-
ous scales and allow us to examine the continuous or
double-porosity representation of flow and transport
in fractured porous media.
2 MODEL AND THEORY
The LB method is a tool for simulating fluid flows
and modeling physics in fluids. Unlike conventional
numerical schemes based on discretizations of macro-
scopic continuum equations, the LB method is based
on microscopic description and mesoscopic kinetic
equations. This feature gives LB method the advan-
tage of studying non-equilibrium dynamics, especially
in the problem of acid stimulation involving inter-
facial dynamics and complex geometries. Since its
introduction as a tool, the LB method has been
proven to be competitive in studying a variety of flow
and transport phenomena (Chen and Doolen, 1998),
including single- or multi-phase flow (Gunstensen and
Rothman, 1993; Zhang et al., 2000) and chemical
dissolution/precipitation (Kang et al., 2002b, 2003)
in porous media. The LB approach is, at present, the
only existing computational method that is capable of
capturing the details of porous media geometry, and
all of the relevant physical and chemical processes,
thus providing necessary data at both the microscopic

443
Copyright 2005 Taylor & Francis Group plc, London, UK

Figure 1. CMT geometry of sandstone of porosity 16%.
and macroscopic scales. Shown in Figure 1 is one of
such rock geometries obtained by computed micro-
tomography at the resolution of microns, on which we
have performed detailed single and multi-phase LB
simulations (Zhang et al., 2000).
In the LB method, the primitive variable is the parti-
cle distribution function, which satisfies the following
evolution equation (e.g., Rothman and Zaleski, 1997;
Chen and Doolen, 1998):
where fi is the particle velocity distribution function
along the i direction, t is the time increment, is
the relaxation time relating to the kinetic viscosity by
eq
v = ( 0.5)RT , fi is the corresponding equilibrium
distribution function, and e
i s are the discrete veloci- bimodal heterogeneous porous medium generated by
ties. The density and velocity of the fluid are calculated
from the particle distribution function via
It is well known that, using the Chapman-Enskog
expansion, the LB equation (1), recovers the correct
continuity and momentum equations at the Navier-
Stokes level (e.g., Chen and Doolen, 1998),
where p = RT is the fluid pressure.
The transport of dissolved chemicals having low
concentrations may be described by another distribu-
tion function gi . This satisfies an evolution equation
444
Copyright 2005 Taylor & Francis Group plc, London, UK
similar to that of fi . It has been shown using the
Chapman-Enskog expansion technique that such a LB
transport scheme recovers the convection-diffusion
equation successfully.
The LB method can be modified to include sur-
face reactions for simulating dissolution/precipitation
in 2D porous media (Kang et al., 2002a, 2003). In
this approach, the surface reactions are accounted for
as boundary conditions to the LB transport equation.
The resulting evolution of pore geometries is coupled
with the LB flow model by updating the geometries at
each time step. Then, the new flow field is passed to the
transport equation.As such, the coupling of flow, trans-
port, and reactions is achieved, leading to the dynamic
evolution of pore geometries.
3 MULTISCALE MODELING OF FLOW
We have developed a unified microscopic/macroscopic
porous media method capable of simulating flow in
various-length-scale porous systems and in systems
where multiple-length scales coexist (Kang et al.,
2002a). The governing equations, omitted here, are
similar to (3)(4) with an additional term to account
for resistance in the porous matrix. Application of
this method to a unidirectional steady flow through
a homogeneous and a heterogeneous porous medium
recovered Darcys law when the effects of inertial
forces and the Brinkman correction can be neglected.
Simulations performed on a fractured porous system
indicated that as far as the overall permeability of the
system was concerned, the current method gave a very
good result and that this method is capable of handling
fractured systems with large length-scale spans.
Here we show an example of flow through a
using a two-stage procedure. The field is 16 m 16 m,
grid size is 81 81. The permeability field is com-
posed of two permeability populations: one with a
low mean permeability (the red material region),
and the other with a high mean value (the blue
region). The permeability also varies within each
region. Overall, the maximum value of the perme-
ability is 6.07 1013 m2 , and the minimum value
is 8.46 1017 m2 . The pressure (head) difference
between the entrance and the exit is 0.5 m. The den-
sity and viscosity of the fluid are 997.81 kg/m3 and
1.00246 103 kg/m/s. There is no analytical solu-
tion for flow through such a porous medium because
of the heterogeneity of the medium, even though the
flow is assumed to be steady and Darcys law is satis-
fied. To verify the validity of the method in simulating
flow through such a medium, we compared our simu-
lation results with those obtained by using the Finite-
Element Heat- and Mass-Transfer code (FEHM). In
both simulations, grid size is 81 81; pressure (head)
16
k 1
34.869
34.3579
30.8353 current method
12.3452
12 5.84217 FEHM
0.8
3.64783
1.22104
0.397418

(p-po)/(pi-po)
0.238475
0.237247 0.6
8 0.21838
0.181986
0.170097
0.158534
0.13355 0.4
0.121968
0.10232
4 0.0792314
0.0355692
0.0156639 0.2
0.0132229
0.0127849

0
0 4 8 12 16 0
0 4 8 12 16
x (m)
Figure 2. The permeability distribution and the pressure
contours obtained by current method (solid lines) and FEHM
(dashed lines). The field is 16 m 16 m. The pressure (head) Figure 3. The dimensionless pressure profile at center line
is 0.5 m at entrance, and 10 m at exit. The permeability has the (y = 8 m) obtained by the current method (solid line) and
unit m2 , and its values are multiplied by 1014 . The Reynolds FEHM (dashed line). pi is the pressure at the entrance, and
number in the current method is 5.96 103 . po is the pressure at the exit.

is specified at entrance and exit; and velocity along

the normal direction of the other two boundaries is set
to zero.
Figure 2 shows the permeability distribution and
the contours of pressure (head) obtained by these two
methods. The permeability values are multiplied by
1014 . The solid lines are the results obtained with
the current method, and the dashed lines are results
obtained with FEHM.The Reynolds number in the cur-
rent method is Re = 5.96 103 . Apparently, at this
Reynolds number, the pressure contours of these two
methods agree with each other very well, and they
are both mainly distributed in two regions where the
permeability values are relatively small. This is cor-
rect since in such regions, the resistance to the flow is
larger and, accordingly, greater pressure drop is needed
to achieve the overall flow rate; hence the pressure
contours are denser than elsewhere. Figure 3 shows
the dimensionless pressure profile along the horizon-
tal center of the domain (at y = 8). The results of these Figure 4. A simple fractured porous medium.
two methods have good agreement.
Figure 4 shows a simple fractured system where a of the porous matrix varies from 0.04 to 0.601, the
fracture is embedded in a porous medium of poros- pore width ranges from 0.0625 to 0.5625 mm, and the
ity . In this particular case, the fracture is of width matrix permeability is one to four orders of magnitude
h = 32 (in lattice unit, lu) compared to the total width less than the corresponding fracture permeabilities.
of 600 lu for the whole system, and the matrix poros- The flow in the system where the fracture and the pore
ity is = 0.601. To explore the effects of the matrix scales coexist can be simulated with two approaches.
porosity (and thus the pore diameter and matrix One is to simulate the system with the conventional
permeability) as well as the fracture width on solute LB model which essentially solves the Navier-Stokes
transport, these parameters and the size of the fractured equation, and the other is to use the unified multiscale
porous system are varied. In the first group of simula- LB model of Kang et al. (2002a) where the fracture is
tions, the fracture aperture is h = 32 lu, corresponding recognized as pore space and the porous matrix is aver-
to 1 mm in linear dimension; in the second group, aged and accounted for with a resistance term in the
the aperture is h = 52 lu or 1.625 mm. The porosity same equation. Good agreements have been founded

445
Copyright 2005 Taylor & Francis Group plc, London, UK

Figure 5. Snapshots of concentration fields for three scenar-
ios of matrix porosity () at two dimensionless times, which
are defined as the plume residence time normalized by the
characteristic time tc = h2 /Dm , where h is the channel width
and Dm is the molecular diffusivity. The Peclet number is
51.35.
between these two approaches although the latter is
more efficient computationally.
4 UPSCALING OF TRANSPORT
Figure 5 shows snapshots of the concentration field
at two dimensionless times for three scenarios (with
matrix porosity = 0, 0.308, and 0.601 and fac-
ture aperture of 1 mm). Time is made dimensionless
with characteristic time tc , defined as tc = h2 /Dm ,
where h is the channel (fracture) width and Dm is the
molecular diffusivity. The Peclet number (defined as
Pe = Uh/Dm ) is 51.35. When the pore matrix becomes
permeable, the plume transfers some of its mass into
= dMf /dt
the matrix as it migrates along the facture and some of
the mass in the matrix transfers back into the fracture
after the main plume passes, creating a long tail. It is
seen that both the mass transfer and plume tailing in
the facture are enhanced as increases. As such, the
overall dispersion of a plume increases as the matrix
becomes more permeable. The spatial moments of the
plume can be analyzed on the basis of these detailed
concentration fields.
Alternatively, the fractured porous media may
be conceptualized with dual-porosity and/or double-
permeability models (e.g., Warren and Root, 1963;
Gerke and van Genuchten, 1993). In such models, the
mass transfer between the facture and matrix systems
is commonly described by the following first-order
model:
where cf and cm are the solute concentration in the
facture and the matrix, respectively, and is the mass
transfer coefficient. In the double-permeability model,
an advection-induced component may be included in
the mass transfer expression. However, in our simu-
lations there is no lateral flow component owing to
the specific boundary conditions. Assuming to be
446
Copyright 2005 Taylor & Francis Group plc, London, UK
10
*
4
2 =0.601,h=32
1
=0.453,h=52
=0.308,h=32
=0.191,h=52
=0.069,h=32
0
0 0.05 0.1 0.15
t*
Figure 6. Time evolution of the mass transfer coefficient
at Pe = 51.35. The coefficient and time are normalized with
respect to tc .
uniform spatially and integrating the relationship over
space leads to
where Mf and Mm are the respective total solute mass
in the facture and the matrix systems and Mf
is the rate of change of mass in the facture.
Figure 6 shows the time evolution of the dimension-
less mass transfer coefficient = tc for different
matrix porosity values and facture apertures. It is seen
that with time the coefficient goes to a constant for
all matrix porosity values under consideration. The
asymptotic mass transfer coefficient increases as the
matrix porosity. The time taken for the coefficient to
reach a constant also increases with the porosity. The
mass transfer between the fracture and the matrix is
usually non-negligible even with porous matrix of low
porosity (e.g., 4% and 6.9%) and thus low permeabil-
ity, in which the contribution of the matrix flow to the
fractured system may be safely neglected (Kang et al.,
2002a).
Figure 7 plots the asymptotic mass transfer coef-
ficients as a function of (h/dg )2 for two fracture
apertures h, where dg is the average grain diameter
(the width of the grain, in our case). It is seen that for
h/dg > 1, the mass transfer coefficient is more or
less a linear function of (h/dg )2 . Our simulations do
not cover the region of h/dg < 1, which is not neces-
sarily linear. When the fracture is wide compared to
the average grain size, one has approximately = c
(h/dg )2 where c is a dimensionless constant, similar to
the formation factor, which depends on porosity, tor-
tuosity, and other factors. It is seen from Figure 7 that
 2 media is non-Gaussian with long tails. The long tailing
1.6
1.2
*
h = 32
0.8 h = 52
0.4
0
1 2 3 4
h2/dg2
Figure 7. Dependency of dimensionless mass transfer coef-
ficient on h2 /dg2 .
the slope c is more or less independent of the fracture
aperture h. In dimensional terms, one has
where cDm may be regarded as the matrix diffusivity,
defined as the molecular diffusivity Dm in free water
corrected by the formation factor c. It is thus seen
that when the fracture aperture is sufficiently large, the
mass transfer coefficient is proportional to the matrix
diffusivity and inversely proportional to the square
of the average grain size of the porous matrix but is
independent of the fracture aperture. Equation (7) is
consistent with the finding of Haggerty and Gorelick
(1995) for diffusion into layers.
5 CONCLUSIONS
The lattice Boltzmann method is not only suitable for
simulating flow in detailed pore geometries but also
for flow in various-length-scale porous systems and in
systems where multiple-length scales coexist. Detailed
lattice Boltzmann simulations of flow and solute trans-
port were performed on simply constructed porous
media embedded with a fracture. The effects of the
porosity (and thus permeability) of the porous matrix
and the fracture aperture on transport were investi-
gated on the basis of these simulations. Both the flow
quantities and the mass transfer coefficients between
the facture and the porous matrix were analyzed. It
has been found that unlike transport in channels or in
pure porous media, solute transport in fractured porous
447
Copyright 2005 Taylor & Francis Group plc, London, UK
stems from mass transfer between the fracture and the
porous matrix and from the contrast in flow velocities
in the two media. It was shown that when the frac-
ture aperture is sufficiently large, the mass transfer
coefficient is proportional to the matrix diffusivity and
inversely proportional to the square of the grain size
of the porous matrix. For such cases, the mass trans-
fer coefficient is independent of the fracture aperture.
Even for low porosity and low permeability matrix
whose contribution to flow in the fractured porous
media can be neglected, mass transfer between the
fractures and the matrix is usually non-negligible.
REFERENCES
Bear, J., Tsang, C.F. & de Marsily, G. 1993. Flow and
Contaminant Transport in Fractured Rock, Academic,
San Diego, Calif.
Berkowitz, B. & Scher, H. 1997. Anomalous transport in
random fracture networks, Phys. Rev. Let., 79(20), 4038.
Chen, S. & Doolen, G. 1998. Lattice Boltzmann Method for
Fluid Flows, Annu. Rev. Fluid Mech., 30, 329.
Gerke, H.H. & van Genuchten, M.T. 1993. A dual-porosity
model for simulating the preferential movement of water
and solutes in structured porous media,Water Resour. Res.,
29, 305.
Haggerty, R.H. & Gorelick, S.M. 1995. Multiple-rate mass
transfer for modeling diffusion and surface reactions in
media with pore-scale heterogeneity, Water Resour. Res.,
31, 2383.
Kang, Q., Zhang, D. & Chen, S. 2002a. Unified lattice
Boltzmann method for flow in multiscale porous media,
Phys. Rev. E., 66, 056307.
Kang, Q., Zhang, D., Chen, S. & He, X. 2002b. Lattice
Boltzmann simulation of chemical dissolution in porous
media, Phys. Rev. E, 036318.
Kang, Q., Zhang, D. & Chen, S. 2003. Simulation of Disso-
lution and Precipitation in Porous Media, J. Geophysical
Research Solid Earth, 108(B10), 2505.
Lee, S.H., Lough, M.F. & Jensen, C.L. 2001. Hierarchical
modeling of flow in naturally fractured formations with
multiple length scales, Water Resour. Res., 37, 443.
Smith, L. & Schwartz, F.W. 1993. Solute transport through
fracture networks, in Flow and Contaminant Transport
in Fractured Rock, edited by J. Bear, C.F. Tsang, and
G. de Marsily, pp. 129167, Academic, San Diego, Calif.
Tsang, Y.W., Tsang, C.F., Hale, F.V. & Dverstorp, B. 1996.
Tracer transport in a stochastic continuum model of
fractured media, Water Resour. Res., 32, 3077.
Warren, J.E. & Root, P.J. 1963. The behavior of naturally
fractured reservoirs, Soc. Petrol. Eng. J., 3, 245.
Zhang, D. & Kang, Q. 2004. Pore scale simulation of solute
transport in fractured porous media, Geo phys. Res. Lett.
31, L12504.
Zhang, D., Zhang, R., Chen, S. & Soll, W.E. 2000. Pore scale
study of flow in porous media: Scale dependency, REV,
and statistical REV, Geo phys. Res. Lett. 27, 1195.