Download as pdf or txt
Download as pdf or txt
You are on page 1of 31

Transport in Porous Media (2019) 130:305–335

https://doi.org/10.1007/s11242-018-1201-4

Porous Media Characterization Using Minkowski Functionals:


Theories, Applications and Future Directions

Ryan T. Armstrong1 · James E. McClure2 · Vanessa Robins3 · Zhishang Liu1 ·


Christoph H. Arns1 · Steffen Schlüter4 · Steffen Berg5,6

Received: 15 May 2018 / Accepted: 15 November 2018 / Published online: 27 November 2018
© Springer Nature B.V. 2018

Abstract
An elementary question in porous media research is in regard to the relationship between
structure and function. In most fields, the porosity and permeability of porous media are
properties of key interest. There is, however, no universal relationship between porosity and
permeability since not only does the fraction of void space matter for permeability but also the
connectivity of the void fraction. With the evolution of modern day X-ray microcomputed
tomography (micro-CT) and advanced computing, it is now possible to visualize porous
media at an unprecedented level of detail. Approaches in analyzing micro-CT data of porous
structures vary in the literature from phenomenological characterization to network analysis
to geometrical and/or topological measurements. This leads to a question about how to
consistently characterize porous media in a way that facilitates theoretical developments. In
this effort, the Minkowski functionals (MF) emerge from the field of statistical physics where
it is evident that many physical processes depend on the geometry and topology of bodies or
multiple bodies in 3D space. Herein we review the theoretical basis of the MF, mathematical
theorems and methods necessary for porous media characterization, common measurement
errors when using micro-CT data and recent findings relating the MF to macroscale porous
media properties. This paper is written to provide the basics necessary for porous media
characterization and theoretical developments. With the wealth of information generated from
3D imaging of porous media, it is necessary to develop an understanding of the limitations
and opportunities in this exciting area of research.

B Ryan T. Armstrong
ryan.armstrong@unsw.edu.au
1 School of Minerals and Energy Resoureces Engineering, University of New South Wales, Sydney
2033, Australia
2 Advanced Research Computing, Virginia Tech, Blacksburg, VA 24061, USA
3 Research School of Physics and Engineering, Australian National University, Canberra, ACT
2601, Australia
4 Helmholtz-Centre for Environmental Research, Theodor-Lieser-Str. 4, 06120 Halle, Germany
5 Shell Global Solutions International B.V., Kesslerpark 1, 2288 GS Rijswijk, The Netherlands
6 Department of Earth Science and Engineering, Department of Chemical Engineering, Imperial College
London, London, UK

123
306 R. T. Armstrong et al.

Keywords Minkowski functionals · X-ray microcomputed tomography · Pore morphology ·


Euler characteristic · Persistence homology

1 Motivation and Aims

The relationship between porous media properties and characteristics of the pore space is one
of the most elementary questions in porous media research. In many porous media disciplines
such as biomedical engineering, hydrology and petroleum engineering, the key properties of
interest are porosity and permeability. The transport of pericellular fluid through biological
tissue is largely important for targeted medical treatments (Anderson et al. 2008). Bone,
for example, is a multi-scale structure of physiological tissue and calcium hydroxyapatite
through which cellular response networks provide feedback loops that influence porosity
and thus, permeability (Knothe 2003; Tate et al. 2009; Knackstedt et al. 2006). Generations
of petrophysicists have studied the relationship between porosity and permeability, which
appears to follow a power law relationship for the same rock with varying degrees of poros-
ity (Tiab and Donaldson 2015). Similarly, capillary pressure of the same sandstone rock can
be scaled by the square root of porosity and permeability (Leverett 1941). However, these
relationships cannot be transferred from one rock type to another and often break down
completely for complex porous systems, such as carbonate rock (Dullien 2012). This clearly
shows that the characteristics of porous materials should be defined through more than poros-
ity and permeability. Permeability depends on more than porosity because not only does the
volume fraction of void space matter but also its connectivity (Blunt 2017). Researchers have
attempted to characterize porous media by multiple different measures (Sahimi and Flow,
2011; Tiab and Donaldson 2015). Most of the measures are specifically related to a property
of interest. Dispersivity, for instance, is closely related to the tortuosity of the rock (Sahimi
and Flow, 2011; Bijeljic et al. 2011), or the permeability of bone tissue is related to the
existence of cellular inclusions (Anderson et al. 2008). Despite decades of research on this
topic, so far no complete set of measures to characterize porous media properties have been
proposed (Liu et al. 2017; Schlüter et al. 2016). Part of the difficulty is that in many cases
properties depend on an unknown set of variables and parameters and their interdependencies
are not fully understood. This occurs because a fundamental theory that covers microscopic
and macroscopic (Darcy scale) properties and respective constitutive relationships derived
from first principles are in many cases not available, or cover only limiting cases.
With these aspirations in mind, it would be beneficial to have a characterization scheme that
fulfills uniqueness and completeness. Geometric characterization of porous media, or fluid
phases therein, can potentially be achieved under these requirements using the Minkowski
functionals (MF) (Mecke 2000). The MF can be used as a characterization tool for many
different types of porous media ranging from a mouse leg to fiber-reinforced concrete to
shale rock to coal to porous glass filters, and example images of these systems are provided
in Fig. 1. The functionals are derived from integral geometry where they are known as intrinsic
volumes or quermassintegrale. They provide a single value measure by integrating over the
surface of an object or the surfaces of many objects (Mecke 1998). In this way, the MF can
be used to quantify the bulk geometry of an object from microscale measurements, providing
a convenient framework for theoretical developments.
The idea of using the MF to study complex morphologies emerged from the field of statis-
tical physics where it was evident that numerous physical processes depend on the geometry
and spatial distributions of bodies in 3D space (Mecke 2000; Schmalzing et al. 1995). The

123
Porous Media Characterization Using the Minkowski… 307

(a) (b) (c)

(e)

(f)
(d)

Fig. 1 Examples of complex porous media: mouse leg (sample courtesy of Prof. Melissa Knothe Tate and Dr.
Tzong-Tyng Hung, University of New South Wales) (a), shale rock (Zhang et al. 2018) (b), coal (Ramandi
et al. 2016) (c), sandstone (Liu et al. 2017) (d), fiber-reinforced concrete (sample courtesy of Prof. Stephen
Foster, University of New South Wales) (e), and two-phase flow (Armstrong et al. 2014) (f, glass = white, gray
= non-wetting phase, dark gray = wetting phase)

MF have been used to study the distribution of galaxies, crystallization in granular packings,
pile stability, spinodal decomposition, phase transitions and percolation phenomenon (Baña-
dos et al. 1994; Schmalzing and Górski 1998; Mecke and Wagner 1991; Mecke and Sofonea
1997; Scheel et al. 2008; Saadatfar et al. 2017; Mecke 2000). A classic example is found in the
work of M. Kac where the question “Can you hear the shape of a drum?” is posed and studied
using MF and the frequencies of normal mode vibrations of the drum membrane(Kac 1966).
Here the drum shape is described through a linear combination of the MF and provides a
direct relationship between structure and function. In addition, the MF can be used to develop
kinematic relationships for the motion and/or evolution of a single body or a complex system
of multiple bodies (Mecke 2000). It is important to note that these developments are more
than an arbitrary correlation since there are geometrical/mathematical reasons as to why a
decomposition is possible. The MF can also be used to study the interrelationship between
structure and function, e.g., complex morphologies that are the result of physical processes
and/or bulk behaviors that depend on microscale structure (Mecke and Sofonea 1997).
We will provide an introduction to the theoretical basis of the MF. We will then address
pragmatic issues of measuring the MF of porous systems using X-ray microcomputed tomog-
raphy (micro-CT) data. The presented examples will be based on using micro-CT data since
it provides an unprecedented level of detail of porous structure (Wildenschild and Sheppard
2013; Blunt et al. 2013). We will also provide examples on how the MF and theoretical frame-
work therein can be applied to study porous systems. Our examples will be based on simple
porous networks or model porous media because these systems are the most instructive for

123
308 R. T. Armstrong et al.

developing concepts and illustrating the utility of the MF. To expand upon these examples,
we will provide a review of selected works where porous rocks and soils, multiphase flow,
granular media packings and biomaterials have been studied using the MF. This paper is not
intended to be a comprehensive review of the MF, applications of MF in statistical physics
nor micro-CT imaging. For these topics the informed reader should refer to the following
publications (Mecke 2000; Mecke and Sofonea 1997; Wildenschild and Sheppard 2013;
Wildenschild et al. 2002; Blunt et al. 2013). This paper is intended to provide an introduction
to porous media characterization using the MF with the main concepts and mathematical
theorems applied to simple porous systems. Numerous works have used the MF to charac-
terize and/or study porous media systems, and it is not possible to acknowledge all of these
excellent publications. We find it instructive to cover a limited selection of publications with
sufficient detail to facilitate future work rather than covering the entire breadth of the avail-
able literature. Our aim is to provide the foundation necessary to engage a reader to apply
the presented concepts to increasingly complex porous systems that cover a broad range of
disciplines.

2 Basic Principles and Applications

We will first cover the basic definitions of the MF, which lay out the mathematics for quan-
tifying an object or collection of objects. Then, a few outstanding theorems are introduced
that demonstrate the utility of the MF, starting with the Gauss–Bonnet theorem, which relates
the total surface curvature of an object to its topology. This provides a type of topological
constraint to a range of possible deformations and/or movements for a given object. Take, for
instance, a pyramid made of flexible poles and now try to bring the vertices together. Intu-
itively, we know from experience that it is impossible to bring the vertices together without
bending the poles. The Gauss–Bonnet theorem, however, explains mathematically how the
curvature that was originally concentrated near the vertices of the pyramid is transferred to
the bent poles as the vertices come together.
Topology is described by the Euler characteristic, which is independent of geometry
meaning that how big or small things are does not matter. A classical example is that a donut
and coffee mug are equivalent in terms of topology. As seen by this example, topology alone
is not enough to uniquely identify an object. A more complete description of an object is
provided by Hadwiger’s theorem, which shows that in d-dimensional space there exists a
basis set of d + 1 intrinsic values that quantify geometry. In three-dimensional (3D) space
these intrinsic values are volume, surface area, integral mean curvature and total curvature or
Euler characteristic, collectively called the Minkowski functionals. The theorem goes further
to show that any other continuous and invariant valuation that can be envisaged must be
uniquely described by a linear combination of the MF since they are a complete set. This
is a powerful statement that is now routinely applied in theoretical physics but not fully
recognized by the porous media community.
Many of the parameters used for solving scientific problems are invariant and continuous.
A common example would be the energy of a system (Hyde et al. 1990). From the previous
pyramid example, we now let the flexible poles bend enough such that a sphere is formed.
The energy stored in the sphere could be derived from fundamental mechanics. But what
if the fundamental mechanics are unknown or yet to be discovered? Hadwiger’s theorem
provides an alternative approach whereby energy can be described by a linear combination
of the intrinsic values (or MF) of our object. The constants of this linear combination could

123
Porous Media Characterization Using the Minkowski… 309

be found by experiments where the force to deform an object is measured along with the
resulting intrinsic volumes. This provides a powerful approach to solving complex physical
problems (Mecke 2000). This outstanding theorem is explained in the following sections
along with other important theorems and concepts required for applying the MF to porous
media-related problems. We then provide an example on how the MF can be measured using
segmented micro-CT data since this is the most common format for studying the structure of
porous media. Furthermore, we present topological characterization using persistent homol-
ogy, which provides a rich and robust method for morphological characterization beyond that
offered by the MF alone. The mathematics of integral geometry explained in the following
section applied to the context of porous media provide many exciting research opportunities.

2.1 Minkowski Functionals: Volume, Surface Area and Curvatures

The MF are geometric measures of size based on set theory (Serra 1983). Here we will cover
the basic formulations and theorems used in porous media research; more in-depth analyses
can be found in Santaló (2004); Schneider (2014) and other references in the following text.
By considering a solid body (X) embedded in Euclidean space () with boundary surface
δ X , integral geometry provides d + 1 functionals where d is the dimension of . Below,
we describe the case d = 3. The first functional M0 is the total volume of X . The second
functional M1 is the integral measure of the surface area of X , defined as,

M1 (X ) = ds (1)
δX
where ds is a surface element on X and the entire surface is denoted as δ X . The third functional
is the integral of mean curvature of δ X , defined as,

M2 (X ) = [1/r1 + 1/r2 ]ds (2)
δX
where r1 and r2 are the principal radii of curvature of surface element ds. Lastly, the fourth
functional is the integral of Gaussian curvature of δ X , defined as,

M3 (X ) = [1/(r1 r2 )]ds = 2πχ(δ X ) = 4πχ(X ) (3)
δX
where χ(δ X ) is the Euler characteristic of the bounding surface and χ(X ) is the Euler
characteristic of the solid object (Ohser and Mücklich 2000; Michielsen and De Raedt 2001);
the details are explained further along in this section. This complete set of MF for any 3D
object measures the size of X in various dimensions: volume (l 3 ), surface area (l 2 ) integral
of mean curvature (l) and a topological invariant called the Euler characteristic. Also note
that the MF are extensive properties and there exist numerous normalization approaches to
account for volume (Schlüter et al. 2011; Liu et al. 2017; Herring et al. 2015; Arns et al.
2001).
The Euler characteristic can also be found as a linear combination of the number of isolated
objects (N), redundant loops (L) and cavities (O), known as the Betti numbers.
χ(X ) = N − L + O (4)
For example, a solid sphere is a single object with zero redundant loops and zero cavities
while a spherical shell has one cavity so that the Euler characteristics of the solid and hollow
spheres according to Eq. 4 are χ(X ) = 1 and χ(δ X ) = 2, respectively. Additional examples
are given in Fig. 2.

123
310 R. T. Armstrong et al.

The factor of 2 between the hollow χ(δ X ) and the solid χ(X ) objects in Eq. 3 is often
found in the porous media literature without much explanation. We rectify this here, starting
with the example of a sphere of radius R, where is it clear that the integral of Gaussian
curvature over the surface is M3 (X ) = 4π R 2 /R 2 = 4π = 4πχ(X ) = 2πχ(δ X ). The
relationship that χ(δ X ) = 2χ(X ) also holds whenever X is a closed bounded solid object in
3D space by the following argument. From the topological (or clay-modeler’s) view point,
any such solid is built by adding some number of handles, L, to the initial solid blob. The
object still has just one piece N = 1 and no cavities, O = 0, so χ(X ) = 1 − L. The surface
δ X is a hollow version of X so has N = 1 but now O = 1. The number of loops doubles
because each handle contributes one loop along the handle and one around the neck, so now
χ(δ X ) = 1 − 2L + 1 = 2 − 2L = 2χ(X ). Intuitively, this makes sense because the solid
object has a surface with one side while the hollow object has a surface with two sides,
resulting in additional loops. It is important to note that as the number of loops increases
the Euler characteristic becomes more negative for both the solid χ(X ) and hollow χ(δ X )
objects.
The relationship of total Gaussian curvature to Euler characteristic, used in Eq. 3 is the
celebrated Gauss–Bonnet theorem. This theorem enforces a type of topological constraint
or kinematic relationship to certain types of deformations, as discussed in Sect. 4.3. The
theorem is formulated for any smooth surface, S, that may be closed or have some number
of boundary curves, B. “Smooth” means that each location on the surface is represented
by continuous and differentiable vector fields U and V so that the two principal radii of
curvature, (r1 , r2 ), are defined at each point on S. The Gauss–Bonnet theorem is stated as,
 
[1/(r1 r2 )]ds + k g dl = 2πχ(S) (5)
S B

where ds is an area element, dl is a line element along the boundary B and k g is the geodesic
curvature (Abbena et al. 2017) along the boundary. The contribution of the geodesic curvature
source term allows for the result to be broadly generalized. The relationship will hold even
when the surface bounding an object is not smooth. For cases when an object is closed with no
boundary the geodesic curvature can be omitted. This is typically the case for our applications
where S = δ X , meaning that the entire surface is smooth with no boundary with another
surface creating a common line of contact, i.e., boundary B. It is intriguing to note that the
integral of local Gaussian curvature over a surface results in a dimensionless term χ(S) that
has no geometrical basis and only depends on the number of objects, loops and cavities. This
type of valuation is distinctly different from those used in our daily life such as distance and
size.
Another defining property of the Minkowski functionals is additivity, which can be demon-
strated by the union (∪) of two bodies (X 1 and X 2 ) in 3D space (). To determine the
Minkowski functional (Mi ) for a union, we simply add the functional for both bodies and
subtract the functional of the intersection (∩), formally defined as,

Mi (X 1 ∪ X 2 ) = Mi (X 1 ) + Mi (X 2 ) − Mi (X 1 ∩ X 2 ) (6)

where this unique feature allows for dealing with boundaries in a concise way. For example, in
porous systems the MF of a given object or the interface between two objects can be uniquely
defined by using additivity, as shown in Sect. 4.2. In addition, additivity makes determination
of the MF on digital images straightforward and embarrassingly parallelizable, as discussed
in Sect. 2.2.

123
Porous Media Characterization Using the Minkowski… 311

Fig. 2 Basic topological features and resulting Euler characteristic for the following solid and hollow objects:
torus, double torus and triple torus (N, L, O)

An extraordinary theorem was given by (Hadwiger 1957) demonstrating that any addi-
tive motion-invariant and conditionally continuous functional F of a set in Euclidean
d-dimensional space is a linear combination of the d + 1 MF. An additional proof of this
theorem is found in Klain (1995). For d = 3 we have,


3
F(X ) = ci Mi (X ) (7)
i=0

where ci is a coefficient that is independent of object (X). Motion-invariant means that F(X )
cannot depend on the orientation of X . The theorem means that the MF make a complete
set and that any other additive functional that meets the criteria of continuous and motion-
invariant can be described by a linear combination of the MF. For example, F could be the
energy associated with a droplet of oil, a cell membrane (Hyde et al. 1990) or any other
type of interface (Mecke 2000). For porous media applications, we often only deal with
homogeneous and isotropic systems where the orientation of the porous media X does not
influence F. In addition, recent studies have used simulated Poisson or Gaussian distribution
models that meet the criteria of motion-invariance to study porous structures (Arns et al. 2001;
Scholz et al. 2012; Mecke 2000). How close to reality these systems represent real porous
media is debatable. Also, how strict the requirement of motion-invariance is for engineering
applications remains an open question. The power of Hadwiger’s theorem is that it proves
that the Minkowski functionals are a fundamental set of quantities from which any other
additive physical functional may be built. One might question how many useful physical
quantities meet the criteria for F, but such parameters are ubiquitous in nature and many
examples exist (Mecke 2000).
Another important formula that predates Hadwiger and likely set the foundation for his
work is Steiner’s formula. This formula relates the Lebesgue measure λ of a convex object in
n-dimensional Euclidean space to geometrical measures of its surface (Federer 1959). The
Lebesgue measure for a bounded and closed object in 3D is volume and area in 2D or simply
M0 . The other geometrical measures are the remaining MF. In general, Steiner defines how
geometrical properties of an object change as the object is deformed by a kernel that creates
a parallel set with positive reach. The formula in 3D is defined as,


3
λ(X ⊕ ∂r ) − λ(X ) = ai Mi r i (8)
i=1

123
312 R. T. Armstrong et al.

where ∂r is a spherical ball with radius, r , and ⊕ denotes the Minkowski sum of X and ∂r .
Steiner’s formula can predict the volume of the parallel set, X ⊕ ∂r , for a particular X . To
understand this formula imagine how the volume of an object would change, λ(X ⊕ ∂r ) −
λ(X ), when a small ball is rolled over the entire surface of an object leaving behind a layer
of material with thickness r . For segmented micro-CT data, a parallel set X ⊕ ∂r could be
generated by running a dilation algorithm with a kernel defined as ∂r . The coefficients ai
are determined by the morphology of X according to the MF, Mi . The constraint of positive
reach means that a ball with a positive radius could be rolled over the surface of X without
intersecting X . This is true for many objects when r is sufficiently small. This is useful
because the volume of a object can be defined by its area, mean curvature and total curvature
meaning that there are only 3 independent variables in 3D space. In specific cases, one of the
independent variables may drop out making for a simplified form. An example is provided
by Weyl’s formula derived for tubes of constant radius, resulting in the mean curvature term
dropping out and the local volume change being predicted from only surface area and Euler
characteristic (Weyl 1939). Furthermore, note that while the MF are independent they are
not orthogonal meaning that for specific applications there could exist correlations between
the MF. Lastly, there is no strict reason why one or more of the coefficients, ai , could not be
zero for specific limiting cases.
A very simple application of Steiners formula is to define how the surface area of a sphere
changes with volume. Take, for example, a sphere, X , of radius 1, which is dilated by ∂r of
radius 1. The MF are simply M0 = 4/3πr 3 , M1 = 4πr 2 , M2 = 8πr and M3 = 4π. The
change in volume caused by dilating the sphere by unit volume 1 is λ(X ⊕ ∂r ) − λ(X ) =
π28/3. Steiner’s formula states that there exists constants ai that satisfy 28 = a1 12 + a2 24 +
a3 12. One possible condition is a1 = 1/3, a2 = 1 and a3 = 0. This solution demonstrates
that the mean curvature of the sphere could be determined by its surface area and change in
volume, stated as,
Vol − M1 r /3
M2 = (9)
r2
where Vol is simply λ(X ⊕ ∂r ) − λ(X ) and r is the radius of the spherical ball used to dilate
our sphere, X . An advanced application of this example is provided by Ohser et al. (2011)
where the curvature of an interface is defined by its change in surface area when dilated by
∂r . Another application is where the length density of plant roots is estimated from surface
area (M1 ) and the integral of mean curvature (M2 ) of segmented micro-CT images of root
networks (Koebernick et al. 2014). For multiphase flow in porous media, this could provide an
alternative means to measure capillary pressure through the Young–Laplace equation when
the direct measurement of mean curvature is not possible or inherently difficult. Additional
applications include the generation of a geometric state function for two-fluid flow in porous
media (McClure et al. 2018), the estimation of free energy stored in a porous system with
two fluids [48] and estimations of percolation thresholds (Mecke and Wagner 1991). Lastly,
it is noteworthy to mention that the decomposition of λ(X ⊕ ∂r ) − λ(X ) into a finite sum of
the MF is a direct consequence of additivity.

2.2 Measuring the Minkowski Functionals from Segmented Micro-CT Data

The MF of any object can be determined by considering the local binary pattern of ones
and zeros (Ohser and Mücklich 2000). This is because additivity allows for the partitioning
of an image into overlapping sets. Consider a 2D image where the MF are effectively area,
perimeter length and Euler characteristic. When the image is partitioned into overlapping

123
Porous Media Characterization Using the Minkowski… 313

Fig. 3 The six possible isotropic configurations considering a two-by-two pixel window in a segmented 2D
image. The light gray region signifies the window of observation, and each of the smaller squares is pixels.
The white and black colors represent a binary image of either one or zero, respectively

sets of two-by-two pixels, there are sixteen possible pixel arrangements ( A) of which there
are six isotropic configurations as defined in Fig. 3. 
The Minkowski functionals are easily determined by summing ( ) the occurrences for
each arrangement (N Ai ),
     
M0 = N A1 + 2 N A2 + 3 N A3 + 4 N A4 + 2 N A5 /4 (10)
    
M1 = N A1 + N A2 + N A3 + 2 N A5 (11)
   
χ= N A1 − N A3 + 2 N A5 /4 (12)

where M0 is area, M1 is perimeter length and χ is Euler characteristic. For A1 , there is


a single pixel resulting in one unit area, while in A4 , there are four pixels resulting in four
unit area. The total summation for all occurrences is then divided by four when consider-
ing a neighborhood connectivity of four. Here each pixel is counted four times due to the
requirement of overlapping windows to account for the perimeter lengths and curvatures that
exist between adjacent windows. Perimeter length is determined by counting the internal
length within each window, considering that the entire window length is a single unit area.
Configurations A1 , A2 and A3 all have a unit length of one, while A5 has a unit length of
two. Since only the internal length of a window is considered, there is no division by four.
To determine curvature, we consider only the internal region of each window and count each
corner as π/2 curvature. The Gauss–Bonnet theorem states that total curvature is equal to
2πχ, and thus, dividing the total curvature by 2π results in Eq. 12. The formulas provided are
for instructional purposes only and thus, are the most basic of various possible formulations.
When an object is digitized into a Cartesian grid, there are biases in considering geometrical
measurements and there are numerous papers that propose approaches to circumvent these
issues providing less biased estimates (Lang et al. 2001; Nagel et al. 2000; Legland et al.
2011; Mantz et al. 2008; Ohser and Mücklich 2000). A similar approach can be applied

123
314 R. T. Armstrong et al.

Table 1 Euler characteristic is measured on a synthetic dataset of spherical objects using various softwares
Software χ Notes on boundary rule

Avizo 9.0 216 Unknown boundary rule


PerGEOS 216 Unknown boundary rule
ImageJ 216 Including boundary objects
ImageJ 64 Excluding boundary objects
Quantim 125 Applies stereological rule to boundaries (Nagel et al. 2000)
Mango 125 Applies stereological rule to boundaries (Nagel et al. 2000)
The measured value depends on how the software handles boundary objects. In the synthetic data, there are
64 spheres enclosed within the image and 152 spheres that touch the boundary in some way or another

to 3D systems by partitioning the image into 23 unit cells resulting in 256 possible voxel
configurations of which there are 22 isotropic configurations (Arns et al. 2001; Lang et al.
2001).
When considering an object or collection of objects that are not closed surfaces, e.g., a
porous media image that is cropped, there are different approaches to handle boundary voxels.
Currently available softwares take slightly different approaches for handling boundary voxels.
For instructional purposes, we created a synthetic dataset (2503 ) with a regular grid of 64
spheres (43 ) that are completely surrounded by background. Another 152 spheres touch the
boundary in one way or another. In total, there are 216 objects. The measurement of Euler
characteristic for the synthetic dataset varies depending on the software and setting used, see
Table 1. For this analysis, we considered that voxels sharing a common face are connected
(six-neighborhood connectivity).
The particle analyzer in ImageJ provides two possible results depending on the user
option to include or exclude boundary objects. Quantim and Mango are consistent because
the stereological rule to count all objects that contact the top, left and front boundary but
to exclude all objects that contact the bottom, right and back boundary (Nagel et al. 2000)
is used. Avizo 9.0 and PerGEOS appear to over count boundary objects. While we are
unaware of documentation that details how Avizo 9.0 and PerGEOS deal with boundaries
when computing Euler characteristic, our results suggest that these algorithms consider nearly
all objects that are in contact with all boundaries. Lastly, another point of consideration is
connectivity. For a 3D Cartesian grid of voxels, either 6-neighborhood or 26-neighborhood
connectivity is commonly considered. For a segmented sandstone image, depending on the
type of connectivity used, we get χ26 = − 4720 or χ6 = − 4779 using Quantim and this
trend would be the case for other softwares as well. Overall, the software and settings used
will influence the value of the Euler characteristic measured, and implications should be
considered on a case-by-case basis.
Regardless of the algorithm used, the measurement of Euler characteristic is inherently
troubled with the occurrence of isolated pixels resulting from segmentation error. Each iso-
lated object counts as a single object, as defined in Eq. 4. For segmented multiphase data,
as presented in Fig. 4a, the wetting phase has numerous disconnected objects (red voxels)
providing a positive Euler characteristic even though the wetting phase is known to be con-
nected across the domain. This is a common problem when measuring the Euler characteristic
of multiphase data and in particular at low wetting phase saturations. This occurs because
wetting fluids tend to exist in the smaller corners and crevices of the rock, which are near the
image resolution (Wildenschild and Sheppard 2013). In Fig. 4b, we performed a sensitivity

123
Porous Media Characterization Using the Minkowski… 315

(a) (b)

Fig. 4 Multiphase image (a) of water (dark gray) and oil (light gray) within a porous glass filter (white). Due
to image resolution, wetting phase films (red voxels) are not well resolved and result in many isolated clusters.
Isolated clusters provide a positive Euler characteristic even though the wetting phase is connected (b). As
small isolated clusters are removed, the Euler characteristic converges to a constant negative value

analysis by removing sequentially larger and larger isolated clusters from a segmented image
of wetting phase at low saturation. Once clusters of 10–100 voxels in size are removed, the
wetting phase provides a constant negative Euler characteristic, as expected since the wetting
phase was known to be percolating across the image domain. This approach has been used
in a few recent publications (Herring et al. 2013, 2015; Berg et al. 2016). The approach,
however, should be performed judiciously since specific porous systems may indeed have
realistic isolated regions that should not be removed.

2.3 Persistent Homology Applied to Micro-CT Data

Homology provides an algebraic, combinatorial, and therefore computable, method for iden-
tifying and counting the Betti numbers (N , L, O) of an object. It is therefore closely related
to the Euler characteristic but calculates each Betti number quantity independently. This pro-
vides an alternative to the method of measuring topology via the total curvature as discussed
in the previous section. Having the values for each Betti number, rather than their alternat-
ing sum, provides more information about topological structure. It enables us to distinguish
between open and closed foams, and between granular and compacted porous materials. In
addition, it allows us to assess topological features at various length scales, which is useful
when studying multi-scale structures or even removing segmentation errors that likely occur
over the length of scale of only a few voxels (Zomorodian and Carlsson 2005; Edelsbrunner
and Harer 2008).
The mathematical machinery is based on cell complexes, which include triangular meshes
used in computer graphics and the cubical grids of digital images. The details of how homol-
ogy computations are performed are quite technical and beyond the scope of this review, but
it provides a mathematically rigorous way to add geometric objects and to detect when two
“topological features” are equivalent or homologous. In essence, two k-dimensional cycles
are equivalent if they form the boundary of a (k + 1)-dimensional piece of the object. So
a pair of points are equivalent if they are the endpoints of an arc, two 1-dimensional loops
are equivalent if they bound an annular section of a surface sitting in the object, and the
inner and outer surfaces of a spherical shell are also equivalent. Homology also lets us add
and subtract k-dimensional cycles and builds up a global picture of how different cycles are

123
316 R. T. Armstrong et al.

related to each other. The k-th Betti number counts the inequivalent k-dimensional cycles
with respect to these notions of equivalence. It is also important to note that homology and
Betti numbers are not additive in the sense that the MF are. When two objects, A and B, are
combined, the homology of A ∪ B cannot be deduced simply from that of A, B and A ∩ B;
the result depends also on the global structure of A ∪ B. A keen reader can find more of the
background information necessary to fully understand this method in the following refer-
ences (Edelsbrunner and Harer 2008, 2010; Carlsson 2009). Another paper directly relevant
to the porous media community is Robins et al. (2016).
When working with data and objects that are noisy, such as micro-CT segmentation error,
or have structure on a range of length scales, such as multi-scale porous media, it is difficult
to quantify topology accurately at a single length scale. Persistent homology (PH) addresses
this problem by using a growing sequence of complexes to capture topological structure with
respect to a parameter (Robins 1999; Edelsbrunner et al. 2002). One way to generate such a
sequence of complexes is as the lower-level cuts of a function, f , defined on the vertices of
a cell complex, C. The lower-level cut contains the cells, σ , for which the function value is
less than a threshold, h:

L f (h) = {cells σ in C such that f (x) ≤ h for each x a vertex of σ.}. (13)

The parameter for persistence is the threshold value, h. Each topological feature (such as
a loop) is born at some parameter value, b, and then later filled in to become a boundary
or “dies” at a larger threshold, d. It is known through Morse theory that these birth and
death values are associated with critical points (maximum, minimum or saddle points) of the
function (Matsumoto 2002; Edelsbrunner et al. 2003). In 3D space, the minimum (index-0)
and maximum (index-3) of a function are easy to envisage while saddle points (index-1 or
index-2) harder to visualize. The index value is the number of principal directions in which
the function decreases near a critical point. In particular, a local minimum always creates
a new isolated piece of L f (h). An index-1 saddle locally forms a one-dimensional bridge
between two regions. If these regions were previously disjoint, then the adding of a saddle
point reduces the number of distinct pieces by one, so it is a “death.” If the regions were
already joined by a different pathway, then adding this 1-saddle creates a new loop and it is
a “birth.” Similarly, adding an index-2 saddle can either fill in a loop or create a new 2-cycle
(i.e., an enclosed cavity). For functions f : R3 → R, a local maximum always fills in a
cavity.
Features that are created and destroyed almost simultaneously are considered noise (d − b
is small), while features that persist over longer parameter ranges are deemed to be more
important (d −b is large). These features and lifetimes are all encoded by persistent homology,
which is formally a set of intervals (bi , di ) where bi is the birth parameter and di the death
parameter of feature i. The set of intervals is called a barcode (Carlsson 2009) or a persistence
diagram (Cohen-Steiner et al. 2007) when the intervals are drawn instead as points in a 2D
scatter plot. Thus, persistent homology could provide a richer picture of pore morphologies
and multi-scale structure than that obtained by measurement of Euler characteristic alone.
For the porous media community, however, this is a new technique and only a few recent
papers have utilized it (Robins et al. 2016; Lee et al. 2017; Hiraoka et al. 2016; Herring et al.
2018). An example of multi-scale analysis is provided by Hiraoka et al. (2016) where PH
is used to study the atomic rings in silica glass and found hierarchical ring structures that
explained the origin of mysterious diffraction peaks (Hiraoka et al. 2016).
A key result for applications is that persistent homology is stable in the sense that if two
functions, ( f , g), defined on the same cell complex C have | f (x) − g(x)| <
for all x ∈ C,

123
Porous Media Characterization Using the Minkowski… 317

then points in the persistence diagrams of f and g are also close (provided we include all
points on the diagonal b = d) (Cohen-Steiner et al. 2007).
To compute persistent homology, we must start with a real-valued function whose level sets
describe something meaningful about the object of interest. To characterize the geometry and
topology of a two-phase structure, we can use a signed Euclidean distance function (SEDT)
defined as follows: Let the grain phase be G, and the pore phase P and the interface between
them be I . Then, the SEDT function f (x) is the distance from x to the closest point on the
interface, with positive values for x in G and negative values in P.

− min y∈I ||x − y||2 if x ∈ P
f (x) = (14)
min y∈I ||x − y||2 if x ∈ G

If the material is imaged using micro-CT, we typically start with a binary segmented image
where each voxel is assigned to either the grain or pore phase. A voxel that is face-adjacent
to one of the opposite phase is then given a distance of 0.5 units. The result is a function
f (x) defined on a cubical grid of points x ∈ Z3 with the property that all local minima of
f lie inside the pore phase and all local maxima in the grain phase. Local minima, x, are at
centers of maximal spheres that fit inside a pore, and their value | f (x)| is the radius of this
sphere. Similarly, local maxima are centers of maximal spheres that fit inside a grain.
A good segmentation remains essential here to obtain a SEDT that accurately reflects
the geometry and topology of the two-phase structure. Mis-identified voxels create an error
in f (x) that depends on the distance from x to the interface I : so the persistent homology
analysis of the SEDT is robust with respect to small perturbations in the position of the
interface, but not with respect to small isolated inclusions of one phase in another.
To illustrate the geometric and topological information captured by persistence diagrams,
we provide the SEDT persistence diagrams computed from a 3D micro-CT image of spher-
ical beads with three different radii (0.35, 0.45 and 0.65 mm). The persistent homology
computations are made using the diamorse implementation available from github (Delgado-
Friedrichs 2016). There are three persistence diagrams presented in Fig. 5, PD0, PD1 and
PD2, encoding birth (b) and death (d) values for topological features in dimension zero, one
and two, respectively. It is a simple property of persistent homology that the values of N , L
and O for the lower-level cut L f (h) are found as the number of points in PD0, PD1 and PD2
with b ≤ h and d > h. Since our function is zero on the interface between pore and grain,
L f (0) contains all of the pore space. For granular geomaterials, we expect L f (0) to have
N = 1, L  1, O = 0 when the pore space is connected with many redundant pathways
(loops) and no isolated grain elements. The diagrams give us much more information than
this however. Given that the pore space is connected, points in PD0 lie in the third quadrant
with b < d < 0. Each point (b, d) signals the existence of a region of pore space associated
with a locally maximal inscribed sphere of radius |b|. The death values |d| in PD0 are throat
radii (found as index-1 saddles of the function f(x)) for those throat connections that join
up all the pores into a spanning tree. Pores are paired with the largest throat that creates a
bridge to another pore of larger radius. The points in PD1 represent loops that can form in the
lower-level cuts of the SEDT. We see that there are (b, d) points in all three quadrants with
b < d. Those with b < d < 0 are loops that are created and filled in within the pore phase:
These occur in highly non-convex pores, such as formed between two beads that almost but
do not quite touch. PD1 points with b < 0 < d represent loops (redundant pathways) in
the pore space. The birth values |b| are throat radii (1-saddles of f(x)) that are the narrowest
constrictions associated with their loop. The death values are associated with index-2 saddles
in the SEDT. Since d > 0, these are in the grain phase and are a grain-contact radius. Those

123
318 R. T. Armstrong et al.

Fig. 5 Persistence diagrams computed from the SEDT of a spherical bead pack imaged with micro-CT. Level
set value at death (d) is in the y-axis and at birth (b) is on the x-axis. PD0 has a few points with d > 0, implying
a disconnected pore space, but since the voxel resolution is 0.02 mm and d − b ≤ 0.05 mm, for those points
we can conclude these are small features near the boundary of the sample and ignore them, i.e., N = 1. PD1
has interesting structure in the b < d < 0 quadrant signaling the presence of many “2-saddle” pores between
beads not-quite-in-contact. Most PD1 points are in the b < 0 < d quadrant, and the total number of points
in this quadrant is the genus, L, of the pore space. A count of these points has L = 602 at a persistence level
(d − b) > 0 mm, and L = 592 at a persistence level of (d − b) > 0.05 mm (2.5 times the voxel resolution).
As we expect, there are no points in PD2 with b < 0 as there are no isolated grains and O = 0. PD2 has three
distinct clusters in the 0 < b < d quadrant picking out the three different bead radii. The horizontal “streaks”
in PD2 are due to beads partially cropped by the rectangular domain of the image

PD1 points with 0 < b < d signal the presence of highly non-convex grains, such as might
be formed in a sandstone when a narrow throat is cemented over, and are mostly absent from
the spherical bead pack. Finally, points in PD2 give us information about the grain geometry.
The death values are associated with local maxima of f (x), and so are radii for maximally
inscribed spheres within the grain phase. The birth values are maximal grain-grain contacts
and would form a spanning tree for the grain-contact network.

3 Porous Media Characterization

Porous systems are ubiquitous in natural and synthetic materials, and the structure of these
materials plays a large role in the resulting macroscale properties. In geomechanics, the con-
tact between grains, and the fluids at these contacts, can influence bulk mechanical properties
(Scheel et al. 2008). In soil sciences, the size of soil aggregates can influence transport pro-
cesses influencing bacterial ecosystems (Schlüter et al. 2018) or water flow, and transport
process through unsaturated soil is largely influenced by pore-scale structure (Kumahor et al.
2015). With porous biomaterials generated for bone ingrowth and/or biomedical scaffold-
ing, structure and connectivity play a critical role in functionality (Jones et al. 2007; Kapfer
et al. 2011; Jones et al. 2004). Many of these systems can be studied by using the MF as a
characterization tool. In Table 2, we provide a broad overview of the various porous media
manuscripts that utilize the MF in one way or another. This is by no means in inclusive
list, but is intended to provide an interesting range of applications that utilize key principles
discussed in this review.
In Falconer et al. (2012), the microscale structure of soil and water distribution are shown
to influence the dynamics of fungal growth (Falconer et al. 2012). Segmented micro-CT
images of soils were used as model domains to simulate water distributions and fungal growth
using the Lattice Boltzmann method coupled with an individual-based model that simulates
the individual interactions between fungal spores rather than considering population-based

123
Table 2 Summary of the manuscripts presented in this review
Application Key principle References

Analysis of multi-scale structure Persistent homology Lee et al. (2017) and Hiraoka et al. (2016)
Assessment of biomaterials Morphology Knackstedt et al. (2006), Knothe (2003), Tate et al. (2009), and Anderson et al. (2008)
Calculating capillary pressure Mean curvature Armstrong et al. (2012), Li et al. (2018), and McClure et al. (2016)
Consolidation of granular media Persistent homology Delgado-Friedrichs (2016), and Saadatfar et al. (2017)
Defining the geometric state of fluids Steiner’s theorem McClure et al. (2018), Liu et al. (2017), Herring et al. (2013), Rücker et al. (2015), and
Schlüter et al. (2016)
Determining the energy of interfaces Hadwiger’s theorem Hyde et al. (1990), Mecke (2000)
Development of geometrical relations Integral geometry Weyl (1939), Ohser et al. (2011), and Federer (1959)
Dynamics of fungal growth in soil Minkowski functionals Falconer et al. (2012), and Grimm (1999)
Estimation of permeability Percolation theory Scholz et al. (2012), and Liu et al. (2017)
Porous Media Characterization Using the Minkowski…

Evaluation of relative permeability Euler characteristic Armstrong et al. (2016), and Liu et al. (2017)
Identification of percolation threshold Minkowski functions and Robins et al. (2011), Robins et al. (2016), Vogel et al. (2010), and Michielsen and De
PH Raedt (2001)
Influence of fertilization on soil structure Minkowski functionals Schlüter et al. (2011), Mantz et al. (2008), and San Jose Martinez et al. (2013)
Interpretation of capillary trapping Morphology and PH Herring et al. (2018), Herring et al. (2013), Herring et al. (2015), and Andersson (2018)
Kinematics of a Haines jump Gauss–Bonnet theorem Sect. 4.3
Linking thermodynamic properties Hadwiger’s theorem McClure et al. (2018), Gray and Miller (2014) Sect. 4.2
Measurement of pore size distribution Minkowski functions Hilpert and Miller (2001), Vogel et al. (2010), San Jose Martinez et al. (2013), Michielsen
and De Raedt (2001)
Processing of segmented images Minkowski functionals Delgado-Friedrichs et al. (2015), Schlüter et al. (2014), Robins et al. (2011)
Quantification of segmented data Set theory Lang et al. (2001), Nagel et al. (2000), Legland et al. (2011), Mantz et al. (2008), Ohser
and Mücklich (2000)
Reconstruction of porous media Minkowski functionals Schlüter and Vogel (2011), Mosser et al. (2017), Jiao et al. (2009)
The list is intended to illustrate a range of possible applications where the MF are useful for porous media research. Many additional applications are available in Mecke (2000)

123
319
320 R. T. Armstrong et al.

statistics (Grimm 1999). Soil pore morphology and resulting water distributions were quan-
tified using the MF. Water content and pore size connectivity played a significant role in
colonization capacity since water blocked off regions of the pore space from fungal growth.
It was shown that rather than the amount of water content it was the location of the water
that played a role in fungal capacity and resulting morphology. Another interesting study
investigates the evolution of soil structure over a long-term fertilization operation. Soil sam-
ples over the summer and spring period were collected from plots that received significantly
different amounts of fertilizer over a period of 106 years (Schlüter et al. 2011). The samples
were imaged with micro-CT and then quantified using the MF. Fertilized topsoils were found
to evolve toward higher porosities and pore connectivities during the growing season than
the non-fertilized control. The soil structure underneath the plow layer had no differences
in terms of pore size distribution and pore connectivity, indicating the soil structure in that
depth is rather stable and has evolved over much longer time scales.
Another interesting application for the MF deals with the statistical analysis of porous
structures (Arns et al. 2001) and the stochastic reconstruction of porous systems. Spatial
statistics provides nearest-neighbor distance distribution functions that can be used to char-
acterize and/or reconstruct a model porous material (Okabe and Blunt 2005, 2007; Øren
and Bakke 2002, 2003; Wang et al. 2018). A few common spatial statistics are the 2-point
correlation function, autocorrelation function and/or variograms. The two-point correlation
is defined as a possibility from one location to the next over a given distance whether the
location will be solid phase or void space (Fredrich et al. 1993). However, these functions
do not necessarily preserve/represent high-order moments of the media, and resulting recon-
structions of porous media based on these statistic have been shown to misrepresent topology
(Adler et al. 1992). The chord length distribution function is another two-point correlation
function (Serra 1982), which has also has been shown to misrepresent the connectivity of
porous systems. Three-point or four-point correlation functions may provide a better rep-
resentation of the system even though higher-order functions are difficult to measure and
incorporate into model reconstructions.
The MF provide an alternative to these approaches and have been used in a few recent
works (Schlüter and Vogel 2011; Mosser et al. 2017). Using MF as the only shape measure
for stochastic reconstruction is insufficient to form percolating, connected structures, since
the MF are local and additive, while percolation is non-local and non-additive. The stochastic
reconstruction of percolating, interconnected channels can be achieved by a combination of
chord length distributions to enforce the corrected separation distance between channels and
the MF to impose the corrected surface area and topology of the channels (Schlüter and
Vogel 2011). Even better results can be achieved by reconstructing long-range connectivity
by using the two-point cluster function, as reported in Jiao et al. (2009). In addition, persistent
homology incorporates higher-order spatial correlations in a natural way (Robins et al. 2011)
and since homology (and the Betti numbers) are not additive functionals, they critically
depend on the global connectivity of an object (Robins et al. 2016).
The MF can also be applied as point processes in 3D space by measuring the functionals
at a given location, (x,y,z), over a given length, d (Michielsen and De Raedt 2001). Since
MF are average properties, they do not provide pore size information, which is known to
be a critical parameter for flow and transport properties. By applying an opening algorithm
(Hilpert and Miller 2001) on the pore space, Vogel et al. 2010 fractioned the pore space
into different minimum pore sizes and measured the MF for each fraction. This provided a
Minkowski function with respect to pore size defined as Mi (d), where d is pore diameter.
With this approach, M0 (d) provides the pore size distribution, M1 (d) provides pore surface
density for a given minimum pore size, M2 (d) provides curvature of a given pore size, and

123
Porous Media Characterization Using the Minkowski… 321

Fig. 6 Persistence diagrams computed from the SEDT of a Berea sandstone core (image dimensions = 4.27 ×
4.54 × 8.14 mm). As there are so many points in the persistence diagrams, we display them as 2D histograms
where the color of each pixel represents the number of points in that patch of the birth–death (b–d) plane.
The d value of the horizontal spike in PD0 is the critical percolating sphere radius for the pore space and
the b value of the “dual” vertical spike in PD2 is the critical percolating sphere radius for the grain phase.
Also notice that for the PD1 diagram of this sandstone there are a much greater proportion of points in the
0 < b < d quadrant than the b < d < 0 one when compared to the bead pack in Fig. 5. This is typical of
consolidated versus unconsolidated granular geomaterials

M3 (d) provides information on critical pore neck size (d) at which M3 (d) = 0 where the
system is presumably no longer connected. An interesting application of this approach was
taken by San Jose Martinez et al. (2013) where they imaged soil samples from a vineyard with
micro-CT to understand how soil management practices influence soil macropore structure
(San Jose Martinez et al. 2013). The study suggested that management approaches whereby
resident vegetation coverage is applied resulted in a larger amount of low density materials
as voids and organic matter that existed within the soil pore structure. Characteristic points
in the M3 (d) function, like the root and the maximum, have also been proposed to result
in a morphological aspect ratio that correlates much better with gas trapping in water-wet
rock than conventional aspect ratios based on local estimates of pore body and pore neck
diameters (Andersson 2018). These global estimates of critical pore neck and pore body
size also coincide with typical normalized mean curvatures of the remaining pore network
resulting from morphological opening.
Robins et al. (2016) demonstrate that pore size distributions, connectivity numbers and the
critical radius of a percolating sphere can be measured by using topological PH. Furthermore,
the PH of distance functions has been applied to granular and porous materials in various
ways. The authors of (Robins et al. 2016) show that the critical percolating sphere radius of
both the pore and grain phases appears as an important length scale in persistence diagrams
of geomaterials and other random geometries. The degree of consolidation of a granular
material can also be seen in PD1 (Delgado-Friedrichs et al. 2014) from the relative number
of points in the first and third quadrants of a persistent diagram. Both of these signatures are
illustrated in the persistence diagrams of the Berea sandstone sample displayed in Fig. 6. The
algorithm developed to identify the critical points of the distance function in Robins et al.
(2011) can be adapted to the image segmentation tasks of skeletonization and partitioning of
complex structures (Delgado-Friedrichs et al. 2015). The topological accuracy provided by
this approach means the pore labeling and pore network are constructed in a precisely dual
way.
In addition, persistence diagrams of a slightly different type of distance function are used
to study frictional athermal bead packings in (Saadatfar et al. 2017). In that study, the centers
and radii of each spherical bead in a monodisperse cylindrical packing are extracted with very
high precision from micro-CT images and then used as a point pattern input for Delaunay and
Voronoi constructions. The associated persistence diagrams have a single cluster of points in

123
322 R. T. Armstrong et al.

PD0 representing the single bead radius, while PD1 captures the geometry of cycles formed by
bead contacts and PD2 contains information about the geometry of the pores. PD2 provides
a particularly clear signature of locally crystalline packing and the deformation pathways
between crystalline and disordered close-packed structures. The densest crystalline packing
of spheres (0.74 volume fraction) is formed by beads at the vertices of regular tetrahedra
and octahedra, and these configurations show up as limit points in PD2. Random close-
packing of spheres (0.64 volume fraction) shows a broad distribution of distorted tetrahedral
configurations, and virtually no octahedral configurations. Partially crystallized packings
(volume fraction between these limits) show highly populated curves in PD2 associated with
particular mechanically realizable deformations of regular tetrahedra and octahedra.

4 From Integral Geometry to Macroscale Properties

A fundamental goal for porous media research is to link pore structure to functions such
as, permeability, relative permeability, diffusivity and/or capillary pressure, which are nec-
essary for engineering applications. In porous media applications, these links must extend
across multiple length scales where microscale parameters are translated to macroscale prop-
erties. The concept of volume averaging (Whitaker 2013) provides the foundation in which
macroscale thermodynamic and conservation equations can be derived from information at
smaller length scales. The MF provide a convenient tool for defining the microscale state of
the system in a macroscale way. In addition, the MF have the support of theoretical tools that
can provide convenient relationships for defining states and/or interrelationships between
states in a morphological sense. In the following subsections, we provide examples of how
the MF and supporting theories can be related to macroscale properties that are important for
porous media applications.

4.1 Porous Media with One Fluid Phase

Absolute permeability (K ) represents the ability of porous media to conduct fluid, which
is defined by Darcy’s Law (Darcy 1856). While permeability is a simple concept, there is
an extraordinary amount of literature that studies it and numerous empirical relationships
that attempt to link the pore-scale structure to permeability (Tiab and Donaldson 2015). One
widely used empirical correlation is the Katz–Thompson model (Katz and Thompson 1986)
shown as,

K = ClC2 (σ/σ0 ) (15)

where C is a constant, lC is a characteristic length scale and σ is the electrical conductivity


of the media saturated with brine with conductivity σ0 . This relationship essentially cap-
tures pore topology through electrical conductivity measurements, and geometry is captured
through the length scale term lc . The σ/σ0 term can be linked to porosity according to Archie’s
law (Tiab and Donaldson 2015) as,

σ/σ0 = (φ − φc )/(1 − φc )μ (16)

where φc is the critical porosity at which the media stops percolating and μ is a critical
exponent. An interesting approach proposed by Scholz et al. (2012) was to incorporate

123
Porous Media Characterization Using the Minkowski… 323

Fig. 7 Comparison between Katz–Thompson model and a topological-based model for 2D Gaussian mixture
models of round, ellipse and square shaped grains. The topological-based model provides a single relationship
near the critical porosity. Data adapted from (Liu et al. 2017)

direct measurements of pore morphology into the Katz–Thompson model. The proposed
formulation was,
K = clc2 (1 − χ0 /N )α (17)
where χ0 is the Euler characteristic of the conducting phase and N is the number of grains
in the granular media. Through topological arguments for 2D systems, it can be shown that
as φ → 1, (1 − χ0 )/N → 1. When grains do not overlap, (1 − χ0 ) is the number of
grains N . Conversely, near the percolation threshold only a single connected pathway will
be established and (1 − χ0 )/N goes to zero. The topological term replacing conductivity in
the Katz–Thompson model provides analogous scaling behavior. The suggested relationship
was shown to hold for Gaussian mixture models of circular and elliptical grains (Scholz et al.
2012, 2015). We regenerated these porous systems with an additional square grain shape
to further test the model (Liu et al. 2017). The permeability of each generated model was
determined by single-phase Lattice Boltzmann simulations, and the results are presented in
Fig. 7. The topological-based model provides a single relationship, whereas Katz–Thompson
provides two unique relationships, one for the spherical and square grains and another for
elliptical grains. While the topological model has not been tested on real porous media nor
has it been attempted on 3D structures to an adequate level of robustness, it does provide
insight into how the scaling behavior of topological and geometrical parameters correlate
with permeability near critical points.

4.2 Porous Media with Two Immiscible Fluid Phases

A key opportunity associated with the geometric characterization of porous media is to use
multi-scale geometric relationships to advance a more comprehensive modeling approach
for two-fluid flow in porous media. For example, Eq. 5 provides a direct link between
microscale geometric invariants—the Gaussian curvature at points on a surface and the
geodesic curvature at points on the common curve—and their macroscopic counterpart, the
Euler characteristic. In the context of two-fluid flow, the implication of the Gauss–Bonnet
theorem is that changes to the overall fluid connectivity result directly from the microscale
changes in Gaussian curvature caused by snap-off and coalescence events within the pore
space. Such a displacement sequence is depicted in Fig. 8, for which the non-wetting fluid

123
324 R. T. Armstrong et al.

Fig. 8 Numerical simulations of wetting phase (transparent) displacing non-wetting phase (red) from a porous
rock (brown). The red fluid is undergoing imbibition and flowing downward through the visible portion of the
sandstone. How the topology and geometry of the NWP change during this process can be characterized by
the MF. The presented data are based on simulations using LBM (Armstrong et al. 2016; McClure et al. 2014)

evolves from a well connected to less-connected state. The fact that such changes can be
quantified completely using only averaged measures is quite powerful. However, in a phys-
ical system geometric changes do not occur in a vacuum; the rate at which the geometric
evolution proceeds is linked to the underlying thermodynamic and mechanical forces that
act on the system. The thermodynamically constrained averaging theory (TCAT) connects
the geometric evolution of the system with appropriate thermodynamic expressions by con-
structing and exploiting a macroscopic entropy inequality for the system (Gray and Miller
2014), an extension of classical approaches used to derive basic constitutive laws based on
the rate of energy dissipation (Groot and Mazur 1962). Key unanswered questions relate to
the link between system dynamics and the geometric evolution. Can approximate models be
constructed by which geometric measures can be tracked reliably at the macroscale?
With TCAT, structural entities at the pore scale are defined and measured in a consistent
manner. Where the pore scale is defined as the length scale at which grains, fluid phase
and interfaces between phases are well resolved. In Fig. 8, the multiphase image can be
decomposed into its pore-scale entities, (1) three-phase contact curve where wetting fluid
(w), non-wetting fluid (n) and solid phase (s) are in common contact, (2) wetting fluid and
non-wetting fluid interface, (3) wetting fluid and solid phase interface and (4) non-wetting
fluid and solid phase interfaces. Each of these entities can be defined in a geometrical way,
and their equivalent macroscale parameter can be defined by the concept of volume averaging
(Whitaker 2013). The macroscale parameters used in TCAT are defined in Table 3 where
subscript i detonates the macroscale phase- averaged value. We will relate a few of these
parameters to the MF in the following paragraphs to illustrate the link between multi-scale
averaging and pore-scale geometry.

123
Porous Media Characterization Using the Minkowski… 325

Table 3 Relevant geometric Quantity Description


measures included in upscaling
theories
Void fraction of pore space

w Volume fraction of wetting phase

n Volume fraction of non-wetting phase

wn Surface area per unit volume for the wn interface

ns Surface area per unit volume for the ns interface

ns Surface area per unit volume for the ws interface
Jwwn Average mean curvature of wn interface
Jsns Average mean curvature of ns interface
Jsws Average mean curvature of ws interface

The MF provide a consistent framework for defining pore-scale entities in a macroscale


way that reflects pore-scale geometries and topologies. For example, the MF of the non-
wetting phase n can be expressed in terms of the parameters listed in Table 3 as,
M0n =
n V
M1n = (
wn +
ns )V
M2n = (Jwwn
wn + Jsns
ns )V
M3n = χ n
where V is total volume. With this notation, superscripts are used to denote a macroscale
property for a given phase. Likewise MF for wetting phase w or solid phase s can also be
defined. The Euler characteristic is not commonly used in upscaling even though it measures
the average state of the system in terms of the number of possible loops, components and cav-
ities that exist, which could be linked to percolation thresholds and absolute permeability, as
previously discussed. Previous works have shown the importance of connectivity when defin-
ing the state of a multiphase system (Hilfer 2006; Picchi and Battiato 2018; Valavanides and
Payatakes 2001). The Euler characteristic could be a natural way to account for connectivity
in a macroscale way.
Interfacial properties formed by the union of two phases can also be defined through the
MF. Interfacial area between WP (w) and NWP (n) phase is defined as,

wn V = (M1w + M1n − [M1w ∪ M1n ])/[2] (18)
where the division by two is required since the interface between phases is counted twice
when adding M1w and M1n . It is inherently convoluted to find the average mean curvature
of the wetting/non-wetting interface (Jwwn ). This can be illustrated by considering the mean
curvature of the flat interface at the union of two hemispheres denoted as H1 and H2 . Here
the average mean curvature is,
J H1 H2
H1 H2 V = V (M2H1 + M2H2 − [M2H1 ∪ M2H2 ])/2 − [πr + 2π] (19)
where r is the radius of each hemisphere. The correction factor of πr + 2π is required
since M2 considers curvature along the perimeter line at the union of the two hemispheres
while the definition of average mean curvature considers only the interfacial curvature. The
correction factor is trivial for the provided example. At the perimeter of the union, there exist
two voxels, i.e., one from each hemisphere, each of which contributes π/2 curvature and an
additional 2π curvature exists in the orthogonal plane to account for the circumference. Once

123
326 R. T. Armstrong et al.

Fig. 9 A classic example of a capillary tube system with a single pore body. The process of drainage is
illustrated in panels a though c. Many quasi-static interface configurations are possible when the three-phase
contact line is pinned at the pore entry , which can be described analytically (Hilpert and Miller 2001)

a complex multiphase system is considered, it would be convoluted to consider the curvatures


along the three-phase contact line to solve for Jwwn . Other approaches have been presented
in the literature whereby average mean interfacial curvature Jwwn is measured directly from a
generated mesh of the fluid/fluid interface (Armstrong et al. 2012; Li et al. 2018). This value
can then be directly related to capillary pressure by the Young–Laplace equation as,

P c = γ Jwwn (20)

where γ is interfacial tension (Bear 2013).

4.3 Kinematic Relationship for a Haines Jump

Here an example of two-phase flow where non-wetting phase (NWP) enters a larger pore body
is discussed. We will develop a geometrical relationship between time, curvature and Euler
characteristic by using the Gauss–Bonnet theorem. This work will provide a simple kinematic
relationship in terms of geometrical quantities that describes a fluid/fluid displacement event
known as a Haines jump (Berg et al. 2013; Haines 1930).
Consider a model that consists of two cylindrical throats that are connected by a single
circular pore body, as shown in Fig. 9. The length, L, of the inlet and outlet throats is equal,
and the width of the inlet throat, r1 , is greater than the outlet throat width, r2 . Here the model is
initially saturated with wetting phase (WP) followed by injection of NWP. Many quasi-static
interface configurations are possible as the pore body region fills with NWP, which can be
described analytically (Hilpert et al. 2003), as displayed in Fig. 9. The MF for this system
can be uniquely defined based on these analytical expressions.
As the NWP advances through the cylindrical pore throat, the MF can be constructed
piecewise because of additivity. This allows us to account for the drained pore throat and the
region of the drained pore body separately. We will consider all possible arrangements of the
NWP as it is pinned at the pore entry. We are not concerned with whether or not a particular
fluid arrangement is mechanically stable. We will consider fluid arrangements from time (t)
at which the three-phase contact points reach the pore entry. Since the solid surface is not
smooth at the pore entry, the contact angle is not well defined. As shown by Hilpert and
Miller 2001 (Hilpert and Miller 2001), the contact point will become pinned at the location
of the kink, and the radius of curvature rm (t) of the NWP meniscus will increase until the

123
Porous Media Characterization Using the Minkowski… 327

pore is completely filled at time t = t jump . For this event the morphological evolution of the
NWP can be defined by the Gauss–Bonnet theorem as,
 
[1/rm (t)]dA + k g ds = 2π (21)
M dM
where rm (t) is the radius of curvature of the fluid/fluid interface at time t, k g is the geodesic
curvature at all sharp corners: two 3-phase contact points and two bottom corners of the pore
throat. Notice that the piecewise integral of curvature over the pore throat is excluded since
it is a flat surface of zero curvature. Furthermore, the total curvature must always equal 2π
since the Euler characteristic for the NWP interface is one and this does not change during
the event. This is an interesting consequence of the GB theorem, which states that under
deformations where topology does not change then total curvature must be conserved. This
is known as an isomorphic deformation, which could occur assuming that the jump does not
facilitate disconnection of NWP in another region of the pore space. For this example, we
consider locality to the single pore body and throat system as displayed in Fig. 9.
The sources of curvature for the NWP correspond to the fluid/fluid interface and any sharp
corner. The fluid/fluid interface provides a total curvature of,

[2πrm (t)a]/[rm (t)] (22)

where a is a fraction that represents the meniscus interfacial area divided by the interfacial area
of an equivalent circle with radius rm (t). Additional curvature is found in the sharp corners
of the NWP. Two 90 degree corners are located at the bottom of the model domain, which
combined provide π curvature. An additional two corners are at the pore entry; however, the
angles formed by the NWP at these points are not constant. We will assume that these angles
are equal and then represent curvature as 2πb where b ranges from zero to one. Then, by Eq.
21, we find,
0.5 = a + b (23)

where a and b represent the fractions of total curvature that are either in the fluid/fluid
interface or at the entry corners of the pore. This simple relationship shows how interfacial
curvature and contact angle are related from a purely geometrical perspective, which could
be considered as a kinematic relationship for a Haines jump for the given model.

4.4 Geometric State of 2-Fluid Systems

The MF have been shown to provide unprecedented insights into the geometric state of 2-
fluid systems (Armstrong et al. 2016; Liu et al. 2017; Khanamiri and Torsæter; Herring et al.
2015; Rücker et al. 2015; Schlüter et al. 2016). To illustrate these insights, a simple network is
presented in Fig. 10 similar to what has been used to model multiphase flow for decades (Dong
and Blunt 2009). The network contains both WP and NWP with a single capillary trapped
NWP droplet. Analogous fluid arrangements have been observed in experimental images
(Tanino and Blunt 2012; Pentland et al. 2011; Iglauer et al. 2011; Andersson 2018). Equation
4 essentially states that the Euler characteristic can be decomposed into three topological
measurements of objects, loops and cavities, which are otherwise known as the Betti numbers
β0 , β1 and β2 , respectively. In Fig. 10a, the non-wetting phase has two objects (β0 = 2), and
no loops (β1 = 0) since a single cut through either object would divide the object. There are
also no cavities within the non-wetting phase (β0 = 0). The resulting Euler characteristic
for the non-wetting phase is two for both Fig. 10a and b. From this simple example, it can

123
328 R. T. Armstrong et al.

Fig. 10 A simple porous network saturated with wetting and non-wetting phases

be seen that the non-wetting phase is not connected across the domain. If additional non-
wetting phase is added to the network, either the disconnected phase would reconnect with
the inlet or a loop would be formed around the four pore bodies. In the prior case, the Euler
characteristic would be zero, while for the latter case, it would be negative one. As more
loops are added or as disconnected regions become reconnected, the Euler characteristic
becomes more negative. The transition from positive to negative Euler characteristic for the
NWP would correspond to a saturation value near the percolation threshold for the presented
network (Vogel 2000; Mecke 1998). In addition, recent work shows this to be important for
electrical resistivity measurements in intermediate to oil-wet systems (Liu et al. 2017).
Both non-wetting phase arrangements presented in Fig. 10 a and b provide the same Euler
characteristic even though the saturations differ. As seen in laboratory experiments, it is
also known that a single saturation can result in various possible capillary pressures (Porter
et al. 2010; Schaap et al. 2007; Karadimitriou et al. 2014). In either case the uniqueness of
the pore-scale morphology is not well defined. As stated by Equation 8, the evolution of
a single functional can be defined by a linear combination of all other functionals (Ohser
et al. 2011). This would suggest that the geometrical state of the network could be defined
by consideration of all four MF.
As provided in the previous subsection, the MF can be linked to multiphase macroscale
properties, such as saturation, specific interfacial area, capillary pressure and topology. By
utilization of these definitions and the Steiner formula, researchers have defined a geometrical
state function for 2-fluid flow (McClure et al. 2018). The findings were based on simulations
using 6 different porous media and over 210,000 fluid configurations. Generalized additive
models (GAM) were generated based on measured fluid morphologies. Three possible models
were considered P c (S w ), P c (S w ,
nw ) and P c (S w ,
nw , χ n ). The resulting relative error for
all three models for one of the porous systems considered in the study is presented in Fig. 11a.
The resulting GAM for P c (S w ,
nw , χ n ) is provided in Fig. 11b. The data are based on two-
phase Lattice Boltzmann simulations on a segmented micro-CT image (McClure et al. 2014;
Armstrong et al. 2016). It can be seen that complete characterization of fluid morphology
provides the lowest residual error. As suggested by Steiner and Hadwiger, the MF can be
interrelated to provide estimates of capillary pressure based on the other 3 functionals. This
illustrates that complete 3D morphological characterization requires all 4 MF. This provides
a geometrical-based function for 2-fluid flow analogous to the constituent relationship of

123
Porous Media Characterization Using the Minkowski… 329

Fig. 11 Residual error when considering generalized additive models (GAM) for estimation of capillary pres-
sure from pore scale morphological characterization (a). The resulting GAM for P c (S w ,
nw , χ n ) showing
a unique relationship for the geometrical state of two-fluid flow

capillary pressure versus saturation except that here a complete morphological description is
provided rather than characterization by only saturation, which is equivalently to M0 .

5 Future Trends and Opportunities

We have shown that the MF provide a consistent and complete formulation for the pore-scale
characterization of porous systems that facilitates theoretical developments. We reviewed
the framework necessary for the pore-scale characterization of digital images collected from
micro-CT data. We provided the basic definition of geometrical measures for n-dimensional
space and showed applications of mathematic theorems used in algebraic topology. We also
showed the ease at which the MF can be measured from micro-CT data and provided insights
into the parameters and issues that influence these measurements. We provided examples
illustrating how the MF have been used to characterize porous systems and/or topological
theorems can be used to provide kinematic relationships. The utility of the kinematic relation-
ships for solving macroscale two-phase flow problems have yet to be demonstrated; however,
we see this as an exciting area of research. In particular, the geometrical state function for
fluids provides a needed closure relationship that uniquely describes capillary pressure. Euler
characteristic provides a natural way to characterize the connectivity of a fluid phase that
could lead to the development of theories that account for connected versus disconnected
phase flow ([110],Valavanides and Payatakes 2001; Hilfer 2006). Further works are required
to bring together the morphological and thermodynamic description of two-phase flow. This
could lead to a system of equations that explain how fluids move through the vector space of
their constrained geometrical states. Recent works that are moving forward in this direction
include ([110,48],Hansen et al. 2016).
The application of PH provided a unique way to characterize porous media and circum-
vents issues with measuring topology based on surface curvatures. PH will likely provide
new ways to categorize porous systems, reconstruct porous media and/or predict macroscale
properties. In particular, the ability of PH to characterize multi-scale structures should be
further assessed for porous systems (Hiraoka et al. 2016). There is a wealth of information
provided by PH for structural analysis and the evaluation of transitions in structures under

123
330 R. T. Armstrong et al.

dynamic processes (Saadatfar et al. 2017). One interesting area of research would be the
coupling of PH with artificial intelligence for categorical analyses and/or image processing
applications where the goal would be to reduce user-bias in image processing and preserve
topological structures that are indeed important for macroscale properties. As a final remark,
we emphasize that although PH analysis of distance functions is widespread (Gyulassy et al.
2007, 2012; Ushizima et al. 2012; Kimura et al. 2018), the same technique can be applied
to any real-valued function. Other relevant applications include the analysis of fluid density
functions in turbulent mixing (Laney et al. 2006; Kramar et al. 2016), and force networks
in 2D granular packings (Kramar et al. 2014; Ardanza-Trevijano et al. 2014). Extensions of
PH to multivariate functions are an area of active research (Carlsson and Zomorodian 2009;
Cerri et al. 2013; Iuricich et al. 2016).
The tools necessary to image and characterize porous systems are readily available, and
there is no doubt that the MF will be a center point for understanding pore-scale mechanisms,
categorizing porous systems and for developing new theories that require data-intensive
approaches. A major barrier to these studies and the uptake of developed theories to indus-
try practices is the computational power required. Any high-fidelity model that relies on
microscale data of geometrical states will require significant computational power and data
storage requirements. The required data will not be extracted by conventional work flows
and thus, will require innovative approaches, astute assumptions for limiting cases and new
technologies. One potential approach is to develop traditional laboratory techniques such
that they can provide microscale insights (Liu et al. 2017) and/or develop softwares that
directly provide microscale measures (McClure et al. 2014). Furthermore, fast micro-CT
imaging using synchrotron facilities coupled with direct analysis tools that provide real-
time data would provide the resources required to test and develop new theories. Lastly,
the multi-disciplinary nature of this research ranges from advanced mathematics in topol-
ogy, thermodynamics, mass and momentum transport, various engineering disciplines, image
processing, X-ray physics, supercomputing and computational fluid dynamics. Collaborative
efforts are essential for these developments.

Acknowledgements We thank the Tyree X-Ray Laboratory in the School of Minerals and Energy Resources
Engineering, UNSW for assistance with image collection and data processing. Professors Stephen Foster
(fiber-reinforced concrete) and Melissa Knothe-Tate and Dr. Tzong-Tyng Hung (mouse leg) are acknowledged
for graciously sharing their microtomography data. We thank Ji-Youn Arns and Zhenghuai Guo for their
persistence in contrast optimization for the mouse leg and visualization of the fiber-reinforced concrete samples.
Funding was provided from the Australian Research Council Discovery Grant DP160104995. VR is supported
by ARC Future Fellowship FT140100604. An award of computer time was provided by the Department of
Energy INCITE program. This research also used resources of the Oak Ridge Leadership Computing Facility,
which is a DOE Office of Science User Facility supported under Contract DE-AC05-00OR22725.

References
Abbena, E., Salamon, S., Gray, A.: Modern Differential Geometry of Curves and Surfaces with Mathematica.
CRC Press, Boca Raton (2017)
Adler, P.M., Jacquin, C.G., Thovert, J.F.: The formation factor of reconstructed porous media. Water Resour.
Res. 28(6), 1571 (1992)
Anderson, E.J., Kreuzer, S.M., Small, O., Tate, M.L.K.: Pairing computational and scaled physical models
to determine permeability as a measure of cellular communication in micro-and nano-scale pericellular
spaces. Microfluid. Nanofluid. 4(3), 193 (2008)
Andersson, L.: Defining a novel pore-body to pore-throat morphological aspect ratio that scales with residual
non-wetting phase capillary trapping in porous media. Adv. Water Resour. (2018)

123
Porous Media Characterization Using the Minkowski… 331

Ardanza-Trevijano, S., Zuriguel, I., Arvalo, R., Maza, D.: Topological analysis of tapped granular media using
persistent homology. Phys. Rev. E 89(5), 052212 (2014). https://doi.org/10.1103/PhysRevE.89.052212
Armstrong, R.T., Porter, M.L., Wildenschild, D.: Linking pore-scale interfacial curvature to column-scale
capillary pressure. Adv. Water Resour. 46, 55 (2012)
Armstrong, R.T., Georgiadis, A., Ott, H., Klemin, D., Berg, S.: Critical capillary number: desaturation studied
with fast X-ray computed microtomography. Geophys. Res. Lett. 41(1), 55 (2014)
Armstrong, R.T., McClure, J.E., Berrill, M.A., Rücker, M., Schlüter, S., Berg, S.: Beyond Darcy’s law: the
role of phase topology and ganglion dynamics for two-fluid flow. Phys. Rev. E 94(4), 043113 (2016)
Arns, C.H., Knackstedt, M.A., Pinczewski, W.V., Mecke, K.R.: Euler-Poincaré characteristics of classes of
disordered media. Phys. Rev. E 63(3), 031112 (2001)
Bañados, M., Teitelboim, C., Zanelli, J.: Black hole entropy and the dimensional continuation of the Gauss-
Bonnet theorem. Phys. Rev. Lett. 72(7), 957 (1994)
Bear, J.: Dynamics of Fluids in Porous Media. Courier Corporation, North Chelmsford (2013)
Berg, S., Ott, H., Klapp, S.A., Schwing, A., Neiteler, R., Brussee, N., Makurat, A., Leu, L., Enzmann, F.,
Schwarz, J.O., et al.: Real-time 3D imaging of Haines jumps in porous media flow. Proc. Nat. Acad. Sci.
110(10), 3755 (2013)
Berg, S., Rücker, M., Ott, H., Georgiadis, A., van der Linde, H., Enzmann, F., Kersten, M., Armstrong, R.,
Becker, J., Wiegmann, A., et al.: Connected pathway relative permeability from pore-scale imaging of
imbibition. Adv. Water Resour. 90, 24 (2016)
Bijeljic, B., Mostaghimi, P., Blunt, M.J.: Signature of non-Fickian solute transport in complex heterogeneous
porous media. Phys. Rev. Lett. 107(20), 204502 (2011)
Blunt, M.J.: Multiphase Flow in Permeable Media: A Pore-Scale Perspective. Cambridge University Press,
Cambridge (2017)
Blunt, M.J., Bijeljic, B., Dong, H., Gharbi, O., Iglauer, S., Mostaghimi, P., Paluszny, A., Pentland, C.: Pore-scale
imaging and modelling. Adv. Water Resour. 51, 197 (2013)
Carlsson, G.: Topology and data. Bull. Am. Math. Soc. 46(2), 255 (2009)
Carlsson, G., Zomorodian, A.: The theory of multidimensional persistence. Discrete Comput. Geom. 42(1),
71 (2009). https://doi.org/10.1007/s00454-009-9176-0
Cerri, A., Fabio, B.D., Ferri, M., Frosini, P., Landi, C.: Betti numbers in multidimensional persistent homology
are stable functions. Math. Methods Appl. Sci. 36(12), 1543 (2013). https://doi.org/10.1002/mma.2704
Cohen-Steiner, D., Edelsbrunner, H., Harer, J.: Stability of persistence diagrams. Discrete Comput. Geom. 37,
103 (2007)
Darcy, H.: Les fontaines publiques de la ville de Dijon: exposition et application... (Victor Dalmont, 1856)
Delgado-Friedrichs, O., Robins, V., Sheppard, A.: In: 2014 IEEE International Conference on Image Processing
(ICIP) (2014), pp. 4872–4876. https://doi.org/10.1109/ICIP.2014.7025987
Delgado-Friedrichs, O.: Diamorse—Digital image analysis using discrete Morse theory and persistent homol-
ogy (2016). https://github.com/AppliedMathematicsANU/diamorse
Delgado-Friedrichs, O., Robins, V., Sheppard, A.: Skeletonization and partitioning of digital images using
discrete morse theory. IEEE Trans. Pattern Anal. Mach. Intell. 37(3), 654 (2015). https://doi.org/10.
1109/TPAMI.2014.2346172
Dong, H., Blunt, M.J.: Pore-network extraction from micro-computerized-tomography images. Phys. Rev. E
80(3), 036307 (2009)
Dullien, F.A.: Porous Media: Fluid Transport and Pore Structure. Academic Press, Cambridge (2012)
Edelsbrunner, H., Harer, J.: Persistent homology-a survey. Contemp. Math. 453, 257 (2008)
Edelsbrunner, H., Harer, J.: Computational Topology: An introduction. American Mathematical Society, Prov-
idence, Rhode Island (2010)
Edelsbrunner, H., Letscher, D., Zomorodian, A.: Topological persistence and simplification. Discrete Comput.
Geom. 28(4), 511 (2002). https://doi.org/10.1007/s00454-002-2885-2
Edelsbrunner, H., Harer, J., Zomorodian, A.: Hierarchical Morse–Smale complexes for piecewise linear 2-
manifolds. Discrete Comput Geom 30(1), 87 (2003). https://doi.org/10.1007/s00454-003-2926-5
Falconer, R.E., Houston, A.N., Otten, W., Baveye, P.C.: Emergent behavior of soil fungal dynamics: influence
of soil architecture and water distribution. Soil Sci. 177(2), 111 (2012)
Federer, H.: Curvature measures. Trans. Am. Math. Soc. 93(3), 418 (1959)
Fredrich, J., Greaves, K., Martin, J.: Pore geometry and transport properties of Fontainebleau sandstone. Int.
J. Rock Mech. Min. Sci. Geomech. Abstr. 30, 691–697 (1993)
Gray, W.G., Miller, : In: AGEM2 ) (Springer, (ed.) Introduction to the Thermodynamically Constrained Aver-
aging Theory for Porous Medium Systems. Advances in Geophysical and Environmental Mechanics and.
Mathematics, C.T.: In (2014)
Grimm, V.: Ten years of individual-based modelling in ecology: what have we learned and what could we
learn in the future? Ecol. Model. 115(2–3), 129 (1999)

123
332 R. T. Armstrong et al.

Groot, S.D., Mazur, P.: Non-Equilibrium Thermodynamics. North-Holland Publishing Company, Oxford
(1962)
Gyulassy, A.G., Duchaineau, M.A., Natarajan, V., Pascucci, V., Bringa, E.M., Higginbotham, A., Hamann,
B.: Topologically clean distance fields. IEEE Trans. Vis. Comput. Gr. 13(6), 1432 (2007). https://doi.
org/10.1109/TVCG.2007.70603
Gyulassy, A., Bremer, P., Pascucci, V.: Computing Morse–Smale complexes with accurate geometry. IEEE
Trans. Vis. Comput. Gr. 18(12), 2014 (2012). https://doi.org/10.1109/TVCG.2012.209
Hadwiger, H.: Vorlesungen tiber inhalt, oberfläche und isoperirnetrie (1957)
Haines, W.B.: Studies in the physical properties of soil. V. The hysteresis effect in capillary properties, and
the modes of moisture distribution associated therewith. J. Agric. Sci. 20(1), 97 (1930)
Hansen, A., Sinha, S., Bedeaux, D., Kjelstrup, S., Savani, I., Vassvik, M.: arXiv preprint arXiv:1605.02874
(2016)
Herring, A., Robins, V., Saadatfar, M., Young, B. Knackstedt, M., Sheppard, A.: (2018)
Herring, A.L., Harper, E.J., Andersson, L., Sheppard, A., Bay, B.K., Wildenschild, D.: Effect of fluid topology
on residual nonwetting phase trapping: Implications for geologic CO2 sequestration. Adv. Water Res.
62, 47 (2013)
Herring, A.L., Andersson, L., Schlüter, S., Sheppard, A., Wildenschild, D.: Efficiently engineering pore-scale
processes: the role of force dominance and topology during nonwetting phase trapping in porous media.
Adv. Water Resour. 79, 91 (2015)
Hilfer, R.: Macroscopic capillarity and hysteresis for flow in porous media. Phys. Rev. E 73(1), 016307 (2006)
Hilpert, M., Miller, C.T.: Pore-morphology-based simulation of drainage in totally wetting porous media. Adv.
Water Resour. 24(3–4), 243 (2001)
Hilpert, M., Miller, C.T., Gray, W.G.: Stability of a fluid-fluid interface in a biconical pore segment. J. Colloid
Interface Sci. 267(2), 397 (2003)
Hiraoka, Y., Nakamura, T., Hirata, A., Escolar, E.G., Matsue, K., Nishiura, Y.: In: Proceedings of the National
Academy of Sciences p. 201520877 (2016)
Hyde, S., Barnes, I., Ninham, B.: Curvature energy of surfactant interfaces confined to the plaquettes of a
cubic lattice. Langmuir 6(6), 1055 (1990)
Iglauer, S., Paluszny, A., Pentland, C.H., Blunt, M.J.: Residual CO2 imaged with X-ray micro-tomography.
Geophys. Res. Lett. 38(21) (2011)
Iuricich, F., Scaramuccia, S., Landi, C., L., : De Floriani, in SIGGRAPH ASIA 2016 Symposium on Visual-
ization (ACM, vol, vol. SA ’16, p. pp. 5:1–5:, 8, New York, NY, USA (2016). https://doi.org/10.1145/
3002151.3002166
Jiao, Y., Stillinger, F., Torquato, S.: A superior descriptor of random textures and its predictive capacity. Proc.
Nat. Acad. Sci. 106(42), 17634 (2009)
Jones, A.C., Milthorpe, B., Averdunk, H., Limaye, A., Senden, T.J., Sakellariou, A., Sheppard, A.P., Sok,
R.M., Knackstedt, M.A., Brandwood, A., et al.: Analysis of 3D bone ingrowth into polymer scaffolds
via micro-computed tomography imaging. Biomaterials 25(20), 4947 (2004)
Jones, A.C., Arns, C.H., Sheppard, A.P., Hutmacher, D.W., Milthorpe, B.K., Knackstedt, M.A.: Assessment
of bone ingrowth into porous biomaterials using MICRO-CT. Biomaterials 28(15), 2491 (2007)
Kac, M.: Can one hear the shape of a drum? Am. Math. Mon. 73(4), 1 (1966)
Kapfer, S.C., Hyde, S.T., Mecke, K., Arns, C.H., Schröder-Turk, G.E.: Minimal surface scaffold designs for
tissue engineering. Biomaterials 32(29), 6875 (2011)
Karadimitriou, N., Hassanizadeh, S., Joekar-Niasar, V., Kleingeld, P.: Micromodel study of two-phase flow
under transient conditions: quantifying effects of specific interfacial area. Water Resour. Res. 50(10),
8125 (2014)
Katz, A., Thompson, A.: Quantitative prediction of permeability in porous rock. Phys. Rev. B 34(11), 8179
(1986)
Khanamiri, H., Berg, C.F., Slotte, P.A.,Torster, O., Schlẗer, S.: Description of free energy for immiscibletwo-
fluid flow in porous media by integral geometry andthermodynamics. Water Res. Res. (in press)
Khanamiri, H.H., Torsæter, O.: Water Resour. Res
Kimura, M., Obayashi, I., Takeichi, Y., Murao, R., Hiraoka, Y.: Non-empirical identification of trigger sites
in heterogeneous processes using persistent homology. Sci. Rep. 8(1), 3553 (2018). https://doi.org/10.
1038/s41598-018-21867-z
Klain, D.A.: A short proof of Hadwiger’s characterization theorem. Mathematika 42(2), 329 (1995)
Knackstedt, M.A., Arns, C.H., Senden, T.J., Gross, K.: Structure and properties of clinical coralline implants
measured via 3D imaging and analysis. Biomaterials 27(13), 2776 (2006)
Knothe, M.T.: Whither flows the fluid in bone? An osteocyte’s perspective. J. Biomech. 36(10), 1409 (2003)

123
Porous Media Characterization Using the Minkowski… 333

Koebernick, N., Weller, U., Huber, K., Schlüter, S., Vogel, H.J., Jahn, R., Vereecken, H., Vetterlein, D.: In situ
visualization and quantification of three-dimensional root system architecture and growth using X-ray
computed tomography. Vadose Zone J 13(8) (2014)
Kramar, M., Goullet, A., Kondic, L., Mischaikow, K.: Quantifying force networks in particulate systems. Phys.
D: Nonlinear Phenom. 283, 37 (2014). https://doi.org/10.1016/j.physd.2014.05.009
Kramar, M., Levanger, R., Tithof, J., Suri, B., Xu, M., Paul, M., Schatz, M.F., Mischaikow, K.: Topology in
dynamics, differential equations, and data. Phys. D: Nonlinear Phenom. 334, 82 (2016). https://doi.org/
10.1016/j.physd.2016.02.003
Kumahor, S., de Rooij, G., Schlüter, S., Vogel, H.J.: Water flow and solute transport in unsaturated sanda—
comprehensive experimental approach. Vadose Zone J. 14(2) (2015)
Laney, D., Bremer, P.T., Mascarenhas, A., Miller, P., Pascucci, V.: Understanding the structure of the turbulent
mixing layer in hydrodynamic instabilities. IEEE Trans. Vis. Comput. Graph. 12, 1053 (2006)
Lang, C., Ohser, J., Hilfer, R.: On the analysis of spatial binary images. J. Microsc. 203(3), 303 (2001)
Lee, Y., Barthel, S.D., Dłotko, P., Moosavi, S.M., Hess, K., Smit, B.: arXiv preprint arXiv:1701.06953 (2017)
Legland, D., Kiêu, K., Devaux, M.F.: Computation of Minkowski measures on 2D and 3D binary images.
Image Anal. Stereol. 26(2), 83 (2011)
Leverett, M., et al.: Capillary behavior in porous solids. Trans. AIME 142(01), 152 (1941)
Li, T., Schlüter, S., Dragila, M.I., Wildenschild, D.: An improved method for estimating capillary pressure
from 3D microtomography images and its application to the study of disconnected nonwetting phase.
Adv. Water Resour. 114, 249 (2018)
Liu, Z., Herring, A., Robins, V.: R. In: International Symposium of the Society of Core Analysts, Armstrong
(2017)
Liu, Z., Herring, A., Arns, C., Berg, S., Armstrong, R.T.: Pore-scale characterization of two-phase flow using
integral geometry. Transp. Porous Media 118(1), 99 (2017)
Mantz, H., Jacobs, K., Mecke, K.: Utilizing Minkowski functionals for image analysis: a marching square
algorithm. J. Stat. Mech. Theory Exp. 2008(12), P12015 (2008)
Matsumoto, Y.: An Introduction to Morse Theory. AMS Bookstore, Providence, RI (2002)
McClure, J.E., Prins, J.F., Miller, C.T.: A novel heterogeneous algorithm to simulate multiphase flow in porous
media on multicore CPU–GPU systems. Comput. Phys. Commun. 185(7), 1865 (2014)
McClure, J.E., Berrill, M.A., Gray, W.G., Miller, C.T.: Influence of phase connectivity on the relationship
among capillary pressure, fluid saturation, and interfacial area in two-fluid-phase porous medium systems.
Phys. Rev. E 94(3), 033102 (2016)
McClure, J.E., Armstrong, R.T., Berrill, M.A., Schlüter, S., Berg, S., Gray, W.G., Miller, C.T.: Geometric state
function for two-fluid flow in porous media. Phys. Rev. Fluids 3, 084306 (2018). https://doi.org/10.1103/
PhysRevFluids.3.084306
Mecke, K.R.: Integral geometry in statistical physics. Int. J. Mod. Phys. B 12(09), 861 (1998)
Mecke, K.R.: Statistical Physics and Spatial Statistics, pp. 111–184. Springer, Berlin (2000)
Mecke, K.R., Sofonea, V.: Morphology of spinodal decomposition. Phys. Rev. E 56(4), R3761 (1997)
Mecke, K., Wagner, H.: Euler characteristic and related measures for random geometric sets. J. Stat. Phys.
64(3–4), 843 (1991)
Michielsen, K., De Raedt, H.: Integral-geometry morphological image analysis. Phys. Rep. 347(6), 461 (2001)
Mosser, L., Dubrule, O., Blunt, M.J.: arXiv preprint arXiv:1712.02854 (2017)
Nagel, W., Ohser, J., Pischang, K.: An integral-geometric approach for the Euler–Poincaré characteristic of
spatial images. J. Microsc. 198(1), 54 (2000)
Ohser, J., Mücklich, F.: Statistical Analysis of Microstructures in Materials Science. Wiley, Hoboken (2000)
Ohser, J., Redenbach, C., Schladitz, K.: Mesh free estimation of the structure model index. Image Anal. Stereol.
28(3), 179 (2011)
Okabe, H., Blunt, M.J.: Pore space reconstruction of vuggy carbonates using microtomography and multiple-
point statistics, Water Resour. Res. 43(12) (2007)
Okabe, H., Blunt, M.J.: Pore space reconstruction using multiple-point statistics. J. Pet. Sci. Eng. 46(1–2),
121 (2005)
Øren, P.E., Bakke, S.: Process based reconstruction of sandstones and prediction of transport properties. Transp.
Porous Media 46(2–3), 311 (2002)
Øren, P.E., Bakke, S.: Reconstruction of Berea sandstone and pore-scale modelling of wettability effects. J.
Pet. Sci. Eng. 39(3–4), 177 (2003)
Pentland, C.H., El-Maghraby, R., Iglauer, S., Blunt, M.J.: Measurements of the capillary trapping of super-
critical carbon dioxide in Berea sandstone, Geophys. Res. Lett. 38(6) (2011)
Picchi, D., Battiato, I.: The Impact of Pore-Scale Flow Regimes on Upscaling of Immiscible Two-Phase Flow
in Porous Media. Water Resour. Res. 54(9), 6683–6707 (2018)

123
334 R. T. Armstrong et al.

Porter, M.L., Wildenschild, D., Grant, G., Gerhard, J.I.: Measurement and prediction of the relationship
between capillary pressure, saturation, and interfacial area in a NAPL-water-glass bead system. Water
Resour. Res. 46(8) (2010)
Ramandi, H.L., Mostaghimi, P., Armstrong, R.T., Saadatfar, M., Pinczewski, W.V.: Porosity and permeability
characterization of coal: a micro-computed tomography study. Int. J. Coal Geol. 154, 57 (2016)
Robins, V.: Towards computing homology from finite approximations. Topol. Proc. 24, 503 (1999)
Robins, V., Wood, P.J., Sheppard, A.P.: Theory and algorithms for constructing discrete Morse complexes
from grayscale digital images. IEEE Trans. Pattern Anal. Mach. Intell. 33(8), 1646 (2011)
Robins, V., Saadatfar, M., Delgado-Friedrichs, O., Sheppard, A.P.: Percolating length scales from topological
persistence analysis of micro-CT images of porous materials. Water Resour. Res. 52(1), 315 (2016).
https://doi.org/10.1002/2015WR017937
Rücker, M., Berg, S., Armstrong, R., Georgiadis, A., Ott, H., Schwing, A., Neiteler, R., Brussee, N., Makurat,
A., Leu, L., et al.: From connected pathway flow to ganglion dynamics. Geophys. Res. Lett. 42(10), 3888
(2015)
Saadatfar, M., Takeuchi, H., Robins, V., Francois, N., Hiraoka, Y.: Pore configuration landscape of granular
crystallization. Nat. Commun. 8, 15082 (2017)
Saadatfar, M., Takeuchi, H., Robins, V., Francois, N., Hiraoka, Y.: Pore configuration landscape of granular
crystallization, Nature. Nat. Commun. 8, 15082 (2017). https://doi.org/10.1038/ncomms15082
Sahimi, M.: Flow and Transport in Porous Media and Fractured Rock: From Classical Methods to Modern
Approaches. John Wiley & Sons, Hoboken (2011)
San Jose Martinez, F., Muñoz, F., Caniego, F., Peregrina, F.: Morphological functions to quantify three-
dimensional tomograms of macropore structure in a vineyard soil with two different management regimes.
Vadose Zone J. 12(3) (2013)
Santaló, L.A.: Integral Geometry and Geometric Probability. Cambridge University Press, Cambridge (2004)
Schaap, M.G., Porter, M.L., Christensen, B.S., Wildenschild, D.: Comparison of pressure-saturation character-
istics derived from computed tomography and lattice Boltzmann simulations. Water Resour. Res. 43(12)
(2007)
Scheel, M., Seemann, R., Brinkmann, M., Di Michiel, M., Sheppard, A., Breidenbach, B., Herminghaus, S.:
Morphological clues to wet granular pile stability. Nat. Mater. 7(3), 189 (2008)
Schlüter, S., Vogel, H.J.: On the reconstruction of structural and functional properties in random heterogeneous
media. Adv. Water Resour. 34(2), 314 (2011)
Schlüter, S., Weller, U., Vogel, H.J.: Soil-structure development including seasonal dynamics in a long-term
fertilization experiment. J. Plant Nutr. Soil Sci. 174(3), 395 (2011)
Schlüter, S., Sheppard, A., Brown, K., Wildenschild, D.: Image processing of multiphase images obtained via
X-ray microtomography: a review. Water Resour. Res. 50(4), 3615 (2014)
Schlüter, S., Berg, S., Rücker, M., Armstrong, R., Vogel, H.J., Hilfer, R., Wildenschild, D.: Pore-scale dis-
placement mechanisms as a source of hysteresis for two-phase flow in porous media. Water Resour. Res.
52(3), 2194 (2016)
Schlüter, S., Henjes, S., Zawallich, J., Bergaust, L., Horn, M., Ippisch, O., Vogel, H.J., Dörsch, P.: Denitrification
in soil aggregate analogues-effect of aggregate size and oxygen diffusion. Front. Environ. Sci. 6, 17 (2018)
Schmalzing, J., Kerscher, M., Buchert, T.: arXiv preprint astro-ph/9508154 (1995)
Schmalzing, J., Górski, K.M.: Minkowski functionals used in the morphological analysis of cosmic microwave
background anisotropy maps. Mon. Not. R. Astron. Soc. 297(2), 355 (1998)
Schneider, R.: Convex Bodies: the Brunn–Minkowski Theory, 151st edn. Cambridge University Press, Cam-
bridge (2014)
Scholz, C., Wirner, F., Götz, J., Rüde, U., Schröder-Turk, G.E., Mecke, K., Bechinger, C.: Permeability of
porous materials determined from the Euler characteristic. Phys. Rev. Lett. 109(26), 264504 (2012)
Scholz, C., Wirner, F., Klatt, M.A., Hirneise, D., Schröder-Turk, G.E., Mecke, K., Bechinger, C.: Direct
relations between morphology and transport in Boolean models. Phys. Rev. E 92(4), 043023 (2015)
Serra, J.: Image analysis and mathematical morphology, pp. 424–478. (1982)
Serra, J.: Image Analysis and Mathematical Morphology. Academic Press Inc, Cambridge (1983)
Tanino, Y., Blunt, M.J.: Capillary trapping in sandstones and carbonates: dependence on pore structure. Water
Resour. Res. 48(8) (2012)
Tate, M.L.K., Steck, R., Anderson, E.J.: Bone as an inspiration for a novel class of mechanoactive materials.
Biomaterials 30(2), 133 (2009)
Tiab, D., Donaldson, E.C.: Petrophysics: Theory and Practice of Measuring Reservoir Rock and Fluid Transport
Properties. Gulf Professional Publishing, Houston (2015)
Ushizima, D., Morozov, D., Weber, G.H., Bianchi, A.G.C., Sethian, J.A., Bethel, E.W.: Augmented topological
descriptors of pore networks for material science. IEEE Trans. Vis. Comput. Gr. 18(12), 2041 (2012).
https://doi.org/10.1109/TVCG.2012.200

123
Porous Media Characterization Using the Minkowski… 335

Valavanides, M., Payatakes, A.: True-to-mechanism model of steady-state two-phase flow in porous media,
using decomposition into prototype flows. Adv. Water Resour. 24(3–4), 385 (2001)
Vogel, H.: A numerical experiment on pore size, pore connectivity, water retention, permeability, and solute
transport using network models. Eur. J. Soil Sci. 51(1), 99 (2000)
Vogel, H.J., Weller, U., Schlüter, S.: Quantification of soil structure based on Minkowski functions. Comput.
Geosci. 36(10), 1236 (2010)
Wang, Y., Rahman, S.S., Arns, C.H.: Super resolution reconstruction of μ-CT image of rock sample using
neighbour embedding algorithm. Phys. A Stat. Mech. Appl. 493, 177 (2018)
Weyl, H.: On the volume of tubes. Am. J. Math. 61(2), 461 (1939)
Whitaker, S.: The Method of Volume Averaging, vol. 13. Springer Science & Business Media, Berlin (2013)
Wildenschild, D., Sheppard, A.P.: X-ray imaging and analysis techniques for quantifying pore-scale structure
and processes in subsurface porous medium systems. Adv. Water Resour. 51, 217 (2013)
Wildenschild, D., Vaz, C., Rivers, M., Rikard, D., Christensen, B.: Using X-ray computed tomography in
hydrology: systems, resolutions, and limitations. J. Hydrol. 267(3–4), 285 (2002)
Zhang, Y., Mostaghimi, P., Fogden, A., Sheppard, A., Arena, A., Middleton, J., Armstrong, R.T.: Time-
lapsed visualization and characterization of shale diffusion properties using 4D X-ray microcomputed
tomography. Energy Fuels 32(3), 2889 (2018)
Zomorodian, A., Carlsson, G.: Computing persistent homology. Discrete Comput. Geom. 33(2), 249 (2005)

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

123

You might also like