Rock Properties From Micro-Ct Images Digital Rock Transforms For
Rock Properties From Micro-Ct Images Digital Rock Transforms For
Rock Properties From Micro-Ct Images Digital Rock Transforms For
PII: S0309-1708(19)30143-5
DOI: https://doi.org/10.1016/j.advwatres.2019.103419
Reference: ADWR 103419
Please cite this article as: Dr Nishank Saxena , Amie Hows , Ronny Hofmann , Faruk O. Alpak ,
Jesse Dietderich , Matthias Appel , Justin Freeman , Hilko De Jong , Rock properties from micro-
CT images: digital rock transforms for resolution, pore volume, and field of view, Advances in Water
Resources (2019), doi: https://doi.org/10.1016/j.advwatres.2019.103419
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.
1
ROCK PROPERTIES FROM MICRO-CT IMAGES: DIGITAL ROCK TRANSFORMS FOR
RESOLUTION, PORE VOLUME, AND FIELD OF VIEW
Nishank Saxena, Amie Hows, Ronny Hofmann, Faruk O. Alpak, Jesse Dietderich,
2
ABSTRACT
Digital Rock Physics is a promising approach for achieving more, cheaper, and faster rock
potential, we must correctly interpret the digitally derived properties in the context of the
limitations imposed by imaging constraints. To this end, we show that a combination of limited
image resolution, a biased segmentation of images with coarse resolution, and a finite field of
permeability (up to a factor of 10) calculated using the Digital Rock Physics (DRP). We
demonstrate these imaging limitations can be overcome by identifying good measures of image
resolution and representative elementary volume and applying appropriate transforms. These
transforms expand the operating envelop of DRP. Transforms for finite image resolution and
limited field of view can be estimated directly from the micro-CT images. However,
suggest how insights from these transforms can be used to define operating envelopes and
optimize imaging resolution and field of view to achieve more reliable results from digital rock
3
1. INTRODUCTION
Digital Rock Physics (DRP) or simply Digital Rock (DR) is a rapidly advancing technology
that relies on digital images of rocks to simulate multiphysics at the pore-scale and predict
images of rocks allow for the quantification of any potential links between effective properties of
rocks and rock texture shaped by geologic forces. For the energy industry, DRP aims to achieve
more, cheaper, and faster results as compared to conventional laboratory measurements (Andrä
et al., 2013a, 2013b, Arns et al., 2005, 2002; Blunt et al., 2013; Dvorkin et al., 2011, 2008;
Kanckstedt et al., 2001; Knackstedt et al., 2008; Knackstedt et al., 2009; Øren et al., 2006;
Saxena and Mavko, 2016; Saxena et al., 2017). A typical workflow of DRP includes, pore-scale
imaging, image processing and segmentation, numerical simulations, and rock property analysis
(Figure 1). In this paper, we discuss a critical but often overlooked step which is the post
processing and transformation of the computed results generated by the DRP workflow to
Porosity and absolute permeability are among the most fundamental rock properties, and thus
accurate estimation of porosity and permeability using images is crucial for maturing the Digital
and permeability and those computed using micro-CT images for reservoir rocks. This bias in
numerical simulation engines or the choice of boundary conditions because these can account for
up to 10-30% variation in computed permeability (Saxena et al., 2017a). This bias also cannot be
4
differences are within 1-2 porosity units as long as all phases are segmented consistently (Saxena
et al., 2017b).
The objectives of this paper are multifold. First, we illustrate the discrepancy between image-
derived and laboratory measured properties using a considerable dataset. Second, we investigate
the reasons for the observed discrepancy between image-derived and laboratory measured
permeability. We apply the transforms on the dataset and observe significant improvements in
The organization of this paper is as follows: we begin the paper with a brief description of
various microstructures and rocks considered in this study. Next, we present expressions for
various transforms that are required for micro-CT computed properties and discuss the
implications of these transforms. We conclude with insights and recommendations for estimating
microstructures including 4 monodispersed grain packs, 5 outcrop rocks that were previously
used for several benchmark studies (H. Andrä et al., 2013; Saxena et al., 2017a, 2017b), and a
global database of reservoir rocks (further details are in Appendix A). All rocks contain
negligible amounts of clay (< 2%). Laboratory measured porosity is the so-called total porosity
(i.e., the ratio of the entire pore space in a rock to its bulk volume). Permeability was measured
in the laboratory by pushing brine through the rock under laminar flow conditions. For digital
rock computations, mini-plugs were extracted from the original plugs used for laboratory
5
measurements and were subsequently imaged with a state-of-the-art micro-CT detector at
roughly the same image resolution (voxel size ~ 2 µm; Table 1).
All three-dimensional micro-CT images were filtered with a proprietary non-local means
filter algorithm (Alon Arad; personal communication) to boost signal to noise ratio. Upon
filtering these images were segmented using spatial Fuzzy C-Means clustering (sFCM) (Chuang
et al., 2006) algorithm into hydraulically conductive pores and non-conductive minerals (Figure
2). The sFCM algorithm combines spatial statistics for each voxel with the Fuzzy C-Means
degree of membership to segment the image into its constituent phases. Volume fraction of pores
estimated from the segmented image is referred here as the image porosity. Permeability was
computed using a single-phase Lattice Boltzmann Method (LBM) solver. Further details on the
numerical engine, convergence criteria, and boundary conditions can be found in Appendix B.
For a more general introduction to the LBM, we refer to Benzi et al., (1992), Succi (2001),
Keehm et al., (2001), Saxena et al. (2017a) and Alpak et al., (2018). The computed permeability
permeability are shown in Figures 3 – 5. We note a systematic offset when comparing laboratory
measurements of porosity and permeability with those computed using micro-CT images. We
find that digital rock analysis leads to systematic underestimation of porosity by down to a factor
factor of 10, for samples with laboratory measured permeability less than 100 mDarcy. This
observation is consistent with the recent study by Saxena et al. (2018) who predicted that due to
imaging considerations the lower bound for simulating accurate permeability using micro-CT
6
images of 2 µm voxel size is around 100 mDarcy. This lower bound drops to 25 mDarcy for
There are many key considerations when comparing laboratory data with digital rock results.
Firstly, the quality of digital rock simulations is determined by a variety of factors, including the
degree of resolved rock components, the quality of image reconstruction, and the signal to noise
ratio in images. Secondly, digital rock computations are impacted by every step of the workflow
(Figure 1) including any biases in image segmentation that can lead to under or overestimation of
porosity. Finally, the two sets of results were obtained at different scales (or field of view) since
laboratory measurements were performed on 40 mm cylindrical core plugs whereas digital rock
computations were carried out on 4 mm size cubes extracted from the larger core plugs used for
laboratory measurements. This means that digital rock simulations were computed on 1/1000th of
the rock volume used for laboratory measurements. This substantial difference in the physical
size of the analyzed sample creates a bias in the computed results even for perfectly homogenous
samples with no geological layering within the plug. These biases must be corrected before
comparing image simulation results with laboratory measurements. To this end, we propose the
following relations
𝜙∞ = 𝜙𝐼 ⁄(𝛼𝑅 𝛼𝑆 𝛼𝑉 ) , (1)
𝑘∞ = 𝑘𝐼 ⁄(𝛽𝑅 𝛽𝑆 𝛽𝑉 ) . (2)
In equations 1 and 2, 𝜙 and 𝑘 denote porosity and permeability, respectively. Subscript “I”
denotes properties computed using micro-CT images. Subscript “∞” denote the corresponding
properties of a hypothetical image of infinite resolution and infinitely large field of view that has
been segmented perfectly. To estimate total porosity (𝜙𝑇 ), excess pore volume within porous
clays identified in images must be added to the corrected porosity 𝜙∞ . Parameters 𝛼 and 𝛽 are
7
the correction factors, and subscripts “𝑅”, “𝑆”, and “𝑉” denote transforms for image resolution,
bias in estimating pore volume due to uncertainty in image segmentation, and field of view (or
upscaling), respectively. There may be additional correction factors required for rocks whose
properties are sensitivity to various other factors, including differences in stress state, swelling of
clays, electrochemical effects, and severe heterogeneity to name but a few. We ignore these
additional transforms as these do not apply to the rocks analyzed in this study. In the next
section, we will derive expressions for parameters 𝛼 and 𝛽 that depend on microstructural (or
geometric) attributes that can be extracted directly using the original micro-CT image.
We begin with estimating the 𝛼 parameters. Due to limited image resolution, a portion of
rock porosity is always unresolved. This portion is significant when porosity is estimated using
x-ray computed tomography images of rocks. Recently, Saxena et al. (2019) derived the
𝐺𝑀
−
𝛼𝑅 = 𝑒 𝑙𝑜𝑔10 (𝑁𝑀 )
. (3)
The parameter 𝛼𝑅 describes how porosity contained within self-similar pore structures in rocks
can be predicted using an image even when these structures are below image resolution, and
therefore can be used to correct for the “missing” sub-resolution porosity. For further details on
derivation of the result in equation 3 refer Appendix C. The parameter 𝑁 denotes the ratio of
pore throat size (entry pore throat 𝐷𝑑 for mercury in a MICP experiment) and image voxel size
(∆𝑥), and thus a better measure of the degree to which the entry pore throat is resolved compared
to voxel size. Parameter 𝐺 is the so-called curvature of a MICP curve (Thomeer, 1960).
Parameter 𝐺 is precisely 0 for pipes, approximately 0.2 for sandstones, and approximately 0.3 for
8
carbonates. Subscript “M” denotes the laboratory-measured parameter, whereas subscript “I”
denotes the value inferred from an image; at infinite image resolution the two sets are the same.
MICP curves upon the required closure transforms (Thomeer, 1960). An example is shown in
Figure 6a for which 𝐺𝑀 ≈ 𝐺𝐼 = 0.3 and 𝑁𝑀 ≈ 𝑁𝐼 = 7. Note that it is important that the
fractional bulk volume of mercury (i.e., mercury volume / rock volume) and not for saturation
curves (i.e., mercury volume / pore volume) (Figure 6b) because raw image porosity is not equal
to total porosity due to the resolution limit. Upon analysis of laboratory-measured MICP data
and multiple micro-CT images (acquired at various resolutions) of the grain packs and the
outcrop rocks we find that 𝐺𝑀 ≈ 𝐺𝐼 and 𝑁𝑀 < 𝑁𝐼 (Figure 7). Thus, entry pore throat size
estimated using micro-CT images is generally larger than that inferred from laboratory
measurements. This bias occurs due to over segmentation of pores in coarse resolution images
(especially when 𝑁 < 5) and thus can be addressed using a more robust segmentation approach.
From this analysis we propose the following empirical relation to calculate 𝑁𝑀 from 𝑁𝐼 (Figure
7):
We now focus on image segmentation correction for porosity (i.e., factor 𝛼𝑆 ). This correction
depends on the quality of image segmentation which itself is sensitive to image resolution, noise,
and segmentation method (Berg et al., 2018; Saxena et al., 2017b). To achieve perfect
segmentation (i.e., 𝛼𝑆 = 1) it is necessary to correctly identify all types of rock pores, as well as
optimally partition pores and grains that are visible in an image. This can only be estimated if
9
checked for perfect segmentation. The result in equation 3 can yield constraints on image
porosity and segmentation, if 𝜙∞ can be inferred using laboratory measured porosity (e.g.,
MICP, NMR). This is because we can predict the expected image porosity (𝜙𝑀 ) using the
following relation:
𝐺
−
𝜙𝑀 = 𝜙∞ 𝑒 𝑙𝑜𝑔10 (𝑁𝑀 )
. (5)
constraints for image segmentation, the preferred protocol should be to improve the quality of
Finally, the field of view correction for porosity is expected be minimal (i.e., 𝛼𝑉 = 1) if large
enough field of view is selected for computation of porosity. Various studies have shown that
fluctuations in porosity values stabilize rapidly with increasing in sample volume (H. Andrä et
al., 2013; Keehm, 2003; Mu et al., 2016; Saxena et al., 2018). The necessary field of view to
minimize the impact of this transform (i.e., 𝛼𝑉 = 1) is achieved if 5 grains are captured in each
dimension (i.e. ~125 grains in a 3D volume). However, if 𝛼𝑉 ≠ 1, this would imply a very small
In this section, we derive expressions for the 𝛽 parameters that are needed to transform
permeability estimated using a micro-CT image. The image resolution correction factor (𝛽𝑅 )
follows directly from the observation that if the image resolution is poor (i.e., 𝑁 < 5) then MICP
simulations overestimate the entry pore throat size, and that permeability depends on the square
10
𝑁 2 𝑁 2
𝛽𝑅 = (𝑁 𝐼 ) ≈ (𝑁 −2𝑒 −0.4(𝑁
𝐼
𝐼 −1)
) . (5)
𝑀 𝐼
For cases where segmentation is inherently uncertain, we propose a transform for image
segmentation that is based on the observation that when a micro-CT image is segmented with
various segmentation algorithms (Figure 8), then the computed permeability values (𝑘) exhibit
3
𝜙 1−𝜙𝑟𝑒𝑓 2
𝑘 = 𝑘(1) (𝜙 ) ( ) . (6)
𝑟𝑒𝑓 1−𝜙
∆𝜙 bias in porosity due to incorrect image segmentation leads to the following bias in
permeability
3𝑘𝑟𝑒𝑓
∆𝑘 ≈ ∆𝜙 . (7)
𝜙𝑟𝑒𝑓
For example, consider a rock whose porosity is 20 porosity units (p.u.) (i.e., percentage of total
rock volume) and permeability is 100 mDarcy. If the bias in estimating porosity due to image
segmentation is +3 p.u., then the bias in permeability will be +45 mDarcy. Therefore, when pore
𝜙 3 1−𝜙 2
𝛽𝑆 = (𝜙 𝐼 ) ( 1−𝜙𝑀 ) , (8)
𝑀 𝐼
where, 𝜙𝑀 is the rock porosity containing only pores larger than the image voxel size. It follows
directly from equation 8 that the transforms related to image segmentation can only be applied if
11
Correction for field of view or rock volume is within the purview of the topic of
Representative Elementary Volume (REV) (Bear, 1975; Hill, 1963; Torquato, 1991). Here we
restrict the discussion to homogenous rocks that are completely devoid of fine scale laminations,
dual pore throat systems, or other microstructural details that cannot be captured in a digital
image of 4 mm. Various authors (Mu et al., 2016; Saxena et al., 2018) have shown that
permeability computed for concentric cubes of segmented images do not completely stabilize
even when digital images are larger than 4 mm (in one direction) for even homogenous rocks
(e.g., Fontainebleau). Therefore, we must consider how the vast scale differences between
image-derived and laboratory measured permeability impact the comparison especially given
that digital rock images have disproportionately fewer pore throats (i.e., the narrowest
constrictions in the largest pores) compared to those present in 40 mm size plugs used for
laboratory measurements. The MICP experiment is a good analog for a comparison (Figure 9).
The pressure at which mercury begins to enter a digital rock, in a numerical simulation, is
considerably lower compared to the entry pressure inferred from laboratory measurement of
MICP. This is because mercury first enters a digital rock sample via “outer” pore bodies instead
of pore throats because they can be accessed at much lower pressure compared to the pore
throats – compensating for the biases introduced by the outer pore bodies to flow properties is
referred here as closure correction. Although this phenomenon also occurs in a physical MICP
measurement, the volumetric contributions of the outer pore body is negligible compared to the
total volume of mercury injected in the physical sample simply due to considerably larger sample
size. This relative contribution is larger for rocks with lower permeability (Figure 9). We define
the ratio of “outer” pore body size to image voxel size as 𝑁𝐼𝑁𝐶 . For a numerical simulation on a
12
digital image, the apparent pore throat size (𝑁𝑒𝑓𝑓 ) must be between entry pore throat (𝑁𝐼 ) and
pore body size (𝑁𝐼𝑁𝐶 ), and thus we can put strict bounds on the correction factor:
where 𝜔 is the ratio of 𝑁𝐼𝑁𝐶 and 𝑁𝐼 . It is apparent that parameter 𝛽𝑉 must decline with larger
field of view, and this decline would be steeper for permeable rocks. Therefore, simply assuming
𝛽𝑉 ≈ 𝜔2 would lead to an overcorrection of image permeability for all rocks. We now consider
an alternative approach that approximates pores as sinusoidal channels with diameters 𝑁𝐼𝑁𝐶 and
𝑁𝐼 and use empiricism to analyze how effective channel width changes with field of view and
𝛽𝑉 = 𝜔 2𝑝 . (10)
If the digital rock image has sufficiently large field of view to reproduce the fluid flow observed
in a laboratory experiment then 𝑝 ≈ 0 and 𝛽𝑉 ≈ 1, hence no field of view transforms are required.
However, if flow is dominated by the outer pore body then the calculated permeability will not
be representative, and thus 𝑝 = 1. This situation might occur if there are too few grains in the
digital rock such that fluid flow entering the rock pores is overestimated. This effect could be
Next, we empirically establish the variation of coefficient 𝑝 with parameters that can be
extracted directly using the original micro-CT image of a rock. We back calculate the parameter
𝑝 using the laboratory measured permeability after compensating for both resolution and
segmentation transforms (equations 5 and 8). This analysis was repeated for grain packs and
outcrop rocks of various sizes (Figure 10). We find that the following empirical relation
13
−2
𝑝 = √1 − 𝑒 −𝑎(𝐷𝑑 𝜙∞𝑁𝑅𝐸𝑉 ) , where 𝑎 = 3 .
0.5
(11)
In equation 11, 𝑁𝑅𝐸𝑉 denotes the ratio of field of view and the effective grain size 𝐷𝑒𝑓𝑓 (as
defined by Saxena et al., 2018) such that it approximates the grains across the flow direction.
The expression in equation 11 is not unique and is merely a good fit. The magnitude of 𝛽𝑉
increases with the ratio of 𝑁𝐼𝑁𝐶 and 𝑁𝐼 , decreasing pore throat size, decreasing porosity,
segmentation, and field of view, should be kept at a minimum and the preferred protocol is to
compensate for these biases by improving the image when possible rather than apply a correction
to the computed permeability. For instance, the need for any resolution and segmentation
transform can be minimized by re-imaging the rock at a higher resolution which can also
improve image quality for image segmentation. Similarly, it is possible to compensate for
closure correction effects using appropriate boundary conditions for flow around the outer pore
bodies of the digital image so that the field of view correction can be minimized. Further detailed
We now apply the transforms to estimate porosity and permeability for the database of
reservoir rocks. These rocks were not used for any calibration to derive the expressions or fit
parameters for the transforms. We also include results from the recent digital rock study by
Chhatre et al. (2018). Values for image resolution transforms (𝛼𝑅 , 𝛽𝑅 ) were estimated directly
using the segmented micro-CT images. The comparison between laboratory measured porosity
and permeability and those computed using Digital Rock improves upon application of the
14
image-derived resolution transform (Figures 11 and 12). This comparison further improves when
field of view transforms and image segmentation transforms are applied (Figures 13 and 14). The
trend between porosity and permeability, after transforms for resolution and field of view,
compares well with those measured in the laboratory (Figure 15). This comparison improves
further when laboratory-measured porosity is used to compensate for the segmentation bias in
porosity (Figure 16). For rocks with low permeability values (< 5 mDarcy), even upon
corrections for image resolution and field of view, we still note relatively large disagreement
between image derived and laboratory measured permeability. There can be several reasons for
this discrepancy, including larger uncertainty in segmenting poorly resolved images with
relatively small pore throats, inability of the resolution and field of view transforms to capture
the required microstructural details, and of course, relatively large uncertainty in measuring low
permeability in laboratory.
The transforms discussed in this paper should be applied prior to comparing image computed
properties with those measured in the laboratory. Still, prior to any calculations, the rock samples
should be imaged at sufficiently high image resolution and large field of view to minimize the
magnitude of any corrections. Unfortunately, the magnitude of corrections due to limited image
resolution can only be minimized at the expense of corrections required for limited field of view.
This is because current micro-CT detectors can only capture a limited number of voxels and can
image small features at higher resolution or large features at coarser resolution. This tradeoff
between image resolution and field of view is insignificant for porosity, as REV requirements
can be generally met even for a very small field of view (Saxena et al., 2018). Therefore, to
estimate porosity from images, it is advisable to image rocks at the highest available image
15
resolution (as per equation 3) so that the magnitude of correction can be minimized. However,
for permeability, the tradeoff between image resolution and field of view can be significant
(Saxena et al., 2018). Still an optimal combination of the imaging parameters exists for which the
We now illustrate that using rough estimates of expected rock properties it is possible to
establish an optimal combination of imaging parameters (i.e., field of view and voxel size). We
consider a case where the micro-CT detector can capture up to 20003 voxels (i.e., M = 2000). We
consider four rocks: a high permeability sample (𝑘∞ = 1000 mDarcy, 𝜙∞ = 0.3, 𝐺𝑀 = 0.1), an
intermediate permeability sample (𝑘∞ = 100 mDarcy, 𝜙∞ = 0.2, 𝐺𝑀 = 0.2), a low permeability
sample (𝑘∞ = 10 mDarcy, 𝜙∞ = 0.15, 𝐺𝑀 = 0.3), and a very low permeability sample (𝑘∞ = 1
To calculate the magnitude of the transforms for permeability, for the four rocks, we need to
calculate factors 𝛽𝑅 and 𝛽𝑉 for the combinations of voxel size (∆𝑥; assumed to vary between 0.1-
10 microns) and field of view (M x ∆𝑥). Factors 𝛽𝑅 and 𝛽𝑉 depend on parameters 𝑁𝑀 , 𝐺𝑀 , 𝑁𝑅𝐸𝑉 ,
and the ratio 𝑁𝐼𝑁𝐶 ⁄𝑁𝐼 . These parameters can be estimated since for the four rocks we know
porosity and permeability. Here we assume 𝑁𝐼𝑁𝐶 ⁄𝑁𝐼 = 𝜔 = 5. The parameter 𝑁𝑀 can be
estimated using pore throat size 𝐷𝐷 for an image of voxel size ∆𝑥 using 𝑁𝑀 = 𝐷𝐷 ⁄∆𝑥. Pore
throat size can be empirically related to porosity and permeability using Thomeer’s permeability
𝜙∞ 𝐷𝐷 2
𝑘∞ = 38068𝐺𝑀 −4/3 ( 213
) . (12)
In equation 12, 𝑘∞ is laboratory measured permeability (in mDarcy), 𝐷𝐷 is entry pore throat size
(in µm), and 𝜙∞ is laboratory-measured porosity (in fractions). Using equation 12, we can
estimate 𝐷𝐷 for a given combination of porosity (𝜙∞ ) and permeability (𝑘∞ ). The parameter
16
𝑁𝑅𝐸𝑉 can be calculated using 𝑁𝑅𝐸𝑉 = M x ∆𝑥 /𝐷𝑒𝑓𝑓 . We can infer 𝐷𝑒𝑓𝑓 using the following
empirical relation between grain size and pore throat size suggested by (Saxena et al., 2018):
Hence, using equations 12 and 13, we can estimate 𝛽𝑅 and 𝛽𝑉 for the four rocks as a function of
voxel size (∆𝑥) and field of view (M x ∆𝑥). As expected, we find that the magnitude of the
required corrections for high permeability rock are not significant (~10%), and the optimal voxel
size for imaging is ≈ 2.7 microns, which results in a maximum field of view of 5.4 mm in each
dimension (Figure 17). The intermediate permeability rock can be imaged at roughly the same
voxel size as the higher permeability rock, but the magnitude of corrections is greater (~40%).
The magnitude of corrections is significantly larger for the low permeability rock (~ factor of 2).
Moreover, the low permeability rock should be imaged at higher resolution (i.e., voxel size < 2
microns). Interestingly, the very low permeability rock should be imaged at a very high
resolution (voxel size < 1 micron) to minimize already very large corrections (Figure 17). One
that a larger number of voxels can be captured so that images of higher resolution can be
obtained for large enough field of view. An example is shown in Figure 18 which presents the
same information as in Figure 17 but for a micro-CT detector that can capture up to 50003 voxels
(i.e., M = 5000). This analysis can be performed to infer optimal imaging parameters for a given
7. CONCLUSIONS
Modern imaging can identify features that are larger than the image voxel size (e.g., µm for
X-ray computer tomography, nm for scanning electron microscopy). A portion of rock porosity
17
is always unresolved due to finite image resolution regardless of the imaging technique. This
portion, however, is significant when porosity is estimated using x-ray computed tomography
(micro-CT) images of rocks. Therefore, micro-CT imaging is not a suitable technique for
estimating total porosity. As a minimum, image-derived porosity must be corrected for limited
image resolution before it can be compared with laboratory-measured total porosity. We use
concepts of capillary physics in rocks to quantify the impact of image resolution on image-
derived porosity and develop novel transforms to derive the corrected porosity that compensate
for the limited image resolution without the need for higher resolution imaging that is only
possible at the expense of image field of view or physical laboratory measurement. Furthermore,
the sub-resolution pore volume, predicted by our method, can also be used to correct the fluid
saturation inferred from multiphase flow simulations on segmented micro-CT images for the
missing pore volume. Image resolution also impacts permeability. Coarser image resolution
leads to an artificial increase in pore throat size in segmented images and thus permeability is
overestimated. We find that finite sample size or field of view of digital rocks also leads to a
laboratory permeability which are performed on samples with significantly larger field of view
compared to numerical simulations. We quantify this effect and suggest relevant transforms
applicable to relatively homogenous rocks. We also suggest a transform that can account for the
porosity is available.
We conclude that these transforms are critically important for estimating meaningful
properties of conventional sandstone and carbonate reservoir rocks and can allow us to further
mature Digital Rock as a technology for existing and future fields. Still, whenever possible, the
18
application of transforms should be kept at a minimum and the preferred protocol is to
compensate for these biases by improving the image when possible rather than apply a correction
to the computed property. Additionally, understanding the sensitivities of the transforms allows
us to balance the tradeoff between imaging resolution and field of view to achieve more accurate
8. ACKNOWLEDGEMENTS
We thank L. Taras Bryndzia, Saad Saleh, Dmitry Shaporov, Chaitanya Pradhan, Majeed
Shaikh, Kunj Tandon, Umang Agarwal, Steffen Berg, Stefan Hertel, Marisa Rydzy, and Ove
19
9. REFERENCES
Adler, P.M., Jacquin, C.G., Quiblier, J.A., 1990. Flow in simulated porous media. Int. J. Multiph.
Flow 16, 691–712. https://doi.org/10.1016/0301-9322(90)90025-E
Alpak, F.O., Gray, F., Saxena, N., Dietderich, J., Hofmann, R., Berg, S., 2018. A distributed
parallel multiple-relaxation-time lattice Boltzmann method on general-purpose graphics
processing units for the rapid and scalable computation of absolute permeability from high-
resolution 3D micro-CT images. Comput. Geosci. https://doi.org/10.1007/s10596-018-
9727-7
Andrä, Heiko, Combaret, N., Dvorkin, J., Glatt, E., Han, J., Kabel, M., Keehm, Y., Krzikalla, F.,
Lee, M., Madonna, C., Marsh, M., Mukerji, T., Saenger, E.H., Sain, R., Saxena, N., Ricker,
S., Wiegmann, A., Zhan, X., 2013a. Digital rock physics benchmarks-part II: Computing
effective properties. Comput. Geosci. 50, 33–43.
https://doi.org/10.1016/j.cageo.2012.09.008
Andrä, Heiko, Combaret, N., Dvorkin, J., Glatt, E., Han, J., Kabel, M., Keehm, Y., Krzikalla, F.,
Lee, M., Madonna, C., Marsh, M., Mukerji, T., Saenger, E.H., Sain, R., Saxena, N., Ricker,
S., Wiegmann, A., Zhan, X., 2013b. Digital rock physics benchmarks-Part I: Imaging and
segmentation. Comput. Geosci. 50, 25–32. https://doi.org/10.1016/j.cageo.2012.09.005
Andrä, H., Combaret, N., Dvorkin, J., Glatt, E., Han, J., Kabel, M., Keehm, Y., Krzikalla, F.,
Lee, M., Madonna, C., Marsh, M., Mukerji, T., Saenger, E.H., Sain, R., Saxena, N., Ricker,
S., Wiegmann, A., Zhan, X., 2013. Digital rock physics benchmarks-Part I: Imaging and
segmentation. Comput. Geosci. 50. https://doi.org/10.1016/j.cageo.2012.09.005
Arns, C.H., Bauget, F., Limaye, A., Sakellariou, A., Senden, T., Sheppard, A., Sok, R.M.,
Pinczewski, V., Bakke, S., Berge, L.I., Oren, P., Knackstedt, M., 2005. Pore-Scale
Characterization of Carbonates Using X-Ray Microtomography. SPE J. 10, 26–29.
https://doi.org/10.2118/90368-PA
Arns, C.H., Knackstedt, M. a., Pinczewski, W.V., Garboczi, E.J., 2002. Computation of linear
elastic properties from microtomographic images: Methodology and agreement between
theory and experiment. Geophysics 67, 1396. https://doi.org/10.1190/1.1512785
Bear, J., 1975. Dynamics of Fluids in Porous Media. Soil Sci. https://doi.org/10.1097/00010694-
197508000-00022
Benzi, R., Succi, S., Vergassola, M., 1992. The lattice Boltzmann equation: theory and
applications. Phys. Rep. https://doi.org/10.1016/0370-1573(92)90090-M
Berg, S., Saxena, N., Shaik, M., Pradhan, C., 2018. Generation of ground truth images to validate
micro-CT image-processing pipelines. Lead. Edge 37, 412–420.
https://doi.org/10.1190/tle37060412.1
Berryman, J.G., Blair, S.C., 1986. Use of digital image analysis to estimate fluid permeability of
porous materials: Application of two-point correlation functions. J. Appl. Phys. 60, 1930–
1938. https://doi.org/10.1063/1.337245
20
Blunt, M.J., Bijeljic, B., Dong, H., Gharbi, O., Iglauer, S., Mostaghimi, P., Paluszny, A.,
Pentland, C., 2013. Pore-scale imaging and modelling. Adv. Water Resour. 51, 197–216.
https://doi.org/10.1016/j.advwatres.2012.03.003
Brooks, R., Corey, A., 1964. Hydraulic properties of porous media. Hydrol. Pap. Color. State
Univ. 3, 37 pp. https://doi.org/10.13031/2013.40684
Chhatre, S.S., Sahoo, H., Leonardi, S., Vidal, K., Rainey, J., Braun, E.M., Patel, P., 2018. A
Blind Study of Four Digital Rock Physics Vendor Laboratories on Porosity, Absolute
Permeability, and Primary Drainage Capillary Pressure Data on Tight Outcrops.
Chuang, K.-S., Tzeng, H.-L., Chen, S., Wu, J., Chen, T.-J., 2006. Fuzzy c-means clustering with
spatial information for image segmentation. Comput. Med. Imaging Graph. 30, 9–15.
https://doi.org/10.1016/j.compmedimag.2005.10.001
Churcher, P.L., French, P.R., Shaw, J.C., Schramm, L.L., 1991. Rock Properties of Berea
Sandstone, Baker Dolomite, and Indiana Limestone, in: SPE International Symposium on
Oilfield Chemistry. pp. 431–466. https://doi.org/10.2118/21044-MS
Dvorkin, J., Armbruster, M., Baldwin, C., Fang, Q., Derzhi, N., Gomez, C., Nur, B., Nur, A.,
Mu, Y., 2008. The future of rock physics: Computational methods vs. lab testing. First
Break 26, 63–68.
Dvorkin, J., Derzhi, N., Diaz, E., Fang, Q., 2011. Relevance of computational rock physics.
Geophysics 76, E141–E153. https://doi.org/10.1190/geo2010-0352.1
Hill, R., 1963. Elastic properties of reinforced solids: Some theoretical principles. J. Mech. Phys.
Solids 11, 357–372. https://doi.org/10.1016/0022-5096(63)90036-X
Hilpert, M., Miller, C.T., 2001. Pore-morphology-based simulation of drainage in totally wetting
porous media. Adv. Water Resour. 24, 243–255. https://doi.org/10.1016/S0309-
1708(00)00056-7
Hinch, E.J., 1977. An averaged-equation approach to particle interactions in a fluid suspension.
J. Fluid Mech. 83, 695–720. https://doi.org/10.1017/S0022112077001414
Howells, I.D., 1974. Drag due to the motion of a Newtonian fluid through a sparse random array
of small fixed rigid objects. J. Fluid Mech. 64, 449–476.
https://doi.org/10.1017/S0022112074002503
Kanckstedt, M. a., Sheppard, a. P., Sahimi, M., 2001. Pore network modelling of two-phase
flow in porous rock: The effect of correlated heterogeneity. Adv. Water Resour. 24, 257–
277. https://doi.org/10.1016/S0309-1708(00)00057-9
Keehm, Y., 2003. Computational Rock Physics: Transport Properties in Porous Media and
Applications. Stanford Univ.
Keehm, Y., Mukerji, T., Nur, A., 2001. Computational rock physics at the pore scale: Transport
properties and diagenesis in realistic pore geometries. Lead. Edge.
https://doi.org/10.1190/1.1438904
21
Knackstedt, Mark, Latham, S., Madadi, M., Sheppard, A., Varslot, T., Arns, C., 2009. Digital
rock physics: 3D imaging of core material and correlations to acoustic and flow properties.
Leas. Edge 28, 28–33. https://doi.org/10.1190/1.3064143
Knackstedt, M, Latham, S., Madadi, M., Sheppard, A., Varslot, T., Arns, C., 2009. Digital rock
physics: 3D imaging of core material and correlations to acoustic and flow properties. Lead.
Edge 28, 28–33. https://doi.org/10.1190/1.3064143
Knackstedt, M.A., Arns, C., Madadi, M., Sheppard, A.P., Latham, S., Sok, R., Bächle, G., Eberli,
G., 2008. Elastic and flow properties of carbonate core derived from 3D X ray‐ CT images.
SEG Tech. Progr. Expand. Abstr. 2008 27, 1804–1809. https://doi.org/10.1190/1.3059394
Leverett, M.C., 1941. Capillary Behavior in Porous Solids. Trans. AIME 142, 152–169.
https://doi.org/10.2118/941152-G
Mu, Y., Sungkorn, R., Toelke, J., 2016. Identifying the representative flow unit for capillary
dominated two-phase flow in porous media using morphology-based pore-scale modeling.
Adv. Water Resour. 95, 16–28. https://doi.org/10.1016/j.advwatres.2016.02.004
Øren, P., Bakke, S., Rueslåtten, H., 2006. Digital core laboratory: Rock and flow properties
derived from computer generated rocks. Int. Symp. Soc. Core Anal. 1–12.
Rumpf, H.C.H., Gupte, A.R., 1971. Einflüsse der Porosität und Korngrößenverteilung im
Widerstandsgesetz der Porenströmung. Chemie Ing. Tech. - CIT 43, 367–375.
https://doi.org/10.1002/cite.330430610
Saenger, E.H., Enzmann, F., Keehm, Y., Steeb, H., 2011. Digital rock physics: Effect of fluid
viscosity on effective elastic properties. J. Appl. Geophys. 74, 236–241.
https://doi.org/10.1016/j.jappgeo.2011.06.001
Sain, R., Mukerji, T., Mavko, G., 2014. How computational rock-physics tools can be used to
simulate geologic processes, understand pore-scale heterogeneity, and refine theoretical
models. Lead. Edge 33, 324–334. https://doi.org/10.1190/tle33030324.1
Saxena, N., Hofmann, R., Alpak, F.O., Berg, S., Dietderich, J., Agarwal, U., Tandon, K., Hunter,
S., Freeman, J., Wilson, O., 2017a. References and benchmarks for pore-scale flow
simulated using micro-ct images of porous media and digital rocks. Adv. Water Resour.
109, 211–235. https://doi.org/https://doi.org/10.1016/j.advwatres.2017.09.007
Saxena, N., Hofmann, R., Alpak, F.O., Dietderich, J., Hunter, S., Day-Stirrat, R.J., 2017b. Effect
of image segmentation & voxel size on micro-CT computed effective transport & elastic
properties. Mar. Pet. Geol. 86, 972–990. https://doi.org/10.1016/j.marpetgeo.2017.07.004
Saxena, N., Hows, A., Hofmann, R., O. Alpak, F., Freeman, J., Hunter, S., Appel, M., 2018.
Imaging & computational considerations for image computed permeability: operating
envelope of digital rock physics. Adv. Water Resour. 116, 127–144.
https://doi.org/10.1016/j.advwatres.2018.04.001
Saxena, N., Hows, A., Ronny, H., Justin, F., Matthias, A., 2019. Estimating Pore Volume of
Rocks from Pore-Scale Imaging. Transp. Porous Media 129, 403–412.
22
https://doi.org/10.1007/s11242-019-01295-x
Saxena, N., Mavko, G., 2016. Estimating elastic moduli of rocks from thin sections: Digital rock
study of 3D properties from 2D images. Comput. Geosci. 88, 9–21.
https://doi.org/10.1016/j.cageo.2015.12.008
Saxena, N., Mavko, G., Hofmann, R., Srisutthiyakorn, N., 2017c. Estimating permeability from
thin sections without reconstruction: Digital rock study of 3D properties from 2D images.
Comput. Geosci. 102, 79–99. https://doi.org/10.1016/j.cageo.2017.02.014
Succi, S., 2001. The lattice Boltzmann equation for Fluid Dynamics and Beyond. Oxford Univ.
Press. https://doi.org/10.1016/0370-1573(92)90090-M
Swanson, B.F., 1981. A Simple Correlation Between Permeabilities and Mercury Capillary
Pressures. J. Pet. Technol. 33, 2498–2504. https://doi.org/10.2118/8234-PA
Thomeer, J.H., 1983. Air Permeability as a Function of Three Pore-Network Parameters. J. Pet.
Technol. 35. https://doi.org/10.2118/10922-PA
Thomeer, J.H.M., 1960. Introduction of a Pore Geometrical Factor Defined by the Capillary
Pressure Curve. J. Pet. Technol. 12, 73–77. https://doi.org/10.2118/1324-G
Torquato, S., 2002. Random Heterogeneous Materials, Applied Mechanics Reviews.
https://doi.org/10.1007/978-1-4757-6355-3
Torquato, S., 1991. Random Heterogeneous Media: Microstructure and Improved Bounds on
Effective Properties. Appl. Mech. Rev. 44, 37. https://doi.org/10.1115/1.3119494
23
10. APPENDIX A: DESCRIPTION OF GRAIN PACKS AND ROCKS
Each grain pack is composed of nearly spherical grains of glass. We refer to these packs as
GP1, GP2, GP3, and GP4. Listed in increasing order of sphere diameter. For reference solution
of grain pack permeability we use Thomeer’s permeability relation (Thomeer, 1983). Other
reference solutions for sphere packs lead to similar permeability values (e.g., Hinch, 1977;
The five outcrop rocks selected for this study span a range of grain sizes from 0.05 to 0.5 mm
and range in porosity from 0.05 to 0.25 (Table 1). These samples were selected because they
cover a range of compositions and textures that may be encountered. Rocks B1 and B5 are from
the Berea formation (Churcher et al., 1991) which is a sub-angular to sub-rounded Mississippian
age sandstone. Rock F is from the Fontainebleau formation which is a sub-rounded to rounded
Oligocene age sandstone. This sample was well cemented with low porosity. Rocks G1 and G2
are from the Castlegate formation of Utah which is a sub-angular to sub-rounded Mesozoic
sandstone. XRD analyses indicates that the Castlegate sandstone sample consists mainly of
The reservoir rocks samples (A, S, J, C, and D) come from various rock formations around
the world and are of different geologic origin and ages (Table 1). The texture and microstructure
of these reservoir rocks is considerably more complicated when compared to grain packs or the
outcrop rock samples. We also include three samples from Chhatre et al. (2018) include an
Austin chalk sample (AC), Indiana limestone sample (IL), and Scioto sandstone sample (T). We
did not have any images for these samples, but used the reported image porosity, laboratory
measured porosity, and MICP simulation results reported in Chhatre et al. (2018).
24
11. APPENDIX B: THE LBM SOLVER AND BOUNDARY CONDITIONS
method (LBM). The LBM implementation is based on the D3Q19 approach (Ladd, 1994; Ladd
and Verberg, 2001) and the Multiple-Relaxation-Time (MRT) technique that capture complex
flow physics (Lallemand et al., 2003; Lallemand and Luo, 2000). The MRT method is described
in detail in D’Humières et al. (2002) and Premnath and Abraham (2007). All flow computations
were performed using the periodic boundary conditions for the main flow direction and the
directions normal to the flow. The computations were run on a large Linux-based CPU-GPU
HPC cluster.
25
12. APPENDIX C: IMPACT OF IMAGE VOXEL SIZE ON POROSITY
We derive the relation between image voxel size and image-derived porosity using the
concept of capillary pressure (Saxena et al., 2019). Mercury Injection Capillary Pressure (MICP)
curves, that are routinely measured in laboratory, describe the relation between the pressure
applied on the mercury to enter pore throats (to overcome capillary pressure) and the fractional
bulk volume (i.e., volume of invaded mercury divided by rock volume) that is occupied by
mercury at that pressure. Various empirical models (Brooks and Corey, 1964; Leverett, 1941;
Thomeer, 1960) can be used to fit the measured MICP data. These models can be expressed in
−𝐺𝑀
𝑃
𝑙𝑜𝑔10 ( )
𝜙𝑃 = 𝜙∞ 𝑒 𝑃𝑑
. (C-1)
In equation C-1, 𝜙𝑃 is the fractional bulk volume occupied by mercury at pressure 𝑃, and 𝜙∞ is
the fractional bulk volume occupied at infinite pressure. Parameter 𝑃𝑑 is the pressure required to
enter the dominant pore throats and is also sometimes referred as the “entry pressure”. Parameter
𝐺𝑀 describes the shape of the transition zone in a MICP curve, thus providing a measure of rock
complexity and distribution of pore throat size. Parameters 𝑃𝑑 and 𝐺𝑀 can be directly inferred
from a laboratory-measured MICP curve (Swanson, 1981; Thomeer, 1983, 1960). Alternatively,
these parameters can also be inferred using image-based simulations (Hilpert and Miller, 2001).
MICP parameter 𝜙∞ is usually slightly smaller than the total rock porosity that is measured using
helium gas. This is because helium can invade secondary pores which remain inaccessible by
mercury (highest pressure achieved in a typical laboratory is around 100,000 psi). Pore throat
size (𝐷) penetrated at a given pressure (𝑃) is given by the Young-Laplace equation (𝑃
26
= 4σcos 𝜃/𝐷, where σ is mercury-air surface tension ≈ 480 [dyne/cm] and 𝜃 is the contact angle
Using equation C-1 we now derive the relation between image voxel size and image-derived
porosity. The highest pressure that can be achieved when simulating the MICP curve using an
Substituting equation C-2 in equation C-1, and rewriting equation C-1 in terms of the “entry” or
dominant pore throat size (𝐷𝑑 = 4σcos 𝜃/𝑃𝑑 ) we get a relation between 𝜙∞ and the expected
image porosity 𝜙𝑀
−𝐺𝑀
𝐷
𝑙𝑜𝑔10 ( 𝑑 )
𝜙𝑀 = 𝜙∞ 𝑒 ∆𝑥 . (C-3)
Here we refer to a quantity 𝑥 with subscript “𝑀” we are referring to the ground truth of the
quantity 𝑥 with no associated errors due to image acquisition or image processing. We refer to
image-derived properties with subscript “𝐼”. Ideally, 𝑥𝑀 and 𝑥𝐼 should be equal if correct image
processing/segmentation is followed. This is seldom the case and hence the distinction. We now
𝐷𝑑 4σcos 𝜃
𝑁𝑀 = = . (C-4)
∆𝑥 ∆𝑥𝑃𝑑
𝑁𝑀 corresponds to the number of voxels available to resolve the dominant pore throat of size 𝐷𝑑 .
For the remainder of the paper, we will refer to 𝑁𝑀 as the pore-throat resolution parameter as it
depends on the image resolution, as well as the size and the shape of the dominant pore throat.
−𝐺𝑀
𝜙𝑀 = 𝜙∞ 𝑒 𝑙𝑜𝑔10(𝑁𝑀) , (C-5)
27
Transforms in equation C-5 apply if one can infer the curvature parameter 𝐺𝑀 using the
simulated MICP curve. This should be possible if the pressure in a MICP curve at the inflection
Symbol Description
𝝓𝑰 Image porosity
𝝓𝑻 Total porosity
𝒌𝑰 Image permeability
28
𝝓𝑷 Fractional bulk volume occupied by mercury at pressure 𝑃
𝛔 Surface tension
𝜽 Contact angle
𝑫𝒅 Pore throat size ((dominant pore throat size) corresponding to entry pressure that
𝑵𝑰 Ratio of dominant pore throat of size and image voxel size inferred from an image
29
14. FIGURES
Figure 1: A typical digital rock physics (DRP) workflow with various steps as shown in legend
(Figure modified from Saxena et al., 2017a).
30
Figure 2: Rock image segmented into pores and non-porous components using the FCM (Chuang
et al., 2006) method. Image computed and laboratory measurements are also shown for
comparison.
31
Figure 3: Laboratory-measured porosity (total porosity, on x axis) versus image-computed
porosity (on y axis). Rock images were segmented using the FCM (Chuang et al., 2006)
method. Solid line is x = y.
32
Figure 4: Laboratory-measured brine permeability (on x axis) versus image-computed
permeability (on y axis) for various sandstones. Solid line is x = y.
33
Figure 5: Laboratory-measured permeability (on y axis) versus porosity (on x axis) for
sandstones (black symbols corresponding to various rocks). Also, shown are reference
solutions for porosity and permeability for grain packs (GP symbols in black). All images
were segmented using the FCM (Chuang et al., 2006) method.
34
Figure 6: (a) Laboratory-measured (black) and image-computed (blue) MICP curves for a
sandstone sample. Infered parameters used for the transforms are shown.(b) Same results as in
(a) plotted in saturation domain.
35
Figure 7: Variation of 𝑁𝐼 versus 𝑁𝑀 ; also shown is the emprical relation that can be used to
estimate 𝑁𝑀 from 𝑁𝐼 . Symbols indicate various rocks.
36
Figure 8: Image-computed permeability (on y axis) and porosity (on x axis) using images
segmented with various segmentation algorithms (blue circles). Results shown for different
sandstones and grain packs. Segmentation model results shown in red curves.
37
Figure 9: Capillary pressure versus fractional bulk volume occupied by mercury. Plot compares
the results for high, medium, and low permeability rocks. The percent contribution of outer
pore bodies increase with decreasing rock permeability.
38
Figure 10: Parameter p versus image derived attributes.
39
Figure 11: Laboratory-measured porosity (total porosity, on x axis) versus resolution corrected
image porosity (on y axis). Solid line is x = y. The errorbars indicate the impact of
uncertainity in estimating the parameters requred to calculate transforms from images.
40
Figure 12: Laboratory-measured permeability (on x axis) versus resolution corrected image
permeability (on y axis). Solid line is x = y. The errorbars indicate the impact of uncertainity
in estimating the parameters requred to calculate transforms from images.
41
Figure 13: Laboratory-measured permeability (on x axis) versus image permeability that is
corrected for resolution and field of view (on y axis). Solid line is x = y. The errorbars indicate
the impact of uncertainity in estimating the parameters requred to calculate transforms from
images.
42
Figure 14: Laboratory-measured permeability (on x axis) versus image permeability that is
corrected for resolution, segmentation, and field of view (on y axis). Solid line is x = y. The
errorbars indicate the impact of uncertainity in estimating the parameters requred to calculate
transforms from images.
43
Figure 15: Laboratory-measured permeability (on y axis) versus porosity (on x axis) for
sandstones (black symbols corresponding to various rocks). Also, shown are reference
solutions for porosity and permeability for grain packs (GP symbols in black). Image-
computed porosity and permeability, corrected for resolution and field of view, are shown in
blue symbols. The errorbars indicate the impact of uncertainity in estimating the parameters
requred to calculate transforms from images.
44
Figure 16: Laboratory-measured permeability (on y axis) versus porosity (on x axis) for
sandstones (black symbols corresponding to various rocks). Also, shown are reference
solutions for porosity and permeability for grain packs (GP symbols in black). Image-
computed permeability, corrected for resolution, segmentation, and field of view, are shown
in blue symbols. The errorbars indicate the impact of uncertainity in estimating the parameters
requred to calculate transforms from images.
45
Figure 17: Example of using transforms to guide the identification of optimal imaging parameters
for estimating permeabilty from micro-CT images for rocks of different properties as shown
in legend. Note the larger vertical scale for plot d.
46
Figure 18: Example of using transforms to guide the identification of optimal imaging parameters
for estimating permeabilty from micro-CT images for rocks of different properties as shown
in legend.
47
TABLES
48
IL Carbonate - 1 0.2 4.4 14.2 28.7 0.13 0.18 0.17 24.9 7.7 6.6 5.9
SS Carbonate - 1 0.15 4 5.3 42.9 0.15 0.2 0.18 8.1 4.9 3.2 1.4
Table 1: Laboratory-measured as well as image-derived properties for various grain packs and
reservoir rocks. All physical lengths are in microns, porosity values are in fraction, and
permeability values are in mDarcy.
49