An Integrated Workflow to Optimize Discontinuity Attributes for the Imaging

4
7 of Faults
EP

TOC

Start Brouwer, Friso Huck, Arnaud

SS

dGB Earth Sciences USA dGB Earth Sciences

Author
1 Sugar Creek Center Blvd, Suite 935 Nijverheidstraat 11-2
Search Sugar Land, Texas 77478, USA 7511 JM Enschede - The Netherlands
C

email: friso.brouwer@dgbes.com email: arnaud.huck@dgbes.com

Help
G

Print 8.5
x 11 Abstract
In the first part of this paper, we give an introduc- In the second part of the paper, we describe an
1

tion in seismic attributes for fault detection to set the integrated workflow to produce optimized discontinuity
01

context to understand imaging of faults with seismic volumes for fault detection First we discuss several pre-
attributes as an integrated workflow. Several well- processing options to optimize the data for the
application of the discontinuity attribute. Second we
2

known attributes for discontinuity analysis are dis-

show the effects of several parameter settings of the
cussed: coherency, similarity, variance, and semblance.
similarity attribute. Finally, we discuss the application
ht

After discussing these attributes and their different rep- of postattribute filters (filters applied to the discontinu-
resentation of the seismic and geological characteristics ity attribute) to demonstrate how the attribute image
of a dataset, we focus on the dip-steered similarity attri- can be enhanced by improving the completeness, reso-
rig

bute. lution, and contrast of the similarity attribute.

Introduction
y
op

Discontinuity volumes are an important tool, increasingly used in automated workflows for seismic
often used by interpreters to assist manual fault inter- interpretation. However, in our experience, there are too
pretation. In addition, discontinuity volumes are many instances where discontinuity attribute analysis
C

Attributes: New Views on Seismic Imaging—Their Use in Exploration and Production 496
 M

yields sub-optimal or disappointing results, especially parameters used to compute the discontinuity
4 when applied by non-experts. In our opinion, improve- attribute, and image enhancement applied post-
3
EP

ment can be achieved easily by applying a more processing to the discontinuity volume.
thoughtful and integrated approach. However, in the 2. Provides an overview of typical methods that can
7 available literature, although there are many papers on be used in the different stages of this workflow.
7
individual attributes and filters, there is a lack of papers While we focus on developing a workflow with
SS

describing a multistep workflow for optimizing discon- the similarity attribute, the workflow or structure
tinuity attributes in a systematic, high-level, holistic described can be applied to all attributes in the class of
manner. In this paper, our aim is to fill that gap and to discontinuity attributes, as well as related structural
C

provide the non-expert user with a guide that: attributes, such as the class of curvature attributes. In
1. Provides a framework for a systematic workflow addition, while here we focus on fault detection, the
G

using a holistic approach, in which all optimiza- same workflow is valid when using discontinuity attri-
tions are considered with the final application of butes to target other geological features, such as incised
the discontinuity volume in mind. This includes stratigraphic features, certain forms of reefs, or collapse
1

conditioning the seismic data, optimization of the (paleokarst) structures.

01

Discontinuity and Other Seismic Attributes for Highlighting Faulting

2

Rationale behind using seismic attributes to highlight faulting

ht

A proper interpretation of faults is essential in 1. The interpretation is often a low resolution and
any seismic interpretation for the purpose of explora- simplified version of the actual fault (system).
tion for, or development of, hydrocarbon reservoirs. First, the actual geologic faults or fault zones are
rig

The conventional workflow for fault interpretation is a only partially imaged in the seismic data. Then,
manual picking of fault sticks by the seismic interpreter the interpretation of the seismic fault system is
on seismic lines and time slices: see Brown (2004). In generally a subsampled and simplified represen-
y

the next step, the individual faults sticks are sorted per tation of the fault system as it is visible in the
op

fault and interpolated to construct fault planes. The seismic data. Compounded, this leads to a geo-
main disadvantages of this manual interpretation work- logically unrealistic simple model of the subsur-
flow are: face.
C

Brouwer and Huck 497

 M

2. Manual interpretation is time consuming. Obviously, there are disadvantages and problems
4 using seismic attributes. The most obvious two are:
3. Manual interpretation is subjective.
3
EP

4. In certain orientations of reflector dip, fault dip,

1. Seismic and geologic artifacts that are high-
7 and the section or slice through the seismic vol-
lighted by the same seismic attribute as faults can
7 ume, the discontinuities in the seismic data repre-
senting faults are difficult to discern on cyclic confuse the interpreter. Examples are surface
SS

seismic images. static effects and stratigraphic incisions that may

have discontinuity as an attribute. This is the flip
Using seismic attributes to highlight faulting in
combination with manual fault interpretation mitigates side of working objectively with the measured
C

these problems: data, as a human seismic interpreter will subjec-

tively filter such artifacts when manually inter-
G

1. A proper fault attribute displays the fault struc-

preting faults. Gaining more experience with
tures in nearly the same resolution as the seismic
data. Details such as Riedel structures, splinter attributes, a seismic interpreter will learn to apply
1

faults, and fault relay ramps show and are easier a similar subjective filter working with seismic
attributes. Additionally, there are several tech-
01

to interpret. Such images make it possible to infer

secondary geological attributes such as paleo- niques to minimize the presence of artifacts dis-
stress regimes and vertical conductivity of faults cussed later in this paper.
2

to hydrocarbons and other fluids.

2. Some faults (portions) are obscured even when
2. The manual interpretation of faults is sped up by using seismic attributes. Such obscuring can be
ht

using seismic attribute displays. In addition, auto-

caused by several factors. For example, the seis-
mated methods can be used to create a model of
the faulting, as demonstrated by Pedersen et al. mic data may be of poor quality, geologic factors
rig

(2002), Dorn et al. (2005), and Jacquemin and can cause certain attributes to fail (including the
Mallet (2005). fault dip-slip being much lower than seismic res-
3. A seismic attribute provides an objective transla- olution), or incidental juxtapositions of similar
y

tion of seismic data to a geologically meaningful seismic phase and seismic amplitude on both
op

image. sides of the fault will locally obscure the fault.

4. Faults that have been obscured due to the cyclic Several techniques exist to minimize these prob-
nature of the seismic data become more visible. lems and will be discussed later in this paper.
C

Brouwer and Huck 498

 M

The use of discontinuity attributes for fault detection

4
3 The most obvious characteristic of geologic Correcting multitrace (discontinuity) attributes for
EP

faults is lateral discontinuity of the geologic strata. The the seismic reflector dip
7 equivalent seismic representation is the discontinuity of
7 A prerequisite to the creation of multitrace attri-
seismic reflectors. Following this reasoning, multiple
SS

butes, such as discontinuity attributes, is the knowledge

seismic attributes highlighting seismic discontinuity, or of the dip of the seismic events; if horizontal seismic
its inverse, the seismic continuity, have been developed. events are assumed, the attribute will return faulty inac-
Examples are: cross-correlation (Bahorich and Farmer, curate values in the presence of seismic dip. Several
C

1995), semblance (Marfurt et al., 1998b), similarity discontinuity attributes, for example coherency, calcu-
(Tingdahl and de Rooij, 2003), variance (van Bemmel late the dip instantaneously as the attribute is
G

and Pepper, 2000), and eigenstructure (Marfurt et al., calculated. Using dip-steered similarity, a dip-steering
1999). Note that this is not an exhaustive list, and other volume is extracted a priori and stored as a separate
volume. This has the advantaged that the dips can be
1

seismic attributes that highlight seismic discontinuity

exist. With the proliferation of attributes to measure validated and conditioned easily. This is discussed later
01

seismic discontinuity, it is impossible to provide detail in the paper; we will first explain the basic principles
on each method. Below, we will explain dip-steered for calculating the dip from the seismic volume.
similarity in detail, and provide a high-level compari- Mathematically, geological dip is defined as a
2

son with some of the other discontinuity attributes. In unit vector in 3D space, normal to the geologic layer
the remainder of the article, we will use dip-steered plane. One can translate the concept of geological dip to
ht

seismic dip, which has its direction thus defined by two

similarity as the principle discontinuity attribute unless
numbers, the polar dip and the azimuth. The polar dip is
noted otherwise. However, the reader should note that
rig

the largest dip of the seismic reflector, measured from

the choice for dip-steered similarity is coincidental with the horizontal and increasing down with positive val-
our experience and that other (discontinuity) attributes ues. The azimuth is the direction of the dip vector,
can be substituted in the workflows described in this measured from the north. As the seismic data exists
y

article. In our opinion, most discontinuity attributes only on a 3D grid, the seismic dip is calculated in refer-
op

provide comparable results. For detailed explanations ence to the seismic survey (inline, crossline, Z), and
of the other methods, we refer to their principle papers stored as inline dip and crossline dip, which are equiva-
mentioned before. lent to polar dip and azimuth.
C

Brouwer and Huck 499

 M

The dip is extracted from the seismic amplitudes Creation of discontinuity attributes using dip-
4 steered seismic amplitude extraction
using a 3D probe centered at the actual sample location.
3
EP

Several methods have been developed to convert the

Discontinuity attributes are nearly always created
7 collected seismic amplitudes into dip values. A first by performing mathematical operations between lateral
7 method consists uses a Fourier-Radon transform trace segments and extracted using local time gates.
SS

(Tingdahl and de Groot, 2003). Locally, the seismic This applies to variance, coherency, similarity, and
amplitudes are transformed from the time domain into semblance-based computations that form one family of
discontinuity attributes.
the wave number frequency domain and then to the
C

slowness frequency domain. Integrating over the fre- One of properties controlling the output of a dis-
continuity attribute is the choice of the operator applied
quency axis and then picking the maximum amplitudes
G

to the trace segments, to estimate their dissimilarity.

in the transform domain provides the dip from the slow- Variance, coherency, similarity, and semblance each use
ness coordinates of the picked maximum. A second a different operator. The variance is a simple statistical
1

method is the extraction of the dip from the calculation property of the seismic amplitudes. The coherency is
01

of the amplitude or phase gradient (A. Wilson, personal based on a dot-product cross-correlation computation.
communication, 2008). This method is based on the The similarity uses the Euclidian distance, while the
semblance computes a squared sum between the trace
strong intrinsic anisotropy of the seismic data: the
2

segments or vectors.
amplitudes or phases vary much quicker perpendicular
A noticeable difference between them is that
ht

to the layer plane than along a layer. Therefore, the esti-

while coherency and similarity work between multiple
mation of the direction of maximum gradient allows the pairs of vectors and then are recombined using another
derivation of the reflector dip. This method is much
rig

measure, variance and semblance can natively incorpo-

faster than a Fourier-Radon transform as it requires nei- rate more than two traces in their computation. For dip-
ther an oversampling of the input seismic nor a costly steered similarity, the vector pairs used for computing
the attribute are most often the trace segment of the
y

3D FFT transformation. A third method of dip calcula-

center trace, and a selection of neighboring traces. Note
tion used by introduced by Randen et al. (2000) to
op

that the vertical location of the trace segments center

create a dip volume is based on the gradient structure varies from trace to trace when using a dip-steering cor-
tensor. rection, illustrated in Figure 1.
C

Brouwer and Huck 500

 M

Similarity attribute parameterization have a major effect on the attribute, as shown in

4 Figure 3.
Most significant variations between the represen-
3
EP

tations of the similarity attribute are due to three 3. Choice of lateral and vertical analysis windows:
7 attribute parameters: The distance between the trace locations that are
7 1. Choice in statistics: In the similarity attribute, compared, directionality of the extracted trace
SS

several measurements of similarity in different locations, and the vertical trace segments that are
directions and trace pairs are recombined. The used around the center point are major factors in
statistic chosen to recombine the different mea- the attributes’ appearance.
surements affects the output. Figure 2 shows the
C

difference between the maximum and minimum Between the different discontinuity attributes, it
output in an 8-trace pairs similarity attribute. is our experience that the choice of attribute (operator)
G

2. Handling of the seismic dip: Usually the disconti- is of lesser importance, and a correct parameterization
nuity attribute is corrected for the seismic dip. of the attribute is of greater importance to create the
1

This dip can be conditioned to represent a local best possible results. The testing of attribute parameter
dip or a subregional dip. This conditioning will is discussed in more detail later in this paper.
01

Other attributes for fault detections

2

Though discontinuity is the principal attribute for property that is mathematically independent of discon-
fault detection, there are other attributes, and methods tinuity. Curvature is especially useful in highlighting
ht

based on multiple attributes to highlight faults. As the faults having a dip-slip that is small compared to the
workflow presented in this paper is applicable to these seismic wavelength. These faults often look like flex-
alternative attributes and methods, we provide an over- ures of the seismic reflector, are not detected by
rig

view of the main alternatives. discontinuity attributes, but can be picked-up by the
curvature attribute. A second application of curvature
Curvature for fault detection arises when the geological strata are
y

folded at the sides of faults due to fault drag, as seen in

op

Seismic curvature is a measure of the amount of Figure 4. Note that contrary to the discontinuity attri-
bending in a seismic reflector (Fig. 4). A nice feature of bute, the curvature attribute normally has its extreme
curvature for fault detection is that it uses a seismic values at the sides of the fault plane instead of in the
C

Brouwer and Huck 501

 M

center. So a filter or numerical manipulation should be ferent extractions will capture different features of the
4 used to place the strongest response in the center of the data and then are recombined in a single, optimized
3
EP

fault plane. fault volume. Examples are the cross-plotting approach,

described by Chopra and Marfurt (2009), and the neural
7 Multiattribute methods network method described by Tingdahl and de Rooij
7
(2005).
SS

Often, single attributes do not optimally highlight

targeted geological features such as faults. In some Choosing the right fault attribute
cases, non-target seismic or geologic features are high-
lighted leading to confusion. In other cases, the entire Proper choice of the principal fault attribute, dis-
C

target feature is not highlighted, leading to oversights. continuity, curvature, or "multiattribute," can be
G

Often both problems occur at once. A solution is to essential for achieving satisfactory results. Therefore,
recombine several parameterizations of an attribute we present in Table 1 a number of criteria that may help
and/or recombine different types of attributes. The dif- the reader to select the proper approach.
1
01

Optimization of Discontinuity Attributes by Applying an Integrated Multistep Workflow

Although many tools for discontinuity attribute 3. Postprocess image enhancement filters applied to
2

analysis exist, the approach of many practitioners con- the discontinuity or generalized fault attribute.
sists of choosing one particular attribute and applying it The best strategy to create an attribute having
ht

to unconditioned input data using the default attribute optimal fault imaging is to sequence systematically
parameters. In our experience, this approach often leads through these steps and if necessary iterate. However,
to suboptimal, if not disappointing, results. Major
rig

before starting at all, the interpreter should hold back,

improvements over a default “push-button” approach to think the problem through, and design a plan of attack.
(discontinuity) attributes will come from optimization This strategy implies reconsidering the exact objective
applied in three steps: of the attribute analysis, identification of problems with
y

the input dataset, and identifying opportunities (A) to

1. Conditioning of the seismic and steering data.
op

solve these problems and (B) to improve further the

2. Customization and optimization of the attribute results in context of the ultimate objectives using the
parameterizations. particular "toolbox" the interpreter has available.
C

Brouwer and Huck 502

 M

Objectives, present applications, and future applications of discontinuity attributes

4
3 To understand the need for optimization, and • Auto-tracking of seismic horizons: The disconti-
EP

which method(s) should be applied, it is important to nuity volume acts as a boundary to prevent the
7 establish the relation between the optimization methods auto-tracker from continuing through fault planes
7
and ultimate objective(s) of the discontinuity attribute. to an incorrect reflector in another fault block.
SS

Here, we give an overview of typical objectives and Figure 5 provides an example of this application.
associated considerations for optimization: For this purpose continuity of the discontinuity
volume is the most important, as a single gap can
• Assisting manual interpretation of faults: The
C

allow the auto-tracker to break through into a

main consideration is to map all faults (complete-
neighboring fault block. Artifacts are less of a
G

ness of the attribute) and create a visually pleas-

ant output for the interpreter to work with. It is consideration, as gaps in horizons due to false
less important to remove all artifacts (uniqueness positives can easily be addressed in a postauto
1

of the attribute) as the interpreter will be able to tracking interpolation.

01

recognize many of the artifacts. However, it is • Prediction of permeability and vertical connectiv-
preferential to remove artifacts that have a fault- ity (D. Burch, personal communication, 2008):
like appearance on time slices, such as linear Discontinuity attributes can be used to predict
2

channels. zones of enhanced permeability both in lateral

• Automated fault extraction using discontinuity and vertical direction. Examples are in optimiza-
ht

attributes: The most important consideration is tion of well completions in unconventional

the removal of noise and artifacts that do not rep- hydrocarbon reservoirs and predicting water
rig

resent faults. Some applications may require breakthrough in the Barnett shale gas reservoir. In
numerical connectivity between discontinuity the critical application of reservoir optimization
patches representing the same fault. However, a one needs a very accurate imaging of fault-
y

complete representation of all faults by disconti- related seismic discontinuities. Thus, one needs
op

nuity voxels would not be necessary as gaps in to address the conflicting requirements of imag-
the fault surfaces would be easily interpolated ing all faults and avoiding inclusion of noise and
postextraction. artifacts in the discontinuity volume.
C

Brouwer and Huck 503

 M

Step 1: Conditioning of the input volumes

4
3 Important improvements of discontinuity attri- applying a cut-off or visually by adjusting the color bar
EP

butes can be achieved by proper conditioning of the clipping.

7 input seismic data. In the case of dip-steered disconti-
7 Several filtering methods may be applied to
nuity attributes, the dip-steering should also be lower noise levels in the seismic data. We apply a
SS

conditioned to achieve optimal results. We will discuss method based on a dip-steered statistical filter operator.
four types of filtering: Alternate approaches are based on a Radon transform
1. Improving the seismic data by removing disconti- type filter (Marfurt, 1998a), on singular value decom-
C

nuities not related to geology. position of the seismic data (Al-Bannagi et al., 2005;
2. Improving the steering cube to correctly repre- Guo et al., 2009) or on adaptative subtraction algo-
G

sent the sub-regional geological dip. rithms (Davogustto et al., 2009).

3. Additional conditioning of the seismic data: fre-
In Figure 6, an example is presented of the
quency enhancement filters and fault enhance-
1

removal of the acquisition footprint using a dip-steered

ment filters.
median and averaging filter. The filter operator is
01

4. Removing stratigraphically oriented noise

aligned with the seismic reflectors using the same dip-
steering principle as applied to the similarity. First, a
Removing artifacts from the seismic data
median filter is applied to the seismic data. Note that
2

Most disturbing to geometric attributes such as the median operator is a step-preserving operator.
discontinuity or curvature are acquisition footprint, Therefore, it will remove outliers but will retain steps in
ht

coherent noise, and certain processing artifacts, such as data values as encountered near faults. The median
“smiles” due to over migration or uncollapsed operator, however, may also retain some of the steps or
rig

diffractions. periodic variations encountered in acquisition foot-

The challenge is to find a seismic filter that sub- prints. In this case, an optional dip-steered averaging
dues "artifact" discontinuities while leaving filter can be applied in sequence. It is key to keep the
discontinuities due to faults unaltered. Note that one operator size of the averaging filter to a minimum, as
y

does not necessarily need to remove the noise alto- the averaging operator is not step-preserving and a
op

gether. Once the dynamic range of the fault-related small operator size minimizes smoothing of the seismic
discontinuities has become separable from the dynamic data over faults. The choice whether to apply a final
range of the artifacts, then one can isolate the faults by averaging step is best made empirically by comparing
C

Brouwer and Huck 504

 M

test results on slices similar to what is shown on Figure 8. An alternative approach is to apply a low pass
4 Figure 6. filter in the frequency-wave number domain.
3
EP

Improving the steering cube

7 Improving the seismic data to emphasize
7 Described earlier in this paper, many discontinu- discontinuities
SS

ity attributes use the seismic reflector to correct trace

segment extraction at step-out traces of the seismic dip. Contrary to removing noise, one can also apply
This can be done a priori, such as with the dip-steered certain operators to the seismic data in which the objec-
similarity, or simultaneously, such as with the dip-scan tive is to improve the imaging of the faults. In our
C

coherency (Marfurt et al., 1999). The first method has workflow, we typically apply two steps: optimization of
the advantage that the dip-field used can be conditioned the frequency content using seismic spectral blueing
G

for optimal results. The main optimizations that one can (SSB) and applying a fault enhancement filter to the
apply are removal of noise in the dip field and remov- seismic data.
ing local information of dips in favor of subregional
1

Often, seismic data are not optimized for the fre-

dips. The latter gives a better result as the local dip cal- quency content. If well data and density and P-wave
01

culated near a fault may be unreliable, as most logs are available, one can apply a geologically (well-
calculations of the seismic dip fail near a seismic dis- log) driven optimization of the frequency spectrum in
continuity. More important, the local dip, if correct, seismic data using SSB. The theoretical justification for
2

may adjust the extraction of trace segments at opposite this method is given in Velzeboer (1981) and Walden
sides of the faults such that the almost identical trace and Hosken (1985) and a practical application is pre-
ht

segments are extracted and thus the fault is obscured. sented in Blache-Fraser et al. (2004). In the context of
This is the case when the fault has a small dip-slip com- discontinuity attributes, the significance of SSB is that
pared to the seismic resolution, or if there is
rig

the seismic visibility of faults having small dip-slip will

considerable folding of the seismic reflectors near the be enhanced. These small faults can have large impact
faults; see Figure 7. Simultaneous dip-scan methods on the production behavior of hydrocarbon reservoirs
will generally be limited to local dip and may suffer because of reservoir compartmentalization and/or verti-
y

from this drawback. See Figure 3 for an illustration of cal communication of fluids along the fault plane.
op

the effect of using local versus subregional dip-fields. Therefore, frequency optimization should be a default
There are different methods for filtering the dip- step in fault detection for reservoir optimization.
field. We generally apply statistical filters, as shown in Figure 9 provides a comparison of a seismic section
C

Brouwer and Huck 505

 M

before and after optimization, illustrating the improve- Removal of stratigraphically oriented artifacts
4 ment in fault visibility. around the seismic zero-crossing
3
EP

Once the best frequency content has been estab-

lished, one can apply a fault enhancement filter to the Often, discontinuity attributes show artifacts
7 seismic data. This step is optional and can be beneficial aligned with seismic zero-crossing or low amplitude
7 reflectors. This effect is stronger when relatively small
if there is a low seismic data quality in the direct vicin-
SS

ity of the faults. The fault enhancement filter is an time windows are used in the discontinuity attribute.
adaptive, conditional filter that sharpens discontinuities Though often there is no real hindrance, these strati-
near suspected faulting and smoothes the seismic graphically aligned anomalous zones do not represent
C

reflectors at all other locations. First, a quality volume faults; thus ideally they should be removed. An
is processed to tag a certain location as part of (near of) improvement can be achieved by using the analytic
G

a fault or part of non-faulted seismic data. If the loca- trace as input instead of the real trace (Chopra and Mar-
tion is near or on a fault, a sharpening filter is applied furt, 2007). Figure 10 shows the result of statistically
based on the gradient observed in the similarity. Else- combining dip-steered similarities with Real and Hil-
1

where, a smoothing filter is applied to the seismic data. bert transformed traces as input to the discontinuity
This workflow has the effect that the seismic amplitude attribute, which minimizes the stratigraphically aligned
01

data become more continuous everywhere, except at anomalies. It also shows the result of an alternative sta-
faults where they become strongly discontinuous. A tistical filter applied to the discontinuity output, rather
2

workflow with a similar effect is described in detail by than the seismic input. Which filter to use is best deter-
Fehmers and Hoecker (2003). mined empirically.
ht

Step 2: Optimization of attribute parameters

Once the input data are optimized, the next step is localization and resolution. A drawback of
rig

to optimize the discontinuity attribute itself. This is an smaller time gates is the enhancement of strati-
empirical process, in which one needs to experiment graphically aligned artifacts, including both seis-
with the parameters of the discontinuity attribute: mic and numerical noise due to zero-crossings,
y

low reflectivity zones, and stratigraphically

op

• Time gate: The time-gate chosen will determine aligned features attributable to geology such as
the vertical localization of the discontinuity attri- erosional edges and mass transport systems. The
bute. Smaller time gates will result in higher second drawback of short time windows is often
C

Brouwer and Huck 506

 M

the loss of vertical continuity in the fault image. but laterally shifted next to it. This shift can be
4 Larger time windows provide, due to averaging, compensated during postprocessing; however,
3
EP

minimal stratigraphically aligned noise and will this is easy to avoid by setting up another sort of
provide larger vertical continuity of the faults but input geometry. Figure 14 presents an example of
7 will exhibit more vertical smearing. The best discontinuity attribute using the four pairs of
7
time-gate should be determined empirically and traces that coincide with the inline and crossline
SS

is dependent on the characteristics of the dataset directions. This attribute still features an effective
and the particular interpretation objective. The size of 1 bin but with connectivity between the
best practice to determine this setting is to mea- bins. The roughness seen previously decreases a
C

sure the dominant period of the seismic signal bit, and the peak amplitude actually corresponds
and then define several test time-gates as ratios of now to the location of the fault plane. Although
G

the dominant period. In the example of Figure 11, the roughness and interconnectivity are coming
we use 0.5, 1, and 2 times the dominant period T from the larger number of pairs (8 instead of 4),
as time gate. the positioning of the response is again directly
1

• Geometry and step-out: The most conventional related to the symmetry of the extraction loca-
01

choice for trace pairs would be two perpendicular tions. It is also possible to use the four diagonal
pairs towards larger inlines and crosslines. Also pairs in order to capture a network of faults ori-
one can imagine the same setting but in the ented along the survey diagonal directions. Nev-
2

reverse direction, or looking at the diagonal ertheless, the gain in detection of diagonal faults
directions. Figure 12 shows the result of such a will have to be balanced with the larger size
ht

discontinuity attribute on vertical and horizontal (+40%) of the extraction array and proportionally
sections. Such trace position settings generate the lower crispness of the attribute. Lastly, it is also
most detailed discontinuity attribute, with an possible to combine the eight pairs of traces
rig

effective size of one bin. However, its display around the central location (Fig. 15). This extrac-
may look aliased with an interpolated (non- tion setup provides the optimum continuity of the
blocky) display. But the largest drawback can be attribute response but at a subsequently lesser
y

seen when displaying the discontinuity attribute level of detail and more contrast between the dis-
op

on top of the seismic (Fig. 13). The actual posi- continuities and the background. In the case
tions of the maximum of the discontinuity attri- where the interpreter wants to focus selectively
bute values are not located at the fault location, on one directional fault set, one can limit the
C

Brouwer and Huck 507

 M

trace pairs to lie in certain directions (Fig. 16). faults as it will always take the similarity of a
4 The directional quality can even be improved trace pair that crossed the plane of maximum dis-
3
EP

more by certain postprocessing filters. continuity (usually a fault). However, it is more

sensitive to noise and will deliver a less "crisp"
7 Generally, it is advantageous to work with the
image. See Figure 2 for an example of the effect
7 smallest step-out distance for the trace pairs as possible.
However, if one has very low dip faults, such as the sole of the output statistic. Between the statistics
SS

of a growth fault or low angle thrust faults, it may be options available, it is generally advised to use
necessary to use larger step-outs to properly image the statistics that selects the trace pair having maxi-
faults. mum discontinuity (minimum similarity), as
C

missing fault (segments) are unacceptable.

• Output Statistic: In many discontinuity attributes
an intermediate result is produced that contains a Finally, the choice of input volume can make
G

set of discontinuity measures between many pairs import differences. Different volumes, such as prestack
of traces. As the final result should be one mea- versus poststack processed volumes or near versus far
sure, this set is often reduced using a statistical stack volumes, do often provide different seismic
1

operation. In the case of similarity one can images of the geologic fault system. In the case that dis-
01

choose minimum, maximum, or average. Choos- continuity attributes have different inputs provide
ing the maximum similarity will result in a complimentary images of the fault system, one can
"cleaner" image, which will not detect all faults, combine them using one of the multiattribute methods
2

as often the maximum pair will not cross the fault presented in Table 1. Before combining discontinuity
but is parallel to the fault. The minimum similar- attributes from different input volumes, care should be
ht

ity will provide a more complete image of the take to align the attributes in temporal and lateral sense.

Step 3: Postprocessing image enhancement

rig

Once one or more discontinuity attributes have ment of discontinuity type attributes. Which method to
been processed, several techniques are available to apply depends on specific problems encountered, the
y

reduce any remaining problems, to improve the general objectives of the discontinuity volume, and if the objec-
tive warrants the effort.
op

quality for interpretation or subsequent workflows, and

to create special purpose volumes. The following list • Stratigraphically oriented noise can be minimized
summarizes a number of options for image enhance- using a vertically oriented statistical filter. We use
C

Brouwer and Huck 508

 M

this as a first step when finalizing the discontinu- filtering (FCF) and fault-dip oriented statistical
4 ity volumes. The improvement of this step is filtering (FSF) and. The sequential application of
3
EP

illustrated in Figure 10 next to an alternative filter FCF and FSF can address issues as stratigraphi-
that can be applied preprocessing. cally oriented noise, vertical fault continuity, and
7 • It is not uncommon to arrive at different versions staircase behavior in one step. This is a powerful
7
of discontinuity attributes that each have different workflow that demands more user interaction
SS

complimentary images of the fault system. To during parameterization and has more computa-
simplify the interpretation, it is often a consider- tional intensity. It is most applicable if the discon-
ation to blend them into one volume but with tinuity volume is meant to be used in a
C

preservation of the optimal characteristics of each subsequent automated workflow, in which there
volume. The simplest approach is to apply a is the need for the volume to be noise free,
G

brute-force merge using an averaging or statisti- smooth, and continuous. Donias et al. (2007),
cal operator between the volumes. However, proposes a similar workflow.
more advanced methods are available, as summa- • Other authors have reported good results using
1

rized in Table 1. techniques from the field of image processing;

01

• To increase resolution and remove remaining for example Barnes (2005).

noise, a ridge-enhancement filter can be applied,
originally developed by de Rooij and Tingdahl Ridge enhancement filter
2

(2002). This filter is especially useful in structur- We frequently use the ridge-enhancement filter
ally complex areas having dense fault systems (REF). The filter is applied to a discontinuity attribute
ht

and in areas with strong stratigraphically oriented or other fault attribute. The filter is an adaptive second-
noise. Details of this filter are discussed below. order derivative filter applied in different lateral direc-
• The interpreter can separate different fault trends
rig

tions. Between the probed directions, the optimum

using directional decomposition of the disconti- direction, typically perpendicular to the fault plane, is
nuity attribute. This can be useful in both manual chosen as output. The advantages of using a second-
and automated fault interpretation. The method is order derivative are: (1) laterally continuous disconti-
y

described in more detail further below. nuity anomalies are obscured, minimizing stratigraphic
op

• The quality of faults can be considerably noise; and (2) the operator is symmetrical and provides
improved for the purpose of manual and auto- a single extreme output at the center of the fault plane,
mated interpretation using fault-dip conditional whereas first order derivatives are asymmetrical and
C

Brouwer and Huck 509

 M

result in extreme values of opposite sign at the sides of data-driven workflows, the fault-dip conditional filter
4 the fault plane. The advantage over the common Lapla- and fault-dip statistical filter are two workflows that
3
EP

cian filter, also based on second order derivatives, is the can significantly improve the quality of output disconti-
adaptive selection of the optimal direction of the opera- nuity volumes. The complete workflow has the
7 tor. An example of the effect of the REF is shown in following steps:
7
Figures 17C and 17D.
SS

1. Calculate a steering volume from the discontinu-

ity data. Typically, the volume will contain the
Directional decomposition of the discontinuity
dips of the discontinuities for which the disconti-
attribute
nuity attribute is anomalous and will contain the
C

Often the interpreter wants to separate one dis- dips of the seismic reflectors (stratigraphic dip)
tinct fault set at a time for interpretation. Directional elsewhere. The output of this step is shown in
G

decomposition of the fault attribute provides that possi- Figure 17.

bility. The first step of this workflow is often done 2. Apply the FCF. Generally, there is a significant
1

during attribute optimization, as described in the previ- difference between minimum fault dip and maxi-
ous section, by selecting only a directional subset of all
01

mum stratigraphic dip. In this case, one can use

available trace pairs in the dip-steered similarity. Then, the dip-difference to remove stratigraphically ori-
a final separation can be obtained by applying the ented discontinuities and only pass steeply dip-
ridge-enhancement filter, limited to a subset of its azi-
2

ping discontinuities typically associated with

muthal probes, followed by an averaging filter in the faulting.
perpendicular direction. The decomposition of the dis-
ht

continuity attribute in two directional attributes, each 3. Apply the FSF. This filter improves the vertical
representing one fault set is demonstrated in Figure 16, continuity of the faults by filling smaller gaps in
rig

showing the benefit for interpretation in a structurally the fault plane. In addition, the typical staircase
complicated setting. appearance of discontinuity attributes is mini-
mized. which is visually more pleasing and may
Fault dip conditional filter and fault dip statistical be useful when building geological models. The
y

filter filter works by applying a multibin statistical

op

operator typically with a median, average, or

If there is a need to make very high quality fault minimum operator, at the location of the faults.
volume that can be used in downstream automated The fault-dip steering volume is used to align the
C

Brouwer and Huck 510

 M

filter operator with the fault plane. A typical

4 result of this step is shown in Figure 17.
3
EP

7 Review and iterate through preprocessing, parameter optimization and postprocessing

7 For benchmarking purposes, Figure 18 shows a Going through the flow of data preconditioning,
SS

detail and the typical characteristics of a good quality attribute parameter optimization, and finally postpro-
discontinuity volume for fault detection. Typical quali- cessing, additional insights are gained on how to
ties include: (1) no stair case behavior; (2) the improve the results, as the initial result is not always the
best possible result. Generally after reviewing the initial
C

discontinuity attribute is nicely centered at the fault

result, one may want to improve and re-apply earlier
plane even if the reflectors are interfingering due to
steps. Experimentation using key lines and slices,
G

poor seismic data quality; (3) artifacts, particularly with instead of the whole volume, can be time-saving. A sys-
respect to stratigraphic features, are subdued; and (4) tematic approach and note keeping will lead the
for fault positions having an isophase juxtaposition, the
1

interpreter to a set of personal best practices that will

fault is still indicated (the complete faults are imaged). save time and improve results in future projects.
01

Conclusions
2

Using seismic (discontinuity) attributes for fault may be perplexing, and it is helpful to adhere to the fol-
imaging has several advantages. Manual fault picking lowing overall workflow:
ht

can be executed with greater speed and accuracy. In 1. Preprocessing: identify obvious problems of the
addition, several methods for automated fault extraction input seismic data and look for other optimiza-
rig

and horizon picking are enabled by good fault tions that could be applied to the input data.
attributes. Based on this analysis, design a preprocessing
strategy to precondition seismic and steering data
There exist many seismic attributes for imaging and select the most appropriate tools that are
y

reflector discontinuity for the purpose of mapping available.

op

faults, as well as many optimization tools to precondi- 2. Empirically optimize the parameter settings for
tion input data and enhance a discontinuity volume the attribute by on-the-fly testing on key sections
postprocessing. The abundance of methods available and slices. Note that different parts of the seismic
C

Brouwer and Huck 511

 M

data volume may require different optimizations. preter’s capability to understand both the complete
4 One can accommodate for that requirement by workflow and its steps in the context of the final objec-
3
EP

using multiattribute methods such as neural net- tive. True understanding, as opposed to a "push-button"
works, or one can merge different attributes approach, will enable the interpreter to identify prob-
7 mathematically; for example, using stratigraphi- lems, possible improvements, and apply the attributes
7
cally derived constraints. and optimization methods in a proper manner that
SS

3. Analyze the output of the optimized discontinuity increases the quality of the final result more than the
attribute, identifying any remaining problems and choice of one particular algorithm over another. This
possible improvements that can be made. Based does require that the interpreter invests into achieving a
C

on the analysis, design a postprocessing strategy certain knowledge level to understand, create, optimize,
and select again the most appropriate tools avail- enhance, and interpret fault attributes. However, once
G

able. knowledge and experience have been obtained, fault

4. QC the final result and iterate through the steps if interpretation workflows can be performed much
necessary. quicker and provide more important detail about the
1

Of course, the computer and interpreter time and fault regime at the same time. In addition, a number of
01

expenses that needs to be, or can be, devoted to the novel automated workflows come within reach of the
workflow depends on the final objective and its signifi- interpreter once he is able to produce a high quality dis-
cance. In addition, it is important that for most steps continuity volume. This next generation of
2

and problems encountered, multiple tools exist that interpretation tools will ensure the interpreter will
often provide comparable outcomes. Ultimately, the spend less time mapping and more time understanding
ht

outcome of the workflow is determined by the inter- the geology.

References
rig

Al-Bannagi, M., S., K. Fang, P.G. Kelamis, and G.S. Doug- Annual Conference and Technical Exhibition,
lass, 2005, Acquisition footprint suppression via the Extended Abstracts, p.318.
y

truncated SVD technique: Case studies from Saudi Bahorich, M., and S. Farmer, 1995, 3-D seismic discontinuity
Arabia: The Leading edge, v. 24, p. 832-834. for faults and stratigraphic features: The coherence
op

cube: SEG Expanded Abstracts, v. 14, p. 93-96.

Barnes, A.E., 2005, Fractal analysis of fault attributes Blache-Fraser, G., and J. Neep, 2004, Increasing seismic res-
derived from seismic discontinuity data: EAGE 67th olution using spectral blueing and colored inversion:
C

Brouwer and Huck 512

 M

Cannonball field, Trinidad: SEG Expanded Abstracts, Marfurt, K.J., R.M. Scheet, J.A. Sharp, and M.G. Harper,
4 v. 23, p. 1794-1797. 1998a, Suppression of the acquisition footprint for
3 Brown, A.R., 2004, Interpretation of Three-Dimensional seismic sequence attribute mapping: Geophysics, v.
EP

Seismic Data: SEG investigations in Geophysics, no. 63, p. 1024–1035.

7 9, p. 55-74 Marfurt, K.J., R.L. Kirlin, S.L. Farmer, and M.L. Bahorich,
7 Chopra, S., and K.J. Marfurt, 2007, Seismic attributes for 1998b, 3-D seismic attributes using a semblance-
SS

prospect identification and reservoir characterization: based coherency algorithm: Geophysics, v. 63, p.
Society of Exploration Geophysicists, 464p. 1150–1165.
Chopra, S., and K.J. Marfurt, 2009, Detecting stratigraphic Marfurt, K.J., V. Sudhakar, A. Gersztenkorn, K.D. Crawford,
features via crossplotting of seismic discontinuity and S.E. Nissen, 1999, Coherency calculations in the
C

attributes and their volume visualization: The Leading presence of structural dip: Geophysics, v. 64, p. 104–
Edge, v. 28, p. 1422-1426. 111.
G

Davogustto, O., Suarez, Y., and Marfurt, K., J., 2009, Foot- Pedersen, S.I., T. Randen, L. Sonneland, and O. Steen, 2002,
print removal using adaptative subtraction algorithms Automatic fault extraction using artificial ants: 64th
for seismic attribute quality enhancement. A case
1

Meeting, EAGE Expanded Abstracts, G037.

study of Anadarko Basin Red Fork incised valley sys-
Randen, T., M. Monsen, C., Signer, A. Abrahamsen, J.O.
01

tem: SEG Expanded Abstracts, v. 28, p. 3390-3394.

Hansen, T. Saeter, J. Schlaf, and L. Sonneland, 2000,
Donias, M., D. Ciprian, Y. Berthoumieu, O. Lavialle, S. Guil-
Three-dimensional texture attributes for seismic data
lon, and N. Keskes, 2007, New fault attribute based on
analysis: SEG Expanded Abstracts, v 19, p. 668-671.
2

robust directional scheme: Geophysics, v. 72, p. 39-

46. de Rooij, M., and K. Tingdahl, 2002, Meta-attributes - the
key to multivolume, multiattribute interpretation: The
Dorn, G., A., H.E. James, and L. Evins, L., 2005, Automated
ht

Leading Edge, v. 32, p. 1050-1053.

Fault Extraction (AFE) in 3-D seismic data: CSEG
National Convention, 247-250. Tingdahl, K.M., and P.F.M. de Groot, 2003, Poststack dip and
rig

Guo, H., K.J. Marfurt, K., J., and J. Liu, 2009, Principal com- azimuth processing: Journal of Seismic Exploration, v.
ponent spectral analysis: Geophysics, v. 74, p. 35-45. 12, p. 113-126.
Fehmers, G., C., and C.F.W. Hoecker, 2003, Fast structural Tingdahl, K.M., and M. de Rooij, 2005, Semi-automatic
interpretation with structure-oriented filtering: Geo- detection of faults in 3D seismic data: Geophysical
y

physics, v. 68, no. 4, p. 1286-1293. Prospecting, v 53, p. 533–542.

op

Jacquemin, P., and J. Mallet 2005, Automatic fault extraction Van Bemmel, P., and R. Pepper, 2000, Seismic signal pro-
using double Hough Transform: SEG Expanded cessing method and apparatus for generating a cube of
Abstracts, v. 24, p. 755-758. variance values: United States Patent 615155.
C

Brouwer and Huck 513

 M

Velzeboer, C.J., 1981. The theoretical seismic reflection Walden, A.T., and J.W.J. Hosken, 1985, An investigation of
4 response of sedimentary sequences: Geophysics, v. 46, the spectral properties of primary reflection coeffi-
3 p. 843-853. cients: Geophysical Prospecting, v. 33, p. 400-435
EP

7
7
SS
C
G
1
2 01
ht
y rig
op
C

Brouwer and Huck 514

 M

4 Table 1. Selection of the best fault imaging methodology.

3
EP

Criterion Best method

7 Most faults are well defined by discontinuities in the seismic reflector. Use a discontinuity type attribute.
7
Most faults are defined by seismic reflector flexures or seismic reflector
Use a curvature type attribute.
SS

folding at the sides of the faults.

The faults are not completely imaged by a single attribute, but the faults
Use cross-plotting and/or mathematical
are well imaged by a pair of attributes (i.e. discontinuity and curvature)
recombination to combine the two
C

or a single attribute type with two different parameterizations (i.e. short

attributes.
time window and long time window).
G

The expression of faults in the seismic data is very variable or weak

Use an interpreter driven neural
and three or more attributes or attribute parameterizations are needed
network.
to adequately image the faulting.
1
01

3
2
ht
y rig
op
C

Brouwer and Huck 515

 M

4
3
EP

7
7
SS
C
G

Non-steered: all samples Full-steered: the depth is

have the same depth determined by the dip field
1
01

Figure 1. The concept of dip-steering. 3D-view of the trace segments center location, without (left) and with (right) the
use a full-steering correction.
3
2
ht
y rig
op
C

Brouwer and Huck 516

 M

4
3
EP

7
7
SS
C
G
1

Maximum Similarity
01

Minimum Similarity
Figure 2. Side-by-side comparison of the minimum statistic (left) and maximum statistic (right) applied to a set of eight
similarity measurements, taken from trace pairs oriented in different directions. It is seen that the “minimum opera-
2

tor” renders a more complete image of the faults (red arrows), while the maximum operator renders an image having
high resolution (blue arrow) and less spurious highlights (green arrows).
ht

3
y rig
op
C

Brouwer and Huck 517

 M

4
3
EP

7
7
SS
C
G
1

Using regional dip Using local dip

01

Figure 3. Side-by-side comparison of dip-steered similarity using the regional dip field as steering (left) and a dip-
steered similarity using the local dip field as steering (right). Close inspection reveals that using the local dip fields
2

leads to a number of faults and fault segments not being imaged (see red arrows).
3
ht
y rig
op
C

Brouwer and Huck 518

 M

4
3
EP

7
A
7
SS
C
G

B
1
2 01
ht
rig

Figure 4. Curvature (red: positive curvature, antiform; green: negative curvature, synform) overlain on seismic section
in grey scale. Many faults are characterized by a positive/negative pair of curvature extremes on the side of the fault.
This pattern can be caused by sub-resolution fault dip-slip (arrow A) expressed as seismic flexures, geological folding
y

expressed as seismic reflector faults (arrow B) or a combination thereof.

3
op
C

Brouwer and Huck 519

 M

4
3
EP

7
7
SS
C
G
1
2 01
ht
rig

Figure 5. Automated horizon tracking in the presence of faulting, using a combination of seismic amplitude and dis-
continuity volumes as input to the horizon tracking algorithm. Each fault block is seeded independently by the discon-
tinuity volume preventing autotracker-tracking through the faults in the absence of an explicitly interpreted fault
y

model.
op

3
C

Brouwer and Huck 520

 M

4
A B
3
EP

7
7
SS
C
G

D C
1
2 01
ht
rig

Figure 6. Removing acquisition footprint, showing dip-steered similarity on the slice and input seismic on the sections.
Starting in the left upper corner, clockwise: (A) original seismic amplitude; (B) dip-steered median filter (DSMF),
y

operator size 1; (C) DSMF, operator size 2; (D) DSMF, operator size 2 followed by dip-steered averaging filter, size 1.
3
op
C

Brouwer and Huck 521

 M

4
3
EP

7
7
SS
C
G
1
01

Figure 7. Seismic data plotted using a variable area with wiggle overlay display. Note how the dip is used to extract
trace segments used for the discontinuity measurement over the fault. In the left image the local dip is used. The abnor-
2

mal dips due to fault drag near the fault plane cause a local correction such that nearly identical trace segments are
extracted. As a result the attribute will not yield an indication of discontinuity. Instead, if a subregional dip is used, as
ht

in the right image, two very different trace segments are extracted, and the fault will be highlighted by the discontinu-
ity attribute.
3
y rig
op
C

Brouwer and Huck 522

 M

4
A B
3
EP

7
7
SS
C
G

D C
1
2 01
ht
rig

Figure 8. Conditioning of a dip-field for dip-steering. Starting in left upper corner, clockwise: (A) input seismic data;
(B) raw dip-field, noise indicated by black arrows; (C) noise removed by vertically oriented filter, local dips indicated
y

by red arrows; (D) local dip removed by lateral filter.

3
op
C

Brouwer and Huck 523

 M

4
3
EP

7
7
SS
C
G
1
01

Figure 9. Frequency enhancement through the Seismic Spectral Blueing enhancing the definition of a series of faults in
a reservoir formation.
3
2
ht
y rig
op
C

Brouwer and Huck 524

 M

4 A B C
3
EP

7
7
SS
C

Figure 10. Comparison of two methods to minimize stratigraphically aligned artifacts. From left to right, (A) depicts
G

the original discontinuity attribute with some stratigraphically aligned artifacts indicated by red arrows. (B) The sta-
tistical recombination of two discontinuity attributes, one based on the real seismic volume and one based on the Hil-
bert transformed seismic volume. (C) Result obtained by applying a vertically-oriented statistical filter to the original
1

discontinuity attribute.
3
2 01
ht
y rig
op
C

Brouwer and Huck 525

 M

4 A B C
3
EP

7
7
SS
C

0.5T T 2T
G

Figure 11. Effect of the time window on the output of the discontinuity attribute. Time windows are best reference to
the dominant period T of the seismic data in the interval under consideration. The left to right time gate is 0.5T (A), 1T
(B) and 2T (C). Red arrows indicate locations that show artifacts or excessive smearing.
1

3
2 01
ht
y rig
op
C

Brouwer and Huck 526

 M

4 CrL
Blocky display “Natural” display:
with interpolation
3
EP

7
7
Inline Inline
SS

Z Slice
InL Z Slice
C

Effective size:
G

1 bin
1
01

Figure 12. Dip-steered similarity on an inline and intersecting time slice, with two pairs of perpendicular traces as
2

input. Note the roughness of the interpolated display due to the small bin size (red arrows).
3
ht
y rig
op
C

Brouwer and Huck 527

 M

4 CrL
3
EP

7
7
SS

InL
C
G

(Fault attribute
display is blocky)
1

Figure 13. Dip-steered similarity on a line overlaying seismic data, computed using two pairs of perpendicular traces
01

as input. Note the position of the amplitudes, always on one side of the fault.
3
2
ht
y rig
op
C

Brouwer and Huck 528

 M

4 CrL Blocky display “Natural” display:

3 with interpolation
EP

7
7
Inline
SS

Inline
Z Slice Z Slice
InL
C
G
1
01

Figure 14. Dip-steered similarity on an inline and intersecting time slice using four pairs of perpendicular traces as
2

input. Note the reduced roughness compared with Figure 12.

3
ht
y rig
op
C

Brouwer and Huck 529

 M

4 CrL
Blocky display “Natural” display:
3 with interpolation
EP

7
7
Inline
SS

Inline
InL Z Slice Z Slice
C
G
1
01

Figure 15. Dip-steered similarity on an inline and intersecting time slice using eight pairs of traces as input. Note the
2

reduced roughness compared with Figures 12 and 14.

3
ht
y rig
op
C

Brouwer and Huck 530

 M

4
3 A B
EP

7
7
SS
C
G

D C
1
2 01
ht
rig

Figure 16. Directional decomposition of a discontinuity attribute shown on a time slice. Clockwise starting top left. (A)
the original discontinuity volume; (B) Northeast-southwest decomposition; (C) Northwest-southeast decomposition;
(D) transparent overlay of the two trends showing their interdependency.
3
y
op
C

Brouwer and Huck 531

 M

4 A B
3
EP

7
7
SS
C
G

D C
1
2 01
ht
rig

Figure 17. The result of applying structurally oriented filters using the fault dip. Clockwise from upper left. (A) Dis-
continuity attribute used as input; (B) Crossline dip component of the dip in the discontinuity volume. (C) The discon-
tinuity volume after application of FCF and FSF. (D) After the additional application of the ridge enhancement filter.
3
y
op
C

Brouwer and Huck 532

 M

3
EP

7 A C
B
SS
C
G
1
01

Figure 18. Illustration of a high-quality discontinuity volume. The volume does not exhibit staircase behavior and is
nicely centered about the fault plane, even in the presence of interfingering reflectors (A). Discontinuities not associ-
ated with faults are suppressed (B). Even at positions where reflectors show an local apparent continuity (i.e., a numer-
2

ical continuity, not representing a geological continuity) the discontinuity volume provides a fault indication (C).
3
ht
y rig
op
C

Brouwer and Huck 533