You are on page 1of 12

CONTENTS

Quantifying and predicting AUTHORS


Creties Jenkins  DeGolyer and MacNaughton,
naturally fractured reservoir Dallas, Texas; cjenkins@demac.com
Creties Jenkins is a senior staff geologist for De-
behavior with continuous Golyer and MacNaughton where he specializes
in reservoir characterization, geocellular modeling,
fracture models and resource estimation in clastic reservoirs, in-
cluding coalbed methane and shale gas accumu-
lations. He received an M.S. degree in geology and
Creties Jenkins, Ahmed Ouenes, Abdel Zellou, and a B.S. degree in geological engineering from the
Jeff Wingard South Dakota School of Mines and Technology.

Ahmed Ouenes  Prism Seismic, Greenwood


Village, Colorado; ouenes@prismseismic.com
ABSTRACT Ahmed Ouenes is the president of Prism Seismic.
Previously, he was the chief reservoir engineer at
This article describes the workflow used in continuous fracture
(RC)2 where he developed the first commercial
modeling (CFM) and its successful application to several projects.
software for the CFM technology. Ahmed’s main
Our CFM workflow consists of four basic steps: (1) interpreting key interest is the development of improved reservoir
seismic horizons and generating prestack and poststack seismic attri- characterization technologies especially for frac-
butes; (2) using these attributes along with log and core data to build tured reservoirs. Ahmed graduated from Ecole
seismically constrained geocellular models of lithology, porosity, water Centrale de Paris and holds a Ph.D. in petroleum
saturation, etc.; (3) combining the derived geocellular models with engineering from New Mexico Tech.
prestack and poststack seismic attributes and additional geomechan- Abdel Zellou  Prism Seismic, Centennial,
ical models to derive high-resolution three-dimensional (3-D) fracture Colorado; abdel@prismseismic.com
models; and (4) validating the 3-D fracture models in a dynamic res-
Abdel M. Zellou is director of consulting at Prism
ervoir simulator by testing their ability to match well performance. Seismic. He has worked as a consultant on nu-
Our CFM workflow uses a neural network approach to integrate merous fractured reservoirs all over the world
all of the available static and dynamic data. This results in a model and contributed to the drilling of many successful
that is better able to identify fractured areas and quantify their im- wells. He codeveloped ReFract, a leading frac-
pact on well and reservoir flow behavior. This technique has been tured reservoir software using patented technology.
successfully applied in numerous sandstone and carbonate reservoirs Abdel graduated from New Mexico Tech with an
M.Sc. degree in petroleum engineering.
to both understand reservoir behavior and determine where to drill
additional wells. Three field case studies are used to illustrate the Jeff Wingard  DeGolyer and MacNaughton,
capabilities of the CFM approach. Dallas, Texas; jwingard@demac.com
Jeff Wingard is a senior staff reservoir engineer at
DeGolyer and MacNaughton where he has devel-
INTRODUCTION oped and evaluated geocellular and simulation
models for waterflood, miscible gas, and thermal
The characterization of naturally fractured reservoirs is a recurring Enhanced Oil Recovery projects. He earned a B.S.
degree in chemical engineering from the Massa-
challenge for oil and gas companies. Reservoir models built to assist
chusetts Institute of Technology in 1980 and a Ph.D.
with field development planning and depletion optimization need to in petroleum engineering from Stanford University
accurately incorporate the effects of natural fractures in the near- in 1988.
wellbore regions and to also predict their distribution in interwell
areas.
ACKNOWLEDGEMENTS
The AAPG Editor thanks the following reviewers
for their work on this paper: Soren Christensen,
Copyright ©2009. The American Association of Petroleum Geologists. All rights reserved. Stuart D. Harker, Tony Morland, Ronald A. Nelson,
Manuscript received February 10, 2009; provisional acceptance April 16, 2009; revised manuscript and an anonymous reviewer.
received May 31, 2009; final acceptance July 13, 2009.
DOI:10.1306/07130909016

AAPG Bulletin, v. 93, no. 11 (November 2009), pp. 1597–1608 1597


Many attempts have been made in the past to achieve and other geological factors are commonly known to
this, but most of them have proven to be inaccurate and control the location and intensity of fracturing. These
unreliable. For example, many companies are still using factors, known as fracture drivers, can be identified
simple methods such as reservoir curvature to identify not just from wellbore data but also from seismic data,
areas that are likely to be fractured. Such a simplified ap- which is the key to predicting fracture occurrence
proach ignores lithology, bed thickness, and other factors throughout the reservoir. Through the CFM approach,
that are generally known to influence fracture develop- these fracture drivers are related to fracture indicators,
ment. Similarly, geostatistical techniques such as kriging which include the interpretation of fractures from core
and sequential Gaussian simulation (SGS) are commonly descriptions, image logs, and production logs that dem-
inadequate for distributing fracture properties in inter- onstrate the existence of a fracture at a specific location.
well areas because these properties do not change in a Once this relationship is established, the fracture driv-
regular manner away from well control. ers can be used to predict the location of fractures else-
Geomechanical models attempt to construct a frac- where in the reservoir.
ture description and mechanical property distribution Common geological fracture drivers include facies
from the tectonic history, but these models commonly types, porosity, and reservoir zonation. Geomechanical
fail to capture the complex heterogeneity and anisotro- fracture drivers include deformation, slopes and curva-
py of the fracture system (Mace et al., 2004). Discrete tures of structural surfaces in multiple directions, and
fracture modeling (Dershowitz et al., 1998; Sabathier proximity to faults. Seismic fracture drivers include elas-
et al., 1998), which has become very popular in recent tic properties, acoustic impedance, and spectral imaging
years, is constrained by the assumption that a limited attributes. In combination, these fracture drivers are a
number of macroscale fractures, as observed in cores much more powerful tool for characterizing the fracture
and image logs, control fluid flow. A review of discrete system than if used alone.
fracture modeling methods and their characteristics can The CFM approach can be applied in both two and
be found in the works of Dershowitz et al. (2004) and three dimensions (2-D and 3-D). When working in 2-D,
Bourbiaux et al. (2005). the fracture indicator is represented by a single value at
To capture fracture effects at multiple scales and each well, and the fracture drivers are a series of 2-D con-
simultaneously integrate all of the available information, tour maps. Ouenes et al. (1998) and Zellou and Ouenes
including core, log, seismic, and well test data, an alter- (2003) presented a 2-D CFM example where Dakota
native technique (Ouenes et al., 1995; Ouenes, 2000) Formation sweet spots were identified in the San Juan
called continuous fracture modeling (CFM) can be used. Basin. When working in 3-D, fracture indicators are as-
This article provides a review of the CFM approach signed to model cells intersected by wellbores, and the
by describing its key characteristics and illustrating these fracture drivers consist of distributed properties condi-
with three case studies. In each of these studies, objec- tioned to log, core, and seismic data. Table 1 lists those
tive criteria are used to evaluate the predictive capabil- fracture drivers most commonly used in building a geo-
ity of the generated model, which is critical to ensure its cellular model for CFM and some of the methods used
reliability. This can be done by comparing the model to obtain the seismic fracture drivers. The table also con-
properties to the actual properties found in new wells tains a list of common fracture indicators available at
or in existing wells that were not used to construct the the wellbore.
model (also known as blind wells). Another method of A neural network is used to find possible relation-
comparison is to test the validity of the model in a numer- ships between the fracture drivers and fracture indica-
ical simulator to see if well performance can be reason- tors observed at the wellbore. The neural network
ably matched without significant model adjustments. approach first ranks all the fracture drivers according
to how reliably they correlate with the fracture indica-
tors. The modeler then reviews the results in light of
what is known about the significance and robustness
CONTINUOUS FRACTURE MODELING APPROACH of each fracture driver, and how the ranking compares
to what is understood about the physical distribution of
The CFM approach does not focus on the fractures fractures in the reservoir. The best-correlated fracture
themselves but instead on the factors that control where drivers that make the most sense are then subjected to
fracturing occurs. Lithology, structure, proximity to faults, a training and validation process before being used to

1598 Continuous Fracture Modeling (CFM)


Table 1. Summary of the Most Common 3-D Fracture Drivers Used in the CFM Approach along with Possible Fracture Indicators
Available at the Wellbore

Structural and Geomechanical Seismic Attributes Used as Fracture Indicators in


Geological Fracture Drivers Fracture Drivers Fracture Drivers the Wellbore
Core- and log-derived Structural curvature (dip Amplitude and amplitude- Fracture count from
porosity and permeability and slope) based attributes image logs
Lithofacies volumes (shale, Structure and reservoir depth Impedance derived from Fracture count from core
dolomite, calcite) poststack inversions descriptions
Log data (gamma ray, Deformation Frequency-dependent spectral Fluid entry or exit locations
density, resistivity, etc.) imaging attributes from production logs
Fluid saturations Stress and strain fields Statistical spectral imaging attributes Well-test permeabilities
Correlation framework Distance to faults Elastic properties derived from Difference between deep and
prestack inversions shallow resistivity-log values
Vertical and lateral trends Bed or layer thickness Azimuthal anisotropy attributes Drilling losses

generate a suite of equiprobable fracture-density real- through high-resolution seismic stochastic inversion
izations. These realizations can then be screened and (Haas and Dubrule, 1994). In this process, synthetic seis-
calibrated to permeability-thickness (kh) data or direc- mograms are generated from the pseudoimpedance log
tional permeability data obtained from well tests. The and compared to the actual seismic trace at a given well lo-
resulting effective permeability model can then be ex- cation. The synthetic seismogram that results in the best
ported for use in numerical simulation and development match to the actual seismic trace is retained as the inver-
planning. A simplified workflow summarizing the entire sion solution at that location. The vertical resolution of
3-D CFM approach is shown in Figure 1. the simulated data is determined by the selection of the
vertical cell size (determined by the end user of the
model) and not by the frequency content of the seismic
data. The result of the stochastic seismic inversion is a
VALUE OF SEISMIC DATA IN CONTINUOUS 3-D volume with a seismic-like areal resolution and a
FRACTURE MODELING loglike vertical resolution that honors both the log data
and the seismic data.
The CFM approach relies heavily on the use of seismic An example showing the soundness of this tech-
data to provide key fracture drivers. Structural fracture nique comes from a recent work in a Hungarian reser-
drivers such as dip magnitude and interpreted faults can voir (Zellou et al., 2006) where the data from five wells
be derived from basic seismic interpretation, and seis- were used to generate a high-resolution seismic imped-
mic attributes can be derived from seismic volumetric ance volume with a sample rate of 0.5 ms (2–3 m [6–
curvatures (Al-Dosari and Marfurt, 2006). Bed thickness 10 ft]). The resulting impedance volume was then com-
variations can be determined from isochrons or from pared to the actual impedance values at four blind wells
the tuning frequency derived from spectral imaging. Li- resulting in a correlation coefficient of 0.77 (Figure 2).
thology, porosity, and other rock properties can be de- This goodness of fit illustrates the potential ability of
rived indirectly from high-resolution seismic attributes the high-resolution inversion to accurately predict imped-
obtained in prestack and poststack inversions calibrated ance values in the 3-D seismic volume. For the fracture
against log and core data from wells. Under good condi- modeling effort in this example, a high-resolution imped-
tions, these seismic attributes can be generated at a ver- ance was derived using all nine wells.
tical resolution of 2–3 m (6–10 ft), which, in our experi- The predictive capabilities of CFM depend, in large
ence, is comparable to the thickness of intervals that part, on the quality of the seismic and well-log data used
control fluid flow in fractured reservoirs and to the ver- to extract the relevant seismic attributes. A good-quality
tical scale at which geocellular models are built. 3-D seismic survey and a large number of available wells
The key to deriving attributes at a 2–3-m (6–10-ft) with good-quality sonic and density logs recorded over a
scale is to integrate the seismic traces with the log data long interval provide the best input data.

Jenkins et al. 1599


Figure 1. Simplified workflow showing the key components of the 3-D CFM approach, which includes the generation of seismic attri-
butes, the generation of geologic and geomechanical fracture drivers constrained by seismic data, the construction of a fracture indicator
log at the wells, and the quantitative integration of all these data with artificial intelligence tools (neural networks).

The quality of the seismic data also has a significant olution of 2–3 m (6–10 ft) for more than a decade (Lo
impact on the quality of the structural framework and and Bashore, 1999; Torres-Verdin et al., 1999; Dasgupta
the resulting 3-D geocellular grid, which are built in the et al., 2000; Robinson, 2001; Sullivan et al., 2004; Francis,
time domain using interpreted horizons and faults. For 2005; Raguwanti et al., 2005). Commercial software
example, poor seismic data may result in the inability using this approach has also been created. Despite these
to capture critical faults, which will negatively impact advances, many geoscientists are still unfamiliar with the
the acoustic impedance inversion. Figure 3 shows a technique and erroneously believe that seismic attributes
cross section of the impedance for the Hungarian res- are limited to a one-quarter wavelength resolution. In the
ervoir (Zellou et al., 2006) described previously. With- CFM approach, many seismic attributes are computed
out a proper structural framework that considers all near the seismic detection limit (2–3 m [6–10 ft]), which
the faults and an inversion algorithm that is able to could be five to seven times smaller than the seismic res-
honor the sharp offsets near each of the faults, the de- olution (15–20 m [49–66 ft]).
rived impedance will not be very useful in the CFM The use of high-resolution attributes is critical to
workflow. achieve good predictive capabilities with the CFM ap-
High-resolution seismic attributes have been pro- proach. Boadu (1998) showed that, in theory, a neural
duced for almost 15 yr using Haas and Dubrule’s sto- network can accurately predict the fracture density
chastic or geostatistical inversion techniques (Haas and using limited seismic attributes. The following case stud-
Dubrule, 1994; Dubrule et al., 1998). Many geoscien- ies each include the use of one or more high-resolution
tists have been using seismic attributes simulated at a res- seismic attributes in the CFM approach.

1600 Continuous Fracture Modeling (CFM)


Figure 2. Predicted
(x axis) versus actual (y axis)
impedance at four blind
wells not used in a seismic
inversion. The predicted
impedance at the wells
was extracted from an im-
pedance cube computed
with a resolution of 0.5 ms
or 2–3 m (6–10 ft) and
compared to the actual
logs averaged at 2 m (from
Zellou et al., 2006; re-
printed by permission of
the Society of Petroleum
Engineers).

FIELD CASE STUDIES OF CONTINUOUS ture density and to create reservoir simulation models
FRACTURE MODELING in the South Arne field, a complex chalk reservoir. Nu-
merous fracture drivers, including a high-resolution
South Arne Chalk Field, Danish North Sea acoustic impedance inversion and several spectral imag-
ing attributes, were combined with porosity log values
In this project (Christensen et al., 2006), the CFM ap- at 15 wells to construct a porosity model. The result-
proach was used to predict the 3-D distribution of frac- ing model compared very favorably with three blind

Figure 3. Cross section


along an inline showing
the derived impedance at
a resolution of 0.5 ms. Be-
cause the structural in-
terpretation captures the
existing faults and their
sharp offsets, the derived
seismic attributes are more
representative and there-
fore will be more useful
in the CFM workflow (from
Zellou et al., 2006; re-
printed by permission of
the Society of Petroleum
Engineers).

Jenkins et al. 1601


Figure 4. Validation of
predicted porosities in the
South Arne field of the
North Sea. The red curves
are the predicted porosity
values from the seismic
attributes. The black curves
are the upscaled porosity
values from the logs
(Christensen et al., 2006;
reprinted by permission
of the Society of Petroleum
Engineers). MD = mea-
sured depth.

wells that were not included in the model construction ity) are individual property models that were each gen-
(Figure 4). Four vertical appraisal wells drilled subse- erated within a single geocellular modeling project. The
quent to model construction were used to further validate effective permeability and porosity models were subse-
the inverted impedance and derived porosity models. quently exported for use in numerical simulation.
After validation, a fracture-density model was gener- As shown for two wells (Figure 5), the resulting
ated using porosity, seismic data, and geomechanical simulation runs matched the individual well perfor-
fracture drivers. Two fracture indicators were used to mance without making any substantive adjustments.
calibrate the fracture model and quantify the effects of The matches in these two wells, which had moderately
fractures: core permeability and the fracture density esti- simple completions, indicate that the derived fracture-
mated from image logs. A matrix permeability model density model was able to quantify the complex frac-
was also generated using the porosity model as the domi- ture network at the appropriate resolution. Other wells
nant fracture driver. This matrix permeability model in the field, with much more complex completions, were
was subsequently combined with the fracture-density more difficult to match. Nonetheless, the CFM-derived
model to generate an effective permeability model for effective permeability model used in the simulation work
numerical simulation. The resulting effective perme- was far better than the model previously used and re-
ability from fracture permeability enhancement for a quired smaller adjustments to obtain matches.
matrix volume with N fractures in the same direction The history match was further optimized by varying
was described by the following equation. the initially estimated average fracture aperture to find
the best overall match to effective permeability. The his-
tory match approach resulted in a geologically meaning-
Keff ¼ Km þ f  a3h  84:44  106 ð1Þ
ful result using apertures ranging from 30 to 40 mm,
which provides a good starting point for future history
where Keff is the effective permeability of the combined matching once more well tests become available.
matrix and fracture flow, (in millidarcies), Km is the ma- The fracture-density model was validated on a well-
trix permeability (in millidarcies), f is the fracture density by-well basis by comparing fracture density in the model
(in number of fractures per meter), and ah is the average with the actual counts of conductive fractures. The re-
fracture aperture open for flow (in microns). sults are shown in Figure 6 for three random horizontal
The four models discussed above (porosity, matrix wells that were available during the study. The compar-
permeability, fracture density, and effective permeabil- ison shows fair to good agreement between the predicted

1602 Continuous Fracture Modeling (CFM)


Figure 5. Initial history matching results for two wells showing good agreement between predicted and actual data for water produc-
tion (blue), pressure (orange), and gas-oil ratio (red). The oil rate curve (green) is specified in the model (Christensen et al., 2006;
reprinted by permission of the Society of Petroleum Engineers).
Jenkins et al. 1603
Figure 6. Count of actual con-
ductive fractures (black) from
image logs in horizontal bore-
holes compared to the predicted
fracture density (red) from CFM
modeling (Christensen et al.,
2006; reprinted by permission
of the Society of Petroleum
Engineers).

and actual occurrence of fractures. Subsequently, two cator. The CFM approach was used to quantify the re-
horizontal wells were drilled in 2007 after the comple- lationship between the fracture drivers and the fracture
tion of the study. Fracture-density logs extracted from indicator, and then to predict fracture density in every
the model showed a favorable comparison to the actual cell of the model.
fracture-density logs from these wells. Fifty equiprobable fracture model realizations were
generated, and three of these were selected as the base
Maloichskoe Carbonate Reservoir, case, downside case, and upside case for further analy-
Western Siberian Basin sis. A connectivity analysis based on displaying the CFM
results as discrete fractures and pipes was then con-
Unlike the South Arne field, where image logs and core ducted to identify those areas most likely to contain
permeability were used as the key fracture indicators, connected fractures. A well was subsequently drilled
this project (Pinous et al., 2006) used a simple differ- using this information, and based on a comparison of
ence in the values of the shallow and deep resistivity the image log from this well to the modeling results, a
logs as the key fracture indicator. good correspondence between the predicted and actual
Fracture drivers included seismic inversion and spec- location of fractured intervals existed (Figure 7).
tral imaging volumes, curvature in multiple directions, The predicted fracture model showed that the up-
distance to faults, porosity, and deformation. Overall, per interval would be highly fractured, the middle inter-
27 fracture drivers were evaluated and ranked based on val would be poorly fractured, and the lower interval
how accurately each was correlated to the fracture indi- would be moderately fractured. This is very similar to

1604 Continuous Fracture Modeling (CFM)


Figure 7. Comparison be-
tween the model-derived frac-
ture density, derived from the
difference between the shallow
and deep resistivity logs, and
the measured Formation Micro-
Scanner (FMS) log in the Ma-
loichskoe field. Note that, at a
depth of 2850 m (9350 ft), the
model predicted a 2-m (6-ft)-
thick interval with no fractures
that was confirmed by the FMS
log (Pinous et al., 2006; reprinted
by permission of the Society of
Petroleum Engineers). MD =
measured depth.

the interpretation subsequently derived from the image multiple fracture-density models. An average model
log in this well. Although variations between the pre- was then chosen, and data from three well tests were used
dicted and actual fracture density are present, the com- to convert the model into fracture permeability. This
parison is quite good given that the key fracture indica- was then used as an input for numerical modeling. Dur-
tor was the difference between a shallow and a deep ing the history matching process, a match was achieved
resistivity log. The fracture-density model was built using by making adjustments to relative permeability. The
a geocellular grid containing cells 2 m (6 ft) thick, and fracture permeability distribution itself was not altered.
thus it was possible to generate a predicted fracture- After history matching was completed, various new dril-
density log having a 2-m (6-ft) resolution, as shown in ling locations were planned and tested with the model.
Figure 7. Based on the results of this work, the Sabria 11 well
was drilled and completed in 2007. A comparison be-
Sabria Quartzite Field, Tunisia tween the actual and predicted porosity for this well
(Figure 8) shows good agreement, as does the actual
This project began with a high-resolution inversion and and predicted impedance for this well. In addition, the
the generation of spectral imaging cubes. A structural actual oil rate of Sabria 11 is in line with the predicted
framework was then built in both time and depth, and range of rates forecasted by the numerical model.
two key seismic attributes, impedance and tuning fre-
quency, were resampled to the geocellular grid. Geosta-
tistical techniques were then used to distribute porosity, IMPACT OF NEW SEISMIC TECHNOLOGIES ON
permeability, water saturation, and shale content into CONTINUOUS FRACTURE MODELING
this grid. These four geological parameters, the two seis-
mic attributes, and structural information were used as Since its inception in 1995 (Ouenes et al., 1995), the
fracture drivers. main focus of the CFM approach has been to incorpo-
The fracture indicator log was provided by core de- rate multiple seismic attributes to improve the 3-D de-
scriptions from three wells. The CFM approach was used scription of fractures. The technique has been applied to
to rank the fracture drivers according to how reliably multiple fields, and its value has been demonstrated by
they matched this fracture indicator and to generate the ability of the resulting models to predict the fracture

Jenkins et al. 1605


The initial application of CFM technology used only
the seismic amplitude (Zellou et al., 1995), but this was
later improved by adding coherency (Gauthier et al.,
2000). Improvements were also made by adding imped-
ance (Laribi et al., 2003; Zellou et al., 2003) and prestack
azimuthal attributes (Boerner et al., 2003; Wong, 2003)
to the list of fracture drivers. In many instances, the addi-
tion of spectral imaging attributes (Ouenes et al., 2004;
Christensen et al., 2006; Pinous et al., 2006; Zellou et al.,
2006) to the previous seismic attributes greatly improved
the predictability of the models.
Recently, the use of volumetric curvature (Al-Dosari
and Marfurt, 2006), which is a significant improve-
ment over curvature methods applied to surfaces and
coherency-type seismic attributes, provided some of
the best seismic attributes for fractured reservoirs where
faults are major factors. The use of elastic properties de-
rived from high-resolution prestack inversion is one of
the best seismic inputs for the CFM approach. Unfortu-
nately, deriving such seismic attributes requires the
shear component from dipole sonic logs, which is rarely
available in fractured reservoirs.
The absence of seismic attributes or the use of poor-
quality attributes reduces the predictive capabilities of
the CFM models. Although a CFM model can be built
without seismic data, it will likely do a poor job of pre-
dicting fracture density at undrilled locations. This may
also be the case if the seismic data are of poor quality,
especially if this leads to an inaccurate structural inter-
pretation that fails to include some of the faults. If a large
number of wells are present, this will help compensate
for poorer quality seismic data and the resulting model
will have better predictive capabilities.
The presence of good seismic data alone is not suf-
ficient to ensure a predictive CFM model, and other im-
Figure 8. Comparison between actual and predicted porosity portant conditions must be satisfied. One of these is the
values in a new well in the Sabria Quartzite field showing a higher need to extract high-resolution attributes capable of
porosity interval from 3814 to 3930 m (12,513 to 12,894 ft). The capturing the vertical variations at a scale of a few me-
well (left) was only logged to 3930 m (12,894 ft), but the model ters. A seismic attribute at an initial seismic resolution of
(right) extracted values along the entire borehole and thus shows 4 ms (which represents 15 to 25 m [49 to 82 ft] of thick-
predicted values of lower porosity deeper than 3930 m (12,894 ft) ness) will not be very useful for the CFM approach. An-
(Ouenes et al., 2008; reprinted by permission of the Society of other important condition is that enough good-quality
Petroleum Engineers). data exist to generate an accurate fracture indicator
log. Ideally, this would include cores, logs, well tests,
and production logs with each demonstrating the exis-
density in new wells and closely match the performance tence of productive fractures at the same location in a
of individual wells. Because of the flexibility of the CFM given wellbore. With less data and/or conflicting infor-
approach, it will continue to benefit from new seismic mation, confidence in the calibration of fracture indica-
technologies that provide better descriptions of subsur- tors to fracture drivers is reduced as is the predictive
face properties. capability of the CFM approach.

1606 Continuous Fracture Modeling (CFM)


SUMMARY Dershowitz, W., P. La Pointe, and T. Doe, 2004, Advances in dis-
crete fracture network modeling: Proceedings of the U.S. Envi-
ronmental Protection Agency/National Groundwater Associa-
A successful modeling approach for naturally fractured tion Fractured Rock Conference, p. 882–894.
reservoirs needs to predict where naturally fractured Dubrule, O., M. Thibaut, P. Lamy, and A. Haas, 1998, Geostatistical
reservoir characterization constrained by 3D seismic data: Pe-
intervals will be encountered in undrilled wells and troleum Geoscience, v. 4, no. 2, p. 121–128.
do so at a scale (2–3 m [6–10 ft]) that is comparable Francis, A., 2005, Limitations of deterministic and advantages of sto-
to core, log, and well-test data. A successful approach chastic seismic inversion: Canadian Society of Exploration Geo-
physicists Recorder, v. 30, no. 2, p. 5–11.
must also be able to provide an effective permeability
Gauthier, B. D. M., A. M. Zellou, A. Toublanc, M. Garcia, and J. M.
model for numerical simulation that will match the per- Daniel, 2000, Integrated fractured reservoir characterization: A
formance of individual wells. case study in a North Africa field: Society of Petroleum Engi-
The CFM approach meets these criteria based on neers Paper No. 65118, 11 p.
Haas, A., and O. Dubrule, 1994, Geostatistical inversion: A sequen-
the results of multiple case studies. The approach pro- tial method of stochastic reservoir modeling constrained by seis-
vides a method for geoscientists to quantitatively inte- mic data: First Break, v. 12, no. 11, p. 561–569.
grate all of the available data by simultaneously relating Laribi, M., et al., 2003, Integrated fractured reservoir characteriza-
tion and simulation: Application to Sidi El Kilani field, Tunisia.:
multiple fracture drivers to any type of fracture indica- Society of Petroleum Engineers Paper No. 84455, 13 p.
tor. This integration is achieved by the quantitative and Lo, T.-W., and W. Bashore, 1999, Seismic constrained facies model-
simultaneous use of multiple 3-D property volumes, ing using stochastic seismic inversion and indicator simulation:
A North Sea example (abs.): Society of Exploration Geophysi-
especially seismic attributes, to predict the interwell
cists Expanded Abstracts, v. 18, p. 923–926.
occurrence of fractures and build effective permeabil- Mace, L., L. Souche, and J.-L. Mallet, 2004, 3D fracture characteriza-
ity models that incorporate them. tion based on geomechanics and geologic data uncertainties: 9th
As new seismic processing and interpretation tech- European Conference on the Mathematics of Oil Recovery:
Houten, Netherlands, European Association of Geoscientists
nologies continue to evolve, the resulting seismic attri- and Engineers, 8 p.
butes will be used to further improve the ability of the Ouenes, A., 2000, Practical application of fuzzy logic and neural net-
CFM approach to quantify the location and effects of works to fractured reservoir characterization: Computers and
Geosciences, v. 26, no. 9, p. 953–962.
fractures on well performance. Ouenes, A., S. Richardson, and W. Weiss, 1995, Fractured reservoir
characterization and performance forecasting using geomechan-
ics and artificial intelligence: Society of Petroleum Engineers
Paper No. 30572, 12 p.
REFERENCES CITED Ouenes, A., A. M. Zellou, P. M. Basinski, and C. F. Head, 1998,
Practical use of neural networks in tight gas fractured reservoirs:
Al-Dosari, S., and K. Marfurt, 2006, 3D volumetric multispectral es- Application to the San Juan Basin: Society of Petroleum Engi-
timates of reflector curvature and rotation: Geophysics, v. 71, neers Paper No. 39965, 8 p.
no. 5, p. 41–51, doi:10.1190/1.2242449. Ouenes, A., G. Robinson, and A. Zellou, 2004, Impact of pre-stack
Boadu, F. K., 1998, Inversion of fracture density from field seismic and post-stack seismic on integrated naturally fractured reser-
velocities using artificial neural networks: Geophysics, v. 63, voir characterization: Society of Petroleum Engineers Paper
no. 2, p. 534–545, doi:10.1190/1.1444354. No. 87007, 11 p.
Boerner, S., D. Gray, D. Todorovic-Marinic, A. Zellou, and G. Ouenes, A., G. Robinson, D. Balogh, A. Zellou, D. Umbsaar, H.
Schnerck, 2003, Employing neural networks to integrate seis- Jarraya, T. Boufares, L. Ayadi, and R. Kacem, 2008, Seismically
mic and other data for the prediction of fracture intensity: So- driven characterization, simulation and underbalanced drilling
ciety of Petroleum Engineers Paper No. 84453, 13 p. of multiple horizontal boreholes in a tight fractured quartzite
Bourbiaux, B., R. Basquet, J. M. Daniel, L. Y. Hu, S. Jenni, A. Lange, reservoir: Application to Sabria field, Tunisia: Society of Petro-
and P. Rasolofosaon, 2005, Fractured reservoirs modeling: A re- leum Engineers Paper No. 112853, 12 p.
view of the challenges and some recent solutions: First Break, Pinous, O., E. P. Sokolov, S. Y. Bahir, A. Zellou, G. Robinson, T. Royer,
v. 23, no. 9, p. 33–40. N. Svikhnushin, D. Borisenok, and A. Blank, 2006, Applica-
Christensen, S. A., T. E. Dalgaard, A. Rosendal, J. W. Christensen, tion of an integrated approach for the characterization of a natu-
G. Robinson, A. Zellou, and T. Royer, 2006, Seismically driven rally fractured reservoir in the West Siberian basement (ex-
reservoir characterization using an innovative integrated ap- ample of Maloichskoe field): Society of Petroleum Engineers
proach: Syd Arne field: Society of Petroleum Engineers Paper Paper No. 102562, 7 p.
No. 103282, 11 p. Raguwanti, R., A. Sukotjo, B. Wikan, and H. Adibrata, 2005, Inno-
Dasgupta, S., M. Hong, P. La Croix, L. Al-Mana, and G. Robinson, vative approach using geostatistical inversion for carbonate res-
2000, Prediction of reservoir properties by integration of seismic ervoir characterization in Sopa field, south Sumatra, Indonesia
stochastic inversion and coherency attributes in super giant (abs.): 2005 AAPG International Conference and Exhibition,
Ghawar field (abs.): Society of Exploration Geophysicists Ex- Paris, France, http://www.searchanddiscovery.net/documents
panded Abstracts, v. 19, p. 1501–05, doi:10.1190/1.1815691. /abstracts/2005intl_paris/raguwanti.htm.
Dershowitz, B., P. Lapointe, T. Eiben, and L. Wei, 1998, Integration of Robinson, G. C., 2001, Stochastic seismic inversion applied to reser-
discrete feature network methods with conventional simulator voir characterization: Canadian Society of Exploration Geo-
approaches: Society of Petroleum Engineers Paper No. 49069, physicists Recorder, v. 26, no. 1, p. 36–40.
9 p. Sabathier, J. C., B. J. Bourbiaux, M. C. Cacas, and S. Sarda, 1998, A

Jenkins et al. 1607


new approach of fractured reservoirs: Society of Petroleum En- acterization using neural networks and fuzzy logic: Three case
gineers Paper No. 39825, 11 p. studies, in M. Nikravesh, L. A. Zadeh, and F. Aminzadeh, eds.,
Sullivan, C., E. Ekstrand, T. Byrd, D. Bruce, M. Bogaards, R. Boese, Soft computing and intelligent data analysis in oil exploration:
and P. Mesdag, 2004, Quantifying uncertainty in reservoir Developments in Petroleum Geoscience 51, p. 583–602.
properties using geostatistical inversion at the Holstein field, Zellou, A., L. Hartley, E. Hoogerduijn-Strating, S. Al Dhahab, W.
deepwater Gulf of Mexico (abs.): Society of Exploration Geo- Boom, and F. Hadrami, 2003, Integrated workflow applied to
physicists Expanded Abstracts, v. 23, p. 1491–1494, doi:10 the characterization of a carbonate fractured reservoir: Qarn Alam
.1190/1.1851128. field: Society of Petroleum Engineers Paper No. 81579, 8 p.
Torres-Verdín, C., M. Victoria, G. Merletti, and J. Pendrel, 1999, Trace- Zellou, A., G. Robinson, T. Royer, P. Zahuczki, and A. Kiraly, 2006,
based and geostatistical inversion of 3-D seismic data for thin-sand Fractured reservoir characterization using post-stack seismic at-
delineation, an application in San Jorge Basin, Argentina: The tributes: Application to a Hungarian reservoir (abs.): European
Leading Edge, v. 18, no. 9, p. 1070–1076, doi:10.1190/1 Association of Geoscientists and Engineers Annual Conference,
.1438434. Vienna, Austria, 4 p.
Wong, P. M., 2003, A novel technique for modeling fracture intensity: Zellou, A. M., A. Ouenes, and A. K. Banik, 1995, Improved fractured
A case study from the Pinedale anticline in Wyoming: AAPG Bul- reservoir characterization using neural networks, geomechan-
letin, v. 87, no. 11, p. 1717–1727, doi:10.1306/07020303005. ics, and 3-D seismic: Society of Petroleum Engineers Paper
Zellou, A., and A. Ouenes, 2003, Integrated fractured reservoir char- No. 30722, 11 p.

1608 Continuous Fracture Modeling (CFM)

You might also like