Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

An Application of Geostatistics

and Fractal Geometry for


Reservoir Characterization
Yngve Aasum, SPE, and Mohan G. Kelkar, U. of Tulsa, and Surendra P. Gupta, SPE,
Amoco Production Co.

Summary. This study presents an application of geostatistics and fractal geometry concepts for 2D characterization of rock proper-
ties (k and cf» in a dolomitic, layered-cake reservoir. The results indicate that lack of closely spaced data yield effectively random distri-
butions of properties. Further, incorporation of geology reduces uncertainties in fractal interpolation of wellbore properties.

Introduction
Previous work l ,2 showed that miscible flood performance is af- logical data to be pooled. These data are classified as either hard
fected by the degree and scale of small-scale heterogeneities, in- or soft. More significance is assigned to the hard data (more cer-
formation which, unfortunately, is not obtained in conventional tain data, such as core data) than to the soft data (less certain data,
reservoir-characterization processes. 3 Two methods-geostatistics such as geophysical data or qualitative interpretations). The simu-
and fractal geometry4-13-may help to obtain such information and lator uses an indicator formalism and probability kriging to esti-
also to estimate associated uncertainties in the reservoir characteri- mate variables. 8
zation. This study evaluates how these two techniques character-
ize a reservoir unit (Unit A pilot area) while honoring the geological Applications of Geostatistics. The length of the permeability scale
information (100 to 200 values each of k and cf> in each of 18 wells contributes to the heterogeneity in a porous medium and signifi-
in Unit A, where distances between wells vary from"'" 85 to 4,000 cantly affects miscible displacements. 1.2 Limited research has been
ft). Fig. 1 shows the Unit A well locations. This study does not conducted to establish the correlation structure of permeability and
extend to using reservoir descriptions 3 to predict and/or to inter- porosity in porous media. Two studies are presented here.
pret any field performance. Da Costa e Silva 6 estimated the spatial structure of permeability
The geostatistical technique was used areally within individual and porosity in a North Sea reservoir (26,247 x65,617 ft) using
layers. More specifically, the correlation structure was obtained data from 30 wells. k and cf> data were obtained from well tests and
for individual layers based on data from the 18 wells in Unit A. were assumed to represent the reservoir's producing zones. Per-
Then, the constraints of the spatial correlation structure and the avail- meability and porosity were found to have similar spatial structures,
able k and cf> data were used with a conditional simulation tech- with anisotropy and scale lengths near 13,000 ft. In some direc-
nique to create several possible reservoir descriptions for the tions it was indicated that structures may exist on a scale smaller
individual layers in the central five-spot pilot. The injector-to- than the smallest lag length or shortest distance between the wells.
Kriging results from the reservoir gave a smooth or "flat" reser-
injector distance is 288 ft for the five-spot pilot. Adverse mobility
voir description.
ratio miscible displacements (M = 5) were simulated for the gener-
Goggin et al. 4,5 conducted outcrop studies of Eolian sandstone
ated set of reservoir descriptions.
deposits near Page, AZ, to obtain a detailed description of perme-
The fractal-geometry approach was applied across a vertical cross
ability heterogeneity at various scales. About 2,000 permeability
section between Wells 1 and 5. Available k and cf> data were used
measurements were taken with a calibrated minipermeameter on
to create a fractal cross section between the two wells that still hon- concentric grid patterns, and statistical analysis was done to infer
ored the geologic layers from the k/cf> plots. Single-phase first- the scale dependence of permeability variation. The spatial corre-
contact-miscible displacements (M=5) for N Pe :50.01 were simu- lation structure of permeability was obtained by variogram analy-
lated across the fractal cross section. t
sis of the measured data. These studies revealed that the
heterogeneity scales closely relate to the geologic scales. The spa-
Background tial correlation structure of permeability identified by the scale
Geostatistics. Geostatistics is a branch of applied statistics that han- length, anisotropy ratio, and the inclination angle of measurement
dles spatially distributed variables. 14 ,15 Recent interest in the was found to be scale-dependent in Eolian systems.
petroleum industry has been to use geostatistics concepts to charac-
terize and to evaluate petroleum reservoirs.4-7 Basic geostatistical Fractal Geometry. Mandelbrot 20 ,21 developed the concept of
concepts can be found in several publications. 16-19 Ordinary krig- fractal geometry, which can be used to describe irregular and frag-
ing, a geostatistical method developed by Matheron,15 is often mented patterns in nature. Fractals are sets of continuous and non-
used as a mapping technique in which the regionalized variable is differentiable shapes with a geometric dimension that strictly exceeds
estimated from an assessment of spatial continuity. The uncertainty the topological (Euclidean) dimension. An important feature of frac-
of any estimated value can be determined. Kriging, however, re- tals is their resemblance at various scales, which means that the
sults in a "smooth" realization of the regionalized variable. 8 characteristics of the fractals can be identified at all scales. Both
We used conditional simulation in this study because it gives a deterministic and random fractals exist, with random fractals being
better visual idea of the uncertainties involved in the estimations. more useful for the characterization of the irregularities of natural
Spatial conditional simulation is a Monte Carlo simulation tech- lines and surfaces. The characteristics of random fractals, however,
nique that honors the available data, their location, and the varia- are statistical when subject to scaling. Various statistical models
ble's spatial structure. Several realizations (sample outputs with exist to describe natural phenomena. One ofthe most useful math-
ematical models available for describing natural random fractals
different seed values) will indicate the uncertainties involved. These
(mountainous terrain, clouds, etc.) is the fractional Brownian mo-
realizations, or reservoir descriptions, can be used in reservoir flow
tion, Ism' 22,23 which is an extension of Brownian motion. Sever-
simulators to provide a range of possible outcomes. Alabert 8 de-
al algorithms exist for creating traces of fractional Brownian
veloped a conditional simulator that allows information from such motion. 22
different sources as core data, well logs, and geophysical and geo-
Applications of Fractal Geometry. Hewett 9 laid the groundwork
Copyright 1991 Society of Petroleum Engineers for applying fractal geometry concepts to reservoir descriptions and

SPE Formation Evaluation, March 1991 11


11 12 '3 .30
o o o
0 0 0
.25
.. • •
0
o •
~ e lII!

7
o 8
o .20
1

14
265
000
o
4
0
----
.c
?-
.15
15
o o o
3 .10 Legend
9 10
o o o ergodic values
.05 • nonergodic values
-model

.OO+----r----r---,---~----r---~
16 17 '8 O. 200. 400. 600. 800. 1000. 1200.
o o o
h (Lag) - ft
Fig. 1-Unit A model study area. Fig. 2-Variograms for In k, Layer 1.

evaluation. The motivation for his work was the observation that thickness. Layer 6 is among the fastest layers of the producing zone;
the structure of the spatial correlations in the permeability distri- i.e., a layer with a high klct> ratio ("" 200 md), whereas a "slow"
bution largely affects fluid flow in heterogeneous porous media. layer has a low klct> ratio ("" 5 md). The average thickness for Lay-
Hewett claimed that property distributions in sedimentary environ- er 6 is 6 ft. Both Layers 1 and 6 are identified in all 18 wells, sug-
ments show a fractal character with long-range correlations. With gesting geological continuity. The spatial structure and the scale
the assumption that a smoothed trace of vertical wellbore data lengths for permeability and porosity are estimated with data from
(porosity) could approximate the correlating structure of porosity all 18 wells in Unit A, and this information is honored when vari-
in the horizontal direction, Hewett outlined an approach to assess ous equally likely reservoir descriptions are generated by condi-
the effect of fractal property distributions on reservoir performance. tional simulation.
Hewett and Behrens lO advocated two uses of conditional simu- The area to be simulated is the main five-spot (shown in Fig.
lation. One use was to scale flow processes and properties. They 1), which includes six wells (Wells I through 6). The conditional
reviewed different scales within the reservoir and indicated that frac- simulations are performed separately for porosity and permeabil-
tal distributions have scaling properties. Different numbers oflay- ity distributions. The conditional simulator used in this study was
ers were used in the simulation to observe the effect scaling had developed by Alabert 8 and uses the concept of indicator
on assigned reservoir fluid and rock properties. The second use kriging 25 ,26 to perform Monte Carlo simulations. For more details
was to estimate uncertainties of outcomes owing to incomplete data about the simulation procedure, refer to Alabert. 8 Details of our
sampling. Flow simulations in various possible reservoir descrip- study can be found in Ref. 27.
tions (realizations of property distributions) generate a probability The conditional simulator can accommodate soft data (qualita-
distribution of reservoir performance that may be used to evaluate tive information), but the complete data set in this investigation was
the risk associated with a project. assumed to consist of hard data (known with 100% reliability). In
Other authors presented methods for assessing reservoir perform- this study, the structures are assumed to be isotropic because the
ance that assumed fractal distributions of k and ct>. These methods areal data are scarce.
were validated by examples from several EOR projects. l1 - 13 The various numbers of gridblocks, nx Xny, used (12 x 12,
These authors concluded that the methodology can be used to predict 24 x 24, and 36 x 36) were transformed into 25 x 25, 49 x 49, and
field-wide project performance. 73 x73 gridpoints, respectively, to satisfy a second-degree poly-
nomial approximation in the tracer model. 24 We assumed that
Approach and Results nz=l. ,;lx, Lly, and.lz were adjusted according to the values of
The goal of this study is to characterize the Unit A model study n x ' ny, and n z ' respectively.
area with geostatistics and fractal-geometry concepts. The objec- Unfortunately, very few hard data were available for this study.
tive is to characterize two layers of the reservoir areally with spa- Only six data points were used as constraints in terms of values
tial conditional simulation and a vertical cross section between two and locations of the data. Each of the six data points represents
wells with fractal geometry. Unit A (Fig. 1) is chosen for this study the average layer value of Wells 1 through 6.
because of data availability at various spacings. The area is known We analyzed the results ofthe conditional simulations by inves-
to have a dolomitic layer-cake structure that, in general, extends tigating the effect of realizations and the number of gridblocks used
from well to well. Impermeable thin layers of anhydrite separate to simulate miscible displacement in the five-spot pattern. Effluent
the geological layers. Existing information about the klct> ratio within concentration profiles and fractional recoveries were investigated
the 18 area wells identifies the nature of the layered system. Per- in the analysis.
meability and porosity values exist for each foot in the producing
zone for each of the 18 wells (100 to 200 values for each well), Geostatistics/Conditional Simulation. Areal Correlation Struc-
and these values are used in the characterization procedures. Flow ture. We transformed permeability, the natural logarithm of per-
simulations were performed with a finite-element tracer model. 2.24 meability, and porosity to indicator variables (D's and 1's) using
Results are essentially free of numerical dispersion and grid- four cutoffs (five classes). All data values within each layer (Lay-
orientation effects. A second-degree polynomial option was chos- ers 1 and 6) were used for each of the 18 wells. Indicator vario-
en for this study. grams were then constructed to represent each cutoff for each
transformed variable. Both ergodic and nonergodic variograms 28
Application of Geostatistics. Two layers are characterized by con- were used. The nonergodic variogram is less restrictive than the
ditional simulation. Layer 1 is a tight zone with a 25-ft average ergodic variogram because it does not implicitly assume that the

12 SPE Formation Evaluation, March 1991


.4- o
>
c: 0.8
W
.3
..
o >
0

-- A
~
,,-----------
()
W
c: 0.6

.. o.
.c
c-
.2 /
,"...... ...J
«
z
0 0.4 Legend
I o i=
() o 25 X 25 GRID POINTS, SEED • 1
/ «
/
Legend c:
U. 0.2
A 25 X 25 GRID POINTS. SEED • 2

+ 49 X 49 GRID POINTS, seED • 1

.1 - I 0 ergodic values X 73 X 73 GRID POINTS. seED - 1

,fJ .. nonergodic values


" ---model (ergodic) 0.5 1.5 2 2.5 3 3.5 4 4.5
" - model (nonergodic) PORE VOLUMES INJECTED
.0 +----...---,_,r-----.---..,-----t
O. 400. 800. 1200. 1600. 2000.
h (Lag) - ft Fig. 4-Effect of number of gridblocks and realization (seed
value) on recovery for Layer 6 (ae=5.00 ft, N pet =0.17, and
Fig. 3-Variograms for In k, Layer 6. M=5).

lag mean of the variable is stationary.28 In other words, the er- Variance restrictions in the indicator formalism dictate the max-
godic variogram assumes that the expected mean of the variable imum level (sill plus nugget) of the model (this value is 0.23 in
does not change with location. The nonergodic variogram allows Fig. 2). While using the ergodic variogram, we observed that the
for variations in the expected mean of the variable at different lo- maximum level was too low compared with the experimental vari-
cations. ogram values. This problem was eliminated when we used the
Semivariograms indicated either the presence of nugget effects 19 nonergodic variograms because the sill of the variogram coincided
or possible correlation structures of 1,000 to 1,500 ft. A pure nug- with the data variance for these variograms (Fig. 2).
get is present when no identifiable correlation structures are visi- Fig. 3 shows ergodic and nonergodic variograms that express
ble in the variogram. Structures may exist, however, at spacings the structural nature of the log of permeability in Layer 6,- which
smaller than the distance between the wells. Fig. 2 shows areal er- is one of the faster layers. Considerable scatter exists in the data
godic and nonergodic variograms of log of permeability for one for the ergodic version, but a general trend of a correlated struc-
of the slowest layers (Layer 1) in Unit A. Fig. 2 indicates a pure ture can be observed. The correlation length is approximately 1,200
nugget effect on both the ergodic and nonergodic variograms. Thus, ft when fitting the variogram values with a spherical model (Fig.
no spatial correlation was observed from the available log of per- 3), but the nonergodic variogram shows that the data are essential-
meability data for Layer 1. Pure nugget effects also were observed ly uncorrelated (a pure nugget effect was observed). Although not
in the permeability and porosity variograms for Layer 1. Structures shown in Fig. 3, the nonergodic variogram does not indicate any
may be present on a smaller scale, a distance smaller than the spatial correlation for permeability or porosity.
smallest average lag ( "" 200 ft) in the semivariogram, but these struc- The difference between the ergodic and nonergodic variogram
tures could not be captured because the data were limited at spac- results illustrates the importance of carefully examining the data
ings smaller than the distance between the wells. The permeabilities set. Tables 1 and 2 list the average porosity and permeability values
and the porosities are effectively random in Layer I because of the for the 18 wells for Layers I and 6, respectively. Note that the first
horizontal character of the variograms. few points in the variograms correspond to data at the inner five-

TABLE 1-AVERAGE POROSITY AND PERMEABiLITY TABLE 2-AVERAGE POROSITY AND PERMEABILITY
VALUES FOR LAYER 1 OF 18 WELLS IN UNIT A VALUES FOR LAYER 6 OF 18 WELLS IN UNIT A

~ k ~ k
Well (%) (md) Well (%) (md)
1 6.04 0.38 1 16.73 17.94
2 6.73 0.34 2 17.82 17.18
3 6.31 0,26 3 18.67 26.43
4 7.06 0.18 4 17.31 13.74
5 6.50 0.25 5 12.50 16.33
6 6.83 0.40 6 17.84 22.90
7 6.33 0.34 7 12.50 2.21
8 4:67 0.17 8 15.01 8.18
9 5.84 0.28 9 12.85 2.36
10 4.16 0.31 10 15.44 10.33
11 6.25 0.31 11 15.06 8.56
12 6.09 0.32 12 13.49 4.42
13 6.71 0.39 13 14.22 5.23
14 5.27 0.30 14 13.66 4.40
15 3.96 0.18 15 11.52 0.96
16 6.12 0.25 16 14.06 10.93
17 5.17 0.26 17 10.31 0.61
18 6.32 0.33 18 13.00 3.30

SPE Formation Evaluation, March 1991 13


k/q,
20 40 60 80 100 120 140 160 180 20o

- f-

UnitA, Well 1 =I
=
-
l-
f-
f-
1

2A
2B
2C

3
4A

i .....
2E
fA
10 100
8 .... -
- h(Lag)-ft

Fig. 5-Well 1. klq, plot. Fig. 6-Well 1. RIS plot of porosity.

spot region, composed of Wells 1 through 6, which have inter- data can sometimes reveal more complex structures, causing greater
distances comparable with the first few lags. Table 1 shows that uncertainty in performance predictions. No small-scale structures
the average values do not vary much for Layer 1. Note that Layer exist in any of the possible reservoir descriptions of Layers 1 and
1 is very tight and may not have any practical oil-recovery value, 6 other than those caused by the size of individual gridblocks, which,
but Layer 1 is included for comparison with Layer 6. The average in effect, act as the scale length of the system in the absence of
values in Table 2, however, generally are significantly larger for information about small-scale structures (Fig. 4).
the inner five-spot wells (Wells 1 through 6) than for the other wells As expected, results depend on the specific situation. The results
in the area. The difference in the ergodic and the nonergodic vari- in this study, however, show that without any information about
ograms for Layer 6 appears to result from the trend of higher mean correlation structures on scales smaller than the system dimensions,
values in the five-spot area, which is corrected in the nonergodic the system behaves effectively randomly. Fine-grid miscible-
version. The correlation structure observed with the ergodic vari- displacement simulations in such a system are shown to be indepen-
ogram is therefore most likely an artifact of not accounting for the dent of the realization. The indicator kriging technique used in this
differences in the mean at different locations in Unit A. study appears to have more potential if more data (constraints) are
Miscible Displacement in Layers Characterized by Indicator included. Additional quantitative data may be introduced if mul-
Kriging. The reservoir property distributions (k and </» were ob- tipattern rather than single-pattern fields are used. Qualitative data,
tained with indicator kriging, which honors the data at the wells. such as seismic data, may also help to estimate the correlation struc-
Information of the correlation structure for the 18 wells in Unit ture of the data.
A was used to characterize Layers 6 (fast) and 1 (slow) in the cen-
tral five-spot pattern with respect to permeability and porosity. Application of Fractal Geometry. Fractal cross sections were
Several realizations (possible reservoir descriptions) were gener- created between Wells 1 and 5 with respect to both porosity and
ated with various levels of detailed information (different number permeability. Each well has permeability and porosity values avail-
of gridblocks). Adverse mobility ratio (M=5) miscible displace- able on a foot-by-foot basis. Data in Well 1 were aligned with data
ments were simulated for both layers in the five-spot pattern. The in Well 5 to satisfy the geologic layering structure, which can be
effect of the realizations on miscible displacement was negligible. correlated from kl</> plots that represent each of the two wells.
Fig. 4 shows that for 25 x25 gridpoints, fractional recoveries are The fractal cross section was generated by fractional Brownian
quite insensitive to the realization for Layer 6. Breakthroughs occur motion, fBm' 9,22 The intermittency coefficient, H, determines the
at nearly the same amount of PV's injected in Layer 6, regardless structure (degree of roughness) offBm traces. Therefore, it is de-
of the permeability and porosity realizations over which miscible sirable to obtain an H value that represents the data to be modeled
flow is simulated. Although not shown, the shapes of the effluent by the iBm' The H value can be extracted from a given record of
concentration profiles are also practically identical, with almost no data by use of several available methods. 9 In this study, two
irregularities, which is similar to the effect that heterogeneities with methods were used: the rescaled range, RIS, analysis and vario-
zero scale length have on effluent profiles of miscible displacements gram analysis of a normalized graded sequence. 9
for the same mobility ratio. 27 The same trends were observed for The two methods were applied to the entire record of vertical
fractional recoveries and effluent profiles in Layer 1. Several pos- wellbore data of each of the 18 wells in Unit A, with porosity and
sible reasons exist for the lack of sensitivity to the seed value or permeability (In k) data records treated separately.
realization on the miscible-displacement simulations in Layers 1 The correlation structure of the data in each well was determined
and 6. by extraction of H values. Then, assumptions were made about
No small-scale structure was detected in the variogram analysis which value of H to use to generate the fractal cross section. In
of Layer 1 or 6. For both layers, the lack of horizontal data on this investigation, the H value in the horizontal direction is pre-
a small scale resulted in an effectively random property distribu- sumed to equal the H value extracted from numerically integrated
tion on such a scale, and therefore, all responses from these reali- vertical data. This assumption must be validated with core data from
zations appear almost identical. Goggin et at. 4' found that when horizontal wells. This assumption is supported by the theory that
they decreased the measurement scale, structures were revealed in the statistics of spatial distribution of sediments are governed areally
the random data variation. Hewett et af. 10 noted that additional and vertically at any given point in time by the statistics of the proc-

14 SPE Formation Evaluation, March 1991


1000 ~----------------------------------.1 Z

::
o
>-
II: ~
..
0.8 0.81-
W
>

~
100lT
Z
0 W
() ()
W
II:
...J
0.6
* 06
5
()
Legend
..+t:J# « en
;S 100; vV' Z
0 0.4
H = 1.00 (.), H .1.00 (k), SEED 1
0.4~
en
?- y"'> F
()
H.O.83 (.p). H.,O.81 (kJ, SEED 1
z
A"
y
«
H.0.83 (</I), H .. 0.81 (k). SEED 2
o
02~
H .. 0.83 (.). H .. 0.70 (k), SEED 1
II:
10; / LL 0.2
12.:..~!_~~1;...~_~:~~.~~!:_~_~_~~_2
~O.6~~~H"'O~J~~t:P~
W
/:(' :::E
V/ H ,. SLOPEl2 • 0.150
~= ~.~ .f!), _~,.~.8.',,<k!.:... S~.~D.-! 15
0.5 1.5 2 2.5
PORE VOLUMES INJECTED
10 100

h(Lag)·ft Fig. 8-Effect of fractal cross section (geologic layers hon·


ored) on recovery and effluent profiles (ac = 1.00 ft,
Fig. 7-Vertlcal semivariogram of In k, Well 1. N pe , =0.009, and M=5).

esses that form them (weather and climate). 9 From outcrop studies Three different fractal cross sections were generated. Each cross
of Page sandstone, Goggin29 observed that horizontal structures section varies according to how the layers are defined. In general,
are similar to vertical structures, but horizontally the structures are each layer was represented by an arithmetically averaged value of
more "stretched out" or larger. a layered property. Then the average layer value was interpolated
The arithmetic average of all H values obtained from Wells 1 to assign values to the top and bottom of each layer to satisfy tracer
and 5 was used as the base case for the vertical cross section. This model requirements for the quadratic polynomial option (nodes are
vertical fractal cross section between Wells 1 and 5 was geI).erated located at the center and ends of every gridblock 24 ). The fractal
by creating/Bm values between corresponding data points of each interpolation followed.
well. The iBm values were synthesized with a successive random The first fractal cross section was divided into 16 geologic layers
addition procedure that is explained in detail in Refs. 9 and 22. based on available k/cp plots. Thirty-three values were calculated
The resulting fractal cross-sectional reservoir descriptions (k and for the quadratic polynomial option. The second cross section con-
cp) were used to simulate single-phase tirst-contact-miscible displace- sists of even more detailed geologic layering. Each of the 16 lay-
ment for M=5. ers in Fig. 5 are subdivided into two layers of equal thickness,

25
5
..-- ABOVE
0.800
0.700
0.600
0.900
0.900
0.800
0.700

-
0.500 0.600

45
III 0.400 0.500
0.300 0.400
0.200 0 ..300
,.-.. ITillITlTIl 0.100 0.200
65
~ D BELOW 0.100
"'--'
,..c 85

105

125

145
(J lr(J rx(J <o(J <b(J (J(J lr(J rx(J <o(J <b(J
\ \ \ \ \

L Cft)
Fig. 9-Contour map of miscible displacement for fractal cross section (M= 5 and Qf = 0.089).

SPE Fonnation Evaluation, March 1991 15


5
-- ABOVE
0.800
0.900
0.900

..--
0.700 0.800
25 0.600 0700
0.500 0.600
0.400 0.500
45 0.300 0.400
m 0.200 0.300
~
65 EEJ 0.100 0.200
ct:!
"'--"'
D BELOW O. 100

,..0
85

105

125

145
() 't() rx() x:P <0() '\()() '\'t() '\ rx() '\x:P '\<0()
L (ft)
Fig. 10-Contour map of miscible displacement for fractal cross section (M=5 and Q, =0.44).

resulting in 32 layers, such that 65 values are present in the verti- ed sequence of log of permeability in Well 1. The linear fit, which
cal direction. Fractal cross sections that do not conform to the avail- was carried out on the first 50% of the semivariogram points (cor-
able geologic information about the layering structure were also responding to small and medium ranges) resulted in H=0.750. Re-
created. For these cases, layers of equal thickness were created and sults of analyses from the other 17 wells showed acceptable
data within the arbitrarily assigned layers averaged before the fractal straight-line fits.
cross sections were generated. Seventeen layers were created (with
35 nodes) from an almost complete set of data in each well (140 Miscible Displacement in a Fractal Cross Section. Fine-grid simu-
of 145 total). lations of miscible displacement for M = 5 were performed in various
Two aspects of the fractal porosity and permeability distributions fractally interpolated cross sections of permeability and porosity.
were investigated; the effect of the intermittency coefficient, H, The grid nodes were 33 x 65 (33 vertical and 65 horizontal nodes)
and the effect of different realizations of the permeability and porosi- for the case where 16 geologic layers were honored, and 65 x 65,
ty distributions. Effluent concentration profIles and fractional recov- where each of the 16 geologic layers was subdivided into two lay-
ery curves were investigated to analyze the effects of Hand ers. For the case where geologic layers were not honored, 35 x65
realizations on miscible displacement in a fractal vertical cross sec- grid nodes were used.
tion between Wells 1 and 5. Concentration contour maps were valu- The variation of H between 0.70 and 1.0 and different realiza-
able in the analysis of the displacement process. tions had a negligible effect on the relatively fine-grid simulation
of miscible displacement in the vertical fractal cross section be-
Fractal Geometry. Intermittency Coefficient, H. On the basis of tween Wells 1 and 5 when the geologic layers based on kl4J ratios
the variogram analysis of a normalized graded sequence and the were honored. The realization affected the miscible displacement
RI S analysis of the original vertical wellbore data for each of the when the geologic layers were not honored by arbitrarily dividing
18 wells in Unit A, the intermittency coefficient was found to vary the pay zone into layers of equal thickness, but the value of H had
between 0.70 and 0.94 for both porosity and the log ofpermeabil- little effect.
ity. This range of H values corresponds to other reported 9,21-23 Fig. 8 shows recoveries of miscible displacements in a geologi-
H values that describe natural phenomena. Fig. 6 shows the plot cally appropriate vertical fractal cross section between Wells 1 and
of the rescaled range of porosity; the slope has the value of H, which 5. Each seed value represents a different realization. The values
is 0.832. This type of time-series analysis, RIS, is most applicable of H ranging from 0.70 to 0.94 did not affect the recoveries. Simi-
when data are abundant (several thousand). The slope is closer to lar results were observed by Tang et at. 13 They observed negligi-
the actual value of H as lag lengths increase,9 but because the ble effects on fractional flow curves from flow simulations in cross
available data were relatively few «200), as much available in- sections in which H varied between 0.62 and 0.78 in the fractal
formation as possible was used; i.e., all data points were used when interpolations. Our study shows that increasing the intermittency
fitting the straight line. coefficient to H= 1.0 also has no significant effect on the recov-
The logarithm of the permeability values was used instead of the ery. H = 1.0 corresponds to the linear interpolation between Wells
permeability values, but the techniques used to obtain the H value 1 and 5, indicating that for this particular layer-cake reservoir with
were the same as those used to analyze the porosity data. Fig. 7 large kl4J contrasts, the fractal approach gives results similar to those
shows a log-log plot of the semivariogram of the normalized grad- obtained by linear interpolation. Emanuel et al. 11 observed that

16 SPE Fonnation Evaluation, March 1991


2
0
~
a:
>-
a: I- 0.8
0.8 2
W W
> 0
0 2
0 0.6
W 0.6 0
a: 0
....J en
« en Legend
2 w
....J
0.4
0 0.4
2
H"" 1.00 (,,), H = 1.00 (k), SEeD 1
~
0 0 H=O.71 (41). H_O.70 (k), seeD 1
« u; + <.,).
a:
u.. 0.2
Legend 2 0.2 H=O.71 H=O.70 (k). SEeD 2

W H = 0.94 ("). H,: 0.94 (k), seED 1


32 LAYERS
~
~.~~~~J!~~ __~_~~~~j!<l:_~_~_'=~_3
16 LAYERS 0
0.5 1.5 2 2.5
05 1. 2 2.5

PORE VOLUMES INJECTED PORE VOLUMES INJECTED

Fig. 12-Effect of fractal cross section (detailed geologic lay-


Fig. 11-Effect of number of geologic layers on recovery ers honored) on effluent profiles (at=1.00 H, N p8 , =0.009,
(at=1.00 H, N Pe , =0.009, and M=5). and M=5). .

in a highly structured reservoir with a high-permeability channel,


varying H between 0.65 and 0.85 did not change the overall per- -,------------------.,-12o
~
meability distribution significantly. From klc/> plots representing
the injector (Well 1), shown in Fig. 5, and the producer (Well 5), >-
a: 0.8 0.8 I-
w Z
we see that one distinct layer has a much larger klc/> ratio than the > W
0 o
remaining layers, suggesting that fluid flow will concentrate in this 0
high-speed flow channel. Fig. 8 also displays the effluent concen-
W
a: 0.6 06
5o
....J
tration profiles, which indicate that the breakthroughs are practi- «Z en
cally identical and that the effluent concentrations show quite similar Legend en
0 0.4 0.4~
behavior regardless of the value of H or the realization. ~ H=O.71 (fJ,), H",O.70 (k). seeD 1 z
0
« o
The insensitivity of these miscible displacements to a change in a: +
H=O.71 ("). H=O.70 (k), seeD 2
u;
the intermittency coefficient and/or the realization can be explained u.. 0.2 H=O.71 ("). H=O.70 (k). SEED 3 0.2Z
H = 0.71 (.pl, H = 0.70 (k). seED 4
W
as follows. The fractal cross section depends on the values of k ~
~~~:~~J!t_~~_~:~~~~_~~..! is
and c/> at each of the two wells. The cross section of porosity and
0.5 1 1.5 2.5
permeability is made up of a series of/Bm values, with each/Bm
PORE VOLUMES INJECTED
generated starting with two endpoints, one data point from each
well. The shape or structure of the iBm between the endpoints is
Fig. 13-Effect of fractal cross section (geologic layers not
determined by H and the variance of the porosity or permeability honored) on recovery and effluent profiles (at = 1.00 H,
values at each end. The iBm sequence becomes progressively N p8 , =0.009, and M=5).
smoother as H increases in value, and the limiting case is H = 1,
which corresponds to a linear interpolation. The iBm sequence also
becomes smoother if the variance between the two endpoints is A more detailed geologic approach was also investigated. Each
smaller. Fig. 5 shows that one particular layer is dominant with of the 16 distinct layers identified in Wells 1 and 5 were subdivid-
respect to the klc/> ratio. The irregularities created between the end- ed in two equal-thickness layers (see Fig. 5 for Well 1) to create
points when H < 1 therefore will have a suppressed effect on the 32 layers. Fractal cross sections were generated for different H
miscible displacements. Also, when geologic layers are accounted values and various realizations from 65 porosity values and 65 per-
for, similarities of the values observed at each end of the sequence meability values in each well. Miscible displacement was simulat-
result in a smaller variance, further reducing the variability of the ed from Well 1 to Well 5 for the various cross sections (realizations).
interwell values. Different realizations change the order of the ir- Fig. 11 compares the fractional recoveries of the 16 and 32 layers,
regularities somewhat but have no significant effect on the displace- respectively. A negligible difference exists in the recoveries from
ment (see Fig. 8 for recoveries and effluent concentration profiles) the different number of layers. As observed for 16 layers, chang-
because of the overpowering effect of the layered stratifications in ing the realization and the H value for the 32-layer system had very
determining the flow pattern. little effect on the fractional recoveries (not shown). Although lit-
Fig. 5 shows the klc/> ratios for Wellt. Note the particularly large tle difference is observed in breakthrough values, the shapes of the
klc/> ratio for Layer 6 and the large contrasts in the klc/> ratio among effluent profiles are not perfectly identical, particularly for differ-
the various layers. Figs. 9 and 10 show the fine-grid simulation ent realizations (Fig. 12), because as more layers are used, more
of miscible displacement in a fractal cross section between the in- fractal interpolations are obtained, which introduces more detailed
jector (Well 1) and the producer (Well 5) for adverse mobility ra- structure to the cross section. Thus, the combined intermittency
tio (M=5) at various stages (PV injected). Fig. 9 is a concentration coefficient and different realizations result in larger effects than were
contour map ofthe displacement after 0.089 PV was injected. Note noted in the case with fewer layers.
the dominance of one particular layer (corresponding to Layer 6 Fractal cross sections of porosity and permeability were also made
with large klc/> ratio in Fig. 5). Most of the injected fluid flows into without regard to available geologic-layer sequence information.
this particular layer, thereby controlling the channeling. The dis- In these cases, the layers were assigned an equal thickness. Conse-
placement is clearly layer-like (Fig. 10), suggesting that the created quently, some of the large kl c/> contrasts were diminished because
fractal cross section reflects the available geologic information. Ear- layers that originally had a large klc/> were combined with layers
ly breakthrough occurred in the fast layer (Fig. 10) with a lot of that had a smaller klc/> to create a dampened effect of klc/> contrasts
in-situ fluid left behind in other layers. over the complete cross section.

SPE Fonnation Evaluation, March 1991 17


Miscible-displacement simulations in these fractal vertical cross honored and have great contrasts in speed (k/q, ratio) as in Unit
sections reveal quite different results compared with those in ver- A; the scale over which contrast in layers is observed is much greater
tical cross sections where geologic layers were honored. The ef- than the scale on which the fractal variations are defined; and the
fect of realization on the fractional recoveries is evident, which variance between the layer interpolation points is small. Inclusion
Hewett and Behrens 10 also observed for cases where the geologic of geology reduces the uncertainty in the reservoir performance
layering structure is not well defined. At relatively low H values predicted with fractal geometry.
(H=0.71 and H=0.70 for porosity and permeability, respectively),
four different realizations of porosity and permeability distributions Nomenclature
are generated and a quite distinct range of possible recoveries from
b = system width, ft
miscible-displacement simulation is indicated (Fig. 13). Fig. 13
fBm = fractional Brownian motion
shows that the effect of H is larger than is indicated from recover-
h(Lag) = separation distance, ft
ies in fractal cross sections between Wells 1 and 5 for which geo-
H = intermittency coefficient
logic layer information is honored. However, the effect of H is still
less than the overall effect of different realizations. When the value k = absolute permeability, md
of H is large (H=0.94) for both porosity and permeability, a slight L = system length, ft
improvement in recovery occurs relative to the low H (H=0.71 M = mobility ratio
and 0.70) case for the same realization. The improved recovery nx = number of gridblocks in the x direction
could be caused by flow being less tortuous because fewer irregular- ny = number of gridblocks in the y direction
ities exist whenH=0.94 than when H=0.70 and 0.71. Fig. 13 also n z = number of gridblocks in the z direction
shows the various effluent concentration profiles that correspond NPe = transverse Peclet number, LOI/b 2
to the recoveries. It is evident that a reduction in the contrast of Q~ = cumulative PV injected
the k/q, ratio in the layers has decreased the controlling power of R = range
the one single fast layer observed previously when geologic layers S = standard deviation
were honored. Different breakthroughs and various shapes of the x,y,z = Cartesian coordinates, ft
effluent concentration profiles indicate that there is a definite vari- d.x = gridblock size in x direction, ft
ation in the miscible displacements for the assorted realizations in .:iy = gridblock size in y direction, ft
Fig. 13 (i.e., when the available geologic layers are not recognized). .:lz = gridblock size in z direction, ft
The immediate implications from the study of miscible displace- OIf = longitudinal dispersivity, ft
ments in cross sections generated by fractal interpolation is that if 01 t = transverse dispersivity, ft
the geologic layering structure is honored and the layers exhibit 'Y(h) = semivariogram value
a distinct difference in speed (k/q, ratio), fractal and linear interpo-
q, = porosity
lations will give similar results. This is a cumulative result of two
distinct effects. First, the contrast in k/ q, values occurs on a scale
Superscript
much larger than the scale on which fractal variations are defined.
Thus, because of size, the larger scale dominates the displacement = average
characteristics of the reservoir. Second, because of the similarities
of the structures of the two endpoints, the variance between the Acknowledgment
two locations is relatively small. As a result, the generated fractal Yngve Aasum thanks Amoco Production Co. for providing the fa-
distributions are quite smooth. cilities, computer time, and financial support to conduct this MS
If the geologic layering structure is not well-defined or the con- thesis study.
trasts in the speed of the layers are not severe, fractal interpola-
tions may be the best approach to use to obtain detailed information References
within a single layer or zone because it can forecast reservoir per- 1. Kelkar, B.G. and Gupta, S.P.: "The Effects of Small-Scale Heterogenei-
formance with associated uncertainties. This information cannot be ties on the Effective Dispersivity of Porous Medium, " paper SPE 17339
obtained with conventional linear interpolation. presented at the 1988 SPEIDOE Symposium on EOR, Tulsa, April
17-20.
Conclusions 2. Kelkar, M.G. and Gupta, S.P.: "A Numerical Study of Viscous Insta-
bilities: Effect of Controlling Parameters and Scaling Considerations,"
Geostatistical Method SPERE (Feb. 1991) 121-28.
1. To quantify correlation structures on a reservoir scale, data 3. Chopra, A.K., Stein, M.H., and Ader, J.C.: "Development of Reser-
from various scales must be available. In this case, we did not have voir Descriptions To Aid in Design of EOR Projects," SPERE (May
enough information on a small scale to quantify such a structure. 1989) 143-50; Trans., AIME, 287.
2. Data must be examined carefully before variograms are con- 4. Goggin, D.J. et al.: "Patterns of Permeability in Eolian Deposits: Page
structed. Nonergodic variograms were found to be more appropri- Sandstone (Jurassic), Northeastern Arizona," SPEFE (June 1988)
297-306.
ate if considerable differences or trends in mean values existed. 5. Goggin, D.J. et at.: "Permeability Transects in Eolian Sands and Their
3. Relatively fine-grid, single-phase, first-contact-miscible dis- Use in Generating Random Permeability Fields," paper SPE 19586
placement (M = 5) simulation in layers of a five-spot characterized presented at the 1989 SPE Annual Technical Conference and Exhibi-
by conditional simulation indicated effectively random behavior of tion, San Antonio, Oct. 8-11.
the permeability and porosity distributions and negligible uncer- 6. Da Costa e Silva, A.J.: "A New Approach to the Characterization of
tainty in displacement predictions. The conditional simulation tech- Reservoir Heterogeneity Based on the Geomathematical Model and Krig-
nique used may have the most potential for cases where more ing Technique," paper SPE 14275 presented at the 1985 SPE Annual
Technical Conference and Exhibition, Las Vegas, Sept. 22-25.
data-quantitative and/or qualitative-can be obtained between the
7. Carr, L.A. et at.: "A Complex Reservoir Characterized by Three-D
wells within the simulated system. Seismic, Geostatistical Reservoir Description, and Sponge-Core Anal-
ysis," SPEFE (Sept. 1989) 335-42; Trans., AIME, 287.
Fractal-Geometry Method 8. Alabert, F.: "Stochastic Imaging of Spatial Distributions Using Hard
1. The intermittency coefficient, H, for the 18 wells in Unit A and Soft Information," MS thesis, Stanford U., Stanford, CA (1987).
was found to be between 0.70 and 0.94 for both porosity and the 9. Hewett, T.A.: "Fractal Distributions of Reservoir Heterogeneity and
log of permeability in the vertical direction. These values agree Their Influence on Fluid Transport," paper SPE 15386 presented at
the 1986 SPE Annual Technical Conference and Exhibition, New
with other H values reported in the literature. Orleans, Oct. 5-8.
2. Simulation of single-phase first-contact-miscible displacement 10. Hewett, T.A. and Behrens, R.A.: "Conditional Simulation of Reser-
(M=5) with a relatively fme grid indicated that the fractal approach voir Heterogeneity With Fractals," SPEFE (Sept. 1990) 217-25; Trans.,
and linear interpolation give similar results if geologic layers are AIME,289.

18 SPE Formation Evaluation, March 1991


II. Emanuel, A.S. et al.: "Reservoir Performance Prediction Methods
Based on Fractal Geostatistics," SPERE (Aug. 1989) 311-18; Trans.,
Authors
AIME,287.
12. Mathews, J.L., Emanuel, A.S., and Edwards, K.A.: "Fractal Methods
Improve Mitsue Miscible Predictions," JPT (Nov. 1989) 1136-42;
Trans., AIME, 287.
13. Tang, R.W., Behrens, R.A., and Emanuel, A.S.: "Reservoir Studies
Using Geostatistics To Forecast Performance," paper SPE 18432
presented at the 1989 SPE Reservoir Simulation Symposium, Houston,
Feb. 6-8.
14. Davis, J.C.: Statistics and Data Analysis in Geology, John Wiley &
Sons, New York City (1986).
15. Matheron, G.: "Principles of Geostatistics, " Economic Geology (Dec.
1963) 58, No.8, 1246-66.
16. <?Iea, R.A .. : Me~uring Spatial Dependence With Semivariograms, Spa-
Kelkar Aasum Gupta
tial AnalysIs Senes 3, Kansas Geological Survey, Lawrence, KS (1975).
Yngve Aasum is a PhD-degree candidate in petroleum en-
17. LaPointe, P.R. and Hudson, J.A.: Characterization and Interpretation
gineering at the U. of Tulsa, from which he holds BS and MS
of Rock Mass Joint Patterns, special paper 199, GSA, Boulder, CO
degrees, also in petroleum engineering. His industry experi-
(1985) II.
ence includes assignments at Amoco Production Co. 's Re-
18. d~ Marsily, G.: Quantitative Hydrology, Academic Press, New York
search Center in Tulsa, OK, and at Phillips Petroleum Co. in
CIty (1986) Chap. II.
Stavanger. Mohan G. Kelkar is an associate professor of
19. Journel, A.G. and Huijbregts, C.J.: Mining Geostatistics, Academic
petroleum engineering at the U. of Tulsa. His research in-
Press, London (1978).
t~r~sts.include reservoir characterization and production op-
20. Mandelbrot, B.B.: Fractals: Form, Chance, andDimension, W.H. Free-
timization techniques. He holds an MS degree in petroleum
man & Co., San Francisco (1977).
engineering and a PhD degree in chemical engineering from
21. Mandelbrot, B.B.: The Fractal Geometry of Nature, W.H. Freeman
the U. of Pittsburgh. He served as faculty sponsor to the U.
& Co., New York City (1982).
of Tulsa SPE Student Chapter during 1984-89. Surendra P.
22. The Science of Fractal Images, H.O. Peitgen and D. Saupe (eds.),
Gupta is a research supervisor in the Rock Properties Group
Springer-Verlag, New York City (1988). at Amoco Production Co. in Tulsa. His previous work involved
23. Mandelbrot, B.B. and Van Ness, J.W.: "Fractional Brownian Motions
chemical and gas-drive EOR processes. Gupta holds a PhD
Fractional Noises, and Applications," SIAM Rev. (Oct. 1968) 10, No:
degree in chemical engineering from Purdue U. He has served
4,422-37. on several SPE committees, including the Fluid Mechanics
24. Young, L.C.: "A Finite-Element Method for Reservoir Simulation," and Oil Recovery technical meeting committees (1984-88) and
SPEl (Feb. 1981) 115-28. the Editorial Review Committee (1983-84).
25. Journel, A. G.: "Nonparametric Estimation of Spatial Distributions,"
Math. Geol. (1983) 15, No.3, 445-68.
26. Journel, A.G.: "Constrained Interpolation and Qualitative SI Metric Conversion Factors
Information-The Soft Kriging Approach," Math. Geol. (1986) 18, No.
ft x 3.048* E-Ol
3, 269-86.
27. Aasum, -:.: "Effect of Heterogeneities on Miscible Displacements," md x 9.869 233 E-04
MS theSIS, U. of Tulsa, Tulsa, OK (1989).
28. Isaaks, E.H. and Srivastava, R.M.: "Spatial Continuity Measures for
Probabilistic and Deterministic Geostatistics," Math. Geol. (1988) 20, ·Conversion factor is exact. SPEFE
No.4, 313-41. Original SPE manuscript received for review April 22, 1990. Paper accepted for publica·
29. Goggin, D.J.: "Geologically-Sensible Modelling of the Spatial Distri- tIon Oct. 31, 1990. Revised manuscript received Dec. 10, 1990. Paper (SPE 20257) first
bution of Permeability in Eolian Deposits: Page Sandstone (Jurassic), presented at the 1990 SPE/DOE Seventh Symposium on Enhanced Oil Recovery held in
Northern Arizona," PhD dissertation, U. of Texas, Austin, TX (1988). Tulsa, April 22-25.

SPE Formation Evaluation, March 1991 19

You might also like