See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/317379480

Enhancement of fault interpretation using multi-attribute analysis and

artificial neural network (ANN) approach: A case study from Taranaki Basin,
New Zealand

Article  in  Exploration Geophysics · August 2017

DOI: 10.1071/EG16072

CITATIONS READS

7 656

2 authors:

Priyadarshi Chinmoy Kumar Animesh Mandal

National Geophysical Research Institute Indian Institute of Technology Kanpur
17 PUBLICATIONS   44 CITATIONS    40 PUBLICATIONS   188 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

PNF Interpretation, North Sea View project

Fault Interpretation in Southern Taranaki basin, New Zeland View project

All content following this page was uploaded by Priyadarshi Chinmoy Kumar on 02 November 2017.

The user has requested enhancement of the downloaded file.

CSIRO PUBLISHING
Exploration Geophysics
http://dx.doi.org/10.1071/EG16072

Enhancement of fault interpretation using multi-attribute analysis

and artificial neural network (ANN) approach: a case study
from Taranaki Basin, New Zealand

Priyadarshi Chinmoy Kumar1 Animesh Mandal2,3

1
CSIR-National Geophysical Research Institute, Hyderabad 500007, India.
2
Department of Earth Sciences, Indian Institute of Technology, Kanpur 208016, India.
3
Corresponding author. Email: [email protected]

Abstract. Enhanced seismic data conditioning and multi-attribute analysis through non-linear neural processing workflows
has been applied to 3D seismic data over 215.10 km2 area of the Opunake prospect located in the south-eastern offshore
Taranaki Basin. The present work aims to delineate faults and the related detail of structural features from the study area.
Post-stack seismic data conditioning techniques such as dip-steering and structural filtering are applied to enhance the
lateral continuity of seismic events and eliminate random noises from the data with the objective of improving the visibility of
faults in the data volume. The conditioned data is then used to extract several attributes, such as similarity, dip variance,
curvature, energy and frequency, that act as potential contributors for enhancing the fault visibility. A fully connected
multilayer perceptron (MLP) network is developed to choose the proper combination of attributes for fault detection.
These seismic attributes (known as the test datasets) are then trained at identified fault and non-fault locations using this
network. The network comprises of 11, 7, and 2 nodes in the input layer, hidden layer and output layer, respectively. The
neural training resulted in an overall minimum root mean square (RMS) misfit and misclassification (%) ranging from 0.54
to 0.67 and 18.67 to 10.42, respectively, between the trained and the test datasets. The neural training generates a fault
probability attribute that produces an improved fault visibility capturing the minute details of the seismic volume as
compared with the results of individual seismic attribute. Thus, the present work demonstrates an enhanced and robust
workflow of fault prediction and visualisation for detail structural interpretation from 3D seismic data volume.

Key words: 3D visualisation, attributes, fault probability cube, interpretation, neural network.

Received 25 June 2016, accepted 17 May 2017, published online 3 August 2017

Introduction control the sealing mechanism across the geological structures,

Seismic attribute analysis contributes significantly in the success identification of plausible trapping zones for hydrocarbon
of 3D seismic interpretation workflows in the petroleum industry. bearing sediments, and indication of structural arrangement
Its analysis brings out an enhanced interpretation of the of subsurface reservoir. This kind of analysis also helps to
subsurface geology and reservoir property by defining their estimate the storage and migration of hydrocarbon resources
structural and depositional environment. Thus, deciphering in the subsurface. However, identification of these structural
detailed structural features hidden within the data by the use of complexities from the subsurface and their seamless interpretation
seismic attributes is an important approach for understanding is an important and challenging task for seismic interpreters.
the subsurface structural configuration, as well as monitoring The advent of 3D seismic visualisation and advanced attribute
the static and dynamic behaviour of the reservoir. Over the past processing techniques have helped the interpreters to take up
two to three decades several seismic attributes have been defined these challenges and contribute towards successful structural
for imaging structural aspects from seismic data. Some of these interpretation from seismic data.
include, seismic coherency (Bahorich and Farmer, 1995; Marfurt However, interpreters very often encounter difficulties in
et al., 1998, 1999), similarity (Tingdahl, 1999; Tingdahl et al., imaging geologic structures from the data. It is due to the fact
2001; Tingdahl and de Rooij, 2005), dip and azimuth (Luo et al., that these features are characterised by discontinuous seismic
2002; Chopra and Marfurt, 2007; Kumar, 2016), volumetric reflections and are most commonly obscured by seismic noises,
curvature (Roberts, 2001; Al-Dossary and Marfurt, 2006) and e.g. dispersion effects, diffractions etc. Thus, a robust data
other attributes (Chopra and Marfurt, 2007). These attributes conditioning technique and attribute analysis workflow is
mostly define the geometrical and structural characteristics, e.g. needed to bring out an enhanced image of these complex
shape, angle and continuity, of the seismic events. Thus, they structural features by suppressing the noise level. In the
deliver a substantial contribution for interpreting geological context of attribute analysis, proper selection of a set of
structures, such as, faults, folds, fracture networks from seismic structural attributes plays significant role as no single attribute
data. Imaging and analysing these structural complexities of the effectively images these structural complexities. Thus, a set of
subsurface provides clues for understanding several key facts logically chosen attributes is needed to generate a better and
related to reservoir geology, e.g. position of juxtaposed beds that more accurate image of complex geological structures. This

Journal compilation Ó ASEG 2017 www.publish.csiro.au/journals/eg

B Exploration Geophysics P. C. Kumar and A. Mandal

approach has been documented by several authors (Russell et al., of the Southern Taranaki Basin (STB) based on the results of
1997; Meldahl et al., 1999; Meldahl and Heggland, 2001; fault analysis studies and estimation of fault displacement
Aminzadeh and de Groot, 2004; Tingdahl and de Rooij, 2005; histories. They also demonstrate the presence of fault networks
Connolly and Garcia, 2012; Brothers et al., 2014; Connolly, (that includes Cape Egmont Fault and its splays) from 3D seismic
2015; Kluesner and Brothers, 2016) and has proven to be a very surveys and have mapped them using seismic coherency cubes.
successful and robust scheme for imaging faults, gas chimneys The existing literature (Cohen et al., 2006; Giba et al., 2010;
and stratigraphic features. Reilly et al., 2015) describes the nature and style of faulting,
Our study area (Figure 1) lies in the south-eastern part of fault displacement models in the southern part of the Taranaki
offshore Taranaki Basin that is characterised by complex fault Basin. However, fault interpretation techniques and their accurate
systems. The study area is cut by the famous Cape Egmont Fault imaging workflows are not well documented. The prevailing
Zone (CEFZ) and its splay. This fault system is a major structural literature talks about the use of high resolution seismic data for
boundary between the tectonically stable Western Platform and interpreting such an extremely complex terrain but none of
the tectonically mobile Taranaki Graben (Nodder, 1993). Seismic them have focused on developing robust seismic interpretation
imaging of these complex fault systems in such a highly deformed procedures for drawing out an enhanced image of the subsurface
basin has always been a challenging task for the interpreters. geological structures of the basin. Therefore, the present work
Nodder (1993) talks about the Cape Egmont Fault (CEF) systems focuses on this issue of providing a detail and robust workflow
and their imaging using high resolution 2D seismic data. for structural interpretation from 3D seismic data volume. To
Recently, Reilly et al. (2015) describes the tectonic evolution address the above issue we make use of data conditioning

172°E 173°E 174°E 175°E 176°E

36°S 36°S

6000

37°S 37°S

4500

3000
NORTHERN UNIT

38°S 38°S

1500
(Topography-Bathymetry (m))

39°S 39°S

–1500
WESTERN UNIT

OPUNAKE 3D

–3000

40°S 40°S

Scale (km)
0 50 –4500
SOUTHERN UNIT

–6000

41°S 41°S
172°E 173°E 174°E 175°E 176°E

Fig. 1. Location map of the study area – Opunake 3D seismic block (marked by a black rectangle). It covers an area of
215.10 km2 and lies in the southern part of the Taranaki Basin.
Enhanced fault interpretation Exploration Geophysics C

techniques that optimally conditions the seismic data and of the offshore Taranaki Basin has been documented by several
brings out sharp and enhanced fault signatures by suppressing authors, e.g. Pilaar and Wakefield (1978), Knox (1982), Schmidt
unwanted information that masks these geological structures. and Robinson (1989), King and Thrasher (1992), Bennett et al.
These conditioned datasets are then used for extracting different (1992), Palmer and Andrews (1993), Nodder (1993), Holt and
structural attributes that defines the geometric behaviour of Stern (1994), King and Thrasher (1996), Giba et al. (2010, 2012,
these structures. Finally, a meta attribute known as the fault 2013) and others. Tectonically it can be classified into four
probability attribute (the attribute that defines all possible fault principal structural units (Figure 2): (1) the tectonically stable
locations from the data volume) has been defined by combining Western Platform; (2) tectonically mobile Eastern Belt (Bennett
these multiple attributes that optimally map the structural et al., 1992) or the Taranaki Graben that is separated by the Turi
details. We use this meta seismic attribute for predicting faults Fault Zone into north (extensional) and south (contractional)
from seismic data and the results are presented through several components (Nodder, 1993); (3) the Southern Zone,
display techniques. characterised by reverse faulting and structural inversion
This study provides/prescribes an optimal workflow for (Knox, 1982); (4) the early–mid Miocene Tarata Thrust Zone
interpreting complex subsurface structures and also demonstrates which runs parallel to the eastern margin of the basin and the
the importance of the fault probability meta seismic attribute Taranaki Fault – an important Miocene basement structure
(Tingdahl and de Rooij, 2005) in preparing an enhanced (Nodder, 1993). All of these structural elements within the
structural image of the subsurface. basin have been developed due to continual tectonic events
and clockwise movement of the plate boundaries (Walcott,
1987; Palmer and Bulte, 1991; King and Thrasher, 1992).
Geologic setting and data The period from Late Cretaceous to Paleocene witnessed the
The Taranaki Basin, a major Cretaceous–Cenozoic sedimentary development of large amount of normal faults exhibiting N–S to
basin is situated along the western flank of the New Zealand NE–SW trend within the basin (Reilly et al., 2015). Several
subcontinent. The basin exhibits an extension from the Taranaki half graben structures also developed during this period, which
Fault Zone along eastern margin and moves beyond the could clutch sedimentary rocks of ~4 km thick (Thrasher, 1990;
continental shelf-break including the onshore Taranaki Peninsula Palmer and Bulte, 1991; King and Thrasher, 1996). However, the
(Smith et al., 1989; Palmer and Bulte, 1991; King and Thrasher, basin continued to remain tectonically dormant during Early to
1992) (Figure 2). It is well known for its enormous hydrocarbon Mid-Eocene period. Shortening of the basin commenced during
potential with complex subsurface geology (Nodder, 1993) and the Late Oligocene and continued up to the Late Miocene (King
is a storehouse of thick sedimentary sequence (up to ~9 km of and Thrasher, 1992; Holt and Stern, 1994; Stagpoole and Nicol,
thick sedimentary deposit) (Reilly et al., 2015), exhibiting an 2008). The shortening activity led to the development of
overall Cretaceous to Recent sedimentary record throughout extensional fault systems in the northern and central parts
the basin. Since our study area lies in the offshore part of the along with the development of several Plio-Pleistocene normal
basin, we present here an overview of the tectonic aspect of the faults within the basin (King and Thrasher, 1996; Nicol et al.,
offshore Taranaki Basin. The structural arrangement (Figure 2) 2005; Giba et al., 2010; Mouslopoulou et al., 2012).

172°E 173°E 174°E 175°E 176°E

Scale (km)
0 50
39°S 39°S

North Wanganui
Basin
WESTERN UNIT
lt
au
tF
on

South Wanganui
m

Fault
Eg

Basin
pe
ult
Ca

Manaia
Fa

40°S 40°S
gi
an
ur

lt

Taranaki Fault
u
h

Fa

lt
Ka

Fau
ri
aa

ate
M

SOUTHERN UNIT
tum
Survil

Mo
le Fau
lt

41°S 41°S
172°E 173°E 174°E 175°E 176°E
Legend
3D seismic block Normal fault
Reverse fault

Fig. 2. Geo-tectonic map of Taranaki Basin, showing different structural features of the basin.
The seismic block (present study area) is indicated by a black rectangle (modified after Reilly
et al., 2015).
D Exploration Geophysics P. C. Kumar and A. Mandal

The seismic data of the present study area is a stacked data along the spatial direction of the geological objects of
migrated seismic volume covering an area of 215.10 km2 interest (Meldahl et al., 2001). The technique provides a
acquired in a 6.25 m
25 m bin grid with 4.0 ms sampling consistent means of detecting, separating and identifying
interval. The total record length of the seismic data is 6.0 s geological features, like, gas chimney, fault layers etc., which
with 60 fold coverage and data polarity is SEG normal varies throughout the data volume (Meldahl et al., 2001). Thus,
(increase in acoustic impedance boundary is reflected as peak adhering to these principles, a set of multiple directive seismic
or positive amplitude). attributes (i.e. the attributes that are extracted along the dip
and azimuth of the seismic events) are computed and then
Methodology recombined through a non-linear multilayered perceptron
(MLP) network (Figure 5) to generate a new attribute that
The methodology adopted for this study has been presented by
optimally enhances the fault detection from the seismic data.
a workflow as given in Figure 3. It begins with the seismic data
The present study has been carried out by using the neural network
conditioning technique which is carried out with an objective of
and dip-steering modules of OpendTect v5.0.11 software.
enhancing structural objects, e.g. faults from the seismic data by
suppressing noise that hinders their effective recognition. For Conditioning seismic data volume
this, the conditioning is done in three different phases. In the first
phase, steered seismic data volumes (Tingdahl, 1999; Tingdahl Effective recognition of faults from seismic data is a crucial
et al., 2001) are prepared using original seismic data (Figure 4a). task for seismic interpreters and forms the backbone for
In the second phase, a median statistical filter known as dip- the structural interpretation workflow. However, noises like
steered median filter (DSMF) is applied over the original data dispersion, scattering effects, diffractions etc., masks the useful
using the pre-processed steering cube. This helps in removing the reflections in the data that corresponds to the subsurface
random noise from the data and maintains the lateral continuity of geology. Due to these reasons manual as well as automatic
the seismic events (Figure 4b). The first output at this stage is a interpretations of geological structures always becomes a
DSMF seismic data volume (Figure 4b). An intermediate filter problematic issue. Thus, to overcome such issues, it is always
called as dip-steered diffusion filter (DSDF) is then applied to necessary to routinely preprocess the seismic data and optimally
enhance the sharpness of the geological features. The second condition it for a better interpretation of the subsurface geology.
output at this stage is a DSDF data volume. Thus, in the third or Here we explain the data conditioning techniques used for this
final phase, these seismic volumes (DSMF and DSDF) are study, which helps in increasing the signal-to-noise ratio and
combined by a logical expression to generate seismic data enhances the structural visibility of geological features from
through the fault enhancement filter (FEF). The FEF data the data.
(Figure 4c) exhibit a significant improvement in delineating
fault features from the original seismic data by enhancing their Dip extraction and creating steering cube
sharpness and reducing the random noise that mask these The workflow at this stage begins with a technique known as
features. This conditioned data volume (FEF seismic) is then dip-steering proposed by Tingdahl (1999) and Tingdahl et al.
taken as an input for attribute extraction. The attribute extraction (2001). The technique uses local dip and azimuth of the seismic
process used in our study honours the philosophy of seismic events to track the event locally with respect to the trace segments
object detection technique proposed by Meldahl et al. (1999), under investigation. When the process is carried over the entire
and Meldahl and Heggland (2001). The main objective of this seismic data volume (i.e. the original seismic data) a dip-azimuth
technique is to extract seismic information from the acquired volume is obtained that contains the local dips of the seismic

INPUT SEISMIC VOLUME

Phase-based dip algorithm

DETAILED STEERING CUBE BACKGROUND STEERING CUBE

Coarser filtering

DSMF SEISMIC SIMILARITY DSDF SEISMIC

Attributes
FEF SEISMIC ENERGY
Attributes
FREQUENCY
STEERED SIMILARITY NON-STEERED
DIP VARIANCE SIMILARITY
CURVATURE

ANN FAULT DELINEATION

Fig. 3. Workflow adopted for the study. In the first phase the data is conditioned to obtain the FEF data. In the
second phase, this data is used for attribute extraction. Then the extracted attributes undergo artificial neural
processing through a non-linear neural model to image the faults from seismic data volume.
Enhanced fault interpretation Exploration Geophysics E

XLINE XLINE
(a) S 3000 4000 5000 6000 N (b) S 3000 4000 5000 6000 N
0 0

0.25 0.25
TWT (s)

TWT (s)
0.50 0.50

0.75 0.75

1 km 1 km

1.0 1.0

Positive Positive
Amplitude

Amplitude
Negative Negative

XLINE
(c)
S 3000 4000 5000 6000 N
0

0.25
TWT (s)

0.50

0.75

1 km

1.0

Positive
Amplitude

Negative

Fig. 4. (a) Original seismic data. The seismic section is displayed up to 1 s length and belongs to inline 2369. Note that random noises have masked the
fault features. (b) DSMF seismic data. The seismic section belongs to the same inline 2369. Note that random noises have been reduced and the fault features
have enhanced their visibility as compared to the earlier in Figure 4a. (c) FEF seismic data. The seismic section belongs to the same inline 2369. Note that the
fault zones have been sharpened as compared to that in Figure 4a, b. The sharpness is indicated by blue arrows.

reflectors and associated discontinuities. The resulting volume steering cube is called the detailed steering cube and helps in
is called a steering cube (Tingdahl, 1999; Tingdahl et al., 2001; defining the local dips of the seismic reflectors. However, for
Tingdahl and de Groot, 2003; Jaglan et al., 2015; Qayyum et al., defining the overall trend, the detailed steering cube is taken as an
2015). The computation of the steering cube forms the heart of input and coarser filtering is applied using a filtering step out
the workflow. The steering cube is computed using the phase- (inline : crossline : sample, 5
5
5). The resulting steering cube
based dip algorithm. Then the steering cube is filtered using after coarser filtering is now called the background steering cube
a median filter both locally and regionally to check the data and contains the overall dip trends of the seismic reflectors and
quality and study the improvements in the data for either of the outlines the background structural information. This steering
two cases. A mild filtering step out (inline : crossline : sample, cube is taken as an input for data enhancement through a
1
1
3) is used to understand the local variations. The resulting process of structural filtering.
F Exploration Geophysics P. C. Kumar and A. Mandal

Attribute 1 Network connections

Attribute 2

DECISIONAL OUTPUTS
Attribute 3
INPUTS

Yes

No

Attribute n

INPUT LAYER HIDDEN LAYER OUTPUT LAYER

Fig. 5. Multilayer perceptron network. The network is a non-linear model consisting of three
different layers: (1) the input layer, (2) the hidden layer and (3) the output layer. The colour codes
(white through grey to black) used here to represent the synaptic weights given to each input. The
white colour indicates the least weight and the black colour indicates the highest weight. Finally,
a decisional output is obtained in the form of ‘yes’ or ‘no’.

Structural filtering direction of the fault plane and thereby enhances the fault
In this study, structural filtering is performed by using a structure- positions and makes the fault visibility sharper. This operation
oriented filter (SOF). The key objective behind this filtering resulted in DSDF seismic data volume.
technique is to differentiate between the dip-azimuth of the Finally, both of these data volumes are logically combined
seismic reflectors and the overlying noise (Chopra and with the help of a computed similarity attribute volume (discussed
Marfurt, 2007). This process helps in removing random noises later) for obtaining a FEF that can perform the edge preserving
from seismic data and enhances the lateral continuity of smoothing operation. For logically combining these two data
the seismic events (Hocker and Fehmers, 2002). Thus, it volumes a cut-off value 0.5 is used, such that the resulting
ultimately helps to provide sharp definition of the geological enhanced seismic data volume would be able to showcase the
structures. The filter operates on three basic principles: (1) seismic reflectors in a prominent way and structures like faults
analysing the orientation of the seismic reflectors, (2) detecting would be enhanced indicating their sharper signature (Jaglan
the edges or reflection terminations, and (3) preserving the edges et al., 2015; Qayyum et al., 2015). The obtained FEF seismic
at these terminations, thereby smoothing the reflectors. Based on data (Figure 4c) volume (a conditioned seismic data) is
these principles, we have filtered the seismic volume in three thereafter used as an input for attribute extraction.
different stages and this has resulted in three different conditioned
seismic volumes. These filtering stages are (1) smoothing the Attribute extraction
seismic reflectors through dip-steered median filtering (DSMF), Structural or geometric attributes plays a significant role in
(2) enhancing the fault positions through dip-steered diffusion imaging geologic structures from seismic data. We here
filtering (DSDF), and (3) logically merging the first two steps for present a brief review of a set of geometric attributes that have
producing FEF seismic data. been used for this study. We have also discussed about their
DSMF is a statistical filter applied on the seismic data volume extraction process.
using a pre-processed steering cube for obtaining a smoothed
seismic volume where the continuity of the seismic reflections is
Similarity
improved by suppressing background random noises (Jaglan
et al., 2015; Qayyum et al., 2015). The filter applies median Similarity quantifies the similar nature of two segments of the
statistics over seismic amplitude following the seismic dips. Here seismic traces u (x, y, t)(Tingdahl, 2003). If we consider the
this operation is performed using a 3
3 median filtering step-out samples of trace segments to be the coordinates of vectors in
and the DSMF seismic data volume is obtained (Figure 4b). In the hyperspace, then the similarity is defined as the Euclidean
next step, we move forward for improving the reflections closer distance between the vectors, normalised over the vector
to fault zones. The seismic data generally possess a diffusing lengths (Tingdahl, 2003; Tingdahl and de Groot, 2003;
character closer to these zones. To enhance the visibility of these Tingdahl and de Rooij, 2005). Thus, if (S) is the similarity
zones and to obtain a sharp definition of geologic structures, an between two trace segments at (xA, yA) and (xB, yB), centred at
intermediate filter DSDF is used. The filter evaluates the quality of the time t, then as per the definition:
the seismic data in a dip-steered circle. The central amplitude is ja  bj
replaced by the amplitude where the quality is deemed best S ¼1 ð1Þ
jaj þ jbj
(Jaglan et al., 2015; Qayyum et al., 2015). As a result, good
quality of seismic data shifts from either side of fault in the where
Enhanced fault interpretation Exploration Geophysics G

2 3 2 3
uðxA ; yA ; t þ t1 Þ uðxB ; yB ; t þ t1 Þ The dip-steered similarity is extracted using vertical time gates
6 uðxA ; yA ; t þ t1 þ dtÞ 7 6 uðxB ; yB ; t þ t1 þ dtÞ 7 like 24 ms (short window), 48 ms (mid window) and 64 ms (long
6 7 6 7
6 ... 7 6 ... 7 window). Different window lengths help in capturing the effect
a¼6 7; b ¼ 6 7
6 7 6 7 of fault signatures throughout the data length. Apart from
4 uðxA ; yA ; t þ t2  dtÞ 5 4 uðxB ; yB ; t þ t2  dtÞ 5
dip-steered similarity we have also extracted the non-steered
uðxA ; yA ; t þ t2 Þ uðxB ; yB ; t þ t2 Þ similarity (the similarity attribute that is computed without
dt is the sampling interval, t1 is the relative start time of the taking steering data) to check the effectiveness of steering that
comparison window, t2 is the relative stop-time of the comparison helps in best imaging the fault signatures.
window and u is the amplitude value in the seismic data cube.
The similarity outputs result in values ranging from zero to Dip variance
one, where the values close to 0 are represented as low similarity
events and the values close to 1 are indicated as high similarity Faults are very often characterised by significant changes in
events. When the similarity attribute is computed using dip- dip. For capturing this variation, statistical variance in the dips
steering seismic (Tingdahl, 2003) as an input, the output is can be computed and analysed. This is called as dip variance and
called dip-steered similarity. The dip-steered similarity is is expressed as (Tingdahl, 2003):
1 X X
expressed as (Tingdahl, 2003): xs ys X t2
varðpx Þ ¼ ðpx ðx þ a; y þ b; t þ tÞ  px Þ2
jadip  bdip j n  1 b¼x a¼ys t¼t1
Sdip ¼ 1  ð2Þ s
jadip j þ jbdip j
ð3Þ
where
where
!
2 3 2 3 X
xs X
ys X
t2
uðxA ; yA ; tA þ t1 Þ uðxB ; yB ; tB þ t1 Þ px ¼ px ðx þ a; y þ b; t þ tÞ =n ð4Þ
6 uðxA ; yA ; tA þ t1 þ dtÞ 7 6 uðxB ; yB ; tB þ t1 þ dtÞ 7 b¼xs a¼ys t¼t1
6 7 6 7
6 ... 7 6 ... 7
adip ¼ 6 7; bdip ¼ 6 7 and t1 is the relative start time of the sub-cube, t2 is the relative
6 7 6 7
4 uðxA ; yA ; tA þ t2  dtÞ 5 4 uðxB ; yB ; tB þ t2  dtÞ 5 stop-time, n is the total number of samples in the sub-cube, xs
uðxA ; yA ; tA þ t2 Þ uðxB ; yB ; tB þ t2 Þ and ys are the lateral step-out values from the centre of the small
sub-cube in x and y directions, respectively, and p is the dip
tAand tB are the dip-steered times going from the position (x, y, t) cube. The dip variance is estimated taking dip angle attribute
to the traces at (xA, yA) and (xB, yB), respectively, and u is the and steering seismic as inputs. The above equation outputs the
amplitude value in the seismic data cube. dip variance in the x direction. Similarly, the dip variance in the y

INLINE
(a) 2600 2400 2200 (b)
N N

E E

6000

5000
XLINE

4000

3000

1 km
1 km

0 0.25 0.5 0.75 1 Negative Positive 0 0.25 0.5 0.75 1

Similarity Amplitude Similarity

Fig. 6. (a) Steered similarity attribute. The computed attributed is displayed for time slice, t = 0.5 s. Low
similarity values close to 0 are observed along the fault zones indicated by green arrows. (b) Steered similarity
attribute co-rendered with FEF seismic data. The attribute is co-rendered by reducing 30% transparency to
obtain the correlation with the seismic data. The attribute correlation is indicated by green arrows.
H Exploration Geophysics P. C. Kumar and A. Mandal

direction var (py) can be calculated. The overall dip variance is amplitude and changes in the frequency content of the
then calculated by the average of var (px) and var (py) which is seismic data. Hence to capture these effects, attributes like
expressed as: energy and frequency are also extracted and used for the
analysis.
varðpx Þ þ varðpy Þ
pvar ¼ ð5Þ
2 Energy
The dip variance refers to this average value (a statistical The energy attribute at (x, y, t) is defined as the sum of the squared
property) which helps to identify the fault locations. Fault amplitudes within a given window (Tingdahl, 2003). It is
zones are also characterised by low energy, loss in signal expressed as:

INLINE INLINE
(a) 2600 2400 2200 (b) 2600 2400 2200
N
N
E E

6000 6000

5000 5000
XLINE

XLINE

4000 4000

3000 3000

1 km 1 km

0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1

Similarity Similarity

Fig. 7. (a) Steered similarity attribute displayed for the time slice t = 0.5 s. A more individuality, sharpness and
enhancement of the faults are observed and are indicated using green arrows. The splay of the Cape Egmont Fault
Zone (CEFZ) (indicated by the orange arrow) are indicated using blue oval yellow arrows. (b) Non-steered similarity
attribute displayed for same slice t = 0.5 s that exactly cuts these structural features. Though faults are imaged but their
sharpness is not clearly and distinctively revealed. This comparison indicates the robustness and competence of using a
steering data for imaging the structural features of the subsurface.

(a) INLINE
2600 2400 2200 N
E (b) (c)
6000 X Y U V
0 0

5000
0.25 0.25
XLINE

TWT (s)
TWT (s)

4000 0.50 0.50

3000 0.75 0.75

1.0 1.0
0 0.25 0.5 0.75 1 Fault Fault
Similarity

Fig. 8. (a) Steered similarity attribute and random lines. The result is displayed for time slice t = 0.5 s that exactly cuts these structural features. To correlate
the fault signatures with the seismic data, random lines XY and UV perpendicular to the strike of the faults are prepared. (b) Random line XY is taken
perpendicular to the splay of CEFZ. Note the sharpness of these faults. (c) Similarly, random line UV is drawn perpendicular to the Cape Egmont Fault to image
and correlate it from the seismic data. The identified faults are marked as black lines in the random seismic sections.
Enhanced fault interpretation Exploration Geophysics I

X
t2
the responses of frequency loss due to the presence of faults in
E¼ uðx; y; t þ tÞ2 ð6Þ the data. The AFS is expressed as (Tingdahl and de Rooij, 2005):
t¼t1 !
XN
2 2
XN
where t1 is the relative start time of the window and t2 is the AFS ¼ pðoi Þ oi = pðoi Þ2 ð7Þ
relative stop-time of the energy window. The energy attributes i¼1 i¼1
are extracted using vertical time gates like 32 ms (short window)
and 80 ms (long window), such that their effect could be where p (o) is the complex response of the trace segments in the
analysed along the entire data length. Fourier domain and N is the length of the signal. The peak shape of
the response of the trace segment is characterised by FSF and is
Frequency expressed as:
 XN   XN 
Average frequency squared (AFS) and frequency slope fall-off FSF ¼ 1 þ pmax  2 2
pðoi Þ = pmax þ2
pðoi Þ2
(FSF) attributes are types of frequency attributes that are i¼1 i¼1
optimum choice (Tingdahl and de Rooij, 2005) for capturing ð8Þ

XLINE
(a)
S 3000 4000 5000 6000 N
0

0.25
TWT (s)

0.50

0.75

1 km

1.0

0 0.0005 0.001 0.0015 0.002

Dip variance
(b)
S N
0

0.2

0.4
TWT (s)

0.6

0.8

1 km
1.0

Negative 0 0.0005 0.001 0.0015 0.002

Positive
Amplitude Dip variance

Fig. 9. (a) Dip variance attribute. The computed attributed is displayed for inline 2369. Note that high
variance in the dip is observed along the fault as we move from the south to north. The colour scale use to display
the dip variance attribute is in pastel, where red indicates high variance in the dip. (b) Dip variance attribute co-
rendered with seismic amplitude. The co-rendering is done for the same seismic section by reducing 25%
transparency in the dip variance attribute. High correlation and significant variation in the dip is observed.
J Exploration Geophysics P. C. Kumar and A. Mandal

where pmax is the highest peak in the frequency domain. The FSF Artificial neural network
value will be close to 1 when there is maximum concentration The concept of artificial neural networks (ANNs) is inspired
of energy at the peak of the trace segment and in other cases the by research that tries to mimic the behaviour of biological
value will be close to 0. neuron (McCulloch and Pitts, 1943). The basic principle that a
Apart from these attributes, other attributes, such as, dip angle, mathematical neuron or ANN follows, is that a weighted sum
curvature and reference time have been taken into consideration of inputs is fed into an activation function which is non-linear
as they also help to efficiently capture the discontinuity effect in nature in order to rescale the sum. A constant bias (node or
from the data. Since attribute extraction is a time consuming threshold) is applied in this process to maintain and control the
process, all of the attributes are first extracted and tested for connection between the input and the processor. Finally, a
several inlines and crosslines. Then the results are quality decisional output is expected from this operation. In the entire
checked and finally they are computed over the entire seismic process, the weights assigned to each input units plays a pivotal
data volume to obtain volumetric seismic attributes. role, as the stronger the weight is, the better would be the

XLINE
S 3000 4000 5000 6000 N
0

0.25
TWT (s)

0.50

0.75

1 km

1.0

Low High
Energy

Fig. 10. Energy attribute. The computed attributed is displayed for inline 2369. Note that there is sudden
change in the energy from high to low at the fault zones. These are indicated by black arrows.

XLINE
S 3000 4000 5000 6000 N
0

0.50
TWT (s)

0.10

0.15

1 km

0.20

Fault Yes Fault No

Fig. 11. Fault location selection for neural training. The location of fault presence is indicated as ‘fault-yes’
(green coloured picks) and the absence of fault is indicated as ‘fault-no’ (red coloured picks).
Enhanced fault interpretation Exploration Geophysics K

connection between the input and the processor. The decisional finally establishes a relationship between the input sets (seismic
output obtained from the network is critically examined. If the attributes) and the output sets (fault-yes and fault-no). The
response output is found to be different than that of the expected training is carried out iteratively and subsequently weights
output, the internal parameters like inputs, weights (or known as are adjusted to obtain a minimum error between the predicted
synaptic weights) etc., are rechecked and rescaled to establish a network result and the known output. The training results are
better match between the obtained and targeted outputs. In this quality checked over few key seismic lines. Once a satisfied
study, a non-linear neuron model called the MLP network is used. output is obtained the NN is applied over the entire seismic
This network system consists of three different layers: an input volume. This results with probability attribute called as the fault
layer, a hidden layer and an output layer (Figure 5). The input probability attribute volume. This attribute volume contains
data fed into the network system flows between these layers and values ranging from 0 to 1, where the values corresponding
performs the operation. to 1 indicates high probability of fault and those close to
For carrying out the neural network (NN) operation, the test 0 corresponds to lower fault probability. This fault attribute
datasets (or input sets), i.e. the seismic attributes (as discussed volume is viewed over seismic sections and volumetric
earlier) are prepared first. These seismic attributes are selected displays. The results of the fault probability attribute are
based on the fact of their ability to capture the geometric finally validated with the existing geology and documented
characteristics of geological structures from the seismic data. literature (if any) of the area to draw out enhanced structural
The attributes are initially tested along few key seismic lines interpretation.
to demonstrate their ability in demarcating known faults
more clearly. Once, a satisfactory result is achieved, then the
attribute sets are chosen as an input for the NN process. After Results
preparing the test datasets, training datasets are selected. These The analysis of non-linear multi-attribute volumes brought
are the example or pickset locations corresponding to the fault- about a momentous improvement in the fault signatures
yes and fault-no positions picked by the interpreter from observed throughout the seismic data volume. It is observed
the seismic data volume. This helps the NN to get guidance that FEF seismic data (Figure 4c) prepared through the seismic
for discriminating the character found in the input sets. Fault- data conditioning technique has sharpened the fault zones
yes positions are the most obvious positions characterised throughout the data by removing unwanted seismic noise that
by low similarity, variable dip, low energy and frequency masked these geological structures. Seismic attributes derived
loss properties. All other locations are grouped into fault-no by using the FEF data as an input, resolve the fault signatures
positions. Around 500 training locations per seismic line are and create a pathway for enhanced interpretation of structural
prepared from the seismic data volume that covers an area of features. The similarity attribute (Figure 6a) exposes the
215 km2. The NN starts to learn from these example locations, and discontinuous nature of faults. These features are associated

Increasing order

Energy

Least Highest
Similarity contribution contribution

Steered-similarity

Cube-similarity

Dip variance
Fault yes
Dip angle

Reference time

Fault no
Frequency slope fall

Average frequency squared

Most positive curvature

Most negative curvature

Input layer; 11 nodes Hidden layer; 7 nodes Output layer; 2 nodes

Fig. 12. Neural network topology used for multi-attribute processing. The colour scale white through grey to black
specifies the relative contribution or weights of each input node to the classification result. The black ones provide highest
contribution in training followed by deep grey. Description of the input attributes (as shown in the left) is given in the text.
The network consists of 11 nodes in the input layer, 7 nodes in the hidden layer and 2 nodes in the output layer. The
predicted output (as shown in the right) indicates the presence (fault-yes) and the absence (fault-no) of faults respectively.
L Exploration Geophysics P. C. Kumar and A. Mandal

with low similarity values. To test the efficacy of this similarity (a)
attribute, the attribute was co-rendered (Figure 6b) with the 1.00
seismic amplitude volume. The fault signatures observed from
the amplitude volume and attribute volume exhibit good
correlation. The similarity attribute was also analysed by 0.75
visualising it over the time slice t = 0.5 s (Figure 7a). This

Normalised RMS
attribute clearly brought out the structural trend of the faults as
south-west to north-east which is in conformation with the
general trend of the CEF. These fault systems are the splay of 0.50
the famous Cape Egmont Fault that cuts our study area. Again,
to test the efficacy of steering, the non-steered similarity
attribute (Figure 7b) was analysed and its results were
0.25
compared with that of steered similarity (Figure 7a). We
observed that the sharpness and clarity of the faults are much
more pronounced with the similarity attribute that uses steering
data as an input. The non-steered similarity attribute could image 0
the fault signatures, however, enhancement of fault imaging
could not be achieved. To critically examine and visualise (b)
the nature of faulting, random seismic sections were prepared
perpendicular to the strike of these structural features
(Figure 8a–c). This revealed the typical normal faulting nature 40
of the geological beds within the prospect. From the dip variance
attribute, it is observed that the central part of the faults is Misclassification (%)
characterised by high dip as compared to that of the
surroundings (Figure 9a). The above facts are clearly
visualised by displaying the dip variance attribute in vertical
20
sections and co-rendering (Figure 9b) them with the original
seismic data of the prospect area. The energy response (Figure 10)
observed near the fault locations in the seismic data is very
low due to the fact that these features are characterised
by amplitude distortions. When the energy attribute was co-
rendered with the seismic data, very low energy (Figure 10) is 0
apparent near the fault zones. Train data Test data
The neural training of these attributes with fault and non-fault
locations (Figure 11) resulted in an enhanced fault probability Fig. 13. (a) Normalised RMS misfit. The neural training resulted in an
overall minimum RMS misfit ranging from ~0.54 to 0.67 for both the trained
attribute. The non-linear MLP network (Figure 12) for the
and test datasets. Training is stopped when both the curves run parallel to
present work comprises of 20 fully connected nodes; 11, 7,
each other. (b) Misclassification percentage (%). Minimum misclassification
and 2 nodes of which are associated with the input, hidden value (%) obtained for both the trained and test dataset ranges between 18.67
and output layers, respectively. The network performed its to 10.42%.
operation by randomly splitting the data into test and train
sets. 30% of the data was used for testing and 70% of the data (Figure 15). The fault probability cube was individually co-
was used for training. The neural training was carried out rendered with the seismic amplitude volume attribute to observe
iteratively. This updates the connection weights as well as the correlation of fault signatures from the fault cube with that of
overcoming problems related to local minima. It is observed the fault signatures from the amplitude cube (Figure 16a–c).
that dip variance, energy and similarity as compared to other
attributes delivered maximum contribution for training the
network. We also observed that after 19 iterations, the overall Discussion
root mean square (RMS) misfits for both trained and tested data Nodder (1993) documented the presence of the CEFZ normal
attain a minimum value between 0.54 to 0.67 (Figure 13a). The fault from 2D seismic profiles and claims it as the most active
minimum misclassification value (%) obtained for both the zone of deformation in the offshore Taranaki Basin. Cohen et al.
trained and test datasets varies from 18.67 to 10.42% (2006) shows the presence of the CEFZ and its splay from the
(Figure 13b). The neural training finally resulted in the fault interpretation of 3D seismic data of the Opunake prospect.
probability attribute that was then made to process over the Giba et al. (2010) talks about the presence of normal faults
entire seismic volume to obtain the fault probability attribute in the STB. These are the extensional features (part of crustal-
volume or the fault cube. The fault probability volume extension), extending ~350 km along the basin. These faults
(Figure 14) is an enhanced fault attribute volume that stores all mainly exhibit NE–SW trends and dip steeply. Reilly et al.
possible fault signatures from the data. This attribute volume (2015) documented that the Cape Egmont Fault is the largest
outputs values between 0 and 1 where, 1 demonstrates higher in the STB, with a vertical displacement of up to ~2500 m.
confidence of fault locations. The cube is analysed by preparing These fault signatures were imaged using only coherency cube
3D display (Figure 15) and overlaying it on the amplitude derived from 3D/2D seismic data. However, the reliability of
seismic section. The presence of these structures has led to the structural delineation using a single seismic attribute may be
compartmentalisation of the subsurface. Several horst-graben debatable. Though all of these workers have talked about the
features are also observed (Figure 15). Some of these faults are fault histories, fault nature and tectonic evolution of the basin,
found to propagate all the way to the sea bed. Several small none of them have yet discussed fault enhancement techniques
parallel faults referred to as step faults are also observed or workflow as a means of improving structural interpretation
Enhanced fault interpretation Exploration Geophysics M

XLINE
S 3000 4000 5000 6000 N
0

0.25
TWT (s)

0.50

0.75

1.0

0 0.25 0.5 0.75 1

Fault probability

Fig. 14. Fault probability attribute. The attribute is displayed for inline 2369. High fault probabilities are
indicated by deep black colour and lower fault probabilities are indicated by white colour. The attribute is
displayed in grey scale. The attribute provides a better visibility of the faults and also enhances their continuity.

0
INLINE
XLINE
0

0.5
0.5
TWT (s)

1.0 1.0

1.5
1.5

1 km 2.0
0 0.25 0.5 0.75 1

N
2.0 Fault probability
Z

Fig. 15. 3D view of fault probability attribute. The attribute is overlain with seismic amplitude section. The
seismic amplitude section is displayed in grey colour scale and the fault probability attribute is displayed in white to
red through black colour scale. The red colour indicates the higher probability of fault locations. Note the faulting
activity resulted in several subsurface compartments in form of horst-graben features. Several small parallel step
faults are also observed. Note that faults appear to propagate to the seabed. This is highlighted by black oval.

for such a complicated geological terrain. Our workflow of steering data as one of the inputs provides a better pathway
for imaging faults from the Opunake prospect not only for performing enhanced structural interpretation. This fact is
concentrates on fault imaging but also orates about the efficacy supported by Figure 7a, b, where we have shown the efficacy of
of using several seismic attributes in enhancing the accuracy of steering and non-steering attributes. The existence of CEF and
mapping these structural features. We have used several related its splay are also observed from the results of the coherency
attributes to define the geometry of these fault signatures. The use attribute showcased by Reilly et al. (2015). The present workflow
N Exploration Geophysics P. C. Kumar and A. Mandal

(a) (b)

XL
XL INE
E INE NE
IN I
INL INL

0.6 0.6
(s)

(s)
TWT

TWT
1.2 1.2

N N
E E
Z Z
Negative Positive 0 0.25 0.5 0.75 1
Amplitude Fault probability

(c)

XL
INE
E
INLIN

0.6
(s)
TWT

1.2
N
E
Z
0 0.25 0.5 0.75 1
Negative Positive
Fault probability
Amplitude

Fig. 16. (a) 3D volumetric display of seismic data volume. Note the fault signatures as indicated by green arrows. (b) 3D volumetric display of fault cube
obtained from multi-attribute neural processing. The red colour indicates the high probability of faults in the seismic data volume. (c) The fault cube is
overlain with the seismic amplitude cube to examine the correlation of faults obtained through neural processing. The fault cube exhibits a significant
visibility of each and every faults present in the seismic data volume.

for generating the similarity attribute demonstrates, however, the interpretation and analysis is simpler. We also suggest that the
efficacy in enhancing the structural features (Figure 7a) in terms use of several structural attributes effectively helps in imaging
of increasing sharpness and structural visibility. This is due to faults by increasing fault interpretation efficiency. Additionally,
the fact that the attributes prepared for such structural imaging, as when these attributes are combined and trained in conjunction
demonstrated in this study, honour the structural orientation of with the interpreter’s assumed fault locations, this produces an
the geological bodies. Moreover, structural conditioning of the efficient way of locating these structures. Thus, we profess that
seismic data (Figure 4c) has enhanced the signal-to-noise ratio such type of analysis is an efficient structural interpretation
and the visibility of geological structures. Different seismic workflow for the study area and is an optimum solution for
attributes prepared using this conditioned data, efficiently any geologically complex and structured terrain.
reveals several geometrical characteristics (e.g. trend, dip,
continuity etc.) of these structures. Attributes also help in
unveiling the impact of these geological structures over the Conclusions
seismic signal (e.g. loss in the frequency content of the signal, The Taranaki Basin is an extremely deformed sedimentary
or loss in signal energy) (Figure 10). The workflow delivered basin with a complex subsurface structural geology. Therefore,
through this case study justifies the importance of using imaging structures from this basin need careful attention.
structural filtering and multi-attribute analysis techniques for Interpretation of such a complex geological terrain needs an
better detailing the subsurface geologic structures, like faults. improved and more robust scheme which can highlight
We recommend that when fault imaging or fault analysis from small changes caused by these structures. These facts are
the seismic data is an objective, robust structural filtering must clearly demonstrated by the work presented in this study. The
be carried out, such that the geological structures are free from major difficulty interpreters usually face during any structural
unwanted information (noises) that usually mask them in the interpretation is due to poor data quality that leads to inaccurate
data. Once the data is pre-conditioned, subsequent steps of fault imaging of the structures masked by unwanted seismic noise.
Enhanced fault interpretation Exploration Geophysics O

The data conditioning scheme that we have presented through Giba, M., Walsh, J. J., and Nicol, A., 2012, Segmentation and growth of an
this case study, clearly unveils the robustness of the workflow obliquely reactivated normal fault: Journal of Structural Geology, 39,
and provides a better solution for enhancing subsurface structural 253–267. doi:10.1016/j.jsg.2012.01.004
features from seismic data. The structural filtering techniques Giba, M., Walsh, J. J., Nicol, A., Mouslopoulou, V., and Seebeck, H., 2013,
Investigation of spatio-temporal relationship between normal faulting
used here contains several filtering steps which generates a filtered
and arc volcanism on million-year timescales: Journal of the Geological
seismic data with enhanced and sharpened fault signatures. The Society, 170, 951–962. doi:10.1144/jgs2012-121
second phase of our workflow talks about the use of multi- Hocker, C., and Fehmers, G., 2002, Fast structural interpretation with
attribute concepts through a non-linear training model. structure-oriented filtering: The Leading Edge, 21, 238–243. doi:10.1190/
Attribute extraction and neural training performed by using 1.1463775
this enhanced data, helps in efficiently capturing the geometric Holt, W. E., and Stern, T. A., 1994, Subduction, platform subsidence and
behaviour of the faults in the form of a fault probability cube. This foreland thrust loading: the late Tertiary development of Taranaki Basin,
probability cube is an optimum attribute that best locates faults New Zealand: Tectonics, 13, 1068–1092. doi:10.1029/94TC00454
from the seismic data. Literature pertaining to our study area Jaglan, H., Qayyum, F., and Huck, H., 2015, Unconventional seismic
discusses faults, their analysis and interpretation, but does not attributes for fracture characterization: First Break, 33, 101–109.
King, P. R., and Thrasher, G. P., 1992, Post-Eocene development of the
discuss the interpretation workflow. The workflow presented here
Taranaki Basin, New Zealand: convergent overprint of a passive margin,
streamlines the interpretation by making it very efficient and in J. S. Watkins, Z. Feng, and K. J. McMillen, eds., Geology and
robust to be applicable to such a multifaceted geological terrain. geophysics of continental margins: American Association of Petroleum
Thus, the results and even the techniques of this study provide a Geologists Memoir 53, 93–118.
pathway for carrying out efficient and accurate future exploration King, P. R., and Thrasher, G. P., 1996, Cretaceous-Cenozoic geology and
activities within the prospect. petroleum systems of the Taranaki Basin, New Zealand: Institute of
Geological and Nuclear Sciences, Lower Hutt (NZ), Monograph 13.
Kluesner, J. W., and Brothers, D. S., 2016, Seismic attribute detection of
Conflicts of interest
faults and fluid pathways within an active strike-slip shear zone: new
The authors declare no conflicts of interest. insights from high-resolution 3D P-CableÔ seismic data along the Hosgri
Fault, offshore California: Interpretation, 4, SB131–SB148. doi:10.1190/
INT-2015-0143.1
Acknowledgements Knox, G. J., 1982, Taranaki Basin, structural style and tectonic setting:
The authors thank dGB Earth ScienceÔ for proving academic licence of New Zealand Journal of Geology and Geophysics, 25, 125–140.
OpendTectÔ software to the Department of Earth Sciences, IIT Kanpur. doi:10.1080/00288306.1982.10421405
New Zealand Petroleum and Minerals, Ministry of Business, Innovation and Kumar, P. C., 2016, Application of geometric attributes for interpreting faults
Employment, New Zealand Government, is highly acknowledged and from seismic data: an example from Taranaki Basin, New Zealand: 86th
thanked for providing seismic datasets for pursuing this academic research. Annual International Meeting, SEG, Expanded Abstracts, 2077–2081.
The authors would also like to express their great appreciation for the Luo, Y., Marhoon, M., Al Dossary, S., and Alfaraj, M., 2002, Edge-preserving
comments of the Associate Editor, and two anonymous reviewers, whose smoothing and applications: The Leading Edge, 21, 136–158. doi:10.1190/
painstaking reviews have led to substantial modification and a drastic 1.1452603
improvement of the manuscript. The first author thanks the Department of Marfurt, K. J., Kirlin, R. L., Farmer, S. L., and Bahorich, M. S., 1998, 3-D
Science and Technology (DST) for awarding him the Inspire Fellowship. seismic attributes using a semblance-based coherency algorithm:
Geophysics, 63, 1150–1165. doi:10.1190/1.1444415
Marfurt, K. J., Sudhaker, V., Gersztenkorn, A., Crawford, K. D., and Nissen,
References
S. E., 1999, Coherency calculations in the presence of structural dip:
Al-Dossary, S., and Marfurt, K. J., 2006, 3-D volumetric multi-spectral Geophysics, 64, 104–111. doi:10.1190/1.1444508
estimates of reflector curvature and rotation: Geophysics, 71, P41–P51. McCulloch, W. S., and Pitts, W., 1943, A logical calculus of the ideas
doi:10.1190/1.2242449 immanent in nervous activity: The Bulletin of Mathematical Biophysics,
Aminzadeh, F., and de Groot, P. F., 2004, Soft computing for qualitative and 5, 115–133. doi:10.1007/BF02478259
quantitative seismic object and reservoir property prediction. Part 1: Meldahl, P., and Heggland, R., 2001, Identifying faults and gas chimneys
neural network applications: First Break, 22, 49–54. using multiattributes and neural networks: The Leading Edge, 20,
Bahorich, M., and Farmer, S., 1995, 3-D seismic discontinuity for faults 474–482. doi:10.1190/1.1438976
and stratigraphic features: the coherence cube: The Leading Edge, 14, Meldahl, P., Heggland, R., Bril, B., and de Groot, P., 1999, The chimney cube,
1053–1058. doi:10.1190/1.1437077 an example of semi-automated detection of seismic objects by directive
Bennett, C., Gregg, R., and King, P., 1992, Petroleum geology of the attributes and neural networks: Part I; methodology: 69th Annual
Taranaki basin, with emphasis on the north-eastern quadrant: International Meeting, SEG, Expanded Abstracts, 931–934.
Petroleum Exploration in New Zealand News, 32, 14–32. Meldahl, P., Heggland, R., Bril, B., and de Groot, P., 2001, Identifying fault
Brothers, D. S., Ruppel, C., Kluesner, J. W., Brink, U. S., Chaytor, J. D., Hill, and gas chimneys using multi-attributes and neural networks: The
J. C., Andrews, B. D., and Flores, C., 2014, Seabed fluid expulsion along Leading Edge, 20, 474–482. doi:10.1190/1.1438976
the upper slope and outer shelf of the US Atlantic continental margin: Mouslopoulou, V., Nicol, A., Walsh, J. J., Begg, J. G., Townsend, D. B., and
Geophysical Research Letters, 41, 96–101. doi:10.1002/2013GL058048 Hristopoulos, D. T., 2012, Fault-slip accumulation in an active rift over
Chopra, S., and Marfurt, K. J., 2007, Seismic attributes for prospect thousands to millions of years and the importance of paleoearthquake
identification and reservoir characterization: SEG. sampling: Journal of Structural Geology, 36, 71–80. doi:10.1016/j.jsg.
Cohen, C. R., Christianson, L. J., Bates, C. R., Laney, R. P., and Morton, G. A., 2011.11.010
2006, Pogo New Zealand’s 3D seismic: new standards and structural/ Nicol, A., Walsh, J., Berryman, K., and Nodder, S., 2005, Growth of a normal
stratigraphic insights in the Taranaki Basin: Pogo Producing Company. fault by the accumulation of slip over millions of years: Journal of
Connolly, D. L., 2015, Visualization of vertical hydrocarbon migration in Structural Geology, 27, 327–342. doi:10.1016/j.jsg.2004.09.002
seismic data: case studies from the Dutch North Sea: Interpretation, 3, Nodder, D. S., 1993, Neotectonics of the offshore Cape Egmont Fault
SX21–SX27. doi:10.1190/INT-2015-0007.1 Zone, Taranaki Basin, New Zealand: New Zealand Journal of Geology
Connolly, D., and Garcia, R., 2012, Tracking hydrocarbon seepage in and Geophysics, 36, 167–184. doi:10.1080/00288306.1993.9514566
Argentina’s Neuquen basin: World Oil, 233, 101–104. Palmer, J. A., and Andrews, P. R., 1993, Cretaceous–Tertiary sedimentation
Giba, M., Nicol, A., and Walsh, J. J., 2010, Evolution of faulting and and implied tectonic controls on structural evolution of Taranaki Basin,
volcanism in a back-arc basin and its implication for subduction New Zealand, in P. F. Ballance, ed., South Pacific sedimentary basins:
process: Tectonics, 29, TC4020. doi:10.1029/2009TC002634 Elsevier, Sedimentary Basins of the World, 2, 309–328.
P Exploration Geophysics P. C. Kumar and A. Mandal

Palmer, J. A., and Bulte, G., 1991, Taranaki Basin, New Zealand, in Stagpoole, V. M., and Nicol, A., 2008, Regional structure kinematic history
K. T. Biddle, ed., Active margin basins: American Association of of a large subduction back thrust: Taranaki Fault, New Zealand: Journal
Petroleum Geologists Memoir 52, 261–282. of Geophysical Research. Solid Earth, 113, B01403.
Pilaar, W. F. H., and Wakefield, L. L., 1978, Structural and stratigraphic Thrasher, G. P., 1990, Tectonics of the Taranaki Rift: New Zealand Oil
evolution of Taranaki Basin, offshore North Island, New Zealand: Exploration Conference Proceedings, Ministry of Commerce, Wellington,
Australian Petroleum Exploration Association Journal, 18, 78–93. 124–133.
Qayyum, F., Catuneanu, O., and Bouanga, C. E., 2015, Sequence stratigraphy Tingdahl, K. M., 1999, Improving seismic detectability using intrinsic
of a mixed siliciclastic-carbonate setting, Scotian Shelf, Canada: directionality: Technical Report, Earth Sciences Center, Goteborg
Interpretation, 3, SN21–SN37. doi:10.1190/INT-2014-0129.1 University, B194.
Reilly, C., Nicol, A., Walsh, J. J., and Seebeck, H., 2015, Evolution of Tingdahl, K. M., 2003, Improving seismic chimney detection using
faulting and plate boundary deformation in the Southern Taranaki Basin, directional attributes, in M. Nikarvesh, F. Aminzadeh, and L. A. Zadeh,
New Zealand: Tectonophysics, 651–652, 1–18. doi:10.1016/j.tecto.2015. eds., Soft computing and intelligent data analysis in oil exploration:
02.009 Elsevier, Developments in Petroleum Science 51, 157–173.
Roberts, A., 2001, Curvature attributes and their application to 3-D Tingdahl, K. M., and de Groot, P. F., 2003, Post-stack dip and azimuth
interpreted horizons: First Break, 19, 85–100. doi:10.1046/j.0263- processing: Journal of Seismic Exploration, 12, 113–126.
5046.2001.00142.x Tingdahl, K. M., and de Rooij, M., 2005, Semi-automatic detection of faults
Russell, B., Hampson, D., Schuelke, J., and Quirein, J., 1997, Multiattribute in 3D seismic data: Geophysical Prospecting, 53, 533–542. doi:10.1111/
seismic analysis: The Leading Edge, 16, 1439–1444. doi:10.1190/ j.1365-2478.2005.00489.x
1.1437486 Tingdahl, K. M., Bril, A. H., and de Groot, P. F., 2001, Improving
Schmidt, D. S., and Robinson, P. H., 1989, The structural setting and seismic chimney detection using directional attributes: Journal of
depositional for the Kupe South filed, Taranaki Basin: New Zealand Oil Petroleum Science Engineering, 29, 205–211. doi:10.1016/S0920-
Exploration Conference Proceedings, Ministry of Commerce, Wellington, 4105(01)00090-0
151–172. Walcott, R. I., 1987, Geodetic strain and the deformational history of the
Smith, E. G. C., Stern, T., and Reyners, M., 1989, Subduction and back-arc North Island of New Zealand during the late Cenozoic: Philosophical
activity at the Hikurangi convergent margin, New Zealand: Pure and Transactions of the Royal Society of London. Series A, Mathematical
Applied Geophysics, 129, 203–231. doi:10.1007/BF00874627 and Physical Sciences, 321, 163–181. doi:10.1098/rsta.1987.0009

www.publish.csiro.au/journals/eg

View publication stats