1

Proceedings of the 6
th
Nordic Symposium on Petrophysics, 15-16 May 2001, Trondheim, Norway
Nordic Energy Research Programme Norwegian U. of Science and Technology
Estimating the Directional Permeability of Reservoir Sandstones Using
Image Analysis
P.A. Lock, X.D. Jing, and R.W. Zimmerman
Department of Earth Science and Engineering, Imperial College, London SW7 2BP, UK, [email protected],
[email protected], [email protected]
Introduction
The goal of this project is to develop a method for predicting the permeability of a rock
sample using only a small number of SEM images, without any computationally intensive or
time-consuming procedures. The pore structure is idealised as consisting of a cubic network
of pore tubes, with the tubes having an arbitrary distribution of cross-sectional areas and
shapes. The hydraulic radius approximation is used to compute the individual conductivities
of the pores, based on their observed areas and perimeters. The effective medium theory of
Kirkpatrick (1973) is then used to estimate the effective pore conductivity.
The methodology has been applied to several reservoir sandstones from the UK
continental shelf, having permeabilities in the range of 10-600 mD. It is found that for those
samples exhibiting obvious anisotropy, the method yields accurate estimates of the
permeabilities, as well as a qualitative indication of anisotropy.
Calculational Procedure
The first step in the analysis is the estimation of the apparent area and perimeter of each
individual pore that can be recognized in the image. When using the intrinsically isotropic
Kirkpatrick equation, data only from that face of the cubic sample that is normal to the flow
is used as input. The application of a stereological correction factor yields the actual area and
perimeter values, from which the pore conductivity can be determined. A hydraulic
constriction factor is then used to account for variations in pore radius along the pore length.
The resulting set of conductances is truncated to eliminate the contribution of non-conducting
features, and is then input into Kirkpatrick's EMT to produce an effective pore conductance,
G
eff
. The permeability is then calculated, after using another stereological correction to
convert the apparent number density of pores in the field of the image to the actual density
perpendicular to the flow direction. The complete procedure can be summarized as follows:
1. Take BSEI photographs of polished sections
2. Digitize pore images
3. Apply gray-level thresholding procedure to identify pores
4. Compute perimeter and area of each pore with image analyzer
5. Apply stereological correction and hydraulic constriction factor to estimate the
hydraulic conductivity G
i
of individual pores
6. Employ areal thresholding procedure to truncate data set
7. Use Kirkpatricks equation to obtain G
eff
8. Compute the areal density of pores inside the image
9. Assuming a cubic lattice, calculate k
Lock et al., Estimating the Directional Permeability of Reservoir Sandstones
Proceedings of the 6
th
Nordic Symposium on Petrophysics
15-16 May 2001, NTNU, Trondheim, Norway
www.ipt.ntnu.no/nordic
2
Results
The predictions were within a factor of two of the measured values in 16 of the 18 cases.
There seems to be a slight trend of overprediction in the low permeability range, and
underprediction in the high permeability range, although this bias may be due to the small
number of samples. The average percentage error is 23%, or 47% if absolute values of the
error are considered.
The method, although based on an inherently isotropic model, gives qualitative
indications of anisotropy, provided that the measured anisotropy is sufficiently substantial to
exceed the error bars of the measurements. If we assume that each permeability
measurement has an error of about 30%, two orthogonal permeability measurements should
differ by at least 60% in order for the core to be considered unambiguously anisotropic. By
this criterion, the only unambiguously anisotropic core is 4, which has a measured anisotropy
of about {6:5:2}, and a predicted anisotropy of {5:3:2}. In this case, not only were each of
the three principal permeabilities predicted to 25%, but the ordering of the permeabilities was
also predicted quite accurately.
Table 1 lists the predicted permeabilities of the eighteen samples, compared with the
values measured in the laboratory. The notation gives a core identifier and thin section
number, where the X,Y,Z labeling refers to the direction that lies perpendicular to the slice
of the thin section.
It was thought that the use of an inherently anisotropic model such as that of Bernasconi
(1974) would lead to an improvement in the predictions. When using the anisotropic
Bernasconi formulation, data from three orthogonal faces are used, and the three
permeabilities are computed in a coupled manner. Our results, however, showed that the
permeability estimations of the Bernasconi model were only slightly different from that of the
Kirkpatrick equation; moreover, they did not lead to systematic improvements of the
predictions.
Table 1. Compilation of results for the eighteen UKCS cores.
CORE G
eff
(m
4
)
No. of
pores
No. of
images
Total area
(m
2
)
k
pred
(mD)
k
measured
(mD)
Error
(%)
1X 1.14 x 10
-21
3495 2 1.52 x 10
-5
29 18 59
1Y 2.42 x 10
-21
7397 5 3.80 x 10
-5
52 18
191
2X 41.2 x 10
-21
1843 5 3.80 x 10
-5
222 137 62
2Y 34.4 x 10
-21
1244 3 2.28 x 10
-5
209 119 76
2Z 57.3 x 10
-21
1579 5 3.80 x 10
-5
265 109
143
3X 29.3 x 10
-21
2310 5 3.80 x 10
-5
198 202 -2
3Y 22.6 x 10
-21
2652 5 3.80 x 10
-5
175 196 -11
3Z 23.5 x 10
-21
2600 5 3.80 x 10
-5
179 222
-19
4X 9.2 x 10
-21
3229 5 3.80 x 10
-5
87 83 4
4Y 5.3 x 10
-21
2876 4 3.04 x 10
-5
56 72 -23
4Z 2.67 x 10
-21
1785 2 1.52 x 10
-5
35 28
25
5X 39.4 x 10
-21
1656 4 3.04 x 10
-5
239 470 -49
5Z 51.3 x 10
-21
1829 5 3.80 x 10
-5
275 500 -45
6X 26.8 x 10
-21
2394 5 3.80 x 10
-5
188 275
-32
6Z 27.8 x 10
-21
1854 4 3.04 x 10
-5
189 228 -17
7X 9.09 x 10
-21
3605 5 3.80 x 10
-5
96 109 -12
7Y 10.6 x 10
-21
2448 4 3.04 x 10
-5
95 109
-12
7Z 14.4 x 10
-21
2179 4 3.04 x 10
-5
115 69 66
Lock et al., Estimating the Directional Permeability of Reservoir Sandstones
Proceedings of the 6
th
Nordic Symposium on Petrophysics
15-16 May 2001, NTNU, Trondheim, Norway
www.ipt.ntnu.no/nordic
3
10
100
1000
10 100 1000
measured permeability mD
p
r
e
d
i
c
t
e
d

p
e
r
m
e
a
b
i
l
i
t
y

m
DBERNASCONI
NETSIM
KIRKPATRICK
Fig. 1. A comparison of the permeability predictions using Kirkpatrick's isotropic EMT, Bernasconi's
anisotropic EMT, and exact network calculations using NETSIM.
The use of the Kirkpatrick equation to estimate the effective conductivity was tested
against the network simulation code NETSIM (Jing, 1990), which performs an essentially
exact calculation by solving the flow equations in the entire network of tubes. This test was
thought to be necessary, because previous studies of the validity of the Kirkpatrick equation
have all been based on idealized conductivity distributions. Reassuringly, we found that the
effective medium predictions agreed to within a few percent with the exact network
calculations.
A cross-plot of the predicted permeability against the laboratory measurements is shown
in Fig. 1, with lines that indicate errors of a factor of two in either direction.
Discussion and Conclusions
We have developed a network model that allows predictions of the hydraulic permeability of
consolidated sedimentary rocks such as sandstones, based on image analysis of polished
sections of a rock core sample. The method sometimes overpredicts, and sometimes
underpredicts the measured permeability, with an (absolute value) error that is on average
only 47%. To put this in perspective, we point out that, for example, McPhee and Arthur
(1994) made a series of gas permeability measurements on a Clashach quarried sandstone
with an average permeability of 693 mD, and found an error of 32%, based on the standard
deviation of all the measurements. Although these values are not directly comparable, the
mean (absolute) relative error in our measurements was 47%, and the median (absolute)
relative error was 23%. So, it seems that the errors in our predictions are comparable to the
error inherent in the laboratory measurements.
Furthermore, the use of more computationally intensive procedures such as the inherently
anisotropic effective medium theory of Bernasconi or essentially exact network calculations
fail to produce any systematic improvement on the predictions obtained using the isotropic
effective medium theory of Kirkpatrick.
It should be emphasized that our model is based entirely on measured attributes of the
pore space, and contains no adjustable parameters. The question may be raised as to why our
estimates are so much more accurate than those made by Koplik et al. (1984), using a broadly
similar procedure. The main difference seems to be our inclusion of the constriction factor
Lock et al., Estimating the Directional Permeability of Reservoir Sandstones
Proceedings of the 6
th
Nordic Symposium on Petrophysics
15-16 May 2001, NTNU, Trondheim, Norway
www.ipt.ntnu.no/nordic
4
and the stereological correction factors for pore size and number density, each of which lower
the estimated pore conductance by about a factor of 2. Taken together, the absence of these
corrections may explain why their estimates were too high by about an order of magnitude.
Acknowledgement
This project was funded by Enterprise Oil, plc.
References
Bernasconi, J., Conduction in anisotropic disordered systems: effective medium theory, Phys. Rev. B, 9, 4575-
79, 1974.
Jing, X.D., The Effect of Clay, Pressure and Temperature on the Electrical and Hydraulic Properties of Real
and Synthetic Rocks, Ph.D. Thesis, Imperial College, London, 1990.
Kirkpatrick, S., Percolation and conduction, Rev. Mod. Phys., 45, 574-88, 1973.
Koplik, J., Lin, C., and Vermette, M., Conductivity and permeability from microgeometry, J. Appl. Phys., 56,
3127-31, 1984.
McPhee, C.A., and Arthur, K.G., Relative permeability measurements: an inter-laboratory comparison, SPE
Paper No. 28826, Eur. Petrol. Conf., London, 1994.