t Special section: Seismic attributes

Seismic attribute detection of faults and fluid pathways within an active

strike-slip shear zone: New insights from high-resolution 3D P-Cable™
seismic data along the Hosgri Fault, offshore California
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Jared W. Kluesner1 and Daniel S. Brothers1

Abstract
Poststack data conditioning and neural-network seismic attribute workflows are used to detect and visualize
faulting and fluid migration pathways within a 13.7 km2 3D P-Cable™ seismic volume located along the Hosgri
Fault Zone offshore central California. The high-resolution 3D volume used in this study was collected in 2012 as
part of Pacific Gas and Electric’s Central California Seismic Imaging Project. Three-dimensional seismic reflec-
tion data were acquired using a triple-plate boomer source (1.75 kJ) and a short-offset, 14-streamer, P-Cable
system. The high-resolution seismic data were processed into a prestack time-migrated 3D volume and pub-
lically released in 2014. Postprocessing, we employed dip-steering (dip and azimuth) and structural filtering
to enhance laterally continuous events and remove random noise and acquisition artifacts. In addition, the struc-
tural filtering was used to enhance laterally continuous edges, such as faults. Following data conditioning, neu-
ral-network based meta-attribute workflows were used to detect and visualize faults and probable fluid-
migration pathways within the 3D seismic volume. The workflow used in this study clearly illustrates the utility
of advanced attribute analysis applied to high-resolution 3D P-Cable data. For example, results from the fault
attribute workflow reveal a network of splayed and convergent fault strands within an approximately 1.3 km
wide shear zone that is characterized by distinctive sections of transpressional and transtensional dominance.
Neural-network chimney attribute calculations indicate that fluids are concentrated along discrete faults in the
transtensional zones, but appear to be more broadly distributed amongst fault bounded anticlines and struc-
turally controlled traps in the transpressional zones. These results provide high-resolution, 3D constraints on the
relationships between strike-slip fault mechanics, substrate deformation, and fluid migration along an active
fault system offshore central California.

Introduction Over the past 20 years, a number of well-known seis-

Seismic attribute calculations are considered to be es- mic attributes have been developed for the analysis of
sential steps in 3D seismic interpretation workflows em- seismic discontinuities related to geologic structures.
ployed by the petroleum exploration industry. The results This includes attributes such as coherency (Bahorich
can accelerate, and in many cases, provide quantitative and Farmer, 1995; Marfurt et al., 1998, 1999), similarity
justification for the interpretation of a variety of geologic (Tingdahl, 1999; Tingdahl et al., 2001; Tingdahl and de
features. Nevertheless, academic researchers aiming to Rooij, 2005), volumetric curvature (Roberts, 2001; Al-
understand near-surface processes and offshore hazards Dossary and Marfurt, 2006; Chopra and Marfurt,
(e.g., by active tectonics and slope stability studies) rarely 2010), and other geometric attributes that perform
have access to high-resolution 3D seismic data. Thus, the mathematic calculations between various lateral trace
application of seismic attribute workflows to such prob- segments and across defined time gates (e.g., Chopra
lems and data remains largely unexplored. New 3D seis- and Marfurt, 2005, 2007). Standard seismic discontinu-
mic technology, such as the high-resolution P-Cable™ ity attribute calculations without structural informa-
system (e.g., Brookshire and Scott, 2015), and expanded tion, such as coherency and similarity, are frequently
access to publically available 3D data sets are providing equated to fault-attribute calculations. However, both
opportunities to develop advanced 3D seismic attribute approaches may struggle to differentiate between geo-
workflows to better understand the near-surface proc- logic features, such as faults, erosional incisions, gas
esses and improve hazard assessments. chimneys, steeply dipping strata, karst collapse struc-

1
Pacific Coastal and Marine Science Center, U.S. Geological Survey, Santa Cruz, California, USA. E-mail: [email protected]; dbrothers@usgs.
gov.
Manuscript received by the Editor 11 August 2015; revised manuscript received 9 October 2015; published online 15 February 2016. This paper
appears in Interpretation, Vol. 4, No. 1 (February 2016); p. SB131–SB148, 11 FIGS., 2 TABLES.
http://dx.doi.org/10.1190/INT-2015-0143.1. Society of Exploration Geophysicists and American Association of Petroleum Geologists

Interpretation / February 2016 SB131

 tures, and also data artifacts. Detection and isolation of of seismic attribute calculations. To suppress the un-
a desired geologic target can be improved by combining wanted noise, we apply a data conditioning workflow
structurally orientated calculations with discontinuity that consists of structural directionality calculations
attribute calculations (e.g., Gersztenkorn and Marfurt, and structural filters. The filtered data are then input
1999; Marfurt et al., 1999; Tingdahl, 1999; Tingdahl et al., into attribute calculations, including maximum curva-
2001; Tingdahl and de Groot, 2003). Yet, undesirable ture and two supervised neural-network meta-attribute
seismic discontinuities (e.g., data artifacts) can still workflows tailored to detect probable faults and chim-
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

be present, possibly complicating the calculated attrib- neys within the 3D volume. The meta-attribute results
ute result and potentially confusing the interpreter show significant noise tolerance and improvement over
(Marfurt and Alves, 2014). single discontinuity attribute calculations.
In an effort to better detect geologic targets and iso- Last, the results from this study provide high-resolu-
late their signals from nondesired geologic features tion 3D constraints on the relationship between fault-
and/or data artifacts, the combination of multiple ing, deformation, and apparent fluid migration within
attributes, or parameterizations of attributes, has been an active strike-slip fault system. The neural-network
shown to provide substantial benefit to the seismic in- fault attribute results provide 3D delineation of the
terpreter. This multiattribute approach was developed geometry and internal structure of restraining and re-
using crossplotting techniques (Chopra and Marfurt, leasing bends, step-over zones, convergence zones,
2009) and soft computing neural-network methodology and paired fault bends (PFBs). Analysis of the neu-
(Russell et al., 1997; Meldahl et al., 1999, 2001; Aminza- ral-network chimney attribute suggests that substrate
deh and de Groot, 2004, 2006; Tingdahl and de Rooij, fluid migration pathways and traps are largely con-
2005). The artificial neural-network approach, also re- trolled by fault mechanics because quantifiable dif-
ferred to as multiattribute cluster analysis (Marfurt, ferences in chimney distribution are observed between
2014), uses multiple weighted seismic attributes and compressional and extension deformation zones.
can be supervised and trained by the interpreter to pro-
vide optimized detection of targeted geologic features. Geologic setting
The approach includes the detection and isolation of The Hosgri Fault forms the southern half of the Hos-
geologic features that produce particular changes in gri/San Gregorio strike-slip fault system, which extends
the seismic response, including faults (Tingdahl and for approximately 400 km along the central California
de Rooij, 2005; Brouwer and Huck, 2011), gas chimneys coast between Point Arguello and San Francisco (Fig-
(Heggland, 2005; Ligtenberg, 2005; Connolly and Gar- ure 1; Sorlien et al., 1999; Lettis et al., 2004; Dickinson
cia, 2012; Kluesner et al., 2013; Brothers et al., 2014; et al., 2005; Johnson and Watt, 2012). The Hosgri/San
Connolly, 2015), salt bodies (Aminzadeh and de Groot, Gregorio is the westernmost active fault zone within
2005), and various lithofacies (Ohl and Raef, 2014). If the broader San Andreas Fault System, a broadly dis-
appropriately developed, parameterized, and opti- tributed, northwest–southeast-oriented transform plate
mized, the combination of multiple attributes into boundary along the West Coast of the United States.
“meta-attributes” can produce results with greater con- The San Andreas Fault proper, located to the east of
trast and fewer unwanted signals than individual attrib- the Hosgri Fault, is the dominant fault within the trans-
utes (e.g., Tingdahl and de Rooij, 2005; Mirkamali et al., form boundary and has an estimated slip rate of approx-
2013). Furthermore, several attributes can be visually imately 30‒36 mm∕yr (McCaffrey, 2005; Meade and
blended for multiattribute display, providing visual en- Hager, 2005; Titus et al., 2011). Faults to the west of
hancement for a more informed interpretation of the the San Andreas Fault carry an estimated approxi-
seismic data (Marfurt, 2015). mately 4‒5 mm∕yr of dextral slip (McCaffrey, 2005;
In this study, we use a neural-network multiattribute Meade and Hager, 2005); Johnson et al. (2014) and
approach in an attempt to optimize the detection and PG&E (2014) use offset seabed morphology to con-
visualization of faults and fluid pathways within a tec- strain the late Pleistocene to present slip rate at approx-
tonically active strike-slip shear zone. The high-resolu- imately 2.6 and 1.75‒1.90 mm∕yr, respectively, for the
tion 3D seismic data volume used in this study was section of the Hosgri Fault that passes through the
collected in shallow water across a portion of the Hos- study area.
gri Fault Zone (HFZ), located offshore central Califor- Our investigation focuses on the HFZ offshore of
nia (PG&E, 2014). Seismic acquisition was carried out Point Sal (Figure 1) using 3D seismic-reflection data
using the P-Cable system, a 3D seismic acquisition sys- collected by Pacific Gas and Electric (PG&E) in 2012
tem used to bridge the resolution gap between tradi- (Ebuna et al., 2013; PG&E, 2014). Previous work in
tional 2D/3D seismic reflection operations and high- the Point Sal area, based on single-channel high-resolu-
resolution subbottom acoustic data (e.g., Petersen et al., tion seismic-reflection and marine magnetic data, sug-
2010; Nishenko et al., 2012; Ebuna et al., 2013; Brook- gests that multiple fault strands characterize the HFZ,
shire and Scott, 2015). High-frequency noise, possibly including an intersection with the northwest-trending
from surface statics or navigation errors, is present Lions Head Fault (LHF; Figure 1; Johnson and Watt,
throughout the publicly released 3D volume. This 2012). Hansen et al. (2004) and Johnson and Watt
high-frequency noise limits the accuracy and quality (2012) map a Hosgri Fault “section boundary” near

SB132 Interpretation / February 2016

 the convergence of the Hosgri and LHF based on prestack time migration (Fugro Consultants Inc., 2012;
changes in fault trend and local deformation patterns. PG&E, 2014).
The portion of the HFZ offshore of Point Sal under- The vertical resolution in the shallow substrate is ap-
lies the Santa Maria River delta and is blanketed by a proximately 2 m based on the dominant frequency and
southward-thinning wedge of latest Pleistocene and estimated 1600 m∕s P-wave velocity (PG&E, 2014). It
Holocene sediment. A prominent erosional uncon- was determined that source characteristics and near-
formity (inferred to be the transgressive surface) sepa- seafloor reverberation limited the vertical resolution
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

rates these young sediments from underlying older at the seafloor to approximately 5 ms, or approximately
Quaternary deposits (Johnson and Watt, 2012; PG&E, 4 m (PG&E, 2014).
2014). A deeper angular unconformity separates the in-
ferred Quaternary deposits from Neogene bedrock Data conditioning and seismic attribute workflow
(Johnson and Watt, 2012; PG&E, 2014). Faulting and de- Postprocessing data conditioning
formation commonly warp, cut, and offset Quaternary Following PG&E’s public release of data acquired by
strata. Local basins and growth folds occur within and the Central Coastal California Seismic Imaging Project
adjacent to the HFZ (Johnson and Watt, 2012). The (e.g., PG&E, 2014), the 3D seismic volume located off-
2012 high-resolution 3D seismic volume data from shore Point Sal was downloaded in SEGY format from
the offshore of Point Sal similarly reveal
that this portion of the HFZ is character-
ized by fault bends, en echelon fault pat-
terns, and structures that result from a
relatively complicated deformation his-
tory (PG&E, 2014).

3D seismic reflection data

acquisition and processing
In 2012, a 13.7 km2 3D P-Cable seis-
mic volume was collected as part of
PG&E’s Central California Seismic Im-
aging Project (PG&E, 2014). The high-
resolution seismic reflection data were
acquired using a triple-plate boomer
source (1.75 kJ input power), with a
seismic frequency range of approxi-
mately 100–700 Hz and a peak fre-
quency of 200–225 Hz (PG&E, 2014).
Seismic traces were recorded using
an array of 14, eight-channel (6.25 group
spacing) digital streamers connected to
a single P-Cable towed perpendicular
to the ship’s heading (Nishenko et al.,
2012; Ebuna et al., 2013). The 14 stream-
ers were spread apart by towed para-
vanes providing a 14 × 8 (112 total)
channel receiver array with a 43.75 m
crossline width and a 50 m inline
length. Sail lines were spaced 37.5 m
apart and oriented northwest–south-
east (338°/158°, Figure 1) and included
a shot interval of 3.125 m, a sample
interval of 0.25 ms (4 kHz sample
rate), and a 0.75 s record length. This
acquisition geometry produced eight-
fold stacked sections with a 3.125 ×
3.125 m bin size (PG&E, 2014). Seismic
processing steps consisted of quality Figure 1. Regional map showing the location of the 3D seismic reflection data
volume for offshore central California. The red rectangle denotes the location
control checks, tidal corrections, veloc- and extent of the 3D seismic volume along a portion of the Hosgri Fault. The
ity analysis, NMO correction, stack- black lines highlight the approximate locations of fault traces. The small blue
ing, surface-related multiple elimination, box in the inset in the top right corner shows location of the regional map along
deconvolution, crossline statics, and the California coast.

Interpretation / February 2016 SB133

 the United States Geological Survey (USGS)’s National structural information was particularly useful in this
Archive of Marine Seismic Surveys. After preliminary study because of the dense network of faults and
analysis of data quality, a data conditioning workflow steeply dipping strata associated with deformation.
was developed to further enhance the quality and inter- The supervised neural-network module used in this
pretability of the seismic volume. Initially, a dip-steering study is a fully connected multilayer perceptron with
(structural steering) volume was calculated that con- one hidden layer. The algorithm used back propagation
tains the local dip and azimuth of seismic events at with momentum and weight decay. Attribute weights
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

every sample position (Tingdahl et al., 2001; Tingdahl were multiplied by a weight decay factor to produce
and de Rooij, 2005). The steering algorithm used a 1 × smoother functions with improved generalization prop-
1 × 1 (inline, crossline, and sample interval) calculation erties (Aminzadeh and de Groot, 2004; Brouwer et al.,
step-out. In addition, a 1 × 1 × 1 step-out median filter 2011). Supervised training consisted of user-defined
was applied to the steering calculation results. This pro- picks along a desired target (e.g., faults and chimneys)
cedure produced a smoothed 3D volume of structural and picks along nontarget examples within the 3D seis-
information for all the seismic events. mic volume. The neural network was then trained, and
Next, a dip-steered median filter (DSMF) was para- the data were split into training and testing vector sets
meterized to reduce random noise. The filter utilized a (Figure 5). The error of the training vectors was used to
1 × 1 inline-crossline step-out and no time gate/window update the weights, and the test vectors were used to
(0 ms). To evaluate noise suppression and potential sig- check performance and avoid overfitting. Training
nal removal, the filtered results (Figure 2c) were sub- was stopped when the error on the test vectors was
tracted from the input (Figure 2a) to evaluate the minimal, the point at which the neural network has op-
signal removed during filtering (Figure 2b). As is shown timal generalization capabilities (Aminzadeh and de
in Figure 2b, a significant amount of noise oriented in Groot, 2004; Brouwer et al., 2011). The generated
the inline and crossline directions was isolated and re- meta-attribute results represent a measurement of rel-
moved, whereas the minimal desired signal was sup- ative probability for the desired target (e.g., chimney,
pressed. The structurally filtered results provided a fault, mass slide deposit, and salt body) that ranges be-
reduction of random and apparent acquisition noise,
tween zero and one, with one representing the highest
while preserving edges and enhancing laterally continu-
probability.
ous events (Brouwer and Huck, 2011).
To enhance the detection and interpretability of
Neural-network fault cube
faults, a dip-steered diffusion filter was applied to the
The 32 attributes, including multiple instances of
DSMF data (Figure 2d). The dip-steered diffusion filter
used a 1 × 1 (inline and crossline) step-out with no time some attributes with different parameters, were used
gate (0 ms). The structurally steered diffusion filter re- as input nodes into the calculation of the neural-net-
placed low-quality traces with surrounding traces of work fault cube meta-attribute. The hidden layer of
better quality and enhanced continuous edges, such the supervised neural-network consisted of 16 nodes.
as reflector offsets caused by faults. Figure 3c shows Table 1 lists the various attributes and weights used
the sharpening of fault zones after the application of during neural-network training. Various parameteriza-
dip-steered median and diffusion filtering. tions of the similarity attribute were used as weighted
Last, a predictive deconvolution filter was parame- input nodes, as well as attributes, such as curvature,
terized and applied to the DSMF data in an attempt to noise, average frequency, and others (Table 1). Multi-
suppress the apparent reverberation below the sea- ple attributes utilized structural information and vari-
floor reflection (PG&E, 2014). Throughout the 3D ous step-out ranges. Approximately 2000 picks of fault
seismic volume, the seafloor reflection suffers from and 3300 picks of nonfault examples were used for
a wide wavelength, and often masks near-surface phe- supervision of the neural-network training (Figure 5).
nomena (e.g., bright spots in Figure 4). The approach Training was stopped when the normalized root-mean-
effectively suppressed the reverberation and also square (rms) error of the train and test sets reached a
sharpened subbottom reflectors, yielding a slight in- minimum value (i.e., before the rms error trend shifted
crease in the resolvability of the seismic stratigraphy upwards, indicating overfitting), as well as the misclas-
(Figure 4). sification percentage. Input vectors were balanced us-
ing 1% noise. Post training, a seismic volume was
Seismic attribute workflows generated that contained fault meta-attribute probabil-
We employed two separate attribute workflows to ity measurements output from the meta-attribute cal-
the preconditioned seismic data for the detection of culation. Fault and nonfault probabilities were then
faults and fluid-migration pathways (i.e., seismic chim- projected along inlines, crosslines, and z-slices within
ney zones). The workflows used in this study utilized the 3D seismic volume. This allowed for the examina-
dip steering and neural networks. Dip steering im- tion and correlation of probable faults to reflector off-
proved the geometric attribute accuracy and object de- sets observed on amplitude data, helping to evaluate
tection power by providing structural information (e.g., the validity of the neural-network fault attribute re-
dip and azimuth) during attribute calculations. The sults (Figure 2h).

SB134 Interpretation / February 2016

 Neural-network chimney cube training. Similar to the fault meta-attribute calculation,
The 36 attributes were used as input nodes into the the hidden layer with the chimney meta-attribute calcu-
supervised neural-network chimney calculation, and lation consisted of 18 nodes, or half of the input nodes.
each node was weighted during the neural-network Table 2 lists the different attributes and assigned
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 2. Multiple panels showing examples of the data conditioning and neural-network attribute steps along a subsection of a
z-slice at 174 ms. (a) Input data, (b) noise extracted, and the inset shows expanded view of noise oriented in inline and crossline
directions, (c) results after noise suppression using a DSMF, (d) dip-steered diffusion filtering applied to DSMF results, (e) dip-
steered similarity attribute results, (f) neural-network fault attribute results, (g) neural-network chimney attribute results coren-
dered with the fault-attribute results, (h) combination of fault and chimney attribute results corendered with DSMF amplitude data.
(i) DSMF maximum curvature attribute with 50% transparency superimposed over the neural-network fault attribute results. Fig-
ure location is indicated by TH dashed red lines in Figure 6. SFS, southern fault strand and WSF, western fault strand.

Interpretation / February 2016 SB135

 weights used during the training procedure. The chim- fault cube to delimit fault and fold geometries and to
ney meta-attribute input nodes used attributes tailored extract information on sense of vertical throw across
to criteria that characterize seismic chimneys, such faults (e.g., Figure 2i; Roberts, 2001).
as verticality (vertically parameterized attributes), dis-
continuity, chaos, frequency anomalies, and others Interpretation
(Brouwer et al., 2011). Approximately 1800 picks of Fault meta-attribute and maximum curvature
chimney and 2400 picks of nonchimney were used to Results from the neural-network fault attribute cal-
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

help to train the neural-network calculation. Similar culation show improvement in fault detection over
to the neural-network fault attribute approach, training single discontinuity attribute calculations within the
was stopped when the rms error reached minimum val- high-resolution 3D seismic volume. The fault-probabil-
ues along with a misclassification percentage of user ity attribute results provide better isolation of the
picks. Input vectors were balanced using 1% noise. A desired fault detection while also maintaining or creat-
few of the highest weighted attributes observed in ing greater contrast along imaged fault strands. Figure 2
the neural-network chimney calculation were parame- underscores such improvement in fault detection and
terized similarity attribute calculations, the “simple isolation along a subsample of the 174 ms z-slice also
chimney attribute,” average frequency, median filter po- shown in Figures 6 and 7. Using the single dip-steered
lar-dip, and variance (Table 2). The simple chimney attribute approach, multiple unwanted seismic discon-
attribute utilized a combination of similarity calcula- tinuities are enhanced, including lateral discontinu-
tions tailored for verticality measurements. ities caused by erosional incision and folded, steeply
dipping strata (Figure 2e). By applying the trained
Maximum curvature neural-network meta-attribute calculation, portions of
In addition to the neural-network attribute work- the unwanted discontinuities are suppressed, which im-
flows, maximum curvature (K max ) was calculated as- proves the resolvability of desired fault strands and
suming a velocity of 1600 m∕s. An inline-crossline smaller scale fault complexities, including features such
step-out of 10 was used during the calculation, and a as subtle fault bends, junctions, and step-over segments
DSMF was applied to the results. The maximum curva- (Figure 2f). In addition, fault strands with weak contrast
ture calculation can be used in conjunction with the using the single attribute calculation show a higher con-

Figure 3. Example of a 2D crossline illustrating the improvement of fault detection using structural filtering. (a) Input data,
(b) DSMF data, (c) DSMF data with dip-steered diffusion filtering. Insets below show enlarged portions of the profiles outlined
by the dashed black boxes above. Note the improved imaging and sharpening of faults. The 2D profile location is outlined by the
dashed horizontal red line (crossline) in Figure 6. VE, vertical exaggeration.

SB136 Interpretation / February 2016

 trast and are more continuous in the neural-network zone was also identified by PG&E (2014) and was in-
fault attribute result (Figure 2). A color scheme that ferred to be the near-surface expression of a blind
grades into transparency for probabilities less than ap- thrust fault (BTF). The NFS of the HFZ extends to
proximately 60% allows for improved visualization, the northwest corner of the 3D volume; however, lack
evaluation, and interpretation of the fault attribute of coverage to the east prevents additional analysis of
results superimposed on filtered amplitude data (Fig- the BTF intersection (Figure 6).
ure 2h). Maximum curvature corendered with the fault
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

attribute provides additional structural information, Chimney meta-attribute

such as fault throw. Figure 2i highlights such patterns, The neural-network chimney attribute results pro-
displaying the upthrown and downthrown sides of de- vide additional insight into potential fluid/gas migration
tected fault traces along a horizontal z-slice at 174 ms. zones present within the 3D seismic volume. As with the
Projection of the fault probability attribute calcula- fault attribute results, the chimney results can be pro-
tion across horizontal z-slices reveals a complex net- jected along a z-slice for plan-view analysis of the
work of faults that span the length of the 3D seismic distribution of probable fluid-migration pathways. Fig-
volume (Figure 6c). A z-slice at 174 ms two-way travel- ure 2g shows an enlarged section of the neural-network
time reveals single fault strands at the northern and chimney attribute calculated along a portion of the
southern edges of the 3D volume. These faults are re- 174 ms z-slice. A gradational color scale with transpar-
ferred to as the northern and southern fault strands ency reveals only the highest (approximately 90% and
(NFS and SFS), and they appear to be the master above) chimney probabilities, allowing for visual in-
strike-slip faults of the HFZ within the survey region. spection of several attribute calculations along the
There is an approximately 6° difference in strike be- same surface (Figures 2h and 7a). By projecting the
tween the NFS and SFS (330° and 336°, respectively; chimney results along vertical inlines and crosslines,
PG&E, 2014) and a complicated network of splayed it becomes evident that zones of high chimney proba-
and convergent fault strands throughout the intervening
area (Figure 6c). The westernmost fault strand (WFS)
splays off of the SFS and can be traced for more than
5 km to the north within the central portion of the seis-
mic survey. The trace of the WFS is characterized by a
series of right steps that appear to be associated with
small pull-apart structures as highlighted by the maxi-
mum curvature attribute as downthrown fault zones
(Figure 2i). In addition, shorter, secondary faults splay
off of the WFS to the north. The HFZ widens (approx-
imately 1 km) and becomes highly distributed to the
north, where fault identification becomes complicated
by broad patches of high discontinuity (Figure 6c) and
complex curvature results (Figure 7b). Just north of the
broadest point, secondary fault strands quickly con-
verge to the west, where the fault attribute results be-
come chaotic and dispersive. A large discontinuity zone
with strong curvature amplitudes converging from the
southeast also complicates this region, intersecting the
chaotic fault zone within the central portion of the 3D
volume (Figure 7b).
The large discontinuity zone converging from the
southeast is identified as the LHF (Figure 6c). The in-
tersection of the LHF coincides with the northern ter-
mination of the WFS and associated splayed faults in
the southern half of the 3D volume. High curvature am-
plitudes are observed in this region (Figure 7b). To the
west of the LHF and HFZ intersection point, a continu-
ous and sharply defined fault emerges from the western
edge of the 3D volume and converges with the NFS.
This fault strand is interpreted to be a PFB due to fault Figure 4. Subsection of a 2D crossline showing indicators of
convergence, and it will be described in greater detail in gas/fluid and results from poststack predictive deconvolution.
(a) DSMF data, (b) DSMF data with poststack predictive de-
the following sections. North of the confluence between convolution. Note the apparent flat spot and possible velocity
the PFB and the NFS, another fault zone emerges from push-down. The location is outlined by the solid black line in
the southeast and intersects the HFZ near the northern Figure 6. The intersection of Figure 10 is delineated by the
edge of the 3D volume (Figures 6c and 7b). This fault solid blue line. VE, vertical exaggeration.

Interpretation / February 2016 SB137

 bility commonly terminate at the base of shallow bright and 8). Folding and an apparent flat-spot at the apex of
spots (e.g., Figure 8). an anticline are also observed in this region (Figure 4),
Visualization of the chimney attribute results in con- and this is suggestive of significant fluid-focusing and
junction with the fault attribute results reveals a strong trapping. Furthermore, zones with anticlinal folding,
correlation between faults and potential fluid-migration such as the LHF and PFB zones, commonly show strong
pathways throughout the study region. For example, maximum curvature amplitudes (Figure 7b). Additional
the chimney attribute results projected along a z-slice zones of high chimney probability are observed along
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

at 174 ms shows a strong connection between detected the NFS and along the projected BTF as it emerges from
faults and chimney distribution (Figure 6d). Within the the southeast (Figure 6d).
southern half of the 3D volume, high-probability chim-
ney zones are predominantly concentrated along dis- Relationship among faults, deformation,
crete faults. Figure 2g and 2h is the example of high- and probable fluid pathways within
probability chimney zones present along portions of de- an active strike-slip shear zone
tected fault strands, and Figure 9b is a 3D perspective Attribute results indicate a strong connection be-
view of the same region. To the north, where the WFS- tween the dominant mode of deformation and the prob-
splayed fault zone widens to approximately 1 km, the ability for gas chimneys and/or fluid pathways. In the
chimney signals become more disperse, with some re- following interpretive sections, we break the 3D study
gions exhibiting large chimney clusters that span across area into three zones: (1) a transtensional zone that
separate splayed fault strands (Figure 6d). characterizes the southern end of the survey region,
The confluence of the LHF and the HFZ/NFS is asso- (2) a transpressional zone located near the center of
ciated with broad patches of high chimney probability the 3D volume, and (3) the LHF/BTF intersection zone
that nearly extend to the western edge of the 3D volume within the northern half of the study area (Figures 7c
(Figures 6d and 9d). Although the dense chimney zone and 9). For additional structural details, mapped hori-
diminishes as the NFS and the PFB become sharply de- zons, and estimated slip rates, see PG&E (2014). In
fined and discrete fault strands, high chimney probabil- addition, Figure 6–28 in PG&E (2014) provides a com-
ities and clusters of shallow bright spots are observed in parison with previous lower resolution mapping within
and around the confluence of the two faults (Figures 6d the region.
1) Examination of offset reflectors
in the southern region of the 3D seis-
mic volume and maximum curvature
results suggests that the faults have
oblique kinematics, with observed
dip-slip in the normal direction (Fig-
ures 3, 7b, and 10). Continuous,
parallel, and subparallel reflectors
within a sedimentary basin located
near the southern end of the survey
(Figure 10) are cut by faults branch-
ing off of the WFS (Figures 2h and
6b); reflectors are displaced in the
horizontal and normal directions,
creating small, localized depressions
along the fault strands. These down-
thrown regions are highlighted on
z-slices by corendering the fault
attribute with maximum curvature
(Figures 2i and 7b). Based on these
observations, and comparison with
interpretations by PG&E (2014), this
section of the HFZ appears to
be dominated by transtensional de-
formation (Figure 7c). Chimney
attribute results generally show the
Figure 5. Example of neural-network attribute training window. The top-left highest probability along discrete
panel shows normalized rms values during training. The bottom-left panel shows
the misclassification percentage of user picks during training. The right panel
faults in places dominated by trans-
shows attribute node weights during training, with warmer colors representing tension, suggesting that upward
higher weights. This example was paused during training, and the attribute fluid migration is mostly confined
weights do not correlate to those in Tables 1 and 2. to fault-generated permeability path-

SB138 Interpretation / February 2016

 ways (Figures 2g and 9b). High chimney probabil- 2) Across the central part of the 3D survey inline and
ities are observed along the northern extension of crossline, seismic profiles show evidence of folding
the SFS (Figures 6d and 9b), but they diminish along and uplift within the HFZ (e.g., Figures 9c and 10).
the fault toward the north, where the deformation The uplifted strata form anticlinal folds that are cut
zone between the SFS and the WFS widens to more by various fault strands of the HFZ. Within this zone,
than 1 km. As chimney probabilities decline north-
ward along the SFS, more disperse zones of strong
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

chimney signal appear in the substrate between indi- Table 2. List of attributes used in the neural-network
chimney attribute calculation. Attribute names are
vidual faults (Figure 6d). This broad zone of higher provided in the left column, and attribute weight
chimney probabilities also coincides with the onset percentages used during training are listed in the
of compressional folding (Figures 7c and 10). right column.

Attribute nodes Weight

Table 1. List of attributes used in the neural-network percentage
fault attribute calculation. Individual attributes or
parameterizations of attributes are listed in the left
column. The right column lists attribute weights used Similarity parallel [−40, 40] 32.1
during the neural-network training. Similarity diagonal [−40, 40] 29.4
Down 100 ms similarity parallel [−40, 40] 16.9
Attribute nodes Weight percentage Up 100 ms similarity parallel [−40, 40] 27.6
Down 100 ms similarity diagonal [−40, 40] 17.6
Similarity long-window parallel 14.9 Up 100 ms similarity diagonal [−40, 40] 17.1
Similarity long-window diagonal 20 Similarity all directions [−40, 40] 54
Similarity long-window all directions 23.3 Down 100 ms similarity all directions 16.8
Similarity midwindow parallel 35.6 [−40, 40]
Similarity midwindow diagonal 22.8 Up 100 ms similarity all directions [−40, 40] 21.8
Similarity midwindow all directions 38.6 Simple chimney attribute 44.3
Similarity short-window parallel 37.6 No neural-network input polardip 18.1
Similarity short-window diagonal 32.7 rms [−40, 40] 30.4
Similarity short-window all directions 37.4 Down 100 ms rms [−40, 40] 35.9
Similarity midwindow double step-out 13.3 Up 100 ms rms [−40, 40] 8.7
parallel No neural-network input filter residual 18.4
Similarity midwindow double step-out 23.6 No neural-network median dip filter 6.9
diagonal step-out = 3
Similarity midwindow double step-out all 33.1 Noise rms [−40, 40] filter residual 39
directions
Similarity midwindow no-steering parallel 28.4 Down 100 ms noise 28.2
Similarity midwindow no-steering diagonal 22.6 Up 100 ms noise 33.2
Similarity midwindow no-steering all 29.1 No neural-network input rms [40, 40] 26.5
directions median
Most-positive curvature 12.8 dip filter
Signal/noise 36.7
Most-negative curvature 10.7
Down 100 ms signal/noise 18.2
Dip curvature 8.6
Up 100 ms signal/noise 27.1
Strike curvature 6.1
TWT 37.3
Curvatureness 21.8
Average frequency 40
TWT 10.8
Average frequency with simple time 34.1
Polar dip 16.3 correction
rms short 40.5 Median filter polar dip 100
rms long 10.4 No neural-network low-pass 10.6
Median dip filter step = 3 10.6 No neural-network rms low-pass 18.7
Filter residual 9.4 No neural-network high-pass 9.3
Noise 100 No neural-network rms high-pass 28.9
Signal 13.6 Frequency washout ratio 17.6
S/N 72.7 Variance [−40, 40] polar dip 56.8
Stability of steering 19.6 Frequency washout ratio with simple time 15.9
Average frequency 6.5 correction
Down 100 ms variance [−40, 40] polar dip 30.5
Average frequency with simple time 9.3
correction Up 100 ms variance [−40, 40] polar dip 18.5

Interpretation / February 2016 SB139

 the major unconformity that separates older Neo- PG&E, 2014), and in some regions, this unconform-
gene rocks from younger Quaternary sediments is ity nearly reaches the seafloor (Figure 10). Relative
clearly folded and uplifted (e.g., Figure 6–33 in to the southern extent of the 3D volume, chimney

Figure 6. Multiple z-slice panels at 174 ms

showing data enhancement and attribute re-
sults. Panels have been rotated 22° clockwise
from true north. (a) Input data, (b) DSMF
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

and diffusion filtered data, (c) neural-network

fault attribute results, (d) neural-network chim-
ney attribute results corendered with fault-
attribute results. The dashed red lines show
the location of Figures 2, 9a, and 9b. The
dashed purple lines show the location of Fig-
ure 8c and 8d. The dashed black lines show
location of Figure 8. The solid blue line de-
notes the location of Figure 10, and the solid
black line displays the location of Figure 4.
NFS, northern fault strand; SFS, southern fault
strand; WFS, western fault strand; EZ, exten-
sional zone; SL, structural leakage; CZ, com-
pression zone; LHF, Lions Head Fault; PFB,
paired fault bend; BTF, blind thrust fault;
HO, horizontal offsets; DSMF, dip-steered
median filter; DSDF, dip-steered diffusion fil-
ter; NNFC, neural-network fault cube; and
NNCC, neural-network chimney cube.

SB140 Interpretation / February 2016

 attribute results show a widespread distribution of
high probabilities in the areas between faults (Fig-
ure 6d), particularly within fault-bounded structures
typically associated with transpression, such as
growth folds, anticlines, and steeply dipping reverse
faults (e.g., Figure 9d). Maximum curvature ampli-
tudes are also very high in this region, and these re-
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

sults help to characterize the folded regions on

z-slices (Figure 7b). Based on these observations,
the central region appears to be dominated by trans-
pressional faulting and folding (Figure 7c), an inter-
pretation also proposed by PG&E (2014) (e.g.,
Figure 6–33 in PG&E, 2014).
3) The northern part of the 3D volume is complicated
by the convergence of the LHF and BTF with the
NFS. Where these faults converge with the NFS,
transpressional deformation is observed (Figures 7c,
9c, and 10), and the convergence appears to have
generated a type of “PFB,” similar to the models pro-
posed by Mann (2007) and Johnson and Watt (2012).
For example, farther north along the HFZ, Johnson
and Watt (2012) propose that a PFB formed west of
the HFZ due to indentation of a rigid basement block
between the HFZ and the shoreline fault (Figure 1).
Similarly, we propose that the convergence of the
LHF and the HFZ (or more specifically, the NFS)
forms an impinging “block zone” and developed
a PFB to the west of the convergence zone. The
leading restraining and trailing releasing bends
associated with this zone of impingement have sub-
sequently led to compression and extension (Fig-
ure 10) within the region bounded by the PFB
and the NFS (Figure 7c). High chimney probabilities
and numerous shallow bright spots are associated
with the compressional portion of the PFB (Fig-
ures 7c, 6d, and 9d). To the north, where the inter-
preted BTF (PG&E, 2014) converges with the NSF,
another zone of compression and uplift is observed
(Figures 7c and 10). Chimney attribute results indi-
cate a more fault-focused distribution in this region
except on the northernmost portion of the 3D vol-
ume, where the fault, chimney, and curvature attrib-
utes exhibit a dispersive pattern. Although limited
by data coverage, numerous shallow bright spots
are observed above these high-probability chimney
zones.
Along the portion of the HFZ imaged in this study,
Figure 7. Some z-slices at 174 ms showing corendered neural-
substrate fluid concentration appears to be closely tied
network attribute results, maximum curvature, and structural
to the kinematics of strike-slip faults and the associated interpretations. (a) Neural-network fault and chimney attribute
structural deformation. Transtension along the HFZ in results with a transparency color scheme on top of structurally
the southern portion of the 3D survey area promotes filtered seismic amplitude data, (b) DSMF maximum curvature
normal-oblique motion along the fault strands, resulting attribute with 50% transparency superimposed over the neural-
in subsidence and sedimentary infill (Figures 9a and network fault attribute results. (c) Structural interpretation
10). Subsidence displaces fluid-rich rocks located be- of faults and regions of compression and extension. NFS,
northern fault strand (green); SFS, southern fault strand (blue);
low the Neogene-Quaternary unconformity (i.e., the WFS, western fault strand (red); LHF, Lions Head Fault
Monterey Formation; Pisciotto and Garrison, 1981; (dashed purple); PFB, paired fault bend (dark blue); and
Isaacs et al., 1983) farther away from the seafloor. Com- BTF, blind thrust fault (dashed orange). The black lines re-
paction of the relatively thick section of concordant present smaller unnamed faults within the survey region.

Interpretation / February 2016 SB141

 parallel and subparallel bedded sedimentary infill may higher probability for chimneys in areas dominated by
force fluids to migrate horizontally toward the fault- transpression (Figure 11).
bounded margins of the transtensional zones. Because Transpressional deformation is most apparent along
extensional faults have lower confining normal stress strike-slip restraining bends and fault-convergence
(e.g., Means, 2012), it is conceivable that they act as zones, such as the NFS-LHF and subsequent PFB. Other
least resistance fluid-flow pathways and provide a fo- studies of offshore California have also documented
cused migration route to the seafloor. In contrast, trans- seafloor seepage along strike-slip fault restraining
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

pression in the central and northern regions of the 3D bends and step-overs (e.g., Hein et al., 2006; Paull et al.,
study area is linked to folds, anticlines, and uplifted sec- 2008; Maloney et al., 2015). Similarly, fault-controlled
tions of older fluid-rich rocks, bringing the Neogene- seepage patterns have been recently observed within
Quaternary unconformity to just below the seafloor the Marmara Sea along the North Anatolian Fault Sys-
in a few locations. The transpressional folds appear tem. Seeps appear to be concentrated along discreet
to create fluid traps and localized reservoirs except faults in transtensional zones, but are widely distributed
where they are cut by multiple strands of the HFZ (e. in transpressional zones (i.e., anticlinal features; Dupré
g., Figure 9d). The combination of transpressional fold- et al., 2015). However, through the use of advanced
ing and strike-slip faulting appears to promote struc- attribute analyses, this study provides high-resolution
tural trapping and fluid-escape pathways along faults insight into 3D patterns of structural deformation (i.
that cut through the traps. Statistical analysis of chim- e., formation of traps and fluid reservoirs) and associ-
ney attribute results along subsections of the transten- ated fluid migration pathways along a transpressional
sional and transpressional zones reveals a substantially section of an active strike-slip fault.

Figure 8. A 3D perspective view of the PFB region showing the neural-network fault and chimney attribute results. Fault and
chimney attribute results along a z-slice at 174 ms show relationship between faulting and chimneys. Chimney attribute results are
also projected onto vertical inlines and crosslines. Note the termination of a dense zone of high-chimney probabilities below
shallow, abruptly ending high-amplitude reflectors interpreted as bright spots. Location provided by dashed black lines in Figure 6.
NSF, northern fault strand and PFB, paired fault bend.

SB142 Interpretation / February 2016

 Potential pitfalls and future work compared with single-attribute approaches; Figure 2e
Although the neural-network seismic attribute work- and 2f), reflections associated with high-angle dipping
flows provided new insights into the distribution of strata can be of lower quality and lower amplitude, and
probable faults and fluid-migration pathways, some they may fail to register in the structural (steering) cal-
potential pitfalls or limitations must be discussed. Sim- culation. The result is unwanted discontinuity returns
ilar to the “structural leakage” artifact discussed by during structurally steered fault-attribute calculations.
Marfurt and Alves (2014), regions with steeply dipping Fortunately, steeply dipping strata are commonly asso-
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

strata or a poor seismic signal can complicate dis- ciated with circular-to-elliptical fold structures and typ-
continuity attribute calculations, and such zones are ically show a contourlike pattern on horizontal z-slices,
present within the transpressional segments of the allowing them to be easily identified and interpreted as
HFZ in this study (e.g., Figures 6c and 10). Such fea- unwanted artifacts (e.g., “SL” in Figure 6c).
tures can exaggerate the probability values in the Because the neural-network chimney attribute calcu-
fault-attribute results. Although the neural-network re- lation is heavily dependent on measurements tailored
sults reduce unwanted discontinuities (especially when to verticality, regions of low seismic quality from

Figure 9. Three-dimensional perspective views showing 3D volume rendering of neural-network chimney attribute results along
the transtensional and transpressional zones within the 3D seismic volume. (a) View of inline, crossline, and z-slice (174 ms) within
the transtensional region, (b) transtensional region with fault attribute projected along the z-slice and a 3D volume rendering of the
chimney attribute results, (c) view of filtered data within a transpressional region. (d) Transpressional region with fault-attribute
results along the z-slice and 3D rendering of the chimney attribute above. Note the differences in abundance and distribution of
high chimney probabilities between the two different regions. WFS, western fault strand and SFS, southern fault strand. The lo-
cations are outlined by the dashed red (panels a and b) and purple (panels c and d) lines in Figure 6.

Interpretation / February 2016 SB143

 steeply dipping strata can also produce unwanted high ered seismic indicators of fluid flow and leakage
chimney probabilities. Again, structural leakage zones anomalies (e.g., Løseth et al., 2009). Our results indicate
may complicate the targeted attribute analysis and lead that regions of high chimney probability commonly ter-
to unrealistic interpretations. Although the chimney minate upwards into the base of shallow bright spots
attribute results calculated in this study along the (Figures 8 and 9d). This correlation helps to strengthen
HFZ likely contain regions of unwanted high probabil- the viability of the neural-network chimney attribute re-
ities, these zones of apparent structural leakage can sults presented herein. However, the validity of the
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

also be identified by the interpreter and discarded. Seis- chimney attribute results would be improved with
mic chimneys and bright spots are commonly consid- additional criteria, such as geochemical data and

Figure 10. Two-dimensional inline through the HFZ. Inline shows three defined structural regions: the transtensional zone in the
southern portion of the survey, transpressional zone near the central region, and the LHF/BTF intersection zone in the northern
half of the survey. Note regions of apparent subsidence (transtensional) and uplift (transpressional). The location of the inline is
delineated by the solid blue line in Figure 6. N-Q, Neogene-Quaternary; LHF, Lions Head Fault; BTF, blind thrust fault; PFB, paired
fault bend; and VE, vertical exaggeration.

Figure 11. Histogram plots of the neural-net-

work chimney attribute results from the z-sli-
ces shown in Figure 9. Top histogram shows
the chimney probability values from the trans-
tensional region (Figure 9b), whereas the bot-
tom histogram plot shows the probability
values from the transpressional zone (Fig-
ure 9d). Note the abundant signal near zero
in the transtensional zone, indicating numer-
ous chimney values with low probabilities.
In contrast, the histogram plot from the trans-
pressional region shows an abundant signal
near one, indicating numerous chimney sig-
nals with high probabilities.

SB144 Interpretation / February 2016

 high-resolution seafloor and water column imagery Seismic Imaging Project. We also thank S. Nishenko,
(e.g., Connolly, 2015). G. Greene, and P. Hogan. We are grateful to OpendTect
Future work will hopefully include the collection and and dGB Earth Sciences for the use of their software
interpretation of seafloor multibeam and water column and associated plug-ins. S. Johnson and J. Watt pro-
backscatter data across the 3D seismic survey area. vided very helpful and insightful discussions on the
Integration of the seismic attribute results with high- HFZ and its associated geologic background. We thank
resolution seafloor and water column backscatter D. Ebuna for a helpful review and for providing addi-
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

information may provide a link between surface and tional details on the acquisition of the 3D P-cable data.
subsurface processes, such as substrate fluid migration In addition, P. Hart is thanked for a constructive review.
and seabed fluid expulsion (e.g., Berndt, 2005; Ligten- Any use of trade, firm, or product names is for descrip-
berg, 2005; Judd and Hovland, 2007; Gay et al., 2012; tive purposes only and does not imply endorsement by
Kluesner et al., 2013; Brothers et al., 2014). Further- the U.S. government.
more, the integration of surface and water-column data
sets with 3D seismic data could provide a more robust References
analysis and evaluation of the accuracy of the 3D seis-
Al-Dossary, S., and K. J. Marfurt, 2006, 3D volumetric multi-
mic attribute results.
spectral estimates of reflector curvature and rotation:
Geophysics, 71, no. 5, P41–P51, doi: 10.1190/1.2242449.
Conclusion
Aminzadeh, F., and P. F. M. de Groot, 2004, Soft computing
In this study, we present a poststack data condition-
for qualitative and quantitative seismic object and res-
ing and neural-network attribute workflow for the de-
ervoir property prediction. Part 1: Neural network ap-
tection of faults and potential fluid-pathways in a
high-resolution 3D P-Cable seismic volume collected plications: First Break, 22, 49–54.
across the HFZ. Structural calculations and structural Aminzadeh, F., and P. F. M. de Groot, 2005, A neural net-
filtering improved data quality by removing noise pre- work based seismic object detection technique: 85th
dominantly orientated in the inline and crossline direc- Annual International Meeting, SEG, Expanded Ab-
tions. Following data conditioning, a neural-network stracts, 775–778, doi: 10.1190/1.2144442.
fault attribute workflow has helped to delineate a com- Aminzadeh, F., and P. F. M. de Groot, 2006, Neural net-
plicated network of faults. Our results provided better works and other soft computing techniques with appli-
noise tolerance and suppression of unwanted seismic cations in the oil industry: EAGE Publications B.V.
discontinuities relative to single discontinuity attribute Bahorich, M. S., and S. L. Farmer, 1995, 3-D seismic discon-
approaches. Application of a neural-network chimney tinuity for faults and stratigraphic features: The coher-
attribute workflow provided additional insight into ence cube: The Leading Edge, 14, 1053–1058, doi: 10
probable fluid pathways through the 3D seismic vol- .1190/1.1437077.
ume. Visualizing the fault and chimney neural-network Berndt, C., 2005, Focused fluid flow in passive con-
attribute results using transparency color schemes al- tinental margins: Philosophical Transactions of the
lowed for enhanced visual interpretation of the relation- Royal Society of London A: Mathematical, Physical
ship between faulting and probable fluid pathways. and Engineering Sciences, 363, 2855–2871, doi: 10
Combining the maximum curvature attribute and .1098/rsta.2005.1666.
structural interpretations with the neural-network attri- Brookshire, B., and L. Scott, 2015, Reducing risk in off-
bute results provided additional insight, revealing a shore planning and development: GEO ExPro Maga-
strong relationship between deformation associated zine, 12, 72–74.
with strike-slip fault kinematics and the distribution
Brothers, D. S., C. Ruppel, J. W. Kluesner, U. S. ten Brink, J.
of probable fluid pathways. Our results suggest that
D. Chaytor, J. C. Hill, B. D. Andrews, and C. Flores,
transtension, subsidence, and formation of small sedi-
2014, Seabed fluid expulsion along upper slope and
mentary basins result in focused fluid migration along
discreet faults. In contrast, transpressional deformation outer shelf of the U.S. Atlantic continental margin: Geo-
along restraining bends and fault-convergence zones physical Research Letters, 41, 96–101, doi: 10.1002/
leads to the formation of structurally controlled traps, 2013GL058048.
localized reservoirs, and distributed updip fluid migra- Brouwer, F. C. G., D. Connolly, and K. M. Tingahl, 2011, A
tion. Strike-slip fault strands associated with these guide to the practical use of neural networks: 31st An-
transpressional zones cut through the structural traps, nual Gulf Coast Section Society of Sedimentary Geol-
providing leakage pathways toward the seafloor. It is ogy, 440–472.
likely that other strike-slip faults offshore California Brouwer, F. C. G., and A. Huck, 2011, An integrated work-
are characterized by similar relationships between fault flow to optimize discontinuity attributes for the imaging
kinematics and fluid migration. of faults: 31st Annual Gulf Coast Section Conference So-
ciety of Sedimentary Geology, 496–532.
Acknowledgments Chopra, S., and K. J. Marfurt, 2005, Seismic attributes — A
We thank PG&E for publicly releasing the 3D seismic historical perspective: Geophysics, 70, no. 5, 3SO–
data volume as part of the Central Coastal California 28SO, doi: 10.1190/1.2098670.

Interpretation / February 2016 SB145

 Chopra, S., and K. J. Marfurt, 2007, Seismic attributes for Heggland, R., 2005, Using gas chimney in seal integrity
prospect identification and reservoir characterization: analysis: A discussion based on case histories, in P.
SEG. Boult, and J. Kaldi, eds., Evaluating fault and cap rock
Chopra, S., and K. J. Marfurt, 2009, Detecting stratigraphic seals: AAPG Hedberg Series, 237–245.
features via crossplotting of seismic discontinuity Hein, J. R, W. R. Normark, B. R. McIntyre, T. D. Lorenson,
attributes and their volume visualization: The Leading and C. L. Powell, 2006, Methanogenic calcite, 13C-
Edge, 28, 1422–1426, doi: 10.1190/1.3272695. depleted bivalve shells, and gas hydrate from a mud
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Chopra, S., and K. J. Marfurt, 2010, Integration of coher- volcano offshore southern California: Geology, 34,
ence and volumetric curvature images: The Leading 109–112, doi: 10.1130/G22098.1.
Edge, 29, 1092–1107, doi: 10.1190/1.3485770. Isaacs, C. M., K. A. Pisciotto, and R. E. Garrison, 1983, Fa-
Connolly, D. L., 2015, Visualization of vertical hydrocarbon cies and diagenesis of the Miocene Monterey Forma-
migration in seismic data: Case studies from the Dutch tion, California: A summary: Developments in Sedi-
North Sea: Interpretation, 3, SX21–SX27, doi: 10.1190/ mentology, 36, 247–282, doi: 10.1016/S0070-4571(08)
INT-2015-0007.1. 70094-3.
Connolly, D. L., and R. Garcia, 2012, Tracking hydrocarbon Johnson, S. Y., S. R. Hartwell, and P. Dartnell, 2014, Offset
seepage in Argentina’s Neuquen basin: World Oil, July of latest Pleistocene shoreface reveals slip rate on the
2012, 101–104. Hosgri strike-slip fault, offshore Central California: Bul-
Dickinson, W. R., M. Ducea, L. I. Rosenberg, G. H. Greene, letin of the Seismological Society of America, 104,
S. A. Graham, J. C. Clark, G. E. Weber, S. Kiddler, 1650–1662, doi: 10.1785/0120130257.
W. G. Ernst, and E. E. Brabb, 2005, Net dextral slip, Neo- Johnson, S. Y., and J. T. Watt, 2012, Influence of fault trend,
gene San Gregorio-Hosgri Fault Zone, coastal Califor- bends, and convergence on shallow structure and geo-
nia: Geological evidence and tectonic implications: morphology of the Hosgri strike-slip fault, offshore cen-
Geological Society and America Special Paper 391, 1–43. tral California: Geosphere, 8, 1632–1656, doi: 10.1130/
Dupré, S., C. Scalabrin, C. Grall, J. Augustin, P. Henry, A. GES00830.1.
M. Şengör, N. Görü, M. Çağatay, and L. Géli, 2015, Tec- Judd, A., and M. Hovland, 2007, Seabed fluid flow: The im-
tonic and sedimentary controls on widespread gas emis- pact on geology biology and the marine environment:
sions in the Sea of Marmara: Results from systematic, Cambridge University Press.
shipborne multibeam echo sounder water column imag- Kluesner, J. W., E. Silver, S. Nale, N. Bangs, and K. McIn-
ing: Journal of Geophysical Research: Solid Earth, 120, tosh, 2013, Fluid migration along a dense, intersecting
2891–2912, doi: 10.1002/2014JB011617. array of faults on the outer-shelf of Southern Costa Rica:
Ebuna, D. R., T. J. Mitchell, P. J. Hogan, S. Nishenko, and Insights from 3D seismic attributes and multibeam data:
H. G. Greene, 2013, High-resolution offshore 3D seismic Mineralogical Magazine, 77, 1480, doi: 10.1180/minmag
geophysical studies of infrastructure geohazards: 26th .2013.077.5.11.
Symposium on the Application of Geophysics to Engi- Lettis, W. R., K. L. Hansen, J. R. Unruh, M. K. McLaren, and
neering and Environmental Problems, 311–320. W. U. Savage, 2004, Quaternary tectonic setting of
Fugro Consultants Inc., 2012, Software validation of south-central coastal California: U.S. Geological Survey
Uniseis and 3D data qualification of 2010–2011 high- Bulletin 1995-AA.
resolution marine survey data, Offshore Diablo Canyon Ligtenberg, J. H., 2005, Detection of fluid migration
Power Plant, Central Coastal California Seismic Imag- pathways in seismic data: Implications for fault seal
ing Project: Report No. PGEQ-PR-03 (Rev0), FSI Project analysis: Basin Research, 17, 141–153, doi: 10.1111/j
No. 2011-4493, 1–67. .1365-2117.2005.00258.x.
Gay, A., R. Mourgues, C. Berndt, D. Bureau, S. Planke, D. Løseth, H., M. Gading, and L. Wensaas, 2009, Hydrocarbon
Laurent, S. Gautier, C. Lauer, and D. Loggia, 2012, leakage interpreted on seismic data: Marine and Petro-
Anatomy of a fluid pipe in the Norway Basin: Initiation, leum Geology, 26, 1304–1319, doi: 10.1016/j.marpetgeo
propagation and 3D shape: Marine Geology, 332, 75–88, .2008.09.008.
doi: 10.1016/j.margeo.2012.08.010. Maloney, J. M., B. M. Grupe, A. L. Pasulka, K. S. Dawson, D.
Gersztenkorn, A., and K. J. Marfurt, 1999, Eigenstructure- H. Case, C. A. Frieder, L. A. Levin, and N. W. Driscoll,
based coherence computations as an aid to 3-D struc- 2015, Transpressional segment boundaries in strike-
tural and stratigraphic mapping: Geophysics, 64, slip fault systems offshore southern California: Implica-
1468–1479, doi: 10.1190/1.1444651. tions for fluid expulsion and cold seep habitats: Geo-
Hansen, K. L., W. R. Letis, M. K. McLaren, W. U. Savage, physical Research Letters, 42, 4080–4088, doi: 10
and N. T. Hall, 2004, Style and rate of Quaternary defor- .1002/2015GL063778.
mation of the Hosgri Fault Zone, offshore south-central Mann, P., 2007, Global catalogue, classification and tec-
California: U.S. Geological Survey Bulletin 1995-BB, tonic origins of restraining- and releasing bends on ac-
1–33. tive and ancient strike-slip fault systems: Geological

SB146 Interpretation / February 2016

 Society of London, Special Publications, 290, 13–142, Pacific Gas and Electric Company (PG&E), 2014, Offshore
doi: 10.1144/SP290.2. low-energy seismic reflection studies in Estero Bay, San
Marfurt, K. J., 2014, Seismic attributes and the road ahead: Luis Obispo Bay, and Point Sal Areas: PG&E Tech-
84th Annual International Meeting, SEG, Expanded Ab- nical Report, GEO.DCPP.TR.14.02, http://www.pge.com/
stracts, 4421–4425, doi: 10.1190/segam2014-1644.1. includes/docs/pdfs/safety/systemworks/dcpp/report/Ch3.
Marfurt, K. J., 2015, Techniques and best practices in multi- GEO.DCPP.TR.14.02_R0_txt.w.ITR.pdf.
attribute display: Interpretation, 3, B1–B23, doi: 10 Paull, C. K., W. R. Normark, W. Ussler, D. W. Caress, and R.
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

.1190/INT-2014-0133.1. Keaton, 2008, Association among active seafloor defor-

Marfurt, K. J., and T. M. Alves, 2014, Pitfalls and limita- mation, mound formation, and gas hydrate growth and
tions in seismic attribute interpretation of tectonic accumulation within the seafloor of the Santa Monica
features: Interpretation, 3, SB5–SB15, doi: 10.1190/ Basin, offshore California: Marine Geology, 250, 258–
INT-2014-0122.1. 275, doi: 10.1016/j.margeo.2008.01.011.
Marfurt, K. J., R. L. Kirlin, S. L. Farmer, and M. S. Bahorich, Petersen, C. J., S. Bünz, S. Hustoft, J. Mienert, and D. Klae-
1998, 3-D seismic attributes using a semblance-based schen, 2010, High-resolution P-Cable 3D seismic imag-
coherency algorithm: Geophysics, 63, 1150–1165, doi: ing of gas chimney structures in gas hydrated sediments
10.1190/1.1444415. of an Arctic sediment drift: Marine and Petroleum Geol-
Marfurt, K. J., V. Sudhaker, A. Gersztenkorn, K. D. Craw- ogy, 27, 1981–1994, doi: 10.1016/j.marpetgeo.2010.06
ford, and S. E. Nissen, 1999, Coherency calculations in .006.
the presence of structural dip: Geophysics, 64, 104–111, Pisciotto, K. A., and R. E. Garrison, 1981, Lithofacies and
doi: 10.1190/1.1444508. depositional environments of the Monterey Formation,
McCaffrey, R., 2005, Block kinematics of the Pacific-North California, in R. E. Garrison, R. G. Douglas, K. E. Pis-
America plate boundary in the southwestern United ciotto, C. M. Isaacs, and J. C. Ingle, eds., The Monterey
States from inversion of GPS, seismological, and geo- Formation and related siliceous rocks of California:
logic data: Journal of Geophysical Research: Solid Pacific Section of the Society of Economic Paleontolo-
Earth, 110, 1–27, doi: 10.1029/2004JB003307. gists Mineralogists Symposium Special Publication, 97–
Meade, B. J., and B. H. Hager, 2005, Block models of 122.
crustal motion in southern California constrained by Roberts, A., 2001, Curvature attributes and their applica-
GPS measurements: Journal of Geophysical Research: tion to 3D interpreted horizons: First Break, 19, 85–
Solid Earth, 110, 1–19, doi: 10.1029/2004JB003209. 100, doi: 10.1046/j.0263-5046.2001.00142.x.
Means, W. D., 2012, Stress and strain: Basic concepts of Russell, B., D. Hampson, J. Schuelke, and J. Quirein, 1997,
continuum mechanics for geologists: Springer Science Multiattribute seismic analysis: The Leading Edge, 16,
& Business Media, 1–339. 1439–1444, doi: 10.1190/1.1437486.
Meldahl, P., R. Heggland, B. Bril, and P. F. M. de Groot, Sorlien, C. C., M. J. Kamerling, and D. Mayerson, 1999,
1999, The chimney cube, an example of semi-automated Block rotation and termination of the Hosgri strike-slip
detection of seismic objects by directive attributes and fault, California, from three-dimensional map restora-
neural networks. Part 1: Methodology: 79th Annual tion: Geology, 27, 1039–1042, doi: 10.1130/0091-7613
International Meeting, SEG, Expanded Abstracts, 931– (1999)027<1039:BRATOT>2.3.CO;2.
934, doi: 10.1190/1.1821262. Tingdahl, K. M., 1999, Improving seismic detectability us-
Meldahl, P., R. Heggland, B. Bril, and P. F. M. de Groot, ing intrinsic directionality: M.S. thesis, Earth Sciences
2001, Identifying fault and gas chimneys using multi- Centre, Göteborg University, paper no. B194.
attributes and neural networks: The Leading Edge, Tingdahl, K. M., Ø. Steen, P. Meldahl, and J. H. Ligtenberg,
20, 474–482, doi: 10.1190/1.1438976. 2001, Semi-automatic detection of faults in 3-D seismic
Mirkamali, M. S., N. Keshavarz Fk, and M. R. Bakhtiari, signals: 81st Annual International Meeting, SEG, Ex-
2013, Fault zone identification in the eastern part of panded Abstracts, 1953–1956, doi: 10.1190/1.1816520.
the Persian Gulf based on combined seismic attributes: Tingdahl, K. M., and P. F. M. de Groot, 2003, Poststack dip
Journal of Geophysics and Engineering, 10, 1–11, doi: and azimuth processing: Journal of Seismic Explora-
10.1088/1742-2132/10/1/015007. tion, 12, 113–126.
Nishenko, S., P. Hogan, and R. Kvitek, 2012, Seafloor map- Tingdahl, K. M., and M. de Rooij, 2005, Semi-automatic de-
ping for earthquake, tsunami hazard assessments: Sea tection of faults in 3D seismic data: Geophysical Pro-
Technology, 53, 15–20. specting, 53, 533–542, doi: 10.1111/j.1365-2478.2005
Ohl, D., and A. Raef, 2014, Rock formation characterization .00489.x.
for carbon dioxide geosequestration: 3D seismic ampli- Titus, S. J., M. Dyson, C. DeMets, B. Tikoff, F. Rolandone,
tude and coherency anomalies, and seismic petrophys- and R. Bürgmann, 2011, Geologic versus geodetic defor-
ical facies classification, Wellington and Anson-Bates mation adjacent to the San Andreas fault, central Cali-
Fields, Kansas, USA: Journal of Applied Geophysics, fornia: Geological Society of America Bulletin, 123,
103, 221–231, doi: 10.1016/j.jappgeo.2014.01.017. 794–820, doi: 10.1130/B30150.1.

Interpretation / February 2016 SB147

 Jared Kluesner received a B.S. Danny Brothers received a B.A.
(2006) from Indiana State University (2004) from the University of Colo-
and a Ph.D. (2011) from Scripps Insti- rado, Boulder, and a Ph.D. (2009)
tution of Oceanography. In 2012, he from the Scripps Institution of Ocean-
became a postdoctoral scholar at ography. He then moved to Woods
the University of California Santa Hole, MA, to work as a Mendenhall
Cruz (UCSC), where he conducted re- Postdoc Fellow, which started his ca-
search on the geophysical detection reer at the USGS. He is a research
Downloaded 02/17/16 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

and visualization of faults and fluid- geophysicist at the USGS in Santa

migration pathways along the Costa Rican margin. In Cruz, California. His research is primarily focused on
2014, he became a researcher at UCSC and led a co- two elements: (1) understanding the tectonics, sedimen-
operative project with the USGS Florida Water Science tary processes, and morphological evolution of continental
Center studying karst structures aquifer systems using margins and (2) the development of new approaches to
2D/3D seismic data and seismic attribute analysis. In quantify submarine earthquake, landslide, and tsunami
2015, he joined the USGS Pacific Coastal and Marine Sci- hazards in U.S. territory. His research involves a significant
ence Center in Santa Cruz, California. His current research amount of seagoing geophysical fieldwork, mostly in the
interests include swath sonar and 2D/3D seismic reflection form of high-resolution seismic reflection and swath
acquisition, processing, and visualization techniques. bathymetry methods.

SB148 Interpretation / February 2016