Hindawi

Geouids
Volume 2017, Article ID 7240524, 24 pages
https://doi.org/10.1155/2017/7240524

Research Article
Lattice Boltzmann Simulations of Fluid Flow in
Continental Carbonate Reservoir Rocks and in Upscaled Rock
Models Generated with Multiple-Point Geostatistics

J. Soete,1 S. Claes,2 H. Claes,1 N. Janssens,1 V. Cnudde,3 M. Huysmans,1,4 and R. Swennen1

1
Earth and Environmental Sciences, KU Leuven, Celestijnenlaan 200E, 3001 Heverlee, Belgium
2
Civil Engineering, KU Leuven, Kasteelpark Arenberg 40, 3001 Heverlee, Belgium
3
ProGRess, Geology and Soil Science, Ghent University, Krijgslaan 281, S8, 9000 Ghent, Belgium
4
Hydrology and Hydraulic Engineering, Vrije Universiteit Brussel, Pleinlaan 2, 1050 Brussel, Belgium

Correspondence should be addressed to J. Soete; jeroen.soete@kuleuven.be and S. Claes; steven.claes@kuleuven.be

Received 12 April 2017; Accepted 5 July 2017; Published 24 September 2017

Academic Editor: Yi Wang

Copyright 2017 J. Soete et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Microcomputed tomography (CT) and Lattice Boltzmann Method (LBM) simulations were applied to continental carbonates
to quantify fluid flow. Fluid flow characteristics in these complex carbonates with multiscale pore networks are unique and the
applied method allows studying their heterogeneity and anisotropy. 3D pore network models were introduced to single-phase
flow simulations in Palabos, a software tool for particle-based modelling of classic computational fluid dynamics. In addition,
permeability simulations were also performed on rock models generated with multiple-point geostatistics (MPS). This allowed
assessing the applicability of MPS in upscaling high-resolution porosity patterns into large rock models that exceed the volume
limitations of the CT. Porosity and tortuosity control fluid flow in these porous media. Micro- and mesopores influence flow
properties at larger scales in continental carbonates. Upscaling with MPS is therefore necessary to overcome volume-resolution
problems of CT scanning equipment. The presented LBM-MPS workflow is applicable to other lithologies, comprising different pore
types, shapes, and pore networks altogether. The lack of straightforward porosity-permeability relationships in complex carbonates
highlights the necessity for a 3D approach. 3D fluid flow studies provide the best understanding of flow through porous media,
which is of crucial importance in reservoir modelling.

1. Introduction understanding the reservoir behavior. The sedimentology of

continental carbonates has been widely studied [713], but
Porosity and permeability control the storage and fluid flow
recently the focus of continental carbonate studies shifted
in reservoir rocks. An example of potential reservoir rocks is
towards the rocks petrophysical properties like porosity,
continental carbonates, such as travertines (a term here used
sensu lato [1]), which are highly heterogeneous as a result of permeability, and acoustic velocities [1422]. Noteworthy
their geological evolution, influenced by sedimentary origin, is the fact that the permeability inside rocks is strongly
diagenetic processes, and burial history. The latter processes dependent on the geometric and topological properties of the
influence the size and shape of pores, producing some of the porous medium at microscopic scales [23].
most complex pore networks recorded in sedimentary rocks. Many efforts have been made to reconstruct digital pore
Recent hydrocarbon discoveries highlighted the continen- networks based on computer tomography (CT) to study the
tal carbonate reservoir potential in the presalt exploration, fluid flow in rock samples [2429]. In this study, the 3D pore
offshore Brazil [2, 3] and in the Namibe basin, Angola [4 networks were acquired with different CT systems, which
6]. Quantitative data about lithofacies occurrence, distribu- allowed scanning a range of sample volumes at different
tions, and their related porosity and permeability are key to resolutions. After segmentation, the obtained pore networks
2 Geofluids

were loaded into Palabos, a software tool for classic computa- types present in the quarries, that is, the subhorizontal,
tional fluid dynamics (CFD). The source code and scripts are reed, cascade, and waterfall facies (Figure 1), as defined by
available at the Palabos website [30]. Claes et al. [1]. All of these facies types have pores at least
In order to evaluate the reservoir potential of continental varying from nanometer to centimeter scale and could have
carbonates, a critical assessment has to be made regarding relevant contributions to fluid flow in this material. For a
the scale at which petrophysical measurements should be sedimentological background of the samples (out of the scope
performed. This requires covering several spatial scales. of this study), the reader is referred to earlier published
Multiscale flow modelling is defined as any method used to literature [1, 14, 41, 42].
explicitly represent the flow properties at more than one scale The facies types were usually characterized by different
within a reservoir [31]. Reservoir models typically cover at meso- and macropore networks, here defined as pores that
least twelve orders of magnitude, ranging from pore to core have sizes of 1100 m and >100 m, respectively. Pseud-
to interwell to full field simulations. Four general scale orders ofenestral and interpeloidal pores, aligned along the layer-
can be recognized, which are pore to lithofacies, lithofacies ing, dominate the subhorizontal (Figure 1(a)) and cascade
to geomodel, and geomodel to reservoir model. Multiple- (Figure 1(b)) facies. In the latter, interlayer and shelter pores
point geostatistics (MPS) gained importance in the recent were also present between shrub crusts [1]. The reed facies
years for modelling geological structures at different spatial (Figure 1(c)) was named after its characteristic phyto-moldic
scales, from m to m [32, 33]. For example, in the field of porosity. In the waterfall facies (Figure 1(d)), hanging plants
hydrology, MPS allows building models on decameter scale on steep slopes became encrusted and generated highly
[34], while Okabe and Blunt [35] used the technique to model porous zones in which well-connected phyto-moldic pores
pore space on a micrometer scale. MPS enables simulating and high interstitial pore spaces formed framework porosity.
complex interconnected pore structures by directly inferring It has to be noted that pore sizes of vugs, caverns, and shelters
the patterns from training images and furthermore allows in these continental carbonates can exceed the sample size of
modelling heterogeneity. This allows simulation of complex classical plugs or cores.
interconnected structures [36, 37] and the upscaling of
complex pore systems to scales relevant to flow property 3. Methodology
considerations.
In this study, the CFD were for the first time applied 3.1. Core Analysis. Petrophysical measurements in this study
to representative continental carbonate samples, taken from were conducted on plugs with 2.5 cm and 3.4 cm diameter.
different reservoir analogues, in order to study micro- to These cylindrical plugs were taken in the core laboratory with
mesoscale fluid flow in lithofacies-specific pore networks. a Hilti water-cooled diamond coring drill. After plugging,
A major benefit to these models is that the samples are the samples were cut and polished to make the ends flat
differentiated based on sedimentological interpretation. Fur- and parallel. In the first step of the research, samples were
thermore, it was verified that scan resolution and also sent to Panterra Geoconsultants (Leiderdorp, Netherlands)
porosity and tortuosity influence the simulated permeability. for porosity-permeability analyses. The effective porosity
In heterogeneous carbonates, pores sizes range over several in the plugs is measured by means of helium expansion
orders of magnitude and pores at different scale influence porosimetry. Gas permeability, in this case with nitrogen gas
the interconnectivity. It is thus necessary to study the pore (N2 ), is measured in a steady-state permeameter.
network at different scales. Not only voxel and pore net-
work based stochastic reconstruction methods [38, 39] but 3.2. Computer Tomography (CT). CT imagery is frequently
also simulated annealing and Markov Chain Monte Carlo applied in recent geomaterials research and industrial appli-
methods [40] have been proposed to merge data obtained cations [4346]. In this study, CT was used to simulate
from multiscale imaging. Here, a MPS scale-independent petrophysical properties in porous continental carbonate
workflow was proposed which integrates small- and large- media. An inherent characteristic of CT is the relationship
scale pore patterns, thus generating large volume porosity between resolution and sample size. The resolution of the
models at high resolution. MPS contributes to the com- scan () is a combination of the pixel size () of the detector,
prehension of reservoir behavior in rocks by solving the the magnification of the object (), and the size of the X-
common reservoir upscaling problem. To verify whether ray focal spot (). Equation (1) gives the relationship between
generated pore patterns were accurate, a comparison between these parameters:
simulated and measured permeabilities on matching artificial
and original samples was made. The latter assured that MPS 1
= + (1 ) ,
can in future studies be used to optimize digital pore networks
and the accuracy of simulated permeabilities. (1)
SDD
= ,
SOD
2. Research Material
where SSD is the source-detector distance and SOD is the
Continental carbonate samples from outcrops in the Ballk source-object distance.
area (Turkey) and Sutto and Budakalasz (Hungary) were Different and new CT scanners generate scans at different
selected for analysis. The continental carbonates are of Qua- and ever improving scan resolutions, which can influence
ternary age. The dataset covered the four dominant facies the estimated porosity. For example, the HECTOR scanner
Geofluids 3

Subhorizontal facies 5 cm Cascade facies 50 cm

(a) (b)

Reed facies 10 cm Waterfall facies 30 cm

(c) (d)

Figure 1: Dominant lithofacies types in the studied continental carbonate quarries. (a) Subhorizontal facies with pseudofenestral porosity
(arrows). (b) Cascade facies with dendrite crusts (arrow 1) and shelter pores (arrow 2). (c) Reed facies with reed moldic porosity (arrows). (d)
Waterfall facies with plant framework porosity (arrows).

developed at UGCT [47] has a 240 kV setup which allows the porosity and thus unrepresentative simulated permeabilities.
scanning of large core samples with a diameter of 10 cm at a The REV of 10 samples, with a scan resolution of 16 m,
28 m resolution. This resolution is normally only reachable was determined according to the statistical REV calculation
on smaller plug samples using standard micro-CT scanners. method described in Claes [49], which follows the approach
HECTOR CT scans were processed in Octopus [48]. In of Bear [23]. The latter author used the chi-square criterion
addition to the HECTOR CT, samples in this study were (2 ) as a measure of porosity fluctuation inside a selected
also scanned with a Phoenix Nanotom S instrument (GE sample volume. The smaller this 2 value was, the closer
Measurement and Control Solutions, Wunstorf, Germany), the correspondence with the volume of the REV was. For
equipped with a 180 kV/15 W high-performance nanofocus each sample, the size of the REV was calculated 100 times,
X-ray tube and a 2304 2304 pixel Hamamatsu detector [21]. according to the procedure of Claes [49]. QQ plots of these
Radiographs were reconstructed with the volume processing 100 REV simulations and value calculations of the chi-
software Phoenix datos|x (GE Measurement and Control square goodness of fit test are used to verify whether the
Solutions, Wunstorf, Germany) and images with an isotropic REV simulations are log-normally distributed. 95% confi-
voxel size of 2, 4, 12, or 16 m were exported. Slices were dence intervals for the REV were subsequently calculated.
segmented in Matlab. Volume of Interest (VOI) selections and This method allowed determining objectively whether the
further analysis were done in Avizo Fire. characteristic meso- and macropores in the selected samples
were representative.
3.3. Representative Elementary Volume. The Representative
Elementary Volume (REV) is a crucial concept when eval- 3.4. Multiple-Point Geostatistics (MPS)
uating petrophysical properties of reservoir rocks. The REV
or unit cell is the smallest volume over which a porosity The Single Normal Equation Simulation (SNESim) Algorithm.
measurement can be made which yields a value representative Traditional geostatistical simulation algorithms can be subdi-
of the whole. In this study, porosity and the pore network are vided into two groups based on their principal method: pixel
used as starting point to perform permeability simulations. based group and object based group. The former simulates
Such simulations could only be correct if the REV for porosity one pixel at a time, while the latter fits an object or pattern
was reached. When the REV for the porosity and pore types onto the simulation grid in one step. One of the important
present is not reached, the pore network will yield variable differences between both methods is the ease with which they
4 Geofluids

can handle conditioning data. Pixel based methods are easily (c) Draw a simulated value () from the condi-
adapted to incorporate hard conditioning data. In contrast, tional distribution and add it to the dataset.
object based methods faithfully fit the large-scale patterns but
are difficult to condition to local data if that data is abundant. Several parameters play an important role in the obtained
The multiple-point concept proposed by Journel [50] and simulation results. They were varied in order to improve the
Guardiano and Srivastava [51] combines the strengths of both results of the simulations. Liu [53] and Meerschman et al.
methods. The main difference between MPS and traditional [54] provide an extensive analysis of the influence of the
geostatistical methods is the use of a training image (TI). This different parameters on the simulations. The most important
TI is a representation of the expected geometry and spatial parameters for upscaling using MPS are described below.
distribution of the objects present in the actual field and In order to capture the large-scale structural information
does not need to be a real image of the field. The simulation in MPS simulations, the original algorithm was adapted
is performed pixelwise with the conditional probabilities to allow the use of multigrids. Tran [55] introduced this
respecting the conditional proportion of the TI. The TI must technique to create a large-scale template with a reasonably
at least be the size of the REV to capture pore geometries and small number of nodes. It was important to keep the number
connections. of nodes limited because otherwise the search tree would
Figure 2 provides a schematic overview of the different become too large, resulting in an exponential increase of
elements used in a MPS approach, as well as its strengths calculation time. In a first step, nodes on the coarsest grid
for simulating complex geological structures. Figure 2(a) were simulated using the rescaled template. Subsequently, the
represents the original image, in which three facies types can nodes on the second coarsest grid were simulated and so on.
be recognized. In Figure 2(b), the principal components of The relationship between different grids was expressed in
the original image are retained and a training image is created. (3). Figure 3 shows an example in which three multiple grids
Details concerning facies 2 (grey), with the lowest occurrence are used ( = 3):
probability in Figure 2(a), are not retained in the training = {21 1 , . . . , 21 } . (3)
image and hence are not present in the simulated images
(Figures 2(d) and 2(e)). Figure 2(c) shows the conditioning In MPS, the TI is used to provide detailed information on
data (e.g., well data) used in the simulations. Figures 2(d) and which the simulation is based. This information is derived
2(e) show the result of MPS simulations using the SNESim from the patterns present in the TI. However, often the
algorithm and the classical Sequential Indicator Simulation marginal distributions of the different categories in the TI
(SISim) algorithm, respectively. In Figure 2(d), the connec- are not equal to those of the real sample, which needs
tivity of facies 3 (black) is better preserved compared to to be simulated; that is, the percentages of occurrence of
the results in Figure 2(e). This observation indicates the the different categories diverge between the TI and the
advantage of a multiple-point approach in contrast to two- desired simulation. Strebelle and Journel [32] implemented
point correlations used in SISim. a method to adapt this discrepancy of the marginal category
In this study, the SNESim algorithm implemented by distribution. However, if the marginal probability of the TI
Strebelle [52] was used. Using this algorithm, the TI was and the desired marginal probability are not close enough to
scanned once and all conditional proportions were stored in a each other, the algorithm would not reproduce the desired
search tree data structure. In the next step, these proportions proportions of each category present in the simulated images.
were used to create the simulated values. At each simulation The servo corrector, introduced by Strebelle [52], bends the
node , the search template was used to retrieve the running simulated marginal probability further towards the
conditional data event Dev(), which is defined as target proportions by introducing a parameter . The larger
is, the stronger the impact of the applied correction is (see
Dev () = { ( + 1 ) , . . . , ( + )} , (2) the following equation):
where ( + ) is a filled-in nodal value. new ( | ) = ( | ) + ( () ()) ,
(4)
SNESIM Algorithm = [0, 1] ,
1
(1) Define search template and construct search tree
specific to template . where () is the proportion which was calculated based
on the original sample and all previously simulated nodes. It
(2) Relocate hard data to the nearest simulation grid node is, however, important to note that () and the MPS can-
and freeze them during simulation. not completely be decoupled; hence, the target proportions
(3) Define a random path visiting all locations to be should not be too different from the TI [53].
simulated and for each location do the following:
MPS in Upscaling. Figure 4 provides a schematic outline of the
(a) Find conditioning data event Dev () defined workflow. Figure 4(a) is a detailed image (TI) obtained using
by template . CT. Hence, this dataset, with an isotropic voxel resolution
(b) Retrieve conditional probability distribution of 4 m, holds detailed information about micrometer scale
function (CPDF) {() = | Dev ()} from porosity, present in a specific facies type. Figure 4(b) origi-
. nated from a medical CT dataset (200 by 200 by 500 m3 )
Geofluids 5

True image Large-scale training

100 100

Facies 3
26.5%
Facies 3
32.3%
North

North
Facies 2
13%

Facies 1 + 2
67.7%
Facies 1
60.5%

0 0
0 100 0 100
East East
(a) (b)
32 sample data
100

Facies 3
80 28.1%

60
Facies 2
12.5%
40

20
Facies 1
59.4%

0
0 20 40 60 80 100
(c)
SNESim realization SISim realization
100 100

Facies 3 Facies 3
27.3% 27.5%
North
North

Facies 1 + 2
Facies 1 + 2 72.5%
72.7%

0 0
0 100 0 100
East East
(d) (e)

Figure 2: Example of MPS simulation of a meandering river system (after Strebelle [52]).
6 Geofluids

introduced parameter in the simulation workflow can be

interpreted as follows: a small value will assign the property
index to a larger amount of pore pixels around the pore center
(Figure 5(c)), while a large value preserves more details
about the center of the pores itself (Figure 5(d)).
By combining the patterns of the TI and the condition-
ing data, computer-generated rock samples were retrieved.
Hence, this allowed investigating several other petrophysical
parameters such as total and effective porosity, tortuosity,
and permeability. Training images used in this upscaling
workflow do not meet the ideal REV criterion that was
previously mentioned. This option was willingly chosen
because of computational cost, which is linearly related to
the cube of the TI edge length. In the SNESim algorithm,
all patterns are stored in the RAM of the computer and
additional different patterns increase the CPU requirements.
n = 1, nest grid Therefore, an arbitrary size of 2503 voxels has been chosen
n = 2, medium grid for training images. This corresponds to a size of 1 mm3 at a
n = 3, coarsest grid resolution of 4 m. These TI volumes have been subjectively
Figure 3: Example of the template using three multiple grids (after
chosen in zones that are relevant for the simulated properties.
Remy et al. [64]). The difference with the concept of an REV is that the volume
of the REV should theoretically be able to select a random
zone in the sample which should always be representative.
This has now been changed to an operator controlled step,
of the same sample. Medical CT usually incorporates a much which should not influence the ultimate results.
larger sample volume, as inferable from the difference in scale
in Figures 4(a) and 4(b). The medical CT datasets provided 3.5. Lattice Boltzmann Method (LBM) Simulations. Station-
information about the pore network on a larger spatial scale ary, pressure-driven fluid flow through porous media was
than the CT and was used as conditioning data in the simulated in Palabos by imposing a constant pressure gradi-
simulations. Thus, the TI (2D slice in Figure 4(c)), derived ent between the inlet and outlet of the pore network. Instead
from the CT scans, is representative of the nonresolved of solving Navier-Stokes equations, LBM solves the discrete
matrix of the medical CT (2D slice in Figure 4(d)). The TI Boltzmann equation by simulating flow for Newtonian fluids
porosity patterns were used to simulate micrometer scale with collision models. In the applied single-phase fluid
matrix porosity, which is below the resolution of the medical simulations under laminar flow conditions, the permeability
CT. This approach allowed retaining the connectivity of was described by Darcys law (see the following equation):
the pore network over a larger spatial scale, resulting in
more reliable pore network reconstructions (Figure 4(e)) as V
well as more accurate permeability simulations, at least if = , (5)
(/)
samples obeyed the REV criterion. In order to determine
the conditioning data, the assumption that the centers of where is the permeability; is the fluid viscosity; V is the
the larger pores stayed the same and were recognizable in flow velocity; and / is the pressure difference over the
all used datasets was made, that is, in TI, condition data, length of the pore network.
and simulated results, regardless of the resolution. Centers The network geometry was stored in an input or geometry
of the largest pores were obtained by applying the workflow file and was based on a stack of binary images (Figure 6(a)).
described below. In the input file, value zero (blue) was assigned to every
The dataset with the largest voxel size was segmented fluid voxel. Value two (red) described internal rock voxels,
(Figures 4(b) and 5(a)) and the distance map of the pore neighboring only other rock voxels, and value one (green)
facies was calculated (Figure 5(b)). This resulted in a dataset was assigned to boundary voxels that touched both fluid and
in which the center of the pore had the highest value. rock voxels (Figure 6(b)). In the geometry file, fluid pathways
Subsequently, the position of the pore center was calculated had to be sufficiently resolved to avoid nonhydrodynamic
by using a regional maximum algorithm. The approach not effects [25, 29]. This was ensured by stepwise refining of the
only resulted in the position of the pore centers but also binary image stacks. First, pore objects with an equivalent
preserved information about the pore sizes. In order to retain diameter shorter than two times the length of a voxel side
more information about the larger pores in the dataset, a were deleted. In a second step, only the effective porosity, that
threshold distance was introduced, which was calculated as is, all the interconnected pore space at the scanned resolution,
times the maximum distance. was selected.
Only distance map values higher than times the In this study, the standard Bhatnagar-Gross-Krook
maximum distance were replaced by an index representative (BGK) collision operator was applied, together with the
for the property under investigation. The effect of the newly D3Q19 lattice. Flow direction during the simulations was
Geofluids 7

Training image 3D Condition data 3D

(mm)
20

0 (mm)
1 mm
0 20

(a) (b)

Training image 2D Condition data 2D

200 m 1.5 mm

(c) (d)

Computer-generated rock

1.5 mm

(e)

Figure 4: Schematic outline of the MPS upscaling workflow. (a) 3D training image at high resolution and (b) 3D conditioning data at low
resolution. Colors indicate pore connectivity. (c) 2D slice through the TI shown in (a) and (d) through the conditioning data shown in (b).
Pores are given in blue. (e) Computer-generated rock slice, with the conditioning data (white) plotted on top of the generated porosity (blue).
8 Geofluids

Original image
Distance map

(a) (b)

Condition data Condition data

= 0.25 = 0.75

(c) (d)

Figure 5: Workflow of generating the conditioning data used in the MPS approach. (a) The original image. (b) The calculated distance map
based on the original image shown in (a). (c) The generated conditioning data for value of 0.25 and (d) for value of 0.75.

limited to a finite vector set, representing the particle travel inlet and outlet of the pore network. Fluid velocity vectors
directions. In case of the D3Q19 lattice, there are 18 discrete with terminal points that coincided with the fluid/rock
lattice velocities for a fluid particle at rest (Figure 7). Palabos interface underwent a bounce back with no slip boundary
comes with several lattice models, including the D3Q15, conditions. This means that particle displacement from a fluid
D3Q19, and D3Q27 lattices. The D3Q15 lattice was not used node to a solid surface (Figure 8(a)) resulted in a bounce
here because of the less accurate results and numerical back vector with the same initial point and direction but the
instabilities at high Reynolds numbers [56, 57]. The D3Q27 opposite sense (Figure 8(b)). The initial vectors (A, B, and
lattice is more complex, when compared to the D3Q19 lattice. C) and bounce back vectors (D, E, and F) neutralized each
It provides eight additional particle transport directions, other and resulted in a zero velocity for the displacement
namely, to the corners of the cubic lattice nodes. Because towards the solid surface. The standard deviation of the
of this, the accuracy of the model will increase, but so will average energy was calculated while running simulations in
the computational needs per iteration step. Both models Palabos. As also mentioned by Degruyter et al. [25], steady
were tested on the continental carbonates and yielded similar state was reached when the standard deviation of the average
results, in agreement with earlier publications [56, 57]. energy fell below a given threshold value. In this study,
Hence, continuing working with the D3Q19 lattice to limit the threshold value was set to 103 , with a maximum of
the computational cost of the permeability simulations was 20000 iteration steps. Similar to Degruyter et al. [25], it was
decided. checked whether the permeability stayed constant when the
The initial fluid velocity in the simulation was set to applied pressure gradient was varied over several orders of
zero. The simulated flow accelerated during the iteration steps magnitude. This ensured that laminar flow was established
under influence of the fixed pressure gradient between the during the simulations [58, 59].
Geofluids 9

500

250

0 250 500 0 250 500

(a) (b)

Figure 6: Input files, with (a) binary CT image and (b) conversion to geometry file in which fluid voxels are blue (value 0), internal rock
voxels are red (value 2), and boundary voxels are green (value 1).

15 where is the effective length of a voxel side; is the lattice

14 5 viscosity; and V is the fluid flow rate.
11 Permeability was simulated along the length axis (-axis)
18 of the plugs and spatial transformation of the binary images
of nine samples allowed simulating permeability in the hori-
zontal and directions, orthogonal to -axis. Permeability
10 3
7
measurements were carried out on CT-derived pore networks
2
and rock models generated in the MPS approach with cubic
1 (500 px3 ) or cuboid volumes of 500 by 500 by 1000 pixels or
8 800 by 800 by 450 pixels, respectively.
4 9

3.6. Tortuosity. The porosity in rocks is defined by the

17 amount of void spaces present in a sample. These pores
12 could be connected in permeable pathways. Then, complexity
6 13
(sinuosity) of the pathways for fluids through these pores
16 is described by a property known as the tortuosity. This
property provides information on the interconnectedness of
Figure 7: D3Q19 lattice with 18 discrete velocity directions for a
the pore objects as part of a pore network and is used in
particle at rest.
transport models for porous media [6163]. The tortuosity ()
is defined by the length of the flow path ( ) and the shortest
During the LBM simulations in Palabos, dimensionless trajectory ( ) between a defined inlet and outlet plane in the
lattice units are used to describe fluid flow properties, which direction of the applied pressure gradient (see the following
are easily converted to any kind of physical system [57, 60]. equation, [63]):
In this study, the dimensionless lattice units were converted
to Darcy units by multiplying with the square of the effective = lim . (7)

length of a voxel side. Knowing that the permeability in
Darcys law is proportional to the ratio between fluid flow The tortuosity equals 1 for a straight path through the sample
rate and the applied pressure gradient () between the inlet ( = ) and will be infinite for a cyclic path ( ).
and outlet () of the pore network, the permeability () was
described as follows: 4. Results
2
V The results below are presented in order of the subsequent
= , (6)
(/) steps in the simulation approach. First, the LBM method
10 Geofluids

D E
F

A B C

Before particle displacement After particle displacement

(a) (b)

Figure 8: Bounce back conditions of fluid particles for a node at the fluid/rock interface. (a) Colored vectors are streamed to neighboring
nodes and black vectors towards the node at the fluid rock interface. Particles travelling into the solid surface are represented by vectors A,
B, and C. (b) Black vectors of the neighboring nodes have been streamed to the node at the fluid/rock interface and, correspondingly,
colored vectors haven been streamed to neighboring nodes. Vectors A, B, and C cannot be streamed towards a neighboring node, since
particle displacement is interrupted by the solid surface. The latter vectors were therefore bounced back according to vectors D, E, and
F with, respectively, the same directions but in the opposite sense.

simulations were conducted on pore networks derived from 4.2. Representative Elementary Volume. Pore REVs of ten
natural samples and the effect of the scan resolution on CT-scanned plugs (3.4 cm diameter) at 16 m resolution
CT based porosity and simulated permeability was checked. were calculated (Table 1). For all the samples, the REV was
Subsequently, MPS was used to integrate high-resolution reached. The QQ plots of 100 REV simulations per sample
porosity patterns into lower-resolution, large-volume CT (not shown) had determination coefficients of at least 0.94.
scans. LBM simulations were used to calibrate several MPS The value of the chi-square goodness of fit test was always
parameters and to verify whether the generated pore net- well above the 0.05 hypothesis threshold. The latter tests
works are a good representation of the real pore system. implied that the null hypothesis was never rejected, that
the calculated REVs were lognormally distributed, and that
4.1. Spatial Resolution Effect on the Recorded Pore Network. A upper and lower bounds of the 95% confidence intervals for
comparison was made between datasets obtained by medical the mean size of the REV could be calculated.
CT with 230 m resolution (Figures 9(a) and 9(e)) and the The confidence intervals for the side length and volume
HECTOR scans with 28 m resolution (Figures 9(b) and 9(f)) of the cubic REV are listed in Table 1. The scanned sample
of two different continental carbonate facies types: a reed volumes were all representative for the pore types and sizes
facies and a subhorizontal sample. It is clear that much more that they contained. The largest REVs were found in samples
detail is visible in the HECTOR scans. Figures 9(c) and 9(d) originating from the reed lithofacies, which is in agreement
show the comparison of the visible porosity in both scans with their heterogeneous pore networks that were observed
of both facies types. The mean porosity difference between by Claes [49]. The smallest REVs were observed in the sub-
both scans of the reed facies type was 1.5% (Figure 9(c)). The horizontal and cascade facies. The pore network had a more
subhorizontal facies type was characterized by a larger mean homogeneous distribution throughout these lithofacies. The
porosity difference of 2.4% (Figure 9(d)). This was explained volumes over which the permeability was simulated were
by pore volumes below the resolution of medical CT, which chosen with the size of the REVs kept in mind. Simulation
were more abundant in the latter facies type. cell volumes for samples from the reed facies, for example,
In order to assess the influence of resolution on smaller which was the facies with the largest REV size, were always
samples, a plug with 7 mm diameter was scanned at 3 different 1180 mm3 , while for the subhorizontal facies, which in general
resolutions: 4 m (Figure 10(a)), 12 m (Figure 10(b)), and had a smaller REV size, simulation cells of 512 mm3 could be
16 m (Figure 10(c)). Figure 10(d) shows a comparison of used. The simulation cells were always at least over two times
the calculated porosity in function of plug height. The the REV volume.
porosity difference between the 4 m and 16 m scan varied
between 1.4 and 2.35% along the height of the sample. 4.3. Permeability Simulations. In total, 18 VOIs were prepared
These observations illustrated the importance of resolution in from continental carbonate plugs. The simulated perme-
correctly characterizing the pore networks. Below, a workflow abilities along the vertical and orthogonal horizontal axes
is proposed to combine information of different datasets are given together with other sample characteristics, such
which have a different resolution in order to improve the as lithofacies types and CT calculated porosities (Table 2).
accuracy of larger samples. The permeabilities obtained from LBM simulations are in
Geofluids 11

Reed continental carbonate Reed continental carbonate

(a) (b)
50 50

40 40
Core height (mm)
Core height (mm)

30 30

Reed facies Subhorizontal facies

20 20

10 10

0 5 10 15 20 0 5 10 15 20 25 30
Porosity (%) Porosity (%)

Hector CT Hector CT
Medical CT Medical CT
(c) (d)

Subhorizontal continental carbonate Subhorizontal continental carbonate

(e) (f)

Figure 9: Influence of the resolution on the measured porosity network in whole cores with a diameter of 9.7 cm. ((a) and (b)) Medical CT
and HECTOR slice of a reed continental carbonate with 230 m and 28 m resolution, respectively. (c) Comparison of the calculated porosity
in function of the core height for a reed continental carbonate. (d) Comparison of the calculated porosity in function of the core height for
a subhorizontal continental carbonate. ((e) and (f)) Medical CT and HECTOR slice of a subhorizontal continental carbonate, with 230 m
and 28 m resolution, respectively.
12 Geofluids

4 m 12 m

(a) (b)

3
Core height (mm)

1
16 m

0 5 10 15 20 25
Porosity (%)

4 m
12 m
16 m
(c) (d)
Figure 10: Influence of the resolution on the measured porosity network in a miniplug with a diameter of 7 mm. Comparison of the same
slice scanned at (a) 4 m, (b) 12 m, and (c) 16 m resolution. (d) Comparison of the calculated porosity in function of the plug height.

Table 1: REV size calculations.

REV side in mm REV volume in mm3
Sample Facies QQ correlation value
Lower 95% CI Upper 95% CI Lower 95% CI Upper 95% CI
1 Subhor. 0.99 0.45 1.58 2.62 3.96 18.01
2 Subhor. 0.97 0.81 2.42 4.94 14.26 120.53
3 Subhor. 0.95 0.75 1.10 5.63 1.31 178.57
4 Subhor. 0.97 0.67 1.14 3.94 1.47 61.26
5 Reed 0.96 0.44 1.72 4.56 5.13 95.09
6 Reed 0.97 0.59 2.78 8.26 21.55 562.70
7 Reed 0.94 0.36 3.83 5.95 56.24 210.17
8 Cascade 0.97 0.18 1.96 4.71 7.56 104.73
9 Cascade 0.98 0.85 2.26 3.74 11.54 52.12
10 Waterfall 0.99 0.67 1.45 3.19 3.02 32.58
Note. With subhorizontal facies (Subhor.), quantiles (QQ), the statistical value, and the confidence interval (CI).
Geofluids 13

Table 2: Permeability simulations.

Simulation Facies sim, (mD) sim, (mD) sim, (mD) (%) , (%)
1 Subhor. 0 1456 28.6 9.2 0 /
2 Subhor. 0.1 14.8 99.5 6.2 5.0 4.9
3 Subhor. 0 / / 0.8 0 /
4 Subhor. 0.4 22.0 9.6 3.4 1.7 3.3
5 Subhor. 1.8 / / 7.2 4.8 2.9
6 Subhor. 100 / / 12.3 11.7 2.4
7 Cascade 0.3 / / 7.1 5.2 2.2
8 Cascade 0.7 11.7 0.5 7.9 7.1 3.3
9 Cascade 0 / / 5.3 0 /
10 Reed 3.6 / / 6.9 4.5 3.1
11 Reed 449 30780 17641 25.9 23.5 2.3
12 Reed 422 / / 25.8 23.5 2.3
13 Reed 1.9 0 0 6.5 3.7 1.9
14 Reed 2325 0 179 12.2 1.7 1.6
15 Reed 62261 / / 14.4 12.4 1.4
16 Waterfall 10.9 / / 14.1 11.2 2.9
17 Waterfall 1931 84.3 178 32.7 32.0 1.5
18 Waterfall 1.0 0 0 11.1 4.6 2.1
Note. With vertical permeability simulations (sim, ) and orthogonal horizontal permeability simulations (sim, and sim, ), the CT calculated porosity (),
the connected porosity in the direction (, ), the tortuosity in the direction ( ), and not-measured values or values that could not be determined (/).

good agreement with physical laboratory measurements 100000

conducted on continental carbonate samples (Figure 11). The 10000
laboratory core analyses were conducted on 5 cm3 or 10 cm3
cylindrical plugs, while volumes analyzed with LBM varied 1000
between 0.5 cm3 and 1.2 cm3 .
Permeability (mD)

100
Despite the volume difference in order of magnitude, a
similar spread for porosity and permeability was observed for 10
the core measurements and LBM simulations, with the latter
1
plotting along the best fit curve to the core measurement data.
The laboratory plug measurements include the data published 0.1
in Soete [22], which indicated that the subhorizontal facies
0.01
data plot mainly below 15% porosity and 100 mD. This is in
agreement with the findings for both vertical and horizontal 0.001
permeability in this study, where only one simulation for the 0 10 20 30 40 50
subhorizontal facies yielded permeability > 100 mD and a Porosity (%)
median of 12 mD was found, while never exceeding 12.3% CT porosity and simulated permeability
porosity. Core porosity and permeabilty measurements
For the cascade facies, phyto-rich samples were avoided
and the simulations focused on samples containing shrub Figure 11: Porosity-permeability cross-plot for laboratory core
crust lithotypes. Dominant cascade pore types were therefore measurements and LBM simulations. The black line is the best fit
similar to those observed in the subhorizontal facies and curve to the data and the dashed lines delineate the zone defined by
90% of the data points closest to the best fit curve.
yielded simulation results that were in the same range.
The reed and waterfall facies were both characterized by
a strong increase in calculated CT porosity. In the reed
facies, this increase was due to the presence of reed moldic added to the dataset of this study. Waterfall pore types
porosity [1, 21], which formed elongated tubes that locally captured in the VOIs possessed grass and bryophyte moldic
ran through the simulation volumes. These tubes provided pores that together formed plug scale framework porosity.
high permeability pathways. The simulated permeability in Framework porosity, formed by plants larger than grass
the reed facies exceeded core lab results from continental and bryophytes, was not included in this study. This explained
carbonates from the Ballk area (Turkey), published in Soete the somewhat lowered permeability results for the LBM
[22], but that was expected, since highly permeable reed approach when compared to laboratory-measured perme-
samples from Sutto and Budakalasz in Hungary [42] were ability for the waterfall facies [22]. The similarities between
14 Geofluids

Table 3 scanning equipment. An MPS workflow (see Section 3.6)

that integrates high-resolution pore network details into large
(a) Effect of resolution and volume on permeability (16 m to 4 m)
volume datasets and that captures the variability of the pore
Sample 16 m 4 m VOI16 m VOI4 m network can provide a solution and is thus proposed in this
1 1581.7 2215.4 1180 mm3 18.4 mm3 paper.
2 22 112.8 1180 mm3 18.4 mm3 The total porosity () and connected porosity in
3 34.8 2.8 1180 mm3 18.4 mm3
direction (, ) were calculated from CT-scanned samples.
For some samples, , was slightly smaller, which meant
4 100.5 10.7 1180 mm3 18.4 mm3
that the majority of recorded pore objects in the CT scans
(b) Effect of changing resolution on permeability (4 m to 2 m) contributed to the connectivity. Other samples displayed
Sample 4 m 2 m VOI4 m VOI2 m large differences between total and connected porosity, for
example, in simulation 14 (12.4 versus 1.4%). This indicated
5 63.8 60.4 2.3 mm3 2.3 mm3
pore network heterogeneity and isolated porosity in the
6 15.4 15.7 2.3 mm3 2.3 mm3 simulated flow direction at the resolution of the acquired
Note. 16 m , 4 m , and 2 m represent simulations at resolution of 16, 4, CT images. The simulated permeability along -axis (sim, )
and 2 m. Table 3(a) shows effect of resolution and volume on permeability
of the samples was plotted against and , (Figure 12).
and Table 3(b) shows difference for simulations at 2 and 4 m resolution
over identical volumes. Porosity was in this case plotted on the vertical axis of
the graph to clearly show the lowering from to , . A
power-law relationship was observed between sim, and , ,
measured and simulated permeabilities lead to the verifi- with a determination coefficient of 0.75. This implied that
cation that the analyzed tomographic volumes qualify as good permeability estimates could be obtained from CT
permeability REVs. The CT scans were thus at high enough porosity calculations at 16 m resolution. However, it has to
spatial resolution to capture the primary permeability con- be kept in mind that pores below the 16 m resolution could
tributors of the pore network. influence flow properties in continental carbonates but were
Miniplugs (7 mm diameter) were taken from some of the not considered in these permeability estimations. Two high-
samples to further investigate the effect of sample size and permeability outliers with relatively low porosities can be
spatial resolution on simulation results. Miniplugs were CT- observed and will be further discussed in the next section
scanned at 4 m resolution and permeability was simulated (Section 4.4).
over a volume of 18.4 mm3 , which was in most cases below the
calculated pore REV of the samples. It was shown in Figure 10 4.4. Tortuosity. The tortuosity () was calculated and plot-
that porosity increased for a resolution step from 16 to ted against permeability (Figure 13). Tortuosity controlled
4 m. Based on the increasing porosity, higher permeabilities permeability in the studied complex pore networks. The
were expected for scans at 4 m resolution. Despite the observed power-law relationship between and sim, was
higher porosity and sharper CT images, the volumes of consistent with findings of studies treating volcanic rocks,
4 m scans were too limited to capture the variability of where increasing tortuosity results in lower permeabilities
the pore network. The 4 m miniplug simulations (no REV [25, 26, 65]. Highest tortuosities were reported for the
reached), although within the order of magnitude, yielded subhorizontal and cascade facies, while the waterfall and reed
permeabilities that overestimated and underestimated the facies had the lowest tortuosities (Table 2). Tortuosity of the
16 m plug simulations, in which the REV was reached pore network helped to understand discrepancies between
(Table 3(a)). porosity and permeability. Two outliers from the porosity-
In addition, two miniplug VOI scans were conducted simulated permeability regression line (sample 14 and 15
at 2 m resolution, the highest achievable resolution for the in Figure 12) were highlighted in the tortuosity-simulated
Phoenix Nanotom S instrument in these rocks. The 2 m permeability cross-plot (Figure 13). Both samples are part of
VOI scans were projected back in their respective 4 m the reed facies and , consists of only a few reed molds.
miniplug scans. Permeability simulations for the latter VOIs The limited, patchy reed mold presence did not result in
were not expected to be representative, since both 2 m and high overall porosities for the samples. Despite the limited
4 m scans were based on volumes below the REV size for porosity, both samples are characterized by low tortuosities,
porosity. The latter simulations were purely conducted to indicative for the straight, high velocity flow paths that the
demonstrate the effect of spatial resolution. The simulations reed molds provide through the sample.
that were conducted over the exact same volumes but at
different spatial scales yielded similar results (Table 3(b)). The 4.5. Simulating Permeability in Orthogonal and Directions.
latter demonstrated that an increase in resolution from 4 For nine samples, including four from the subhorizontal
to 2 m did not significantly improve the accuracy of the and cascade facies and five from the reed and waterfall
digital pore network. Based on the above observations, it was facies, the permeability was also simulated in the horizontal
concluded that the best digital representations of the pore and directions, orthogonal to the vertical simulations
network were in this case achieved in 4 m resolution CT along -axis, totaling 18 horizontal simulations. These oper-
scans but with VOIs that equal at least VOI16 m . Achieving ations were limited to nine samples because of the high
4 m resolution while scanning large VOIs is at this point computational cost of the rotation and simulation proce-
impossible and would need significant improvement of CT dures. The simulations sim, , sim, , and sim, for the nine
Geofluids 15

40 Subhorizontal &
100000 cascade Reed & waterfall
35
10000
30 1000
100
25
Porosity (%)

Equation: y = 5.39x0.21 10
20 Determination coefficient: 0.75 1

Sample 15
15 0.1 Z
X
0.01 Y

Sample 14
10
2 outliers 0

SU18

Al18

BU02

EC30

Al09

FA155

BU11

CK31

CA31
5

0
0 0.01 0.1 1 10 100 1000 10000 100000 Figure 14: Simulated permeability sim, (red), sim, (green), and
Simulated permeability, K,z (mD) sim, (blue) for four subhorizontal and cascade samples and five
reed and waterfall samples.
Calculated porosity ()
Connected porosity (c,z )

Figure 12: Cross-plot of the calculated total and connected CT rock fabrics, which formed barriers and resulted in low
porosity (resp., blue and red symbols) versus the permeability. vertical permeabilities (<2 mD). A pore network volume
rendering of a subhorizontal facies sample, which included
gastropod moldic pores (Figure 15(a)), demonstrated this low
100000
Sample 15 connectivity in direction. The alignment of the pores along
10000
the laminations provided better horizontal connectivity.
Sample 14 For the reed facies, permeability was strongly dependent
on reed moldic pores. The 3D visualization of flow paths
Simulated permeability (mD)

1000
inside a reed sample (Figure 15(b)) demonstrated the hetero-
100 geneous nature of fluid flow through these porous media.
High velocities were achieved within only a couple of reed
10 tubes, which dominated the fluid flow. Depending on the
orientation of the reed molds, that is, eroded (horizontal)
1 or in-growth position (vertical), the direction of the highest
permeability differed; for example, in sample Al09, which
0.1
contained eroded reed, the highest permeability was achieved
in the horizontal direction (Figure 14).
0.01
Equation: y = 86242x9.378 Grass and bryophyte framework pores in the waterfall
0.001
Determination coefficient: 0.66 facies formed vertically oriented pore networks. The small
1 2 3 4 5 6 framework pipes yielded high permeabilities in the vertical
Tortuosity () direction, while the horizontal connectivity was dependent
on the connections between individual pipes. The hori-
Figure 13: Tortuosity-simulated permeability cross-plot. A moder- zontal permeability was therefore lowered in samples from
ate power-law relationship is observed between these parameters.
the waterfall facies, for example, samples CK31 and CA31
(Figure 14). The streamline rendering (Figure 15(c)) shows
the low tortuous path in direction, with relatively straight
samples were plotted and the influence of different facies connections between the bottom and top plane of the sample.
types and related pore types on the results was checked
(Figure 14). For the subhorizontal and cascade facies, the 4.6. Permeability Simulations on the MPS Generated Rock
highest permeabilities were observed in the horizontal direc- Models. In Section 3.3, the spatial resolution effect on sim-
tion, often being over an order of magnitude higher than ulated permeability on the scanned sample set was shown.
the vertical permeability of the same sample. Even in sample The results demonstrated that (1) high resolution scans were
Al18, where an open network in the vertical direction was needed to capture micrometer scale pores which impacted
absent, high horizontal permeabilities were encountered. flow properties and that (2) lower-resolution CT scans should
Pseudofenestral, interpeloidal, interlayer, and shelter porosi- be performed on sufficiently large volumes to fully capture
ties in the subhorizontal and cascade facies were described the heterogeneity of the pore network. In order to overcome
as pore types that appear aligned with the horizontal or the volume versus resolution problem of CT scanners, MPS-
inclined laminations. In the vertical direction, these pore generated rock models are used as input for permability
types were usually discontinuous and overgrown by younger simulations. The latter allowed assessing the applicability of
16 Geofluids

500

800 800
250

400 400

0
Z
X Y
CT scan resolution: 16m
(a)
500
500

250 250

Z Z

0 250 500 X 0 400 800

Y X Y
6 6
7.7e5 4e5 0 7.4e 4e 0

Velocity magnitude Velocity magnitude

(b) (c)

Figure 15: (a) Pore network volume rendering of a sample from the subhorizontal facies (800 by 800 by 500 pixels). The majority of the
pores are aligned horizontally. Notice that some gastropod molds are present (see arrows). (b) 3D streamline velocity distribution within a
sample from the reed facies, with preferential flow in direction, along vertical tubes, from = 0 to = 500 pixels. (c) 3D streamline velocity
distribution within a sample from the waterfall facies, with preferential flow in direction from = 0 to = 500 pixels. The velocity magnitude
is given in nondimensional lattice units.

the MPS workflow in generating larger-volume rock models conditioning dataset had a resolution of 16 m. Hence, the
at higher resolutions. resulting simulated dataset contained the volume of the larger
In order to assess the influence of the above-introduced 16 m CT scans, with a higher 4 m resolution. The TI
conditioning data MPS parameter on the rock models, had a size of 2503 pixels or 10003 m. The search template
a series of simulations were performed using different distance is 25 pixels. The servo corrector was kept zero and six
values (Figures 16(a)16(d)), while keeping all other param- multigrids were used. The TI volume was chosen with extra
eters equal. The simulations were performed on continental care in order to represent the typical patterns or in this case
carbonate samples from the reed facies. This facies type pore structures. The resulting simulated samples are shown
was chosen because of its typical abundant open porosity in Figure 16. The shape of the pores corresponded well with
network and the characteristic elongated shape of the pore pore shapes that were present in the TI. Rod- and blade-
bodies. The latter resulted in permeability values of 50 mD shaped pores made up 50% of the pores. Because permeability
and higher [41]. The TI had a resolution of 4 m and the was measured in the direction of the aligned pore shapes, a
Geofluids 17

permeability value of around 1000 mD was expected based on doubled compared to the first set of simulations. This allowed
petrophysical measurements indicated in Figure 11. simulating artificial samples at a larger spatial scale with
Analyzing the shape of the pores, according to the a resolution of 28 m. Because of the porosity difference
classification of Claes et al. [66], provided an excellent tool between the TI and the generated rock sample, a servo
for the quality control of the simulation. Similar pore shape factor of 0.75 was used. If a lower value was used, the
occurrence percentages between the physical sample to be porosity value of the simulated rock samples became too
simulated and the MPS simulated sample were indicative high, resulting in visually unrealistic models and unrealistic
of good quality simulations. The results also indicated high simulated permeability values. For example, when the reed
dependency on conditioning data in order to retain the sample (Figure 18(c)) was simulated using a servo factor of
desired larger-scale connectivity in the simulated sample. The 0, the calculated permeability became 22 D. This observation
connectivity of the pore network had a direct influence on the illustrated the importance of the servo factor, which should
simulated permeability values of the simulated rock samples. be taken higher when the resolution difference increases. A
The permeability values increased when more conditioning clear difference between the simulated porosity networks of
data, that is, lower values, were used in the simulation. Sim- both the reed and subhorizontal facies types was observed.
ulated permeability values were in the expected range for reed The moldic reed pores were retained in the simulation. In
continental carbonate samples, but the most representative contrast, typical sequences of porous and less porous layers in
permeabilities were found for = 0.25, that is, the largest subhorizontal samples were observed in simulated samples.
amount of conditioning data. Large amounts of conditioning The simulation-obtained permeabilities were in the
data were needed to simulate complex carbonate rocks. expected order of magnitude compared to measured values.
In order to further test the applicability of the workflow, Moreover, the anticipated difference between the subhorizon-
three adjacent volumes of a sample from the subhorizontal tal and reed facies type pore networks was clearly apparent.
facies were simulated using a TI (2503 pixels) with a resolu- In the subhorizontal sample, no open network was present
tion of 4 m. Figure 17(a) depicts the conditioning data (1600 in direction ( shown in Figure 18(a)). Figures 18(b) and
1600 800 pixels) used in a simulation of this facies type, 18(d) show the simulation results in which the same TI
which had a resolution of 16 m and value of 0.25. Three was used as in Figures 18(a) and 18(c), respectively, but the
different zones were generated using the same TI and the conditioning data were derived from a different sample of a
results (8003 pixels) were depicted in Figures 17(b), 17(c), and similar facies type. Simulations with the same TI but different
17(d), respectively. CD demonstrated that the quality of the HECTOR based TI
The TI was selected inside a porous layer characteristic was sufficient to be used on multiple samples.
for this facies type. The resulting subvolumes had a side
length of 3.2 mm and hence depicted only one sedimentary 5. Discussion
lamina. This explained why the typical horizontal layering of CT scanning and computational fluid dynamics (CFD)
these samples, with porous and less porous layers, was not proved to be excellent tools for the investigation of flow
clearly visible in the simulated volumes. They would only properties in complex continental carbonate pore networks.
have become visible when the spatial scale of both the TI Different flow properties and paths were observed for differ-
and simulation grid was increased. The resulting simulated ent facies types and their characteristic pore types. Consistent
permeability values ranged from 0 mD, in case no connected with De Boever et al. [16] and Claes et al. [66], the permeabil-
network was present, to 203 mD. Hence, the permeability ity was higher for samples from the reed and waterfall facies.
values were highly dependent on the provided conditioning The subhorizontal and cascade facies were characterized by
data, as was already demonstrated in the reed facies samples lower permeabilities and often yielded higher permeabilities
in Figure 16. Consistent with the laboratory permeability horizontally in contrast to the vertical direction.
measurements, the overall values were lower compared to the
The main factors influencing the permeability in conti-
reed facies (1367 mD for = 0.25, Figure 16(d)). The lower
nental carbonates were connected porosity (, ) and tortu-
permeabilities in the vertical direction for the subhorizontal
osity (). For both variables, a power-law relationship with
facies were in line with the findings in Section 3.5, where
the permeability was established. However, the determination
permeability was simulated in horizontal and vertical direc-
coefficients (0.75 and 0.66, resp.) revealed data scatter; hence,
tions. The simulated rock models preserved the characteristic
permeability predictions based on either of these variables
laminated pore structure of the subhorizontal facies, with
alone were not very accurate. and , counteracted
limited connectivity in the vertical direction. each other at several occasions, like when porosity would
The unique datasets generated by the HECTOR scans overestimate simulated permeability of a sample and when
provided an excellent case study to evaluate the potential of tortuosity would underestimate it and vice versa. Therefore,
the proposed workflow. Because of their detailed resolution more accurate permeability predictions for a given sample
(28 m), HECTOR scans delivered excellent TI. The medical were obtained by using (8), which includes both , and :
CT scan data provided the larger-scale conditioning data
and had a resolution of 230 m in and directions and = 431219.4 + (1.384 ) , .
0.2056
(8)
500 m in direction. This allowed performing petrophysical
simulations on core samples. In this case, the difference A cross-plot of predicted versus LBM-simulated permeability
in TI resolution and conditioning data was more than (determination coefficient of 0.85) illustrates how and
18 Geofluids

K = 0.3 mD K = 43 mD

1 mm
1 mm

1 mm 1 mm

(a) (b)
K = 190 mD K = 1367 mD

1 mm
1 mm

1 mm 1 mm
(c) (d)

Figure 16: Computer-generated models of a reed facies sample with different values, which controls the level of conditioning data used. (a)
No conditioning data; (b) = 0.75; (c) = 0.50; and (d) the largest amount of conditioning data with = 0.25 (colors indicate connectivity). The
more the conditioning data approaches the truth, the more the simulated permeability approaches the laboratory measured permeability.

, interact in the determination of the permeability of size. In general, with increasing sample size, new porosity
the samples under investigation (Figure 19). LBM-simulated REVs would be encountered whenever new large-scale pore
permeability was slightly overestimated by the predicted types would be introduced. Macroscopic observations also
permeability at low values. The predicted values tend to showed that large-scale framework, cavern, and vug porosity
eliminate the most extreme values, limiting the 0.110000 mD were mainly interconnected through the pores encountered
simulated permeability range to 0.354500 mD in the estima- within the studied plugs. The main pore scales governing
tions. connectivity and fluid flow at the scale of the carbonate body
The REV measurements and CFD simulations demon- were therefore expected to be included in this study.
strated the effect of sample scale and spatial resolution on In general, the following can be concluded: the larger the
the permeabilities. Claes (2015) confirmed that different REV CT analyzed volume and the finer the scan resolution, the
sizes can be found at different scales, corresponding to more realistic the representation of the pore network. The
different geological length scales. Here, only pores below MPS workflow proposed in this manuscript allows generating
plug size were considered in REV analyses. For this pore artificial rock samples which can be used to investigate
scale, the porosity REV was reached in all 16 m VOIs, over the influence of rock heterogeneity on fluid flow dynamics.
which permeability simulations were performed. Large-scale Moreover, this technique permits bridging the gap between
framework pores, caverns, and decimeter-sized vugs were, different datasets with different resolutions. The workflow
however, also encountered in the field but do not form a was used on datasets where the resolution of the conditioning
part of this research. Increasing the sample scale to include data was four and eight times the resolution of the TI. In both
decimeter-sized vugs would drastically increase the REV cases, the simulations yielded realistic results.
Geofluids 19

No open network

Condition data
D

B C

1 mm
1 mm
(a) (b)
K = 77 mD K = 203 mD

1 mm
1 mm

1 mm
1 mm

(c) (d)

Figure 17: Computer-generated models of a subhorizontal facies sample with = 0.25. (a) Conditioning data, with three different subvolumes
B, C, and D indicated. (b) Simulated rock model based on the conditioning data in subvolume B; (c) simulated rock model based on
the conditioning data in subvolume C; and (d) simulated rock model based on the conditioning data in subvolume D. The same TI is
used in each simulation. Similar colors indicate pores that are connected.

The amount of conditioning data, determined by the 6. Drawbacks and Future Perspectives
value, and the use of the servo factor proved to be impor-
tant parameters to obtain realistic simulations. Comparing As with all modelling techniques, there are some drawbacks
measured and simulated permeabilities showed that = to this research even though efforts are made to reduce
0.25 yielded the most realistic pore networks models for the them. In this paper, the choice for fluid flow modelling
investigated continental carbonate samples. Large amounts of techniques and upscaling approaches is based on the need for
conditioning data were necessary to model pore networks of correctly representing the porous network. This is achieved
continental carbonate samples in a representative way. The by avoiding simplifications as much as possible. In first
MPS workflow was applied to upscale from miniplug (diam- instance, this is evidenced by the chosen CFD method. The
eter 7 mm) to plug (diameter of 3.81 cm) and to large core LBM approach is chosen as it offers a good trade-off between
volumes (diameter of 9.7 cm), but Zhang [67] also proved the computational power and representativeness. In LBM, the
applicability of MPS datasets on larger spatial scales such as fluids are represented as small packages in individual voxels
borehole imaging and facies distribution models in reservoir rather than as individual molecules. The actual pore network
modelling. is used in the simulations rather than simplifications that pore
The LBM simulations for generated rock models are topology models typically use. In order to obtain realistic
plotted on top of the measured relationship between porosity permeability simulations, representative pore networks are
and permeability for continental carbonate plug samples preferred, even though simplifications would decrease the
(Figure 20). The simulated volumes were indicated in red. The computational cost drastically.
simulated values follow the general trend and are therefore Secondly, the choice was made to improve resolution of
considered to be realistic. spatially larger models rather than simplifying the problem
20 Geofluids

K = 455 mD No connected network

Subhorizontal

Same TI
New sample
10 mm

10 mm
10 mm 10 mm
(a) (b)

K = 4172 mD K = 1613 mD
Reed

Same TI
New sample 10 mm
10 mm

10 mm 10 mm
(c) (d)

Figure 18: Computer-generated models based on a TI of a HECTOR scan (resolution of 28 m) and conditioning data based on medical CT
data. (a) Subhorizontal sample with TI and conditioning data based on the same sample. (b) Subhorizontal sample with TI and conditioning
data based on a different sample. (c) Reed sample with TI and conditioning data based on the same sample. (d) Reed sample with TI and
conditioning data based on a different sample. The colors indicate pore connectivity.

100000 through classical upscaling techniques that smoothen out

small-scale variations. This choice is made based on the ratio-
LBM-simulated permeability (mD)

10000 nale that small-scale features, variations, and heterogeneities

hold the information that can determine large-scale reservoir
1000 properties. This rationale is not only relevant in, for example,
unconventional reservoirs, but also in carbon capture and
100 storage (CCS), in which the smallest pores are used for
capillary capture of fluids.
10 The use of multiple-point statistics in upscaling is still a
relatively new approach and as such a methodology which is
1 not yet perfected. Known issues include the determination of
Equation: y = 0.3402x1.1321
Determination coefficient: 0.85
input parameters in the SNESim algorithm and its compu-
0.1 tational requirements. A lot of parameters, like the number
0.1 1 10 100 1000 10000
of multigrids, search template size, and parameter, have
Predicted permeability based on equation (8), to be adjusted in the SNESim algorithm, which introduces
including c,z and (mD)
subjectivity into the models. To avoid subjectivity as much
Figure 19: Predicted versus simulated permeabilities. The predicted as possible, the workflow is here calibrated with physical
permeabilities are based on an average between the permeability laboratory experiments. Visual inspection is always applied
calculated from tortuosity and connected porosity for each sample. on the results to ensure that (1) a good fit between simulated
Geofluids 21

100000 input data such as training images or conditioning data could

nevertheless be created through photogrammetry or LiDAR
10000
[68, 69] or through methods that use 2D input data like field
1000 pictures to generate 3D volumes, similar to Okabe and Blunt
[35]. Generating models that include macropores observed at
Permeability (mD)

100 outcrop scale, that is, vugs, framework porosity, caverns, and
fractures, can significantly increase the sensitivity towards
10
reservoir scale fluid flow. Outcrop scale and plug to core
1 scale models, however, have to be considered separately,
since incorporating all elements of a multiscale pore system
0.1 would make the models extremely large and computationally
expensive.
0.01 The future of MPS-driven upscaling techniques should
0.001
aim to reduce subjective parameters as much as possible.
0 10 20 30 40 50 Both this research and the research by Zhang [67] indicate
Porosity (%) that MPS is a potent method in upscaling of reservoir
properties, but neither of these approaches provides the most
MPS-generated rock porosity and simulated permeability
solid solution. For fluid flow experiments on increasingly
Core porosity and permeabilty measurements
larger models, it might be required to use intensive parallel
Figure 20: Porosity-permeability plot of continental carbonate computing, apply modified versions of the LBM, or use pore
samples: Palabos permeability simulation results on computer- network models with microlinks [24] in order not to nullify
generated rock volumes are indicated in red. the increased resolution.

7. Conclusion
properties and physical measurements is obtained and that
(2) models are geologically relevant. Transport properties are fundamental in the characterization
Apart from visual inspection, it is observed that the - of reservoir bodies. In this study, computed tomography
parameter for conditioning data is best kept constant for of continental carbonate samples and Lattice Boltzmann
specific upscaling problems. A -parameter of 0.25 is an ideal Method (LBM) permeability simulations were applied to
value when the resolution for the TI and conditioning data obtain 3D quantifications of the pore network and flow
differs with a factor of 4. A -parameter of 0.75 proved to be paths. Working with 3D datasets was an absolute necessity to
more appropriate for a factor of 8 resolution difference. This understand flow in complex continental carbonates, because
difference in the -parameter is logical given the difference accurate estimates of properties which govern fluid flow
in the amount of overlap in the samples. Another drawback in a rock, that is, connectivity, tortuosity, and so forth,
is that even when applying the -factor, it sometimes occurs could not be obtained from 2D observations. The simulated
that new larger pores are unnecessarily generated. The larger permeabilities were in good agreement with physical plug and
pores are, however, expected to be sufficiently represented in core permeabilities. Power-law relationships were observed
the lower-resolution scans. The training images are chosen between permeability and connected porosity on one hand
at 2503 voxels, which is smaller than the REVs, to limit the and permeability and tortuosity on the other hand. Estimates
computational cost. The selection of these smaller training of rock permeability for a sample were strongly improved
images was user-controlled in order to select highly repre- by including both tortuosity and connected porosity. Both
sentative zones. The quality of the TI therefore will not be parameters balance one another, and so their integration
compromised by the smaller size. allows better prediction of the fluid flow through the complex
The final drawback of any upscaling technique that porous network of continental carbonate samples.
aims to increase resolution, while also increasing scale, is The subhorizontal and cascade facies were associated
computational cost. For example, increasing the resolution with pseudofenestral and interpeloidal pores and yielded the
by a factor of 4 increases the required storage space by a lowest porosities and permeabilities. The reed and waterfall
factor of 43 . Such volumes might easily exceed the abilities facies, with, respectively, reed moldic and framework poros-
of standard workstations to model fluid flow. Because of the ity, were characterized by much better reservoir properties.
computer power limitation, this research focused on CT and The resulting porosities easily surpassed 30% and exceeded
HECTOR scans, which yield sample volumes smaller than 50 D, in both measured and simulated permeabilities. Facies
the outcrop scale macropores that are regularly present in and pore type-dependent heterogeneity was observed. Better
carbonates. Such oversized pores potentially influence large- horizontal connectivity and permeability for the subhorizon-
scale fluid flow and can thus be important components in tal and cascade facies were in contrast to the generally higher
reservoir models. Even though these pores have not been vertical permeabilities that were observed for samples from
treated here, they could be incorporated into models with the reed and waterfall facies.
the proposed reservoir scale upscaling purpose, as it is The proposed multiple-point geostatistics (MPS) work-
intrinsically scale-invariant. There is, ad hoc, no method that flow is an additional tool to solve common problems in the
allows imaging of oversized pores three-dimensionally. MPS upscaling of reservoir properties. An additional advantage
22 Geofluids

of this scale-independent approach is its applicability on deposition, Geological Society Special Publication, vol. 370, no.
2D as well as 3D datasets. MPS does not introduce general 1, pp. 3947, 2012.
averaging that is commonly used in upscaling. [4] A. Saller, S. Rushton, L. Buambua, K. Inman, R. McNeil, and J.
The workflow allowed upscaling to realistic models at A. D. Dickson, Presalt stratigraphy and depositional systems
different scales by starting from detailed, high-resolution in the Kwanza Basin, offshore Angola, AAPG Bulletin, vol. 100,
information and incorporating it into larger-scale models. no. 7, pp. 11351164, 2016.
Moreover, using recent advances in fluid flow models, this [5] S. Schroder, A. Ibekwe, M. Saunders, R. Dixon, and A. Fisher,
workflow has the potential to replace expensive permeability Algalmicrobial carbonates of the Namibe Basin (Albian,
measurements on large core samples and can serve as input Angola): Implications for microbial carbonate mound develop-
for simulations of multiphase fluid flow. The results have also ment in the South Atlantic, Petroleum Geoscience, vol. 22, no. 1,
pp. 7190, 2016.
indicated that the TI image can be used for different samples
of the same facies type. This would allow scanning samples [6] I. Sharp, K. Verwer, H. Ferreira et al., Pre- and Post-Salt
Non-Marine Carbonates of the Namibe Basin, Angola, in
more quickly using conventional medical CT scanners. When
Microbial Carbonates in Space and Time: Implications for Global
a more detailed description of a part of the pore network is
Exploration and Production, pp. 52-53, The Geological Society,
desired, a pore model can be generated using this workflow. London, UK, 19-20 June 2013.
The here-presented innovative and integrated LBM-MPS
[7] T. D. Ford and H. M. Pedley, A review of tufa and travertine
workflow was applied in continental carbonate rocks with deposits of the world, Earth-Science Reviews, vol. 41, no. 3-4,
complex pore networks. The methodology is directly appli- pp. 117175, 1996.
cable to other lithologies, comprising different pore types, [8] B. W. Fouke, J. D. Farmer, D. J. Des Marais et al., Depositional
pore shapes, and pore networks altogether. The lack of facies and aqueous-solid geochemistry of travertine-depositing
straightforward porosity-permeability relationships in com- hot springs (Angel Terrace, Mammoth Hot Springs, Yellowstone
plex carbonates highlights the necessity for a 3D approach. National Park, U.S.A.), Journal of Sedimentary Research, vol. 70,
Studying fluid pathways in 3D provides the best possible no. 3, pp. 565585, 2000.
understanding of flow through porous media and will be of [9] L. Guo and R. Riding, Hot-spring travertine facies and
crucial importance in reservoir modelling. sequences, Late Pleistocene, Rapolano Terme, Italy, Sedimen-
tology, vol. 45, no. 1, pp. 163180, 1998.
Conflicts of Interest [10] S. Kele, M. Ozkul, I. Forizs et al., Stable isotope geochem-
ical study of Pamukkale travertines: New evidences of low-
The authors declare that there are no conflicts of interest temperature non-equilibrium calcite-water fractionation, Sed-
regarding the publication of this paper. imentary Geology, vol. 238, no. 1-2, pp. 191212, 2011.
[11] H. M. Pedley and M. Rogerson, Tufas and Speleothems Unravel-
ling the Microbial and Physical Controls, 362, Geological Society,
Acknowledgments London, UK, 2010.
The authors thank the owners and managers of the respective [12] A. Pentecost, Travertine, Springer, 2005.
quarries for allowing them to work in actively excavated [13] M. Ozkul, S. Kele, A. Gokgoz et al., Comparison of the
environments, which provided insight into the 3D pore Quaternary travertine sites in the Denizli extensional basin
networks of the continental carbonate rocks. They would based on their depositional and geochemical data, Sedimentary
Geology, vol. 294, pp. 179204, 2013.
like to thank Herman Nijs for aiding in practical matters
during the sample preparation processes. The authors would [14] H. Claes, M. Degros, J. Soete et al., Geobody architecture,
genesis and petrophysical characteristics of the Budakalasz
like to acknowledge the Hercules Foundation (Flanders) for
travertines, Buda Hills (Hungary), Quaternary International,
founding the micro- and nano-CT project for the hierarchical pp. 122, 2016.
analysis of materials. J. Soete was funded by a Ph.D. grant
[15] H. Claes, M. Marques Erthal, J. Soete, M. Ozkul, and R.
from Agentschap voor Innovatie door Wetenschap en Tech- Swennen, Shrub and pore type classification: Petrography of
nologie (IWT), Flanders, Belgium. Last but not least, they travertine shrubs from the Ballk-Belevi area (Denizli, SW
would like to thank Professor Mehmet Ozkul and Sandor Kele Turkey), Quaternary International, vol. 437, pp. 147163, 2017.
for their help and the interesting discussions during the field [16] E. De Boever, A. Foubert, D. Oligschlaeger et al., Multi-
trips in Turkey and Hungary. scale approach to (micro)porosity quantification in continental
spring carbonate facies: Case study from the Cakmak quarry
References (Denizli, Turkey), Geochemistry, Geophysics, Geosystems, vol.
17, no. 7, pp. 29222939, 2016.
[1] H. Claes, J. Soete, K. Van Noten et al., Sedimentology, three- [17] M. M. Erthal, E. Capezzuoli, A. Mancini, H. Claes, J. Soete, and
dimensional geobody reconstruction and carbon dioxide origin R. Swennen, Shrub morpho-types as indicator for the water
of Pleistocene travertine deposits in the Ballik area (south-west flow energy - Tivoli travertine case (Central Italy), Sedimentary
Turkey), Sedimentology, vol. 62, no. 5, pp. 14081445, 2015. Geology, vol. 347, pp. 7999, 2017.
[2] J. G. S. Terra, A. R. Spadini, A. B. Franca et al., Classificacoes [18] S. Khatib, P. Rochette, M. C. Alcicek, A.-E. Lebatard, F. Demory,
classicas de rochas carbonaticas, Boletin Geociencias Petrobras, and T. Saos, Etudes stratigraphique, sedimentologique et
vol. 18, no. 1, pp. 929, 2010. paleomagnetique des travertins de Kocabas, Bassin de Denizli,
[3] V. P. Wright, Lacustrine carbonates in rift settings: The Anatolie, Turquie, contenant des restes fossiles quaternaires,
interaction of volcanic and microbial processes on carbonate LAnthropologie, vol. 118, no. 1, pp. 1633, 2014.
Geofluids 23

[19] A.-E. Lebatard, M. C. Alcicek, P. Rochette et al., Dating [36] J. Caers and T. Zhang, Multiple-point geostatistics: a quan-
the Homo erectus bearing travertine from Kocabas (Denizli, titative vehicle for integrating geologic analogs into multiple
Turkey) at at least 1.1 Ma, Earth and Planetary Science Letters, reservoir models, in Integration of Outcrop and Modern Analogs
vol. 390, pp. 818, 2014. in Reservoir Modeling, G. M. Grammer, P. M. Harris, and G. P.
[20] P. Ronchi and F. Cruciani, Continental carbonates as a hydro- Eberli, Eds., vol. 80, pp. 383394, AAPG Memoir, 2004.
carbon reservoir, an analog case study from the travertine of [37] A. Journel and T. Zhang, The necessity of a multiple-point
Saturnia, Italy, AAPG Bulletin, vol. 99, no. 4, pp. 711734, 2015. prior model, Mathematical Geology, vol. 38, no. 5, pp. 591610,
[21] J. Soete, L. M. Kleipool, H. Claes et al., Acoustic properties in 2006.
travertines and their relation to porosity and pore types, Marine [38] Z. Jiang, M. I. J. Van Dijke, K. S. Sorbie, and G. D. Couples,
and Petroleum Geology, vol. 59, pp. 320335, 2015. Representation of multiscale heterogeneity via multiscale pore
[22] J. Soete, Pore network characterization in complex carbonate sys- networks, Water Resources Research, vol. 49, no. 9, pp. 5437
tems, a multidisciplinary approach [Ph.D. Thesis], Department 5449, 2013.
Geography-Geology, Faculty of Sciences, KU Leuven, Belgium, [39] K. M. Gerke, M. V. Karsanina, and D. Mallants, Universal
2016. stochastic multiscale image fusion: An example application for
[23] J. Bear, Dynamics of Fluids in Porous Media, Dover Publications, shale rock, Scientific Reports, vol. 5, Article ID 15880, pp. 113,
Mineola, New York, USA, 1972. 2015.
[40] J. Yao, C. Wang, Y. Yang, R. Hu, and X. Wang, The construction
[24] T. Bultreys, L. Van Hoorebeke, and V. Cnudde, Multi-scale,
of carbonate digital rock with hybrid superposition method,
micro-computed tomography-based pore network models to
Journal of Petroleum Science and Engineering, vol. 110, pp. 263
simulate drainage in heterogeneous rocks, Advances in Water
267, 2013.
Resources, vol. 78, pp. 3649, 2015.
[41] E. De Boever, A. Foubert, B. Lopez et al., Comparative study
[25] W. Degruyter, O. Bachmann, and A. Burgisser, Controls
of the Pleistocene Cakmak quarry (Denizli Basin, Turkey) and
on magma permeability in the volcanic conduit during the
modern Mammoth Hot Springs deposits (Yellowstone National
climactic phase of the Kos Plateau Tuff eruption (Aegean Arc),
Park, USA), Quaternary International, 2016.
Bulletin of Volcanology, vol. 72, no. 1, pp. 6374, 2010.
[42] A. Torok, A. Mindszenty, H. Claes, S. Kele, L. Fodor, and
[26] W. Degruyter, A. Burgisser, O. Bachmann, and O. Malaspinas,
R. Swennen, Geobody architecture of continental carbonates:
Synchrotron X-ray microtomography and lattice Boltzmann
Gazda travertine quarry (Sutto, Gerecse Hills, Hungary),
simulations of gas flow through volcanic pumices, Geosphere,
Quaternary International, pp. 122, 2016.
vol. 6, no. 5, pp. 470481, 2010.
[43] V. Cnudde and M. N. Boone, High-resolution X-ray computed
[27] J. Domitner, C. Holzl, A. Kharicha et al., 3D simulation of tomography in geosciences: a review of the current technology
interdendritic flow through a Al-18wt.%Cu structure captured and applications, Earth-Science Reviews, vol. 123, pp. 117, 2013.
with X-ray microtomography, in Proceedings of IOP Conference
Series: Materials Science and Engineering, vol. 27, June 2011. [44] K. A. Alshibli and A. H. Reed, Advances in Computed Tomogra-
phy for Geomaterials: GeoX 2010, Wiley-ISTE, 2010.
[28] H. Dong, S. Fjeldstad, L. Alberts, S. Roth, S. Bakke, and P.
[45] D. Wildenschild and A. P. Sheppard, X-ray imaging and
ren, Pore network modelling on carbonate: A comparative
analysis techniques for quantifying pore-scale structure and
study of different micro-CT network extraction methods, in
processes in subsurface porous medium systems, Advances in
Proceedings of the International Symposium of the Society of Core
Water Resources, vol. 51, pp. 217246, 2013.
Analysts, pp. 112, 2008.
[46] K. Remeysen and R. Swennen, Application of microfocus
[29] B. Ferreol and D. H. Rothman, Lattice-Boltzmann simulations
computed tomography in carbonate reservoir characterization:
of flow through Fontainebleau sandstone, Transport in Porous
Possibilities and limitations, Marine and Petroleum Geology,
Media, vol. 20, no. 1-2, pp. 320, 1995.
vol. 25, no. 6, pp. 486499, 2008.
[30] FlowKit Ltd, Palabos, Parallel Lattice Boltzmann Solver, 2011-
[47] B. Masschaele, M. Dierick, D. V. Loo et al., HECTOR: A 240kV
2012, http://www.lbmethod.org/palabos.
micro-CT setup optimized for research, Journal of Physics:
[31] P. Ringrose and M. Bentley, Reservoir Model Design: A Practi- Conference Series, vol. 463, no. 1, Article ID 012012, 2013.
tioners Guide, Springer, Dordrecht, The Netherlands, 2015. [48] J. Vlassenbroeck, M. Dierick, B. Masschaele, V. Cnudde, L. Van
[32] S. B. Strebelle and A. G. Journel, Reservoir Modeling Using Hoorebeke, and P. Jacobs, Software tools for quantification
Multiple-Point Statistics, in Proceedings of the SPE Annual of X-ray microtomography at the UGCT, in Proceedings of
Technical Conference and Exhibition, Society of Petroleum the 10th International Symposium on Radiation Physics (ISRP
Engineers, New Orleans, Louisiana, USA, 2001. 10), vol. 580, pp. 442445, Nuclear Instruments and Methods
[33] G. Mariethoz, J. Straubhaar, P. Renard, T. Chugunova, and P. in Physics Research, Section A: Accelerators, Spectrometers,
Biver, Constraining distance-based multipoint simulations to Detectors and Associated Equipment, 2007.
proportions and trends, Environmental Modelling and Soft- [49] S. Claes, Pore classification system and upscaling strategy
ware, vol. 72, pp. 184197, 2015. in travertine reservoir rocks [Ph.D. Thesis], Department
[34] M. Huysmans and A. Dassargues, Modeling the effect of clay Geography-Geology, Faculty of Sciences, KU Leuven, Belgium,
drapes on pumping test response in a cross-bedded aquifer 2015.
using multiple-point geostatistics, Journal of Hydrology, vol. [50] A. G. Journel, Geostatistics: roadblocks and challenges, in
450-451, pp. 159167, 2012. Geostatistics Troia 92, A. Soares, Ed., vol. 5 of Quantitative
[35] H. Okabe and M. J. Blunt, Prediction of permeability for Geology and Geostatistics, pp. 213224, Kluwer Academic Pub-
porous media reconstructed using multiple-point statistics, lications, Dordrecht, The Netherlands, 1993.
Physical Review E, vol. 70, no. 6, Article ID 066135, p. 066135/10, [51] F. B. Guardiano and R. M. Srivastava, Multivariate Geostatis-
2004. tics: Beyond Bivariate Moments, in Geostatistics Troia 92, A.
24 Geofluids

Soares, Ed., vol. 5 of Quantitative Geology and Geostatistics,

pp. 133144, Kluwer Academic Publications, Dordrecht, The
Netherlands, 1993.
[52] S. B. Strebelle, Sequential Simulation Drawing Structures from
Training Images [Ph.D. Thesis], Stanford University, 2000.
[53] Y. Liu, Using the Snesim program for multiple-point statistical
simulation, Computers and Geosciences, vol. 32, no. 10, pp.
15441563, 2006.
[54] E. Meerschman, G. Pirot, G. Mariethoz, J. Straubhaar, M. Van
Meirvenne, and P. Renard, A practical guide to performing
multiple-point statistical simulations with the Direct Sampling
algorithm, Computers and Geosciences, vol. 52, pp. 307324,
2013.
[55] T. T. Tran, Improving variogram reproduction on dense
simulation grids, Computers and Geosciences, vol. 20, no. 7-8,
pp. 11611168, 1994.
[56] R. Mei, W. Shyy, D. Yu, and L.-S. Luo, Lattice Boltzmann
method for 3-D flows with curved boundary, Journal of
Computational Physics, vol. 161, no. 2, pp. 680699, 2000.
[57] J. Latt and M. J. Krause, OpenLB User Guide, 20062015,
http://optilb.com/openlb.
[58] T. J. Pedley, Introduction to fluid dynamics, Scientia Marina,
vol. 61, no. 1, pp. 724, 1997.
[59] R. W. Fox, A. T. McDonald, and P. J. Pritchard, Introduction to
Fluid Mechanics, John Wiley and Sons, 7th edition, 2006.
[60] J. Latt, Choice of units in lattice Boltzmann simulations, 2008,
http://www.palabos.org.
[61] Y. Watanabe and Y. Nakashima, RW3D.m: Three-dimensional
random walk program for the calculation of the diffusivities in
porous media, Computers and Geosciences, vol. 28, no. 4, pp.
583586, 2002.
[62] Y. Nakashima and T. Yamaguchi, DMAP.m: A Mathematica
program for three-dimensional mapping of tortuosity and
porosity of porous media, Bulletin of the Geological Survey of
Japan, vol. 55, no. 3-4, pp. 93103, 2004.
[63] C. J. Gommes, A.-J. Bons, S. Blacher, J. H. Dunsmuir, and
A. H. Tsou, Practical methods for measuring the tortuosity
of porous materials from binary or gray-tone tomographic
reconstructions, AIChE Journal, vol. 55, no. 8, pp. 20002012,
2009.
[64] N. Remy, A. Boucher, and J. Wu, Applied Geostatistics with
SGeMS: A Users Guide, Cambridge University Press, 2009.
[65] C. Bouvet de Maisonneuve, O. Bachmann, and A. Burgisser,
Characterization of juvenile pyroclasts from the Kos Plateau
Tuff (Aegean Arc): Insights into the eruptive dynamics of a large
rhyolitic eruption, Bulletin of Volcanology, vol. 71, no. 6, pp.
643658, 2009.
[66] S. Claes, J. Soete, V. Cnudde, and R. Swennen, A three-
dimensional classification for mathematical pore shape descrip-
tion in complex carbonate reservoir rocks, Mathematical Geo-
sciences, vol. 48, no. 6, pp. 619639, 2016.
[67] T. Zhang, MPS-Driven Digital Rock Modeling and Upscaling,
Mathematical Geosciences, vol. 47, no. 8, pp. 937954, 2015.
[68] D. Hodgetts, Laser scanning and digital outcrop geology in the
petroleum industry: A review, Marine and Petroleum Geology,
vol. 46, pp. 335354, 2013.
[69] A. Pickel, J. D. Frechette, A. Comunian, and G. S. Weissmann,
Building a training image with Digital Outcrop Models,
Journal of Hydrology, vol. 531, pp. 5361, 2015.
 International Journal of Journal of
Ecology Mining

Journal of The Scientific

Geochemistry
Hindawi Publishing Corporation
Scientifica
Hindawi Publishing Corporation Hindawi Publishing Corporation
World Journal
Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of
Earthquakes
Hindawi Publishing Corporation
Paleontology Journal
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of
Petroleum Engineering

Submit your manuscripts at

https://www.hindawi.com

*HRSK\VLFV
International Journal of

Hindawi Publishing Corporation Hindawi Publishing Corporation

http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 201

Advances in Journal of Advances in Advances in International Journal of

Meteorology
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Climatology
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Geology
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Oceanography
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Oceanography
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014

$SSOLHG Journal of
International Journal of Journal of International Journal of (QYLURQPHQWDO Computational
Mineralogy
Hindawi Publishing Corporation
Geological Research
Hindawi Publishing Corporation
Atmospheric Sciences
Hindawi Publishing Corporation
6RLO6FLHQFH
+LQGDZL3XEOLVKLQJ&RUSRUDWLRQ 9ROXPH
Environmental Sciences
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 KWWSZZZKLQGDZLFRP http://www.hindawi.com Volume 2014