ABSTRACT
Fracture characterization protocols that reduce sampling bias are
likely to yield higher quality input for exploration and development
decisions when dealing with naturally fractured reservoirs. A new
set of estimators for fracture density, intensity, and mean trace
length corrects for sampling biases and provides a useful integrated
description for bulk aspects of a fracture network. These estimators
are based on counts of intersections between fracture traces and
circular scan lines and of trace terminations in circular windows.
Application to synthetic fracture patterns with known parameters
validates the use of the new estimators, which are then applied to
natural fault trace maps from seismic volumes and joint trace maps
from rock pavements. The new estimators are distribution independent and eliminate the effects of orientation, censoring, and
length biases, which limit the effectiveness of other sampling techniques. Estimator accuracy improves as sample size increases, particularly for larger circles that exceed a fracture-defined block size.
Estimator accuracy for mean trace length improves when the sample exceeds threshold count values for fracture terminations based
on guidance from the analysis of similar synthetic patterns. These
new estimators also provide both inputs and independent checks of
predictions for fracture-generator programs used to model fracture
populations in a rock volume.
INTRODUCTION
Quantification of fracture parameters such as density, size, and intensity aids in the assessment of hydrocarbon flow and storage in
fractured reservoirs (Reiss, 1982; Nelson, 1985; Dershowitz and
LaPointe, 1994; Narr, 1996) but is complicated by the difficulty of
deciding on the best approach for incorporating fractures into a
reservoir model. Problems arise from the lack of consensus as to
Copyright 2002. The American Association of Petroleum Geologists. All rights reserved.
Manuscript received January 24, 2000; revised manuscript received January 10, 2001; final acceptance
June 15, 2002.
2089
AUTHORS
M. B. Rohrbaugh Jr. Tennessee
Department of Environment and
Conservation, Division of Underground
Storage Tanks, 540 McCallie Avenue, Suite
550, Chattanooga, Tennessee, 374022013
M. Bruce Rohrbaugh Jr. received his B.S.
degree in geology from West Virginia
University in 1997 and his M.S. degree in
structural geology from the University of
Tennessee, Knoxville in 2000. His research
interests include hydrogeology, application of
computers to solving geologic problems, and
structural geology. He is currently employed
as a geologist with the Tennessee Department
of Environment and Conservation.
W. M. Dunne Department of Geological
Sciences, 306 G&G Building, University of
Tennessee, Knoxville, Tennessee, 37996-1410;
wdunne@utk.edu
William M. Dunne, although born in the
United States, received his B.S. degree and
Ph.D. in geology from the University of Bristol,
England. He joined the Department of
Geological Sciences at the University of
Tennessee in 1988 as an associate professor
and is now is a professor and department
head. His research interests include fracture
characterization particularly in younger rocks,
deformation in thrust belts from large to small
scale, and deformation analysis of
sedimentary rocks.
M. Mauldon Department of Civil and
Environmental Engineering, Virginia Tech, 200
Patton Hall, Mail Code 0105, Blacksburg,
Virginia, 24061; mauldon@vt.edu
Matthew Mauldon, although born in England,
has geology (B.A.) and civil engineering (M.S.)
degrees and a Ph.D. in civil engineering from
the University of California at Berkeley. He
spent eight years on the faculty at the
University of Tennessee, where he
collaborated with Dunne and Rohrbaugh.
Mauldon is now an associate professor in the
Via Department of Civil and Environmental
Engineering at Virginia Tech, where he
teaches and conducts research in the areas of
rock mechanics, engineering geology, and
geotechnical engineering.
ACKNOWLEDGEMENTS
Acknowledgment is made to the donors of
The Petroleum Research Fund, administered
by the American Chemical Society, for partial
support of this research. Additionally, the
Geological Society of America and the Southeastern Section of the Geological Society of
America are thanked for their partial support
of the field work. We would also like to thank
M. F. Schaeffer for permission to use the fracture trace map from Rocky Creek, South Carolina, and Camilo Montes, Yen-Yit Chan, You Li,
and Jim Calcagno for programming assistance.
Steve Laubach, John Lorenz, and Bill Dershowitz are thanked for their insightful and
constructive reviews.
which parameters to quantify, the difficulties in measuring the parameters, and intrinsic sampling biases. To address these problems,
we offer a set of estimators for fracture parameters based on the
use of circular scan lines and windows where a scan line is the perimeter and the window is the interior of a circle. These estimators
reduce sampling biases and provide a useful integrated description
of bulk aspects of a fracture network. Use of these estimators requires only counts of the number (n) of fracture traces intersecting
the circumference and/or counts of the number (m) of fracture
traces terminating in the circle interior. Performance of the new
circular scan line/window estimators is evaluated for both synthetic
and natural fracture patterns to demonstrate that estimates converge on true values for a fracture trace population, to demonstrate
that the new estimators outperform or match existing estimators,
and to discuss issues of estimator performance.
Fracture Parameters
A redundant, commonly mutually inconsistent vocabulary exists to
describe the amount of fracturing in a rock. Rather than reviewing
this terminology, we define three key parameters of a fracture pattern: density, size, and intensity (Table 1).
Density
Fracture density is commonly treated as the number of observed
isolated fractures or fracture segments per unit length, area, or volume (Dershowitz and Herda, 1992; Ghosh and Daemen, 1993).
This is a scale-dependent quantity that we call apparent density.
Fracture density is defined in this article as the number of fractures
per unit length, area, or volume, enumerated in terms of unique
points, such as fracture centers (Mauldon, 1998; Mauldon and Dershowitz, 2000). Apparent density overestimates density (Kulatilake
and Wu, 1984; Mauldon et al., 2001), and the magnitude of this
overestimation increases as sample size decreases (Figure 1B). For
example, the apparent density (number of visible traces divided by
circle area) of fractures of Set 1 in Figure 1A is 0.0014 per m2 for
a circular window of radius 75 m and increases to 0.0020 per m2
for a smaller window of radius 25 m. If trace centers (dots in Figure
1A) are counted to estimate true density for Set 1, the estimates
(count divided by circle area) are 0.0010 per m2 for both the circle
of radius 25 m and the circle of radius 75 m. In practical applications, one half the number of fracture trace terminations is used as
an unbiased estimate of the number of trace centers, because centers cannot be identified unless both ends of a trace are visible
within the window (Mauldon, 1998; Mauldon et al., 2001).
Size
Fracture size is defined in one, two, or three dimensions, as fracture trace length, area, and volume, respectively. For composite
connected fractures, an investigator should decide whether to
characterize individual segments or the entire composite. Typically
2090
Definition
Density
Linear
Areal (q)
Volumetric
Size
Linear (l)
Areal
Volumetric
Linear
Areal (I)
Volumetric
Intensity
Estimator
q m/2pr 2
l (pr/2)(n/m)
I n/4r
*Where L is a dimension of length and r is radius. See Figure 3 for illustration of m and n.
Two common sampling methods are used for estimating fracture parameters: straight scan lines and
areal sampling (Figure 2) (LaPointe and Hudson,
1985; Priest, 1993; Wu and Pollard, 1995, Becker and
Gross, 1996; Marrett et al., 1999; Ortega and Marrett,
2000). Straight scan lines sample the fractures they
intersect and are used to systematically record fracture
characteristics, such as number, orientation, aperture,
and so on. (Priest and Hudson, 1981). Areal sampling
involves mapping the fracture trace pattern and recording desired fracture characteristics at locations in
the map area (e.g., Priest, 1993; Wu and Pollard,
1995).
Rohrbaugh et al.
2091
Figure 2. Circular scan line/window (dotted circle), areal (irregular white window), and straight scan line (dotted line) sampling of a fracture trace population. Solid lines represent visible
fracture traces, and dashed lines represent covered fracture
traces.
Sampling Biases
Orientation bias occurs where scan lines or the long
axes of inequant sampling areas are not perpendicular
to a fracture set. In both cases, intensity is underestimated (Terzaghi, 1965; Priest, 1993; Mauldon and
Mauldon, 1997). Scan line estimates are corrected by
dividing by the cosine of the angle between the scan
line and the normal to the fracture set (Terzaghi, 1965;
Peacock et al., in press). However, as this angle approaches 90, the cosine approaches zero, and corrected estimates approach infinity, significantly overestimating intensity (Priest, 1993). No procedures are
currently available for correcting orientation bias from
inequant sampling areas.
Censored fracture traces extend beyond the exposure or seismic coverage, so that one or both ends
are not visible (Figure 1A) (e.g., Cruden, 1977;
Baecher and Lanney, 1978; Einstein and Baecher,
1983; Kulatilake and Wu, 1984; LaPointe and Hudson,
1985; Pickering et al., 1995; Mauldon, 1998; Marrett
et al., 1999). Such traces are referred to as singly or
doubly censored, respectively. Using censored traces to
estimate density and size directly, rather than using estimators such as those in Table 1, leads to overestimates of density and underestimates of size (Figure 1).
For example, a count of all visible trace segments in a
circle of radius 25 m (4 segments vs. 2 centers) of Set
1 (Figure 1A) overestimates density by a factor of two
(0.002 vs. 0.001 per m2). Similarly, using the censored
2092
NEW ESTIMATORS
Because neither areal sampling nor straight scan lines
are completely satisfactory in terms of efficiency, accuracy, and lack of bias, this article examines the use
of circular scan lines and windows for characterizing
intensity, density, and mean trace length (Figure 2).
Data from circular scan lines and windows are applied
to estimators (Table 1) that do not require knowledge
of fracture spacing, trace length, or orientation distributions and are, therefore, distribution independent
(Mauldon, 1998; Mauldon et al., 2001).
means for the estimates (Figure 4) converge on the input values for intensity, density, and mean trace length
and correspond closely to them after about 40 samples,
whereas individual estimates fluctuate about the mean.
These results show that the new estimators deal successfully with orientation, censoring, and length biases.
Similar results were found for all other synthetic cases.
Comparison to Other Estimators
Having established that the new estimators overcome
sampling biases, their accuracy was compared to that
from straight scan lines and/or areal sampling of a synthetic pattern. To achieve equivalent sampling comparison, estimates were obtained using 100 circular
scan lines/windows of radius 10 L , 100 straight scan
lines of the same length (2p (10 L) 63 L), and
areal samples consisting of 100 circle interiors.
For example, a fracture set with orientation of 060
30, mean trace length of 40 20 L, intensity of
1.95 L/L2, and density of 0.049 per L2 was generated
(Figure 5A). Intensity estimates from both circular
scan lines and areal samples accurately estimate intensity, but map construction for areal samples is likely to
Rohrbaugh et al.
2093
Figure 5. Estimator performances for a synthetic fracture trace pattern. (A) Part of a synthetic fracture pattern used to test relative
performance of new and old estimators. Pattern characteristics are described in text. (B) Intensity estimates from circular scan lines
(thick black line), circular areas (light gray line), and straight scan lines (thin line). Control value is 1.95 L/L2. (C) Density estimates
from circular windows (thick black line) and circular areal sampling of apparent density (light gray line). Control value is 0.049 per
L2. (D) Mean trace length estimates from circular windows (thick black line) and circular areal sampling of apparent trace length
(light gray line). Control value is 40.0 L. Open circles in (B), (C), and (D) are the running means for appropriate estimators.
DISCUSSION OF ESTIMATOR
PERFORMANCE
Effects of Block Size on Estimates
Block size in two dimensions is the unfractured area
bounded by fracture traces from two or more sets
2095
2096
Late Namurian
(Carboniferous), Upper
Sandstone Group
Name, Location,
Figure No.
Joints 010020
Faults WSW
Faults variety of
orientations
Joints cross075
master165
Sets
Single set
Orthogonal pattern
(200 & 290); other
younger sets
Orthogonal pattern
(200 & 290) and
one younger joint
set
Somewhat polygonal
Pattern Geometry
Same as Amroth
Origin
5000 ft grid based on Nevada State Plane coordinates, Nevada State Plane projection.
M. F. Schaeffer, 1998,
unpublished data, 25
cm, 1:120, 1446.6 m2
(exposed area), 922.2
m2 (exposed, stippled
area)
2097
Figure 6. Fracture trace maps. (A) Llantwit Major. (B) Amroth. (C) Sleipner Vest (modified from Ottesen Ellevset et al., 1998). (D)
Cartier Trough (modified from Walsh et al., 1996). Continued.
igneous processes (Figure 6E, G); and patterns in crystalline rocks that vary from a single fracture set in a
granite (Figure 6F) to a complex array of fractures that
2098
Figure 6. Continued. (E) Telpyn Point. (F) Ward Lake (modified from Segall and Pollard, 1983). (G) P100 (modified from Barton et
al., 1993). (H) Rocky Creek (modified from M. F. Schaeffer, 1998, unpublished data). Light-gray regions in (B), (F), (G), and (H)
represent visually identified homogeneous subdomains, and black regions in (F) and (H) are unexposed. See Table 2 for geologic
information for trace maps.
Table 3. Accuracy of Intensity Estimates Using Ten Circular Scanlines for a Variety of Natural Fracture Patterns
Estimate Accuracy, Scanline Radius
Natural Fracture Patterns
1.0 m
1.5 m
2.0 m
Intensity** (m/m2)
Y
14.2
Y
5.4
N
49.5
Y
9.9
Y
15
Y
9.0
N
50.7
Y
7.6
Y
6.0
Y
8.8
Y
12.6
N
63.1
Y
10.6
Y
13.3
Y
14.0
Y
2.9
Y
9.7
Y
4.8
N
18.6
Y
7.7
N
35.9
Y
0.1
Y
2.5
Y
3.4
N
23.4
Y
8.2
Y
13.7
High
2.00
High
3.40
Low
0.55
Mod
1.80
Low
0.50
High
3.60
Mod
1.00
High
2.80
High
3.30
*Y(yes)/N(no) indicates whether estimate is/is not within 15% of the areal value; / denotes an over/underestimate.
**Actual percent error for the intensity estimate are shown in the second row for each fracture system. Low/moderate/high intensity indicates values in the ranges
(0:1), (1:2), or (2), respectively.
larger sampling circles are restricted because of pavement width (Figure 6). The subdomain at Ward Lake
eliminates a low intensity region for the fracture set
but contains heterogeneities due to fracture clusters
and covered areas that prevent random circle placement. Thus, the geometry of older joints, joint persistence, and joint clustering, as expected, generate spatial heterogeneity. More important, based on these
results, we believe that a sampling strategy of 10 circles
with a size that exceeds the block size or fracture spacing but is substantially smaller than the minimum dimension of a sample area will yield an intensity estimate within 15% of the actual intensity for a
homogeneous fracture pattern.
Minimum Count of m for Reliable Tracelength Estimates
Of the three estimators discussed here, the mean tracelength estimator exhibits the greatest variability. Low
trace densities, large fracture trace lengths relative to
circle size, or small total sample area can lead to small
counts of fracture endpoints (m). Such small counts
may be an issue for the petroleum industry be-
2101
input mean trace-length values, is graphed as a function of m counts in Figure 9. For this graph , an accurate
result was defined as within 15% of the input value.
Using this criterion, virtually all sampling strategies
that yielded m counts of 30 or greater produced accurate estimates. These results (Figure 9) suggest that
for a specified fracture intensity, a threshold value of
m counts exists for which accurate estimates are a near
certainty.
This analysis could not be extended directly to natural fracture patterns, because, due to problems of censoring and length bias, no direct technique exists to
accurately estimate mean trace length for comparison
to the results of the circle-based estimator. Although
threshold values of m counts for accurate estimates
could be determined for our synthetic patterns, our
models had Poissonian trace center distributions,
which is unlikely in nature. Still, results from the synthetic patterns should provide guidance, such as in the
case considered here, where a fracture pattern with
large fractures and an intensity of 2 m/m2 that is being
sampled by small circles (or small-diameter boreholes)
needs an m count greater than 30 to yield an accurate
estimate at the 15% level.
2.
3.
4.
5.
CONCLUSIONS
6.
1. New circle-based estimators for fracture intensity,
density, and mean trace length virtually eliminate
orientation, censoring, and length biases, which se2102
REFERENCES CITED
Baecher, G. B., and N. A. Lanney, 1978, Trace length biases in joint
surveys: Proceedings of the 19th U.S. Symposium on Rock Mechanics, v. 1, p. 5665.
Barton, C. C., E. Larsen, W. R. Page, and T. M. Howard, 1993,
Characterizing fractured rock for fluid-flow, geomechanical,
and paleostress modeling: methods and preliminary results from
Yucca Mountain, Nevada: United States Geological Society
Open-File Report 93-269, 62 p.
Becker, A., and M. R. Gross, 1996, Mechanism for joint saturation
in mechanically layered rocks: an example from southern Israel:
Tectonophysics, v. 257, p. 223237.
Cruden, D. M., 1977, Describing the size of discontinuities: International Journal of Rock Mechanics and Mining Sciences and
Geomechanical Abstracts, v. 14, p. 133137.
Dershowitz, W. S., and H. H. Herda, 1992, Interpretation of fracture
spacing and intensity, in J. R. Tillerson and W. R. Wawersik,
eds., Proceedings of the 33rd U.S. Symposium on Rock Mechanics: Rotterdam, Balkema, p. 757766.
Dershowitz, W. S., and P. R. LaPointe, 1994, Discrete fracture approaches for oil and gas applications, in R. A. Nelson and S.
Laubach, eds., Rock mechanics: Rotterdam, Balkema, p. 19
30.
Dunne, W. M., and C. P. North, 1990, Orthogonal fracture systems
at the limits of thrusting: an example from southwestern Wales:
Journal of Structural Geology, v. 12, p. 207215.
Einstein, H. H., and G. B. Baecher, 1983, Probablistic and statistical
methods in engineering geology, specific methods and examples, part 1: exploration: Rock Mechanics and Rock Engineering, v. 16, p. 3972.
Ghosh, A., and J. J. K. Daemen, 1993, Fractal characteristics of rock
discontinuities: Engineering Geology, v. 34, p. 19.
Gilmour, H. M. P., D. Billaux, and J. C. S. Long, 1986, Models for
calculating fluid flow in randomly generated three-dimensional
networks of disk-shaped fractures: theory and design of
FMG3D, DISCEL and DIMES: Berkeley, California, Lawrence
Berkeley Laboratory, Earth Sciences Division, LBL-19515,
144 p.
Jian, X., R. Olea, and Y. Yu, 1996, Semivariogram modeling by
weighted least square: Computers and Geosciences, v. 22,
p. 387397.
Kulatilake, P. H. S. W., and T. H. Wu, 1984, The density of discontinuity traces in sampling windows (technical note): International Journal of Rock Mechanics and Mining Sciences and
Geomechanical Abstracts, v. 21, p. 345347.
Kulatilake, P. H. S. W, D. N. Wathugala, and O. Stephansson, 1993,
Joint network modeling with a validation exercise in Stripa
Mine, Sweden: International Journal of Rock Mechanics and
Mining Sciences and Geomechanical Abstracts, v. 30, p. 503
526.
Kulatilake, P. H. S. W., R. Fiedler, and B. B. Panda, 1997, Box fractal
dimension as a measure of statistical homogeneity of jointed
rock masses: Engineering Geology, v. 48, p. 217229.
LaPointe, P. R., 1988, A method to characterize fracture density and
connectivity through fractal geometry: International Journal of
Rock Mechanics and Mining Sciences and Geomechanical Abstracts, v. 24, p. 421429.
LaPointe, P. R., 1993, Pattern analysis and simulation of joints for
rock engineering, in J. A. Hudson, ed., Comprehensive rock
engineering, volume 3rock testing and site characterization:
New York, Pergamon Press, p. 215239.
LaPointe, P. R., and J. A. Hudson, 1985, Characterization and interpretation of rock mass joint patterns: Geological Society of
America Special Paper 199, 37 p.
Rohrbaugh et al.
2103
2104