SPE 91940

Dynamic Behavior of Discrete Fracture Network (DFN) Models

Hugo Araujo, Pablo Lacentre, Tomás Zapata, Aldo Del Monte, and Francisco Dzelalija, Repsol-YPF; James Gilman, SPE,
and Hai-Zui Meng, SPE, iReservoir.com; and Hossein Kazemi, SPE, and Erdal Ozkan, SPE, Colorado School of Mines

Copyright 2004, Society of Petroleum Engineers Inc.

construction of the ultimate reservoir flow model would
This paper was prepared for presentation at the 2004 SPE International Petroleum Conference depend on calibration against reservoir performance as well as
carefully designed flow tests.5 In fact, an ultimate reliability
in Mexico held in Puebla, Mexico, 8–9 November 2004.

This paper was selected for presentation by an SPE Program Committee following review of
information contained in a proposal submitted by the author(s). Contents of the paper, as
standard for viability of any reservoir characterization model
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to is calibration against dynamic data from several wells. This
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at implies that reservoir characterization life cycle is an iterative
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
process as new static and dynamic data become available.
for commercial purposes without the written consent of the Society of Petroleum Engineers is Experience in naturally fractured reservoirs (NFR) has
prohibited. Permission to reproduce in print is restricted to a proposal of not more than 300
words; illustrations may not be copied. The proposal must contain conspicuous indicated that low permeability formations often produce
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
fluids through open fractures, which intersect the wellbores.
This paper illustrates fluid flow through individual fractures
intercepted by a vertical well, initially estimated using a
Abstract parallel plate method, finding that contribution of each
This work shows that discrete fracture network modeling is fracture need to be calibrated to a production test. The
very desirable for the characterization of naturally fractured composite summation of individual fractures along the
reservoirs but it is only a highly subjective starting point. wellbore allowed the construction of a synthetic PLT profile.
Thus, calibration against short and long term pressure transient A horizontal sidetrack from the vertical well was planned
tests is most crucial. This paper shows how the dynamic to improve the well’s productivity. A discrete fracture network
behavior of a discrete fracture network model of Margarita gas (DFN) model, generated around the area of influence of the
field compared against pressure transient measurements in a well, was used to identify the best direction for the sidetrack
sidetrack delineation-well. The performance comparison of a interval (perpendicular to the most open main fracture planes)
very fine-grid reservoir model, which included the discrete and to predict the type of the fractures to be intercepted. After
fracture network information, versus a much coarser upscaled the sidetrack was drilled, fracture types (hierarchy by aperture)
grid model is also documented. were identified and classified using an ultra sonic borehole
image tool. The DFN fracture model was calibrated against a
Introduction pressure buildup test conducted in the sidetrack well, using
Geological reservoir characterization is the most crucial first dual-porosity numerical simulation over a very fine grid
step in construction of a credible reservoir flow model for model. The calibration studies showed that the aperture-based
naturally fractured reservoirs. The most notable approaches effective transmissivity was significantly greater than the
for constructing a geological model of a reservoir include actual well test transmissivity while the estimated fracture
classical deterministic methods, where geologists make the storativity was significantly lower than the actual well test
best interpretation from existing data and build a model of the value. Dynamic upscaling of the fine-grid model was
reservoir. In the last several years, however, deterministic performed to evaluate whether the upscaled models would
approaches have been complemented by quantitative preserve the well/reservoir behavior of the fine-grid models.
geostatistical approaches such as multipoint statistics (MPS)
and discrete fracture network (DFN) modeling.1, 2, 3 Margarita Field General Characteristics
The MPS generates a depositional and lithofacies model of The Margarita structure is located in the southern Bolivian
a reservoir by incorporating geological architecture and Subandean, on the structural trend of the Suaruro Range, 35
properties of the rock fabric and is touted as a very promising km to the west of Villamontes town.6-8 The Margarita Field
stepping stone in construction of viable numerical reservoir lies in the Caipipendi Block (Fig. 1) operated by Repsol-YPF
flow models.1 However, for reservoir modeling purposes, the with 37.5% equity. Partners are BG International (37.5%) and
flow units and fracture flow paths must be included in the Pan American BP (25%). The field is located in the northern
depositional model.4 The DFN modeling is often used to part of an elongated anticline oriented NNE-SSW, and is 30
accomplish the latter. Both the deterministic and geostatistical km long and 9 km wide. The closure of the field consists of several
techniques, however, are highly subjective and, while such compartments separated by reverse faults.
fundamental geological modeling techniques are often the Well MGR-X1 in the Huamampampa formation at 4404
necessary starting points, any revisions toward the mMD is the discovery well of the Margarita Field. Log
2 SPE 91940

analysis, core and side core samples, and well testing data f = fracture friction factor, dimensionless
indicate that the Huamampampa formation is a low porosity Lf = fracture length, ft
(5% or less), quartzite sand formation, with natural fractures ρ  = fluid density, lb/ft3
allowing fluid production from an overpressured reservoir. The friction factor is a function of the Reynolds Number, NRe,
Microfractures and macrofractures are the only porosity calculated from the following equation:
inherent to this reservoir. Microfractures could be the
NRe = 4 (b/2) ρ V C2/µ, (3)
equivalent of the matrix porosity while macrofractures become
to be the main carrier paths.6-8 Where,
V = fluid velocity, ft/s
Discrete-Fracture Hierarchy at the Wellbore C2 = constant, (124)
Image log interpretation was used to determine the open, µ = fluid viscosity, cp
semi-open and closed fractures intersecting the wellbore and Jones9 experimentally determined in the laboratory, the
the approximate aperture of the fractures. Fractures were fracture-roughness behavior and the relationship between the
classified into four types according to the following arbitrary Reynolds Number, NRe, and roughness/aperture ratio (θ/b)
range of apertures: with the friction factor.
Aperture < 0.01 mm 1 gray Although the measurement of the roughness/aperture ratio
0.01 < aperture <0.10 2 light green and the Reynolds numbers is the desirable approach, because
0.10 < aperture < 0.30 3 dark green of the lack of experimental data for the cores in this field, we
0.30 < aperture 4 red used the relationships published by Jones9. Flow rate
Gray fractures are defined as the preexisting ones with no calculation using the cubic equation provides the first
preferential direction, red fractures are the main fractures with approximation that is used to calculate the turbulence based on
preferential direction perpendicular to the field structure, and the fluid velocity and Reynolds number relationship. An
green fractures are the secondary fractures that are assumed to iterative process, as Jones suggested in his paper, continue
terminate at the intersection with a higher hierarchy fracture until the sum of the individual rates of the fractures matches
(see the outcrops section). Open and partially open fractures the well production test results. The observed pressure drop
interpreted from the image logs in the vertical well are illustrated in during the well test is assumed to be governed by the fracture
Fig. 2. It was observed that the fracture density had an inverse network within the radius of influence of the well, which
relationship to the fracture aperture; that is, higher fracture density
defines a length for all the discrete fractures. During the
was observed for smaller fracture aperture. In fact, the fracture-
aperture-width distribution obeyed a power-law, fractal model.
calculation, the fracture apertures are changed as well as the
number of fractures until the sum of the discrete fractures
Fluid Flow through Discrete Fractures matches with the production test data. Plotting the production
Fracture flow calculations through single fractures can be rate for each individual fracture against the depth of fractures
modeled based on laminar flow of fluids through parallel and the summing of the rates from the bottom to the top of the
plates and a well-known equation derived by Poiseuille, often producing interval, provides a synthetic production profile
called as the cubic equation or Poiseuille Law. with the total rate at the top of the interval (Fig. 3). It is
desirable to have this profile from a production test to improve
q = C1 hf ∆P ωf3/µL, (1) the fracture calibration but there was no such test in this well.
Where, However, the approximate synthetic profile is useful to
q = fracture flow rate, bbl/day understand the fracture behavior at the wellbore. Because
C1 = conversion factors constant fractures intersecting the wellbore are interconnected with the
hf = fracture height, ft other fractures within the drainage radius of the well, it is
∆P = pressure drop through fracture, psi necessary to investigate the influence of the fracture
distribution in the network in more detail.
ωf = fracture width, mm
µ = fluid viscosity, cp
Discrete Fracture Network (DFN) Model Construction
L = fracture length, ft
The discrete fracture network concept was used to understand
Because Poiseuille Law does not take into account the fracture
and quantify the fracture model properties (fracture
roughness, the calculated flow rate results are optimistic.
orientation, aperture width, fracture length, and fracture
Jones9 experimentally determined that fracture roughness
distribution in the field) and to use this knowledge in the
plays an important role causing turbulence and additional
design of the sidetrack Well X1ST to improve the productivity
pressure drop. His equation, which includes friction factor due
of the vertical well.7 The information obtained from the
to roughness, is: 9
sidetrack Well X1ST established that the horizontal section of
q = C hf (∆P b3/f Lf ρ) 0.5, (2) the well was to be oriented towards the azimuth 25°,
perpendicular to the main fracture planes with the largest
Where,
apertures. This orientation was expected to cause the
q = flow rate through single fracture, bbl/day
horizontal well to intersect more main fractures which are
C = conversion factor (5.06x104)
hf = fracture height, ft expected to be the most productive fractures.
Several important fundamental concepts prevailed in the
∆P = pressure drop, psi
construction of the DFN model of the Margarita Field: 7 a) the
b = fracture aperture, in
overpressure phenomenon, which generated the preexisting
SPE 91940 3

fractures, b) the translation phenomenon of the reservoir assigned a smaller fracture aperture on the basis that there is a
bodies, which generated new fractures and reactivated the smaller set of fractures not detected by the logs but present in
preexisting fractures and enlarged them by coalescence, and c) the fracture network. Fracture porosity and permeability of a
the outcrops analog model, which was used due to several grid cell was determined according to the aperture size of
corresponding events related to the field structure. fractures in the cell. Core data micro-analysis provided the
Additionally, well information, such as well log image fracture degradation quality index (DQI) with depth, which
interpretation, core analysis, sidewall core microanalysis, and allowed modification of the fracture properties (based on
the geomechanical interpretation were used to correlate the aperture) with depth for each grid cell in each layer. The
fracture distribution. The outcrops study performed in the loading sequence GR-LG-DG-R (Fig. 6) resulted in a high-
Abra del Condor region8, is considered applicable for fracture permeability system dominated by the red fractures. Because
modeling for this field because of the similarity of the of the high permeability of the red fractures and their large
geometry and structural style between the Margarita field length, flow was predominantly linear. The loading sequence
structure and the outcrops.7, 8 R-DG-LG-GR resulted in a very-low-permeability system
The main event observations were noted along the scan dominated by the gray fractures of small aperture. Several
lines over the detailed zone with an area of 500 m x 500 m. combinations of loading sequence were tried and,
These events were classified in major, medium, and minor subsequently, flow simulated until the model permeability
fracture events, color-coded with red, green and black, approached the permeability obtained from the well tests. The
respectively, in Fig. 4. Red color events register large fractures final selected sequence was GR-DG-R-LG.
or faults with large offsets up to 7 m (1 to 2 m average), and
happen with a mean periodicity of 76 m.8 The green color Tuning Fracture Properties with Well Test
events correspond to fractures of smaller scale and are Flow rate history during the well test (Fig. 7) was input for the
associated with the red fractures as “splays”. They would be numerical simulation. Simulation experiments were conducted
associated to shear movements of the major fractures. Green- with this gas rate target to obtain the pressure response as a
fracture sets have very low offsets and an average periodicity function of time for the 2m x 2m simulation grid (Figs. 8 and
of 2.7 m. Another characteristic of these events is that green 9). The simulated pressure responses were analyzed with a
fractures form an approximately 30° to 40° angle well test interpretation package comparing both pressure and
(counterclockwise) with major fractures (red). When these sets derivative behavior with the observed data. The analysis of the
of fractures intersect, the red fractures become dominant and green simulated responses provided the system permeability,
fractures disappear. The length of the red fractures is also larger than formation damage, and fracture system parameters, λ and ω.
that of the green fractures. The black fractures are considered of Type Comparison of these results with the numerical model input
I, or pre-deformation fractures. Geometric distribution of fracture
provided a good understanding of several fracture-network
network in the outcrops was plotted to represent the spacing laws
governing each type of the fracture sets. Most of the natural fractures characteristics discussed below.
observed in the outcrops were sub-vertical (high inclination angle) To approach the observed well test estimates of
with the maximum frequency trend oriented to the azimuth 015°- permeability, λ, and ω, it was necessary to decrease the
025°. fracture permeability, increase the fracture porosity, and
The fracture propagation process followed the increasing the fracture compressibility. Fracture permeability
geomechanical rules and the geometric distribution laws was found to be optimistic and significant reductions were
applied for each of the fracture aperture sets as well as the applied. Effective fracture permeability in the numerical
deformation premises of the geologic model, outcrop data and model is expressed by:
well data. Due to the CPU limitations, the DFN model limits
kef = kf*φf. (4)
the sets of fractures only to the surrounding area of the
horizontal section of the well MGR-X1. Fracture porosity not only plays an important role in the above
expression but is also important for the storativity calculation
DFN Numerical Simulation Model by the following equation11:
The discrete fracture network model was gridded on a dual
ω= (φct)f /[(φ ct)f+(φ ct)m]. (5)
porosity numerical simulation model with a 2m x 2m cell size
and approximately 10 m length thickness for each of the 9 Fracture porosity was increased within its realistic ranges.
layers penetrated by the side track well, making a total of 1.5 Fracture compressibility was found to play a very important
million-cell model (Fig. 5). The matrix has constant properties role in tuning the fracture storativity, ω. Since the fracture
per layer and for the fracture system. The DFN model porosity had an upper limit, increasing the total fracture
provided four sets of color-coded vertical fracture distributions compressibility, ctf, was the only way to achieve the high ω
(each set with different properties as a function of depth) to value observed during the well test.
resemble sub-vertical fractures of the corresponding hierarchy. The total fracture compressibility is expressed by the
A grid cell was allowed to contain only one fracture type. following relationship:
Assigning the fracture properties to each fine-grid cell was
ctf = cf + (Sg cg)f + (Sw cw)f . (6)
accomplished by loading the different fracture sets in a
successive order. Whenever two different types of fractures The fracture compressibility, cf, in Eq. 6 can be approximated
entered a cell, the loading order followed the sequential by
hierarchy to discard the earlier and leave the latter. If none of cf = (ν/E) (1/φf), (7)
the four sets of fractures entered a given cell, the cell is
4 SPE 91940

Where by the logs and the small fractures play an important role in
cf = fracture compresibility, psi-1 transferring the fluids from the matrix to the fracture network.
ν = Poisson ratio
E = Young’s modulus, psi Upscaling the Fracture Model
φf = fractura porosity, fraction. Preliminary upscaling experiments were conducted with the
Typically for a Poisson’s ratio of 0.25, Young’s modulus of reservoir outer cells from 2 m to 10m (see Fig. 5). Geometrical
106 psi, and a mean fracture porosity of 0.001, the fracture upscaling was applied to these external cells. The simulated
compressibility, cf, is 3.9x10-4. pressure and derivative responses indicated lower fracture
Then, Eq. 6 indicates that the total fracture compressibility permeability. Therefore the fracture permeability was scaled
is dominated by the gas compressibility and fracture down by a factor of 2.5 by trial and error to maintain the
compressibility. Because the reservoir is overpressured, the network characteristics reflected within the fine grid area.
gas compressibility and thus the total fracture compressibility Because larger cell sizes would be required for the large-
may be high enough to yield an ω value (Eq. 5) that matches scale field simulation, upscaling for 32-m, 64-m, and 96-m
the observed value from the well test. This conclusion may cell sizes were tested. The near-well grid cells were
have very important implications for the future reservoir- maintained at the original size in order to preserve the dual-
depletion-behavior studies since potential fracture closure may porosity signal near the well. Fig. 11 shows the scaled up 32-
affect the future permeability. mx32-m grid.
We modeled the variability of the matrix to fracture inter- The fracture porosity was upscaled by arithmetically
porosity flow parameter, λ , by changing the shape factor, σ , averaging the cells subject to the upscaling rules. The fracture
and km/kf ratio. The shape factor was calculated by using permeability, on the other hand, is a scale-dependent property
Kazemi’s definition: 10 and several upscaling methods were applied to determine the
degree of distortion. Arithmetic, geometric, and dynamic
σ = 4 (1/Lx2+1/Ly2+1/Lz2), (8) averaging methods were used for each layer. No attempt was
Where made to reduce the number of layers.
σ = matrix blocks shape factor Simulation runs indicate that the geometric averaging
Lx, Ly, Lz = block size dimensions. method resulted in pessimistic values, less than 10 times the
This definition uses the dimensions of the matrix blocks for original fracture permeability of the fine grid model. The
the largest fracture sets (red fractures). The dimension of the arithmetic averaging method resulted in optimistic values,
matrix blocks estimated from this modeling study was 20 m, more than 30 times the original fracture permeability of the
which is consistent with the average distance of the red fine grid model. The dynamic averaging method was
fractures in the DFN model. determined to be the best approximation. The larger the
upscaling ratio, the better was the calculated fracture
λ = σ rw2 km/kf , (9) permeability from the simulation match results. For the
where upscaled grid sizes of 10 m and 32 m, the correction was 2.5
λ = inter-porosity flow parameter, dimensionless and 1.4 times the original value, respectively. For the grid
rw = wellbore radius, ft sizes of 64 m and 96 m, the upscaling did not required
km = matrix permeability, md correction. This can be explained by the fact that the fracture
kf = fracture permeability, md. connections are more critical in fine grids than in larger grids.
Any changes on all parameters that are part of the model have Besides the comparison of the average fracture
to be checked to maintain consistency between the analytical permeability, long-term runs were performed to evaluate the
solution and the numerical model. rate and pressure variation during depletion. Fig. 12 shows
Formation damage was tuned by reducing the permeability that very small deviation of production rates was found
near the wellbore and excluding 30% of the fractures between the fine grid and the upscaled grid models. Fig. 13
intersecting the wellbore. Of the 400-m open hole exposed to illustrates the cumulative production during the depletion
flow, only 70% was allowed to produce. It was found also that period. Similarly, as shown in Fig. 14, pressure variation at
no more than one cell (2 m) should be affected by reducing its several points, emulating interference tests, indicate that the
permeability to emulate the formation damage; otherwise the largest variation is in Layer 10 (first layer for fractures). These
behavior at early time would change in a way not observed in tests ensured that not only the estimated properties are in
the test results. agreement but also the rate behavior, cumulative production,
Finally, the smallest fracture sets filling up all cells not and pressure distribution during depletion satisfy the
intersected by any of the four fracture sets provided by the expectations.
DFN model were found to be of larger permeability than that
originally assigned. These fractures have lower porosity and Conclusions
permeability than the gray fractures. They play an important This work indicates that a naturally fractured reservoir may be
role in connecting the matrix with gray and larger fractures. adequately represented with a matrix system and a discrete
Without these fractures, the system could not recover its fractures network (DFN) in a highly detailed, dual-porosity,
pressure within the buildup period of the well test. This fine-grid model. This work also shows that the DFN modeling
indicates that the field has many small fractures not detected is a good starting point to characterize naturally fractured
reservoirs but it is essential to calibrate the fracture properties
with transient well tests of short and long term.
SPE 91940 5

Calibration of the DFN model against well test data 7. Zapata, T. and Araujo, H.: “Discrete Fracture Characterization
established that: applied to the Margarita field in Bolivia”, AAPG VIII Simposio
a. Red fractures in this model are of smaller length than Boliviariano Petroleum Exploration in the Subandean Basins
that originally thought. “Unraveling the Potential” September 21-24, 2003, Cartagena,
Colombia.
b. Preliminary fracture properties for different sets (red, 8. Florez J., Aydin A., and Antonelli M., “Characterization of
dark green, light green, gray, and background) based fracture faults in Huamampampa, Icla and Santa Rosa
on the apertures provided optimistic results and it was formations, near Tarija: A surface analog to Margarita
crucial to calibrate them against dynamic data from Reservoir, Rock Fracture Project Stanford University report for
well tests. Repsol-YPF, 2001.
c. Although the major, large fractures are main conduits 9. Jones, T.A., Wooten, S. S., and Kaluza, T. J.: SPE 18175,
for flow, the smaller, minor fracture sets appear to be “Single-Phase Flow through Natural Fracture,” SPE 63rd Annual
not only more abundant than originally estimated but Technical Conference and Exhibition, Houston, TX, Oct. 2-5,
also more important for the fluid exchange between 1988.
10. Kazemi, H., Merrill, L. S., Porterfield, K. L. and Zeman, P. R.:
the matrix and the fractures. “Numerical Simulation of Water-Oil Flow in Naturally
The high storativity value determined from the well test Fractured Reservoirs,” Soc. Pet. Eng. Jour. (Dec. 1976) pp.
(ω=0.7) was largely due to the influence of high total 317-326.
compressibility of the fractures dominated by the gas 11. Warren, J.E. and Root, R.J.:”The Behavior of Naturally Fractured
compressibility of the overpressured reservoir. Reservoirs” SPEJ, (Sep 1963) 245-255.
The upscaling studies indicated that it is possible to 12. Gilman J.R. and Kazemi H. “Improvements in Simulation of
represent a detailed DFN model with a large-scale grid model Naturally Fractured Reservoirs,” Soc. Pet. Eng. Jour. (Aug.
as long as the upscaling process is performed in a dynamic 1983) pp.695-707.
fashion and preserving the matrix to fracture transfer function,
σ, of the fine grid model.
A procedure has been established to validate the DFN
models with dynamic tests using an appropriate fine-grid,
dual-porosity simulation model. The characterization can be
improved as new dynamic data from more wells become
available for use in calibration. This indicates that the
characterization of natural fractures is an iterative process.

Acknowledgments
The authors would like to thank Repsol-YPF, operator, and
British Gas (BG) and Pan American Energy (PAE) partners
for the permission to publish this paper.

References
1. Horne, R. and Raghavan, R.: “Reservoir Description and
Dynamics,” JPT (March, 2004), pp. 32-33.
2. Liu, Y., Harding, A., Abriel, W. and Strebelle, S.: “Multiple-point
simulation integrating wells, three-dimensional seismic data, and
geology,” AAPG Bulletin (Month, 2004) pp. 905-921.
3. Dershowitz, W., LaPointe, P., Eiben, T., and Wei, L.: “Integration
of Discrete Fracture Network Methods with Conventional
Simulator Approaches,” SPE Res. Eval. & Eng., (April, 2000)
pp. 165-170.
4. Adibrata, B. W. H. and Hurley, N. F.: “ Flow-unit modeling using
neural networks, logs, and cores in a vuggy dolomite reservoir,
Dagger Draw Field, New Mexico,” SPWLA 44th annual
Logging Symposium (June 22-25, 2003) .
5. Dreier, J., Ozkan, E. and Kazemi, H.: SPE 92039, “New
Analytical Pressure-Transient Models to Detect and
Characterize Reservoirs With Multiple Fracture Systems,” SPE
International Petroleum Conference, Puebla, Pue., Mexico, 7-9
November 2004.
6. Araujo, H and Clivio, J.L.: SPE 69509, "Tectonic Influence in
Productive Potential of the MGR-X3 Well in the Margarita
Field, Bolivia, SPE Latin American and Caribbean Petroleum
Fig. 1 – Location Map of Margarita Field, Caipipendi Block,
Engineering Conference, March25-28, 2001, Buenos Aires, Bolivia.
Argentina.
6 SPE 91940

Fig. 2 – Fracture planes intersecting the wellbore at Well

MGR-x1, Huamampampa formation, Margarita Field.

Fig. 4 – Scan lines on the Abra del Condor outcrops

4380

4400

4420

4440
Depth (m)

4460

4480

4500

4520

4540
0 2 4 6 8 10 12 14
Production Rate (MMscf/d)

Fig. 3 – Actual flow rate (red) from individual fractures versus Fig. 5 – Grid System: Fine-grid, 2m x 2m; coarse-grid, 10m x 10m.
depth, and synthetic cumulative production (blue) versus depth .
SPE 91940 7

Pressure, bars

Depletion
from layer 7
completions

Depletion from upper

layer completions

Fig. 6 – Fine-grid fracture population in Layer loaded in the GR- Fig. 9 – Pressure distribution showing elliptical drainage shape
LG-DG-R sequence. Gray fractures appear in white color. due to the fracture anisotropy.

Log(perm)

Fig. 7 – Flow rate history of the well test and the last pressure
build-up period used for model comparison.

Fig. 8 – Actual pressure response and derivative data from the Fig. 10 – Fracture permeability distribution. Light green
well test (black and blue lines) and numerical simulation pressure corresponds to the smallest fractures with permeability lower
and derivative (red and blue dots). than that of gray fractures).
8 SPE 91940

Fig. 11 – 32m x 32 m outer coarse-grid with 2m x 2m near-well fine

grid.
Fig. 13 – Long-term depletion cumulative production for the fine-
and coarse-grid models.

Fig. 14 – Pressure change during the long-term depletion at Point

Fig. 12 – Long-term depletion flow rate for the fine- and coarse-
3 for the fine- and coarse-grid models.
grid models.