Advances in Discrete Fracture Network Modeling

William S. Dershowitz, Paul R. La Pointe, and Thomas W. Doe, Golder Associates Inc.

Abstract
This paper describes recent advances in discrete fracture network (DFN) modeling and analysis. These
advances represent a significant convergence between DFN and EPM methods for flow, combined with
significant increases in geological realism, and more appropriate solute transport conceptual models.

Introduction
Since its introduction in the late 1970s, considerable controversy has raged in the hydrogeologic community
over the value of discrete fracture network (DFN) approach, and particularly about the merits of the DFN
approach as compared to the stochastic continuum approach. Much of this controversy has been related to the
fundamental concept of the representative elementary volume (REV) which underpins all continuum
approaches (Bear, 1972).
According to the REV concept, there exists a scale at which individual heterogeneities and discrete features can
be ignored, due to a process of averaging to produce an effective continuum medium. The fundamental
motivation of discrete fracture network (DFN) modeling is the recognition that at every scale, groundwater
transport in fractured and carbonate rocks tends to be dominated by a limited number of discrete pathways
formed by fractures, karts, and other discrete features. The DFN approach can thus be defined as analysis and
modeling which explicitly incorporates the geometry and properties of discrete features as a central component
controlling flow and transport.
Discrete Fracture Network related research carried out over the past twenty years has focused on identifying
those individual discrete features and karts which provide discrete connections which carry the most important
portion of flow. The task of hydrogeology then becomes to characterize these individual pathways, rather than
the average or REV properties of the medium. To achieve this, it is necessary to understand the geometry and
properties of the discrete fractures and karts which form these pathways.
The development of this discrete feature conceptual model is the essence of the DFN approach. The DFN
approach provides the three dimensional framework of discrete features which concentrate flow and transport,
and also the flow barriers such as faults and argillaceous layers which provide partial or complete seals. This
DFN conceptual model can be referred to as a discrete feature hydrostructural model (Winberg et al., 2003).
Once the DFN hydrostructural model has been implemented, and the discrete features of importance are know
deterministically, or at least stochastically, the DFN can then be solved to address the specific flow and
transport engineering issues of concern.
This paper highlights several trends in DFN modeling including:

Convergence of DFN and Continuum Methods,

Increasing Geologic Realism, and

Multiple Immobile Zone Transport

Convergence of DFN and Continuum Methods

DFN analysis is fundamentally about the development of an appropriate hydrostructural model which considers
the role of known and unknown discrete features in controlling flow and transport. As such, DFN analysis is as
essential for conventional continuum approaches using volume elements as it is for flow and transport models
built on pipe or plate discrete feature elements. If you dont know the nature of discrete feature transport
pathways or flow barriers, you cant carry out conventional continuum modeling any more than you can DFN
flow and transport modeling.
A recent example of this is shown in Figures 1, 2, and 3. Figure 1 illustrates the hydrostructural model
developed to describe a 200 m scale block of sp granite within the sp hard rock laboratory. Figures 2 and
3 illustrate DFN and EPM implementations of this model. The EPM implementation of the model (GomezHernandez et al.,2003) fits the definition of a DFN model, since it explicitly considers the discrete features

882

identified within the discrete feature hydrostructural model. However, the EPM implementation uses volume
elements at the scale of 5 m, which is coarse considering the 2 cm in total thickness of the fractures being
modeled.
Comparison of Figures 2 and 3 demonstrates that the debate over EPM vs. DFN is no longer about the value of
the DFN hydrostructural approach, but only about the most efficient implementation. Both EPM and DFN
models were built on a common DFN hydrostructural model, developed using DFN analysis techniques.
During recent years, this has led increasingly to the development of hybrid DFN/EPM models such as
ConnectFlow (Serco, 2004) and FracWorks XP (Dershowitz et al., 2004). There are three types of integration
for DFN/EPM models.

Layered DFN/EPM Models,

EPM Implementation of DFN Hydrostructural Models, and

Nested DFN/EPM Models.

Layered DFN/EPM Models

In many geological environments, heterogeneously connected karstic or fractured rocks occur within
stratigraphic columns containing units best represented by continuum elements. Layered DFN/EPM Models
meet this by incorporating both EPM volume elements, and DFN (pipe or plate) elements. Figure 4 illustrates a
layered DFN/EPM model, with the EPM representing primarily sandstone layers, and the DFN representing
fractured granite. The advantage of this approach is that it is able to more accurately model the response of the
groundwater table and shallow wells using continuum EPM elements, while still using the DFN for evaluating
connectivity between wells in the fractured granite.
The key to implementation of the layered DFN/EPM approach is the linkage between DFN nodes, which occur
at the intersections between and edges of polygonal fracture elements, and EPM nodes, which occur at the
corners of tetrahedral or hexahedral volumes.
In general, the DFN elements and EPM elements are on different scales such that there may be many DFN
nodes corresponding to a single EPM node, or vis-versa. The linkage between DFN and EPM nodes is achieved
by recognizing this difference in discretizations, and providing for a node based linkage. This type of linkage is
illustrated in Figure 5. In this example, DFN nodes within a distance d from any EPM node are linked to the
closest EPM node. This linkage can be achieved by setting the linked EPM and DFN nodes to the same head
value, or by placing a pipe element of specified transmissivity between the nodes, if there is some special
material at the DFN/EPM interface.
EPM Implementation of DFN Hydrostructural Models
Approaches for implementation of DFN hydrostructural models using EPM technologies include

Explicit modeling of a key faults and fractures,

Geostatistical simulation of faults and fractures, and

Oda Tensor approaches.

EPM models have always been able to represent a limited number of faults and fractures explicitly using the
same volumetric elements that are used to represent other geologic materials. Where these few discrete features
carry the vast majority of flow and transport, these EPM models could be considered a fairly straightforward
DFN hydrostructural model implementation. In examples reviewed for this paper, examples were found where
over 95% of the flow occurred in the explicitly modeled discrete features, such that the EPM model was
functionally identical to a DFN.
EPM models of discrete features have been implemented using finite element (Figure 6), finite difference
(Figure 7), and finite volume methods. Of these, the finite element method is attractive since the grid can be
adjusted to conform to the fracture geometry. In contrast, the finite difference method generally requires a

883

stairstepped approximation. The finite volume method is frequently significantly less computationally efficient
than either the finite element or finite difference methods, but has advantages for modeling coupled processes.
Geostatistical Simulation of Faults and Fractures
Traditional geostatistical simulation derives variograms from the spatial pattern of hydraulic properties, and
then attempts to reproduce this pattern by kriging from these variograms. Given the complex, three dimensional
nature of faults and fractures, this method is unlikely to be able to regenerate the pattern of semi-planar discrete
features.
During the past ten years, advances in geostatistical modeling has made it possible to implement geostatistical
models which reproduce and also enhance DFN hydrostructural models.
a) indicator kriging cells can be marked based on the occurance of fractures or flow barrier faults, and the
geostatistically derived property fields can be adjusted accordingly. This produces fields which correspond to
the location and properties of the discrete features.
b) co-kriging cell hydraulic properties can be adjusted based on a combination of both the hydraulic test
results which provide permeability and storativity, and the geophysical fields which are indicators for the
location of faults.
c) numerical inversion geostatistical EPM model cell hydraulic properties can be optimized by a combined
inversion of the underlying DFN hydrostructural model, geophysical data, and hydraulic data.
The effectiveness of the integrated hydrostructural, hydrologic, and geophysical numerical inversion approach
is illustrated in Figure 6 (Will et al., 2003).
Oda Tensor Approaches
The Oda (1985) method for calculating EPM properties from DFN fractures is now twenty years old, and has
been re-derived with variations many times. The advantage of the Oda approach is that it can obtain EPM
properties for grid cells based directly on the geometry and properties of the fractures within those cells.
The Oda approach begins by generating the full, three dimensional DFN. While DFN flow and transport
modeling is limited to on the order of 104 to 105 fractures by computational constraints, the Oda approach can
be applied to patterns of 107 or more fractures. The Oda approach overlays an EPM grid on the fractures, and
derives EPM properties for each grid cell based on the DFN contained in that cell.
The orientation of fractures in each grid cell is expressed as a unit normal vector n. Integrating the fractures
over all of the unit normals N, Oda obtained the mass moment of inertia of fracture normals distributed over a
unit sphere:

N = ni n j E ( n ) d

(1)

where:
N
ni, nj
E(n)

=
=
=

number of fractures in .
the components of a unit normal to the fracture n
probability density function that describes the
number of fractures whose unit vectors n are
oriented within a small solid angle d
entire
solid
angle
corresponding
to
the
unit sphere

surface

of

For a specific grid cell with known fracture areas, Ak, and transmissivities, Tk, obtained from the DFN model,
an empirical fracture tensor can be calculated by adding the individual fractures weighted by their area and
transmissivity:

884

1 N
Fij = f k Ak Tk nik n jk
V k =1

(2)

where:
Fij
V
N
fk
Ak
Tk
nik, njk

=
=
=
=
=
=
=

fracture tensor
grid cell volume
total number of fractures in grid cell
percolation factor for fracture k (generally assumed to equal 1)
area of fracture k
transmissivity of fracture k
the components of a unit normal to the fracture k

Odas permeability tensor is derived from Fij by assuming that Fij expresses fracture flow as a vector along the
fractures unit normal. Assuming that fractures are impermeable in a direction parallel to their unit normal, Fij
must be rotated into the planes of permeability

k ij =

1
(Fkk ij Fij )
12

(3)

where:
kij
Fij

ij

=
=
=

permeability tensor
fracture tensor
Kroenecker delta

The Oda approximation derives an equivalent permeability tensor, according to a specific grid. This tensor can
also be derived by using a numerical permeameter, in which a boundary condition is applied to the edges of the
grid cell (see e.g., Long, 1984). At the extreme of fine discretization (mm scale grid cells), the Oda
approximation reproduces the underlying discrete fracture hydrostructural model, both for flow and for
transport. However, at the practical and much coarser discretizations of tens or hundreds of meters, the Oda
approximation produces results which are less accurate.
The Oda tensor approach thus represents a balance between the accuracy of directly modeling each of the
structures in the hydrostructuctural model, against the computational efficiency of coarser numerical model
discretizations.
Nested EPM/DFN Models
The inherent compromise of using EPM models to represent DFN hydrostructural models can be reduced
through the recent development of nested EPM/DFN Models. These models combine the use of DFN elements
in the locations where fracture geometry is of most concern, such as at intersections with boreholes and tunnels,
with EPM elements at less sensitive locations.
Figure 9 illustrates a nested EPM/DFN model. This model utilizes a coarse Oda based EPM grid for most of
the model, while maintaining a direct implementation of the DFN hydrostructural model using DFN elements in
the direct vicinity of the boreholes of interest.
The key to the implementation of Nested EPM/DFN models is the same as for layered EPM/DFN models, i.e.,
the linkage between the EPM and DFN elements, both for flow and for transport. The approach described
above for layered EPM/DFN models has also been used successfully for nested EPM/DFN models.

Advances in Geological Realism

885

Discrete fracture network hydrostructural models are implemented by integration of

geology

geophysics, and

hydrogeology

Recent advances in geophysics such as 3D VSP and AVO/AVAZ have made it possible to develop DFN
hydrostructural models with explicitly incorporate the 3D geometry of structures greater than on the order of
100 m (Parney and La Pointe, 2002). This increasingly explicit information about fault geometry perhaps
surprisingly increases the need for geologically and hydrologically based methods to a) derive the properties of
those identified structures, and b) extrapolate from the seismic scale structures to the sub-seismic features.
In sites where geophysical characterization of structures is either economically or technically infeasible,
geological and hydrogeological methods are also necessary for deriving the location and properties of both
seismic and sub-seismic structures.
Below, we present three key recent advances in the geologic realism of DFN hydrostructural models:

DFN Fault Models

Carbonate Solution Features

Geocellular and Bootstrap Models

DFN Fault Models

Figure 10 (after Caine, 1999) presents a schematic view of the internal structure of faults. Faults generally
combine conductive and flow barrier features a flow barrier at the fault core, combined with conductive
fractures, particularly in the fault damage zone. Recent advances in DFN modeling provide approaches for
implementing the full range of conductive, flow barrier, and partially sealing fault behaviors:

Purely conductive faults can generally be represented as a planar or tessellated non-planar discrete
feature. The variability within the fault can be represented by applying varying transmissivities to the
elements defining the fault,

Purely flow barrier faults can be represented by truncating the fractures intersecting at their fault
intersection, or reducing fracture transmissivity in the vicinity of the fault intersection (Figure 11),

Faults flowing primary in the footwall and hangingwall damage zones can be modeled by generating
special fracture sets for these regions, or by representing this transmissivity within the fault surface
itself, and

Partially sealing faults can be represented by reducing the transmissivity of fractures where they
connect to or pass through the fault.

These fault barrier models are an essential aspect of an increasing percentage of DFN modeling, as an
increasing proportion of sites show evidence of hydraulic compartmentalization. Figure 12 illustrates an
example of hydraulic compartmentalization at the Block Scale experiment in Sweden. Despite a strong
hydraulic gradient toward the northeast due to the presence of unlined tunnels, the salinity indicates the
presence of a NNW oriented flow barrier compartmentalizing the system into saline and fresh groundwater
regimes.
DFN Carbonate Solution Feature Models
While some fractured rocks can be treated as single porosity materials, carbonate rocks are frequently three
porosity systems, combining a significant matrix permeability with fractures and solution features such as vugs
and solution enhanced discrete pathways. These features are particularly difficult to model with EPM
approaches. DFN approaches for simulating carbonate solution features include the following:
Vugs: Vugs are solution features which can range in size from millimeters to tens of meters, with on the order
of 100% porosity. Within DFN models, vugs can be modeled as three dimensional discrete features using either

886

volume element or as a storage interaction term in the approximate dual porosity approach. Figure 13 illustrates
an approach for modeling vugs as storage terms interacting with fractures. In this approach, the storage in the
vug interacts with the fracture through a 1-D interaction term.
Wormhole Channels: Solution enhanced fracture intersections are major form of transport pathway in
fractured carbonates. These channels frequently have apertures 10 to 50 times larger than those of the
intersecting fractures, such that the channel itself has a transmissivity 103 to 105 that of the fractures (assuming
the cubic law applies). This can now be implemented in DFN models by adding pipe elements at karst enhanced
fracture intersections. The extra storage and transmissivity of the wormhole channels can then be represented
by a separate parameter. The geometry of the wormhole channels can then be defined by a geological
simulation process reflecting the geologists conception of solution evolution.
Karstic Porosity on Fracture Planes: One of the key features of karstic carbonates is that certain portions of
fracture planes have dramatically increased transmissivity and storage due to local solution enhancement of the
fracture aperture. This may be part of a continuous pathway (as in wormholes), or it may be a local effect. For
example, at the Yates Carbonate Oil Reservoir in Texas, vuggy areas on fracture plan account for most of the
locations at which oil flows freely to the production wells (Wadleigh, 1998).
The karstic porosity on fracture planes can be modeled by tessellating the fracture surfaces, and applying the
appropriate spatial pattern of karstic porosity to the planes. The pattern of karst porosity can be described by
geostatistical or fractal methods, or on a geological basis if there is a geological theory for the spatial
distribution of karstic porosity (Figure 14).
In general, the karstic porosity on the fracture plan is described by three parameters the karsted percentage of
the fracture plane area (karstic persistence) , the aperture distribution for the karsted regions, and the spatial
pattern (e.g., range or correlation length for a geostatistical field).
Geocellular and Bootstrap Models
Fractured rock DFN hydrostructural models covering areas of more than about 1 km2 generally have a spatial
variability in the fracture pattern. The orientation, size, and intensity of fracturing tend to vary across the site,
such that statistically homogeneous models are inadequate. The need for such models has been filled over the
past five years through the development of geocellular and boostrap models.
Geocellular models fill the entire 3D space with information extrapolated from geology, geophysics, and
hydrogeologic. This information is commonly stored in Geomatic/GIS systems such as for example ArcInfo
(ESRI, 2004), Petrel (Schlumberger, 2004), EarthVision (Dynamic Graphics, 2004), or StrataModel (Landmark,
2004). In these systems, the rock volume is discretized to volumetric cells, and each cell is linked to
appropriate properties. In a geocellular DFN model, the fractures and faults, and their properties are generated
directly from that cell-based information. For example, at the Yates field, fracture intensity was directly tied to
shale content. The geocellular DFN model therefore generated fracture intensity P32 (m2/m3) at every location
based on the relative shale content.
Another application of the geocellular approach is in generating 3D DFN hydrostructural models based on an
interpreted 3D stress or strain field. In this approach, the geologic or rock mechanics analysis interprets the
present or recent stress or strain field throughout the rock mass. This field may be based on, for example, stress
cell measurements, or a palinspastic reconstruction of the strain history (La Pointe et al., 2003). In either case,
the geocellular grid then contains a stress or strain tensor, and the fracture orientation, intensity, and
transmissivity can be generated at each location in the rock contingent on the local tensor information. An
example stress tensor grid based DFN model is shown in Figure 15.
Bootstrap models start with know values at specific locations (generally geophysical measurements or
boreholes), and use this information directly to extrapolate the three dimensional population, without creating
an intermediate synthetic geocellular field. The bootstrap algorithm is summarized as follows. When
generating a fracture at a specific location, the model looks for data which is relevant, and ranks that data by its
relative relevance. For example, for generating a fracture orientation, the most relevant data would be near by,
in the same stratigraphic unit, and on the same side of any significant structural feature such as a fault. The
fracture orientation at that location would then be generating by choosing a single measured value from the
table of relevant measurements, with a probability weighted by relevance. Once a particular value was chosen,

887

a small dispersion factor would be applied to account for the stochastic and spatial variability. This process is
illustrated in Figure 16.
The advantage of the bootstrap approach is that it directly utilizes available data, rather than an interpolated
field, as in the geocellular approach. This works well when there is good spatial coverage for data, or where
there is no appropriate method for populating a geocellular field.

Multiple Immobile Zone Transport

The key advance in fractured rock transport analysis over the past ten years is the gradual realization of the
importance of the interaction between mobile and immobile zones in fractured rock. The conventional EPM
approach assumes a single, mobile (advective) porosity, with retardation occurring through sorption to grain
surfaces within that porosity. In fractured rock, in situ experiments such as those at sp, Sweden (Winberg at
al., 2003) have demonstrated that the transport in fractured rock is strongly dependent on the interaction
between advective transport in fractures, and the storage of solutes in immobile zones. These immobile zones
include a) gouge, breccia, cataclasite, calcite, and mylonite in the fracture itself, b) fracture coatings, c) the
higher porosity altered/damaged zone surrounding many fractures, and d) intact rock (Figure 17). The
geometry and properties of these immobile zones are characterized as the microstructural model for the
fractures, and are as important as the DFN hydrostructural model discussed above for many transport analysis
applications.
Four basic types of transport approaches have been developed to describe transport in fractured rock with
multiple immobile zones:
Diffusive Exchange (Sudicky and McLaren, 1992). This approach implements the conventional advectiondispersion-diffusion equation in the Laplace domain. In this approach, the exchange between mobile and
immobile zones is exclusively due to a diffusive process expressed by the concentration gradient between the
advective and immobile zones. Multiple immobile zones can be defined both in parallel and in series, allowing
explicit use of measured values for each immobile zone.
Multirate Diffusion (Fleming and Haggerty, 2001, Cvetkovic et al.,1999). Rather than depending on a
characterization of the properties of each of the immobile zones listed above, the multirate approach assumes a
continuous distribution of immobile zone porosities, generally defined by a Pareto or Powerlaw distribution.
This assumption allows a single exchange term between the mobile and immobile zones, while recognizing the
importance of the different characteristics of different immobile zones. In the Cvetkovic et al (1999)
implementation LaSAR, advection and diffusion are described by a lumped immobile zone retention
parameter and by the distribution of inverse aperture weighted advective velocity .
CTRW Continuous Time Random Walk (Berkowitz et al., 2000) treats both diffusion and Gaussian
dispersion as special cases of multi-rate advection, and can reproduce both diffusive and Gaussian transport on
the basis of a single parameter , the power for the assumed powerlaw (Pareto) distribution of velocity.
Advective exchange (AX) (Miller, 1996) assumes that the process of exchange between mobile and immobile
zones occurs at a fixed rate of exchange per meter of advection, rather than at a rate proportional to differences
in concentration. The approach is based on two parameters: , the number of advective exchanges per meter,
and fimm, the ratio of the immobile zone volume to the mobile zone volume.
The above approaches have several common features of interest to practical applications

They all depend on accurate characterization of fracture immobile porosities through an appropriate
microstructural model

They all recognize explicitly that transport is occurring on discrete features, and assume that the
advective velocity within fractures (rather than the effective EPM velocity) is the relevant parameter

They all have been used successfully in predictive simulations of solute transport at 10 to 50 m scales

They generally use as a key universal parameter!

888

Conclusions
This paper has attempted to highlight recent advances in the discrete fracture network approach to flow and
transport modeling for fractured and karstic rock. Major advances have been made in the development and
implementation of hydrostructural models for fracture geometry, and microstructural models, for the immobile
zones influencing solute transport. These advances have made it possible to build more accurate three
dimensional hydrogeologic models, and to better analyse both flow and transport processes.

Acknowledgements
Grateful acknowledgement is made to the major sponsors of the technical developments described in this paper
including

Japan Nuclear Cycle Development Institute, Tokai, Japan

National Petroleum Technical Office, US DOE, Pittsburg
Fractured Reservoir Engineering Development Consortium (Shell, Exxon, Marathon, Agip, and
NorskHydro)
Swedish Nuclear Fuel and Waste Management Co., sp, Sweden

Biographical Sketch
William Dershowitz, Paul Lapointe, and Thomas Doe are the senior staff of the FracMan Technology Group in
Golder Associates Redmond WA office. The authors have an aggregate of over 75 years of experience in
characterizing and simulating flow in fracture networks. The signature product of the FracMan Technology
group is the FracMan family of DFN modeling and data analysis software. These methods have been used
for fractured rock research for radioactive waste research in underground laboratories in Sweden, Japan,
Switzerland, and France as well as practical applications for petroleum, solid waste, and civil engineering.
Corresponding Author and Speaker (platform preferred): William Dershowitz
Address for all authors:
Golder Associates
18300 Union Hill Road Suite 200
Redmond WA 98052 USA
Tel: 425 883 0777; Fax 425 882 5498
Emails: dersh@[Link]; plapointe@[Link]; tdoe@[Link]

889

References
Bear, J., 1972, Dynamics of Fluids in Porous Media. New York, Elsevier.
Berkowitz, B., H. Scher and S. E. Silliman, 2000, Anomalous transport in laboratory-scale, heterogeneous porous media, Water Resources
Research, Vol. 36, No. 1, pp.149-158.
Caine, J, 1999. Fault Zone Architecture and Permeability Structure. PhD Dissertation, Geology. Salt Lake City, University of Utah.
Cvetkovic, V., Selroos, JO, and Cheng, H., 1999, Transport of reactive tracers in rock fractures. Journal of Fluid Mechanics, 378: 335-356.
Dershowitz, W., G. Lee, and T. Foxford, 2004, FracWorks XP Discrete Fracture Modeling Software, User Documentation. Redmond, WA,
Golder.
Dershowitz, W., K. Klise, A. Fox, S. Takeuchi, and M. Uchida, 2002, Channel Network and Discrete Fracture Network Modeling of
Hydraulic Interference and Tracer Experiments and the Prediction of Phase C Experiments. SKB Report IPR-02-29. Stockholm, SKB.
Dynamic Graphics, 2004. Earth Vision User Documentation. Alameda, CA, Dynamic Graphics, Inc.
ESRI, 2004, ArcInfo User Documentation. Redlands, CA., ESRI.
Fleming, S. W., and R. Haggerty, 2001, Modeling solute diffusion in the presence of pore-scale heterogeneity: Method development and an
application to the Culebra Dolomite Member of the Rustler Formation, New Mexico, USA, Journal of Contaminant Hydrology, 48, 253-376,
2001
La Pointe, P, 2004, Development and Validation of a Fractured Reservoir Model for the Circle Ridge Field, Wind River Basin, WY, AAPG
Annual Meeting Abstracts, Houston, AAPG.
Long, J.C.S., 1984, Investigation of Equivalent Porous Medium Permeability in Network of Discontinuous Fractures. PhD Dissertation,
Earth Science Division, Lawrence Berkeley Laboratory, University of California, Berkeley.
Miller, 1996, Advective Exchange: A New Conceptual Approach for Solute Transport in Fractured Rock. Internal Report. Seattle, Golder
Associates Inc.
Oda, M.; 1985, Permeability Tensor for Discontinuous Rock Masses., Geotechnique Vol. 35, pp 483.
Parney, R, and P. La Pointe, 2002. Fractures Can Come into Focus. AAPG Explorer Online, Nov.
Poteri, A., D. Billaux, W. Dershowitz, J J Gomez-Hernandez, V. Cvetkovic, A Hautojrvi, D. Holton, A. Medina, and A. Winberg, 2002,
Final Report of the TRUE Block Scale Project, Report 3, Modeling of Flow and Transport. SKB Technical Report TR-02-15. Stockholm,
SKB.
Schlumberger Information Solutions, 2004. Petrel. Houston, SIS.
Serco Assocurance, 2004. Connect Flow Software. Harwell UK.
Sudicky, E.A. and McLaren, R.G., 1992, The Laplace transform Galerkin technique for large-scale simulation of mass transport in
discretely fractured porous formations, Water Resources Researcj Vol 28, No. 2, 499-514.
Taivassalo, V. and F. Meszaros, 1994, Simulation of Groundwater Flow of the Kivetty Area. YJT Report YJT-94-03. Helsinki, Posiva OY.
Wadleigh, E., 1998. Fracture Images Showing Karsted Fracture Intersections. Midland TX, Marathon Oil Co.
Will, R, R. Archer, and W. Dershowitz, 2003. Integration of Seismic Anisotropy and Reservoir Performance Data for Characterization of
Naturally Fractured Reservoirs Using Discrete Feature Network Models. SPE Paper 84412. Houston, SPE.
Winberg, A. P. Andersson, J Byegrd, A. Poteri, V Cvetkovic, W Dershowitz, T Doe, J. Hermansson, J J Gomez-Hernandez, A Hautojrvi,
D Billaux, E-L Tullborg, P Meier, A Medina, 2003. Final Report of the TRUE Block Scale Project, Report 4, Synthesis of Flow, Transport
and Retention in the Block Scale. SKB Technical Report TR-02-16. Stockholm, SKB.

890

z = -550 m

z = -450 m

z = - 350 m

-2

-2

-8

-14
+2.2

-14
-2

-8

-1.7

-14

Figure 3.
Gomez-Hernandez (2002)
Equivalent Porous Medium Implementation
of TRUE Block Scale Hydrostructural
Model (after Poteri et al, 2003)
Figure 1. Numbered Structures Comprising
the Hydrostructural Model for the TRUE
Block Scale Experiment (after Poteri et al.,
2003)

Figure 4. Layered EPM/DFN Model

EP

Figure 2. Dershowitz et al (2002) Discrete

Fracture Network Implementation of TRUE
Block Scale Hydrostructural Model (after
Poteri et al, 2003)

Lik
d

Figure 5. Linking EPM and DFN Nodes

891

adjusted the DFN hydrostructural model,

then transformed this model to an EPM
using the Oda tensor approach (after Will et
al., 2003)
`

Figure 6 Finite Element Implementation of

DFN Hydrostructural Model (Taivassalo and
Meszaros, 1994)

Figure 7.
Finite Difference EPM
Implementation of DFN Hydrostructural
Model

Figure 9. Nested DFN/EPM Model

Low

Distributed
Conduit

% Core

High

High

High

Permeability
Structures
in Fault Zones

% Damage
Zone

% Damage
Zone

Low
Localized
Conduit

Combined
Conduit-Barrier

Low

Low

% Core

High

Localized
Barrier

after Caine et al. (1996)

Figure 8. Combined Inversion of DFN and

EPM Models Using geophysical and
hydraulic data. BHP and OPR and borehole
pressure and simulated oil production rate,
respectively. The inversion simultaneously

Figure 10. Conceptual Models for Faults in

DFN Hydrostructural Models (after Caine,
1999)

892

Figure 11.
Flow Barrier Fault
implementation
in
DFN
Model.
Hangingwall and Footwall fractures both
terminate against surface representing fault.
Conductivity across fault is therefore
controlled by the permeability of the fault
plane inself.

Figure 12. Representation of Vug Storage

Oorosities in DFN Models

Figure 14. Pattern of Karst Porosity on

Fracture Plane

Figure 12.
Geochemical Evidence of
Hydraulic Compartmentalization at sp
TRUE Block Scale (after Doe et al., 2002).
Contours
should
compartmentalized
concentration of Cl-, despite strong NE
gradient .

Figure 15. Stress Model Based DFN

893

`
Figure 16. Bootstrap DFN Model

Figure 17. Immobile Zones (after Winberg

et al, 2003)

894