Engineering Geology 116 (2010) 166–177

Contents lists available at ScienceDirect

Engineering Geology
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e n g g e o

Application of fracture network model with crack permeability tensor on flow and
transport in fractured rock
Jian-Bang Pan, Chen-Chang Lee, Cheng-Haw Lee ⁎, Hsin-Fu Yeh, Hung-I Lin
Department of Resources Engineering, National Cheng Kung University, Tainan, Taiwan 701

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

Article history: Very efficient and practical models that use a crack tensor in terms of fracture geometry parameters for field
Received 2 March 2010 rock mass are proposed. A framework of the crack tensor approach combined with the identifying fracture-
Received in revised form 19 July 2010 controlled characteristics was successively used in a hypothetical hydro-geological field for determining the
Accepted 4 August 2010
dynamics of flow and transport of nuclides in fractured rock.
Available online 20 August 2010
The modeling results and site-specific data of Lan-Yu Island in Taiwan reveal the characteristics associated
with the geometry of fracture networks under the prescribed boundaries. The simulated fracture networks
Keywords:
Fractures
were transformed into a grid-based simulation procedure under consideration of the delaying effect in
Fracture network model matrix owing to particle migration. Synthetic flow results indicate that the flowing vector is very sensitive to
Crack permeability tensor fracture intensity (P32); high fracture intensity leads to high velocity. Heterogeneity generated by fractures'
Ingrowth-decay transport field will result in the shortest and the longest transmission distance of 12-fold difference. While taking into
Lan-Yu island consideration of the ingrowth-decay transport, the plumes of the daughter product were relatively tortuous
Taiwan and anomalous to predict.
Crown Copyright © 2010 Published by Elsevier B.V. All rights reserved.

1. Introduction other blocks by fractures with various hydraulic behaviors. Thus, the
flow through such a complex system depends on the magnitudes of
Fractured formation, which is the most common and general the permeability in each of the two regimes. However, the
material in rocks, makes it difficult for hydro-geologists, geologists, permeability of the intact rock is relatively low; fractures are regarded
engineers, and scientists to strictly quantify and characterize the as more permeable than the rock matrix by several orders of
flowing mechanism induced by spatial heterogeneity (Berkowitz, magnitude. Consequently, fractures are the major path for fluids to
2002; Neuman, 2005). Geomedia disposal is viewed as the best flow (Priest, 1993; Lee and Farmer, 1993; National Research Council,
method for disposing of radioactive waste. One of the most serious 2001; Berkowitz, 2002; Neuman, 2005).
problems is the isolation of the radioactive waste from the biosphere. When classified according to how heterogeneity is incorporated in
Groundwater flow through geological discontinuities is believed to be the model structure, fluid flow and solute transport fall into one of
the most significant mechanism of radionuclide migration. Theoret- three groups: discrete network models, equivalent continuum
ical models and experimental studies have contributed to the models, and hybrid models (National Research Council, 2001). It
understanding of flow and solute migration fields within fracture could be done via standard approaches such as by increasing the
planes, including the characterization of aperture variation, intersec- permeability of the porous medium to account for the fracture system
tions among various phases, and the transfer between fractures and (or double-porosity models), or by using discrete stochastic fracture
the rock matrix. network models.
The analysis of subsurface fluid flow and mass transport is often In the discrete fracture network model, such as Baecher's model
complicated by the fact that rock masses, especially at shallow depths, (Baecher et al., 1977; Dershowitz, 1984; Dershowitz and Einstein,
may contain various systems of cracks or failure surfaces. The rock 1988; Ivanova, 1998) each fracture is placed in a space centered on
mass can therefore be regarded as an assemblage of intact rock blocks Poisson points and oriented randomly according a specified distribu-
that are separated by discontinuities. These discontinuities may be tion. It is then extended to define a stochastic process for the
faults, joints, fissures, or fractures; the word “fracture” is used here as termination of the fracture shape and size. The usual process is to
the general term. In hydraulic modeling, each solid block can be begin with unbounded Poisson planes and to superimposed on them a
considered as a continuous porous medium that is separated from the series of Poisson lines or the lines of intersection between the
unbounded Poisson planes; this is called the enhanced Baecher model
(Dershowitz, 1984; Dershowitz and Einstein, 1988; Ivanova, 1998).
⁎ Corresponding author. Tel.: + 886 6 2757575x62833; fax: + 886 6 2380421. The main advantage of the discrete fracture model is that it allows the
E-mail address: [email protected] (C.-H. Lee). inspection of the influence of individual fracture parameters on flow.

0013-7952/$ – see front matter. Crown Copyright © 2010 Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.enggeo.2010.08.007
 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177 167

Its disadvantage is that the detailed definition of fracture geometry 2. Methodology

and the hydraulic properties of individual fractures are uncertain.
Researchers more often consider the orthogonal fracture network due 2.1. Study approach
to its simplicity. In the discrete channel model is analogous to the
discrete fracture model except that the fractures are viewed as The crack tensor approach is commonly used in geotechnical,
channels. In the continuum orthogonal fracture mode, the fracture environmental, and civil engineering fields, especially in the estima-
parameters are assumed to behave macroscopically. The formation tion of permeability or conductivity under a rock mass (Wang et al.,
has either isotropic or non-isotropic properties in homogenous media. 2002; Bluma et al., 2005). Firstly, the discrete fracture model counted
The phenomena (such as permeability tensor or crack tensor in the on a statistical characterization of the rock mass (Lee et al., 2007). The
paper) were represented correctly through the continuum equivalent generation of the fracture network was based on realistic distributions
so that three principal values of phenomena properties were coaxial of geometric parameters, including fracture orientation, size, inten-
to orthogonal axial. Long et al. (1982) suggested that if a sufficient sity, and location, as well as hydro-geological parameters, including
number of fractures exist within a rock mass, then the rock mass transmissivity and transport aperture (Lee et al., 1996). A three-
behavior can be simulated using an equivalent porous medium. dimensional fracture network model was generated using the
Oda (1985) proposed a model for establishing the fracture stochastic realization of polygonal fractures when the crack tensor
network to predict rock mass permeability. The fractures are regarded approach (Oda, 1985) was applied to assemble the fracture clusters
as a series of penny-shaped cracks. The crack geometry can generally from the discrete fracture network model. The numerical heat and
be regarded as a function of fracture orientation, fracture trace length, mass transport tool (Zyvoloski et al., 1988; Zyvoloski et al., 2008) was
and fracture aperture. The permeability tensor is viewed as a unique used to model the flow under the prescribed head condition within
function of the crack tensor, and permeability principal axes are the particle tracking for the spatial distribution of particles. Finally,
viewed as being coaxial with those of the crack tensors. In situ the spatial and temporal concentrations of decay nuclides were
measurements of the fracture size and aperture are difficult. The investigated to assess possible contaminated zones.
fracture size may be obtained by mapping and measuring the fracture
trace length, but this method is time-consuming. In addition, the 2.2. Calculation of permeability and crack tensor
fracture trace length depends on the shape and size of the fracture, the
angle between the fracture plane and outcrop plane, and the The presence of cracks can induce anisotropy in the permeability
measurement scale of the sampling window. Hence, learning how tensor for a cracked soil or rock. Cracks or fractures produce a high
to estimate the representative values of fracture geometric para- permeability in the predominant direction of cracking. The coefficient
meters (fracture orientation, size, and aperture) is important. It is the of permeability of the cracked soil or rocks in the direction normal to
following work to describe the network formed by fractures. Several predominant cracking will be small. The permeability's spatial
researchers have used fracture continuum (FC) models to numerically variability and anisotropy has attracted much attention to fractured
simulate flow and transport in fractured media (Neuman, 1987; Wide´n rock and petroleum studies (e.g., Snow, 1969; Hsieh et al., 1985;
and Walker, 1999; Svensson, 2001a,b; Ando et al., 2003; Langevin, Teimoori et al., 2005).
2003; Pohlmann et al., 2004; McKenna and Reeves, 2006; Reeves If a fractured rock is treated as a homogeneous, anisotropic porous
et al., 2008). Instead of representing discrete fractures as continuous medium, the permeability tensor, obeying Darcy's law, can be
line elements in 2-D or planes in 3-D, FC models are based on the formulated as (Bear, 1972):
conversion of discrete fractures, or more commonly fracture zones, to
permeability structures on a model grid. This conceptualization is g ∂ϕ g
supported by field observations that rock volumes are often intersected Vi = − Kij = Kij Jj ð1Þ
μ ∂xj μ
by a few dominant fractures (Neuman, 2005). Often, the choice of
using the discrete fracture network (DFN) or fracture continuum (FC)
model depends on scale; DFN models are favored at smaller scales where Vi is the apparent flow velocity, Jj is the hydraulic gradient, and
where they remain computationally feasible and FC models are used Kij is the permeability tensor.
for larger-scale models. Comparisons between DFN and FC models If a fractured rock mass is assumed to have an impermeable matrix
have shown that both methods are equally capable of capturing key and groundwater is assumed to only flow through the fractures, then
aspects of flow and transport in fractured crystalline rock masses the apparent flow velocity can be defined as:
(Svensson, 2001b; Selroos et al., 2002; Ando et al., 2003). The
assembled procedures to do such a task is called fracture network ð cÞ dV ðcÞ
model. After the model is built, it is coupled to study permeability. In the Vi = ∫ ðcÞ Vi ð2Þ
V Vv
present study, the focus is to determine hydraulic properties and also to
realize in flow and transport to deal with different application
where V (c) is the volume of the fractures and VI(c) is the local velocity
circumstances.
for each fracture. If the fracture is considered as a parallel plate model
A framework from field data is integrated with a numerically
with an aperture e and groundwater moves by laminar flow in the
spatial distribution of fracture parameters in an equivalent fracture
fractures, obeying the cubic law, then the mean velocity VI(c) can be
network model which is combined with discrete fracture networks
formulated as (Snow, 1969):
and the concept of fracture tensors in a fractured rock mass. Based on
the crack patterns and the statistical geometric characteristics of the
fractured rock, the proposed model can be used to estimate the crack ðcÞ 1 g 2 ð cÞ
Vi = e J : ð3Þ
tensor and to predict the permeability tensor. The study area, a 12 μ i
potential disposal site of low-level and intermediate-level radioactive
waste at Lan-Yu island (Fig. 1) composed of moderately fractured Each fracture has a void volume equal to (π/4)r2e. The total void
andesite (Deng et al., 1995) is utilized on performance in this idea. A volume dV(c) associated with the fracture geometry is given by:
hypothetical hydro-geological field is proposed to quantify and
characterize the behavior of flow and transport of parent–daughter
2
radionuclide for the assessment of transport behavior under this ð cÞ πr e
dv = dN: ð4Þ
framework. 4
168 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177

Fig. 1. Location of Lan-Yu island and the sampling area.

(Satellite picture obtained from Google Earth).

Substituting Eqs. (1), (5), and (4) into Eqs.(2) and (3) produces an where ρ is the volume density of fractures, defined as m(v)/Vv.
equivalent permeability tensor Kij for the fracture system: Assuming that the random variables f, a, and e are statistically
independent of one another, E(f,a,e) = E(f) E(a) E(e), where E(f), E(a),
1

and E(e) are the density functions of f, a, and e, respectively, then:
Kij = Pkk δij −Pij : ð5Þ
12
em 3 am 2
In this case, Pij and Kij are both symmetric, having principal values Pij = ρ∫ e EðeÞde∫ a EðaÞda∫ fi fj Eð f Þdω: ð9Þ
0 0 w
in the principal directions. Since Kij is a function of Pij, its principal axes
are coaxial with those of Pij, with the major permeability in the minor
Since fractures generally occur in clusters or sets, the density
principal direction of Pij. If i = j in Eq. (5), the scalar hydraulic
function of E(f) for each set can be obtained in the form of a possible
conductivity Ks is given by:
distribution with the goodness-of-test on fracture orientations as
g 1 described above. However, it is difficult to obtain reliable data for the
Ks = P : ð6Þ density functions E(e) and E(a) for each set. It is a time-consuming
3μ 12 ii
work, especially for measuring the fracture apertures in situ (Snow,
1969). To overcome this difficulty, fracture aperture in the present
In this study, the rock masses are assumed to be composed of a
study is estimated from laboratory tests; the value is constant under a
series of fractures and an impermeable matrix. The crack geometry
fixed normal stress. Since the fracture orientation f, fracture size
can generally be regarded as a function of fracture orientation f,
(area) a, and fracture aperture e are independent, Eq. (9) can be given
fracture size (area) a, and fracture aperture e. In addition, the unit
as:
vector f is normal to the fractures that are orientated inside a small
solid angle dω around f and the diameters and the apertures range
Nf Nf
from a to a + da and from e to e + de, respectively. E(f,a,e)dωdade k 3 k k
Pij = ∑ Pij = ∑ e ρ〈a〉 N ij ð10Þ
is the probability of the unit normals of (f,a,e) fractures. In this case, k=1 k=1
E(f,a,e) is defined over the entire solid angle ω. If m(v) fractures are
located within the flow region of volume Vv, the frequency of the 2 3
2
probability of (f,a,e) fractures whose centers are located within Vv, ∑l ∑lm ∑ ln 1
Nij = ∫ ni nj EðnÞdω = 4 ∑ml ∑mn 5
k
∑m 2
ð11Þ
then let dN be a number of (f,a,e) fractures, to estimate the total of ω Rk
∑nl ∑nm ∑n2
fractures,

ðvÞ am
dN = m Eð f ; a; eÞdωdade: ð7Þ k
〈a〉 = ∫0 aEðaÞda ðk = 1; 2; …nÞ ð12Þ

The crack tensor Pij can be written as:

where Rk is the magnitude of the directional cosine for the kth set. Nf is
em am 3 the number of fracture sets. Nij or Nkij is a symmetric, second-rank
Pij = ρ∫ ∫ ∫ ae fi fj Eð f ; a; eÞdωdade ð8Þ
0 0 ω tensor which can be calculated from density function E(n).
 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177 169

3. Parameter estimation of fractures Since the field data of the fracture orientation are generally scattered
due to geological complexity, the deviation must be reduced to estimate
3.1. Mean orientation of fracture sets the solution of true frequency of each fracture set. Karzulovic and
Goodman (1985) established a method with a least square technique to
Fractures generally occur as fracture sets or clusters. The mean obtain the true frequency of fracture sets of the rock mass. In their
orientation of each cluster is required to estimate various engineering model, when there are Sf non-parallel scanlines, it is possible to obtain
problems. Determining the pattern of a fractured network is an the true frequency of each of the Nf fracture sets. Their approach
essential task for evaluating the contaminant transport in fractured produced the following result, expressed in matrix form:
rock. The statistical description of orientation data in the past was
often expressed as orientations of fracture values of normal in terms ½A½λi  = ½B ð15Þ
of spherical co-ordinates (θ, ϕ, 1) or direction cosines (l, m, n). The
where [A] is a symmetrical matrix with Nf by Nf. The term Akl is defined
mean orientation of each cluster is then taken as the orientation of the
as the value of [A] in row k and column l and is given by:
resultant vector R, which is determined by adding all of the fracture
poles: Sf
Akl = ∑ cos δkl cos δlj ð16Þ
h i j=1
2 2 2
R = ∑l + ∑m + ∑n ð13Þ
where Sf are non-parallel scanlines, δkj and δlj are acute angles
between the j-th sampling line and normal to sets k and l, respectively.
where l = sinϕcosθ, m = sinϕ sinθ, and n = cosϕ. The angles (θ and ϕ)
The matrix [B] has one column and A rows. The term Lk in row k of [B]
are related to the dip angle φ, and the dip direction α, by θ = 2π − α
is given by:
(0 ≤ θ ≤ 2π) and ϕ = φ (0 ≤ θ ≤ π/2). It should be noted that, before
Eq. (13) can be used in estimating the mean orientation for each set, Sf
poles of fractures for each set must be identified. Bk = ∑ λj cos δkj ð17Þ
Therefore, a delineation of the fracture data is necessary before the j=1

mean direction of each fracture set can be calculated. Methods for

where λj is the apparent frequency of the j-th scanline. The solution
delineating fracture sets were developed by Shanley and Mahtab
vector [λi] in Eq. (15) can is obtained from Eq. (17) by applying an
(1976) (using one-level mode analysis) and by Miller (1983) (using
appropriate numerical solution for simultaneous equations.
contingency table analysis). Both models set up the sphere of radius r
Based on the above process, the true frequency of each fracture set
at each point (each pole of the fracture) on a stereonet. All of the
can be estimated. This method can be used if the mean direction of
centers of the sphere containing κ or more observations are labeled as
each set is known. It is possible to use more scanline surveys than the
“dense points.” Finally, all of the non-dense points are added to
number of fracture sets in situ. If the true frequency of each fracture
adjoining points until every point is part of the cluster. Parameters r
set is determined, then the mean spacing of each fracture set (Si) can
(radius or degree) and κ (center density) are found by optimizing the
be obtained using:
objective function in order to illustrate the clustering system. The
value of r is generally selected to be between 10° and 20°. The 1
advantage of the objective function is that, within a site, all data in a Si = : ð18Þ
λi
cluster can be analyzed using a rational and standardized system.
However, clustering is generally related to the method of objective The P32 of a rock mass is used to define the fracture intensity. P32 is
functions used as various means of clustering fractures as ways defined as the total area of fractures per unit volume.
to represent fracture orientation distribution graphically. Several
models, such as Fisher distribution or Bingham distribution, can Af
provide better fits for asymmetric and girdle orientation data. The P32 = ð19Þ
Vt
Fisher distribution is a valuable model for fitting fracture orientation
data due to its simplicity and flexibility (Priest, 1993). More detailed where Af is the total area of the fractures and Vt is the total volume of the
explanations of the corresponding procedures of objection functions region being measured. P32 is dimensional (in units of 1/m), but is scale
for delineating a cluster set of fracture orientations can be found in our independent and invariant with respect to the distribution of fracture
previous research (Lee et al., 1995). Knowing the fracture sets can size. It can therefore be used to classify a given rock mass and hence is the
determine the dominant orientation of fractures which is very useful preferred measure of fracture intensity (Dershowitz and Einstein, 1988).
for planning and designing the principal direction of subsurface And knowing the fracture intensity is aimed for the vulnerability of rock
structures. mass and preventing some important structures form these areas.

4. Fracture frequency, mean spacing, and fracture intensity 5. Fracture aperture

When the fracture orientation is delineated as sets or clusters, the Numerical models have been developed to describe aperture
mean fracture orientation for each set can be estimated from the poles deformation (Barton et al., 1985). A hyperbolic function was suggested
of each set, as shown in Eq. (13). In order to estimate the principle for describing the relationship between stress and closure or opening of
frequency for each set, Priest and Hudson (1981) showed that the fractures in terms of joint wall compressive strength (JCS), joint
fracture frequency λθ along a scanline can be represented as: roughness coefficient (JRC), and normal stress. Barton et al. (1985)
showed that an empirical relationship between the mechanical aperture
Nf em and the hydraulic aperture ec can be expressed in empirical form as:
λθ = ∑ λi cos θi ð14Þ
i=1
e2m
ec = ð20Þ
JRC 2:5
where λi is the true frequency of the i-th fracture set along the
direction of its normal, θi is the acute angle between the scanline and where em and ec are in micro-meters and JRC is in a range of 0 (smoothest
the mean normal to the i-th set, and Nf is the number of fracture sets. surface) to 20 (roughest surface). Eq. (20) is valid only if ec ≤ em.
170 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177

Since the effective stress is normal to the plane of the fracture, the aperture, and transmissivity variation (2) and the arrangement of
amount of normal closure primarily depends on the mechanical the fractures in space, defined by the type of point process their
aperture em and the roughness of the fracture surface represented by locations followed (Baecher model), and the way they truncate each
JRC and JCS. JCS is the unconfined compressive strength of wall rock other. When dealing with a three-dimensional model, many of these
estimated from Schmidt hammer test. JRC can also be easily estimated choices must be made somewhat arbitrarily because only 1- or 2-D
from tilt tests, preferably with direct measurement (Barton et al., 1985), samples were taken from the 3-D system of the rock mass. A detail
using: description and analysis of fracture parameter sampling and variation
can be found elsewhere (Dershowitz, 1984; Dershowitz and Einstein,
αo −ϕor 1988). For all models, fracture termination is specified by the
JRC = 
logð JCSÞ
ð21Þ
σ no
termination probability PT, which is the probability that a fracture
will terminate, given an intersection with another fracture.
where σno = γhucos2α, γ is the rock density, hu is the thickness of the
upper block, ϕor is the residual friction angle of intact rock, and α is the NT
PT = ð24Þ
tilt angle. The empirical value em is given by (Bandis et al., 1983): NFi
 
JRC σ where NT is the number of T intersections for fractures in the present
em = 0:2 c −0:1 ð22Þ
5 JCS set with fractures from previously generated sets, and NFi is the
number of fractures in the set having intersections with fractures from
where σc is the uniaxial compressive strength. Substituting Eq.(22) previously generated sets.
into Eq.(20) gives the hydraulic aperture: This form of termination disturbs the distribution of fracture size,
since portions of fractures are “discarded” if they are beyond the
  2 intersection where termination occurs. This is only a problem for high
1 JRC σc
ec = 0:2 −0:1 : ð23Þ termination percentages (Dershowitz et al., 2004). In the present
JRC 2:5 5 JCS
study, the enhanced Baecher's model was used to obtain the more
realistic representation of robustness in fractures at intersections with
Knowing the hydraulic aperture is aimed for the assessment on pre-existing fractures.
flow's influence on each section of the fractured rock.
7. Flow and transport simulation
6. Generation of enhanced Baecher's model
Using a continuum approach (Oda, 1985; Lee et al., 1995), the
The three-dimensional model shown in Figure 2 (using FracMan permeability of fracture networks was translated onto a grid of
code developed by Golder's Associate) was applied in the present equally-sized cells, and FEHM (finite element heat and mass transfer
study. The choice of a stochastic conceptual model for a three- code developed by Los Alamos National Laboratory and could be
dimensional fracture system involved specifying (1) the individual downloaded by signing some agreement with DOE of USA) was used
characteristics of the fracture such as shape, size, orientation, to solve for fluid flow in both the fracture network and the less
permeable matrix assigned to cells not occupied by fractures. This
resulted in a discrete network solution which is a set of linear
equations of pressures at intersections.
FEHM was modeled within a non-isothermal, multiphase flow and
transport code that simulated the flow of water and air and the
transport of heat and solutes in two- and three-dimensional saturated
or partially saturated heterogeneous media. It also included compre-
hensive reactive geochemistry and transport modules, and particle
tracking capability. Fractured media can be modeled using the
continuum approach used in the present study, discrete fracture,
dual porosity, or dual permeability approaches (Zyvoloski et al.,
2008). The conservation of fluid mass can be written as:

∂Amass ̅
+ ∇⋅ f mass + qmass = 0 ð25Þ
∂t

where:

Amass ϕρ
→ →
f mass ρv
Amass fluid mass per unit volume (M/L3)
→
f mass fluid mass flux (M/L3t)
qmass fluid mass source (M/L3t)
ϕ porosity in the system
ρ fluid density (M/L3).

The velocity of a fluid can be expressed using Darcy's law:

 
→ k →
Fig. 2. Example of a three-dimensional stochastic fracture network generated within a v = − ∇ P−ρ g ð26Þ
100 (m) cube. μ
 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177 171

where:

P fluid pressure (M/Lt2)

μ dynamics of viscosity of the fluid (M/Lt)
k permeability (L2)
→
g acceleration due to gravity (L/t2).

The above equations are for constant-density water, so the results

should be converted to a head solution rather than a pressure

solution.

→
Using the reference density ρ = 1000(kg/m3), g = 9:8 m=s2 , and
elevation z (reference to mean sea level), pressure was converted to
head using h = z + P / (ρg). The conservation of the solute equation is
explicitly coupled to the fluid flow field. The conservation equation for a
single component, non-reactive solute in a constant density fluid in a
saturated medium is:

∂As →
= ∇⋅ðρϕDs ∇Cs Þ−∇⋅ fs −qc ð27Þ
∂t

where:

As ϕCsρ
→ →
fs ρCs v
Fig. 3. Pole diagram of fracture orientation at the Lan-Yu site on a Schmidt net. Numbers
As solute mass storage per unit total volume for aqueous indicates the set number.
component liquid concentration
Cs concentration of solute (mole/ M)
Ds dispersion coefficient (L2/t) fracture network was assumed to be 10 m × 10 m × 10 m. The reference
→
fs dvective mass flux of solute parameters are shown in Table 1.
qc solute source or sink (moles/L3t). A general study of enhanced Baecher's model was proposed to
determine the importance of variability on structures based on spatial
intersections of fractures. Simulated within a 10 m cubic-block region
8. Model construction and application of a 1% gradient from upstream to downstream flow was used to
mimic a small portion of andersite rock mass. Generic boundary
A disposal site of intermediate radioactive waste in freshly fractured conditions were chosen to obtain a globally unidirectional flow forced
andesite located in the southeast part of Lan-Yu Island, about 49 miles from one vertical face to the opposite face by assigning a constant
southeast of Taitung City in the southeast part of Taiwan, was used to head boundary condition at the faces, as shown in Figure 4. Other
verify the proposed method (see Fig. 1). Investigations of the fracture boundaries were sealed by a no-flow condition in order to lower the
geometric parameters, such as the fracture sets, the mean orientation, significance of the boundary effects. Then, the crack tensor approach
the frequency, the spacing, and the trace length, were conducted on the combined with discrete fractures (Oda, 1985) was decided with the
fields listed in Table 1. Four sets of fracture orientation were represented topological characteristics of the fracture-clustered area of a specified
at the sampling area shown in Figure 3 (marked in red). Not only the volume, while other volumes not coincident with fractures were
surface survey on geology but also the hydrogeology experiments were treated as a matrix of low permeability on the magnitude of 10− 13
made here (Lee et al., 1996). The transmissivity was estimated as
8.9 × 10− 4 m/s where the upper and lower bound of transmissivity
distributions was 3.81 × 10− 4 m/s and 2.27 × 10− 5 m/s, respectively.
The average hydraulic conductivity of 8.04 × 10− 5 m/s was converted to
a permeability of 1.02 × 10− 10 m2 at 20 °C. The domain in the 3-D

Table 1
Input parameters considered in this study (Chen et al., 2001).

Set-1 Set-2 Set-3 Set-4

(Trend, plunge) (41.3,16.6) (311.2,29.6) (133.9,20.3) (213.2,19.6)

Orientation distribution Fisher Fisher Fisher Fisher
Dispersion 5 5 5 5
Average fracture length (m) 0.355 0.441 0.322 0.308
Fracture length distribution Lognormal Lognormal Lognormal Lognormal
Fracture length standard 0.093 0.159 0.094 0.189
deviation (cm)
Average fracture aperture 1.91E−4 1.91E−4 1.91E−4 1.91E−4
Fracture aperture Lognormal Lognormal Lognormal Lognormal
distribution
Fracture aperture 1.82E−4 1.82E−4 1.82E−4 1.82E−4
standard deviation
Fracture intensity (PVA) 0.356 0.034 0.365 0.850

P32 is the areal intensity variant with respect to the distribution of fracture size Fig. 4. Modeling domains with no-flow conditions on upper, lower, front and back
(Dershowitz, 1984). boundaries where the flow is forced by a gradient of 1%.
172 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177

Fig. 5. Fracture intensity (P32) distribution.

(m2). Finally, the grids were assigned to the mesh and numerically
solved using the finite element tool (FEHM) to get the subsequent
flow dynamics and transport behavior.

9. Results and discussion

Results presented here correspond to the characteristics of fracture

distribution and features of first-order decay particles of the half-year
life at the outlet of the 10 × 10 × 10 m fractured block for the fracture-
induced flow pattern and hydraulic head distribution difference.
Conditions considered in this study are a unidirectional head gradient
of 1 × 10− 2 m/m (1%) along one direction of the block, and a panel of
particles upstream from the middle of the block.
The simulated fracture intensity distribution is shown in Figure 5.
The distribution was directly numerically transferred from Figure 4
via the crack tensor approach. Figure 6(a), (b), and (c) shows the
permeability distributions in three rectangular directions (Kx, Ky, and
Kz, respectively). The reference system of 1605 planes contains
approximately 8000 elements, allowing for a detailed and realistic
representation of flow and transport within a very complex system. A
solid view of the flow field is shown in Figure 7. The figure shows how
the complexity of the flow field is represented in a detailed manner
using the proposed finite element approach. Homogeneous matrix
properties are considered here but the extension to heterogeneous
cases is not considered; that is, the matrix is thought to have a single
permeability of 10− 13 (m2).
From Figure 7, the overall directional flow is consistent within the
assumed boundary conditions (the flow is universally from upstream
to downstream). Figure 8 shows that the flow moves arbitrarily,
especially as it moves through the higher fracture density distribution.
However, a lot of clusters of higher fracture intensity (P32) lead to a
large vector distribution. Moreover, the much more connected pore
space may be supplied within more chance for flow, on the contrary it
is also shown that the vector will evade from the matrix in the
implicitly of lower fracture intensity.
One thousand non-reactive particles were released as a panel at
(4.5, 9.0, 4.5), as shown in Figure 9. The overall spatially distributed
particles are shown in Figure 10 and Figure 11 for cross-sections at
Fig. 6. Permeability distribution in rectangular x, y, and z directions (units are m2).
Z = 5 and X = 5, respectively. In Figure 10, the plane is depicted as a
 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177 173

Fig. 9. Releasing panel of particles in modeling domain.

Fig. 7. Pressure head distribution of the modeling domain.

same because the directional flow may contribute. The shortest and
horizontal spreading of particles and two very interesting points the longest transmission distance of 12 times the difference under the
addressed here one point is that the vanity of space was the place consideration of non-reactive particles. From Fig. 13, comparison of
where almost few particles want to visit and the other point is that the two modes of discrete fracture network model (DFN), convectional
particles' moving trend is aligned with more right-hand side than left- equivalent porous model (convectional EQM) with the mode was told
hand side. In Figure 11 the particles seem to follow a general path; that the early arriving time happened in DFN and moved quicker
they almost follow a regular pattern of moving straight downward. nearly one order of magnitude because the particles in DFN did travel
These spreading of particles are generally according to the directional only in fractures while the others both considered the matrix's
flow but the moving-late particles do not. The blank space is contribution to transport. From Figure 13, it was also clearly seen that
purposely regarded as relatively less transmissible media of matrix. our model not only caught the early traveling time but also took the
Looking on the migration path shown in Figure 12, the path line is matrix's effect in consideration.
calculated as the fastest-moving and the slowest-moving particles. The decaying particles are introduced to represent the phenomena
The pattern in Figure 12(c) is different from that in Figure 12(a) of the decay and in-growth of reactive particles. Figure 14(a) and (b)
because in the beginning, the slowest chooses the way that is filled shows the plumes of parent product A and daughter product B,
with matrix and it has to spend more time to pass the relatively respectively. Parent product A, which has a half-year half life without
stagnant zone. In Figure 12(b) and (d), the patterns are almost the adsorption, was released as a fixed concentration at the same location

Fig. 8. The fracture intensity distribution associated with the velocity vector.
174 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177

Fig. 10. Particles' spreading at Z = 5.

in Figure 9. Following the distributed concentration of Figure 14, the connection of void space and the inherent decay from the parent
plumes of A are aligned within the dominant flowing direction such as product.
flow in Figure 11, but in Figure 14(b) the plumes of B are moving quite
totally different. The relative concentration of B is smaller than that of 10. Conclusion
A because of the effect on decay, and simultaneously the higher
concentration of B is not obviously under the same way that parent A An approach for developing an applicable model from a site-
did. A possible reason for this anomalous transport is that the complex specific discrete fractured network was demonstrated for a small
structures of fracture intersections give rise to a non-continuous synthetic block on a 10 × 10 × 10 m scale. The initial model included

Fig. 11. Particles' spreading at X = 5.

 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177 175

Fig. 12. The fastest and slowest movement of particles. (a) Fastest movement of particles at Z = 5, (b) Fastest movement of particles at X = 5, (c) Slowest movement of particles at
Z = 5 and (d) Slowest movement of particles at X = 5.

1605 fractures associated with complex patterns and the heteroge-

neity of assigned continuum grids. A theoretical estimated perme-
ability method was developed to deal with the complexity of fractures
in fractured rocks. The system was simplified using a crack tensor in
combination with a discrete fracture network. This idea can be
extended to a framework that couples groundwater flow and
containment transport in fractured rock mass.
The results reveal the characteristics of the geometrical connection
of fracture networks under the prescribed flow field. The simulated
fracture networks were transformed into a grid-based simulation
procedure under consideration of the delaying effect in matrix owing
to particle migration. Synthetic flow results indicate that the flow
vector is very sensitive to fracture intensity (P32); a high fracture
intensity (P32) leads to a high velocity. Dynamics of particles can be
addressed in some significant points that the space discrepancy may
pose some fracture-clustered heterogeneity on transport as well as
the ratio of 12 times differences between the longest and shortest
particles' traveling distance. This effect will be extremely sensitive to
decay-ingrowths transport because the plumes of daughter product
are relatively tortuous and anomalous to predict.
The ability of the system simplified to major transport paths
proves that lower fracture intensities (background fractures) have a
minor effect on the flow and transport. This result is closely related
with low connectivity of the larger fractures for the particular block
and is not expected for a more connected system where one would
infer that (1) increase on flow properties of blocks and (2) consequently
Fig. 13. Comparison of these three different models. increase at least the contact surface with the matrix zones and as
176 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177

financial support coming from National Research Council under the

contract of NSC-98-3114-E-007-015.

References
Ando, K., Kostner, A., Neuman, S.P., 2003. Stochastic continuum modeling of flow and
transport in a crystalline rock mass: Fanay-Augeres, France, revisited. Hydrogeol-
ogy Journal 11 (5), 521–535.
Baecher, G.B., Lanney, N.A., Einstein, H.H., 1977. Statistical Description of Rock
Properties and Sampling, Proceedings of the 18th U.S. Symposium on Rock
Mechanics, vol. 5. American Institute of Mining Engineers, pp. C1–C8.
Bandis, S.C., Lumsden, A.C., Barton, N.R., 1983. Fundamental of rock joint deformation.
International Journal of Rock Mechanics & Mining Sciences & Geomechanics
Abstracts 20, 249–268.
Barton, N., Bandis, S., Bakhtar, K., 1985. Strength. deformation and conductivity
coupling of rock joints, International Journal of Rock Mechanics & Mining Sciences
& Geomechanics Abstracts 22, 121–140.
Bear, J., 1972. Dynamics of Fluids in Porous Media. American Elsevier, New York.
Berkowitz, B., 2002. Characterizing flow and transport in fractured geological media. A
review, Advances in Water Resource 25 (8–12), 861–884.
Bluma, P., Mackay, R., Riley, M.S., Knight, J.L., 2005. Performance assessment of a nuclear
waste repository: Upscaling coupled hydro-mechanical properties for far-field
transport analysis. International Journal of Rock Mechanics and Mining Sciences 42,
781–792.
Chen, R.H., Lee, C.H., Chen, C.S., 2001. Evaluation of transport of radioactive
contaminant in fracture rock. Environmental Geology 41, 440–450.
Christophe, G., Bernard-Michel, Gilles, Hakim, B., 2008. Evaluation of retention
properties of a semi-synthetic fractured block from modelling at performance
assessment time scales (Äspö Hard Rock Laboratory, Sweden). Hydrogeology
Journal. doi:10.1007/s10040-008-0415-x.
Deng, B.W., Lee, C.H., Chang, J.L., 1995. Characterization and interpretation of variability
of fracture pattern in jointed andesites Lan-Yu, Taiwan. Journal of the Chinese
Institute of Engineers 18, 787–799.
Dershowitz, W., Lee, G., Geier, J., Foxford, T., LaPointe, P., Thomas, A., 2004. FracMan,
Interactive Discrete Feature Data Analysis, Geometric Modelling, and Explora-
tion Simulation, User Documentation. Golder Associates Inc., Seattle,
Washington.
Dershowitz, W.S., (1984) Rock joint system, PhD Thesis, Massachusetts Institute of
Technology, Cambridge, Massachusetts.
Dershowitz, W.S., Einstein, H.H., 1988. Characterizing rock joint geometry with joint
system models1,. Rock Mechanics and Rock Engineering 21, 21–51.
Hsieh, P.A., Neuman, S.P., Stiles, G.K., Simpson, E.S., 1985. Field determination of the three-
dimensional hydraulic conductivity tensor of anisotropic media 2. Methodology and
application to fractured rocks. Water Resources Research 21 (11), 1667–1676.
Ivanova, V.M., (1998) Geologic and stochastic modeling of fracture systems in rocks,
PhD Thesis, Massachusetts Institute of Technology, Cambridge, Massachusetts.
Karzulovic, A., Goodman, R.E., 1985. Determination of principal joint frequencies.
International Journal of Rock Mechanics & Mining Sciences & Geomechanics
Abstracts 22, 471–473.
Langevin, C.D., 2003. Stochastic ground water flow simulation with a fracture zone
continuum model. Ground Water 41 (5), 587–601.
Lee, C.H., Farmer, I., 1993. Fluid Flow in Discontinuous Rock. Chapman & Hall, London.
169 pp.
Lee, C.H., Lee, C.C., Lin, B.S., 2007. The estimation of dispersion behavior in discrete
fractured networks of Andesite in Lan-Yu Island, Taiwan. Environmental Geology
52 (7), 1297–1306.
Lee, C.H., Chang, J.L., Deng, B.W., 1995. A continuum approach for estimating
permeability in naturally fractured rocks. Engineering Geology 39, 71–85.
Lee, C.H., Chang, J.L., Hsu, K.T., 1996. Investigation of hydraulic aperture at surface-
exposed rock fractures in situ. Geotechnique 46, 343–349.
Long, J.C.S., Remer, J.S., Wilson, C.R., Witherspoon, P.A., 1982. Porous media equivalents
Fig. 14. Particles within decay. (a) Plumes of parent product A and (b) Plumes of for network of discontinuous fracture. Water Resource Research 18, 645–658.
daughter product B. McKenna, S.A., Reeves, P.C., 2006. Fractured continuum approach to stochastic
permeability modeling, Stochastic Modeling and Geostatistics: Principles, Methods,
and Case Studies, AAPG Comput. Appl. Geol., vol. 3, 2nd ed. Am. Assoc. of Pet. Geol,
Tulsa, Okla, pp. 173–186.
such the retention properties of the block, found on Christophe et al. Miller, S.M., 1983. A statistical method to evaluate homogeneity of structural
(2008). populations. Mathematical Geology 15, 317–328.
Although using a three-dimensional approach is difficult in real National Research Council, 2001. Conceptual models of flow and transport in the
fractured vadose zone. National Academy Press, Washington, DC. 374.
world applications, tentative studies on simple academic cases and Neuman, S.P., 1987. Stochastic continuum representation of fractured rock permeabil-
geometries can still provide valuable insights, following, for instance, ity as an alternative to the REV and fracture network concepts, in Rock Mechanics.
the line of Rubin et al. (1999) or Wu et al. (2004). This integrated In: Farmer, I.W., et al. (Ed.), Proceedings of the 28th U.S. Symposium. A.A. Balkema,
Rotterdam, Netherlands, pp. 533–561.
framework can be used to determine the benefit of using the crack
Neuman, S.P., 2005. Trends, prospects and challenges in quantifying flow and transport
tensor approach to capture the characteristics or non-uniformity of through fractured rocks. Hydrogeology Journal 13 (1), 124–147.
the geologic structure. Oda, M., 1985. Permeability tensor for discontinuous rock masses. Geotechnique 35,
483–495.
Pohlmann, K., Pohll, G., Chapman, J., Hassan, A.E., Carroll, R., Shirley, C., 2004. Modeling
Acknowledgements to support groundwater contaminant boundaries for the Shoal underground
nuclear test, Publ. 45184. Desert Res. Inst, Reno, Nev.
The authors would like to thank "Final LLW and HLW Disoposal Priest, S.D., 1993. Discontinuity Analysis for Rock Engineering. Chapman & Hall.
Priest, S.D., Hudson, J.A., 1981. Estimation of discontinuity space and trace length using
Team" at the Institute of Nuclear Energy Research in Taiwan for scanline survey. International Journal of Rock Mechanics & Mining Sciences &
helping generate the fracture networks and flow simulation and also Geomechanics Abstracts 18, 183–197.
 J.-B. Pan et al. / Engineering Geology 116 (2010) 166–177 177

Reeves, D.M., Benson, D.A., Meerschaert, M.M., 2008. Transport of conservative solutes Teimoori, A., Chen, Z., Rahman, S.S., Tran, T., 2005. Effective permeability calculation using
in simulated fracture networks: 1. Synthetic data generation, Water Resource boundary element method in naturally fractured reservoirs. Petroleum Science and
Research 44, W05404. Technology 23 (5), 693–709.
Rubin, G., Jansen, D., Forkel, C., Köngeter, J., 1999. Simulation of contaminant transport Wang, M., Kulatilakea, P.H.S.W., Uma, J., Narvaiz, J., 2002. Estimation of REV size and
in fractured permeable formations by multiporosity modeling. Journal of three-dimensional hydraulic conductivity tensor for a fractured rock mass through
Hydrology 223, 107–130. a single well packer test and discrete fracture fluid flow modeling,. International
Selroos, J.-O., Walker, D.D., Stro¨m, A., Grylling, B., Follin, S., 2002. Comparison of Journal of Rock Mechanics and Mining Sciences 39, 887–904.
alternative modeling approaches for groundwater flow in fractured rock. Journal of Wide´n, H., Walker, D., 1999. SR 97—Alternative models project: Stochastic continuum
Hydrology 257, 174–188. modelling of Aberg, S.K.B. Rep. (R-99-42).
Shanley, R.J., Mahtab, M.A., 1976. Delineation and analysis of clusters in orientation Wu, Y.S., Liu, H.H., Bodvarsson, G.S., 2004. A triple-continuum approach for modeling flow and
data. Mathematics Geology 8, 9–13. transport processes in fractured rock. Journal of Containment Hydrology 73, 145–179.
Snow, D.T., 1969. Anisotropic permeability of fractured media. Water Resource Zyvoloski, G.A., Robinson, B.A., Viswanathan, H.S., 2008. Generalized dual porosity: a
Research 5, 1273–1289. numerical method for representing spatially variable sub-grid scale processes.
Svensson, U., 2001a. A continuum representation of fracture networks. part I: Method Advances in Water Resources 31, 535–544.
and basic test cases, Journal of Hydrology 250, 170–186. Zyvoloski, G.A., Dash, Z.V., Kelkar, S., 1988. FEHM: Finite-element heat- and mass-
Svensson, U., 2001b. A continuum representation of fracture networks. part II: transfer code, Los Alamos National Laboratory Report, LA-11224-MS.
Application to the Aspo Hard Rock laboratory, Journal of Hydrology 250, 187–205.