Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

SPE-171608-MS

The Application of Discrete Fracture Network Modelling to the Optimization


of Wells in Resource Plays
Simon Emsley, Shihong Chi, Jhon Rivas, and Jim Hallin, ION Geophysical

Copyright 2014, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE/CSUR Unconventional Resources Conference - Canada held in Calgary, Alberta, Canada, 30 September –
2 October 2014.

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 have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper 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 SPE copyright.

Abstract
The drilling and completion of wells in resource plays has tended to be based on drilling a pre-determined
number of wells with uniformly spaced laterals drilled on pre-defined azimuths. The development of
resource plays is now incorporating geophysics. Workflows have been developed that make 3D seismic
data relevant to unconventional resource optimisation and to improve the economics of exploration and
development. This can be achieved by constructing models, incorporating data sets derived from the
seismic volumes and wireline data and then forward modelling and subsequently optimising well
placement and hydraulic fracture design.
The workflow described here starts with a 3D surface-seismic multi-component data set and goes
through to hydraulic fracture simulation and potential well optimization achieved through the construction
of a discrete fracture network (DFN) model. The data sets are derived from azimuthal velocity analysis
and anisotropic parameter calculations; seismic attribute analysis, interpretation and inversion provide a
rich data source to develop a model, driven by deterministic data.
Three theoretical wells were included in the model, the location of one indicated that it was drilled into
the intersection of two faults which could have been avoided. The other two were forward modelled,
simulating hydraulic fracturing; this showed that in areas of high fracture density there was a larger
number of stimulated fractures and predicted micro-seismicity, suggesting the reactivation of pre-existing
fault and fractures. In areas with lower fracture densities, additional fractures were generated suggesting
that hydraulic fracturing would result in the generation of new tensile fractures.
The modelling provided an enhanced understanding of the reservoir, providing greater insight than
would be obtained from co-rendering. Importantly, the model tested different well positions, stage
spacings and drilling azimuths. In this case study the locations and azimuths of the laterals were changed
with differences in the stage-to-stage stimulated volume of between 3 and 30%. This was particularly the
case for the central stages of the modelled wells and this has clear financial implications in the
development of the resource in terms of drilling costs, completions and production.
2 SPE-171608-MS

Figure 1—Hydrocarbon prone shales from the Catskill Mountains, New York (photograph courtesy of Prof. R Swarbrick).

Introduction
The exploitation of resource plays has been seen to be based on drilling a pre-defined number of wells in
a given area and as such the plays are often termed statistical plays. In such a play type the operator
generally drills a large number of wells and these often have similar designs in terms of drilling direction
and hydraulic fracture completion strategy. Although drilling of vertical pilot holes and horizontal laterals
can be performed throughout the leased acreage, the results are often very variable not just from well to
well but along the length of the wells. The production from the wells is seen to decline rapidly over a
relatively short period of time, e.g. a few years and recovery efficiencies are generally low with values
of a few percent being reported. This engineering driven approach seems to neglect the effects and
influence of geology and structure and the consequences these have on performance of the wells and
return on investment. However, generally the wells are drilled orthogonal to the main structural grain,
fault and fracture direction and maximum horizontal stress direction. This neglects the local variations in
stress direction and the changes in dominant fracture directions. The fracture directions can rotate based
on numerous variations in the subsurface structure, including proximity to faults and folds and stress
direction. Fracture systems are often mechanically bound and are concentrated in specific layers not
crossing mechanically harder beds/layers whilst other fracture sets are through going. Figure 1 shows an
example of mechanical layering where the majority of the fractures are seen to be confined to the layers
where thin beds and layering are seen. The photograph in Figure 1 shows a number of different fracture
sets and layering at different scales that would be seen as anisotropy. The figure also illustrates the
challenges in drilling horizontal wells and keeping them in the zone. It also illustrates, albeit at a small
scale, one of the issues in hydraulic fracture completions that of keeping the fractures generated within
the reservoir, rather than breaking through the mechanically stronger rocks above and below the reservoir
unit. Additionally, it can be envisaged that if a lateral porpoises out of zone crossing through the
mechanically stronger layering any completions in such a zone would be unlikely to be effective.
The effectiveness of completions leads to a consideration of economics and success rates. It is
estimated that approximately half of the wells drilled in the United States are uneconomic and that in the
SPE-171608-MS 3

Eagle Ford Shale approximately 36% of perforations are unproductive (PFC, 2014). Some of this lack of
effectiveness is considered to result from wells that were drilled in the incorrect place or were landed in
the wrong zone (PFC, 2014). The approach to completion design, which is often geometric, has a
significant role to play in the underperformance. Hydraulic fracture stages are often equally spaced and
of similar lengths from well to well. This results in variable reservoir quality within the fracture stage and
even different rock types in the same stage as can be envisaged by looking at Figure 1. The statistical
approach to drilling and the geometric approach to completion programmes and designs must therefore be
inappropriate. The importance of local stress direction, natural fracture systems and the lithological
variability are important factors and affect the efficiency of stimulation programmes. Elastic and
geomechanical properties of the rocks are also of considerable importance and influence the completion
performance.
Including an understanding of natural fractures in the design of the stimulation process is important;
not only the fracture direction but also intensity. It has been shown recently in a Marcellus Shale play in
West Virginia that the dominant fracture direction change with depth and thus what might be the
appropriate drilling azimuth for one zone may not be appropriate for another interval. It is well known that
the presence of natural fractures exert an influence on the hydraulic fracture stimulation process (Gale et.
al., 2006). It is considered that the one of the key elements to the successful hydraulic completion is the
effective use of the natural fracture system by the hydraulic stimulation process to enable the well to
connect to a significantly larger drainage volume (Rogers et al, 2010). The interaction between the
hydraulic fracture stimulation process and the natural fracture system is not entirely clear but evidence will
be presented that indicates that where fractures are at a high angle to the well bore, these tend to open
either by dilation or a shearing mechanism, both resulting in an increase in volume. Where fractures tend
to be more oblique to the well path the fractures tend to close as a result of the increasing pressures and
resultant stress modification. It has also been noted that fractures may open to one side of the well and
close on the other side of the well. The parameters that are less easily understood include the ways fluid
and proppant travel through fractures that are open but are not stimulated i.e. there is no volumetric change
in the fracture or increase in aperture. One of the challenges in including fracture information in
completion design is the paucity of data or the paucity of reliable data. Certainly in the early phases in
the development of a resource play, information at the well scale from wireline logs is spare. Often wells
drilled go unlogged but without information on the natural fracture systems it is likely that there will be
continued inefficient and uneconomic completions and a poor return on investment.
This paper discusses a case study that addresses the issues of identifying fractures and fracture systems
through the use of multi-component 3D seismic data and the analysis of these data to derive information
that can be used for modelling. The important aspect to note here is that the information used is
deterministic in nature as opposed to a stochastic approach. These deterministic data are used as inputs
to construct a discrete fracture network (DFN) model. These models are then used to forward model
hydraulic fracture stimulations, with the results being used to analyse inter-well connectivity and
inter-stage connectivity. The modelling can therefore be used to optimize well placement in terms of
lateral separation, drilling azimuth and completion stage spacing.
Allegheny Case Study
The case study described in this paper is based on an 85 square mile multi-component 3D seismic survey
acquired in West Virginia over the Marcellus Shale, the location of the AlleghenySCAN Survey is shown
in Figure 2. The workflow described here starts with the 3D surface-seismic multi-component data sets
(PP and PS data) and goes through to hydraulic fracture simulation and potential well optimization
achieved through the construction of a discrete fracture network (DFN) model.
There are challenges in conducting fracture modelling and in particular the construction of DFN
models. In general there is a paucity of data and a lack of reliable data. The controls of production from
4 SPE-171608-MS

Figure 2—Location of the survey area, the Allegheny Survey (green area towards the bottom left of the figure) is the basis of the case study.

unconventional reservoirs or any fracture dominated reservoir, as opposed to a conventional porous media
reservoir, are the fractures, the fracture network and the connectivity of that system. Hence the hydraulic
fracture stimulation process adopted in the unconventional reservoir development to create, and prop open
fracture networks. Fractures are small and often below the resolution of many logging tools and, as in this
example, logs that may image the fractures are often not acquired or are unavailable. Although fractures
are small, the permeability associated with them can be very high, up to thousands of milliDarcy. The
other issue in fracture modelling is spatial range or sampling; the fractures have dimensions that are much
smaller than any seismic survey resolution. However, in the absence of any other information, seismic
data must be relied upon to provide the required information to generate the DFN models.
Seismic data can be co-rendered, for example curvature data and interval Vp velocity or density
information and this can be used to help define sweetspots. These images generate more readily
interpretable images of the subsurface and an enhanced understanding the reservoir and have been
successfully used to target wells which can be demonstrated to have better cumulative production. While
such methods of visualisation and interpretation yield valuable results they do not consider connectivity
of the fracture systems and compartmentalisation which can be addressed through the construction of a
DFN model.
The integrated and interconnected workflow is summarised in Figure 3. The primary source of data for
the construction of the DFN is the 3D seismic data. During processing, the azimuthal velocity analysis of
the PP data produces information on the Vp fast and Vp slow directions, the magnitudes of the velocities
and on the difference in the velocities, which can be related to fracturing. A similar analysis of the PS data,
examining shear wave splitting, also provides the information related to fracturing, including the fracture
directions.
The full azimuth, 3D multicomponent data provides valuable information on the fracture directions and
also information that can guide estimation of fracture intensity. These data are used as direct deterministic
inputs to the DFN model. The seismic data were interpreted and mapped horizons for the Marcellus
SPE-171608-MS 5

Figure 3—Overview of the workflow used to derive data and generate the inputs to the DFN modelling.

interval and the underlying Onondaga Limestone were generated. These provided input to the construction
of both the geocellular model and the DFN model. Interpreted faults also form one of the deterministic
inputs to the DFN model and are also used to construct the geocellular model, both models are consistent
as they use the same input data sets.
The inputs to the DFN model included:
● Information derived from the seismic volume including, interpreted and mapped horizons and
faults
● PP anisotropy data (magnitude and azimuth)
● HTI anisotropy (Epsilon)
● Seismic attribute volume
● Wireline data
● Seismic inversion results
● Dip information on fractures (from literature; Engelder et al, 2009)
Although the survey area is in close proximity to the Allegheny front, which forms part of the
Appalachian Structural front that formed during the Alleghenian Orogeny, approximately 320 – 250
million years ago, the faulting that affects the Marcellus is generally subtle, showing limited or no
observable offset in the reflectors at seismic resolution. There are exceptions to this, as seen in Figure 4,
however, as an aid to structural analysis and to generate data for the DFN, seismic attribute analysis was
undertaken. Whilst this could have utilised curvature analysis, fault probability analysis was conducted.
A fault probability slice is seen in Figure 4, and displayed in map view in Figure 5.
In Figure 5 the faults that are seen to affect the Marcellus interval in Figure 4 can be seen as strong
sub linear features. Also seen in the slice is a set of faults running from south south-west to north
north-east approximately in the centre of the slice, seen as bright whitish features. This has been
interpreted as a shear zone and runs parallel to the Allegheny front. A large number of smaller blueish
features are seen in the data, which are interpreted as smaller scale fault and fracture systems. A threshold
was applied to the fault probability data and those features above the threshold were classified as faults
and these were extracted and used to generate faults that populated both the geocellular model and used
as an input to the DFN model.
6 SPE-171608-MS

Figure 4 —Section through the 3D seismic volume, coloured by amplitude, the horizontal slice seen in the figure cutting through the Marcellus is a
Fault Probability slice which is shown in Figure 5.

One of the observations that can be made of a lot of DFN models is that the distribution of fractures
in the model space tends to be relatively uniform and that the sizes of the fractures often cover a relatively
small range. The fault probability data appear to indicate that the distribution of fractures is far from
uniform and therefore one of the objectives of the modelling described here is to honour the more random
distribution of fractures.
The large scale faulting that was extracted from the fault probability data formed one of the
fault/fracture sets in the DFN model; these were included in the model as deterministic features. The
production of the second set used the fault probability data as a deterministic input to the fracture
generation process – these data were also used to generate a fracture set in the DFN model. The objective
of generating the second fracture set was to produce fractures that honoured the information contained in
the seismic data that were visualized through fault probability analysis. The second fracture set is shown
in Figure 6 and a comparison of Figure 5 with Figure 6 show a very good correlation.
The processing of the seismic data provides information on both the fast and slow P-wave directions
and the difference in those velocities. In addition, shear wave birefringence data were available providing
directions and time differences. Both sets of data allowed for an understanding of fracture directions.
These data were utilized as a deterministic input to the generation of smaller scale fracturing that was
based on the anisotropy as measured by the multi-component 3D seismic data. The information derived
from the azimuthal velocity analysis was also used as a deterministic input to generate a fracture set. The
third fracture set produced for this model is shown in Figure 7. The complete DFN model is shown in
Figure 8.
SPE-171608-MS 7

Figure 5—Fault Probability slice through the Marcellus interval.

Figure 6 —Fracture set based on Fault Probability Data


8 SPE-171608-MS

Figure 7—Fracture set based on HTI anisotropy computed from the results of the azimuthal velocity analysis conducted during processing

Figure 8 —The complete DFN model; combining the deterministic features and the fractures sets generated based on fault probability data and HTI
anisotropy information.

Once a DFN has been constructed, the model can be validated through history matching to production
data and can be used to examine flow and compartmentalistion. The model can also be used to examine
SPE-171608-MS 9

Figure 9 —DFN model showing the fractures generated based on fault probability, two theoretical laterals and the fractures that intersect the wells
are shown as discs.

well placement, in terms of interwell distances and drilling azimuths; it can also be used to examine
hydraulic stage spacing and stress effects. To enable scenario testing to be conducted, it is necessary to
populate the model with elastic and geomechanical properties, along with the three principle stress
magnitude and directions. It is noted that the local stress directions are very variable and a constant
maximum horizontal stress direction cannot be assumed and if it is, it will result in the generation of
pseudo-planar tabular fractures in hydraulic fracture simulations.
As shown in Figure 3, petrophysics and rock physics analysis of the available well data were
performed. These were used to develop rock physics models and guide the seismic inversion – both
simultaneous and joint seismic inversion of the PP and PS data sets were undertaken. These analyses were
used to develop an understanding of litho-types and total organic carbon (TOC) distribution.
The DFN model was thus populated with Young’s Modulus, Poisson’s Ratio, and the three principal
stress magnitudes along with their directions. Having completed this phase, it was then possible to forward
model hydraulic fracture stimulation programmes. The stages can be defined in the wells, pump schedules
created and the relevant pump pressures and fluid densities can be defined. The process of hydraulic
fracture simulation commences with an examination of the fracture sets in the model that intersect the
wells and a critical stress analysis. The results of determining which fractures intersect the well are shown
in Figure 9, where the individual fractures are represented as discs of equal radius and independent of the
actual size of the fracture.
Without doing any forward modelling or simulation, a number of observations can be made directly
from this figure. The figure illustrates that a large number of fractures intersect the two wells in the model
(only one fracture set is displayed). It also shows that there are more fractures towards the toe and heels
of the wells and it may be anticipated the completions maybe easier or better in those two areas. The figure
10 SPE-171608-MS

Figure 10 —Optimised well placement; in terms of landing position and drilling azimuth of the laterals.

Figure 11—Storage aperture for one of the wells

also shows that fault/fracture systems intersect both wells and that communication maybe anticipated from
one set of stages in one well with stages in the other well.
Forward modelling the hydraulic fracture process was performed on a well-by-well basis and the results
of this are shown in Figure 10. The two well trajectories shown have been optimized; the landing location
has been moved between 100 and 200 feet to the north east for both wells and the drilling azimuth changed
by approximately 10 degrees toward the south. The efficiency of the hydraulic fracture simulation process
was shown to have increased by 3 to 30% stage to stage with this simple refinement in design. The
simulation used local stress orientations and it can be seen that the modelled fractures, both reactivated
and generated are curvi-linear. It also shows that the fractures generated in one well intersect the other
well, also seen is that there are more fractures generated in areas of higher natural fracturing in the model.
SPE-171608-MS 11

Figure 12—Locations of simulated micro-seismic events.

The left image in Figure 10 shows towards the centre of the lateral the fractures generated roughly parallel
to the well and this may be expected as the natural fractures have the same orientation and the local stress
directions are aligned. The modelling showed that the hydraulic fracture stimulation efficiency would be
decreased in the central section of the lateral as the pressures would tend to close the fractures and this
would be exacerbated by rotating the well azimuths a few degrees (~10 degrees) to the east.
The forward modelling of the stimulation process allows for the definition and visualisation of the
storage aperture and this can be thought of as a proxy for propped volume. This, along with the drainage
volume computations, may represent a better estimate of stimulated rock volume (SRV) in comparison
with the SRV defined by the extent of the micro-seismic cloud, which may indicate zones of pressure and
stress modification without any accompanying improvement in production. This is shown in Figure 11,
where the warmer colours show higher values and may be indicative of larger propped volumes. The
modelling also allows for the simulated micro-seismic events to be generated and these can be visualized,
Figure 12.
Having forwarded modelled the micro-seismic events, the locations of these can be compared with the
results of micro-seismic monitoring. The monitoring data can, through hypo-centre location and moment
tensor inversion, be used to provide a direct feedback and calibration of the model and also provide
additional fracture data.
Figure 12 shows clouds of simulated micro-seismic events that would be generated through the
stimulation process for one of the wells. These events are coloured by stage and show that there is a lot
of overlap between stages, which implies that the stages could have be placed further apart and should
have resulted in the same stimulated rock volume. The events also show the curvi-linear nature of a lot
of enhanced and generated fractures.
Conclusions
The case study described here indicates that a realistic discrete fracture network model can be constructed
that honours the information contained and derived from the multi-component 3D seismic data. The
12 SPE-171608-MS

acquisition of 3D seismic data is extremely relevant and of significant importance in developing an


understanding of the subsurface fracture systems and optimising well placement in resource plays. The
modelling provided an enhanced understanding of the reservoir, providing greater insight than would be
obtained from co-rendering. Importantly, the model tested different well positions, stage spacings and
drilling azimuths. In this case study the locations and azimuths of the laterals were changed, with
differences in the stage to stage stimulated volume of between 3 and 30%. This was particularly the case
for the central stages of the modelled wells and this has clear financial implications in the development
of the resource in terms of drilling, completions and production costs. The detailed understanding
developed through such comprehensive examination of the fracture systems and stress directions indicates
not only that drilling laterals in one particular direction is unlikely to produce optimum results but
different drilling azimuths are most likely to be required at different levels and for different shale units.

Acknowledgements
The authors acknowledge ION Geophysical for permission to present this work and colleague Farshid
Forouhideh for his contribution and those who reviewed the paper, Paul Brettwood and Jacques Leveille.

References
Engelder, T., Lash, G. G. and Uzcátegui., R. S., 2009. Joint sets that enhance production from Middle
and Upper Devonian gas Shales of the Appalachian Basin. AAPG Bulletin, V. 93, No. 7 (July 2009) pp.
857–889.
Gale, J F W, Reed, R M and Holder, J. 2006. Natural fractures in the Barnett Shale and their
importance for hydraulic fracture treatments. AAPG Bulletin, Vol. 91, No 4 (April 2007), pp. 603–622.
PFC Energy (2014). Operator Breakeven Analysis.
Roger, S., Elmo, D., Dunphy, R. and Bearinger, D. 2010. Understanding Hydraulic Fracture Geometry
and Interactions in the Horn River Basin through DFN and Numerical Modelling. CSUG/SPE SPE-
137488.

You might also like