Spe 206109 Pa

Electrical, Diffusional, Hydraulic, and
Geometrical Tortuosity Anisotropy

Quantification Using 3D Computed
Tomography Scan Image Data
Andres Gonzalez1, Zoya Heidari1*, and Olivier Lopez2
1
The University of Texas at Austin
2
Equinor
Summary
Sedimentary rocks display complex spatial distribution of both pore space and solid components, impacting the directional dependence
Downloaded from http://onepetro.org/REE/article-pdf/26/01/23/3066722/spe-206109-pa.pdf by Eni SPA E&P Division user on 12 April 2023
of physical phenomena such as electrical conduction, fluid flow, heat transfer, and molecular diffusion. The complexity of the pore space
is often quantified by the concept of tortuosity, which measures the sinuosity of the connecting paths in the pore space. Tortuosity is an
important quantity in formation evaluation as it impacts petrophysical properties such as permeability and formation factor. However,
the existence of various techniques can lead to nonuniqueness in assessment of tortuosity. Furthermore, spatial variation of the solid
components of the rocks occurring at the core-scale domain reflected in the connectivity and distribution of the minerals is typically not
quantified. The objectives of this paper are (a) to quantify tortuosity and tortuosity anisotropy of porous media through estimation of
electrical, diffusional, hydraulic, and geometrical tortuosity at the pore scale and core scale and (b) to compare electrical, diffusional,
hydraulic, and geometrical tortuosity.
We estimate tortuosity in the pore space of microcomputed tomography (micro-CT) scan images and in the most connected and abun-
dant solid phase of whole-core CT scan images. We conduct numerical simulations of electric potential distribution, diffusion, and fluid
flow and velocity distribution to estimate electrical, diffusional, and hydraulic tortuosity, respectively. To calculate geometrical tortuos-
ity, we use the segmented pore space from micro-CT scan images to extract a pore network model and compute the shortest path of all
opposing pores of the samples. Finally, tortuosity values obtained with each technique are used to assess the anisotropy of the samples.
We applied the documented workflow to core- and pore-scale images. The CT scan images in the core-scale domain belong to a silici-
clastic formation. Micro-CT scan images in the pore-scale domain were obtained from Berea Sandstone, Austin Chalk, and Estaillades
limestone formations. We observed differences in estimates of direction-dependent electrical, diffusional, hydraulic, and geometrical
tortuosity for both types of images. The highest numerical differences were observed when comparing streamline electrical and hydraulic
tortuosity with diffusional tortuosity. The observed differences were significant in anisotropic samples. Differences in tortuosity esti-
mates can impact the outcomes of rock physics models for which tortuosity is an input. The documented comparison provides insight in
the selection of techniques for tortuosity estimation. Use of core-scale image data provides semicontinuous estimates of tortuosity and
tortuosity anisotropy, which are typically not attainable using pore-scale images. Additionally, the semicontinuous tortuosity anisotropy
estimates from whole-core CT scan images provide a tool for selection of best locations to take core plugs.
Introduction
Complex spatial distribution of rock components (i.e., pore space and solid matrix) can result in anisotropy of various petrophysical
properties such as permeability and formation factor. Anisotropy is often estimated by measuring the petrophysical property of interest at
different orientations. Then, the differences between the values of the property of interest at different orientations are used as a measure
of anisotropy of that given property. The complexity of the pore space is often described by the concept of tortuosity, which is a quantity
that attempts to capture the sinuosity of the connecting paths within the pore space. Tortuosity is commonly invoked in formation evalu-
ation workflows. Examples include appearance of tortuosity in fluid flow and electrical conduction models (Kozeny 1927; Carman 1937;
Garcia and Heidari 2018). Estimation of tortuosity can be accomplished through experimental measurements, theoretical models, simu-
lation of physical phenomena, and through geometrical calculation in high-resolution image data.
Modern imaging techniques such as X-ray micro-CT and scanning electron microscopy enabled the acquisition of high-resolution
images resolving the pore space of sedimentary rocks and other porous materials at the micron/nanometer scale (Fredrich et al. 1993;
Spanne et al. 1994; Auzerais et al. 1996; Arns et al. 2001; Rassenfoss 2011). The advent of this technology has allowed the acquisition of
3D images representing the pore space of sedimentary rocks. The aforementioned 3D images are often used for the simulation of various
physical phenomena such as fluid flow and electrical current flow at the pore-scale domain (i.e., pore-scale modeling). Such pore-scale
simulations have been used to estimate permeability and permeability anisotropy (Al Mansoori et al. 2014; Sun et al. 2015) as well as
tortuosity of the porous media (Fu et al. 2021).
Tortuosity appears as an input in various rock physics models (Clennell 1997; Ghanbarian et al. 2013; Fu et al. 2021) such as the
Carman-Kozeny equation for estimation of permeability (Kozeny 1927; Carman 1937) as well as in water saturation models (Chen and
Heidari 2015; Garcia and Heidari 2018), highlighting its importance in formation evaluation. However, the existence of various tech-
niques for estimation of tortuosity and their conceptual differences can lead to inconsistencies in estimates of petrophysical properties
through tortuosity-dependent models. Tortuosity estimates through simulation of physical phenomena and use of image analysis
*Corresponding author; email: zoya@utexas.edu

Copyright © 2023 Society of Petroleum Engineers
This paper (SPE 206109) was accepted for presentation at the SPE Annual Technical Conference and Exhibition, Dubai, UAE, 21–23 September 2021, and revised for publication.
Original manuscript received for review 21 February 2022. Revised manuscript received for review 5 July 2022. Paper peer approved 7 July 2022.
February 2023 SPE Reservoir Evaluation & Engineering 23

techniques include hydraulic, electrical, diffusional, and geometrical tortuosity. The aforementioned tortuosity estimates are often used
interchangeably in the literature (Ghanbarian et al. 2013).
In this paper, we document a workflow for the estimation of tortuosity and tortuosity anisotropy through simulation of electric current
flow, diffusion, and fluid flow in porous media. Additionally, we compute tortuosity using geometrical techniques. The purpose of this
workflow is to quantify the numerical differences between tortuosity estimates using various techniques and to identify which tortuosity
definitions can be used interchangeably. Furthermore, we explore the possibility of using whole-core CT scan images to estimate the
tortuosity of the solid components of sedimentary rocks. First, we preprocess the whole-core 3D CT scan images to eliminate undesired
noncore material portions in the raw image slices. Then, we segment the visually identifiable phases of the whole-core 3D CT scan images
using a supervised learning algorithm. Segmented and preprocessed micro-CT scan images were already available. Afterward, we con-
duct simulation of electric potential distribution, diffusion, and fluid flow to obtain electrical, diffusional, and hydraulic tortuosity in all
three Cartesian directions. Simulations are conducted for both whole-core 3D CT scan (in the most continuous phase) and 3D micro-CT
scan (in the pore space) images. Additionally, we estimated geometrical tortuosity in all the 3D micro-CT scan images.
Materials and Method

We introduced a workflow for estimation of direction-dependent tortuosity and anisotropy assessment through numerical simulation of
hydraulic, diffusional, and electrical transport processes using segmented 3D images from whole-core CT scan images and micro-CT scan
images. Additionally, we estimated geometrical tortuosity in segmented images from micro-CT scan images. Fig. 1 illustrates the pro-
posed workflow.
Fig. 1—Workflow for estimation of direction-dependent tortuosity and rock anisotropy.
The main steps of the proposed workflow are summarized as follows:

• Image preprocessing and image segmentation of the visually identifiable phases of whole-core 3D CT scan images.
• Numerical simulation of electric potential distribution, diffusion, and fluid flow in both segmented whole-core CT scan and
micro-CT scan images.
• Estimation of direction-dependent tortuosity using the results of the conducted simulations and estimation of geometrical tortuosity
by means of image analysis and graph representation techniques.
• Estimation of tortuosity anisotropy in the phase of interest (pore space or matrix component) and comparison of tortuosity estimates
obtained via the simulation of the aforementioned physical phenomena and by means of geometrical techniques.
Description of the Data Set. The data set used in this paper consisted of two dual-energy 3D CT scan images from a siliciclastic
formation and three segmented 3D micro-CT scan images from three distinct formations. In the following section, we will describe both
types of image data used in this paper.
Whole-Core Computed Tomography Scan Images. The two dual-energy 3D CT scan images are from a 4 in. whole core acquired in a
siliciclastic depth interval covering approximately 2 m. Fig. 2 shows the initial formation evaluation for the depth interval containing the
2-m whole-core 3D CT scan images. Porosity in the evaluated depth interval ranges between 14.2 and 26.4%, while permeability ranges
between 3.55 and 268 md. The blue-shaded box indicates the location of the two whole-core 3D CT scan images. The main lithology of
the selected depth intervals is a biogenic sedimentary rock composed by sponge silica spicules. The resolution of the used whole-core
3D CT scan images is 0.234 mm/pixel in the x- and y- directions and 0.5 mm/pixel in the z-direction. However, we rescaled the resolu-
tion of the 3D images in the x- and y- directions to match the resolution in the z-direction to preserve a homogeneous discretization of
the simulation domain. Fig. 3 shows 3D renders of whole-core 3D CT scan images Sample 1.1 (Fig. 3a) and Sample 1.2 (Fig. 3b) with
raw dimensions of 512×512×1,778 and 512×512×2,025 (x-, y-, and z-direction, respectively) voxels, respectively. The final dimensions
of the samples after rescaling are 126×126×1,778 and 126×126×2,025 voxels for Sample 1.1 and Sample 1.2, respectively. To conduct
the simulations described in the method section, we partitioned the samples into cubic subsamples of 126×126×126 voxels, resulting in
14 and 16 subsamples for Sample 1.1 and Sample 1.2, respectively. The reasons for choosing the described samples are the following:
24 February 2023 SPE Reservoir Evaluation & Engineering

Fig. 2—Conventional formation evaluation—tracks from left to right: Track 1: depth; Track 2: gamma-ray (GR); Track 3: bulk density
(RHOB) and neutron porosity (NPHI, in water-filled sandstone units); Track 4: shallow (M2R1) and deep resistivity (M2RX); Track
5: photoelectric factor (PEF); Track 6: shear-wave slowness (DTS) and compressional-wave slowness (DTC); Track 7: estimated
total porosity and core porosity measurements; Track 8: water saturation; Track 9: estimated permeability and core permeability;
Track 10: core CT scan image; and Track 11: core photo.
Fig. 3—3D renders of whole-core 3D CT scan image samples: (a) Sample 1.1 and (b) Sample 1.2 (after Gonzalez et al. 2021).
(a) there is significant visual evidence of complex distribution of mineral components and (b) enough gray-scale and textural contrast to
segment different phases.

Microcomputed Tomography Scan Images. The used segmented 3D micro-CT images belong to three formations with different pore
structure. The first 3D image (Sample 2.1) was acquired from the Berea Sandstone formation (Dong 2007; Chi and Heidari 2016), which
has a spatial resolution of 5.345 μm/pixel, porosity of 19.65%, and 400×400×400 voxels. The second 3D image (Sample 2.2) belongs to
the Austin Chalk formation (Prodanovic et al. 2015; Chi and Heidari 2016), which has a spatial resolution of 0.7 μm/pixel, porosity of
29.56%, and 600×600×600 voxels. The third 3D image (Sample 2.3) was obtained from the Estaillades limestone formation (Andrew
2014; Azizoglu and Heidari 2021). This image has a spatial resolution of 3.31 μm/pixel, porosity of 12.73%, and 1,000×1,000×1,000
voxels. In the numerical simulations on the Austin Chalk and the Estaillades limestone samples, we used 400×400×400 voxel subsamples.
Fig. 4 shows the 3D rendering of the segmented 3D micro-CT scan images.
Fig. 4—3D renders of segmented 3D micro-CT scan images: (a) Sample 2.1, (b) Sample 2.2, and (c) Sample 2.3. Gray pixels and
black pixels represent the grains (or matrix) and the pore space, respectively.
Preprocessing of 3D Computed Tomography Scan Images. Whole-Core 3D Computed Tomography Scan Images. Acquisition of
whole-core 3D CT scan images is typically accomplished with the core material (e.g., sandstone) contained in the core barrel used in the
coring process. As such, the raw 3D images contain visual elements that do not correspond to the core material. Examples of noncore
material visual elements are the core barrel used in the coring process and gypsum used to stabilize the core material during the coring
process. Fig. 5a shows an example of a raw image slice, displaying the core material, core barrel, and gypsum. Removal of the noncore
material visual elements is necessary to conduct the subsequent steps of the proposed workflow. The core barrel is displayed as a circular
ring surrounding the core material. Gypsum appears as a bright material between the core barrel and the core material. To remove the
noncore material visual elements, first we used an edge detection technique (Canny 1986) to identify the edges of each whole-core CT
scan image slice. Then, we use the detected edges as an input for a parametric curve detection technique, the Hough transform (Hough
1962). The outputs of the Hough transform are the parameters of the curve that best fit the detected edges. In this case, our objective was
to identify the circumference formed by the core material. We reduced the search space by limiting the potential circumference radii to
the diameter of the retrieved core (i.e., 4 in.). Once we obtained the best fitting circumference, we replaced the value of all the pixels
outside of the detected circumference with a characteristic value and were not utilized for further analysis. Fig. 5 shows all the steps of
the workflow to remove the noncore material elements from the raw whole-core 3D CT scan images. Finally, to conduct our simulations
in a Cartesian coordinate system, we cut a rectangular prism from the cylinder obtained in the previous step. We fitted the largest possible
square within the perimeter of the circumference of each whole-core CT scan slice (red line in Fig. 5c).
3D Microcomputed Tomography Scan Images. Raw micro-CT scan images can display visual artifacts such as beam hardening or
noise after acquisition. These artifacts negatively impact all the subsequent steps in a typical pore-scale modeling workflow. To remove
these artifacts noise removal filters can be applied. Additionally, for segmentation purposes, contrast enhancing techniques are used to
help identifying the different phases (i.e., mineral constituents and pore space) present in the analyzed image. Different phases are typi-
cally defined as regions of the image substantially different from other regions in terms of gray-scale intensity, shape, or image texture.
It should be mentioned that no preprocessing or segmentation of micro-CT scan images was conducted in this work, as preprocessed
and segmented images were available. Segmentation of these types of images can be accomplished with the same methodology used for
whole-core CT scan images.
Segmentation of 3D Computed Tomography Scan Images. The numerical simulations are conducted in the most continuous solid
phase in the case of the whole-core 3D CT scan images and in the pore space in the case of 3D micro-CT scan images. We used a machine
learning-assisted segmentation (Misra and Wu 2019) workflow called WEKA 3D trainable segmentation plug-in (Arganda-Carreras et al.
2017) accessible in the image analysis software ImageJ-Fiji (Schindelin et al. 2012) to segment different phases in whole-core CT scan
images. The segmentation scheme used in WEKA 3D relies on the extraction of image-based features from regions representing each
phase to train a supervised learning algorithm (Wang et al. 2015; Berg et al. 2018; Misra and Wu 2019). We distinguish each phase mainly
by gray-scale intensity differences, which translate into differences in density between the lithologies that compose the evaluated interval.
The selected regions are used to extract the features used in the training step of the segmentation algorithm. Fig. 6 presents an example
of the selected regions of interest for the visually identifiable phases and the segmented image slice. The red, green, and yellow phases
represent rock constituents with increasing density from red to green to yellow.
Tortuosity Estimation. We used the segmented whole-core CT scan and micro-CT scan 3D images as the domain for the simulation of
electric potential distribution, diffusion, and fluid flow through porous media. As previously mentioned, 3D simulations are conducted in
the most continuous solid phase of the whole-core 3D CT scan images and in the pore space of the 3D micro-CT scan images. In the case
of whole-core 3D CT scan images, we assumed that the most continuous phase is electrically conductive for the case of electrical potential

Fig. 5—An example for the process of removing noncore material visual elements: (a) raw CT scan image slice, (b) detected edges,
(c) detected core perimeter, and (d) processed CT scan image slice.
Fig. 6—An example for the CT scan image slice segmentation process: (a) CT scan image slice and (b) segmented CT scan image
slice. The colored rectangles show examples of the visually identified phases.
distribution simulations. The same phase is considered as the space available for fluid flow and diffusion in the cases of fluid flow and
diffusion simulations. All the simulations were conducted using Cartesian coordinate system to discretize the simulation domain. Fig. 7
shows the discretization of the simulation domain using Cartesian coordinates. In the following sections, we explain the method used to
conduct each one of the aforementioned simulations.
Electrical Tortuosity. To estimate the electrical tortuosity, we first estimated the electric potential distribution of the segmented sam-
ples. This was accomplished by solving the steady-state continuity equation for electric charges. The simulations were conducted in
the pore space fully saturated with conductive brine. We assumed that the rock grains/rock matrix are not conductive. We estimated the

Fig. 7—Discretization of the simulation domain in the Cartesian coordinates system.
current density and the distance covered by each electric charge after applying a potential difference using a streamline algorithm. Finally,
we used the computed lengths to estimate electrical tortuosity. Fig. 8 shows the main steps to compute the electrical tortuosity.
Fig. 8—Workflow for estimation of electrical tortuosity. The black and gray regions in the segmented image represent the pore
space and the matrix, respectively.
We estimated the electric potential distribution in the segmented whole-core CT scan and micro-CT scan 3D images by solving the
steady-state continuity equation for electric charges defined as

‍ r J = r rE = 0,‍ (1)
where σ is the electric conductivity, E is the electric potential, and J is the electric current density. First, we discretized the simulation
domain using a Cartesian framework. Afterward, we solved Eq. 1 utilizing the transmissibility method (Garcia and Heidari 2018) by
computing the half-cell transmissibility defined in the x-direction as
y z
Tx = 2x , (2)
‍ x ‍

where Δx, Δy, and Δz are the dimensions of the grid cells in each Cartesian direction. We computed the electrical transmissibility in the
y- and z-directions in an analogous fashion. To conduct the simulations in each Cartesian direction, we applied boundary conditions of a
constant electric potential difference in the two faces perpendicular to the direction of interest and no potential difference in the remaining
faces of the cubic domain. The problem is reduced to an algebraic linear problem, and we solved it using the biconjugate gradient tech-
nique (Barrett et al. 1994). After solving for the electric potential distribution in each Cartesian direction, we estimated the electric current
density as
‍ J = rE,‍ (3)
where σ is the electrical conductivity of the phase available for electrical current flow, and E is the electric potential distribution in each
Cartesian direction. Then, we used the electrical current density as an input for a streamline tracing algorithm. The algorithm we used is
described in the publication by Pollock (1988). The algorithm was developed for tracing particles in groundwater flow modeling. The
method uses the velocity distribution to track the path of individual particles using semianalytical equations. However, in this paper, we
used the electric current density as suggested by Garcia and Heidari (2018) instead of particle velocity. The electric current density is
related to the drift velocity in the x-direction via

Jx = J11 + J12 J11 xx . (4)
‍ ‍
We computed the drift velocity in each Cartesian direction. Then, we used the computed drift velocities to estimate the hypothetic time
that would take for a charge to travel from one end to the other end of the grid cell in each Cartesian direction. Afterward, the lowest time
estimated is used to determine the direction in which the charge is leaving the cell. The calculations are repeated for every particle in the
domain. The outcome is a set of streamlines describing the path of each charge when traveling from the face of the highest electric poten-
tial to the face of the lowest electric potential of the cubic domain. Finally, we computed the electrical tortuosity via
Pns L
s,i
e,i = , (5)
‍ s=1 L ‍
where Ls,i is the length of the streamline of each particle in the i-th direction, ns is the total number of streamlines, and L is the length of
the sample in the i-th direction, which is the direction in which the electric potential difference is applied. Other authors (Herrick and
Kennedy 1994; Fu et al. 2021) defined electrical tortuosity as
r
0
es,i = , (6)
‍ rock ‍
where σ0 is the conductivity of brine, σrock is the conductivity of the rock saturated with brine, and ‍ ‍ is the connected porosity of the
sample. We computed electrical tortuosity using both definitions (i.e., Eq. 5 and Eq. 6). In the case of Eq. 6, we obtained the conductivity
of the rock by dividing the electric current density over the voltage applied across the sample. The electrical tortuosity computed using
Eq. 6 is referenced as simplified electrical tortuosity, and the one computed using Eq. 5 is referenced as streamline electrical tortuosity.
Diffusional Tortuosity. To compute the diffusional tortuosity, we simulate a random walk of n nonabsorbing particles in the segmented
pore space. Fig. 9 shows the workflow for the calculation of diffusional tortuosity.
The process simulated by the random walk algorithm resembles the Brownian motion of particles suspended in a medium. The random
walk of nonabsorbing particles can be considered as a simplified diffusion simulation (Fu et al. 2021). We used Pytrax (Tranter et al.
2019), a readily available open-source Python library that implements a random walk algorithm for estimation of tortuosity. The required
inputs for tortuosity estimation using the Pytrax library are the simulation domain identifying the space available for diffusion and the
space where no diffusion is possible, the number of walkers, and the number of time steps of the simulation. Diffusional tortuosity is
defined as the square root of the ratio of the diffusivity in free space to the diffusivity in the restricted simulation domain (Nakashima and
Kamiya 2007). Diffusivity in the free space is defined as
a2
D0 = , (7)
‍ 6t ‍
where a is the dimension of a cubic voxel, τ is a dimensionless time, and t is the time. On the other hand, the diffusivity of nonabsorbing
particles in the porous media is defined as
D E
2
1 r td
(8)
D t = ,
‍ 6 dt ‍
where r(t)2 is the mean squared displacement of the walkers. Then, the diffusional tortuosity at the limit when both t and τ tend to infinity
is computed as
s v
D0 u a2 dt
u
d = = t D E. (9)
D t 2
d r t
‍ ‍
The inverse of the time derivative of the mean squared displacement derivative is obtained by fitting a straightline to the plot of the mean
squared displacement against time. Similarly, the directional diffusional tortuosity is computed via
v
u a2 dt
u
d,i =t D E,
2
3d r t i
‍ ‍ (10)

Fig. 9—Workflow for estimation of diffusional tortuosity. The black and gray regions in the segmented image represent the pore
space and the matrix, respectively. The red dots and the black paths in the random walk simulation panel represent the walkers
and the path of each walker, respectively.
where r(t)2i is the mean squared displacement in the i-th Cartesian direction (i.e., x, y, or z).
The assumptions of the algorithm code in the Pytrax library include: (a) the walkers do not enter to the solid phase (space where no
diffusion possible), (b) the walkers move only in orthogonal direction (i.e., no diagonal displacement is allowed), (c) a walker reaching
the solid phase will not keep moving, instead it will bounce back to its original position, (d) when a walker reaches the boundary of the
simulation domain it is not contained, instead it keeps moving in mirror copies of the simulation domain to prevent overestimation of the
tortuosity values, and (e) the axial squared displacement is plotted against time and a regression line is fitted and forced to have an inter-
cept of zero, then the square root of the inverse of the slope of the regression line is the directional diffusional tortuosity. To select the
number of walkers and time steps, we conducted a sensitivity analysis to determine the minimum number of time steps and walkers to
traverse the entire sample resulting in a constant value of tortuosity.
Hydraulic Tortuosity. To estimate hydraulic tortuosity, we simulated the process of fluid flow in porous media using Palabos (Latt
et al. 2021) an open-source lattice Boltzmann method solver, commonly used in computational fluid dynamics. The lattice Boltzmann
method is based on the Boltzmann equation from kinetic theory of gases. The method uses discrete particle distribution functions that
represent the density of particles with a certain velocity at a particular position at a given time. Using this approach, macroscopic quan-
tities of interest (e.g., fluid density and momentum density) can be retrieved from the evolution in time and space of the discrete particle
distribution functions. The inputs for the solver we used include a stack of 2D images that define the space available for fluid flow, the
rock matrix, the rock matrix boundaries in contact with the fluid, the dimensions of the sample in all Cartesian directions, and the applied
pressure drop across the desired direction in lattice units. Fig. 10 shows the workflow to estimate hydraulic tortuosity.
Palabos solves the discrete form of the Boltzmann equation defined as

‍ fi x + ci t, t + dt = fi x, t + i x, t ,‍ (11)
where fi is the discrete velocity distribution, x is the position of the particle, t is the time, and Ωi is the collision operator. In our simula-
tions, we used the Bhatnagar-Gross-Krook collision operator defined as
eq
fi fi
i = t, (12)
‍ r ‍
where τr is the relaxation time and fi eq is the equilibrium distribution function computed via
2 !
eq uci uci uu
fi = wi 1 + 2 + 2 , (13)
cs 2c4s 2cs
‍ ‍
where ρ is the macroscopic fluid density, u is the macroscopic velocity vector, ci is the particle velocity, cs is the speed of sound, and wi is
the weighting coefficient defined in the velocity set (velocity discretization). In our simulation, we used D3Q19 velocity set (Mohamad
2011; Krüger et al. 2017) to discretize velocity and space in 3D fluid flow simulations. Further details on the derivation, discretization,
and velocity set in the lattice Boltzmann method can be found in the work documented by Krüger et al. (2017). The output of the fluid
flow simulations includes the velocity distribution in the simulation domain, which can be used to estimate the permeability of the media.
However, we are not interested in the permeability of the media but in the direction-dependent tortuosity. For that purpose, we computed
the direction-dependent hydraulic tortuosity defined by Matyka and Koza (2012) as

Fig. 10—Workflow for estimation of hydraulic tortuosity. The black and gray regions in the segmented image represent the pore
space and the matrix, respectively.
P
n
v
r=1
h,i = , (14)
Pn
vi
‍ r=1 ‍
where v is the velocity magnitude, vi is the velocity in the i-th direction (i.e., x, y, and z), and n is the total number of cells in the Cartesian
simulation domain, the i-th direction corresponds to the direction in which the pressure drop is applied. The fluid used in our simulations
is a Newtonian incompressible fluid. This was enforced by defining a density of the fluid of 1 in lattice units and constant viscosity.
Additionally, we conducted a sensitivity analysis on the pressure differential in lattice units to verify that simulations occurred in laminar
flow regime. This was accomplished by plotting permeability in lattice units against pressure differential in lattice units and verifying the
range of pressure drops for which the permeability remained constant.
Geometrical Tortuosity. To estimate geometrical tortuosity of the pore network, we first extracted a pore network representation of
each sample from the segmented micro-CT scan images. Then, we used a graph representation of the pore network in conjunction with
the Dijkstra algorithm (Dijkstra 1959) to estimate the shortest path between pores located in opposing faces. The Dijkstra algorithm is a
method used in graph representation (data representation by vertices and edges) to compute the shortest distance between two points or
vertices. The algorithm starts by looking to the neighbors of the starting vertex and stores the distance from the starting vertex to its neigh-
bors. Next, the vertex with the shortest distance to the starting vertex is visited, and the distance with the unvisited vertices is computed.
This last step is repeated until the target vertex is reached. Fig. 11 shows schematically what we define as the shortest path between pores
located in opposing faces of the sample. Finally, we computed the geometrical tortuosity via
Png L
g,i
g,i = , (15)
‍ s=1 L ‍
where Lg,i is the length of the shortest path connecting a pair of pores on opposing faces of the sample in the i-th direction, ng is the total
number of connecting paths (paths connecting pores on opposing faces of the sample), and L is the length of the sample in the i-th direc-
tion. The geometry and shapes that result from the segmentation of the distinct solid phases in whole-core CT scan images are not suited
for the extraction of a pore network model. Thus, geometrical tortuosity was not computed for whole-core CT scan images.
The extraction of a pore network representation for each sample was accomplished using image analysis and pore-scale simulation
Python libraries, namely Scikit Image (Van der Walt et al. 2014), Porespy (Gostick et al. 2019), and Openpnm (Gostick et al. 2016). First,
we used the segmented images (segmented micro-CT scan images for each used sample were available) as input for the computation of
the distance transform of the segmented pore space. Then, the distance transform of the segmented pore space and the peaks of the dis-
tance transform were used as arguments for a Watershed segmentation algorithm (Digabel and Lantuejoul 1978). The output of the water-
shed is the pore space of the segmented image dived into pore regions (i.e., regions of the pore space considered to be pores). Finally, we
used a function from the library Porespy to convert pore regions to a pore-network representation by fitting spheres to the pore bodies and
cylinders to the pore throats.

Fig. 11—Examples of pathways connecting opposite sides of an idealized pore network: (a) shortest path between pores in
opposite sides of the pore network and (b) other path between pores in opposite sides of the pore network. The black background
represents the matrix of the rock. White circles and rectangles represent the pore bodies and throats, respectively. The red-dotted
line shows the shortest path, and the blue-dotted line shows another possible path.
For the computation of the shortest path between pores in opposing faces of the sample, we first identified all the pores that belong to
the faces of the cubic sample. This was accomplished by comparing the coordinates of the centers of each sphere (pore) with respect to
the minimum and maximum coordinates in each Cartesian direction among all the pores in the pore network, accounting for a predefined
tolerance value. Then, we used a function from Openpnm to compute the shortest distance between opposing pore pairs using the Dijkstra
algorithm. This process was performed iteratively. Each pore of the starting face is visited once, and all the possible paths connecting the
starting pore to the pores of the opposing face were computed. For each pore, we selected the shortest connecting pathway with an oppos-
ing pore. Finally, we computed the geometrical tortuosity using Eq. 15.
Results
Whole-Core 3D Computed Tomography Scan Images. Segmentation Results. In the case of whole-core 3D CT scan images, we
segmented the samples in four visually identifiable phases. We identified each phase based mainly on gray-scale differences. We trained
the WEKA 3D algorithm individually for each sample and then used the trained algorithm to segment each 3D CT scan image. Fig. 12
shows the segmented phases of Subsample 12 from Sample 1.1. We labeled the most continuous phase as Phase A. This is the phase
that we used for numerical simulations in the whole-core 3D CT scan images. The distribution of the remaining phases (i.e., B, C, and
D) displays heterogeneity in the spatial distribution of the segmented phases despite the apparent continuity of Phase A, which can lead
to rock anisotropy. We observed several subsamples with heterogeneous distribution of the segmented phases in both Sample 1.1 and
Sample 1.2.
Tortuosity Anisotropy Estimation. We conducted numerical simulations of all the physical phenomena documented in the method
section. For this purpose, we divided the segmented whole-core 3D CT scan images in cubic subsamples and used them as the simula-
tion domain. We conducted all the simulations in the most predominant and continuous phase, Phase A (red phase in Fig. 6b), this phase
represents the lithology with second lowest density in the evaluated samples. We verified the continuity of each of the segmented phases
through individual visualization of each segmented phase. Numerical simulations in the most continuous phase were conducted to quan-
tify its spatial distribution. In the case of electric potential distribution, we established a potential difference of 100 V across opposing
faces of the cubic domain in the desired Cartesian direction. The conductivity of the brine in the simulation domain is 20 S/m. In the
random walk simulations, the initial position of the walkers is randomly assigned before the start of the simulations. We ran the random
walk simulations using 1e4 walkers and various number of time steps from 1e4 to 1e7 to guarantee that each walker was allowed to travel
through the entire simulation space. This condition is verified by the convergence of the obtained tortuosity values to an almost constant
value. Fig. 13 shows the behavior of directional diffusional tortuosity against time steps for Sample 1.1 Subsample 8. For all the evalu-
ated samples 1e7 time steps were sufficient to observe convergence in tortuosity values (i.e., the number of time steps is enough for each
walker to travel through the entire sample).
In the case of fluid flow, we ran the simulations at different pressure drops in lattice units to verify laminar flow regime. We verified
this by plotting permeability against pressure drop both in lattice units. Fig. 14a shows the behavior of permeability against pressure
differential in lattice units for Subsample 3 in Sample 1.1. Constant values of permeability for a range of pressure drop values in lattice
units indicate laminar flow regime. For all the evaluated samples, we found that a pressure differential of 5e-5 in lattice units results in
laminar flow regime. Fig. 14b shows the streamlines for the velocity distribution of subsample 3 in Sample 1.1. We used the same simu-
lation parameters for all the simulations in all subsamples from Sample 1.1 and Sample 1.2.
Fig. 15 shows the tortuosity estimates for all subsamples in Sample 1.1 in all three Cartesian directions. In the case of the streamline
electrical tortuosity, the displayed values correspond to the average electrical tortuosity computed over all generated streamlines. In the
other three cases, a single value is obtained for simplified electrical, diffusional, and hydraulic tortuosity directly from the simulation
results. Additionally, an unwrapped CT scan image and a photo of the slabbed whole core are displayed. The black-dashed line number 1
indicates an isotropic interval where the tortuosity computed for Phase A in all three Cartesian directions shows similar values for all the
simulated physical phenomena. On the other hand, the black-dashed line number 4 shows an anisotropic interval where the tortuosity
computed for Phase A in the z-direction differs from the tortuosity values in both x- and y-directions, for all the simulated physical phe-
nomena. The blue-dashed lines number 2 and 3 indicate an interval where Phase A is discontinuous in the z-direction. Discontinuity of
Phase A in any given direction will prevent the simulation process to occur across the sample (e.g., the movement of the walkers in the
case of diffusional tortuosity will be restricted to each of the isolated regions of Phase A in a given subsample). Consequently, we do not
expect to obtain a valid tortuosity value for a subsample for which Phase A is disconnected. In the case of both streamline and simplified

Fig. 12—3D renders of the segmentation results of Subsample 12 from Sample 1.1: (a) Phase A, (b) Phase B, (c) Phase C, and
(d) Phase D.
Fig. 13—Sample 1.1 Subsample 8: directional diffusional tortuosity against number of time steps.
electrical tortuosity, we did not obtain a valid numerical value of tortuosity for intervals 2 and 3. However, in both diffusional and hydrau-
lic tortuosity, we obtained a numerical value in both intervals 2 and 3 which were removed manually. In the case of diffusional tortuosity,
despite obtaining a numerical value, we observed that the coefficient of determination of the linear model fitted to the axial squared dis-
placement in the z-direction was negative, indicating a poor fit of the model to the displacement data. Therefore, the numerical value of
the slope used to estimate diffusional tortuosity does not correspond to the simulation results, and the obtained tortuosity values are not
valid. In the case of hydraulic tortuosity, we also obtained a numerical value. Fluid flow simulations still provide a velocity distribution
array (with negligible values) regardless the discontinuity of the space available to flow. Since hydraulic tortuosity is computed using Eq.
14, it still provides a numerical outcome. Track 4 of Fig. 15 shows image-based rock classes obtained from image features from CT scan
images and the use of a cluster validation index (Gonzalez et al. 2020). The obtained rock classes show a degree of agreement with the
variation of the estimated tortuosity values in the z-direction. However, we do not expect to see a perfect correlation as the image-based
features capture only the spatial distribution of the rock constituents in the plane represented by the 2D CT scan image.

Fig. 14—Sample 1.1, Subsample 3: (a) permeability in lattice units against pressure drop in lattice units and (b) stream lines for
fluid flow in z-direction. Warmer colors indicate higher velocities in lattice units.
Fig. 15—Sample 1.1: Tortuosity estimation. Tracks from left to right: Track 1: depth; Track 2: unwrapped 2D CT scan image; Track
3: slabbed whole-core photo; Track 4: image-based rock classes; Track 5: tortuosity estimates in the x-direction; Track 6: tortuosity
in the y-direction; Track 7: tortuosity in the z-direction. Blue, black, green, and red lines in Tracks 5 through 7 represent streamline
electrical, simplified electrical, diffusional, and hydraulic tortuosity. The black-dashed lines 1 and 4 indicate an isotropic interval
and an anisotropic interval, respectively. The blue-dashed lines 2 and 3 indicate an interval where the most continuous phase is
disconnected in the z-direction.

Fig. 16—Sample 1.2: Tortuosity estimation. Tracks from left to right: Track 1: depth; Track 2: unwrapped 2D CT scan image;
Track 3: slabbed whole-core photo; Track 4: image-based rock classes; Track 5: tortuosity estimates in the x-direction; Track 6:
tortuosity in the y-direction; Track 7: tortuosity in the z-direction. Blue, black, green, and red lines in Tracks 5 through 7 represent
streamline electrical, simplified electrical, diffusional, and hydraulic tortuosity.
Fig. 16 shows the tortuosity estimates for all subsamples in Sample 1.2 in all three Cartesian directions, along with an unwrapped CT
scan image and a photo of the slabbed whole core. The tortuosity values obtained through the introduced simulation techniques show an
overall isotropic depth interval, except for the first 10–15 cm of the retrieved core. Visual analysis of the slabbed whole-core photo and
unwrapped CT scan image displays a more visually homogeneous distribution of the mineral constituents making up the rock when com-
pared with Sample 1.1. It should be noted that no tortuosity values are computed on the lower part of the interval due to severe loss of
continuity of Phase A. Track 4 in Fig. 15 shows the image-based rock classes for the evaluated interval. In this case, we do not observe a
good agreement between the tortuosity estimates and the rock classes. This lack of agreement can be explained by the fact that the image-
based features used for image-based rock classification not necessarily capture the directional nature of the spatial distribution of rock
constituents. However, it should be noted that integration of tortuosity estimates, and image-based features can potentially improve the
estimated rock classes.
The semicontinuous estimates of electrical, diffusional, and hydraulic tortuosity show a similar trend in both examples (Figs. 15 and
16) capturing the vertical variation of the evaluated depth intervals. However, measurable differences in the numerical values of the esti-
mated tortuosity are observed, particularly in the results of Sample 1.1. Figs. 17a and 17b show the mean differences between tortuosity
estimates, computed as the depth-by-depth differences in tortuosity values and averaged over the entire depth interval. The largest differ-
ences are observed between streamline electrical and diffusional tortuosity values in the z-direction in both Sample 1.1 and Sample 1.2
with differences up to 0.51 and 0.03 units, respectively. The smallest differences are observed between streamline electrical and hydraulic
tortuosity with differences of 0.030 tortuosity units in the y-direction and between streamline electrical and diffusional tortuosity of 0.007
tortuosity units in the y-direction for Sample 1.1 and Sample 1.2, respectively. We did not compute the differences between the simplified
electrical tortuosity and the other tortuosity estimates as the simplified electrical tortuosity is virtually equal to diffusional tortuosity. The
largest differences across both samples for all the conducted simulations are observed in the z-direction, which can be related to the
increased heterogeneity in the spatial distribution of the rock components in the z-direction related to the depositional mechanisms of
sediments. Furthermore, the mean difference in tortuosity estimates for Sample 1.2 is two orders of magnitude lower than in the case of
Sample 1.1. This observation can be explained by the homogeneous distribution of Phase A in Sample 1.2, which results in tortuosity
values close to 1.

Fig. 17—Mean differences between hydraulic, streamline electrical, and diffusional tortuosity estimates: (a) Sample 1.1 and
(b) Sample 1.2.
3D Microcomputed Tomography Scan Images. Tortuosity Anisotropy Estimation. We conducted simulations of electric potential
distribution, diffusion, and fluid flow in all three Cartesian directions. Simulations were conducted in the segmented pore space. In
the case of electric potential distribution simulations, we assumed 100% brine saturation and used a 100 V potential difference across
opposing faces of the cubic domain in the Cartesian direction of interest. The conductivity of the brine saturating the pore space is 20 S/m.
In the case of random walk simulations (diffusion), we used 1e4 walkers and 1e7 time steps except for Sample 2.3 for which 3.5e7 time
steps were necessary. We ran the simulations at different pressure drops in lattice units to verify laminar flow regime. For all the evaluated
samples, we found that a pressure differential of 5e-5 in lattice units results in laminar flow regime. For the computation of geometrical
tortuosity, no input parameters are required, other than the segmented image.
Fig. 18 shows tortuosity values obtained for the three samples in each Cartesian direction through numerical simulation of electric
potential distribution (simplified and streamline electrical tortuosity), diffusion, and fluid flow as well as the computed geometrical tortu-
osity. The tortuosity values obtained for Sample 2.1 segmented micro-CT scan image (Fig. 18a) show an isotropic sample, displaying
similar values of tortuosity in all three Cartesian directions for electric, diffusional, hydraulic, and geometrical tortuosity. This behavior
is expected as the Berea Sandstone is a relatively simple formation where no significant anisotropy was expected. Comparison of electric,
diffusional, and hydraulic tortuosity shows a similar behavior as observed for the results of whole-core 3D CT scan images Sample 1.1.
Hydraulic and streamline electrical tortuosity show similar values while diffusional tortuosity shows higher values than hydraulic and
streamline electrical tortuosity. The highest observed difference between hydraulic and diffusional tortuosity in Sample 2.1 is 0.62 units
in the x-direction tortuosity. On the other hand, the largest observed difference between streamline electrical and diffusional tortuosity in
Sample 2.1 is 0.76 units in the z-direction. The largest difference between diffusional and geometrical tortuosity in Sample 2.1 is 0.26 units
in the x-direction.
Fig. 18—Electrical (streamline and simplified), diffusional, and hydraulic tortuosity estimates in segmented 3D micro-CT scan
images: (a) Sample 2.1, (b) Sample 2.2, and (c) Sample 2.3.
In the case of Sample 2.2, the tortuosity values obtained (Fig. 18b) show an anisotropic sample, with measurable differences in the
tortuosity values in all three Cartesian directions for all five types of computed tortuosity values (simplified electrical, streamline electri-
cal, diffusional, hydraulic, and geometrical). Streamline electrical and hydraulic tortuosity show similar values in the x-direction, while
differences of 0.22 and 0.47 units are observed in the y- and z-direction. However, the highest differences are observed when comparing
diffusional with both streamline electrical and hydraulic tortuosity. The highest observed difference between hydraulic and diffusional
tortuosity in Sample 2.2 is 1.21 units in the x-direction. On the other hand, the highest observed difference between streamline electrical
and diffusional tortuosity in Sample 2.2 is 1.66 units in the z-direction. Similarly, the highest difference between geometrical and diffu-
sional tortuosity in Sample 2.2 is 0.76 units in the z-direction.
The tortuosity values obtained for Sample 2.3 (Fig. 18c) show anisotropy in estimates of tortuosity through all the used approaches
(i.e., electrical, diffusional, hydraulic, and geometrical). As in the previous two samples, the highest tortuosity values are observed when
computing diffusional tortuosity. The highest differences between diffusional and the other three types of computed tortuosity occur in the

z-direction, 3.75, 3.26, and 1.65 units for streamline electrical, hydraulic, and geometrical tortuosity, respectively. The values of simplified
electrical tortuosity are almost equal to the diffusional tortuosity values for all three samples.
Discussion
Limitations of the conducted work include the scaling of the results to core- and log-scale domain. In the case of micro-CT scan images,
the size of the scanned samples might be below the size of the representative elementary volume, as such scaling of the results to larger
length scales can be challenging. In the case of the whole-core CT scan images, the resolution of the CT scanner might not be able to
provide clear visual differences between the mineral constituents of the rock, preventing segmentation of the sample into different phases
and the computation of tortuosity. In both whole-core CT scan and micro-CT scan images, it is required that the phase of interest (i.e.,
mineral constituent or pore space) is continuous and connected throughout the simulation domain. Failure to comply with this requirement
can result in numerical errors and nonrealistic tortuosity estimates. In the particular case of hydraulic tortuosity simulations, verification
of laminar flow regime by running simulations at different pressure drops can be time-consuming. Similarly, in the case of random walk
simulations verification of convergence of the tortuosity values by running simulations at different time steps can also be time-consuming.
Finally, in the case of geometrical tortuosity values, the obtained results depend on how well the extracted pore network represents the
actual pore space of the sample. Representation of pore bodies using spheres and pore throats using cylinders is adequate in clastic for-
mations such as sandstones, but it might introduce uncertainty in formations such as carbonates with complex pore geometry.
Numerical differences were observed between the different types of tortuosity estimates. The observed differences can be explained by
differences of the simulated physical phenomena and the assumptions of each type of technique for tortuosity estimation. For instance,
both diffusion and electric potential distribution are mathematically described by Laplace’s equation. Therefore, diffusional and electrical
tortuosity should display similar values as reported in some previous publications (Clennell 1997; Ghanbarian et al. 2013; Fu et al. 2021).
Our results corroborated the previous statement in the case of electrical tortuosity estimated as defined by Herrick and Kennedy (1994)
and diffusional tortuosity. However, the results of streamline electrical tortuosity display values closer to hydraulic tortuosity. The main
difference between fluid flow and electric current flow is the no-slip boundary condition that is present in fluid flow at the interface
between the fluid and the surface of a solid boundary. Nevertheless, the use of streamlines to track the movement of electric charges
resembles the streamlines representing the fluid movement in fluid flow simulations. This resemblance might explain the similarity
between streamline electrical tortuosity and hydraulic tortuosity observed in our results. Geometrical tortuosity estimates do not require
simulation of any physical phenomenon. In the cases documented in this paper, geometrical tortuosity is estimated using a simplified
representation of the pore space. Consequently, the geometrical tortuosity estimates depend on (a) the geometrical shapes used to repre-
sent the pore space and (b) how closely the pore network captures the complexity of the pore space of a given sample. Another possible
cause for the observed numerical differences between tortuosity estimates can be related to the numerical solvers used to simulate various
physical phenomena. Saxena et al. (2017) documented the numerical differences between permeability estimates using different numeri-
cal solvers. They found that as the complexity of the pore space increases, the difference between permeability estimates obtained from
different solvers increases. We expect a similar observation for the case of tortuosity estimates.
The observed differences in tortuosity estimates are relevant for rock physics models that require tortuosity as an input parameter for
the calculation of a petrophysical property. The presented results show that electrical tortuosity estimated using streamlines is similar to
hydraulic tortuosity. On the other hand, electrical tortuosity estimated using the definition by Herrick and Kennedy (1994) shows similar
values as diffusional tortuosity. These observations suggest that the selection of the technique for tortuosity estimation needs to follow a
physics-based approach to minimize the uncertainty in estimation of petrophysical properties using rock physics models that require
tortuosity as an input parameter. Furthermore, the documented observations also suggest that similarities observed between different
tortuosity estimates can be advantageous in scenarios where a particular type of simulation cannot be conducted. For instance, streamline
electrical tortuosity estimates could be used instead of hydraulic tortuosity, in cases where only electrical potential simulations are
possible.
Conclusions
We estimated electrical, diffusional, hydraulic, and geometrical tortuosity in 3D micro-CT scan images and electrical, diffusional, and
hydraulic tortuosity in whole-core 3D CT scan images. We used the direction-dependent tortuosity estimates to assess the anisotropy of
the evaluated samples. In the case of the whole-core 3D CT scan images, the proposed method relies on the existence of at least two
distinguishable phases (mineral constituents) in the raw image slices. This condition is necessary as the objective of the proposed work-
flow is to quantify the spatial distribution of specific rock constituents. The main conclusions are summarized as follows:
• Direction-dependent tortuosity estimates correlated well with the spatial distribution of rock constituents observed in both slabbed
whole-core photos and whole-core 3D CT scan images. Additionally, results of tortuosity from simulation of electric potential distri-
bution, diffusion, and fluid flow showed similar vertical variation in the case of whole-core 3D CT scan images.
• Tortuosity estimates through all the documented techniques in all three Cartesian directions reflected the complexity of the rock fabric
for all three 3D micro-CT scan images from Berea Sandstone, Austin Chalk, and Estaillades limestone formations.
• The observed numerical differences between the techniques used for estimation of tortuosity in both whole-core CT scan images and
micro-CT-scan images highlight the impact of the selection of the technique for estimation of tortuosity. This possibly becomes more
critical when the obtained tortuosity values get used as inputs for estimation of petrophysical properties through tortuosity-dependent
rock physics models.
• Numerical differences between tortuosity estimates were approximately an order of magnitude higher in the case of pore-scale anal-
ysis than in the case of core-scale analysis. Tortuosity values obtained for the micro-CT scan images were in general higher that those
obtained for the whole-core CT scan images. This observation is a consequence of the relatively higher complexity in the spatial
distribution of the pore space in the micro-CT scan images compared with the spatial distribution of the most continuous solid phase
in the whole-core CT scan images.
Nomenclature
a = Dimension of a cubic voxel
ci = Particle velocity, m/s
cs = Speed of sound, m/s
D = Diffusion coefficient of nonabsorbing particles in the porous media, m2/s
D0 = Diffusion coefficient in the free space, m2/s

Dm = Number of dimension where the molecules are moving
E = Electric potential, Volts
fi = Discrete velocity distribution
fieq = Equilibrium distribution function
J = Electric current density, A/m2
Jx = Electric current density in the x-direction, A/m2
K = Boltzmann constant, m2kg/s2K
Lg,i = Length of the shortest path connecting a pair of pores, m
Ls,i = Length of the streamline of each particle, m
L = Length of the sample mm, m
m = Mass of the molecules, kg
n = Total number of cells in the Cartesian simulation domain
ng = Total number of connecting paths
ns = Total number of streamlines
r = Mean squared displacement, m
ρ = Macroscopic density, kg/m3
t = time, seconds
T = Temperature, K
Tx = Electrical transmissibility, Ohm-1
u = Macroscopic velocity vector, m/s
v = Velocity magnitude, m/s
vi = Velocity in the i-th direction, m/s
wi = weighting coefficient
σ = Electrical conductivity, S/m
σrock = Electrical conductivity of the saturated rock, S/m
σx = Electrical conductivity in the x-direction, S/m
σ0 = Electrical conductivity of the brine, S/m
τ = Dimensionless time
τd = Diffusional tortuosity
τd,i = Diffusional tortuosity in the i-th direction
τe,i = Electrical tortuosity in the i-th direction
τg,i = Geometrical Tortuosity in the i-th direction
τh,i = Hydraulic tortuosity in the i-th direction
τr = Relaxation time
‍ ‍ = Connected porosity, v/v
Ωi = Collision operator
Disclaimer
The views and opinions expressed in this paper are those of the authors and do not necessarily reflect those of the Joint Industry Research
Program members or example data owners.
Acknowledgments
Research reported in this paper was funded by the Joint Industry Research Program on “MultiScale Rock Physics” sponsored by
Baker Hughes, BHP, Core Laboratories, Equinor, ExxonMobil, Occidental Petroleum, and Petrobras. Special gratitude goes to Equinor
and partners for providing the data and CT scan images used in this paper, and the permission for publishing this work. The authors
acknowledge the Texas Advanced Computing Center (TACC) at the University of Texas at Austin for providing High Performance
Computing (HPC) resources that have contributed to the research results reported within this paper.
References
Al Mansoori, M., Dernaika, M., Singh, M. et al. 2014. Application of Digital and Conventional Techniques to Study the Effects of Heterogeneity on
Permeability Anisotropy in a Complex Middle East Carbonate Reservoir. Paper presented at the Society of Petrophysicists and Well-Log Analysts 55th
Annual Logging Symposium, Abu Dhabi, United Arab Emirates, 18–22 May. SPWLA-2014-VVVV.
Andrew, M. 2014. Reservoir Condition Pore Scale Imaging of Multiphase Flow Using X-Ray Microtomography. PhD Thesis, Imperial College, London,
UK.
Arganda-Carreras, I., Kaynig, V., Rueden, C. et al. 2017. Trainable Weka Segmentation: A Machine Learning Tool for Microscopy Pixel Classification.
Bioinformatics 33 (15): 2424–2426. https://doi.org/10.1093/bioinformatics/btx180.
Arns, C. H., Knackstedt, M. A., Pinczewski, M. V. et al. 2001. Accurate Estimation of Transport Properties from Microtomographic Images. Geophys Res
Lett 28 (17): 3361–3364.
Auzerais, F. M., Dunsmuir, J., Ferréol, B. B. et al. 1996. Transport in Sandstone: A Study Based on Three Dimensional Microtomography. Geophys Res
Lett 23 (7): 705–708. https://doi.org/10.1029/96GL00776.
Azizoglu, Z. and Heidari, Z. 2021. Interpretation of Multi-Frequency Dielectric Permittivity Measurements for Assessment of Water Saturation in
Carbonate Formations with Complex Pore Structure. Paper presented at the Society of Petrophysicists and Well-Log Analysts 62nd Annual Logging
Symposium, Virtual, 17–20 May. SPWLA-2021-0034. https://doi.org/10.30632/SPWLA-2021-0034.
Barrett, R., Berry, M., Chan, T. F. et al. 1994. Templates for the Solution of Linear Systems: Building Blocks for Iterative Methods. https://doi.org/10.1137/
1.9781611971538.
Berg, S., Saxena, N., Shaik, M. et al. 2018. Generation of Ground Truth Images to Validate Micro-CT Image-Processing Pipelines. Lead Edge 37 (6):
412–420. https://doi.org/10.1190/tle37060412.1.
Canny, J. 1986. A Computational Approach to Edge Detection. IEEE Trans Pattern Anal Mach Intell 8 (6): 679–698. http://dx.doi.org/10.1109/TPAMI.
1986.4767851.
Carman, P. C. 1937. Fluid Flow Through Granular Beds. Trans Inst Chem Eng 15: 150–166.

Chen, H. and Heidari, Z. 2015. Quantifying the Directional Connectivity of Rock Constituents and Its Impact on Electrical Resistivity of Organic-Rich
Mudrocks. Math Geosci 48 (3): 285–303. https://doi.org/10.1007/s11004-015-9595-9.
Chi, L. and Heidari, Z. 2016. Directional-Permeability Assessment in Formations With Complex Pore Geometry With a New Nuclear-Magnetic-Resonance-
Based Permeability Model. SPE Journal 21 (4): 1436–1449. 1449. https://doi.org/10.2118/179734-PA.
Clennell, M. B. 1997. Tortuosity: A Guide Through the Maze. Geol Soc Spec Publ 122 (1): 299–344. https://doi.org/10.1144/GSL.SP.1997.122.01.18.
Digabel, H. and Lantuejoul, C. 1978. Paper presented at the Proceedings of European Symposium on Quantitative Analysis of Microstructures in Material
Science, 85–99, Riederer Verlag, Stuttgart, Germany.
Dijkstra, E.W. 1959. A Note in Two Problems in Connexion with Graphs. Numerische mathematik 1 (1): 269-271.
Dong, H. 2007. Micro-CT Imaging and Pore Network Extraction. PhD thesis, Imperial College, London.
Fredrich, J. T., Greaves, K. H., and Martin, J. W. 1993. Pore Geometry and Transport Properties of Fontainebleau Sandstone. Int J Rock Mech Min Sci
Geomech Abstr 30 (7): 691–697. https://doi.org/10.1016/0148-9062(93)90007-Z.
Fu, J., Thomas, H. R., and Li, C. 2021. Tortuosity of Porous Media: Image Analysis and Physical Simulation. Earth Sci Rev 212: 103439. https://doi.org/
10.1016/j.earscirev.2020.103439.
Garcia, A. P. and Heidari, Z. 2018. Development of a Resistivity Model That Incorporates Quantitative Directional Connectivity and Tortuosity for
Enhanced Assessment of Hydrocarbon Reserves. SPE J. 23 (5): 1552–1565. SPE-181571-PA. https://doi.org/10.2118/181571-PA.
Ghanbarian, B., Hunt, A. G., Ewing, R. P. et al. 2013. Tortuosity in Porous Media: A Critical Review. Soil Sci Soc Am J 77 (5): 1461–1477. https://doi.org/
10.2136/sssaj2012.0435.
Gonzalez, A., Kanyan, L., Heidari, Z. et al. 2020. Integrated Multiphysics Workflow for Automatic Rock Classification and Formation Evaluation Using
Multiscale Image Analysis and Conventional Well Logs. Petrophysics 61 (5): 495–518. https://doi.org/10.30632/PJV61N5-2020a7.
Gonzalez, A., Teymouri, M., Heidari, Z. et al. 2021. Anisotropy Quantification Using High-Resolution Whole-Core CT-Scan Images. Paper presented at
the 2021 SPWLA 62nd Annual Logging Symposium, Virtual, 17–20 May. SPWLA-2021-0087. https://doi.org/10.30632/SPWLA-2021-0087.
Gostick, J., Aghighi, M., Hinebaugh, J. et al. 2016. OpenPNM: A Pore Network Modeling Package. Comput Sci Eng 18 (4): 60–74. https://doi.org/10.
1109/MCSE.2016.49.
Gostick, J., Khan, Z. A., Tranter, T. G. et al. 2019. PoreSpy: A Python Toolkit for Quantitative Analysis of Porous Media Images. J Open Source Softw 4
(37): 1296. https://doi.org/10.21105/joss.01296.
Herrick, D. and Kennedy, W. 1994. Electrical Efficiency A Pore Geometric Theory for Interpreting the Electrical Properties of Reservoir Rocks. Geophysics
59 (6): 918–927.
Hough, P. V. 1962. Method and Means for Recognizing Complex Patterns. U.S. Patent 3,069,654.
Kozeny, J. 1927. Über Kapillare Leitung Des Wassers Im Boden‐Aufstieg. In Versickerung Und Anwendung Auf Die Bewässerung.
Krüger, T., Kusumaatmaja, H., Kuzmin, A. et al. 2017. The Lattice Boltzmann Method, Vol. 10, 4–15. Cham, Switzerland: Springer International Publishing.
Latt, J., Malaspinas, O., Kontaxakis, D. et al. 2021. Palabos: Parallel Lattice Boltzmann Solver. Comput Math with Appl 81: 334–350. https://doi.org/10.
1016/j.camwa.2020.03.022.
Matyka, M. and Koza, Z. 2012. How to Calculate Tortuosity Easily? Paper presented at the Porous Media and Its Applications in Science, Engineering,
and Industry, Vol. 1453, 17–22, Potsdam, Germany. https://doi.org/10.1063/1.4711147.
Misra, S. and Wu, Y. 2019. Machine Learning Assisted Segmentation of Scanning Electron Microscopy Images of Organic-Rich Shales with Feature
Extraction and Feature Ranking. In Machine Learning for Subsurface Characterization, 289.
Mohamad, A. A. 2011. Lattice Boltzmann Method, Vol. 70, Vol. 70. London, UK: Springer.
Nakashima, Y. and Kamiya, S. 2007. Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity and Anisotropic Tortuosity of
Porous Rocks Using X-Ray Computed Tomography Image Data. J Nucl Sci Technol 44 (9): 1233–1247. https://doi.org/10.1080/18811248.2007.
9711367.
Pollock, D. W. 1988. Semianalytical Computation of Path Lines for Finite-Difference Models. Ground Water 26 (6): 743–750. https://doi.org/10.1111/j.
1745-6584.1988.tb00425.x.
Prodanovic, M., Esteva, M., Hanlon, M. et al. 2015. Digital Rocks Portal: A Repository for Porous Media Images, https://doi.org/10.17612/P7CC7K.
Rassenfoss, S. 2011. Digital Rocks Out to Become a Core Technology. J Pet Technol 63 (5): 36–41. SPE-0511-0036-JPT. https://doi.org/10.2118/0511-
0036-JPT.
Saxena, N., Hofmann, R., Alpak, F. O. et al. 2017. References and Benchmarks for Pore-Scale Flow Simulated Using Micro-CT Images of Porous Media
and Digital Rocks. Adv Water Resour 109: 211–235. https://doi.org/10.1016/j.advwatres.2017.09.007.
Schindelin, J., Arganda-Carreras, I., Frise, E. et al. 2012. Fiji: An Open-Source Platform for Biological-Image Analysis. Nat Methods 9 (7): 676–682.
https://doi.org/10.1038/nmeth.2019.
Spanne, P., Thovert, J. F., Jacquin, C. J. et al. 1994. Synchrotron Computed Microtomography of Porous Media: Topology and Transports. Phys Rev Lett
73 (14): 2001–2004. https://doi.org/10.1103/PhysRevLett.73.2001.
Sun, H., Vega, S., and Tao, G. 2015. Study on Permeability Anisotropy in Carbonate Reservoir Samples Using Digital Rock Physics. Paper presented
at the Abu Dhabi International Petroleum Exhibition and Conference, Abu Dhabi, UAE, 9–12 November. SPE-177540-MS. https://doi.org/10.2118/
177540-MS.
Tranter, T. G., Kok, M. D. R., Lam, M. et al. 2019. Pytrax: A Simple and Efficient Random Walk Implementation for Calculating the Directional Tortuosity
of Images. SoftwareX 10: 100277. https://doi.org/10.1016/j.softx.2019.100277.
Van der Walt, S., Schönberger, J. L., Nunez-Iglesias, J. et al. 2014. Scikit-Image: Image Processing in Python. PeerJ 2, e543. https://doi.org/10.7717/peerj.
453.
Wang, Y., Lin, C. L., and Miller, J. D. 2015. Improved 3D Image Segmentation for X-Ray Tomographic Analysis of Packed Particle Beds. Miner Eng 83:
185–191. https://doi.org/10.1016/j.mineng.2015.09.007.

Spe 206109 Pa

Uploaded by

Copyright:

Available Formats

Spe 206109 Pa

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Spe 206109 Pa

Uploaded by

Copyright:

Available Formats

Electrical, Diffusional, Hydraulic, and

Geometrical Tortuosity Anisotropy

*Corresponding author; email: zoya@utexas.edu

February 2023 SPE Reservoir Evaluation & Engineering 23

Materials and Method

Fig. 1—Workflow for estimation of direction-­dependent tortuosity and rock anisotropy.

The main steps of the proposed workflow are summarized as follows:

24 February 2023 SPE Reservoir Evaluation & Engineering

February 2023 SPE Reservoir Evaluation & Engineering 25

26 February 2023 SPE Reservoir Evaluation & Engineering

February 2023 SPE Reservoir Evaluation & Engineering 27

28 February 2023 SPE Reservoir Evaluation & Engineering

February 2023 SPE Reservoir Evaluation & Engineering 29

30 February 2023 SPE Reservoir Evaluation & Engineering

February 2023 SPE Reservoir Evaluation & Engineering 31

32 February 2023 SPE Reservoir Evaluation & Engineering

February 2023 SPE Reservoir Evaluation & Engineering 33

34 February 2023 SPE Reservoir Evaluation & Engineering

February 2023 SPE Reservoir Evaluation & Engineering 35

36 February 2023 SPE Reservoir Evaluation & Engineering

February 2023 SPE Reservoir Evaluation & Engineering 37

38 February 2023 SPE Reservoir Evaluation & Engineering

February 2023 SPE Reservoir Evaluation & Engineering 39

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Fig. 1—Workflow for estimation of direction-dependent tortuosity and rock anisotropy.