3-D Pore-Scale Modelling of Sandstones and

Flow Simulations in the Pore Networks

Stig Bakke and Pl-Eric ren

A new method for generating realistic homogenous and heterogeneous 3-D pore-scale sandstone
models is presented. The essence of our method is to build sandstone models which are analogs of
actual sandstones by numerically modelling the results of the main sandstone-forming geological
processes - sandgrain sedimentation, compaction, and diagenesis. The input data for the modelling
are obtained from image analyses of thin section images of the actual sandstone.
The spatial continuity of the sandstone model in the X, Y, and Z directions is determined using a scale-independent invasion percolation based algorithm. The resulting spatial continuity
function, which is an ellipsoid, may be used as a heterogeneity descriptor for the sandstone model.
Heterogeneity analyses show that compaction reduces the spatial continuity in the horizontal direction more rapidly than in the vertical one.
The architecture and geometry of the network representation of the pore space are determined
by applying various 3-D image analysis algorithms directly on the fully characterised sandstone
model. A 3-D pore network which was generated from thin section data from a strongly water wet
Bentheimer sandstone is used as input to a two-phase network flow simulator. Simulated transport
properties for the sandstone model are in good agreement with those determined experimentally.

Introduction
The existence of a highly permeable void or pore space distinguishes reservoir rocks from other rocks. The pore space
of reservoir rocks is highly chaotic, consisting of a spatial
network of pores in which larger pores (pore bodies) are
connected via narrower pores (pore throats). The architecture and geometry of the pore network and its complementary grain matrix determine several macroscopic properties
of the rock such as absolute permeability, relative permeability, capillary pressure, formation factor, and resistivity
index1.
The problem of predicting macroscopic rock properties
from the underlying microscopic structure and pore-scale
physics has been the subject of extensive investigation. One
commonly applied tool in this investigation has been the use
of network flow simulators (network models). This approach
requires a detailed understanding of the physical processes
occurring on the pore-scale and a complete description of the
morphology of the pore space. The procedure has been applied, with success, to two-phase flow in simple or idealised
porous media2-6 using pore-scale physics identified in
Copyright 1997, Society of Petroleum Engineers, Inc.
Paper first presented at the European 3-D reservoir Modelling Conference held in
Stavanger, Norway, 16-17 April, 1996.
Received for review, September 16, 1996
Accepted for publication, January 23, 1997
Camera-ready copy received, April 2, 1997

SPE 35479
136

micromodels7-9. Recently, much of the pore-scale physics

for three-phase flow have been unveiled10-16, leading to the
development of network models for three-phase flow17-19.
The extension of these techniques to real porous media has
been complicated by the difficulty of adequately describing
the complex 3-D pore structure of real porous rocks.
Information about the pore structure of reservoir rocks are
usually obtained from mercury injection data and image
analysis of thin section images. Excellent overviews of
methods for characterising pore structures are given by
Yanuka et al.20 and Wardlaw1. Mercury injection data provide statistical information about the pore throat size distribution, or more correctly, the distribution of volumes which
may be invaded within specified pore throat sizes. Traditional analyses of pore geometries by image analysis of thin
sections are normally based on Delesse's principle21 which
states that 2-D observations in thin sections are representative for the 3-D sample. This is, however, only valid for scalars, i.e. for areal and volumetrical considerations. The principle is thus valid for porosity and mineralogical measurements, but not for geometrical measurements of pore bodies
and throats since these are vectors.
In order to characterise the geometry of a pore body or
throat from thin section images, it is necessary to define
which of the 2-D porosity elements (porels22) are pore bodies and which are pore throats. Furthermore, the flow direction in each porel must be known because the geometry of a
pore throat has to be measured in a plane normal to the flow
direction. At present, there exists no known techniques for
directly extracting this information from thin sections. Data
obtained from traditional porel structure analysis are thus
SPE Journal, Volume 2, June 1997

S. BAKKE AND P.E. REN

insufficient to directly provide a complete 3-D description of

the complex pore network of real rocks.
Advanced techniques such as micro-CT23,24 and serial
sectioning combined with BSE imaging of the sections25,26
provide a detailed description of the 3-D pore structure of
rocks. These techniques are, however, expensive and not
readily available. Bryant et al.27 constructed 3-D pore networks by simulating the results of rock-forming processes
using a completely defined packing of equal spheres constructed by Finney28,29. The method of reconstructed porous
media30 has also been used to characterise 3-D pore structures. The modelling of porous media has also been the subject of much investigation within the field of chemical engineering31,32.
The present paper presents a new method for generating
realistic and fully characterised 3-D pore networks. The
method is based on numerical modelling of the results of the
main sandstone-forming geological processes. The network
representation of the pore space of the modelled sandstone is
extracted from its complementary mineral matrix network
using image analysis techniques. Fluid flow simulations are
performed directly on the irregular pore network and computed transport properties are compared with available experimental data.

sented by a known mathematical function, grain sizes are

randomly assigned according to the given distribution. Otherwise, all the grain sizes recorded from the thin section
analysis are numbered and ordered according to size. Next, a
random real number between zero and the total number of
recorded grains is picked and the corresponding grain size is
determined by interpolating between the discrete curve
points (Fig. 1). There is no size-limitations for the grain-size
distribution and rocks ranging from siltstones to conglomerates may be modelled.

3-D Process/Stochastic Modelling of Sandstones

Sandstones are the end products of a series of complex geological and hydrodynamical processes which, schematically
described, starts with erosion of quartz-bearing rocks, followed by sandgrain transport via air, water or ice, and, finally, deposition of grains in sedimentary basins. The deposited grains may be reworked by one or several cycles of erosion, transport, and sedimentation. The sandstone generation
process is completed with compaction and different diagenetic processes.
A sandstone sample and its petrographical parameters is a
result of all the geological and hydrodynamical processes
which have affected the sedimentary basin and its environment. We do not attempt to model the detailed dynamics of
these extremely complex processes. Instead we attempt to
reconstruct the 3-D sandstone by simulating the results of the
main sandstone-forming processes using petrographical data
obtained from thin section analyses.
The sandstone reconstruction process consists of three
main modelling steps; (i) sedimentation (sandgrain deposition), (ii) compaction, and (iii) diagenesis.
(i) Sedimentation. The modelling of the sedimentation
process commences with the generation of a grain-size distribution curve. The grain-size distribution is obtained from
BSE images of thin sections of the actual sandstone. Image
analysis techniques are used to measure the diameters of all
the sandgrains in an image (Fig. 1).
At present, all the detrital sandgrains are treated as
spherical quartz grains. Selection of individual grains from a
given grain-size distribution is done randomly in order to
mimic the stochastic nature of the complex geological and
hydrodynamical processes which affect the deposition of
individual sandgrains in a sandbed. The manner in which
random grains are selected depends on the shape of the
grain-size distribution curve. If the curve is closely repre-

SPE Journal, Volume 2, June 1997

Fig. 1Example illustrating the process of randomly

selecting grain sizes from a numbered and ordered list
of measured grain sizes.

777777
r
777777
r
222222222222
222222222
777777
r
222222222222
22222222
222222222
777777
r
222222222222
2222222
22222222
777777
r 222222222
222222222222
22222222
2222222
22222222
222222222
222222222222
22222222
2222222
22222222
222222222
222222222222
22222222
2222222
22222222
222222222
222222222222
22222222
2222222
22222222
222222222
2222222222222
222222222222
22222222
2222222
22222222
222222222
2222222222222
2222222222
222222222222
22222222
2222222
22222222
2222222222222
2222222222
2222222
222222222222
22222222
2222222222
2222222222222
2222222222
2222222
222222222222
22222222
2222222222
2222222222222
2222222222
2222222
22222
222222222222
2222222222
2222222222222
2222222222
2222222
22222
2222222222
2222222222222
2222222222
2222222
22222
2222222222
2222222222222
2222222222
2222222
22222
2222222222
2222222222222
2222222222
2222222
22222
2222222222
2222222222222
22222222222222222
22222
2222222222222
2222222222 2222222222
2222222222
2222222222222
r

Fig. 2Two-dimensional schematic illustrating modelling of sedimentation (low-energy sedimentation). A new

sandgrain with radius r settles on the grainbed. The new
grain is reduced to a point and the radii of the grains in
the grainbed are increased by r. After the new grain
(point) is deposited in a local stable position, all the
grain radii are reset to their original values.

The dynamics and geometry which govern the deposition

of individual sandgrains onto an existing grainbed are fairly
complex and tedious to describe numerically. To simplify
this problem, the radius of each new randomly selected grain
is reduced to zero prior to deposition. The radii of all the
grains in the grainbed are increased by an amount equal to
137

3-D PORE-SCALE MODELLING OF SANDSTONES AND FLOW SIMULATIONS IN THE PORE NETWORKS

the radius of the new grain. This reduces each new grain to a
point and the grainbed to a solid surface (Fig. 2) which,
geometrically and dynamically, are easier to describe. After
the new grain-point is deposited in a stable position, all
grain radii are reset to their original values before the procedure is repeated for the next sandgrain. This procedure continues until a predefined basin subset (bounding box) is filled
with grains. All the sandgrain centre co-ordinates and radii
are recorded and used for sphere rendering (Figs. 3-7).

Fig. 3The low-energy (local minimum) sedimentation

results in an open grainbed with porosity = 37.8%.

The exact location in the grainbed where each new sandgrain is deposited depends on whether the sedimentation
process occurs in a low or high energy environment. For
low-energy sedimentation, it is assumed that the horizontal
velocity of the new sandgrain is zero and that no lateral
forces affect the grainbed. In this case, each new randomly
selected sandgrain (reduced to a point) is dropped onto the
(expanded) grainbed from a random X,Y position. Upon
hitting the (expanded) grainbed, it rolls down the maximum
local gradient until it hits a stable position in a local minimum (Figs. 2 and 3). This leads to an undulating grainbed
surface growth with local peaks and valleys.
Sedimentation of sandgrains in extreme low-energy basins is, however, rare. Sandstones are normally the results of
high-energy depositional events (low-energy deposition results in claystones and siltstones). Sandgrains precipitate
from water or air suspension when the lateral velocity decreases below a threshold value which is proportional to the
grain size33. The sandbed is usually influenced by lateral
forces like streams and waves, and gravity in the case of
dipping beds. These lateral forces move individual sandgrains around until they settle in a stable position where the
effect of the transporting forces is at a minimum.
High-energy sedimentation is modelled by placing each
new randomly selected sandgrain at the lowest available
position (global minimum) in the grainbed (Fig. 4). In contrast to low-energy sedimentation, high-energy sedimentation
138

leads to an almost flat growth of the grainbed surface. The

porosity values for the grainbeds depicted in Figs. 3 and 4
show that high-energy (global minimum) sedimentation results in a denser grainbed than low-energy (local minimum)
sedimentation. In general, modelling of fine sand or silt
deposition is carried out as low-energy sedimentation
whereas sand and coarse sand deposition is modelled as
high-energy sedimentation.

Fig. 4The high-energy (global minimum) sedimentation

with exactly the same input parameters as in Fig. 3 results in a denser grainbed with porosity = 35.3%.

Fig. 5Laminated grainbed model.

The modelling described above produces homogenous or

massive sandstone models. Natural sandstones often show
some degree of sedimentological heterogeneity on different
SPE Journal, Volume2, June 1997

S. BAKKE AND P.E. REN

scales. On the m to mm scale, the pore space itself is always

heterogeneous. On the mm to cm scale, sandstones may be
heterogeneous due to laminations, graded bedding, or to
diagenetic processes.
Laminated sandstones are modelled by placing as many
as three laminas with different and independent grain-size
distributions on top of each other (Fig. 5). Graded bedded
sandstones are modelled as upward fining or upward coarsening beds, reflecting the gradual variation in grainsize with
depth (Fig. 6).

where Z is the new Z-co-ordinate, Z0 is the old one, and is

a compaction factor which varies between zero and one. Fig.
7 and Table 1 shows the effects of compaction on porosity
for different compaction factors.

Fig. 7Effect of linear compaction modelling for compaction factors equal to 0.0, 0.1, 0.2 and 0.3.

Fig. 6Upward coarsening grainbed model.

(ii) Compaction. Compaction, or bulk volume reduction, in

response to vertical stresses from the overburden is an important agent for porosity reduction in quartz sandstones
undergoing burial. At shallow depths, compaction is the result of mechanical processes such as grain rearrangements
and fracturing. At greater depths, grain interpenetration as a
result of pressure dissolution in the presence of K2O-bearing
minerals34 is the dominant mechanism. Compaction may be
responsible for local changes or heterogeneities in the pore
space even in extremely homogenous sandstones. In serial
sectioning models25, and even in thin sections, it is sometimes possible to observe how stable cell structures of quartz
grains have resisted the compaction forces and preserved
open grain structures inside these cells whilst the grain
structure outside the stable cells may have suffered significant compaction.
The results of compaction are modelled as a linear process similar to that described by Bryant et al.27. The Z-coordinate of every sandgrain centre is shifted vertically
downwards by an amount proportional to the original Zposition
Z = Z0(1 - )

SPE Journal, Volume 2, June 1997

(1)

(iii) Diagenesis. Diagenetic processes sometimes introduce

heterogeneities which totally overprint the heterogeneity
pattern of the primary structures of the sandstone. Diagenetic
processes may be very complex, involving several phases of
dissolution of minerals and cement overgrowth, often in interaction with compaction, as a result of changing geological
conditions. Currently, only a simplified subset of known
diagenetic processes are modelled: quartz cement overgrowth and subsequent clay coating of the free surface.
Quartz cement overgrowth is simulated by increasing uniformly the radii of all the sandgrains in the sandbed27 (Figs.
8-10). The amount of quartz cement overgrowth may be
equal to the volume loss from the grain interpenetration during compaction (obeying mass conservation) or it may be
specified independently by a cement thickness input parameter. The latter mimics the migration of SiO2 to or from the
sandstone model.
Clay coating is precipitated randomly on the free quartz
grain, quartz cement, or existing clay surfaces (Figs. 8-11).
The amount of clay to be precipitated is extracted from petrographical analysis. In thin sections, clay minerals are
sometimes observed to be concentrated in a few porels whilst
the rest of the porels are almost clay free. This type of clay
coating is implemented via a clustering routine which increases the likelihood of new clay voxels (3-D array elements or gridcells) to be precipitated on already existing clay
voxels. Currently, we do not distinguish between different
types of clay minerals. The modelled clay is assumed to behave as a mixture of mainly kaolinite and some illite.

139

3-D PORE-SCALE MODELLING OF SANDSTONES AND FLOW SIMULATIONS IN THE PORE NETWORKS

Boundary Conditions. The modelling of the sedimentation

and compaction processes is object-based and is followed by
a 3-D gridding of the model. The diagenesis modelling is
performed on this grid-based model. A 3-D array of e.g.,
100x100x110 voxels defines the spatial limitations or
bounding box for the centres of the grains in the sandbed.
The enlargement of the bounding box in the Z-direction is
due to extra space requirements when modelling compaction.
In nature, of course, such a near cubic grainpack would not
be stable since the vertical angles exceed the natural angle of
repose for grain packs. This is, however, not a problem,
since we are only modelling a small part of a continuous
sandstone.
To reduce possible boundary effects when characterising
the sandstone models, only the core of the sandstone model
is used for further processing and analysis. If, for example,
the working array consists of 100x100x110 voxels, the centre 70x70x70 voxels define the region of interest.
Verification of the Model. The sedimentation modelling is
the most CPU-intensive part of the modelling process. Typical CPU-times for a 285x285x315 grid with an average
grainsize of 30 grid-units is 3 hours for an object-based and
parallelised version of the software running on 4 CPUs on a
SGI Challenge R10000 computer. The compaction and diagenesis parts of the modelling process are run in a separate
loop, typically for about 5 minutes for the model described
above.

Fig. 8Isosurface representation of a sandstone model

generated from the grainbed model shown in Fig. 4.
Quartz is grey, quartz cement is light grey and clay minerals are dark grey.

The sandstone model is verified by comparing the porosity of the model with the porosity of the thin section or of the
plug, and by comparing visually arbitrary 2-D cross sections
through the model (Fig. 12) with the BSE image of the
sandstone sample. The compaction and diagenesis parame140

ters are tuned until the deviations in porosity values and the
visual difference between the 2-D cross sections and the thin
section are at a minimum.

Fig. 9Isosurface chair-mode representation of the

model shown in Fig. 8. Quartz is grey, quartz cement is
light grey and clay minerals are dark grey.

Fig. 10Isosurface chair-mode representation of the

mineral matrix shown in Figs. 8 and 9, rendered together
with its complementary pore space. The pore space is
medium grey, quartz is grey, quartz cement is light grey
and clay minerals are dark grey.

SPE Journal, Volume2, June 1997

S. BAKKE AND P.E. REN

Visualisation of the Model. Scientific visualisation is an

extremely useful tool for examining complex 3-D data and
models. In the present work, Iris Explorer was used to visualise the generated sandstone model and its pore space. Fig. 8
shows an isosurface representation of a sandstone model
which was generated from the simulated grainbed shown in
Fig. 4. The different grey shades represent different mineralogy and cement.
With the present visualisation tool, it is possible to remove parts of the model and study its inside (chair-mode).
Examples of this are illustrated in Figs. 9 and 10. By binarising and inverting the sandstone model, the complementary
pore space may be visualised together with the sandstone
model (Fig. 10). The complete pore space of the sandstone
model is shown in Fig. 11.

Fig. 11Isosurface representation of the pore space of

the model shown in Figs. 8 - 10.

Fig. 12Randomly oriented 2-D cross section through

the sandstone model.

Both orthogonal and randomly oriented 2-D slices

(simulated thin sections) through the 3-D model may be
extracted and visualised (Fig. 12). Simulated thin sections
from the sandstone model may thus be compared with those
SPE Journal, Volume 2, June 1997

from the real sandstone which it is assumed to mimic. This

makes it possible to verify that the simulated thin sections
have the same statistical properties (i.e. petrographical, mineralogical, and geometrical) as the thin sections from the real
rock.
Characterisation
Spatial Continuity. The spatial continuity of the pore space
in the X, Y, and Z directions strongly influences fluid flow in
rocks on all scales. In the present work, the narrowest constriction along the main flow path in a given direction is used
as a measure of spatial continuity. To determine quantitatively the spatial continuity, it is necessary to have a geometrical description of the pore space.
Erosional Geometrical Analysis. Erosion is a well known
2-D image analysis function which shrinks the size of objects
in a binary image according to different pixel masks or
neighbour connectedness35. The erosion function has been
extended to 3-D using the concept of 3-D neighbour connectedness36. Fig. 13 shows that the 3-D erosion technique
provides visual information about the size of the pore bodies
and throats and about the spatial continuity of the pore network. By counting the number of erosion loops which is necessary to erode a specific voxel, quantitative pore size information may be obtained.
The erosional geometrical analysis is carried out as an ultimate 3-D quantitative erosion. Fig. 14 shows a 2-D representation of the results from an erosional geometrical analysis. This erosion technique provides quantitative information
about pore body and throat sizes, but no direct information
about the network architecture or the main flow path through
the pore space.
Invasion Percolation Simulation. The main flow path
through the pore space in a given direction is found by
simulating an invasion percolation type process37 on the
output from the erosional geometrical analysis of the 3-D
pore space model. The invading fluid is injected from one
side plane and percolates through the model according to the
rules of invasion percolation i.e., at every step the highest
numbered pore space voxel connected to the front is invaded. The progress of a simulated invasion process is
shown in Fig. 15.
When the invading fluid reaches the opposite side plane
(i.e. breakthrough), the invaded volume shows the optimal
flow path in the given direction. The narrowest constriction
(lowest voxel value, smallest pore) which the fluid has
invaded is identified. This is the constriction which determines the drainage capillary pressure which is required to
establish a continuos flow through the pore space in the
given direction. The size of the narrowest constriction is used
as a measure of the spatial continuity in the given flow direction. The spatial continuity of the pore space in the other
two main directions is determined in an identical manner.
The resulting spatial continuity function may be represented
as an ellipsoid and it is used as a heterogeneity descriptor for
the pore space (Table 1).
The effect of increasing compaction on the spatial continuity of the grainbed model shown in Fig. 7 is summarised in
Table 1. The grain radii of the grainbed model range from
10 to 20 units with an exponential size distribution. The cement thickness is 1 unit.

141

3-D PORE-SCALE MODELLING OF SANDSTONES AND FLOW SIMULATIONS IN THE PORE NETWORKS

Fig. 13The results of 5 erosion loops on a Bentheimer

sandstone model. The largest pore has a radius of about
5.5 units, as seen in the lower right image (marked by an
arrow). The spatial continuity of the pore space in the
model is zero after 2 erosions.
0000000000000000000000000000000000000000
1111000000000000000000000000000000000000
1111111100000000000000000000000000001111
1111111111000000000000000000000011111111
1111111111111110000000000111111111111111
1111111111111111100001111111111111111111
1111111111111111111111111111111111111111
1111111111111111111111111111111111111111
1111111111111111111111111111111111111111
1111111111111111111111111111111111111111
1111111111111111110001111111111111111111
1111111111111111000000001111111111111111
1111111111110000000000000001111111111111
1111111111000000000000000000001111111111
1111111100000000000000000000000011111111
1111110000000000000000000000000000001111
1110000000000000000000000000000000000011
1100000000000000000000000000000000000000
0000000000000000000000000000000000000000

0000000000000000000000000000000000000000
1111000000000000000000000000000000000000
2222111100000000000000000000000000001111
3333222211000000000000000000000011112222
4444333322111110000000000111111122223333
5555444433222221100001111222222233334444
6666555544333332211112222333333344445555
7777666655444443322223333444444455556666
8888777766554444332223334445555566667777
9987776655443333221112223334445566667788
8876665544332222110001112223334455556677
7765554433221111000000001112223344445566
6654443322110000000000000001112233334455
5543332211000000000000000000001122223344
4432221100000000000000000000000011112233
3321110000000000000000000000000000001122
2210000000000000000000000000000000000011
1100000000000000000000000000000000000000
0000000000000000000000000000000000000000

Fig. 142-D quantitative erosion of a binary image of a

2-D section through an idealised pore structure. In the
left image, zeros are matrix pixels and 1s are pore pixels. The right image shows the number of erosion loops
necessary to erode each pixel. The maximum number of
erosion loops is 9. This means that the radius of the
largest pore body to the left is about 9 units. The radius
of the pore throat is 2 units.

142

Fig. 15Six selected images from a 60-step invasion

percolation simulation (solid) on an ultimately eroded
pore space model (transparent). Picture 4 shows the
situation at breakthrough, i.e., the optimal flow path in
the actual direction.
TABLE 1COMPACTION EFFECT ON POROSITY AND ANISOTROPY
Comp.

Porosity

0.00
0,05
0.10
0.15
0.20
0.25
0.30
0.35
0.40

31.16%
28.52%
26.10%
23.22%
20.82%
18.20%
15.74%
12.94%
10.50%

Spatial Continuity
X
Y
Z
4
6
6
4
5
5
4
4
5
3
3
5
3
3
5
2
3
4
2
2
3
1
1
2
1
1
2

Anisotropy
2Z/(X+Y)
1.20
1.11
1.25
1.67
1.67
1.60
1.50
2.00
2.00

Table 1 shows that the porosity of the grainpack decreases near linearly with increasing compaction factors and
that the spatial continuity in the X and Y directions decreases
more rapidly than in the Z direction. In spite of small fluctuations due to finite cut-off values, the heterogeneity
(anisotropy) increases with increasing compaction factors.
This is in agreement with the results of Bryant et al.27.

SPE Journal, Volume2, June 1997

S. BAKKE AND P.E. REN

Pore Structure.
Network Architecture, the Skeleton. In the present work,
the pore space architecture is defined as a complete 3-D
characterisation of the network structure of the pore space.
Serra38 and Adler30 define the skeleton of the pore space as
the set of points (voxels) at equal distances from two or more
points of the solid wall. The skeleton may thus be thought of
as a spatial representation of the deformation retract, or the
centre line, of the pore network. This spatial centre line will
contain points where 2 or more lines meet. These points are
the network nodes (approximate centre of pore bodies)
which are connected to other nodes via links or pore throats.
In 2-D image analysis, skeleton extraction from binary
images is carried out by means of thinning algorithms39. 2-D
thinning algorithms are mainly based on deleting or preserving the analysed pixel according to the rules for the binary 1and 0-pattern of its surrounding 8 neighbour pixels. In 2-D,
the number of possible 0- and 1-combinations around one
pixel is:

In the original mineral matrix space, the border surface

between two ultimate dilated grains is a grain-to-grain connection surface. Inside the original pore space, however, this
surface contains the pore network skeleton. The skeleton of
the pore network is thus found where two or more such dilated grain border surfaces meet. This line is defined by the
points in space which have neighbour voxels from 3 or more
different ultimately dilated grains (Fig. 16). The points on
this spatial line which have neighbour voxels from 4 or more
different ultimately dilated grains are the nodes in the pore
network.
This ultimate dilation technique gives a good estimate of
the skeleton of a pore network (Fig. 17). The skeleton provides both visual and quantitative information about the connectivity of the pore space. All the nodes (pore bodies) in the
skeleton are mapped and defined by their spatial coordinates. The architectural analysis of the pore network is
completed with mapping of the connected neighbour nodes
for each node (i.e., co-ordination number).

28 = 256
In 3-D, the number of combinations of 0s and 1s for the
26 neighbours surrounding one voxel is:
226 = 67 108 864
which makes definition of a complete set of rules for a 3-D
thinning algorithm very tedious. Thovert et al.40 succeeded
in developing a 3-D thinning algorithm which worked well
on their models. Attempts were made to develop a similar
thinning algorithm in this work. However, visual examination showed that in some complex voxel junctions, the algorithm introduced artificial holes in the pore network
skeleton. This can, for fluid flow purposes, be quite catastrophic because it may result in the definition of artificial
hydraulic circuits.
The logic of the sandstone modelling system offers, however, a solution to the 3-D skeleton or thinning problem. The
inverse function of thinning is dilation. Consequently, an
ultimate thinning of the pore network is exactly the same as
ultimate dilation of the complementary grain network. This is
carried out by first numbering each grain, as shown to the left
in Fig. 16. Next, the numbered grainbed array is subjected to
an ultimate dilation of the grains, as shown to the right in
Fig. 16.

Fig. 16Grainbed with numbered grains, ultimately dilated numbered grainbed to the right. Different grey
shades indicate grains with different numbers.
SPE Journal, Volume 2, June 1997

Fig. 17The skeleton of the pore network shown in Fig.

11.

Pore Geometry. As shown in Fig. 11, the detailed geometry

of the pore space is highly chaotic. In order to simulate fluid
flow in the pore network it is necessary to replace the irregular geometry of pore bodies and throats with an equivalent
but simplified geometry which retains the essential features
relevant to fluid flow. This can be done by recording the
following pore space descriptors: (i) the maximum inscribed
radius in pore bodies (volume calculations), (ii) minimum
inscribed radius in the pore space connecting two pore bodies (drainage capillary pressure and hydraulic conductance),
and (iii) pore shapes (for describing the simultaneous flow of
two or more fluids in the same pore).
The skeleton (Fig. 17) is used as a basis for mapping of
the pore throats between nodes. For each pore throat skeleton
voxel, the pore throat wall is mapped in a plane normal to the
143

3-D PORE-SCALE MODELLING OF SANDSTONES AND FLOW SIMULATIONS IN THE PORE NETWORKS

local direction of the pore throat. The throat radius is measured with a rotating vector which measures the distance from
the skeleton voxel to the pore throat wall in this plane for
each 10 degree 36 times. The area, perimeter and inscribed
radius for the narrowest constriction of the pore throat are
recorded (Fig. 18).
The volume of each pore body is calculated using an average radius from the geometrical centre point which is defined by the skeleton node. The geometry of the pore bodies
are analysed in a manner similar to that of the pore throats
i.e., the pore body wall is searched with a rotating radius
vector in increments of 10 degrees 36 times in a plane (Fig.
18). This plane is rotated 18 times in increments of 10 degrees to cover the entire space angle. Both the radii and the
co-ordinates of the points where the searching vector hits the
pore wall are recorded. Abnormal long radii are assumed to
be measured into pore throats and are eliminated by replacing them with the average value. For every pore body, the
volume and the inscribed sphere radius is recorded.

shape factor ranges from zero for a slit shaped pore to

0.0481 for a pore having the shape of an equilateral triangle.
The analysing technique described above provide a complete and quantitative characterisation of the pore network
architecture and the geometry of the pore bodies and pore
throats. Fig. 19 shows a simplified ball-and-stick representation of a numerical pore network generated from thin section
data from a 2160 mD Bentheimer sandstone.
The flow boundary conditions of the numerical pore network are defined as follows: The flow direction controls
which faces are inlet outlet. All nodes and links on the inlet
face (and the outlet face) are connected to an imaginary node
outside of this face. Currently, the four other faces have noflow boundaries (implementation of periodical flow boundaries is under way). The flow direction may be parallel to the
X, Y or Z axis.

Grain

Skeleton
line

Grain
Pore throat
cross sections

Fig. 182-D schematic showing the principles of geometrical analysis of pore throats. The upper length section shows the skeleton line and the planes for the geometrical analysis. The lower cross section shows how
the skeleton-based geometrical analysis of each of the
cross sectional planes is carried out.

A key characteristic of the pore structure of real porous

media is the angular corners of pores. Angular corners retain
wetting fluid and allow two or more fluids to flow simultaneously through the same pore body or throat. In the present
work, the cross-sectional shape of every pore body and throat
is described in terms of a dimensionless shape factor, G,
which is defined as41
G = A/p2

(2)

where A is the area of the pore body or throat cross section

and p is the corresponding perimeter length. The value of the

144

Fig. 19Schematic ball-and-stick representation of the

pore network of a Bentheimer sandstone with 1262
nodes. The radius of the spheres are proportional to the
average radius of the pore bodies, and the radius of the
cylinders are proportional to the minimum radius of the
pore throats. Cube sides are 1.7 mm long.

Network Modelling of Fluid Flow

The simulation of drainage displacements in the 3-D network
representations of our reconstructed sandstones has been
described previously42, but is repeated briefly here for completeness.
Flow Assumptions. Two-phase flow in the pore network is
modelled using the following assumptions:
1. The flow is everywhere laminar (Poiseuille flow) and
all the fluids are Newtonian, incompressible, and immiscible.
The injected fluid is the non-wetting fluid.
2. Pore bodies and pore throats may be occupied by one
or both fluids simultaneously. In the case of dual occupancy,
non-wetting fluid is the bulk fluid whilst wetting fluid is retained as wetting films in angular corners of pore bodies and
throats.
SPE Journal, Volume2, June 1997

S. BAKKE AND P.E. REN

3. Pore bodies or nodes are divided into two groups - full

nodes and interface nodes. Full nodes (nodes not being invaded) are occupied by a single bulk fluid whilst interface
nodes (nodes undergoing invasion) are occupied by two bulk
fluids. The fluid contents of an interface node change with
time.
4. Fluid pressures are only defined in pore bodies. Pore
bodies are considered to be sufficiently large for the capillary
pressure across an invading interface in the pore body to be
negligible.
For the present discussion, it is assumed that the pore
network is strongly water wet. Mixed wet systems are beyond the scope of the present discussion.

Anw =

r2
Aw
4G

where r is the inscribed radius of the pore body or throat and

rw is the radius of curvature of the film interface which is
given by the Young-Laplace equation
rw =

Pnw Pw

1 + 2 G IJ
rIJ

Fluid Conductance. In order to compute viscous pressure

drops in the network, we must first assign hydraulic conductances to pore bodies and throats. The hydraulic conductance
of a pore body or throat is determined by the geometry of the
cross-sectional area open to flow and the prevailing capillary
pressure. When both fluids are simultaneously present in a
pore body or throat, capillary pressure dictates that the nonwetting fluid occupies the bulk of the pore space and that
wetting fluid is retained in the angular corners, as schematically illustrated in Fig. 20. For strongly wetting conditions,
the cross-sectional areas open to flow for the two fluids (Anw
and Aw) are readily calculated from elementary geometry and
is given by
1

Aw = rw2
4G

SPE Journal, Volume 2, June 1997

nonwetting

(3)

where is the interfacial tension and rIJ is the inscribed radius of the minimum throat constriction. When the above
condition is satisfied, the throat is said to be open to the invading fluid and the invading fluid immediately enters the
throat. Otherwise, the throat is closed and the interface remains stationary at the entrance to the throat.
When a throat IJ opens, the invading interface immediately enters the adjoining node J which then becomes an
interface node. If the interface node is occupied by nonwetting fluid, coalescence occurs and the displacement is
considered to be complete. If it is filled with wetting fluid,
non-wetting fluid displaces wetting fluid from the node. The
rate at which an interface node fills or empties is determined
by the viscous pressure gradients in the two fluids.

(4)

(6)

Invasion of Pore Throats and Bodies. When the invading

non-wetting fluid is located in a pore body I which is connected to a pore throat IJ filled with wetting fluid, capillary
forces prevent the non-wetting fluid from spontaneously entering the throat. The non-wetting fluid can only enter the
throat when the pressure in the non-wetting fluid (Pnw) exceeds the pressure in the wetting fluid (Pw) by a value equal
to the threshold capillary pressure Pc

Pnw , I Pw , J Pc , IJ =

(5)

rw

Fig. 20Schematic distribution of wetting and nonwetting fluids in a triangular pore body or throat.

In the present treatment, we make the simplification that

the conductance of the fluid i which occupies the bulk of a
pore body or throat may be characterised by the concept of a
mean hydraulic radius, Ri, defined as
Ri = 0.5(r + rv ,i )

(7)

where rv,i is an equivalent volume radius which is defined as

the radius of the cylinder whose volume is equal to the volume of fluid i in the pore body or throat, i.e.
rv ,i =

Ai

(8)

If the bulk of the pore space is occupied by non-wetting

fluid, Ai is given by Eq. (5) whilst if it is completely filled
with wetting fluid, Ai=r2/4G. The conductance of fluid i, gi,
is given by Poiseuille's law
gi =

Ri2 Ai
8 i l

(9)

where i is the fluid viscosity and l is the length of the pore

body or throat.
If wetting fluid is present in the corners of a pore body or
throat which is predominantly occupied by non-wetting fluid,
we use:

145

3-D PORE-SCALE MODELLING OF SANDSTONES AND FLOW SIMULATIONS IN THE PORE NETWORKS
2
v ,w

r Aw
8 i l

gw =

(10)

where is a dimensionless flow resistance factor which accounts for the reduced conductance of wetting fluid close to
the pore wall. Finite element solutions43 of the corner flow
problem show that depends on the corner geometry and on
the boundary condition at the non-wetting/wetting fluid interface. For instance, for an equilateral triangular pore body or
throat, =5.3 if there is no flow at the fluid interface whilst
=2.5 for a no stress interface boundary condition.
Phase Pressures. For laminar flow, the volumetric flow rate
of fluid i between two connecting pore bodies I and J, Qi,IJ is
given by
Qi , IJ = gi , IJ (Pi , I Pi , J )

(11)

where gi,IJ is the hydraulic conductance of fluid i between the

centres of pore bodies I and J. For simplicity, we assume that
gi,IJ is the harmonic mean of the conductances of the throat
connecting the pore bodies and the two pore bodies themselves, i.e.

1
1
1 1
1
=
+
+

gi , IJ gi,tr 2 gi, I gi, J

(12)

Since the fluids are incompressible, continuity requires

that at each node

i , IJ

=0

(13)

where J runs over all the pore throats connected to node I.

Eqs. (11) and (13), together with the appropriate initial and
boundary conditions, give rise to a set of linear algebraic
equations for the nodal or pore body centre pressures which
in matrix form may be written as
GP = b

(14)

where G is a sparse matrix whose elements contain the g's, P

is the (unknown) nodal pressures, and b is a source vector
which is essentially zero except for inlet, outlet, and interface nodes. The nodal pressures are determined by an iterative conjugate gradient method.
The advance of the displacement is treated using the usual
quasi-static approximation. During a time step t, the fluids
flow through the network at a constant rate in response to the
computed pressure fields in each of the two fluids. The time
step is selected such that only one interface node is completely filled in every time step. Since a number of interface
nodes may be undergoing invasion simultaneously, we identify all the nodes which are filling and compute the time required to fill each node. The minimum invasion time, t, is
determined and the non-wetting fluid saturation in every
interface node is updated

Snw , I (t + t ) = Snw , I (t ) +

IJ

(15)

where VI is the volume of node I and the sum only includes

non-wetting phase filled pore throats which are connected to
interface node I. In some interface nodes, Snw may decrease.
If Snw reaches zero, non-wetting fluid spontaneously retracts
out of the connecting pore throats and the node is removed
from the list of interface nodes.
Initial and Boundary Conditions. Initially, the network
contains only wetting fluid. Non-wetting fluid is injected at a
constant rate through a non-wetting fluid reservoir which is
attached via pore throats to every pore body along the inlet
side of the network. Fluids escape through the outlet face on
the opposite side of the network where the wetting phase
pressure is arbitrarily set to zero. Impermeable boundary
conditions are imposed along the sides parallel to the main
direction of flow.
Since the wetting fluid is continuous throughout the entire
network, Eq. (14) must be applied at all pore bodies other
than those at the outlet where the pressure is arbitrarily set to
zero. The non-wetting fluid is also continuous. However, it
only occupies part of the network at any instant in time. Eq.
(14) is thus only applied at pore bodies which are occupied
by the non-wetting fluid.
The solution for the two fluid pressures is iterative because the individual fluid pressures are coupled through the
pressures in interface nodes where the bulk fluid pressures
are equal, and because it is necessary to specify a priori
which pore throats are open and which are closed to the invading non-wetting fluid. The procedure for each time step,
t, is as follows:
1. The pressure for each fluid in every node containing
the fluid is computed using Eq. (14).
2. The computed pressures are used to calculate capillary
pressures for each closed throat. If the capillary pressure
exceeds the threshold pressure, the throat is opened and the
pressure solution is re-calculated to check that the throat
remains open. This procedure is iterated until the computed
capillary pressures are everywhere consistent with the
threshold capillary pressure criteria.
3. The flow rates at every interface node is computed using Eq. (11). The flow rates are used to calculate the time
step t such that only one interface node is completely filled
during the time step.
4. The saturation in all the interface nodes is updated using Eq. (15).
5. Repeat the above procedure for the next time step.
Macroscopic Properties. The calculations of macroscopic
transport properties are performed on a central region of the
pore network in order to minimise boundary effects. When
the network is fully saturated with wetting fluid, we specify a
constant pressure drop across the network and calculate the
flow rate Q and the pressure drop P across the region of
interest. The absolute permeability K is determined from
Darcy's law, i.e.
K=

146

t
VI

Q L
A P

(16)

SPE Journal, Volume2, June 1997

S. BAKKE AND P.E. REN

where A is the cross-sectional area of the domain and L is the

length.
Relative permeabilities, kr, are calculated in a similar
manner. At various stages of the displacement we compute
the average pressure drop in the non-wetting and wetting
fluids across the domain. The saturation and the total flow
rates of non-wetting and wetting fluids across the domain are
also calculated. The relative permeability of fluid i is computed using Darcy's law
k r ,i ( S ) =

Qi i L
KAPi

(17)

The effective capillary pressure, Pc(S), in the domain is

defined as the difference in the mean pressures of the nonwetting and wetting fluids. This is computed at various
stages of the simulation.
The electrical conductivity r of the network when it is
fully saturated with a conducting brine is found from Eqs.
(11-14), but with current I substituted for Q, voltage gradient
V for P, and electrical conductance ge for g. We assume
that the pore walls are insulating and that the electrical conductance depends only upon the geometry of the pore body
or throat and is given by
ge =

A w
l

throats. The measured formation factor was 12.0 which is in

excellent agreement with the predicted one, 11.7.
Fig. 22 compares computed and experimentally measured
(porous plate method) drainage capillary pressures. The
predicted capillary pressure curve is in good agreement with
the experimental one. Fig. 23 compares computed and experimentally measured (steady state method) drainage relative permeabilities. The simulated results are in good agreement with the experimental data, both for the wetting and the
non-wetting phases. It should be emphasised, that this
agreement is obtained without the use of any adjustable parameters. The only inputs are readily available thin section
data and knowledge that capillary forces dominate the displacement on the pore-scale.

(18)

where w is the (known) brine conductivity and A is the

cross-sectional area of the pore body or throat. A constant
voltage gradient is applied across the network and the system
is relaxed using a conjugate gradient method to evaluate the
voltages at each pore body. From the voltage distribution
one may calculate the total current and thus the formation
factor, F = w/r.
Simulation Results. Network models have long been used
to study two-phase2-6, and recently three-phase17-19, displacement processes in simple or idealised porous media.
The extension of network modelling techniques to real porous media is complicated by the difficulty of adequately
describing the complex nature of the pore network and seldom do network models claim to be representative of reservoir rocks. The present work is an attempt at overcoming
some of this difficulty and offers the possibility of using
network models in a predictive capacity.
To investigate the possibility of extending predictive capabilities to network models, a capillary dominated (Nca =
10-6) drainage displacement was simulated on the network
representation (Fig. 19) of the reconstructed Bentheimer
sandstone. Fig. 21 shows four snapshots of the progress of
the simulation in a small subsample of the pore network (the
flow is from left to right). The schematic ball-and-stick representation of pore bodies and throats is for illustrative purposes only.
The computed permeability of the pore network was 3010
mD whilst the experimentally measured permeability was
2160 mD. The difference between the predicted and measured permeability is most likely due to uncertainties in the
assignment of hydraulic conductance to pore bodies and

SPE Journal, Volume 2, June 1997

Fig. 21Four snapshots showing non-wetting fluid

(dark grey) displacing wetting fluid (grey) in a small subsample of the pore network for a Bentheimer sandstone,
cube sides are 0.7 mm.

Fig. 22Comparison between simulated and experimentally measured drainage capillary pressure for a strongly
water wet Bentheimer sandstone.

147

3-D PORE-SCALE MODELLING OF SANDSTONES AND FLOW SIMULATIONS IN THE PORE NETWORKS

S=
t=
V=
=
=
=
=
=

Fig. 23Comparison between simulated and experimentally measured drainage relative permeabilities for a
strongly water wet Bentheimer sandstone.

The agreement between predicted and experimental data

provides a strong hint that it may be possible to use network
models to a priori predict petrophysical and transport properties of rocks from the underlying microstructure, at least
for strongly wetted systems.
Conclusions
(1) Process/stochastic modelling of the sandstone-forming
processes sedimentation, compaction, and diagenesis provides a cost-effective and rapid method for generating representative 3-D pore-scale models of homogenous and heterogeneous sandstones from thin section data.
(2) The degree of heterogeneity of the sandstone model is
determined by analysing the spatial continuity of the pore
space in the X, Y, and Z directions using a scale-independent
invasion percolation based algorithm. Spatial continuity
analyses show that compaction reduces the spatial continuity
in the horizontal direction more rapidly than in the vertical
direction
(3) 3-D image analysis techniques may be used to extract
the pore space network of a rock from its complementary
grain matrix network, and to quantitatively characterise the
architecture and geometry of the pore network.
(4) A 3-D pore network generated from thin section data
from a strongly water wet Bentheimer sandstone is used as
input to a two-phase network flow simulator. Predicted
drainage capillary pressure and relative permeabilities are in
good agreement with laboratory measurements.
Nomenclature
A=
F=
g=
G=
K=
kr =
l=
L=
Nca=
p=
P=
Q=
r=
R=
148

area
formation factor
conductance
dimensionless shape factor
absolute permeability
relative permeability
length
length of domain
capillary number
perimeter length
pressure
volumetric flow rate
radius
mean hydraulic radius

saturation
time
volume
dimensionless flow resistance factor
interfacial tension
compaction factor
fluid viscosity
electrical conductivity

Subscripts
c = capillary
e = electrical
i = fluid no. i
I = pore body no. I
J = pore body no. J
tr = throat
r = rock
v = volume-equivalent
w = wetting
nw = non-wetting
Acknowledgements
The authors would like to acknowledge Den norske stats
oljeselskap a.s. (Statoil) for granting permission to publish
this paper.
References
1.

2.

3.
4.
5.

6.
7.
8.

9.
10.

11.

Wardlaw, N.: Quantitative Determination of Pore Structure

and Application to Fluid Displacement in Reservoir Rocks,
in North Sea Oil and Gas Reservoirs - II, Graham and Trotman, (1990).
Jerauld, G. R., and S. J. Salter: The Effect of Pore-Structure
on Hysteresis in Relative Permeability and Capillary Pressure: Pore-level Modelling, Transport in Porous Media, 5,
(1990), 103-151.
Blunt, M. J., and P. J. King: Relative Permeabilities from
Two- and Three-Dimensional Pore-Scale Network Modelling, Transport in Porous Media, 6, (1991), 407.
Blunt, M. J., P. J. King, and H. Scher: Simulation and Theory of Two-Phase Flow in Porous Media, Phys. Rev. A, 46,
(1992), 7680.
Billiotte, J., H. De Moegen, and P. E. ren: Experimental
Micromodelling and Numerical Simulation of Gas-Water Injection-withdrawal Cycles as Applied to Underground Gas
Storage, SPE. Adv. Tech. Ser., 1, (1993), 133.
Chan, D. Y. C., B. D. Hughes, and L. Paterson: Simulating
Flow in Porous Media, Physical Review A, 38, (1988),
4106-4120.
Lenormand, R., C. Zarcone, and A. Sarr: Mechanism of the
Displacement of One Fluid by Another in a Network of
Capillary Ducts, J. Fluid. Mech., 135, (1983), 337.
Lenormand, R. and C. Zarcone: Role of Roughness and
Edges During Imbibition in Square Capillaries, SPE 13264,
Proc. SPE 59th Ann. Tech. Conf. and Exhib. of the SPE,
Houston, Texas, (1984).
Li, Y. and N. Wardlaw: The Influence of Wettability and
Critical Pore-Throat Size Ratio on Snap-Off, J. Colloid and
Interface Science, 109, (1986), 461-472.
Kantzas, A., I. Chatzis, and F. A. L. Dullien: Mechanisms of
Capillary Displacement of Residual Oil by Gravity-Assisted
Inert Gas Injection, SPE 17506, Proc. SPE Rocky Mountain
Regional Meeting, Casper, Wyoming, (1988).
Kantzas, A., I. Chatzis, and F. A. L. Dullien: Enhanced Oil
Recovery by Inert Gas Injection, SPE 17379, Proc.

SPE Journal, Volume2, June 1997

S. BAKKE AND P.E. REN

12.
13.

14.

15.

16.
17.

18.

19.
20.

21.

22.

23.
24.
25.

26.
27.
28.
29.
30.
31.

SPE/DOE Symposium on Enhanced Oil Recovery, Tulsa,

Oklahoma, (1988).
ren, P. E., J. Billiotte, and W. V. Pinczewski: Mobilization
of Waterflood Residual Oil by Gas Injection for Water-Wet
Conditions, SPE Formation Evaluation, 7, (1992), 70-78.
ren, P. E. and W. V. Pinczewski: The Effect of Film Flow
on the Mobilization of Waterflood Residual Oil by Gasflooding, Proc. 6th. European IOR-Symposium, Stavanger, 1,
(1991), 705-716.
ren, P. E. and W. V. Pinczewski: Effect of Wettability and
Spreading on the Recovery of Waterflood Residual Oil by
Immiscible Gasflooding, SPE Formation Evaluation, 9,
(1994), 149-156.
ren, P. E. and W. V. Pinczewski: Fluid Distribution and
Pore-Scale Displacement Mechanisms in Drainage Dominated Three-Phase Flow, Transport in Porous Media, 20,
(1995), 105-133.
Blunt, M. J., D. Zhou, and D. H. Fenwick: What Determines
Residual Oil Saturation in Three Phase Flow?, SPE 27816,
Proc. Improved Oil Recovery Symposium, Tulsa, OK, (1994).
Soll, W. E., and M. A. Celia: A Modified Percolation Approach to Simulating Three-Fluid Capillary PressureSaturation Relationships, Advances in Water Resources, 16,
(1993), 107-126.
ren, P. E., J. Billiotte, and W. V. Pinczewski: Pore-Scale
Network Modelling of Waterflood Residual Oil Recovery by
Immiscible Gas Flooding, SPE 27814, Proc. SPE/DOE 9th
Sympoium on Improved Oil Recovery, Tulsa, Oklahoma,
(1994).
Fenwick, D. H. and M. J. Blunt: Pore Level Modelling of
Three Phase Flow in Porous Media, Proc. 8th European
IOR Symposium, Vienna, Austria, (1995).
Yanuka, M., F. A. L. Dullien, and D. E. Elrick: Geometrical
and Topological Model of Porous Media Using a ThreeDimensional Joint Pore Size Distribution, in "Percolation
Processes and Porous Media" Academic Press Inc., (1986).
Oosthuyzen, E. J.: An Elementary Introduction to Image
Analysis - a New Field of Interest at the International Institute for Metallurgy, Report No. 2058, Randberg, SouthAfrica, (1980).
Etris, E. L., D. S. Brumfield, R. Ehrlich, and S. J. Crabtree:
Relations between Pores, Throats and Permeability: A Petrographic/Physical Analysis of Some Carbonate Grainstones
and Packstones, Carbonates and Evaporites, 3, (1988), 17.
Coles, M. E., P. Spanne, E. L. Muegge, and K. W. Jones:
Computed Microtomography of Resevoir Core Samples,
Proc. 1994 Int. Symp. of SCA, Stavanger, (1994).
Hazlett, R. D.: Simulation of Capillary-Dominated Displacements in Mictotomographic Images of Reservoir
Rocks, Transport in Porous Media, 20, (1995), 21-35.
Holt, R. M., E. Fjr, O. Torster, and S. Bakke:
Petrophysical Laboratory Measurements for Basin and Reservoir Evaluation, Marine and Petroleum Geology, 13,
(1996), 383-391.
Lin, C. and M. H. Cohen: Quantitative Methods of Microgeometric Modelling, J. Appl. Phys., 53, (1982), 4152.
Bryant, S., C. Cade, and D. Mellor: Permeability Prediction
from Geologic Models, AAPG Bull. , 77, 8, (1993), 1338.
Finney, J.: Random Packings and the Structure of the Liquid
State, Ph.D. thesis, University of London, London, England.
(1968)
Finney, J.: Random Packings and the Structure of Simple
Liquids. I. The Geometry of Random Close Packings, Proc.
Roy. Soc., 319A, (1970), 479.
Adler, P. M.: Porous Media. Geometry and Transports. Butterworth-Heinemann Series in Chemical Engineering, (1992).
Reyes, S. C., and E. Iglesia: Effective Diffusivities in Catalyst Pellets: New Model Porous Structures and Transport
Simulation Techniques, J. of Catalysis, 129, (1991), 457.

SPE Journal, Volume 2, June 1997

32.

Spedding, P. L. and R. M. Spencer: Simulation of Packing

Density and Liquid Flow in Fixed Beds, Computers chem.
Engng,, 19, 1, (1995), 43.
33. Sundborg, .: The River Klaralven, a Study in Fluvial
Processes, Geografiska Annaler, Serie A, 38, (1956), 125.
34. Bjrkum, P. A.: How Important is Pressure in Causing Dissolution of Quartz in Sandstones?, J. Sed. Res., In press.
35. Yokoi, S., J. Toriwaki, and T. Fukumura: A Sequential Algorithm for Shrinking Binary Pictures, Systems, Computers.
Controls., 10, 4, (1979), 69.
36. Bakke, S. and C. M. Griffiths: Interactive Quantitative
Matching of Stratigraphic Sequences of Numerical
Lithostates Based on Gene-Typing Techniques, In Correlation in Hydrocarbon Exploration, Norwegian Petroleum Society, Graham & Trotman, (1989), 61.
37. Wilkinson, D. and J. F. Willemsen: Invasion Percolation: A
New Form of Percolation Theory, J. Phys. A: Math. Gen.,
16, (1983), 3365.
38. Serra, J.: Image analysis and mathematical morphology,
Academic Press, New York (1982).
39. Hilditch, C. J.: Linear Skeleton from Square Cupboards, in
Machine Intelligence, 6, Edinburgh Univ. Press, (1969), 403.
40. Thovert, J. F., J. Salles, and P. M. Adler: Computerized
Charcterization of the Geometry of Real Porous Media:
Their Discretization, Analysis and Interpretation, J.of Microscopy, 170, (1993), 65.
41. Mason, G., and N. R. Morrow: Capillary Behaviour of a
Perfectly Wetting Liquid in Irregular Triangular Tubes, J. of
Colloid and Interface Science, 141, (1991), 262-274.
42. ren, P. E., S. Bakke, L. S. Nilsen, and A. Henriquez:
Prediction of Relative Permeability and Capillary Pressure
from Pore-Scale Modelling, Proceedings of the 5th European Conference on the Mathematics of Oil Recovery,
Leoben, Austria, (1996).
43. Ranshoff, T. C., and C. J. Radke: Laminar Flow of a Wetting Liquid Along the Corners of a Predominantly Gasoccupied Noncircular Pore, J. of Colloid and Interface Science, 121, (1988), 392-401.
SPEJ
Stig Bakke is a geologist at Statoil Research Centre, Department of Petroleum Engineering, Postuttak, N-7004
Trondheim, Norway, email: stiba@statoil.no. His research
interests include micro-scale reservoir description, numerical
geological process modelling and 2-D and 3-D image analysis. Before joining Statoil, he worked at IKU, Trondheim.
Bakke holds a MS in geology from NTH.
Pl-Ericren, SPE, is a research scientist at Statoil Research Centre, Department of Petroleum Engineering, Postuttak, N-7004 Trondheim, Norway, email: peoe@statoil.no.
His research interests include physics of multiphase flow,
reservoir engineering, and modelling of wettability. ren
holds a BS in chemical engineering from the University of
Michigan, and a PhD in petroleum engineering from the University of New South Wales.

149