SPE 102165
Characterisation and Modelling of a Fractured Reservoir Using a Novel DFN Approach
F. Gouth and A. Toublanc, Total, and M. Mresah, CPTL
Copyright 2006, Society of Petroleum Engineers
This paper was prepared for presentation at the 2006 Abu Dhabi International Petroleum
Exhibition and Conference held in Abu Dhabi, U.A.E., 5–8 November 2006.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
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
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
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than
300 words; illustrations may not be copied. The abstract must contain conspicuous
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.
Abstract
Characterising and modelling of naturally fractured reservoirs
(NFR) with fracturing at different scales is usually a
challenging task as the specific response from each scale is
difficult to isolate.
We focus on a carbonate reservoir in North Africa, in
production for two years. There is evidence of fracturing at
different scales (from diffuse fractures to conductive faults)
with significant implications for oil production.
We first present the way fractures have been fully
characterised using an extensive integration of static (FMI)
and dynamic (Well test, Pressure Build Up, Mud losses) data.
The advanced use of well test signatures in understanding the
main flow mechanisms occurring within the reservoir is
emphasized.
Then, we detail the way the dynamic model was build
using a novel discrete fracture network (DFN) approach
developed internally. The method populates the static
parameters at the full field scale using a geo-statistical process
guided by a geological driver. Using an innovative technique
(the Elementary Patch), a fast calculation of fracture
permeability in each grid cell and automatic calibration on the
well test data can be made. We also discuss how the best
choice for the simulation model (single porosity vs double
porosity) depends not only on the fracture-to-matrix
permeability ratio but also on the magnitude of the shape
factor and on the production mechanism. The final dual
porosity–dual permeability model integrates the intermediate
scale fractures implicitly and the conductive faults explicitly
whereas the smallest scale is discarded as it has a negligible
impact.
In summary, we present a complete and innovative
workflow process for a geoscience study on NFR and
demonstrate its application in a real case study. The need for
full data integration to achieve a comprehensive understanding
of the flow mechanisms involved during production is
underlined.
Introduction and scope
The full workflow of a fracture study from characterization to
modeling on a field operated by TOTAL in Africa is described
in this paper. Particular attention is given to the integration of
static and dynamic data and the way it allowed us to achieve a
satisfactory characterization of the fracture network. A new
DFN tool developed by TOTAL has been used for the
modeling of this field. For confidentiality purposes, the field
will be designated as reservoir ‘R’ in the paper.
The field is an oil and gas accumulation forming a WE
anticline truncated to the East by a fault (Fig 1). It is located in
a complex folded and faulted transfer zone, active during the
reservoir deposition, of Eocene age. The morpho-structural
pattern of the transfer zone is constrained by N150 normal
faults with possible salt diapir activity paralleling the rifts and
bounding 3 sedimentation domains. They cut folds with N80
axes partially interpreted as reactivation of previous normal
faults. The western domain of the transfer zone is constrained
by tectonic activity during the Eocene reservoirs deposition
with:
• High subsidence under open lagoonal and marine
influence characterized by thick Upper ‘R’ Nummulitic
limestone facies limited to paleo-highs generated by the
growing anticlines.
• Thin, low energy, bioclastic limestone and dolomite in
Lower ‘R’.
The field was appraised by three wells. W01, drilled
vertically on the crest, found gas in the Upper ‘R’ and oil in
the Lower ‘R’. W02, drilled vertically on the flank, was water
bearing. W03, drilled slanted and up dip of W02, crossed the
WOC and triggered the decision to develop the field since it
proved an overall oil thickness of 220m and exhibited
characteristics even better than those encountered in W01. The
total reservoir thickness is around 250m and most of the oil
lies in the Upper ‘R’. The dip varies from 4° westward to 30°
on the northern and southern flanks. Rock data show better
properties in the Upper (φ=17%, k=30mD) than in the Lower
(φ=12%, k=5mD).
This field has been on stream since 2003. It is currently
producing at a plateau of 40000 bopd through 5 horizontal and
1 vertical wells. Additional wells and gas re-injection are
planned. In this context, a good characterization of the
fractures was deemed critical.
2 SPE 102165
Fracture characterisation
In Naturally Fractured Reservoirs, characterization of the
fractures is a difficult task that requires a rigorous workflow
which can be summarized by the following steps.
• First, fractures sets with different orientation and scale
have to be identified. Scales typically range from diffuse
fractures at one extreme to fracture corridors and
conductive faults at the other.
• Then for each fracture set, static parameters i.e. length
and spacing distributions have to be defined. Well
imagery (FMI) is the most effective tool for assessing
spacing distributions but lengths can not be measured
directly in the reservoir. An outcrop analog is often the
best way to get an idea of this parameter.
• Finally, the conductivity of each fracture set must be
estimated. This is usually the most difficult task. MDT is
the only tool enabling a selective fracture measurement
but it is often ineffective as radial flow within the fracture
is concealed by the storage effect. Well test data are the
most effective tool to assess the overall dynamic response
of a fractured reservoir but it is difficult to separate the
individual conductivity of each fracture set.
Static data. For reservoir ‘R’, all the available static data
were reviewed to build a conceptual fracture model. These
included outcrops, well imagery and conventional logs.
Scales on outcrop. Very good outcrops of this Lower
Eocene Carbonate Platform have been found. They show 3
main scales of deformation:
Bed Controlled Fractures (BCF): particularly well
developed in the multi-layered Lower ‘R’ formation, the BCF
are second-order heterogeneities in the thick (200m) and
hyper-compact Upper ‘R’.
Highly Persistent Fractures (HPF): these fractures have a
large horizontal extension (up to one kilometre) and cut the
platform over its whole thickness (> 200m). On the aerial
photographs, the mean spacing between HPF appears to range
from 10 to 50m (2 or 3 sets of Highly Persistent Fractures
generally co-exist.) These HPF are believed to exist in
reservoir ‘R’ and if open, they could act as gas/water conduits
during production.
Faults: only a few faults have been observed on the
outcrop. All these faults are normal faults. Vertical throw is
generally small, always less than 20 meters (i.e. probably sub-
seismic in the reservoir). 10 to 50 cm tectonic breccias always
are associated to the fault planes. As these breccias are poorly
cemented, they could be also efficient gas/water carriers
during oil production.
No “fracture corridors” (i.e. dense clusters of large
extension fractures) were observed on the outcrop. Fracture
clusters have been described in some reservoir R wells (W04
for instance), but these are believed to be in fact Bed
Controlled Fracture sets.
Well imagery. Image log data was available for 8 wells, 3
of them calibrated on cored intervals. Core and image log
studies reported 3 main types of fractures:
• Highly persistent, thick, resistive cemented fractures (20-
40%), either scattered or in small clusters, and often
related to micro-faults that are also cemented.
• Small extension, open, conductive fractures usually
confined in tight beds, in clusters of 10-20ft. These
discontinuous conductive fractures represent 60 to 70% of
the overall natural fracture population.
• Large (decimeter scale in cores) conductive, open, highly
dipping fractures in echelon, scattered in clusters of
discontinuous fractures or alone in more porous facies.
These form 20-30% of the natural fracture population.
Induced fractures are frequent in cores and images on the
vertical wells. Breakout and induced fracture orientation is
related to present day maximum horizontal stress, N160-N170,
in agreement with the preferential wellbore elongation
detected by SCHLUMBERGER CLIBAN module run in wells
W04, W05 and W06. Natural fractures parallel to this
orientation are more likely to be widened by the drilling
process.
Natural fracture orientations found by cluster analysis on
the full data set for reservoir ‘R’ are:
• N70: 31% of all fractures, 77% of which are cemented.
• N150 to N180: 52% of all fractures, mostly open.
• N110: 17% of all fractures, mainly in W08 but some in
W03 & W07.
In terms of fracture density, cemented fractures are either
clustered (10 fractures/ft) in micro-faulted zones or scattered
elsewhere with intervals of 300 to 500ft (up to 1300ft at W08)
without fractures.
Open fractures are mainly clustered (10-15 fr/ft up to
30fr/ft). Clusters are 10-30 ft thick, usually made of
discontinuous fractures, but occasionally of 100 to 200ft.
Continuous fractures (10 to 50%) are scattered in the density
picks or individually in porous facies. Intervals without
fractures range from 50 to 300ft.
Fracture spacing has a reverse exponential distribution
with minimum spacing around 0.5ft, most of the values under
5ft and maximum spacing around 200ft but up to 500ft.
In terms of lithology, best fracture density picks are
usually related to tight stringers but some thick tight levels (at
bottom Upper ‘R’/ top Lower ‘R’) are not or only partially
fractured. Density picks of diffuse, discontinuous conductive
fractures are confined to tight beds while more continuous
fractures are scattered in previous density picks or more
individually in more porous facies.
Upper ‘R’ (more porous) is less fractured overall (0.15
fr/m) than Lower ‘R’ which has thick tight levels (0.25 fr/m).
Static conceptual fracture model. From the above, it is
possible to divide the deformation features into 3 classes (from
the largest to the smallest):
• Faults (seismic and sub-seismic scale).
• Highly persistent fractures (HPF): defined from the
frequency logs as isolated fractures with no other
fractures within 10 ft on either side. Although they seem
to cross the entire platform from outcrop analog, they
could be confined to the sub-units (Upper or Lower ‘R’)
considering the rock mechanical differences based on
porosity logs.
• Bedding confined fractures (BCF): characterised on
image logs by fracture clusters located in thin interbedded
porous and tight rocks. This mechanical context is very
favourable for the development of highly fractured beds.
SPE 102165
Three orientations (2 major and 1 minor) are interpreted
from the image logs:
• N150: regional “background” fracturing that is always
present. These fractures are sub-parallel to present day
maximum horizontal stress and almost always open. They
have a fairly homogeneous density throughout the field
with an average spacing of 300ft.
• N70: these fractures are parallel to the axis of the
structure. They are interpreted as fold related fractures as
their distribution shows a fairly strong variability with
structural location. The spacing ranges from 20 to 150ft.
They are 90% cemented at BCF scale and 60% at HPF
scale.
• N110: This set is a minor one and seems to develop
locally (SW). Although its origin is still unclear, it could
be related to some major faults. They are mostly open.
These 3 orientations are present in both BCF and HPF sets.
Dynamic data. Dynamic data have been carefully looked at
to check the consistency of the previously defined static model
and to characterize the dynamic behaviour of the different
fracture sets. The dynamic data consist of well test (DST) and
PBU data, production data, production logging tool (PLT)
data, core data measurements and mud loss data. They provide
insights into the relative conductivity of each fracture scale.
DST and PBU. DSTs were performed on two of the
vertical exploration-delineation wells, in both reservoir units
(Upper and Lower). No dual porosity effect can be seen on the
tests and the interpreted kh values suggest slightly larger
permeability than those measured on cores (within a ratio of
about 2). At the time the field was developed there was thus
little information on the possible dynamic impact of the
fractures.
The field has been developed with 500 m long horizontal
drains mainly targeting the Upper R unit of better
productivity. Two wells target both the Upper and Lower
units. The wells have been completed with 7’’ liners and up to
10 intervals were perforated along each drain. The initial
completion strategy carefully avoided opening the wells in
fractured zones identified on well imagery. The minimum
stand off between the completion and the fractures was about
30ft. This strategy was aimed of course at avoiding direct
communication with the fracture network and early gas break
though from the gas cap. Since the start of the production, an
intensive monitoring program has been set up on all the
producers. The well completions are equipped with permanent
down-hole gauges and regular PBU have been recorded. A
screening of the available derivative signatures on the early
data (not yet affected by gas break through effects) shows two
major types of behaviour:
• Some wells show two radial stabilizations (Fig 2): the
early stabilization is interpreted as the vertical radial flow
and is dependant on kv.kh whereas the later one is
interpreted as the horizontal flow and yields in kh.h where
h is the investigated reservoir thickness.
• Some wells show linear flows (preceeded sometimes by
bi-linear) (Fig 3). This type of signature is associated with
wells cross-cutting or lying very close to conductive faults
identified by heavy mud-losses while drilling. The linear
3
flow, which is usually associated with the so-called
‘hydraulic fracture model’, corresponds in the present
case to the flow in the vicinity of the conductive faults.
Matching the derivative enables the lateral extension of
the fault (linear) and its conductivity (bi-linear) to be
assessed. The typical values for reseroir R are
respectively about 3000ft and 100000mD.ft.
Some wells that show massive losses while drilling do not
display linear flow on the derivative. The likely explanation is
that the flow mechanism is a competition between two types
of flow: radial around the drain and linear around the fault, as
illustrated in Fig 4. The signature will show the flow that
dominates, i.e. linear if the fault extension is large compared
to the drain length or radial in the opposite case.
PLT. PLTs were run on all the wells shortly after first oil.
The oil production profiles along the well bore show a clear
relationship between productivity and vicinity of a fractured
zone, i.e. the productivity is not controlled by the matrix but
by the fracture parameters. One example is illustrated in Fig 5.
Well W05 shows a strong productivity decline for the intervals
at the toe. The porosity log on the corresponding interval is
uniform. The HPF observed on well imagery at 11400 ft MD
is controlling the productivity of the nearby intervals: the
further from the HPF, the lower the productivity.
Mud losses. Mud losses are an important source of data for
ranking the conductivity of the different fracture sets
encountered during the drilling of the wells. Major losses have
been monitored while drilling across subseismic faults.
Significant losses have also been reported at some, but not at
all HPFs (Fig 6). No measurable losses were associated with
BCF. This information is of key importance for understanding
the hierarchy of conductivity among the different fractures
scales.
Production data. The production data are also key for
understanding the major contribution of the fractures during
production. The major observations are:
• Productivities of the wells put on stream were better than
anticipated.
• When wells were produced beyond their critical rate, gas
breakthrough was observed after a few hours only,
suggesting a very limited pore volume was swept by the
gas before reaching the wells. This limited pore volume
can only be associated with fractures and is hence a good
indicator of continuity within the fracture network. It is
worth emphasizing that this gas coning effect was
reversible. As the well was choked and oil production
decreased, the GOR stabilised at the dissolved gas ratio
level.
Dynamic impact of the different fractures scales. The
static synthesis has identified three scales of fractures that co-
exist in the reservoir:
• BCF seems to have almost no dynamic impact. No mud
losses reported in tight layers with BCF. Permeability
might be slightly enhanced only.
• HPF: Their dynamic impact seems to be significant
according to the mud losses.
• Conductive Faults: They have the largest conductivity but
they are isolated features.
4 SPE 102165

It is important to emphasize that the dynamic impact of a
given fracture scale at the reservoir stage will be dependent on
conductivity but also on connectivity. As a result, three
quantitative parameters will be of the utmost importance to
characterize the possible impact of the fracture sets:
• The conductivity of the fractures.
• The spacing and length of the fractures, which control the
connectivity of the set.
A good example is provided by the comparison of the
relative dynamic impacts of faults and HPF in reservoir R. The
faults will have a very high local impact but since their
spacing is large, they do not form a connected network and
their overall impact at field scale will be moderate. HPFs
which are much denser and which form a connected fracture
network should have a more significant impact.
The comparison between the core data and the
permeability derived from the tests has not revealed a large
discrepancy between matrix and fracture permeability and thus
is not conclusive in proving the dynamic impact of the
fractures. To better estimate the impact of the HPF, we have
checked the possible correlation between parameters
controlling the fracture distribution and dynamic data. On the
graph below, we have plotted the test kh on the y-axis vs the
HPF N70 fracture frequency (FF) on the x-axis. HPF N150 FF
is uniform throughout the reservoir and its analysis is thus of
limited interest.
KH vs N70 frac density
50000
45000
40000
on test data starts to be significant. The denser the
network, the higher the equivalent test permeability.
This kind of relationship has been already published for small
scale (diffuse) fractures in the North Sea (1).
One may want to split the matrix and frature dynamic
contribution on the above plot. The basic idea is to consider a
roughly uniform matrix contribution of about 15000mD.ft
throughout the field (which is suggested by well data) and to
subtract this value from the test kh. This new relationship (Fig
7) has been used later on for the dynamic calibration of the
fracture conductivities in our DFN model.
To bolster the above analysis, we have tried to simulate the
behaviour illustrated by the cross-plot using a DFN. The
procedure was the following:
• For a given HPF N70 FF, create several (10) geostatistical
realisations of the N70-N150 HPF system and calculate
the equivalent permeability tensor on a 150m by 150m
domain (size of a grid block). The fracture lengths of both
sets were taken as 300m..
• Check the maximum and minimum eigenvalues of the
permeability tensors (kmax and kmin) and use flags to
characterize them: “Non” if kmax=0, “Partially” if kmax>0
and kmin=0 and “Fully” for kmin>0.
• Re-do the process for different N70 FF values (from 0 to
0.15 by 0.01 increments).
The results are plotted on the graph below:
Connectivity function of HPF N70 FF
10
9
8
Nb of realisations

35000 7
6
KH test (mD.ft)

30000
5
25000
4
20000 3
15000 2
10000
1
0
5000
0,04

0,05

0,1

0,15
0

0,01

0,02

0,03

0,06

0,07

0,08

0,09

0,11

0,12

0,13

0,14
0
0 0,02 0,04 0,06 0,08 0,1 0,12 0,14 0,16 FF N70
N70 frac density (frac/m)
Fully Partially Non

The plot clearly illustrates a correlation which indicates
that the N70 HPF is the key fracture set that controls the
dynamics of the field. The plot shows two different trends:
• For N70 FF below 0.07frac/m, the plot is flat with a
plateau of about 15000mD.ft. This kh value is roughly
equal to the kh associated with the matrix according to
core data.
• For N70 FF above 0.07frac/m, the plot shows a linear
increase of kh with fracture frequency.
These observations can be analysed the following way:
• The 0.07frac/m N70 FF is interpreted as the percolation
threshold of the HPF network i.e. the value beyond which
the N70 and N150 HPFs are dense enough to build up a
connected fracture network.
• Below the threshold value, the HPFs are isolated in the
matrix and the equivalent test permeability is
representative of the matrix contribution only.
• Beyond the threshold value, the HPFs form a connected
fracture network and the dynamic impact of the fractures
It can be seen that a steep change occurs in the proportion of
“Fully” flags above an FF of 0.05. which can be considered as
the percolation threshold. Beyond this value, the proportion of
tensors with two non-zero components is large enough to
ensure continuity within the fracture network.
Fracture modeling
The fracture characterization described above has helped us to
understand the key features that control the dynamic
behaviour of the field. We have emphasized the important role
played by the HPF in increasing locally the equivalent
permeability and in adding heterogeneity to the sweep
efficiency.
We emphasize also the need for using a dual porosity
model for modelling Reservoir ‘R’ even though the equivalent
permeability of the fractures is only slightly larger than the
permeability of the matrix. As soon as two phase flows are
involved (gas coning from gas cap and later gas injection), it is
of the utmost importance to be able to reproduce with the
simulator, the differential sweep efficiency between the
SPE 102165
fractures and the matrix. A dual porosity-dual permeability
model has therefore been built. An internal DFN tool named
GoFraK has been used to calculate the fracture permeability
grids. In the next sections, we first detail the main features of
the tool, then we explain its use for the modeling of reservoir
‘R’.
GoFraK. GoFraK is a new DFN developed by TOTAL &
FSS International. The full Workflow, from static to dynamic
modelling has now been available for more than a year.
Previous publication (2,3 &4) shows the use of the tool’s static
module on a field case study. The current paper describes the
application of the full GoFraK workflow for modelling of
reservoir R.
Workflow. The main steps are detailed here after:
• Horizons of the main mechanical units and the grid into
GoCad.
• Import the well trajectory as well as the imagery log data
(strike directions and dip) for each fracture set.
• Process the data (automatic correction of the well
deviation) and define histograms distributions for the
different fractures sets.
• Use the principal component analysis module to define
the best suited driver using seismic data, curvature maps.
GoFraK delivers a full field gridded map with the
principal component and bi-variate histograms for each
fracture set.
• Geostatistical conditional simulation of FF to deliver full
field grids of FF for each fracture set.
• Definition of conductivities and lengths per fracture set.
Uniform values throughout the field are usually assumed
but an option allows for the conductivity to be correlated
with spacing via a power-law model.
• Processing for each grid block of the equivalent fracture
permeability tensor and shape factors.
Main strengths of GoFraK. The main strength of GoFraK
compared to its competitors lies in the use of the Elementary
Patch technique which enables to focus on the smallest
periodic element in the fracture network i.e. the patch is build
for each grid block and the equivalent properties are processed
on the patch only as illustrated below:
This method is effective in:
• Saving simulation time in the processing of the grid.
• Assessing the exact permeability tensor i.e. without any
assumptions on the boundary conditions since the
periodicity of the patch ensures a periodicity in the
boundary conditions.
5
A specific conductivity calibration module is also available
in GoFraK. Condutivities of the different fractures sets are
seldom measured data. Practically, the only available dynamic
information comes from test data. GoFraK’s calibration
module calculates the equivalent test permeability within a
user defined investigation area. The value is compared to the
actual data and GoFraK iterates on the conductivity of the
different fracture sets till a good match is achieved. The
simulation of the test data assumes steady-state conditions and
is not suited to reproduce the transient flow observed during a
PBU or DST. This calibration module is associated with a
sensitivity module that allows checking the influence of the
main input parameters on the equivalent fracture permeability.
It requires a manual match but an automatic inversion module
will be available soon.
Reservoir modelling. As mentioned earlier, the HPF has
the major dynamic impact in the reservoir and it was thus
decided to model this specific scale with the DFN. The HPF
spacing is small compared to the grid block size, so they can
be represented implicitly in the simulation grid using GoFraK.
Conductive faults also have a strong dynamic impact, but
due to their size and spacing were modeled explicitly in the
fracture grid. The modeling was limited to the faults showing
significant mud losses.
BCF have not been represented in the dynamic model
because their dynamic impact is not significant.
Static grid building. As HPFs are thought to have a large
vertical continuity, the reservoir has been divided in two
mechanical units: one for the Upper and one for the Lower R.
Well imagery has been analysed and statistics calculated
separately for each mechanical unit. Three sets of HPF were
previously defined with orientations N70, N150 and N110.
The latter is not well developed and mainly located in the
south-western part of the reservoir. It is assumed to have little
impact and was discarded from the modeling. The Fracture
Frequency distributions for the other sets show:
• A quite uniform distribution throughout the reservoir for
the N150 set which is considered as background regional
fracturing. These fractures are open, oriented in the main
stress direction and hence assumed to have the maximum
conductivity.
• A distribution strongly related to the folding for the axial
N70 set. The maximum curvature turned out to be the best
fracture frequency driver. Fig 8 provides the curvature
map showing the largest values are in the area of rapid dip
variation between the crest and the flanks. Fig 9 shows
the correlation between the fracture frequency and the
curvature established on actual data.
The GoFraK geostatistical module was used to populate
the grid in fracture frequency distributions and several
realizations were generated. A ranking of the realisations was
performed based on the frequency predicted by GoFraK in the
vicinity of the wells where no imagery data was available but
whose kh was used to estimate the N70 fractures frequency.
The N70 fracture frequency grid selected for the dynamic
modeling is given in Fig 10.
Dynamic grid building. In GoFraK, each fracture set is
assumed to have constant length and conductivity parameters
6 SPE 102165
throughout the field. These parameters can not be measured
directly. We use the following as a best guess:
• An average length of 1000ft for both orientation sets. This
data is consistent with observations on outcrops.
Furthermore, it turned out to be consistent with the
percolation threshold observed on actual data (see
Fracture characterization above).
• Initial conductivity derived from the equivalent fracture
permeability from well tests i.e. C=kf/FF. A ratio of 10
has been considered between N70 and N150 sets to
account for likely anisotropy effects related to current day
stress (parallel to N150), giving respectively 3000 and
30000mD.ft.
These input parameters have been processed by GoFraK to
generate permeability tensors for each grid block of each
mechanical unit.
Dynamic grid calibration. The calibration module of
GoFraK has been used to match the test permeability within
the user defined investigation area playing on conductivities of
the different sets. Some slight modifications of the initial best
guess values were required. For W07, it was not possible to
match the low permeability of the test with the fracture
frequency associated with the static grid. It is thought that with
the fracture length considered, the percolation threshold is
slightly too optimistic in GoFraK and results in a connected
network in the vicinity of W07 whereas the actual data suggest
the opposite. However, since the only mismatch was observed
for this well, it was decided to not reprocess GoFraK with
different length assumptions. Instead, local modifications to
the fracture grid were made manually.
The principle directions of the permeability tensors were in
line with the orientation of the grid and hence the principal
components were exported directly to the reservoir simulator.
The tensors were calculated for the two mechanical units i.e.
Lower and Upper: there was assumed to be no vertical
variation within a given unit. The resulting map for the
permeability in the N70 direction is shown in Fig11.
History Match. The production history is short but the
early time data are rich in information in terms of critical rate
and GOR variations. First a match in pressure was performed
by adjusting the aquifer activity and the transmissivity
between Upper and Lower. Then, the GOR variations of the
wells were matched. Refined historical data (on a daily basis)
were required to capture the fast production changes. Since the
permeability field had already been calibrated on the well test
data, the viscous forces are correctly simulated in the reservoir
and so the critical coning rate should also be correct.
However, the rise in GOR depends on the exchange between
the fracture and the matrix. Indeed, two mechanisms compete
within the gas cone developing between the gas cap and the
producers:
• The gas tends to flow from the GOC to the well through
the fracture network. The gas front velocity depends on
kf/φf.
• The gas column within the fracture creates a gravity
drainage mechanism which tends to force the gas into the
matrix and expel the oil. This mechanism mitigates the
differential sweep efficiency between the fractures and
the matrix.
We emphasize that the only way to model this reversible gas
coning mechanism was to use a double pososity model. The
principal task of the history match consisted in tuning the
parameters controlling the magnitude of the gravity drainage
mechanism to match the GOR rises. Those parameters are:
• The matrix-block height. The base case value was taken
as the good matrix layer thickness i.e. about 30 ft (Fig
13). Indeed, when the gas flows down the HPF, it will
also diffuse horizontally within the tight layers where
BCF enhance locally the permeability. The gas saturation
increase in the tight layers will partially disconnect the
porous matrix blocks by creating a baffle zone with
lowered oil mobility. This creates a “dynamic block”
effect where the block height is equal to the thickness of
the porous layers.
• The shape factor, which impacts the exchange rate.
It is important to point out that those parameters will have
the same effect on the GOR and as a result multiple matching
solutions are possible. Gas saturation measurements (TDT
logs) in the gas swept area would help to constrain the match
by adding valuable information on the capillary trapping of oil
in the matrix i.e. the block height. Unfortunately, such
measurements are not possible given the layout of the
horizontal development wells in this reservoir. Since the
uncertainty on the matrix-block height is limited compared to
the uncertainty on the shape factor, it was decided during the
history match to tune the latter parameter only.
A global multiplier for the whole field was applied to the
shape factor values calculated by GoFraK. A decrease by a
factor of 50 was required, which means that the gravity
drainage block size is 7 times larger than the block sizes
calculated by GoFraK, which are based on the geometry of the
fracture network, which in turn depends on fracture frequency.
Though this kind of discrepancy is common in fractured
reservoirs, the analysis of the causes is seldom developed.
Naturally fractured reservoirs are a particular case of very
heterogeneous reservoirs. Fractures and matrix are tightly
imbricated and their properties are very different. One way to
tackle the complex issue of reservoir modeling is to consider a
dual porosity model which represents separately the fractures
and the matrix. This means that properties are modeled
separately and pressure and saturation are handled separately.
Though this represents a huge improvement over a single
porosity model, the heterogeneity commonly observed within
the fracture network itself is only partially accounted for. In
the present case, the assumption that all the HPFs within a
given set share the same conductivities and lengths might not
be realistic. This assumption is made because of the limited
amount of available data. We have some evidence from mud
losses of low conductive HPF but the organization of
conductivity within the reservoir is poorly understood. When
two phase flows are involved, the gas will likely move into a
few HPFs leading to an equivalent dynamic block size much
larger than the static one.
The final match was acceptable (typical result given for
well W03 in Fig 12) and most of the major GOR changes have
been well captured. It is important to point out that GOR
changes are related to oil rate fluctuations around the critical
value. One or two wells have a constant GOR slightly higher
than the dissolved GOR. This may be due to gas channelling
SPE 102165 7

effects the casing. The gas saturations after one year of The major outcomes of this study will help to optimize the
production are griven in Figs 14 & 15 for both the fractures way this newly developed reservoir will be produced. This
and the matrix. They show significant differences in term of includes reservoir management activities and the future gas
sweep efficiency. The average gas saturation in the matrix injection project.
behind the front remains high in the model but this might be
different if a block height other than 30 ft was used. For the Acknowledgments
forecast simulations, a range of 15 to 60 ft has been retained. The authors thank NOC, CPTL, Wintershall and Total for the
For each value in this range, the shape factors have been permission to publish this paper.
readjusted to keep a consistent match of the historical GOR.
Nomenclature
Conclusions f,m Fracture, Matrix
This study has shown how the integration of all dynamic and BCF Bed Confined Fractures
static data is essential for the characterization of a complex C Conductivity of a fracture
fractured reservoir. Characterization of fractured reservoirs FF Fracture Frequency
requires a large amount of data as the level of heterogeneity is HPF Highly Persistent Fractures
very high. Unfortunately, those data are seldom available at k Permeability
the time the field development decision is taken. It is thus φ Porosity
important to phase the development and perform an intensive
monitoring program during the early life of the field. References
Specifically, outcrop data was essential in understanding [1] A. Toublanc & al. Ekofisk Field: fracture permeability evaluation
the overall organisation of fracturing with the reservoir and in and implementation in the flow model. Published in Petroleum
interpreting the well imagery data. The latter provided Geoscience, Vol 11 2005, pp 321-330.
information on fracture frequency and orientation for each set [2] B. Gauthier, M. Garcia. Static and dynamic characterization of
fracture pattern in the Arab reservoirs of the ABK Field,
of fractures within the reservoir itself. Mud loss and PLT data offshore Abu Dhabi: from well data to full field modeling. Paper
showed which fracture sets were conductive locally, but well ADIPEC 10-22, October 2002.
test and PBU data were needed to assess the connectivity of [3] Gauthier & Al (2000), ‘Integrated reservoir characterization: a
the fracture sets. Finally, matching GOR variations with rate case study in North African Field’ SPE 65118.
during the early stages of production provided information on [4] Gauthier B, Garcia M, Daniel JM (2002) ‘Integrated reservoir
the effective matrix block size. characterization: a case study in a North African Field’ SPE
Analysis of linear flow regimes observed on PBU 79105.
derivatives enabled fault conductivities (up to 100000 mD.ft)
and extensions (up to 3000ft) to be inferred. However, the
faults do not form a connected network, so have only a local
impact on production.
A correlation between well test derived kh values and
fracture frequency for the N70 HPF set revealed that it is the
HPF that have the most significant impact on production. The
existence of a frequency threshold for connectivity of the
fracture network was inferred from the test data, consistent
with results from the DFN model.
A correlation between HPF fracture frequency and
curvature was demonstrated and used in GoFraK to generate
fracture permeability for input to a dual porosity-dual
permeability simulation model. The elementaty patch
technique and the automatic calibration module in GoFraK
have considerably improved the traditional workflow and have
been very time effective. BCF have not been represented as
this scale did not show a significant dynamic impact but the
matrix block height of the model accounts for a gas
disconnection effect related to BCF. Conductive faults have
been represented explicitely within the fracture grid itself.
Model history matching was relatively easy thanks to the
calibration against test permeability performed in GoFraK.
The only tuning parameters used for matching the GOR rise
are those controlling the gas gravity drainage mechanism i.e.
the block height and the shape factor. A dynamic shape factor,
significantly different from the static shape factor, has been
obtained. The difference is likely related to the heterogeneity
in the dynamic properties of the HPF fracture set itself
throughout the reservoir.
8 SPE 102165

Fault extension Xf

GOC
OWC Radial flow Linear flow
B07

B3 9
9 5 B1a
B06
BD1 platform
Horiz Wellbore
5

B04

B2 B05
Drain Length L
20
B08 18
Pl an e s
(Stri k e) We i g hte d H i s to g ra m

View from Top to Bottom Fault

Axi a l N = 29

Fig 1: Reservoir R TOP structural map Fig 4: Flow models in the vicinity of wells and faults

W05: intervals PI

PI (bopd/psi)
30,0

25,0

20,0
Upper R
10 -2

Lower R
15,0
DP & DERIVATIVE (PSI/STB/D)

10,0

5,0
10 -3

PD =1/2

0,0
1 2 3 4 5 6 7 8 9 10 11 12 13 14

Isolated fractures
10 -4

« Isolated No Fractures
10 -5

10 -3 10 -2 10 -1 10 0 10 1
Delta-T (hr)

poor image quality

Fig 2: PBU derivative signature without conductive fault

ENDWBS
-1
10

Fig 5: PLT per interval along W05 – Imagery logs

DP&DERIVATIVE(PSI/STB/D)

Fractures Flow in - Flow out

200 Clustered fractures (BCF) in bedding Isolated N150 fracture

-2
10

150 No mudlosses Mudlosses

100
50
0
-3
10

-50
-100
-150
-4

-200
10

14050

14100

14150

10 - .5 10 .5 10 1.5
Delta-T (hr)

Fig 3: PBU Derivative signature with conductive fault Fig 6: Mud losses log
SPE 102165 9

KH vs N70 frac density

35000

30000

25000
F
KH frac (mD.ft)

20000 *

15000

10000

5000

L o w H ig h
0
0 0,02 0,04 0,06 0,08 0,1 0,12 0,14 0,16
N70 frac density (frac/m)

Fig 7: KH fracture vs N70 HPF FF Fig 10: Fracture frequency grif for N70

High Low

Fig 8: Maximum curvature maps Fig 11: Axial fracture permeability map

Correlation coeff =0.73

Actual data

Simulation

Fig 9: N70 HPF Fracture Frequency vs maximum curvature Fig 12: GOR match of well W03
 10 SPE 102165

L
Hectometric

Block height P
BCF

HPF

Fig 13: Matrix block height

Fig 14: Gas saturation in the matrix fater 1 year of production

Fig 15: Gas saturation in the fracture after 1 year of production