You are on page 1of 12

Journal of Structural Geology 72 (2015) 55e66

Contents lists available at ScienceDirect

Journal of Structural Geology


journal homepage: www.elsevier.com/locate/jsg

The use of topology in fracture network characterization


David J. Sanderson a, c, *, Casey W. Nixon b
a
Faculty of Engineering and the Environment, University of Southampton, Southampton SO17 1BJ, UK
b
Ocean and Earth Sciences, University of Southampton, Southampton SO14 3ZH, UK
c
Reservoir Development, BP, Chertsey Road, Sunbury-on-Thames TW16 7BP, UK

a r t i c l e i n f o a b s t r a c t

Article history: In two-dimensions, a fracture network consist of a system of branches and nodes that can be used to
Received 8 September 2014 define both geometrical features, such as length and orientation, and the relationship between elements
Received in revised form of the network e topology. Branch lengths are preferred to trace lengths as they can be uniquely defined,
9 January 2015
have less censoring and are more clustered around a mean value. Many important properties of networks
Accepted 21 January 2015
Available online 31 January 2015
are more related to topology than geometry.
The proportions of isolated (I), abutting (Y) and crossing (X) nodes provide a basis for describing the
topology that can be easily applied, even with limited access to the network as a whole. Node counting
Keywords:
Fracture
also provides an unbiased estimate of frequency and can be used in conjunction with fracture intensity to
Network estimate the characteristic length and dimensionless intensity of the fractures. The nodes can be used to
Topology classify branches into three types e those with two I-nodes, one I-node and no I-nodes (or two con-
Geometry nected nodes). The average number of connections per branch provides a measure of connectivity that is
almost completely independent of the topology. We briefly discuss the extension of topological concepts
to 3-dimensions.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction mineral fill, etc.). Fracture sets can be characterized by the statis-
tical distribution of these attributes (i.e. averages, ranges, etc.). The
Networks of discontinuities control many of the physical prop- geometry of fractures is established by measurement, and usually
erties of rocks and their characterization is important in many areas involves the spatial orientation of the fracture surface, together
of geology. The abundance and arrangement of the discontinuities with some physical dimension, such as trace length, area or volume.
may control physical properties, such as stiffness, strength, porosity These topics are widely discussed in the literature (e.g. Priest, 1993;
and permeability (Adler and Thovert, 1999; others). For the pur- Adler and Thovert, 1999).
poses of this paper we will use the term ‘fracture’ to refer to any In general, fractures will interact with each other and the rock
type of discontinuity, only referring to different types of ‘fracture’ matrix to control the physics of the rock (e.g. Zimmerman and
(joints, faults, deformation bands, stylolite, etc.) where this is Main, 2003). Thus, it is not just the geometry of the fractures that
relevant to discussion of the underlying physical process. controls rock properties, but also the relationships between indi-
A fracture network can be regarded as a system of fractures vidual fractures or fracture sets; a simple example of such a feature
developed within the same rock volume. A network may involve a is the connectivity of the fractures. We will use the term ‘topology’
number of distinct fracture sets, which may or may not intersect. to describe the relationships between these geometrical objects.
These generally evolve over time and vary in their spatial distri- Topology has often been used to characterize and describe complex
bution (e.g. Adler and Thovert, 1999). The individual fractures can network structures in the natural, engineering and social sciences
be described by a range of attributes such as orientation, size, age, (e.g. Latora and Marchiori, 2002; Ravasz and Barab asi, 2003;
morphology, etc. Many of these attributes are geometrical (length, Boccaletti et al., 2006). In general, topology is an area of mathe-
angle, etc.) or involve some form of classification (fracture type, matics that deals with the abstraction and generalization of spatial
relationships such as connectivity and continuity. Essentially it
refers to properties that are unchanged by continuous trans-
* Corresponding author. Bldg 7, Room 5030, Faculty of Engineering and the formation (i.e. strains, but not tearing) of the space in which the
Environment, University of Southampton, Southampton SO17 1BJ, UK. objects are embedded. Such transformations change the
E-mail address: d.j.sanderson@soton.ac.uk (D.J. Sanderson).

http://dx.doi.org/10.1016/j.jsg.2015.01.005
0191-8141/© 2015 Elsevier Ltd. All rights reserved.
56 D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66

geometrical properties (lengths and angles), but not the topology; a Manzocchi (2002), we can recognize 3 types of node e isolated tips
concept familiar in many areas of structural geology. As such, the (I-nodes); crossing fractures (X-nodes); and abutments or splays (Y-
description and characterization of topology generally involves nodes or T-nodes). In the case of a random array of lines (Fig. 2a),
dimensionless parameters (Jing and Stephansson, 1997) that are each line will consist of two I-nodes, with additional intersections
invariant to scale changes and continuous transformation. represented by X-nodes. For many ‘natural’ fracture networks,
In this paper, we introduce some simple techniques to charac- however, fractures terminate against (or abut) pre-existing frac-
terize the topology of a fracture network. In 2-dimensions this in- tures, producing many Y-nodes (Fig. 2b). In the case of regular
volves recognition that a network comprises a system of branches lattices, a fourth type of node (V in Fig. 2c) exists at which two tips
and nodes (Fig. 1), which may be used to provide important addi- coincide. This is due to the position of the nodes determining the
tional parameters to characterize the network. We will show that lines in such lattices, whereas in continuum percolation (random
topology helps in the general characterization of fractures and, lines) and natural examples it is the lines that determine the nodes.
most importantly, allows us to make some estimates of the network Thus, the probability of a V-node occurring in a natural network is
properties, even where there is limited access to the network as a negligible.
whole. In this paper we will address the following objectives:

1) To outline simple methods for the topological analysis of frac- 2.2. Node counting
ture networks.
2) To explain how topology links different measures used to The proportion of I-, Y- and X-nodes can be used to characterize
characterize fracture size and abundance; a fracture network (Manzocchi, 2002; Ma €kel, 2007), and these
3) To promote the use of branches, rather than trace lengths, in proportions may be plotted on a triangular plot (Fig. 3). Such a
analysis of fracture systems; classification is topological rather than geometrical as it is un-
4) To provide practical guidelines for the determination of fracture changed by any continuous transformation of the network.
measures, especially in situations where we have limited access Since I- and Y-nodes represent the tip of ONE line, the number of
to the network; lines (NL) is given by:
5) To identify sources of error and bias in these measures, and
NL ¼ 1=2ðNI þ NY Þ (1)
minimize these.
This is a very useful relationship in counting the number of lines
More specifically, topology is fundamental to an understanding in a given area, especially where many of the lines extend outside
of connectivity of individual fractures within the network. the sample area, and is equivalent to the tip-counting procedure
Although we will touch on this subject, it will be the subject of a advocated by Mauldon (1994) and Rohrbaugh et al. (2002) and is
subsequent paper. discussed further in Section 4.
Each branch will have two nodes, with an I-node contributing to
2. Network topology one branch, a Y-node to 3 branches and an X-node to 4 branches.
Hence the number of branches (NB) is
2.1. Basic concepts of lines, nodes and branches
NB ¼ 1=2ðNI þ 3NY þ 4NX Þ (2)
In 3-dimensions, any system of fractures will consist of a
Thus, the ratio of number of branches to lines is
number of finite surfaces or planes that either terminate at tip lines,
or intersect with other planes at intersection (or branch) lines. In 2- NB =NL ¼ ðNI þ 3NY þ 4NX Þ=ðNI þ NY Þ
dimensions, the fracture surfaces intersect a surface of inspection
(often an exposure surface or bedding plane) to form a finite line e ¼ ðPI þ 3PY þ 4PX Þ=ðPI þ PY Þ (3)
the fracture trace (Fig. 1). These fracture traces will either terminate
at tips, or intersect and abut other fracture traces. Where PI, PY and PX represent the proportion of each type of node.
In this paper we consider the topology of a 2-D fracture network Since PI þ PY þ PX ¼ 1 or PX ¼ 1  PI  PY, Eq. (3) can be written more
to consist of lines, nodes and branches, where each line consists of simply as:
one or more branches with a node at each end (Fig. 1). Following
NB =NL ¼ ð4  3PI  PY Þ=ðPI þ PY Þ (4)
Eqs. (1)e(4) allow us to determine the number of lines or
branches by simply counting the number of nodes. In addition,
these equations also allow us to convert the number of lines to the
equivalent number of branches. Since the definition of a complete
branch
Y line or branch requires 2 nodes, it is implicit in Eqs. (1)e(4) that
lines and branches that extend beyond the sampled area are
A branch
counted as ½ and that those cutting through the area of sampling
I will be omitted from the analysis of that area. In this way lines are
X
only counted a total of once in non-overlapping sample areas (see
Y further discussion in Sections 3 and 4).
Y
Whilst node counting within a finite area is a basic geometrical
X
I B descriptor and has dimensions of [L]2, the proportions of nodes,
node and measures derived from these, are basic topological descriptors,
Y since these will not change with transformation of a given sampling
space.
Fig. 1. Fracture trace (AeB), with associated intersecting fractures (dashed), showing
Fig. 4a shows how the ratio NB/NL varies with the topology of the
arrangement of nodes and branches: I-nodes (circles); Y-nodes (triangles); X-nodes network. A system of isolated fractures will be dominated by I-
(diamonds). nodes and NB/NL / 1, whereas a system of long, closely-spaced,
D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66 57

I I
I Y Y X Y Y
X X
X Y X I
I I
X X
Y Y X V I

I
X I Y Y V
X X I Y V
I
I I

(a) (b) (c)


Fig. 2. Network topology defined by the arrangement of lines and nodes: a) random array of lines as generated in stochastic models; b) schematic representation of a fracture
network; c) network generated by random selection of branches on a square grid.

I Deformation band
Fault
Joint
Stochastic line
simulations

Y X
Fig. 3. Triangular plot of the proportion of node types for different networks (after Manzocchi (2002)). Three natural fracture networks (from Sanderson and Zhang (1999)) and two
random line simulations show how the fracture network topology is characterized. The dashed lines show specific numbers of intersections per line (see Fig. 4b), with CL ¼ 2
representing a limit above which a spanning cluster is not possible and CL ¼ 3.57 the value widely reported from random line simulations, see text for details.

cross-cutting fractures will be dominated by X-nodes and NB/NL / poorly connected fracture networks) and 10 (e.g. compare Figs. 3
∞. A system dominated by abutting fracture terminations (Y-nodes) and 4).
will have NB/NL z 3. Natural fracture networks may vary between Another parameter that is easily determined from the node
these extremes, but generally have values between 2 (typical for count is the average number of connections per line (CL). Since a

I I
NB / NL CL 1

2
2

3 3

4 4
5
6
7
8
10 10

20

Y X Y X
(a) (b)
Fig. 4. Triangular plots of node proportions. (a) ratio of number of branches to lines (NB/NL) and (b) average number of connections per line (CL), showing how these vary with
network topology.
58 D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66

connection must be either a Y- or X-node and each of these will


provide a connection on two lines, we have: a)

CL ¼ 2ðNY þ NX Þ=NL (5a)

CL ¼ 4ðNY þ NX Þ=ðNI þ NY Þ (5b)


b)
CL ¼ 4ðPY þ PX Þ=ðPI þ PY Þ (5c)
The number of connections per line (CL) has been widely used as
a measure of connectivity (Balberg and Binenbaum, 1983;
Berkowitz, 1995; Manzocchi, 2002), with CL ¼ 3.57 (Fig. 3) being
Fig. 6. Comparison of branch and fracture traces. (a) Shows a trace map of fractures e
the value simulated for percolation of a system of random lines of
it is a matter of interpretation how one relates different parts of the traces to one
fixed length (Balberg et al., 1984). However, CL is not independent of another. (b) Shows the same network with the unambiguously identified nodes and
topology, as can be seen in Fig. 4b, where networks with no isolated branches.
nodes plot along the XeY axis and would be completely connected,
yet have very different values of CL. is NI: (3NY þ 4NX). Thus the probability of the end of any branch
being an isolated (PI) or connected (PC) node is given by:

2.3. Branch classification


PI ¼ NI =ðNI þ 3NY þ 4NX Þ
PC ¼ ð3NY þ 4NX Þ=ðNI þ 3NY þ 4NX Þ
Every branch has two nodes that can be used to classify it into
one of 6 types: IeI, IeX, IeY, YeY, YeX, XeX. As nodes can be If there is a random assignment of nodes to branches the
grouped into isolated (I) nodes and connecting (C, i.e. X- or Y- probability of each branch type is:
nodes), we can simplify the branches into three main topological
groups: IeI (isolated) branches, IeC (partly connected) branches,
and CeC (doubly connected) branches, as proposed by Ortega and PII ¼ PI2 ; PIC ¼ PI PC ; PCC ¼ PC2
Marrett (2000) for fracture traces. We use this scheme to plot a
These values plot on the ternary branch diagram (Fig. 5a) as a
triangular diagram of the proportion of branch types (Fig. 5a).
single line with increasing values of CB. Any departure from this line
Networks that plot towards the IeI corner of the triangle have a
indicates correlations between the node types at opposite ends of a
high proportion of isolated branches and, hence, a low level of
branch, rather than differences in underlying topology. For
connectivity. Networks that plot towards the CeC corner of the
example, fully connected networks of Y and X nodes have the same
triangle have a high proportion of interconnected branches.
connectivity with CB ¼ 2. This reinforces the idea that the number
From the number of different node types, we can derive an
of connections per branch (CB) is a more useful measure of con-
average number of connections per branch (CB) where:
nectivity than the more widely used measure of connections per
line (CL).
CB ¼ ð3NY þ 4NX Þ=NB (6)
Fig. 5 shows results from two strike-slip fault networks from
This gives a dimensionless number in the range 0e2 that is a north Devon, UK: network A from Westward Ho! (Nixon et al.,
useful measure of connectivity. There is not a one-to-one mapping 2011) and network B from off Hartland Point (Nixon et al., 2012).
between the nodal classification and the branch classification Note that network A (Westward Ho!) is more connected, with
(Fig. 5b) as Eq. (6) is not entirely independent of topology. CB ¼ 1.7, and a high proportion of CeC branches. Network B
Each branch ends at a node, but an I-node produces just one end, (Hartland) is not so well connected, forming only local clusters of
a Y-node produces 3 and an X-node produces 4, giving a total of faults, with CB ¼ 1.1, and a dominance of IeC branches (~50%) with
(NI þ 3NY þ 4NX) ends. The proportion of isolated to connected ends similar proportions of IeI and CeC branches (~25% each).

Fig. 5. (a) Branch classification plot that shows proportions of different branch types with numbers 0 to 2.0 indicating connections per branch (CB). The curve shows results from
random assignment of node types to branches, based on probabilities discussed in text; we would expect samples of branches to cluster around this curve. A and B are examples
from two studies of fault networks from Hartland Point and Westward Ho, respectively. (b) Values of number of connections per branch (CB) on the triangular plot showing
proportions of different nodes, together with corresponding node proportions for networks A and B.
D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66 59

2.4. Why use branches rather than fracture traces? divided into a larger number of branches. This results in the
branches having a narrower range of lengths, the distributions of
There are at least three practical problems routinely encoun- which generally conform to a negative exponential or log normal
tered when characterizing fracture networks in terms of trace (occasionally normal) distribution. These distributions are less
length. affected by restricted sample ranges and hence their means
(characteristic lengths) are more easily determined. Fig. 7 shows
2.4.1. Non-uniqueness of fracture traces some plots of cumulative percentages of branches (continuous
In a network of interconnected fractures, it is often difficult to lines) and traces (dashed lines) lengths of faults from an extensive
recognize a unique fracture trace that connects two terminating sub-marine platform of Hartland Point (Nixon et al., 2012) and a
nodes (either I-type or Y-type), particularly where there is splaying more restricted coastal platform from Westward Ho! (Nixon et al.,
or branching of fractures (Fig. 6a). On the contrary two adjacent 2011). These data support the idea that branch lengths conform
nodes of any type are always connected by a uniquely identifiable best to a log normal or negative exponential distribution (Fig. 7 a,
branch (Fig. 6b), hence, it is much easier to identify fracture b), whereas fault traces conform better to power-law distributions,
branches than entire fracture traces, and this can nearly always be although the traces from Westward Ho! are still best represented
done in a consistent and reproducible manner. by a log normal distribution, probably due to the censoring of their
lengths (Pickering et al., 1995).

2.4.2. Censoring
In exposures of finite size, many fractures extend beyond the 2.5. Why is topology important?
sampling area (Fig. 6a). This is particularly common when exam-
ining data from wells or boreholes. In such cases, the frequency and Two fracture systems could contain the same geometrical ele-
length of fractures will be subject to a sampling bias, generally ments, defined in terms of the orientation and trace length, but
referred to as censoring. A detailed discussion of this problem may have very different topologies. Fig. 8 illustrates a simple example of
be found in many engineering geology publications (e.g. Priest, this. Network A contains two sets of fractures at a high angle to one
1993; Pickering et al. 1995; Riley, 2005). By definition, the length another and arranged with dominantly abutting contacts, as is
of a branch is less than or equal to that of an entire trace, hence the typical of a system of master and cross-joints (e.g. Gross, 1993). We
degree of censoring of the branch lengths will be less than for trace can take the same geometrical elements and place them more
lengths (Fig. 6b). Corrections exist for this censoring effect in the randomly to form network B. The two networks have the same
case of negative exponential and log normal distributions (Priest, geometry, as defined by their number, orientation and mean
1993), but not for power-laws. Thus, correcting the length of frac- length; they thus have the same fracture intensity and character-
ture branches is a more tractable procedure but one we will not istic length, and, hence, the same dimensionless intensity. Clearly
explore here. network A, with a well-developed spanning cluster of fractures, is
much more highly connected than network B. If the fractures were
2.4.3. Branch distributions conductive, network A would be expected to have a much greater
Commonly, fracture trace lengths have been shown to exhibit permeability that B. This illustrates that it is important to include
power-law (e.g. Bour and Davy, 1999; Clark et al., 1999) or negative topology in the characterization of a network.
exponential (Priest and Hudson, 1981) distributions, both of which Another way of looking at this example is that one may have
are difficult to characterize in regions of limited exposure and/or characterized the geometry of the fractures in terms of their ori-
where small fractures may not be consistently sampled (or not entations, lengths, fracture intensity, etc., but that this information
sampled at all). In most fracture systems the longer traces will have is not sufficient to produce a sensible model of the fractures.
the greater number of connected fractures (CL) and, hence, be Network B could be viewed as a stochastic model of the input

Lognormal -ve Exponential Powerlaw


3.0 100.0 100.0

2.0 %N %N
SD
H
1.0 W 10.0 10.0

0.0 W
H H
-1.0 1.0 1.0

-2.0 W

-3.0 0.1 0.1


1 10 100 1000 10000 0 1000 2000 3000 1 10 100 1000 10000
Length (m) Length (m) Length (m)

(a) (b) (c)


Fig. 7. Plots of branch lengths (continuous lines) and trace lengths (dashed lines) of faults from Westward Ho! (W) and off Hartland Point (H). (a) Standard deviation plotted against
log (length) with straight line indicating log normal distribution; note good fit for branches, deviation for longer trace lengths. (b) Log (cumulative percentage) plotted against
length, with straight line indicating negative exponential distribution. (c) Log (cumulative percentage) plotted against log (length) with straight line indicating power-law
distribution.
60 D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66

A - cross-joint network B - stochastic network


I

Y X
(a) (b) (c)

Fig. 8. Comparison of two fracture networks with the same geometry of individual fractures in terms of number, orientation and lengths of the individual factures, but different
topology. Trace-map (a) shows the fractures arranged in a system of cross-joints, (b) has a random (stochastic) placement of the same elements. Nodes are identified as in Figs. 1 and
2. (c) Shows the topology based on the proportions of nodes in (a) and (b).

geometry derived from A, and this sort of realization is easily coded Dimension of Feature
in discrete fracture network models. Clearly such stochastic
0 1 2 3
modelling will fail to capture the topology of the network and,
number length area volume
hence, may fail to model the underlying physical properties of the
system.
P10 P11
Dimension of Sampling Region

line
1
1-D frequency = Dimensionless
3. Measures of fracture abundance and size fracture intensity intensity

In this section, we focus on developing a simple scheme for the


topologically based analysis of fracture networks. The approach P20 P21 P22
area
2

emphasises the use of node counting in the evaluation of the Fracture Dimensionless
2-D
abundance of branches and indicates where some established pa- frequency intensity intensity
rameters need to be interpreted in terms of the topology.
P30 P32 P33
Volume
3

3.1. The Pxy system


3-D Fracture Dimensionless
frequency intensity intensity
Fractures can be sampled in 1-dimension (scan-lines, bore-
holes), 2-dimensions (maps, surface exposures) or 3-dimensions
Fig. 9. Pxy system and terms used for these in this paper. The columns show the di-
(rock volumes). Information on the fractures may simply be their mensions of the features measured; the rows represent the dimension of the sampling
number (or frequency), trace length or surface area. This range of region. The dimensionless intensity is obtained by multiplying the fracture intensity in
information and sampling space has given rise to a wide range of each row by the characteristic fracture length. For further discussion see text.
methodology and associated terms being used to describe fracture
abundance. Much of the methodology was rationalized by
The term fracture intensity is used to express the total trace
Dershowitz and Einstein (1988), and Dershowitz and Herda (1992),
length per unit area, as originally suggested by Dershowitz and
who define a number of measures of fracture abundance by Pxy,
Einstein (1988), see also Rohrbaugh et al. (2002). In this usage
where x denotes the dimension of the sampling region and y the
fracture intensity has dimensions [L1], and the concept is
dimension of the feature being measured (Fig. 9). There is, however,
extended to 1- and 3-dimensions for measures with the same di-
still a lot of inconsistency in the terms applied to these different
mensions (Fig. 9). Note that this term is often referred to as fracture
measures.
density in the geological literature (e.g. Nixon et al., 2012). Spacing
Frequency is used for any sample of the number of fractures (NL),
may be regarded as the inverse of intensity and has an unambig-
and can be normalized by the line length (L), sample area (A) or
uous meaning for fracture networks when expressed in this way.
volume (V) to define:
Spacing as a measure of the distance between fractures only has
h i meaning for a single set of sub-parallel fractures and as such has no
Linear frequency : FL ¼ P10 ¼ NL =L has dimensions L1
clear meaning for random or multiple sets of fractures.
A third term, dimensionless intensity, is reserved for measures
h i
Areal frequency : FA ¼ P20 ¼ NL =A has dimensions L2 that are dimensionless [L0]. Because of the common usage of in-
tensity (as defined above), we will always refer to these dimen-
sionless measures as dimensionless intensity.
Volumetric frequency : FV ¼ P30 In this paper we do not attempt to discuss or justify this ter-
h i minology, but simply seek to adopt a consistent use of three terms
¼ NL =V has dimensions L3
(frequency, intensity and dimensionless intensity) to describe quite
specific aspects of fracture abundance. The term density has been
This follows the original usage of frequency defined by
used by different communities, and at different times, for all three
Dershowitz and Herda (1992) and has been widely applied (e.g.
of these measures, and to avoid this confusion we propose that the
Priest, 1993), but subsequently referred to as fracture density (e.g.
term density be abandoned.
Mauldon, 1994; Rohrbaugh et al., 2002).
D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66 61

Thus in 2-D areas, there are 3 possible measures of fracture the scan-line one obtains the Linear Frequency, which in 1-D is the
abundance: same as the Fracture Intensity (Fig. 9). However, to study fracture
h i networks requires 2- or 3-dimensional sampling. As 3-dimensional
Aerial Frequency P20 ¼ NL =A dimension L2 (7) rock volumes are difficult, often impossible, to sample completely,
much of our characterization and interpretation of fracture net-
h i works depends on 2-dimensional samples, usually in the form of
Fracture intensity P21 ¼ SL=A ¼ NL LC =A dimension L1 trace maps, image logs of well surfaces or slabs from core. The
orientation of the sample plane with respect to the fractures is
(8)
important and needs to be considered carefully in the experimental
. h i design and interpretation of the data (see Section 3.3 and 4). In
Dimensionless intensity P22 ¼ NL L2C A ¼ P21 $LC dimension L0 some cases a 2-dimensional approach may be justified, for example
sampling on bedding planes where fractures are orthogonal to
(9)
bedding, or using dip-sections to sample steeply inclined faults and
fractures with similar strike.
where LC is the characteristic length (LC), and is most simply defined
as the arithmetic mean of the line lengths, hence:
3.2. Parameters for fracture branches
LC ¼ SL=NL (10)
These definitions are interrelated, with values changing by a The parameters P20, P21 and P22 (as defined in the previous
factor of LC as one moves from left to right through each column in section) are all based on determining the number and length of
Fig. 9. It is possible to use aperture rather than LC to form a individual fracture traces. As discussed in Section 2.4, trace lengths
dimensionless intensity, which is essentially equivalent to porosity. are often difficult to measure and subject to bias. For these reasons
Whilst this may have useful applications in fracture characteriza- the concepts of branch length and branch number are introduced
tion, its use in the analysis of topology becomes meaningless. and defined in terms of the network nodes.
In practice, the simplest way to determine P22 and LC is to Branch length (B) is the length between two adjacent nodes that
determine the aerial frequency (P20) and fracture intensity (P21) and are linked by a fracture. Branch number (NB) is the number of such
from Eqs. 7e10, above: branches. The characteristic branch length (BC) is the average branch
length and is best estimated as SLB/NB.
P22 ¼ ðP21 Þ2 =P20 (11) The parameters P20, P21, P22 for lines (above) can be replaced by
their corresponding ‘branch’ values:
and h i
Frequency B20 ¼ NB =A dimension L2 (18)
LC ¼ P21 =P20 ¼ P22 =P21 (12)
h i
In 3-dimensions, there are 3 widely used measures of fracture Intensity B21 ¼ SLB =A ¼ NB BC =A dimension L1 (19)
abundance:
h i . h i
Volumetric Frequency P30 ¼ NL =V dimension L3 (13) Dimensionless intensity B22 ¼ NB B2C A dimension L0

. h i (20)
1
Fracture intensity P32 ¼ SAi =V ¼ NL L2C V dimension L For any network, the sum of the branch lengths and line lengths
(14) are the same. Hence, length and branch determinations of intensity

. h i
Dimensionless intensity P33 ¼ NL L3C V dimension L0
(15)

where LC is a characteristic line length and Ai is the area of indi-


vidual fracture surfaces.
The dimensionless intensity (P33) is related to the fracture in-
tensity (P32) by

P33 ¼ P32 $LC ¼ P31 $L2C ¼ P30 $L3C (15)


Hence:

LC ¼ ðP32 =P30 Þ1=2 (16)

and
.
1=2
P33 ¼ ðP32 Þ3=2 P30 (17)

Fracture attributes can also be sampled in 1-dimension along


scan-lines and this provides the basis for data collection in many Fig. 10. Geometry for the Terzaghi correction in a well or borehole (a) and on a map
applications, particularly in engineering geology and well logging. surface (b). q is the angle between the normal to the fracture and the axis of the sample
By counting the number of fractures and dividing by the length of line or borehole.
62 D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66

are the same (i.e. B21 ¼ P21), but the frequency and dimensionless as illustrated in Fig. 10. The corrected fracture intensity provides an
intensity are different and given by: unbiased estimate of abundance and values determined from 1-, 2-
. dimensional samples should then be equivalent to the 3-
B22 ¼ ðB21 Þ2 B20 (21) dimensional intensity (P32) as discussed in Section 4.
The probability of encountering a fracture is proportional to the
and apparent spacing (S0 ) along a scan-line. This can be related to the
true spacing (S), since S0 ¼ S/cosq (Fig. 10), where q is the angle
BC ¼ B21 =B20 (22) between the pole to the fracture and the scan-line. Orientation bias
is corrected by weighting each fracture by the factor 1/cosq, and has
been used widely in engineering geology, especially in the esti-
3.3. Determination of fracture intensity mation of fracture intensity and spacing. This correction is easily
implemented in 3-dimensions as it is simply the scalar (or dot)
The parameter most easily and uniquely defined from field product of two unit vectors, cos q ¼ n.a, where n is normal to the
measurement of a fracture network is the fracture intensity, with fracture and a parallel to the well axis.
dimensions [L1]. This is usually determined from the number of The Terzaghi correction is often applied to fracture data from
fractures intersecting a line traverse (a form of node counting) or wellbores, usually by assuming that the fractures completely
measurement of the total line length within a trace map area. In intersect the cylindrical volume of the core or wellbore. This is only
both cases, the raw measurements of intensity are biased by the valid where the fracture is large in relation to the well diameter and
orientation of the fractures relative to the sample line or plane. A breaks down for small fractures and those sub-parallel to the
correction procedure, following Terzaghi (1965), should be adopted wellbore axis. This is important for power-law size distributions,

Fig. 11. Example of fracture traces in relation to a circular sample area of diameter 1 unit. (a) Full extent of fractures intersecting the circle; (b) trace length of fractures inside circle;
(c) intersections of fracture traces with circumference of circle; (d) basis of correction factor for a set of equally-spaced parallel lines; (e) barycentres of fracture traces; (f) tips of
fractures within circle.
D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66 63

particularly with high scaling exponents and, hence high pro- overestimate of the frequency and such overestimates are a com-
portions of small fractures. These effects have been analysed by mon feature of “trace counting”. The problem is that if the circle
Davy et al. (2006), who derive a more sophisticated correction represented the physical bounds of our sample (say a core or well
procedure. image), we would not be able to locate the centres and determine
Trace lengths from planar maps can also be corrected in a the correct frequency. One solution is to count the number of
similar way by using the angle (q) between the normal to the fracture tips (i.e. I- and Y enodes; Fig. 11f) within the circle
fracture and sampling plane. For a horizontal map, where d is the (NT ¼ 15) and from this we can use NL ¼ ½ NT (as in Eq. (1)) to
dip of the fracture, the correction is 1/sind, since q ¼ 90  d. estimate the probable number of fractures (Mauldon et al., 2001;
A useful strategy, that also illustrates the value of node counting, Rohrbaugh et al., 2002). Using NT ¼ 15, gives NL ¼ 7.5, which
is to use circular scan-lines for trace maps (Mauldon, 1994; agrees well with the observed 8 centres.
Mauldon et al., 2001; Rohrbaugh et al., 2002, Fig. 11). A circle can Using SL ¼ 4.8, this gives an estimate of the characteristic trace
be considered as representing a line that has all orientations length LC ¼ 4.8/7.5 ¼ 0.64, which looks reasonable from Fig. 11a. In
equally represented (in 2-dimensions) and, hence, the potential to general, this strategy works well if LC is less than the minimum
provide less biased samples than straight lines. The procedure is to distance across the sample area (for equant shaped areas this is
count NE, the number of intersections of fractures with a circle of zA½). Having estimated the intensity and characteristic length, we
diameter D (Fig. 11c). The intensity (P10) is then calculated as are now in a position to determine the dimensionless intensity
from Eq. (9), giving P22 ¼ P21*LC z 3.9.
P10 ¼ ðNE =circumferenceÞ$p=2 (23)
Since the circumference ¼ p D, Eq. (24) simplifies to 4. Discussion e application to well data

P10 ¼ 1=2ðNE =DÞ (24) In this section we outline how topology-based characterization
The factor p/2 in Eq. (23), represents an average correction can be applied to core and image data from wells and boreholes. For
derived as follows (Fig. 11d). Consider equally-spaced lines parallel some fracture attributes, such as orientation and thickness (aper-
to x-axis that are sampled along a circular scan-line of radius r ¼ D/ ture) it is relatively easy to produce estimates from core or image
2. The number of lines intersected (NE) is given by: logs, for others, such as size and connectivity, it is extremely diffi-
cult, especially where the length-scale of the fractures is much
NE ¼ 4r=s ¼ 2D=s (25) greater than the diameter of the well. Some procedures exist that
allow estimation of the length-scale of fractures based on the
0 ) is
where s is true spacing. The apparent intensity (P10 relative frequency of complete fracture traces, average inclination
to well axis and well diameter (Ozkaya, 2003). In all cases we also
0
P10 ¼ NE =circumference ¼ ð4r=sÞ=ð2prÞ ¼ 1=s$2=p ¼ P10 $2=p need to consider the bias introduced by the orientation of the well.

Hence P10 ¼ P100 . p/2, where p/2 is the factor in Eq. (23), and this

leads to Eq. (25). Mauldon et al. (2001) give a more rigorous proof 4.1. Abundance and orientation measures in a well
and show that Eq. (24) gives an unbiased estimate of intensity.
A well or borehole provides a range of sampling opportunities
(Fig. 12):
3.4. Determination of frequency, branch length and dimensionless
intensity a) 1-dimensional samples treat the well as a line that intersects
fractures within the rock volume (Fig. 12a). By counting the
As defined in this paper, the frequency, intensity and dimen- number of fractures (N) and dividing by line length (L) we
sionless intensity are related to one another by the characteristic determine the 1-dimensional intensity (P10). If we can deter-
length, either through the trace length (Eqs. (7)e(18)) or the branch mine the orientation of the fractures relative to the well axis (q)
length (Eqs. (18)e(22)). One way to determine frequency and we may apply the Terzaghi correction, hence P10 ¼ N/(L cosq).
dimensionless intensity would be to determine the characteristic b) 2-dimensional samples examine surfaces within the well, e.g.
trace length (LC) or branch length (BC). This is often impractical due the wall of the wellbore, or the outer surface or slabbed section
to the effects of censoring, although BC than LC. of a core. Fig. 12b shows the situation for a slabbed core where
An alternative is to try to produce an unbiased estimate of fre- we are able to measure the trace length (T ¼ D/cosq) and
quency and then to use this, together with the intensity (B21 ¼ P21), determine P21 ¼ N/(L cosq). Note that the value of P21 ¼ P10 if the
to calculate the remaining parameters. Since the branch parameters Terzaghi corrections are applied.
are determined by the nodes (as in Section 2), it follows that the c) 3-dimensional samples, where we have the opportunity to
number of lines (NL) and branches (NB) can be evaluated in terms of extract information about surface areas and volumes (i.e. mea-
the number (NI, NY, NX) of different node types. sure P32, P30 directly). The recent use of techniques like CT-
Consider the situation of some fractures in a circular sample area scanning has opened up new observational possibilities of this
(Fig. 11a) of diameter D ¼ 1, hence area A ¼ p/4. There are 12 frac- type. From Fig. 12c, fracture area ¼ N (1/4 p T D) and the volume
tures within the circle, although many of these pass out of it of the cylinder ¼ 1/4 p D2. Since T ¼ D/cosq, P32 ¼ N/(L cosq), as in
(Fig. 11a). If we use NL ¼ 12, the frequency will be overestimated. The the 1- and 2- dimensional analysis.
total line length within the circle can be determined unambiguously
as SL z 4.8 (Fig. 11b), hence, the fracture intensity P21 ¼ 4.8/(p/ Thus, the Terzaghi correction produces robust measures of
4) ¼ 6.1. We could also estimate this from the intersections with the fracture intensity that are not dependent on the dimensionality of
circumference of the circle (Fig. 11c, as in Section 3.3). the sampling space. In Fig. 12, only planes (fractures) that cut across
Fig. 11e shows the centres of the traces and there are 8 within the cylindrical volume of the well or core are considered and we
the circle. Hence the true value of NL ¼ 8, and this is the value that can obtain geometrical information on intensity and orientation.
should be used to calculate the frequency (P20 ¼ 8/A z 10.2). In this The measures that are usually missing are those of fracture fre-
case, counting lines that leave the circle produces a 50% quency and characteristic length.
64 D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66

Fig. 12. Comparison of relationships between intensity estimates in 1-, 2- and 3-dimensions from a well-bore.

4.2. Topology from a well can then be used to estimate the characteristic line and branch
lengths (LC ¼ 1.4 m and BC ¼ 0.23 m). These values make sense since
Access to surfaces and volumes allows us to see more of the only two of the 5 fractures have tips in the well volume whereas
relationships between fractures, and hence to access their topology. there are 4 intersections, suggesting branch lengths on the order of
Obviously if the length scale of the fractures is much greater than the well diameter (0.2 m). This illustrates how the branch/node
the diameter of the well this is still a very limited view. Intersection topology can be utilized to extract more information on the fracture
of a fault network by a few wells, provides information on the in- network. The topology itself suggests that the network may well be
dividual faults but little insight into the nature of the fault network connected since there is a dominance of X and Y nodes, but further
as a whole. There are, however, many situations where wells analysis is needed to confirm this.
intersect fracture systems on a scale where it is possible to make
some observations on the nature of the branches and nodes.
Fig. 13 shows a hypothetical example of an interpreted image log 5. Conclusions
with 5 fractures, belonging to two orientation sets e one dipping
steeply ENE the other gently to the SSW. Three of the fractures cut Fracture networks consist of a system of fractures developed
entirely across the well (B, C, D), one tips in the well (E) and the within the same rock volume. In two-dimensions, a system of
other (A) abuts against fracture B. Fracture C cross-cuts both B and branches and nodes can be used to define both the length and
D. Thus there are four nodes (one I-node, one Y-node and two X- orientation of the fractures e two essential features of their ge-
nodes). The resulting topology plots in the lower-right part of the ometry. Branch lengths are preferred to trace lengths as they can be
IYX triangle. The axis of the well intersects all 5 fractures, producing more uniquely defined, have less censoring and are more clustered
an uncorrected estimate of P10 ¼ 5 m1. Application of the Terzaghi around a mean value. It is relatively easy to convert branch mea-
correction gives an estimate of fracture intensity P32 z 7 m1, sures into trace lengths as the topology defines the ratio of the
which can be confirmed by measuring the areas of the ellipses and number of branches to lines.
part-ellipses in Fig. 12c. If required one could apply the Terzaghi Topology essentially describes the relationship between frac-
correction to the orientation data to improve estimates of tures in a network. It has two main impacts on fracture charac-
orientation. terization. Firstly, node counting and branch analysis can help
Counting the nodes and tips (as in Section 2) produces estimates determine the orientation, frequency and intensity of fractures.
of the number of fractures as 1 and number of branches as 6. These Secondly, the proportions of different types of nodes provide a basis
D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66 65

Fig. 13. An example of the extraction of topology from an image log. (a) Interpreted image log showing extracted nodes and branches; (b) 3-D reconstruction of fractures; (c) equal-
area stereogram showing 2 fracture sets; (d) classification of nodal topology.

for describing the topology that can be easily applied even with topology being essential to the interpretation of characteristics
limited access to the network as a whole. such as connectivity. In this paper we emphasise the practicality of
Plots of the proportions of I-, Y- and X-nodes provide a simple capturing aspects of topology, even in cases, such as in a well or
summary of some aspects of the topology and are useful in dis- borehole, where it is not possible to image the network as a whole.
tinguishing different networks. The plot describes topology rather
than geometry since it is unchanged by any continuous trans-
Acknowledgements
formation of the network. The nodes can be used to classify
branches into three types e IeI, IeC and CeC branches. Again this
We would like to thank Tom Manzocchi, Jon Bull, Tom Blen-
provides a useful classification of network topology, but the main
kinsop, Steve Dee and Robert Humphreys for discussion of various
value may be that the average number of connections per branch
aspects of this work. C.W. Nixon acknowledges financial support
(CB) appears to provide a measure of connectivity that is almost
from NERC Case studentship (NE/H524922/1) with BP. We are
completely independent of the topology, although this needs to be
grateful for reviews by Tom Blenkinsop and David Healy that hel-
examined further.
ped clarify parts of the paper.
Node counting provides an unbiased estimate of frequency and
can be used in conjunction with fracture intensity to estimate the
characteristic length and dimensionless intensity of the fractures. References
The latter provides a measure of fracture development that is in-
Adler, P.M., Thovert, J.F., 1999. Fractures and Fracture Networks. Kluwer Academic
dependent of scale and has many practical applications. The frac- Publishers, Dordrecht.
ture intensity (number of fracture intersections per unit length, line Balberg, I., Binenbaum, N., 1983. Computer study of the percolation threshold in a
length per unit area, surface area per unit volume) is the most two-dimensional anisotropic system of conducting sticks. Phys. Rev. B 28,
fundamental measure of fracture abundance and has units of [L]1.
3799e3812.
Balberg, I., Anderson, C.H., Alexander, S., Wagner, N., 1984. Excluded volume and its
Raw measurements can easily be corrected to provide an unbiased relation to the onset of percolation. Phys. Rev. B 30, 3933e3943.
estimate and such estimates are equivalent for sampling in Berkowitz, B., 1995. Analysis of fracture network connectivity using percolation
theory. Math. Geol. 27, 467e483.
different dimensions.
Boccaletti, S., Latora, V., Moreno, Y., Chavez, M., Hwang, D., 2006. Complex net-
Determining both the geometry and topology of fractures allows works: structure and dynamics. Phys. rep. 424, 175e308.
us to improve the characterization of fracture networks, both in Bour, O., Davy, P., 1999. Clustering and size distributions of fault patterns: theory
terms of the efficiency of the methodology and the range of derived and measurements. Geophys. Res. Lett. 26, 2001e2004.
Clark, R.M., Cox, S.J.D., Laslett, G.M., 1999. Generalizations of power law distribu-
parameters. Conversely, in the modelling of fractures it is important tions applicable to sampled fault-trace lengths: model choice, parameter esti-
to honour both the geometry and topology of the network, with mation and caveats. Geophys. J. Int. 136, 357e372.
66 D.J. Sanderson, C.W. Nixon / Journal of Structural Geology 72 (2015) 55e66

Davy, P., Darcel, C., Bour, O., Munier, R., de Dreuzy, J.R., 2006. A note on the angular Nixon, C.W., Sanderson, D.J., Bull, J.M., 2012. Analysis of a strike-slip fault network
correction applied to fracture intensity profiles along drill core. J. Geophys. Res. using high resolution multibeam bathymetry, offshore NW Devon U.K. Tecto-
111, B11408. http://dx.doi.org/10.1029/2005JB004121. nophysics 541e543, 69e80.
Dershowitz, W.S., Herda, H.H., 1992. Interpretation of fracture spacing and intensity. Ortega, O., Marrett, R., 2000. Prediction of macrofracture properties using micro-
In: Tillerson, J.R., Wawersik, W.R. (Eds.), Proceedings of the 33rd U.S. Sympo- fracture information, Mesaverde Group sandstones, San Juan basin, New
sium on Rock Mechanics. Balkema, Rotterdam, Sante Fe, New Mexico, Mexico. J. Struct. Geol. 22, 571e588.
pp. 757e766. Ozkaya, S.I., 2003. Fracture length estimation from borehole image logs. Math. Geol.
Dershowitz, W.S., Einstein, H.H., 1988. Characterizing rock joint geometry with joint 35, 737e753.
system models. Rock Mech. Rock Eng. 1, 1e51. Pickering, G., Bull, J.M., Sanderson, D.J., 1995. Sampling power-law distributions.
Gross, M.R., 1993. The origin and spacing of cross joints - examples from the Tectonophysics 248, 1e20.
Monterey Formation, Santa Barbara coastline, California. J. Struct. Geol. 15, Priest, S.D., 1993. Discontinuity Analysis for Rock Engineering. Chapman & Hall,
737e751. New York, p. 473.
Jing, L., Stephansson, O., 1997. Network topology and homogenization of fractured Priest, S.D., Hudson, J.A., 1981. Estimation of discontinuity spacing and trace length
rocks. In: Jamtveit, B., Yardley, B.W. (Eds.), Fluid Flow and Transport in Rocks. using scanline surveys. Int. J. Rock Mech. Min. Sci. 18, 183e197.
Mechanisms and Effects. Chapman & Hall, Oxford, pp. 191e202. Ravasz, E., Barabasi, A.-L., 2003. Hierarchical organization in complex networks.
Latora, V., Marchiori, M., 2002. Is the Boston subway a small-world network? Phys. Phys. Rev. E 67, 1e7.
A Stat. Mech. Appl. 314, 109e113. Riley, M.S., 2005. Fracture trace length and number distributions from fracture
Ma €kel, G.H., 2007. The modelling of fractured reservoirs: constraints and potential mapping. J. Geophys. Res. 110, B08414. http://dx.doi.org/10.1029/2004JB003164.
for fracture network geometry and hydraulics analysis. Geol. Soc. Lond. 292, Rohrbaugh, M.B., Dunne, W.M., Mauldon, M., 2002. Estimating fracture trace in-
375e403. Special Publications. tensity, density, and mean length using circular scan lines and windows. Bull.
Manzocchi, T., 2002. The connectivity of two dimensional networks of spatially Am. Assoc. Petrol. Geol. 86, 2087e2102.
correlated fractures. Water Resour. Res. 38 http://dx.doi.org/10.1029/ Sanderson, D.J., Zhang, X., 1999. Critical stress localization of flow associated with
2000WR000180. deformation of well-fractured rock masses, with implications for mineral de-
Mauldon, M., 1994. Intersection probabilities of impersistent joints. Int. J. Rock posits. In: McCaffrey, K., Lonergan, L., Wilkinson, J.J. (Eds.), Fractures, Fluid Flow
Mech. Min. Sci. 31, 107e115. and Mineralisation, Geological Society Special Publication, vol. 155, pp. 69e81.
Mauldon, M., Dunne, W.M., Rohrbaugh, M.B., 2001. Circular scanlines and circular Terzaghi, R.D. Sources of error in joint surveys, Geotechnique 15, 287e304.
windows: new tools for characterizing the geometry of fracture traces. J. Struct. Zimmerman, R., Main, I., 2003. Hydromechanical behavior of fractured rocks. In:
Geol. 23, 247e258. Gueguen, Y., Bouteca, M. (Eds.), Mechanics of Fluid Saturated Rocks, Interna-
Nixon, C.W., Sanderson, D.J., Bull, J.M., 2011. Deformation within a strike-slip fault tional Geophysics Series, vol. 89, pp. 361e419.
network at Westward Ho!, Devon U.K.: domino vs conjugate faulting. J. Struct.
Geol. 33, 833e843.

You might also like