University of Southern California Dissertations and Theses

Multi-scale imaging of major fault zones in Southern California

(USC Thesis Other)

Multi-scale imaging of major fault zones in Southern California

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Multi-scale imaging of major fault zones in
Southern California

Pieter-Ewald Share

A Dissertation

Presented to the Faculty of the University of Southern California

Graduate School

In partial fulfillment of requirements for the degree of

Doctor of Philosophy

in

Geological Sciences

August 2018

1
Acknowledgements
First and foremost, I would like to thank my parents for their unwavering love and support
and the sacrifices they have made to help me achieve all that I can in life. From a young age they
taught me to be hard-working, diligent and persistent and to always believe in myself. Needless to
say, without these values I would not have been able to successfully complete a PhD. This
accomplishment belongs to them as much as it does to me.
To my advisor, Yehuda, I owe an enormous amount of gratitude. Throughout my PhD he
helped me stay positive and motivated despite several setbacks and external factors that could have
steered my focus away from work. From him I learned how to do science more concisely and
accurately and that solutions to most problems requires an iterative approach. I also want to thank
all my teachers and mentors throughout my years of schooling. Some of their names I have
ashamedly forgot, but the knowledge and experienced gained from them remains with me forever.
The list of people who have contributed to my PhD research is long and I appreciate and thank
them all for their inputs. The ones most directly involved in my work were: Yehuda Ben-Zion,
Frank Vernon, Zach Ross, Hongrui Qiu, Amir Allam, Fan-Chi Lin, Cliff Thurber, Haijiang Zhang,
Hao Guo, Tom Rockwell, Petr Taborik, Petra Stepancikova, Jakub Stemberk, Charlie Sammis,
Justin Haldar, David Okaya and John Platt. I have learned a lot from all of you. Thanks also to the
USC staff for making life for us students much less stressful. John McRaney, Tran Huynh, Deborah
Gormley, Vardui Ter-Simonian, Cindy Waite, Karen Young, Barbara Grubb, John Yu and Miguel
Rincon, you guys are great.
I was blessed to meet very early in the PhD my best friend and partner, Erin. Her and I have
been together through all the ups and downs of graduate student life and without her these last few
years would have been undoubtedly much duller. She is both a big critic and supporter of my work
and motivates me every day to be a better person. I am truly grateful for all her support.
Lastly, to my friends and family in the US and South Africa. None of this would have been
possible without you. A big thank you to every single one of you.

2
Table of Contents
Abstract ...................................................................................................................................... 3
Introduction ................................................................................................................................ 4
1. Seismic imaging of the Southern California plate-boundary around the South-Central
Transverse Ranges using double-difference tomography ............................................................. 7
2. Bimaterial interfaces in the South San Andreas Fault with opposite velocity contrasts NW and
SE from San Gorgonio Pass ...................................................................................................... 38
3. Internal structure of the San Jacinto fault zone at Blackburn Saddle from seismic data of a
linear array ............................................................................................................................... 48
4. Structural properties of the San Jacinto fault zone at Blackburn Saddle from seismic data of a
dense linear array ...................................................................................................................... 74
Discussion .............................................................................................................................. 101
A1. Chapter 1 supplementary figures ...................................................................................... 104
A2. Chapter 2 supplementary figures ...................................................................................... 112
A3. Sensitivity of fault zone head waves to the fault-normal distance of generating events ..... 114
A4. Chapter 4 supplementary figures ...................................................................................... 119
References .............................................................................................................................. 122

3
Abstract
Knowledge of material and geometric properties along major faults in Southern California is
critical for understanding the large earthquake potential and seismic hazard in the region. Using
different signals and techniques, I image these properties at multiple length-scales (100 km to 100
m) with a focus on the two most seismically active faults in the area, the San Andreas (SAF) and
San Jacinto (SJF) faults.
Analysis of fault zone head waves reveals long (~60 km), continuous segments of the central
SJF, and, the SAF northwest and southeast of San Gorgonio Pass (SGP) with across-fault VP
contrasts of 3-10%. High-resolution travel-time tomography shows distinct along-fault changes in
large-scale (>4 km) structure through SGP, potentially impeding large earthquakes propagating
through the region. It also reveals a northeast-dipping segment of the southern SAF around
Coachella Valley. Broad damage and deformation associated with the SJF and SAF manifest as
high VP/VS anomalies at shallow depth (<5km) and in some areas at greater depth (>10 km) as low
VP/VS anomalies. These variations give insight into the complex interplay between fault-associated
damage, crack geometry, fluid content and lithology near these faults. At smaller scale (~100 m-1
km), the internal structure of the SJF is imaged using characteristics of earthquake waveforms,
including arrival time variations and fault zone trapped waves recorded by dense linear arrays
crossing the fault. These analyses reveal 100-200 m wide and <5 km deep core damage zones
located predominantly northeast of the seismogenic SJF.
The determined fault geometries, bimaterial structures, damage asymmetry and deformation
can influence for example the rupture propagation direction and length (magnitude) of earthquakes
along the SAF and SJF. These results should be incorporated in future seismic hazard analysis to
improve predictions of ground motion from large earthquakes.

4
Introduction
Knowledge of material and geometric properties along major faults in Southern California is
critical for understanding the large earthquake potential and seismic hazard in the region. These
properties reveal information on past activity along faults. In addition, from them, characteristics
of future earthquakes can be inferred. Improved data acquisition and analysis methods combined
with faster and more effective computer resources, allow us to investigate these fault properties on
never-before-seen spatiotemporal scales. Using different signals and techniques, I image structural
properties at multiple length-scales (100 km to 100 m) with a focus on the two most seismically
active faults in the area, the San Andreas (SAF) and San Jacinto (SJF) faults.
Fault zones have various anomalous properties that make them appropriate targets for imaging
studies. In terms of elastic properties, faults can join dissimilar materials (bimaterial interface,
Ben-Zion, 1989; Ben-Zion and Malin, 1991) and therefore separate a “fast” seismic block on one
side from a “slow” seismic block on the other. Moreover, strain localization and earthquake
ruptures cause on- and off-fault damage. These damage zones, which also act as fluid conduits
(e.g., Hardebeck and Hauksson, 1999; Faulkner et al., 2010), usually have anomalously low
seismic velocities, and can vary in size from 10’s of km (e.g., Platt and Becker, 2010) to 10’s of
mm (e.g., Mitchell and Faulkner, 2009).
Local earthquakes near a bimaterial fault generate seismic head waves (FZHW) that travel
exclusively along the bimaterial interface and are recorded at the surface. FZHW contain crucial
information, such as the existence and continuity of large-scale fault interfaces. If a fault is
discontinuous, it limits the spatial extent of a potential earthquake and thus limits its magnitude.
The presence of bimaterial structures also gives insight into the likely propagation direction of an
earthquake rupture (e.g., Shi and Ben-Zion, 2006; Shlomai and Fineberg, 2016), which influences
the resulting amplitude and duration of ground shaking.
Within the broader damage structure, some faults also have cores of intense damage (Mitchell
and Faulkner, 2009; Perrin et al., 2016). Core damage zones coherent over large distances act as
trapping structures and near-fault local earthquakes generate trapped waves (FZTW) that
propagate within these structures (Peng et al., 2003). Analysis of FZTW helps quantify properties
such as the size, attenuation and reduced velocities of core damage zones. These properties provide
critical constraints on where and how earthquakes nucleate and where and why they terminate
5
(e.g., Huang and Ampuero, 2011). They also influence the amount of ground motion expected
from such earthquakes (Avallone et al., 2014).
Starting at the largest-scale and lowest resolution, I summarize in the following chapters
various imaging results for structures around the SAF and SJF. The study region for the work
contained in Chapters 1 and 2 is the South-Central Transverse Ranges (SCTR). This is a
structurally complex area where the SAF interacts with several other major faults, including the
SJF. A long-standing question is whether, despite complexity at the surface, the SAF is continuous
at depth through this region (e.g., Magistrale and Sanders, 1996; Yule, 2009). If continuous, the
potential for a large earthquake propagating along the southern San Andreas is higher (Porter et
al., 2011).
Chapter 1 documents results from a double-difference seismic tomography study. Travel times
between local earthquakes and surface stations/sensors of the regional seismic networks are used
as data during tomographic inversion. For a given earthquake-station pair, the travel time between
the two varies in response to the local seismic velocities. If velocities are locally slow, the travel
time is long, and, in contrast, the travel time is short if fast velocities exist between the earthquake
and station. Such variations in travel times are used to model near- and off-fault velocity anomalies
of length-scales >4 km within the SCTR (~2 km to ~17 km depth). The identified features are then
used to infer properties such as the dips of major faults, regions of most damage and deformation,
and, changes in lithology. Chapter 2 focuses on detecting and analyzing FZHW propagating along
the SAF within the SCTR. This analysis is used to inform large-scale bimaterial structures along,
and continuity of, the SAF. Large-scale structure and fault continuity then give insight into the
dynamics of potential earthquake rupture through the SCTR.
In Chapters 3 and 4 emphases are placed on imaging SJF structure. The SJF is the most
seismically active component of the plate boundary in southern California in recent times
(Hauksson et al., 2012) and has experienced several MW>7.0 earthquakes in the past 4,000 years
(Rockwell et al., 2015 and references therein). Similar to the SAF, the SJF is structurally complex
and highly segmented at the surface (Sanders and Magistrale, 1997).
Local earthquakes recorded by two dense linear arrays crossing the SJF are analyzed for
internal properties (~100 m to 1 km) of the SJFZ and for general larger scale features along its
length. The first array consists of 7 stations, has aperture of ~180 m and recorded continuously for
~18 months (Chapter 3). The second array consists of 134 sensors installed over a ~2 km distance
6
for a roughly month-long period (Chapter 4). Using state-of-the-art automatic seismic phase
detectors (Ross and Ben-Zion, 2014; Ross and Ben-Zion, 2015; Ross et al., 2016) in combinations
with visual inspection, I detect travel time variations, FZHW and FZTW recorded across these
arrays. Guided by the underlying physics, I then analyze and model these data to determine, for
example, velocity variations, attenuation, width and depth of the core and broader SJFZ damage
zone. In addition, I use FZHW to infer continuity at depth along the SJFZ and reveal how its
structural complexity propagates to great depth.
Following these chapters is a discussion on how the different results complement each other
and together give a more holistic understanding of structural properties of the Southern California
plate boundary. I conclude by highlighting potential directions for future research in the area.

7
1. Seismic imaging of the Southern California plate-boundary around the
South-Central Transverse Ranges using double-difference tomography
(Share et al., 2018a)

1.1 Introduction
The South-Central Transverse Ranges (SCTR) section of the plate boundary in Southern
California is highly complex and includes intersections between the San Andreas Fault (SAF), San
Jacinto Fault Zone (SJFZ) and Eastern California Shear Zone (ECSZ) (Fig. 1.1). Understanding
the geometry and continuity of fault structure through the region is key to assessing the potential
for large earthquake rupture along the southern SAF (Yule, 2009; Porter et al., 2011) and SJFZ
(Fialko, 2006; Lindsey and Fialko, 2013; Lozos, 2016). Geological mapping (e.g., Allen, 1957;
Sharp, 1967; Matti et al., 1992; Yule and Sieh, 2003) and paleoseismic surveys (e.g., McGill et
al., 2013; Rockwell et al., 2015 and references therein) show a variety of fault and earthquake
properties. The SAF for example has a significant discontinuity through San Gorgonio Pass (SGP),
where it steps 15 km to the left over a distance of 20 km (e.g., Allen, 1957; Matti et al., 1992) and
transfers strain to a network of dextral-normal and dextral-reverse faults (SGP fault zone, SGPFZ
and Banning/Garnet Hill fault, BF/GHF) south of the step-over (Fig. 1.1 bottom left inset). Fault
complexity at depth, such as the proposed northward dip of the SGPFZ and BF/GHF (e.g.,
Magistrale and Sanders, 1996; Yule and Sieh, 2003), is determined through analyses of earthquake
hypocenters (e.g., Jones et al., 1986; Magistrale and Sanders, 1996), focal mechanisms (Yang et
al., 2012) and potential field data (e.g., Fuis et al., 2012). Fault dip and continuity are also assessed
by analysis of fault zone head waves that refract along major fault bimaterial interfaces (Ben-Zion,
1989). Continuous segments of the SAF via the inactive Mission Creek fault (MCF, Share and
Ben-Zion, 2016) and the SJFZ (Share et al., 2017) through the Anza seismic gap (Sanders and
Kanamori, 1984) have been established through such analysis.
Tomographic images of P and S wave velocities (VP and VS) in volumetric elements within a
study area give regional context and insights on large-scale fault structures. Tomography is
effective at highlighting anomalous properties of the derived velocity models that can be associated
with major faults such as boundaries between different lithological units and damage zones. Fault
damage zones, which also act as fluid conduits (e.g., Ben-Zion and Sammis, 2003; Faulkner et al.,
2010), usually have anomalously low seismic velocities and VP/VS ratios. Tomography and studies
8

Figure 1.1. Location of the South-Central Transverse Ranges (SCTR) where the San Andreas fault
(SAF), San Jacinto fault zone (SJFZ) and Eastern California Shear Zone (ECSZ) intersect. Shown
for reference is Cajon Pass (CP), Elsinore fault (EF) and the town of Palm Springs (PS). Depicted
by circles are local earthquakes (MW>1) from 01/01/10 to 06/30/2015 (Hauksson et al., 2012).
Small black dots are all events in the area and colored larger circles and red triangles are the 12,674
events and 259 stations used during inversion, respectively. The inversion grid consists of nodes
space 1 km apart in the center (small blue box) and down to 15 km depth and node spacing as large
as 2 km outside that region (large blue box). Beige rectangle highlights San Gorgonio Pass (SGP)
at the center of the SCTR and the bottom left inset shows a zoom in of this area with its complex
network of faults. SBB=San Bernardino Basin; CHFZ=Crafton Hills fault zone; SBSSAF=San
Bernardino strand of the San Andreas fault; SGPFZ=San Gorgonio Pass fault zone; mcF=Mill
Creek fault; MCF=Mission Creek fault; BF=Banning fault; GHF=Garnet Hill fault; PMF=Pinto
Mountain fault; CVSSAF=Coachella Valley strand of the San Andreas fault. The geological
basement blocks that make up the region from south to north are the Peninsular Ranges batholith
(PRB), the San Gabriel Mountains block (SGMB) and the San Bernardino Mountains block
(SBMB).
9
of fault zone head and trapped waves have been used to image bimaterial interfaces and localized
damage zones along several faults in California (e.g., Ben-Zion and Malin, 1991; Li et al., 1994;
McGuire and Ben-Zion, 2005; Thurber et al., 2006; Lewis and Ben-Zion, 2010; Bennington et al.,
2013) and elsewhere (e.g., Mizuno and Nishigami, 2006; Calderoni et al., 2012; Najdahmadi et
al., 2016). Similar tools have been used to image fault structures in the southern California plate-
boundary including parts of the SCTR (e.g., Allam et al., 2014b; Zigone et al., 2015; Fang et al.,
2016; Share and Ben-Zion, 2016; Qiu et al., 2017; Share et al., 2017; Qin et al., 2018). Larger scale
velocity models of the entire Southern California (Tape et al., 2010; Lee et al., 2014; Barak et al.,
2015; Shaw et al., 2015) have too low resolution to capture such fault zone features.
In this study we image VP and VS velocity models around the SCTR, with a focus on SGP at
high-resolution (≥1 km), using double-difference tomography (DDT, Zhang and Thurber, 2003;
2006) based on arrival time data associated with local earthquakes. The use of an advanced
automatic detector (Ross and Ben-Zion, 2014; Ross et al., 2016) to pick the onset of P and S waves
allows for rapid extraction of large arrival time datasets from only a few years of data recorded in
recent years by high number of stations of the regional seismic networks. The detector produces
similar amounts of P and S picks (Ross et al., 2016), which leads to improved VS models in
particular and high-quality estimates of VP/VS in the area. We employ a new DDT algorithm that
incorporates absolute arrival times and both event-pair and station-pair double differences (DD)
(Guo and Zhang, 2017; Zhang et al., 2017). The combination of these different data allows for
resolving VP and V S structures from the near-surface where seismic stations are located (station-
pair DD), through regions with little seismicity (absolute data) and down to seismogenic depths
(event-pair DD). The algorithm allows for simultaneous relocation of earthquakes and during this
process event-pair DD determine high-resolution relative earthquake locations, while station-pair
DD allow for better absolute locations (Guo and Zhang, 2017).
The resulting velocity models provide improved high-resolution images of fault structures and
seismic properties in the southern California plate boundary around the SCTR. Incorporating these
velocity models in future calculations of seismic motion from large earthquakes in the area will
provide improved estimates of seismic shaking hazard (Porter et al., 2011 and references therein).
In the next section (1.2), known large-scale geology and tectonics of the SCTR are discussed in
some detail. Sections 1.3 and 1.4 describe DDT methodology and the data selection process,
respectively. Section 1.5 begins with a discussion on model resolution and continues with VP and
10
VS inversion results. At the end of section 1.5, prominent features in the resulting models are
highlighted. These features are interpreted and discussed in section 1.6.

1.2 Geology and seismotectonic setting
Figure 1.1 (bottom left inset) shows three geologic basement types (blocks) that exist in
distinctly different regions within the SCTR but complexly interact around SGP (Yule, 2009). The
Peninsular Ranges batholith (PRB), located south of the northern SJFZ, SGPFZ, BF/GHF and
Coachella Valley strand of the SAF (CVSSAF), consists of Jurassic and Cretaceous granitoid and
prebatholithic metamorphic rocks (Matti et al., 1992; Gutierrez et al., 2010). These rocks,
especially in the western PRB, have some of the highest densities (Langenheim et al., 2004) and
seismic velocities (Barak et al., 2015) in the SCTR. Located north of the San Bernardino strand of
the SAF (SBSSAF), MCF and CVSSAF is the San Bernardino Mountains block (SBMB), which
contains Cretaceous plutonic and Paleozoic to Proterozoic metamorphic rocks. The metamorphic
rocks are more prominent in the eastern SBMB (Gutierrez et al., 2010; Fuis et al., 2017). Also,
SBMB rocks adjacent to the SAF are characterized by high magnetic susceptibilities (Bankey et
al., 2002; Langenheim et al., 2005). The San Gabriel Mountains block (SGMB) is located between
the PRB and SBMB and is divided into lower and upper (outcrops mostly in the east) plates by the
Vincent Thrust (Matti et al., 1992). The lower plate consists of Paleogene to Cretaceous Pelona
schist whereas the upper plate is made up of plutonic (late Cretaceous to Triassic) and Proterozoic
crystalline rocks (Matti et al., 1992; Langenheim et al., 2005; Gutierrez et al., 2010). The lower
plate rocks have below average densities (Anderson et al., 2004; Langenheim et al., 2005) and low
seismic velocities (Allam and Ben-Zion, 2012), particularly within the San Bernardino Basin
(SBB, Anderson et al., 2004). These across-fault geological contrasts suggest major faults in the
area separate regions of different velocities and are therefore important targets for seismic imaging.
The oblique orientation of several active SAF strands (e.g., SBSSAF, SGPFZ and BF/GHF)
with respect to northwest relative plate motion (McGill et al., 2015) and the presence of inactive
ancestral strands of the SAF (e.g., western BF and MCF) drive structural complexity around SGP
and partitioning of strain to surrounding faults. The SBSSAF and CVSSAF only accommodate
limited portions of plate motion (<25 mm/yr, McGill et al., 2013; Gold et al., 2015) and are
proposed to dip northward (Daire and Cooke, 2009; Fuis et al., 2012; Lindsey and Fialko, 2013;
Barak et al., 2015) by as much as 37° and 65°, respectively (Fuis et al., 2012). Some of the
11
remaining strain is released by frequent low to moderate sized events (MW<5.0, Magistrale and
Sanders, 1996; Hauksson et al., 2012) and infrequent moderately large earthquakes (MW>5.0, Jones
et al., 1986; Nicholson, 1996) along the BF/GHF as well as uplift of the surrounding mountains
(Cooke and Daire, 2011). Structural complexity within SGP is highlighted by these diverse
earthquake focal mechanisms (Bailey et al., 2009; Yang et al., 2012) and variable stress drops
(Goebel et al., 2015).
The ECSZ (Spinler et al., 2010) and SJFZ (e.g., Meade and Yager, 2005; Dair and Cooke,
2009) also accommodate significant amounts of plate boundary related strain. The ECSZ has
relatively low average slip rates (<5 mm/yr, Oskin et al., 2008) but high rates of seismicity
(Hauksson et al., 2012), which is mostly concentrated along fault segments that ruptured during
the 1992 MW 6.1 Joshua Tree, 1992 MW 7.3 Landers and 1999 MW 7.1 Hector Mine earthquakes.
The SJFZ is the most seismically active component of the plate boundary in Southern California
in recent times (Hauksson et al., 2012; Ross et al., 2017a) and has experienced several MW>7.0
earthquakes in the past 4,000 years (Rockwell et al., 2015 and references therein). Compared to
the nearby southern SAF, the SJFZ has lower average slip rate (<20 mm/yr versus >20 mm/yr)
despite being more optimally aligned with present-day plate motion (e.g., Harden and Matti, 1989;
Meade and Yager, 2005; Blisniuk et al., 2013; Onderdonk et al., 2015).
The intricate fault structures and seismic activity distributed throughout the SCTR suggest
that extensive deformation and damage is present. High-resolution imaging and analysis of
velocity anomalies associated with these structural complexities will help map them and give
insight into fault zone properties in the region.

1.3 Double-difference tomography
The DDT inverse problem is solved iteratively using a nonlinear least squares approach,
where, per iteration, the approximate solution to a system of linearized equations is used to update
model parameters. This process is then repeated for a user-defined number of iterations. An
example linear equation corresponding to an absolute arrival time at station 𝑘 from event 𝑖 is
(Zhang and Thurber, 2003):
𝑟
$
%
= ∑
()
*
+
(,
-
+
∆𝑥
0
% 1
023
+∫ 𝛿𝑢𝑑𝑠
$
%
+∆𝜏
%
+𝑆
$
, (1.1)
12
where 𝑟
$
%
is the residual between observed and calculated arrival times, 𝑇 is travel time, 𝑥 and ∆𝑥
are hypocenter coordinates and their perturbations, 𝜏 is origin time, 𝛿𝑢 is perturbation to slowness,
𝑑𝑠 is an element of path length, and 𝑆 is station correction, respectively. Station corrections are
added to help compensate for unmodeled slowness structure near the surface. Equations for
residual 𝑟
$
=
corresponding to the same station 𝑘 (as in Equation 1.1) but different event 𝑗 and
residual 𝑟
?
%
for the same event 𝑖 (as in Equation 1.1) but different station 𝑙 are:
𝑟
$
=
= ∑
()
*
A
(,
-
A
∆𝑥
0
=
1
023
+∫ 𝛿𝑢𝑑𝑠
$
=
+∆𝜏
=
+𝑆
$
, (1.2)
𝑟
?
%
= ∑
()
B
+
(,
-
+
∆𝑥
0
% 1
023
+∫ 𝛿𝑢𝑑𝑠
?
%
+∆𝜏
%
+𝑆
?
. (1.3)
By subtracting Equations 1.2 and 1.3 from Equation 1.1 we obtain, respectively:
𝑑
$
%,=
= 𝑟
$
%
−𝑟
$
=
= ∑ E
()
*
+
(,
-
+
∆𝑥
0
%
−
()
*
A
(,
-
A
∆𝑥
0
=
F
1
023
+∫ 𝛿𝑢𝑑𝑠
$
%
−∫ 𝛿𝑢𝑑𝑠
$
=
⋯
+ ∆𝜏
%
−∆𝜏
=
, (1.4)
𝑑
$,?
%
= 𝑟
$
%
−𝑟
?
%
= ∑ E
()
*
+
(,
-
+
−
()
B
+
(,
-
+
F∆𝑥
0
% 1
023
+∫ 𝛿𝑢𝑑𝑠
$
%
−∫ 𝛿𝑢𝑑𝑠
?
%
+𝑆
$
−𝑆
?
, (1.5)
where 𝑑
$
%,=
and 𝑑
$,?
%
are called event-pair DD (Waldhauser and Ellsworth, 2000; Zhang and Thurber,
2003) and station-pair DD (Guo et al., 2017; Zhang et al., 2017). Note that the station correction
terms and the event origin time terms can be canceled in event-pair DD and station-pair DD,
respectively.
Equations 1.1, 1.4 and 1.5 corresponding to all available data are weighted from 0 to 1 (based
on data quality and misfit per iteration) and constitute the rows of:
𝑾𝒅 = 𝑾𝑮𝒎, (1.6)
where 𝑾 is an 𝑁×𝑁 diagonal matrix containing the weights, 𝒅 is the 𝑁×1 residual vector, 𝒎
is the 𝑀×1 model perturbation vector and 𝑮 is the 𝑁×𝑀 matrix of Frechet derivatives. The 𝑾𝑮
matrix in Equation 1.6 is augmented with Tikhonov regularization operators with smoothing (𝜂)
and damping (𝜀) factors. Sharp spatial changes in slowness are penalized more for larger 𝜂 whereas
increasing 𝜀 leads to smaller changes in model parameters per iteration. Finally, the model
perturbation vector 𝒎 is solved using the LSQR algorithm that includes damping (Paige and
Sanders, 1982), and results after final iterations are reported in section 1.5.
The event-pair DD are sensitive to the relative locations between pairs of events

(Waldhauser
and Ellsworth, 2000)

and the velocity structure near the source region (Zhang and Thurber, 2003)
13
whereas station-pair DD can better resolve the absolute event locations (Guo and Zhang, 2017)

and the structure outside the source region.

1.4 Data selection
We choose a model space that has dimensions X=222 km, Y=164 km and Z=34 km (3 km
above to 31 km below sea level), is centered on SGP, and is rotated clockwise by 40° (Fig. 1.1).
Waveform data corresponding to 19,332 M>1 events occurring from 1/1/2010 - 6/30/2015 and
recorded by all available stations located within this space are downloaded from the Southern
California Earthquake Data Center (SCEDC, 2013) and ANZA-YN seismic network database
(Vernon, 1982; Vernon and Ben-Zion, 2010). Accelerometer and geophone data recorded by
available short-period and broadband channels are extracted in 70 s windows. The windows start
10 s before and end 60 s after the origin times reported in the relocated Southern California catalog
(Hauksson et al., 2012).
Next an automatic picking algorithm (Ross and Ben-Zion, 2014; Ross et al., 2016) is used to
detect the onset of P and S arrivals in the extracted waveforms. The algorithm uses short-term
average to long-term average detectors together with kurtosis- and skewness-based detectors to
identify and pick the onset times. The picker produces an SNR value and a kurtosis-based metric
(Ross et al., 2016) for P and S picks, respectively, which are used to select the highest quality pick
for each event-station combination and to assign a weight to the selected data point. Outlier picks
and redundant data are systematically removed through the application of the following steps:
1) We discard picks more than 1 s off from predicted arrival times, using the CVM-H15.1.0
velocity model (Shaw et al., 2015, hereinafter referred to as the Harvard model).
2) Only 1 P and 1 S arrival time is allowed for each event-station combination. Toward that
end, the P and S picks with the largest SNR and kurtosis-based metric, respectively, are
selected.
3) The model space is divided into 8 km
3
cubes, azimuths to observing stations are calculated
for events within each volume, and event-station combinations are randomly selected such
that there is a maximum of 7 P and 7 S arrival times per 10° azimuth.
4) In cases where 1 or 2 stations locate within a 10° bin only 4 P and 4 S data are allowed per
station. This step helps remove redundant data when, for example, an isolated station
records multiple events from a dense cluster.
14
Steps 1-2 produce 586,367 P and 590,421 S picks corresponding to 18,875 events and 262
stations. The choice of 3D model to remove outliers in step 1 is not crucial as using the CVM-
S4.26 (Lee et al., 2014) and Fang et al. (2016) velocity models (hereinafter referred to as SCEC
and Fang2016 models), for example, produce 88% and 90% of the same picks. The maximum
allowed data in step 3 is derived from the average number of stations located within an arbitrary
10° bin measured from the center of the model space. That is, 10°×(262/360°) ~7 stations. This
step limits the number of rays with similar azimuths and therefore homogenizes azimuthal ray
coverage through the model space, which is key to producing undistorted anomalies in travel time
inversions. It also ensures that VP and VS models have similar ray coverage, which facilitates later
comparison between the two and improves VP/VS model quality. Application of steps 3-4 reduces
the amount of data by ~55%. The remaining data (P=242,122 and S=243,445) associated with
12,674 events and 259 stations (Fig. 1.1) constitute the absolute arrival times incorporated during
inversion (Equation 1.1).
The criteria used to select event-pair DD from the absolute arrival times are summarized in
Table 1.1 (top) and are similar to those used by Waldhauser and Ellsworth (2000). These criteria
produce 469,278 P and 534,741 S event-pair data. Station-pair DD are selected from absolute
arrival times in a similar fashion with some control parameters including the minimum and
maximum inter-station offset, minimum inter-station epicentral distance, minimum and maximum
number of station-pair observations per event, and the maximum number of neighboring stations
attached to a single station (Guo and Zhang, 2017). The values used during the station-pair DD
selection process appear in Table 1.1 (bottom). The total selected P and S station-pair data are
816,976 and 837,337.

1.5 Results
We discretize the velocity model such that nodes are 1 km apart in a 142 km by 72 km area
centered on SGP and down to 15 km depth and as much as 2 km apart outside that region (Figs.
1.1 and A1.1). The inversion uses 20 iterations to ensure satisfactory convergence of the final
result. During the first 10 iterations emphasis is placed on fitting the absolute data whereas during
the last 10 iterations emphasis is placed on DD data. Both sets of 10 iterations consist of 7 joint
inversions of velocity and location, and 3 inversions of relocation only. In the following section
15

(1.5.1) we test the resolution of structure in different parts of the model space by inverting synthetic
data using the described settings. In section 1.5.2 we invert the observed seismic data.

1.5.1 Model resolution
Model resolution is assessed first with the well-known checkerboard test where synthetic data
generated from a regular grid of positive (above average) and negative (below average) anomalies
are inverted and the resultant model (recovered model) is compared to the checkerboard model.
Resolution is said to be good in areas where the two models agree and poor where they are not
sufficiently similar. This test is useful but amongst several limitations (Lévêque et al., 1993) there
is no one-to-one mapping between resolution in the recovered model and the model obtained by
inverting actual data (inverted model). This is because the two models have different velocities
and therefore rays associated with the same event-station combinations do not necessarily sample
the same parts of the model space. Given this, we additionally analyze model quality using derivate
weighted sums (DWS, Thurber and Eberhart-Phillips, 1999) and the diagonal of the resolution
Table 1.1. Selection criteria for event-pair (top) and station-pair (bottom) DD from arrival time
data.
Event-pair selection criteria
Maximum distance between even-pair and station 100 km
Maximum hypocentral separation between events 5 km
Maximum number of neighbors per event 10
Minimum number of links required to define a neighbor 8
Minimum number of event-pair observations per station 8
Maximum number of event-pair observations per station 30
Station-pair selection criteria
Minimum distance between stations per event 2 km
Maximum distance between stations per event 100 km
Minimum epicentral distance difference between stations 2 km
Maximum number of neighbors per station per event 40
Minimum number of station-pair observations per event 4
Maximum number of station-pair observations per event 200

16
matrix (e.g., Rawlinson et al., 2014), which are both more direct indicators of resolution in the
inverted model.
To test the various length scales resolvable with the available data, we use checkerboard
models with anomalies that extend 2 to 10 grid nodes in X, Y and Z directions. These anomalies
alternate between 4.75 km/s and 5.25 km/s (5 km/s +/-5%) in the VP models, and those values
divided by 1.73 (average regional VP/VS ratio) in the VS models. Checkerboard models with
anomalies spanning 2 and 6 grid nodes are shown at 10 km depth in Figs. 1.2a-b for VP and 1.3a-
b for VS, and in cross-section at X=8 km in Figs. 1.4a-b for VP and 1.5a-b for VS. Synthetic travel
times corresponding to the data discussed in section 1.4 are extracted from the checkerboard
models. DD are then calculated from the travel times using the parameters in Table 1.1. Next, we
give all synthetic data a weight of 1 and invert them using homogeneous half-space starting models
of 5 km/s for VP and 2.89 km/s for VS (5 km/s divided by 1.73). To recover as best possible small-
scale anomalies and in general resolve sharp velocity contrasts we set a small smoothing parameter
𝜂=1. We then test different damping parameters (𝜀) and choose one that ensures the ratio between
largest and smallest eigenvalues (condition number) of the regularized 𝑾𝑮 matrix (Equation 1.6)
lies between 100-200. It is important to keep this range low but not too narrow, because the true
condition number is different, and likely lower, than the one calculated here with LSQR (Zhang
and Thurber, 2007). We found that using 𝜀=1400 and 1450 during the first and second sets of 7
joint inversions, respectively, and 𝜀=200 during relocation only inversion, produces condition
numbers in the specified range. The results show that with synthetic data we can resolve
checkerboard anomalies spanning as few as 2 grid nodes over a broad region (Figs. 1.2-1.5c).
To facilitate comparison with the inverted models, we repeat this process using the same
synthetic data, but the regularization parameters used in the inversion of actual data (see section
1.5.2). The larger smoothing applied (𝜂=45) results in poorer resolution of small-scale velocity
structures (not shown) and increased misfit. Specifically, we find that only checkerboard
anomalies spanning >4 grid nodes are sufficiently recovered in regions with adequate ray
coverage. As examples, results corresponding to 6 grid node checkerboard models are shown in
Figs. 1.2-1.5d.
Arrival time uncertainty is greater in the present study compared to similar earlier studies
(e.g., Allam and Ben-Zion, 2012), because an automated detector instead of a trained analyst is
used to pick arrivals. To test the resolvable length scales when data contain errors, we invert the
17

Figure 1.2. Checkerboard recovery results at 10 km depth (VP). Checkerboard model used to
generate synthetic data consisting of +/-5% anomalies spanning 2 (a) and 6 (b) grid nodes. c)-d)
Recovered models corresponding to checkerboard models in (a) and (b) and noise-free synthetic
data. e)-f) Recovered models corresponding to checkerboard models in (a) and (b) and noisy data.
More smoothing is applied during inversion of (d) and (f). White contours depict DWS=20 and
magenta contours show !
"
=0.008.

18
same synthetic data again but now with noise added. The amount of noise added to synthetic travel
time data is based on the picking errors normally produced by the automatic detector. The
STA/LTA detector used to pick P arrivals (Ross and Ben-Zion, 2014) performs similar to other
STA/LTA detection algorithms (e.g., Gentili and Michelini, 2006; Nippress et al., 2010) where the
Figure 1.3. Similar to Fig. 1.2 but for VS checkerboard test at 10 km depth.

19
distribution of the difference between “true” and automatic picks has a standard deviation of ~0.1
s. In terms of S arrivals, the employed detector produces automatic picks that are within 0.16 s of
manual picks 75% (1.15 standard deviations for normal distribution) of the time (Ross et al., 2016).
Figure 1.4. Checkerboard recovery results in cross-section at X=8 km (VP). Checkerboard model
used to generate synthetic data consisting of +/-5% anomalies spanning 2 (a) and 6 (b) grid nodes.
c)-d) Recovered models corresponding to checkerboard models in (a) and (b) and noise-free
synthetic data. e)-f) Recovered models corresponding to checkerboard models in (a) and (b) and
noisy data. More smoothing is applied during inversion of (d) and (f). White contours depict
DWS=20 and magenta contours show !
"
=0.008.

Figure 1.5. Similar to Fig. 1.4 but for VS checkerboard test at X=8 km.

20
Subsequently, zero-mean Gaussian noise with standard deviations equal to 0.1 s and 0.14 s (0.16
s divided by 1.15) are added to the synthetic P and S travel time data. In addition, we assign lower
weights to data with more noise added. To simultaneously test how uncertainty in event location
affects structural resolution, we displace hypocenters by amounts equal to the absolute relocation
uncertainties in the Hauksson et al. (2012) catalog. That is, for 90% (1.64 standard deviations for
normal distribution) of the events in the catalog, the horizontal and vertical uncertainties are less
than 0.8 km and 1.5 km. Therefore, horizontal and vertical displacements sampled from zero-mean
Gaussian distributions with standard deviations of 0.5 km and 0.9 km (0.8 km and 1.5 km divided
by 1.64) are added to event hypocenters.
Noisy synthetic data are inverted with the same regularization parameters as before and the
resulting models are compared with those obtained from inverting noise-free synthetic data. First,
we invert using 𝜂=1 and 𝜀=1400 and 1450 during the first and second set of 7 joint inversions,
respectively, and 𝜀=200 during relocation only inversion. The results show these regularization
parameters are inappropriate to use when data contain noise, because they lead to distorted and
incoherent recovered models (compare Figs. 1.2-1.5c with 1.2-1.5e). We subsequently invert the
noisy data using settings that include more smoothing (𝜂=45, see section 1.5.2). In addition, to
suppress the unwanted effects of noise, outliers are removed per iteration by only allowing data
residuals within a certain number of median absolute deviations (MAD) from the median to be
inverted. The number of allowed MAD is gradually decreased from 10 in the first iteration to 3 in
the last iteration. Medians and MAD are calculated independently for P and S data per iteration as
these data have characteristically different errors given the employed detectors. This stricter
regularization and outlier removal process leads to more coherent anomalies being recovered. As
with synthetic data, acceptable recovery is achieved for anomalies spanning >4 grid nodes (Figs.
1.2-1.5f). However, the anomalies are slightly distorted due to the influence of noise and the
recovered area is smaller than before (compare Figs. 1.2-1.5d and 1.2-1.5f). The latter mostly
relates to lower data weights and the removal of outliers per iteration during noisy data inversion.
To further quantify model resolution, we compare the checkerboard results with DWS values
computed per grid node for each model. These values are calculated by summing the 𝑾𝑮 matrix
elements (prior to regularization, Equation 1.6) column-wise. They therefore represent the sum of
all fractional ray lengths (times data weight) in the vicinity of a given grid node. A comparison of
DWS and checkerboard results shows the two are complementary. Specifically, DWS=20 values
21
(white contours in Figs. 1.2-1.5) enclose regions where checkerboard anomalies are well recovered
and appear in subsequent figures. Generally, near the edges of these regions checkerboard
anomalies are only partially recovered (<50% of original amplitudes), whereas at the center
recovered amplitudes are >80% of the original checkerboard amplitudes. For nodes where both VP
and VS models have DWS≥20 the mean and standard deviation of recovered VP/VS values are 1.73
+/-0.01 (noise-free synthetic data) and 1.73 +/-0.02 (noisy synthetic data) (dark blue lines in Figs.
A1.2b-c and A1.3b-c). These means are equal to the true checkerboard model VP/VS (Figs. A1.2a
and A1.3a). Model deviations from 1.73 relate to small-scale (≤3 km) VP/VS anomaly artifacts
caused by different resolutions of the P and S data. These artifacts become especially pronounced
when noisy data are inverted (Figs. A1.2c and A1.3c). To suppress the small-scale artifacts in the
VP/VS model inverted using real data (next section), we apply a 3 by 3 median filter to the depth
slices and cross-section presented. This inevitably decreases the resolution in VP/VS models but
helps avoid incorrect interpretations due to model artifacts. Figures A1.2d and A1.3d demonstrate
the effects of applying a median filter to a model containing small-scale artifacts (Figs. A1.2c and
A1.3c).
In general, DWS is an effective tool for indicating regions of acceptable resolution. However,
structures with similar DWS (and ≥20) are not necessarily equally well resolved. For example, at
10 km depth the model space around Palm Springs (X=50 to 70 km, Y=-20 to 0 km) is not as well
resolved as around the SJFZ trifurcation area (X=70 to 110 km, Y=-30 to -10 km), while both
regions are at the edge of acceptable ray coverage (bordered by DWS=20 contours in Figs. 1.2d
and f and 1.3d and f). One reason for this is that DWS relates to ray density but is insensitive to
the azimuthal distribution of those rays. As a consequence, if a region has large DWS but all
through-going rays have similar azimuths, then imaged anomalies in the region will be distorted
and thus poorly resolved.
We use the resolution matrix to quantify this variability in resolution. Using the notations in
section 1.3, the 𝑁×𝑁 resolution matrix 𝑹 is defined as:
𝑹 = (𝑮
)
𝑾
U3
𝑮+𝜂
V
𝑫
)
𝑫+𝜀
V
𝑰)
U3
𝑮
)
𝑾
U3
𝑮, (1.7)
where 𝑫 is the finite difference approximation of the first spatial derivative of slowness and 𝑰 is
the identity matrix. Present-day state-of-the-art algorithms and computers are sufficient to
calculate the full 𝑹 (Bogiatsiz et al., 2016), however, to save on time and computational cost we
use a stochastic method of estimating the resolution matrix diagonal only (Bekas et al., 2007;
22
MacCarthy et al., 2011). This technique is one of several (e.g., An, 2012; Trampert et al., 2013;
Fichtner and van Leeuwen, 2015) that utilize random vectors to quantify model resolution.
Consider a sequence of 𝑠 random vectors, 𝒗
𝟏
, …, 𝒗
𝒔
, each 𝑁×1 in size, sampled from a standard
normal distribution. An estimate of the diagonal 𝒓
𝒔
of the resolution matrix 𝑹 is (MacCarthy et al.,
2011):
𝒓
𝒔
= [∑ 𝒗
𝒌
`
$23
.×𝑹𝒗
𝒌
].∕[∑ 𝒗
𝒌
.×𝒗
𝒌
`
$23
], (1.8)
where .× and .∕ denote element-wise vector multiplication and division, respectively. To compute
𝒓
𝒔
we therefore need to be able to generate random vectors, which is trivial, and the product 𝑹𝒗
𝒌

of the resolution matrix and the random vectors. The latter is obtained by multiplying Equation 1.7
with 𝒗
𝒌
, and is given by:

𝑹𝒗
𝒌
= (𝑮
)
𝑾
U3
𝑮+𝜂
V
𝑫
)
𝑫+𝜀
V
𝑰)
U3
𝑮
)
𝑾
U3
𝑮𝒗
𝒌
, (1.9)
which is simply the normal equations solution for
𝑚𝑖𝑛fg
𝑾𝑮
𝜂𝑫
𝜀𝑰
h𝑹𝒗
𝒌
−i
𝑾𝑮𝒗
𝒌
𝟎
kf
V
, (1.10)
and can be solved using LSQR. As 𝑠 becomes large, 𝒓
𝒔
approaches the true diagonal of the
resolution matrix 𝑹. To ensure satisfactory convergence of 𝒓
𝒔
we use the upper-bound of 𝑠=512
suggested by MacCarthy et al. (2011) and run 20 realizations to compute 20 independent estimates
of 𝒓
𝒔
, taking in the end the median of those estimates to produce a single robust value for 𝒓
𝒔
.
We compute the resolution matrix diagonal in this manner for the recovered models with
checkerboard anomalies spanning >4 grid nodes. This is done for the regularized 𝑾𝑮 matrix
generated during the final iteration. Comparisons of recovered models and associated resolution
values (𝑟
`
) show that during final iteration regions with 𝑟
`
≥0.008 have the best-resolved
checkerboard anomalies. Figures 1.2d and f and 1.5d and f and A1.2b and c and A1.3b and c show
𝑟
`
=0.008 contours for the recovered models with 6 grid node checkerboard anomalies (magenta
lines). 𝑟
`
values calculated here are relatively low compared to earlier studies using DDT (e.g.,
Zhang and Thurber, 2007; Zhang et al., 2017), and this stems from the large 𝜂 and 𝜀 employed to
suppress the effects of data noise. However, the checkerboard tests show that after 20 iterations
anomalies are sufficiently recovered despite relatively low 𝑟
`
per iteration. Moreover, changes in
𝑟
`
highlight well the variability in resolution here, which was the original purpose of calculating
the resolution matrix diagonals. Recovered anomalies within regions with 𝑟
`
≥0.008 have
23
amplitudes >50% of the original checkerboard anomalies with a maximum recovery of ~90%.
Outside these regions there is a rapid decrease in these percentages.
To summarize, the checkerboard tests indicate that when using the same regularization as in
the inversion of actual data (section 1.5.2), we can resolve structures >4 km. Moreover, analyses
of DWS and resolution matrix diagonals reveal that >4 km anomalies are only well constrained if
they also have associated DWS≥20 and are best resolved if they have 𝑟
`
≥0.008. Lastly,
regularization inevitably produces recovered VP and VS anomalies that are lower in amplitude
when compared to the checkerboard models from which data are generated. However, original
VP/VS values are well recovered, because P and S datasets are similar in coverage and size, and
the same regularization is applied to both datasets. In the next section (1.5.2) and section 1.6 these
findings will be used as guidelines when interpreting the inverted model.

1.5.2 Inversion of observed data
Starting with the Harvard model (Shaw et al., 2015) shown in Fig. A1.4a at 13 km depth, we
systematically test different combinations of 𝜂 and 𝜀 to find values that produce inverted models
containing coherent and geologically feasible features. Simultaneously, we require the condition
number to be around 100 for all iterations. This is near the lower bound of the range used during
inversion of synthetic data and is done to ensure that data uncertainty does not significantly
diminish model quality. Similar to the checkerboard tests using noisy data, we remove outliers per
iteration by discarding data more than 10 MAD from the model median during the first iteration,
and gradually decrease this number to 3 during the last iteration. After extensive testing, we obtain
satisfactory results for regularization parameters around 𝜂=45 and 𝜀=800 and 900 during the first
and second set of 7 joint inversions, respectively, and 𝜀=300 during relocation only inversion. The
initial data RMS for P and S arrival times (0.29 s and 0.42 s) are reduced to P=0.10 s and S=0.14
s after 20 iterations (Fig. 1.6a). The latter are equal to the expected uncertainties in P and S arrival
time picks (section 1.5.1), providing confidence that the chosen regularization parameters do not
lead to significant over-fitting of the data during inversion. Figure 1.6b shows the inversion results
for VP and VS structures at 13 km depth.
Next, we illustrate the obtained results are independent of the starting model by repeating the
same data selection and inversion processes but starting instead with the SCEC (Lee et al., 2014)
and Fang2016 (Fang et al., 2016) models. Figures A1.4b and c show these models at 13 km depth.
24

Figure 1.6. Inversion results at 13 km depth for different starting models. a) Initial (transparent)
and final (not transparent) P (blue) and S (red) model misfits for the Harvard model. b) Inverted
V P (left) and V S (right) structure using the Harvard model as starting model. White contours depict
DWS=20. c)-d) Same layout but for the SCEC model. e)-f) Same layout but for the Fang2016
model.

25
After applying the selection criteria in section 1.4, the numbers of arrival time data to be used
during inversion are 93.2% (SCEC) and 93.1% (Fang2016) of those described in section 1.4.
Therefore, according to our criteria, more data are considered outliers relative to the SCEC and
Fang2016 models compared to the Harvard model. Related to this is the relatively large initial data
RMS for P and S data associated with the SCEC and Fang2016 models, namely, P=0.44 s and
S=0.52 s (Fig. 1.6c) and P=0.43 s and S=0.56 s (Fig. 1.6e), respectively. From these data, we
extract DD using the criteria in Table 1.1. Then, following the same inversion procedure as before
(Fig. 1.6b), we invert the respective data, and after 20 iterations their arrival time misfits decrease
to P=0.11 s and S=0.15 s (SCEC, Fig. 1.6c) and P=0.09 s and S=0.13 s (Fang2016, Fig. 1.6e).
Figures 1.6d and f contain, respectively, updates to the SCEC and Fang2016 models at 13 km
depth following inversion. Generally, in regions with adequate ray coverage (DWS≥20), structures
correlate well between the results using different starting models (compare Figs. 1.6b, d and f).
We compare the depth-dependent variations between the different models by exploring their
respective average velocities at each depth (1D models) pre- and post-inversion. Figure 1.7 shows
these profiles (P=blue; S=red) before (a) and after (b) inversion from 5-15 km depth. This depth
range is chosen because it is where we have greatest sensitivity given the data and methodology
employed. The Harvard and Fang2016 profiles prior to inversion are similar, but both differ from
the SCEC profiles, which generally have higher velocities in the study region and at the depths
presented (Figs. 1.7a and A1.4). Following inversions, the 1D profiles (calculated only for nodes
with DWS≥20) agree more. For example, there is a significant decrease in the velocities of the
SCEC profiles, especially in VP, as they start to converge to the Harvard and Fang2016 profiles.
The Harvard and Fang2016 profiles, in turn, are closer to each other still, and show a parallel
increasing trend in velocity with depth. These comparisons demonstrate that, where we have
acceptable ray coverage and resolution, the results converge to similar solutions irrespective of
starting model. Given this, and that it is best practice in linearized inversions to use a starting
model as close as possible to the final model, we choose to interpret features in results generated
with the Harvard model. This model has the lowest initial data RMS, and prominent features
observed in the inverted model also appear in results obtained from using the SCEC and Fang2016
models (compare Figs. 1.6b, d and f).
Figure 1.8 shows final inversion results (using the Harvard model) at various depths for the
entire model space, and relocated events are compared with locations from the Hauksson et al.
26

(2012) catalog in Fig. 1.9. The mean epicentral difference between the two sets of events is 0.08
+/-1.4 km and the mean difference in depth is 0.28 +/-2.3 km (relocated events are on average
deeper). The majority of relocated and catalog epicenters differ by small amounts (<0.5 km).
Moreover, apart from a few outlier events, large discrepancies (>1.0 km) in epicenters are only
observed for events near the edges of the model space where data constraints are weaker. This is
true for the differences in depth between the two sets of events also. The largest differences in
depth are associated with events near the surface or at the base of the seismogenic zone. If we
exclude these poorly relocated events, then the differences between the relocated and catalog
events are of the same order as the absolute uncertainty in the latter (see section 1.5.1). This means
the absolute locations in the Hauksson et al. (2012) catalog are of high quality and did not need to
be displaced by large amounts to achieve a sufficiently low misfit during inversion. Various cross-
sections through the 142 km by 72 km box centered on SGP (Fig. 1.1) appear in Figs. 1.10-1.12.
We list below several features of interest and number them accordingly in Figs. 1.8 and 1.10-1.12.
From shallow to greater depth, the observed large-scale features are:
1) At depths ≤5 km, the lowest velocities and highest VP/VS values (≥1.8) are associated with
the SBB, San Jacinto Basin (SJB) and Coachella Valley.

Figure 1.7. Average VP (blue) and VS (red) versus depth before (a) and after (b) inversion using
the Harvard (solid lines), SCEC (dotted lines) and Fang2016 (dashed lines) as starting models.

b a
579 11 13 15
Depth (km)
5.7
5.8
5.9
6
6.1
6.2
6.3
6.4
6.5
6.6
3.3
3.35
3.4
3.45
3.5
3.55
3.6
3.65
3.7
3.75
3.8
Pre-inversion
SCEC
Fang2016
Harvard
Vp (km/s)
579 11 13 15
Depth (km)
5.7
5.8
5.9
6
6.1
6.2
6.3
6.4
6.5
6.6
3.3
3.35
3.4
3.45
3.5
3.55
3.6
3.65
3.7
3.75
3.8
Vs (km/s)
Post-inversion
SCEC
Fang2016
Harvard
27

Figure 1.8. Final VP (left), VS (middle) and VP/VS (right) inversion results at depths of 4 km (a),
7 km (b), 10 km (c) and 14 km (d). Prominent features are highlighted and numbered (see text for
descriptions). White contours depict DWS=20 and regions with DWS<20 are shaded. Note the
color bars change with depth.
28
2) Faults such as the SBSSAF and PMF have similarly low velocities and high VP/VS values,
especially where they intersect with other faults. In contrast, the MCF and SGPFZ locate
within regions of average to high velocities and average VP/VS values.
3) Rocks of the PRB have the highest velocities, and at shallow depth, abut the southwest
sides of the SBB and SJB, forming prominent velocity contrasts across the northern SJFZ.
4) At depths ≥6 km, the transition from the western to eastern PRB manifests as a sharp
decrease in velocity from west to east. This contrast is more prominent in VP than in VS,
emphasized by a decrease in average VP/VS of 1.72 (west) to 1.69 (east). Also, the contrast
does not migrate with depth, showing that it is a vertical feature at least down to depths
where we have adequate resolution.
5) The lowest velocities in the eastern PRB are associated with the SJFZ and elongated fault-
normal features that extend to the Elsinore fault (EF). These elongated features are clearest
in the VP model at depths ≥6 km.
6) Rocks of the SGMB lower plate (outcrops mostly in west) and eastern SBMB have low
velocities and high VP/VS values (>1.75).
7) The SGMB upper plate and western SBMB have variable velocities but average to low
VP/VS values.
8) At greater depth, the velocity contrast across the northern SJFZ becomes confined to where
rocks of the western PRB and SGMB lower plate meet.
9) Rocks of the western PRB, which border the SJB at shallow depth (3), intersect the SJFZ
at greater depth (>7 km) and occupy the region known as the Hemet step-over (Marliyani
et al., 2013).
10) The lowest velocities around 10 km depth are associated with the eastern SBMB. East of
Palm Springs these low velocities undergo a sharp transition to fast velocities in the
southwest along a bimaterial contact. This contact extends for >50 km and is parallel to the
SAF but offset to the northeast by ~7 km.
11) At these same depths, low VP and low VP/VS anomalies relate to fault strands of the ECSZ
north of the PMF.
12) From 13 km downwards, low VP and VP/VS anomalies develop beneath the SGPFZ.
13) Low VP and the lowest VP/VS anomalies (~1.6) are associated with the SJFZ trifurcation
area at depths ≥10 km.
29
Focusing next on the region around SGP, we observe the following additional features:
14) At various depths the northern SJFZ and its intersection with the Crafton Hills fault zone
(CHFZ) are characterized by anomalously low VS and high VP/VS (≥1.78). These
anomalies extend vertically downwards and have no apparent dip.
15) The mapped surface trace of the SBSSAF near where it intersects with the MCF has
similarly low VS and high VP/VS at various depths. These anomalies are also near vertical.

Figure 1.9. Comparison between the Hauksson et al. (2012) catalog and relocated events. The
main plot shows differences in event epicenters with warmer colors representing larger differences.
The cross-sections depict differences in depth between the two sets of events projected along the
x-axis (bottom) and y-axis (right). The same color scale as in the main plot applies.

−118.00˚
−118.00˚
−117.50˚
−117.50˚
−117.00˚
−117.00˚
−116.50˚
−116.50˚
−116.00˚
−116.00˚
−115.50˚
−115.50˚
33.00˚ 33.00˚
33.50˚ 33.50˚
34.00˚ 34.00˚
34.50˚ 34.50˚
35.00˚ 35.00˚
0.0 0.5 1.0 1.5 2.0 2.5
location (km)
X−axis X−axis
Y −axis Y −axis
0 20 40 60
km
0
10
20
30
Depth (km)
−100 −80 −60 −40 −20 0 20 40 60 80 100 120
X distance (km)
0
10
20
30
Depth (km)
−60 −40 −20 0 20 40 60 80
Y distance (km)
30
16) Furthermore, the earlier mentioned deep low VP and VP/VS anomaly beneath the SGPFZ
(12) concentrates near its intersection with the SBSSAF.
17) In contrast, a high VP/VS anomaly (>1.76) with apparent northward dip is associated with
the BF/GHF system east of the SGPFZ.
18) At depths ≥13 km the Hot Springs area of the SJFZ is also characterized by high VP/VS
values.
19) Finally, the earlier mentioned deep low VP and VP/VS anomalies beneath the SJFZ
trifurcation area (13) concentrate beneath the Clark fault (CF) and Buckridge fault to the
northeast and southwest of the Coyote Creek fault.
All listed features have length-scales >4 km, acceptable resolution (DWS≥20) and for the
most part rs≥0.008 (Figs. A1.5-A1.8). In the following section we discuss these features and
explore their implications.

1.6 Discussion
We obtain detailed tomographic images of seismic velocities in the southern California plate
boundary around the SCTR using a newly developed DDT algorithm (Guo and Zhang, 2017;
Zhang et al., 2017) that incorporates station-pair DD with the absolute arrival times and event-pair
DD traditionally used. The station-pair data allow for structural resolution near the surface (≤2
km) and better absolute locations of relocated events. We use an automatic detection algorithm to
extract P and S arrival times from event waveforms. This allows for the rapid extraction of large
P and S datasets from only a few years of recordings (01/01/2010-06/30/2015) associated with a
large number of seismic stations. Using these data, we invert for high-resolution VP and VS models
and derive from them high quality VP/VS models. These models contain various interesting
features that are discussed below in light of what they reveal about fault zone structures. We note
in parenthesis and in italics the feature number as listed in section 1.5.2 (Figs. 1.8 and 1.10-1.12)
that is the topic of discussion.
Our results show evidence of contrasting elastic properties across major faults in the area as
well as anomalous velocities along these faults. At shallow depths (≤5 km) the low-density basins
(Anderson et al., 2004; Langenheim et al., 2004) have very low velocities (1). For the SBB in
particular, these low velocities are juxtaposed against high velocities of the dense mafic western
PRB (3) (Silver and Chappell, 1987; Langenheim et al., 2004), and crystalline rocks of the western
31
Figure 1.10. Final VP inversion results at several cross-sections through the 142 km by 72 km area
centered on SGP (small blue box in Fig. 1.1) for depth ranges 2 km – 12 km (a) and 7 km – 17 km
(b). White contours depict DWS=20 and regions with DWS<20 are shaded. Small black dots
denote relocated seismicity within 2 km of each respective cross-section. Black dot-dashed lines
are projections in depth of major fault traces at the surface. Some prominent features are
highlighted and numbered (see text for descriptions). Note the color bars change with depth.
32
SBMB (Gutierrez et al., 2010), forming prominent velocity contrasts across the northern SJFZ and
SBSSAF, respectively. At similarly shallow depths, low velocities are associated with SBSSAF
and CVSSAF related damage and deformation zones. Similar low velocities are not observed for
Figure 1.11. Same as Fig. 1.10 but for VS inversion results.

33
the MCF (especially between the SBSSAF and PMF) or at the center of the SGPFZ/BF/GHF
system (2). The relatively high velocities around these faults are also imaged at lower resolution
in other tomography results (e.g., Zigone et al., 2015; Fang et al., 2016). We interpret these
Figure 1.12. Same as Fig. 1.10 but for VP /VS structure.

34
velocities to be caused by a lack of recent seismicity and related damage (Kendrick et al., 2015)
and crystalline rocks that persist from the western SBMB in the north through the upper plate of
the SGMB to the PRB in the south (Gutierrez et al., 2010).
Fault-related anomalies are observed near SGP at greater depth. Contrasts across the northern
SJFZ and SBSSAF persist at >7 km (albeit weaker) as velocities beneath the SBB remain low
given the presence of Pelona schist within the SGMB lower plate (8). Velocities within the SGMB
upper plate remain high (7) whereas a clear contrast develops across the CVSSAF where
crystalline rocks of the PRB to the south (fast) meet metamorphosed rocks of the eastern SBMB
in the north (slow) (10). The observed sharp change in velocity at ~10 km depth may correspond
to this lithological change across the CVSSAF and the fault is dipping to the northeast if this
inference is correct. A northeast dipping CVSSAF has been proposed by several other authors
(e.g., Fialko, 2006; Lindsey and Fialko, 2013; Share and Ben-Zion, 2016; Fuis et al., 2017). Using
our VP model, a plane dipping by 57° is formed by connecting the main surface trace of the
CVSSAF with the region of maximum change in velocity within the relocated seismicity cluster
at 10 km depth (Fig. 1.9). This value is similar to the 65° proposed by Fuis et al. (2017). At even
greater depth (≥13 km), high velocities around the MCF persist. In contrast, the SGPFZ is
characterized by a low VP anomaly that disappears at the transition to the BF/GHF in the west
(12). This anomaly is amplified near the intersection between the SGPFZ and SBSSAF and is less
prominent in the VS model (12).
Several imaged features south of SGP are worth discussing. The decrease in velocity from the
western to eastern PRB (4) relates to the lower density and more felsic composition of the latter
(Silver and Chappell, 1987; Langenheim et al., 2004). This transition has been observed in
potential field results (e.g., Langenheim et al., 2004) and other tomography studies (e.g., Hearn
and Clayton, 1986; Allam and Ben-Zion, 2012; Barak et al., 2015). However, here it is a vertical
feature (at depths ≥7 km) and does not dip to the east as has been suggested (Langenheim et al.,
2004; Barak et al., 2015). East of this transition the imaged elongated low velocity features (5)
correlate with abnormally high seismic activity between the SJFZ and EF (Fig. 1.9). This suggests
high strain localization in an intra-fault region that may be related to basin structures (Axen and
Fletcher, 1998) or counterclockwise block rotation from bookshelf faulting (e.g., Ron et al., 1984;
Tapponnier et al., 1990) that can produce complex geodetic signals along and in between the SJFZ
and EF.
35
The obtained VP/VS models provide additional constraints on lithology and fault structure.
VP/VS values generally reflect bulk chemical composition (e.g., Christensen, 1996; Brocher,
2005), rock damage and fluid content (e.g., Lockner et al., 1992; Mavko et al., 1998; Hamiel et
al., 2004). Most crustal rocks have VP/VS values in the range 1.7-1.85 (Brocher, 2005). There are
however two notable exceptions (Christensen, 1996). First, pure quartz has a VP/VS of 1.5 and
therefore rocks abundant in quartz (>55% of total volume) also have anomalously low VP/VS
values. Second, anorthosite has very high VP/VS (>1.9), and metamorphic rocks, which contain
anorthosite-rich plagioclase feldspar, usually have increased VP/VS values. These values are
temperature and pressure dependent (Christensen, 1996) but their effects on VP/VS can be
neglected here given the depths under investigation. With regard to less intact rock, very low VS
and high VP/VS (~2 to 5, Brocher, 2005) have been reported for highly porous or fractured fluid-
filled rocks, such as those found in sedimentary basins. Conversely, highly fractured dry rock is
expected to have low VP and low VP/VS (O’Connell and Budiansky, 1974). Anomalously low VP
and low VP/VS values near fault damage zones at depth have also been linked to large aspect ratios
of fluid-filled cracks (thick cracks) (e.g., O’Connell and Budiansky, 1974; Lin and Shearer, 2009).
VP/VS is therefore expected to vary greatly across faults separating predominantly crystalline and
metamorphic units, and even greater along fault zones as the degree and type of cracking and the
amount of fluids occupying those cracks changes laterally and in depth.
Some general trends in VP/VS can be related to lithology in the SCTR. The PRB (3), SGMB
upper plate and western SBMB (7), with their abundant quartz-rich crystalline rocks, are regions
with below average VP/VS. The eastern PRB in particular contains granitic rocks with 60-70%
quartz (Silver and Chappell, 1987) and has the lowest VP/VS values in the study region (5 and 13).
The low VP/VS anomaly beneath the SGPFZ at depths ≥13 km (12) suggests this fault system
intersects similarly quartz-rich crystalline rocks that may be derived from the western SBMB
and/or the SGMB upper plate (Langenheim et al., 2005). In contrast, metamorphic rocks present
in the SGMB lower plate and eastern SBMB have, as expected, above average VP/VS (6).
Overprinted on large-scale variations in VP/VS are anomalies concentrated around major faults
that give insight into the extent and type of damage present. Very high VP/VS values near the
surface are caused by intensely fractured and porous rocks within the SBB, SJB and Coachella
Valley basin (1) and damaged wet rocks near major faults (e.g., O’Connell and Budiansky, 1974;
Mavko et al., 1998). Near vertical low VS and high VP/VS anomalies extend downwards from the
36
surface traces of the northern SJFZ, CHFZ and SBSSAF (14 and 15). This suggests these faults
are near vertical also and represent regions of relatively intense damage that may contain fluid-
filled cracks. The pronounced low VP and low VP/VS anomaly present beneath the mapped
intersection of the eastern end of the SBSSAF and the SGPFZ (16) similarly suggest near vertical
fault structure. Moreover, the low VP/VS values point to a damage regime consisting of thick fluid-
filled cracks (e.g., Lin and Shearer, 2009). Near vertical faults around this intersection contrast
with inferences of dipping faults from analysis of potential field data (Fuis et al., 2012) and
numerical modeling to reproduce geologic uplift and slip rates (Daire and Cooke, 2009). However,
it is supported by suggested fault orientations from studies on seismicity distributions (Carena et
al., 2004). The northeast dipping low VS and high VP/VS anomalies beneath the BF/GHF (17)
corroborate earlier suggestions (e.g., Jones et al., 1986; Nicholson, 1996) that this fault system is
dipping in that direction.
It is of some interest to note the correlation between near-fault VP/VS anomalies at depth and
the basement rocks within which they locate. Very low VP/VS structures are only observed near
faults located within quartz-rich crystalline rocks (e.g., SGPFZ, SJFZ trifurcation area (19) and
ECSZ north of PMF (11)), and not near faults bounding or intersecting metamorphic rock units
(e.g., northern SJFZ, GHFZ, SBSSAF, BF/GHF and ECSZ south of PMF). This may indicate a
dependence of the geometry of fluid-filled cracks on lithology. Thick cracks might preferentially
form in the isotropic mineral texture of crystalline rocks such as granite. Such low VP/VS values
have been measured in cracked silica-rich rocks under high pressure from the KTB-drill holes
(Popp and Kern, 1994). Alternatively, cracks forming in metamorphic rocks may predominantly
be thin and long (and have high VP/VS, O’Connell and Budiansky, 1974) as they are parallel to the
planar foliation in these rocks. Dry cracks can help explain some of these variations, but several
studies suggest fluid-rich fault zones in the region (e.g., Hardebeck and Hauksson, 1999). To
further investigate VP/VS variations near faults within different lithologies, it would be useful to
use other geophysical tools, such as magnetotellurics (Becken and Ritter, 2012), that are sensitive
to the same properties but have inherently different resolutions.
In conclusion, the highlighted features (1-19) suggest that structural complexity of the SGP at
the surface continues at depth. No continuous low-velocity anomaly (potential damage zone) is
observed along the SAF system through SGP. Likewise, a continuous bimaterial structure with a
single velocity polarity is also not detected. Instead, a fast rock body cuts across the SAF system
37
near the surface whereas at depth abrupt west-to-east changes in elastic properties characterize the
transition from the SBSSAF and SGPFZ to the BF/GHF. This significant structural heterogeneity
through SGP is corroborated by recent results from other seismological studies (e.g., Share and
Ben-Zion, 2016). Our results also show strong evidence that faults west of SGP have near vertical
dips whereas faults to the east of that area dip to the northeast. These changes in elastic properties
and fault geometry can help explain the high rates of small to moderate sized events in the area
(Hauksson et al., 2012), large variations of earthquake focal mechanisms and stress drops (Bailey
et al., 2009; Yang et al., 2012; Goebel et al., 2015), and strain partitioning in general. It also
provides a potential structural barrier for through-going earthquake ruptures as suggested by the
1948 MW 6.0 Desert Hot Springs (Nicholson, 1996) and 1986 MW 5.8 North Palm Springs (Jones
et al., 1986) earthquakes, which both ruptured large sections of the BF/GHF but did not extend
into the SGPFZ.
The results presented also give insight into the role of the SJFZ in large earthquakes in
Southern California. The sharp change in velocity across the northern SJFZ (8) points to a potential
bimaterial structure at the core of this fault section. Based on theoretical models and laboratory
experiments (e.g., Weertman, 1980; Andrews and Ben-Zion, 1997; Shlomai and Fineberg, 2016),
ruptures on bimaterial faults tend to have preferred propagation direction. An earthquake with
strong directivity can produce an order of magnitude larger ground motion in the direction of
propagation than in the opposite direction. Directivity is therefore a key component in estimations
of expected ground motion from large earthquakes. On the other hand, the extent of potential
earthquake rupture along the SAF and northern SJFZ is affected by the continuity within and
between the two fault systems. Our results point to along-fault structural heterogeneity in the
seismic zone around the Hemet step-over (9). To inform further both properties of potential
bimaterial structures and fault continuity through CP, analysis of fault zone head waves should be
applied to local earthquake data acquired in the region. The results of this study can also be
improved by performing joint inversions of earthquake arrival times and surface wave dispersion
data (e.g., Fang et al., 2016), as well as body and head wave data (Bennington et al., 2013). Work
on some of these topics will be the subject of follow up studies.

38
2. Bimaterial interfaces in the South San Andreas Fault with opposite velocity
contrasts NW and SE from San Gorgonio Pass
(Share and Ben-Zion, 2016)

2.1 Introduction
The South-Central Transverse Ranges section of the plate-boundary in southern California is
highly complex and includes intersections between the San Andreas Fault (SAF), San Jacinto Fault
Zone and Eastern California Shear Zone (Fig. 2.1). The region around the San Gorgonio Pass
(SGP) is associated with a major "structural knot" consisting, at least in the top few km, of a
complex set of thrust and strike-slip faults (e.g., Matti and Morton, 1993; Spotila et al., 2001; Yule
and Sieh, 2003). A fundamental question for seismic hazard is whether large earthquakes that
rupture the Mojave or Coachella Valley sections of the SAF are likely to propagate through, or be
arrested by, the SGP knot. To address this, we attempt to clarify the existence and properties of a
continuous seismogenic fault surface through the SGP. This is done by performing a systematic
search for fault zone head waves (FZHW) that refract along bimaterial fault interfaces. Such
interfaces influence strongly properties of dynamic ruptures (e.g., Ben-Zion and Andrews, 1998;
Ranjith and Rice, 2001; Shlomai and Fineberg, 2016) and they tend to be localization surfaces of
large earthquakes (e.g., Martel et al., 1988; Bruhn et al., 1994; Sibson, 2003; Dor et al., 2008).
FZHW are generated by earthquakes located near a fault separating different solids and they
propagate along the bimaterial interface (Ben-Zion, 1989; 1990), so they provide direct evidence
for a continuous fault section between the generating events and recording stations. These waves
were used previously to document and image properties of bimaterial fault interfaces along
sections of the SAF and Hayward fault in central CA (e.g., Ben-Zion and Malin, 1991; McGuire
and Ben-Zion, 2005; Allam et al., 2014a), the North Anatolian fault in Turkey (Bulut et al., 2012;
Najdahmadi et al., 2016) and faults in various other locations (e.g., Fukao et al., 1983; Hough et
al., 1994; Yang et al., 2015). In addition to documenting continuous seismogenic fault sections,
bimaterial interfaces have important implications for expected behavior of earthquake ruptures
(e.g., Ben-Zion, 2001, and references therein).
A contrast of elastic properties across a fault produces asymmetric dynamic stress field at the
tips of in-plane ruptures propagating in the opposite along-strike directions (e.g., Weertman, 1980;
Andrews and Ben-Zion, 1997; Ben-Zion and Huang, 2002). Behind the tip propagating with
39
standard subshear velocity in the direction of slip on the compliant side of the fault there is dynamic
reduction of normal stress (and hence frictional strength), while in the opposite direction there is
dynamic increase of normal stress and strength. Consequently, subshear ruptures on bimaterial
interfaces evolve for wide range of conditions to pulses that propagate in the direction of slip on
the side with slower seismic velocity (e.g., Shi and Ben-Zion, 2006; Ampuero and Ben-Zion, 2008;
Brietzke et al., 2009; Shlomai and Fineberg, 2016). For supershear ruptures, the polarity of the
Figure 2.1. A map of the study region (160 km by 20 km black box) centered on the San Gorgonio
Pass and intersection of the San Andreas (SAF) and San Jacinto (SJFZ) Faults. The centers of
triangles and circles mark locations of stations and local earthquakes analyzed in this study. Black
dots denote all M ≥ 1.5 events from 1/1/2009 to 30/6/2015. Yellow (22 northwest, 19 southeast)
and dark green (8 west, 5 east) circles mark events for which fault zone head waves (FZHW) were
and were not detected, respectively. Yellow NW events produce FZHW at station SNO (left red
inverted triangle), while yellow SE events generate FZHW at station MSC (right red inverted
triangle). Blue inverted triangles depict stations with no FZHW detections. CP - Cajon Pass; MCF
– Mission Creek Fault; ECSZ - Eastern California Shear Zone; SGPFZ – San Gorgonio Pass Fault
Zone; PS - Palm Springs.

−117˚30' −117˚00' −116˚30' −116˚00'
34˚00' 34˚00'
34˚30' 34˚30'
0 20 40
PS
SAF/MCF
SAF
SGPFZ
SAF
ECSZ
SJFZ
CP
0
10
20
Depth (km)
−80 −60 −40 −20 0 20 40 60 80
X distance (km)
SVD FHO 5442SNO MSC DEV WWC
A A`
A
A`
40
dynamic changes of normal stress and preferred rupture direction are reversed compared to those
in the subshear regime (e.g., Weertman, 2002; Shi and Ben-Zion, 2006; Rubin and Ampuero,
2007). The dynamic changes of normal stress on bimaterial interfaces make them mechanically-
favored surfaces for rupture propagation (Brietzke and Ben-Zion, 2006). Therefore, imaging
bimaterial interfaces in fault zone structures can provide important information on likely failure
surfaces, dynamic stress fields and rupture propagation directions.
The seismic imaging done in this work uses local earthquake waveforms recorded by stations
of the Southern California Seismic Network (SCSN) around the SAF in the region (Fig. 2.1).
Section 2.2 describes the employed data and methods. In section 2.3 we present evidence for
FZHW propagating along sections of the SAF in the South-Central Transverse Ranges. The results
indicate the existence of two large-scale bimaterial interfaces with opposite sense of velocity
contrast in the opposite along-strike directions from SGP. The implications of the results for large
SAF earthquake ruptures are discussed in section 2.4.

2.2 Data and Methods
Our seismic imaging is based on waveforms of 1144 earthquakes with M > 1.5 that occurred
between January 2009 and June 2015 within a 160 km by 20 km box centered on the SGP (Fig.
2.1). The event sizes, locations and origin times were taken from the relocated catalog of Hauksson
et al. (2012). Waveforms of these events observed at seven continuously recording stations in the
South-Central Transverse Ranges close to SGP (six broadband high gain seismometers, SVD,
FHO, SNO, MSC, DEV, WWC, and one accelerometer, 5442) were downloaded from the
Southern California Earthquake Data Center.
A few basic properties help to identify FZHW. They propagate along a bimaterial interface
with the velocity and motion polarity of the faster rock body and are radiated from the fault to the
side with slower seismic velocity. FZHW are the first arriving phases at locations on the slower
solid with normal distance to the fault less than a critical distance 𝑥
l
given (Ben-Zion, 1989) by
𝑥
l
= 𝑟∙tanqcos
U3
q𝛼
`
/𝛼
w
xx, (2.1)
where 𝑟 is the propagation distance along the fault (both along-strike and up-dip direction) and 𝛼
`
,
𝛼
w
are the average P wave velocities of the slower and faster media, respectively. FZHW are
emergent phases with opposite first-motion polarity than the more impulsive following direct P
41
waves. The differential arrival time between the head and direct P waves increases with
propagation distance along the fault as
∆𝑡 ≈𝑟∙∆𝛼/𝛼
V
, (2.2)
where 𝛼 and ∆𝛼 denote the average and differential P wave velocities, respectively (Ben-Zion and
Malin, 1991). FZHW also have horizontal particle motion with a significant fault-normal
component, since they radiated from the fault, in contrast to the particle motion of the direct P
waves that points in the epicenter direction (Bulut et al., 2012; Allam et al., 2014a).
An automated phase picker (Ross and Ben-Zion, 2014) is applied as the first step to detect
candidate FZHW. The algorithm searches in the early P waveform for a possible emergent phase
followed by a sharper arrival with a time separation between a minimum value (0.065 s
representing the width of a narrow P wave wiggle) and a maximum value (corresponding to a 10%
velocity contrast). Noting that the 7 stations were not all operational during the entire examined
time window, the percentages of automatic detections of FZHW for stations SVD, FHO, 5442,
SNO, MSC, DEV and WWC (relative to all events analyzed per station) are 3.0%, 4.9%, 5.0%,
4.6%, 20.9%, 2.1% and 5.8%, respectively.
Following the automatic detection, the identification of FZHW is improved by visual
inspection and comparing results generated by multiple events at multiple stations. Some candidate
phases found to be similar to the noise preceding the P arrivals are discarded. Emergent first
arrivals from a given event at stations on both sides of the fault (likely generated by a source effect)
are also discarded as possible FZHW. Focusing on events with locations close to the SAF proper
leads to further reduction of candidate FZHW. Waveforms originating from earthquakes close to
the hypocenters of the events generating the remaining candidate FZHW are visually inspected to
identify additional FZHW phases. This helps detecting more phases because the automatic
algorithm is designed to minimize false detections. Station SNO has relatively high noise level
leading to low rate of automatic detections, but visual inspection results in a large increase of
candidate FZHW at SNO. Finally, particle motion analysis of the early P waveforms is used to
confirm the existence of a first phase with a component pointing to the fault followed by a second
phase polarized in the back azimuth direction to the event epicenter. Using these steps, we identify
in the examined data 41 earthquakes (yellow circles in Fig. 2.1) producing clear FZHW at stations
close to the SAF.

42
2.3 Results
Inspecting the locations of events producing FZHW and comparing the seismic waveforms
associated with multi-events and multi-stations, we observe two source regions that consistently
produce FZHW recorded only at stations close to and on one side of the SAF proper. The first
region includes 22 events near Cajon Pass (Fig. 2.2) that generate FZHW recorded only at station
SNO south of the SAF/Mission Creek Fault (MCF) and not at the 4 neighboring stations (Fig.
A2.1) north of the SAF. The second region includes 19 events around the Coachella Valley strand
of the SAF northeast of Palm Springs (Fig. 2.3) generating FZHW recorded only at station MSC
(north of SAF/MCF). This second event group did not produce FZHW at two neighboring stations
(SNO and DEV) south of the SAF and at two stations (FHO and WWC) north of but not
sufficiently near (Equation 2.1) the SAF proper (Fig. A2.2). Significantly, none of the FZHW
generated by the 22 Cajon Pass events and recorded at station SNO were observed at station MSC
(Fig. 2.2, bottom right). Similarly, no FZHW generated by the 19 Coachella Valley events were
observed at station SNO (Fig. 2.3, bottom right). FZHW were also not observed at SNO and MSC
for events closer to the stations than about 20 km, because generation of first-arriving FZHW
requires sufficiently long propagation distance along the bimaterial interface (Equation 2.1).
Waveforms for 8 and 5 of these events (dark green circles in Figs. 2.1-2.3) are shown in Figs. 2.2
and 2.3 (bottom panels), respectively.
To estimate the differential time separation between the FZHW and direct P waves, the
relevant velocity waveforms at stations SNO (Fig. A2.1) and MSC (Fig. A2.2) were aligned on
the second major P swing (typically the absolute P maximum) and integrated to displacement to
enhance the clarity of the initial P waveforms (Figs. 2.2 and 2.3, bottom left). Integration was done
using the cumulative trapezoidal numerical integration tool in MATLAB. Similar alignment was
used for stations with no FZHW detections (Figs. 2.2 and 2.3 bottom right; Figs. A2.1 and A2.2).
After alignment, clear moveouts of FZHW arrivals with increasing hypocentral distance are
observed (green diagonal lines in Figs. 2.2 and 2.3, bottom left). Using Equation 2.2, the average
velocity contrasts estimated from the slopes of the moveout lines are small (~2-3%), yet
compatible with sections of the Hayward fault in California (Allam et al., 2014a) and the North
Anatolian fault in Turkey (Najdahmadi et al., 2016). The actual contrasts are likely to be somewhat
larger, since Equation 2.2 involves propagation distances along the fault and we use for simplicity
the longer hypocentral distances. We also note that the obtained values reflect average velocity
43

Figure 2.2. (Top) A map of the NW portion of the study area with 22 and 8 events (yellow and
dark green circles, respectively) that do and do not generate FZHW at station SNO. None of the
events produce FZHW at station MSC. (Middle panels) Horizontal particle motions for two events
generating FZHW (hw1 and hw2) with increasing hypocentral distance (Hypo. dist.) and one event
(ref) without FZHW. Red dots correspond to a time window starting at the FZHW pick and ending
at the direct P pick, while blue dots correspond to 0.2 s after the direct P pick. The location of
these example events and vertical component seismograms are marked by maroon color in the top
and bottom left panels. (Bottom panels). Vertical component displacement seismograms generated
by the 22 yellow events at station SNO (left) with FZHW detections and at station MSC (right)
without FZHW. For reference, seismograms from the 8 dark green events are also plotted. The
waveforms were detrended and high-pass filtered at 1 Hz with a one-pass Butterworth filter.
Velocity seismograms for stations SNO and MSC and other three stations in the map (SVD, 5442
and FHO) are shown in Fig. A2.1.
−1
−0.5
0
0.5
1
NS
Back azim.=289.4°
Hypo. dist.=57.96 km
Back azim.=291.8°
Hypo. dist.=63.52 km
hw2 hw1
EW
−10 1
EW
−10 1
EW
FZHW
P direct
−0.5 0 0.5
−90
−80
−70
−60
−50
−40
−30
MSC
Time relative to aligned P phase (sec)
P direct pick
−0.5 0 0.5
−80
−70
−60
−50
−40
−30
−20
−10
0
Time relative to aligned P phase (sec)
Hypocentral distance W (−) and E (+) (km)
2.2%
P direct pick
FZHW pick
ref
hw1
hw2
Δt~r(Δα/α )
2
SNO
Back azim.=278.7°
Hypo. dist.=21.11 km
−10 1
ref
FHO
MSC
SVD
5442
SNO
N
10 km
44
contrasts over the top 14 km (Fig. 2.1), and that the contrasts may be double in the uppermost 7
km or so of the crust (e.g., Ben-Zion et al., 1992; Lewis et al., 2007). The velocity contrasts are
estimated assuming an average P wave velocity of 6.2 km/s, which is the average velocity in the

Figure 2.3. Same as Fig. 2.2 but for the 19 events generating FZHW (yellow circles, top panel)
and 5 events not generating FZHW (dark green circles, top panel) at station MSC in the SE portion
of the study area. None of these events produce FZHW at station SNO. The time window used in
the particle motion plots of P direct phases has been decreased to 0.14 s to make the FZHW motion
more visible. Velocity seismograms for stations MSC, SNO, FHO, DEV and WWC are shown in
Fig. A2.2.

−0.5 0 0.5
0
10
20
30
40
50
60
70 MSC
Time relative to aligned P phase (sec)
Hypocentral distance W (−) and E (+) (km)
2.8%
Δt~r(Δα/α )
2
hw2
hw1
ref
P direct pick
FZHW pick
DEV
FHO
SNO
WWC
MSC
N
10 km
−0.5 0 0.5
20
30
40
50
60
70
80
Time relative to aligned P phase (sec)
P direct pick
SNO
−1 −0.5 0 0.5
−0.5
0
0.5
1
EW
NS
Back azim.=116.5°
Hypo. dist.=13.86 km
−0.5 0 0.5 1
−0.5
0
0.5
1
EW
Back azim.=114.1°
Hypo. dist.=37.94 km
FZHW
P direct
−0.5 0 0.5 1
−0.5
0
0.5
1
EW
Back azim.=117.1°
Hypo. dist.=49.70 km
hw2 hw1 ref
45
tomography model of Fang et al. (2016) at the median depth (10 km) of the events producing
FZHW.
The moveouts observed in Figs. 2.2 and 2.3 from events >45 km NW of SNO and >30 km SE
of MSC and close to the bottom of the seismogenic zone provide strong evidence of FZHW
propagating along bimaterial interfaces that extend to at least the depth of the deepest events. Due
to the limited seismicity these interfaces are not sampled continuously at distances <45 km NW
and <30 km SE, respectively. However, they can be inferred because FZHW propagate exclusively
along bimaterial interfaces and would not be observed at stations SNO and MSC in the absence of
continuous interfaces between the generating events and recording stations. Additional evidence
for bimaterial interfaces is given by the horizontal particle motions of the FZHW (red dots, Figs.
2.2 and 2.3, middle panel), which have large components in the fault-normal directions (Bulut et
al., 2012; Allam et al., 2014a). These are distinctly different from the particle motions of the direct
P phases following the FZHW or the P phases for nearby reference events with no FZHW (blue
dots, Figs. 2.2 and 2.3, middle panel). The FZHW and direct P waves do not have opposite polarity
as expected for strike-slip events. This can be explained by the diversity of focal mechanisms in
the area (Bailey et al., 2010) and/or by events that are slightly off the SAF on the side with slower
seismic velocity.
The detection of FZHW at two neighboring stations on the opposite sides of the SAF/MCF
implies the existence of two distinct bimaterial interfaces with opposite sense of velocity contrast
to the NW and SE from SGP. This is supported by the 3D velocity model based on joint
tomographic inversion of body wave arrivals and surface wave dispersion data of Fang et al.
(2016). The tomographic results for the region show (Fig. 2.4) slower seismic velocities SW of the
SAF to the NW of station SNO, while to the SE of station MSC the slower velocity rocks are NE
of the SAF. Additional support is provided by geological mapping indicating that the SAF
separates to the NW of station SNO Mesozoic granite of the San Bernardino Mountains block in
the NE from Cenozoic to Mesozoic Pelona schist in the SW (Matti and Morton, 1993). However,
to the SE of station MSC, Precambrian gneiss and schist in the NE are juxtaposed across the SAF
with Mesozoic plutonic rocks in the upper plate of the San Gabriel Mountains block (Matti and
Morton, 1993) and Mesozoic granite of the Peninsular Ranges block (Gutierrez et al., 2010) to the
SW. The observation of FZHW at stations SNO and MSC imply that these alternating rock bodies
are bounded by sharp bimaterial interfaces with reversed velocity contrast polarity.
46

2.4 Discussion
The FZHW detected and analyzed in this study (Figs. 2.1-2.3) reveal two large-scale
bimaterial interfaces that extend to the bottom of the seismogenic zone and may form the spine of
the SAF in the South-Central Transverse Ranges (pink line in Fig. 2.4). One interface extends
continuously from station SNO to the NW slightly beyond Cajon Pass and separates slower rocks
SW of the SAF from faster block to the NE. The other bimaterial interface extends continuously
from station MSC to the Coachella Valley and has a velocity contrast with opposite polarity. The
reversal of the velocity contrast across the SAF in the area is associated (Fig. 2.4) with a relatively
fast rock body that cuts across the SAF below SGP (and stations SNO and MSC). This fast rock
body broadly correlates with crystalline rocks that persist from the San Bernardino Mountains
block in the north through the upper plate of the San Gabriel Mountains block in SGP to the
Peninsular Ranges block in the south (Gutierrez et al., 2010). The imaged bimaterial interfaces
imply continuous fault sections at depth despite the lack of continuous low magnitude seismicity
(Magistrale and Sanders, 1996; Carena et al., 2004) and surface fault traces (Wesnousky, 2006).
The generation of FZHW from the easternmost events with epicenters >5 km north of the SAF
Figure 2.4. Events generating FZHW (black circles) recorded by stations (large triangles) SNO
(NW events) and MSC (SE events). Small triangles denote stations not recording FZHW. The
locations are overlain on the mean P wave velocities from the 3D velocity model of Fang et al.
(2016), averaged from 3 km (top of the seismogenic zone) to 10 km (median depth of events
generating FZHW). To the NW of station SNO the SW side of the SAF has slower seismic
velocity, while to the SE there is a reversal in the velocity contrast across the SAF. The pink line
and arrows mark, respectively, surface projections of the imaged bimaterial interfaces and
preferred propagation direction of subshear ruptures.

47
trace (Fig. 2.3, top panel) support inferences that the SAF in that area is dipping to the NE (Daire
and Cooke, 2009; Fuis et al., 2012; Lindsey and Fialko, 2013).
The observed FZHW combined with expected behavior of bimaterial ruptures (e.g.,
Weertman, 1980; Ben-Zion and Andrews, 1998; Ampuero and Ben-Zion, 2008) have important
implications for large earthquakes on the SAF in the South-Central Transverse Ranges. The
velocity contrast to the NW of SGP suggests that typical subshear earthquakes that start around
station SNO tend to propagate along the SAF to the NW. This is consistent with observed rock
damage asymmetry across the Mojave section of the SAF (Dor et al., 2006a) and along-strike
asymmetry of aftershocks on that fault section (Zaliapin and Ben-Zion, 2011). The opposite
velocity contrast to the SE of SGP suggests that subshear earthquakes tend to propagate on that
section of the SAF to the SE. This is consistent with observed along-strike asymmetry of
aftershocks (Zaliapin and Ben-Zion, 2011) and small reversed-polarity deformation structures
(Ben-Zion et al., 2012) on the southern SAF. Occasional supershear ruptures on these sections
would tend to propagate in the opposite directions.
The results also suggest that ruptures that enter the SGP area from either direction would likely
be arrested by the reversal of the velocity contrast they would encounter with continuing
propagation. This is because the reversal of the velocity contrast would increase the frictional
strength due to the dynamic increase of normal stress at the rupture tip. The SGP region is therefore
not only a "geometrical knot" but also a "dynamical knot" for earthquake ruptures. One possible
scenario for a very large SAF earthquake is nucleation between stations SNO and MSC that may
lead to simultaneous ruptures on both bimaterial interfaces in the opposite along-strike directions.
It would be useful to substantiate the results of this work with additional high-resolution
seismological imaging of the South-Central Transverse Ranges section of the SAF, and efforts to
obtain additional seismological and geological evidence on preferred rupture directions of
earthquakes to the NW and SE from SGP.

48
3. Internal structure of the San Jacinto fault zone at Blackburn Saddle from
seismic data of a linear array
(Share et al., 2017)

3.1 Introduction
The 230km-long San Jacinto Fault Zone (SJFZ) is the most seismically active fault zone in
southern California (Hauksson et al., 2012) and accommodates a large portion of the plate
boundary motion in the region (Johnson et al., 1994; Fialko, 2006; Lindsey et al., 2014). Extensive
paleoseismic work indicates that the SJFZ has repeatedly produced large (MW > 7.0) earthquakes
in the past 4000 years (Rockwell et al., 2015, and references therein). Variations of lithological
units and geometrical complexities (e.g., Sharp, 1967) produce non-uniform distribution of strain
and seismicity along the length of the fault (Sanders and Kanamori, 1984; Sanders and Magistrale,
1997, Hauksson et al., 2012). Recent tomographic studies (Allam and Ben-Zion, 2012; Allam et
al., 2014b; Zigone et al., 2015) imaged with nominal resolution of 1-2 km large-scale variations
of seismic velocities across the fault and significant damage zones at different locations. Internal
structural components of the SJFZ have been studied using various seismic arrays that cross the
fault at different locations (e.g., Li and Vernon, 2001; Lewis et al., 2005; Yang et al., 2014; Li et
al., 2015; Ben-Zion et al., 2015; Hillers et al., 2016), along with geological mapping of rock
damage and analysis of geomorphologic signals (Dor et al., 2006b; Wechsler et al., 2009).
In the present study we use several seismological techniques to clarify internal components of
the SJFZ at Blackburn Saddle northwest of Anza. The study area is at the head of Blackburn
Canyon near the northwestern end of the longest continuous strand of the SJFZ, the Clark Fault
(Sharp, 1967). The site ruptured during two large earthquakes in the last 250 years, the M 7.2-7.5
1800 event (Salisbury et al., 2012) and the M 6.8 1918 earthquake (Sanders and Kanamori, 1984).
The analyses employ earthquake waveforms recorded for about 1.5 years by a linear seismic array
(BB array, Fig. 3.1 bottom right inset) across the Clark fault in the study area. We use delay times
of P arrivals across the array, fault zone head waves (FZHW) and fault zone trapped waves
(FZTW) to image properties of the fault damage zone and bimaterial fault interface. We focus on
these structural components because they contain information on likely properties of past and
future earthquake ruptures and associated ground motion (e.g., Andrews and Ben-Zion, 1997; Dor
et al., 2006b; 2008; Brietzke et al., 2009; Shlomai and Fineberg, 2016).
49

Figure 3.1. (Top) The study region (110 km by 100 km black box) centred on the Clark fault in
the San Jacinto Fault Zone (SJFZ). The San Andreas (SAF), Elsinore Faults and Eastern California
Shear Zone (ECSZ) are marked. Waveforms from events (light green circles) within the black box
are inspected for FZTW. Events within the red rectangle (110 km by 20 km) are used for delay
time and FZHW studies. The yellow triangle shows the location of the BB array. The black squares
mark the towns of Anza and Hemet. (Bottom) A depth section of events projected along the profile
A-A' on top.

50
FZHW propagate along a fault bimaterial interface with the velocity and motion polarity of
the body waves on the faster side of the interface. These phases are analogous to Pn head waves
in horizontally layered media, and they arrive at near-fault stations on the slower side of the fault
before the direct body waves. FZHW provide the highest resolution tool for imaging the existence
and properties of bimaterial fault interfaces (e.g., Ben-Zion et al., 1992; McGuire and Ben-Zion,
2005). On the other hand, misidentification of FZHW as direct arrivals can introduce biases and
errors into derived velocity structures, earthquake locations and fault plane solutions (e.g.,
McNally and McEvilly, 1977; Oppenheimer et al., 1988; Bennington et al., 2013). FZTW are slow
seismic energy associated with resonance modes within low-velocity fault zone layers. For the
antiplane S case they are analogous to surface Love waves of a horizontally layered structure,
while for the P case they are analogous to surface Raleigh waves or leaky modes (e.g., Ben-Zion
and Aki, 1990; Ellsworth and Malin, 2011). The generation of FZTW requires a sufficiently
coherent zone of damaged rocks that can act as a waveguide (e.g., Igel et al., 1997; Jahnke et al.,
2002). These and other less coherent parts of the fault damage zone also produce delay time of
direct P and S waves propagating through the fault zone structure (e.g., Lewis and Ben-Zion, 2010;
Yang et al., 2014; Qiu et al., 2017).
In the next section (3.2) we provide more detail on the array stations and data. The analysis
techniques and results are described in section 3.3 using four subsections. The first two subsections
contain information on identification of P arrivals from teleseismic and local earthquakes, and
related calculations of delay times across the array. The latter two subsections describe
identification and analyses of FZHW and FZTW. The results are summarized and discussed in the
last section 3.4.

3.2 Instrumentation and data
The BB array is a part of a PASSCAL deployment (YN) within and around the SJFZ (Vernon
and Ben-Zion, 2010). The array comprises of 7 Guralp CMG-40T-1 short period three-component
sensors installed ~30 m apart (locations in Table 3.1). The array is orientated normal to the surface
trace of the Clark Fault and the middle sensor (BB04) is installed on top of the surface trace of the
fault (Fig. 3.1 bottom right inset). The instruments measure ground velocity and have a flat
frequency amplitude response between 1 and 100 Hz with a sampling rate of 200 Hz. Recording
started on 11/18/12 and ended on 4/26/14.
51

A catalog of seismicity up to, 2013 for the SJFZ region (White et al., 2016) is used to extract
local event waveforms from continuous BB recordings. The catalog utilizes the Anza network,
nearby stations of the Southern California Seismic Network and stations from several local
deployments. We extract 80 s long waveforms for events occurring from 11/18/12 to 12/31/13, 10
s before and 70 s after the origin times reported in the catalog. In total, 10,603 of these events are
located within a 110 km by 100 km box centered on the array and aligned with the Clark Fault. S
waveforms generated by the 10,603 events are analyzed in the FZTW study (Fig. 3.1). P arrivals
and waveforms from 8,216 events contained in a smaller region (110 by 20 km, red box in Fig.
3.1) are analyzed in the delay time and FZHW studies.
P waveforms corresponding to all M > 5 teleseismic earthquakes (within 30°-100°) contained
in the Southern California Earthquake Data Center (SCEDC, 2013) that occurred during the study
period are extracted from BB data in 30 s windows. A subset of 79 high quality events with
sufficiently high signal-to-noise ratio is retained for further analysis (Fig. 3.2a).

3.3 Methods and Results
3.3.1 Teleseismic earthquake delay time
The primary factors that contribute to P wave pick time variations between BB stations are
incorrect identification of P wave arrivals, variations due to propagation paths to different stations
and variations in local P velocity structure. The performed delay time analysis provides statistical
information on P arrival times determined from waveforms generated by numerous events, with
the aim of constraining the local velocity structure.
Table 3.1. Locations of the 7 BB stations.
Station name Latitude (degrees) Longitude (degrees) Elevations (m)
BB01 33.66871 -116.79584 1173
BB02 33.66897 -116.79544 1167
BB03 33.66914 -116.79528 1165
BB04 33.66927 -116.79511 1165
BB05 33.66946 -116.79480 1169
BB06 33.66967 -116.79462 1169
BB07 33.66999 -116.79456 1180

52

Figure 3.2. Relative delay time results based on teleseismic P waves. a) Locations of 79 events
used in the study with the red circle representing an example event. b) Velocity waveforms
(vertical component) of the example event normalized by the absolute maximum of BB01 and
aligned relative to the event origin time. Red triangles depict manual P picks. c) Cross-correlation
functions between each BB trace and a template trace (see text for calculation of template, positive
lag=delayed arrival). d) Average relative delay of all events based on manual picking and cross-
correlation. Effective frequency range refers to the dominant frequencies of teleseismic P waves
analyzed here.

53
In the case of teleseismic events, the first step consists of accurately determing P arrival time
differences between stations. This is done in two ways. Firstly, for a given event the first P
maxima/minima coherent across the array are picked and designated P arrivals (P peak picks in
Fig. 3.2b). They are more readily identified than the first arriving P energy because the latter in
almost all cases is low in amplitude and comparable to the noise level. Secondly, relative time
delays between stations are obtained using cross-correlation of the 30 s waveforms (Fig. 3.2c). If
the first maximum/minimum has large amplitude and is coherent while the trailing waveform is
incoherent across the array, then P peak pick differences give the best estimates of relative arrival
times. In contrast, if the first maximum/minimum has low amplitude (more prone to be affected
by noise) but entire waveforms are coherent between stations, then cross-correlation is the best
method to estimate relative arrival times. The two methods jointly provide a robust way of
estimating relative times between stations for a variety of teleseismic P waveforms.
The second step encompasses minimizing time variations due to different propagation paths.
Arrival times at different stations in the absence of shallow lateral velocity changes and topography
are first approximated using TauP (Crotwell et al., 1999) and the IASP91 model. Although the BB
array is located ~1 km above sea level, topography varies little within the array (largest difference
is 15 m, Table 3.1). Therefore, constant horizontal slowness can be assumed and the relative arrival
times at the average elevation across the array are equal to times predicted using TauP. For manual
P picks the predicted arrival times are removed from picked arrival times for each event. Prior to
cross-correlation the same predicted arrival times are used to appropriately shift the time series.
During the final step relative delays are computed and velocity structure is inferred from the
delays. In the case of manual picks, the average of the remaining times for each event is subtracted
to produce relative delay times. Next, means and standard errors of the 79 relative delays at each
station are computed (red curve in Fig. 3.2d). In the case of cross-correlation any trend is removed
from the data and a bandpass filter between 0.2-2 Hz is applied. A template is then created for each
event by summing the seismograms across the array. Next, a cross-correlation function is
calculated and used to measure the relative delay time from the peak correlation lag. Similar to
manual picking, means and standard errors of the 79 relative delays at each station are computed
(blue curve in Fig. 3.2d). Both results using manual picking and cross-correlation show a gradual
increase in relative delay from BB01 to BB07, suggesting an increase in subsurface slowness from
54
southwest to northeast. The observed gradual increase in slowness reflects the relatively low
dominant frequencies of the teleseismic P waves (0.5 – 1.5 Hz).

3.3.2 Local earthquake delay time
We perform a similar analysis using local earthquakes with a slightly modified methodology.
First, we process the early P waveforms for all (>10,000) earthquakes with an automatic algorithm
(Ross and Ben-Zion, 2014). The algorithm uses short-term average to long-term average detectors
together with kurtosis- and skewness-based detectors to identify and pick the onset times of P
waves (Fig. 3.3), and FZHW if present. To avoid ambiguity between FZHW and direct P phases,
if a FZHW pick is made at any BB station for a given event, then that event is discarded for the
delay time analysis. For all remaining events, outlier P picks are systematically removed through
the application of the following steps:
1) We discard P wave picks more than 1 s off from predicted arrival times based on a 1D
model, which is an averaged discretized in 1 km layers of the 3D tomographic model of
Allam and Ben-Zion (2012) for the region.
2) Each observed travel time is normalized by the theoretical ray length computed from the
1D model to obtain an average slowness. This minimizes differences in travel times due to
station separation and homogenizes travel times associated with different event
hypocenters. We then discard picks which have slowness values outside a reasonable range
(0.13 to 0.22 s/km) bounded by the corresponding maximum and minimum velocities (7.73
and 4.57 km/s) of the 1D model.
3) Events with slowness values at fewer than 4 stations are removed to focus on observations
associated with most of the array stations.
4) A final round of outlier removal is applied using statistical inner and outer fencing. That
is, a given slowness value s is considered an outlier and removed if s < Q1-1.5(Q3-Q1) or
s > Q3+1.5(Q3-Q1), where Q1 and Q3 are the slowness values nearest to the first and the
third quartile, respectively.
The picking algorithm initially produced 37,206 P and 2,033 FZHW picks for 6,573 and 758
events, respectively. After completing the aforementioned steps, 17,592 P picks from 2,777 events
remained (histogram in Fig. 3.4). Figure 3.4 displays the mean slowness and associated standard
errors for the BB array. Similar to the teleseismic results, the largest subsurface slowness is
55
observed for BB07. The increase in slowness from BB01 to BB07 is not as smooth as in Fig. 3.2d,
reflecting the higher dominant frequencies of local P waves (5 – 25 Hz) compared to the
teleseismic data and associated higher sensitivity to small-scale heterogeneities. Following an
outlier exclusion process, a significant amount of variability still exists for each station (large error
bars in Fig. 3.4). While picking errors can account for some of this, most of the variability is likely
the result of 3D structure outside the fault zone. One way to more appropriately deal with this is
Figure 3.3. Automatic P picks (red triangles) made on velocity waveforms (vertical component)
generated by an example event with hypocenter ~20 km away (location in inset). Waveforms are
normalized by BB01 and aligned relative to the event origin time. The dashed blue line is aligned
with the pick at BB03 and highlights the arrival time differences between stations.

56
by comparing relative slowness values between stations, rather than absolute slowness.
Subsequently, for each event we estimate the relative slowness by dividing each non-zero slowness
value by the mean slowness across the array of that event. Then we calculate the mean and standard
error of relative slowness at each station individually. These calculations show (Fig. 3.5a, black
line) that BB07 has larger relative slowness compared to BB01 and the error bar for each station
Figure 3.4. Delay time results derived from automatic P picks of local arrivals. Slowness values
are calculated from arrival times (see text) and a mean and standard error of slowness is calculated
at each station individually (black curve and error bars) using 2503, 2458, 2574, 2550, 2495, 2550
and 2462 data points for BB01 to BB07, respectively. Effective frequency range refers to the
dominant frequencies of local P arrivals analyzed here. Top left inset shows the distribution of this
data and the horizontal dashed line in the main plot equals the peak in the distribution.

57
calculation is significantly reduced. In summary, rays propagating to BB07 sample on average
structure that is 0.9-1.2% slower than rays propagating to BB01.
To check whether the results are independent of azimuth, events are partitioned into north,
east and south blocks (Fig. 3.5b) and for each corresponding dataset the mean and standard error
are computed per station. Same computations are not made for events located west of the array
due to a lack in seismicity. The relative slowness of each subset of data is almost identical to the
relative slowness calculated from all data at every station (Fig. 3.5a). This indicates that the general
increase in slowness from BB01 to BB07 is associated with local structure and effects of 3D
variations outside the fault zone are not significantly present in the relative slowness data.

3.3.3 Fault zone head waves
3.3.3.1 Methodology
FZHW are critically refracted emergent phases that travel along a fault bimaterial interface
with the velocity and motion polarity of the faster medium (Ben-Zion, 1989, 1990). They arrive
Figure 3.5. Relative delay time results of local arrivals. a) Estimates based on all events (black
curve) are compared to those using events north (magenta), east (red) and south (cyan) of the array.
b) (Top) Locations of north (magenta), east (red) and south (cyan) event subsets. Yellow triangle
represents the array location. (Bottom) A depth section of events projected along the profile A-A'
on top.

58
before the impulsive direct P waves at locations on the slower medium with normal distance to the
fault less than a critical distance 𝑥
l
given by
𝑥
l
= 𝑟∙tanqcos
U3
q𝛼
`
/𝛼
w
xx, (3.1)
where 𝑟 is the propagation distance along the fault (both along-strike and up-dip direction) and 𝛼
`
,
𝛼
w
are the average P wave velocities of the slower and faster media, respectively (Ben-Zion, 1989).
For events with focal mechanisms coinciding with the fault, FZHW and trailing direct P waves
have opposite first motion polarities (Ben-Zion and Malin, 1991; Ross and Ben-Zion, 2014). Also,
FZHW are radiated from the fault and have horizontal particle motion (HPM) with a significant
fault-normal component (Bulut et al., 2012; Allam et al., 2014a; Share and Ben-Zion, 2016). In
contrast, HPM of direct P waves points in the epicenter direction. The differential time ∆𝑡 between
FZHW and direct P waves increases with propagation distance along the fault and is related to the
average velocity 𝛼 across the fault by (Ben-Zion and Malin, 1991):
∆𝑡 ≈𝑟∙∆𝛼/𝛼
V
, (3.2)
where ∆𝛼 is the differential P wave velocity. Also, ∆𝑡 decreases with increasing normal distance
from the fault to zero at the critical distance 𝑥
l
.

3.3.3.2 Results
Using the criteria in 3.3.1 we focus on determining which of the events previously flagged by
the automatic detector (and discarded in 3.2) produce FZHW. The detector flags P waveforms with
an emergent phase followed by an impulsive arrival with a time separation between a minimum
value (0.065 s representing the width of a narrow P wave wiggle) and a maximum value that
depends on hypocentral distance (e.g., 0.8 s over a distance of 40 km). The latter is calculated
assuming a faster side velocity of 5.5 km/s and a velocity contrast of 10% based on the tomographic
results of Allam and Ben-Zion (2012). We do not require polarity reversal between FZHW and
direct P waves because of the mixed complex focal mechanisms for events in the region (Bailey
et al., 2010).
Arrivals from flagged events recorded at different stations are visually compared to remove
erroneous picks such as emergent early phases similar to the noise. For each event we also inspect
the waveforms recorded at two reference stations that are part of the regional network and are close
to the BB array (BCCC and RHIL, Fig. 3.6). If emergent first arrivals are flagged at both reference
stations, those emergent phases are not FZHW (since head waves exist only on one side of a fault
59
bimaterial interface) and the event is discarded. We then search for events in the catalog within 10
km of the remaining events and examine them visually for possible additional FZHW phases. The
automatic detector uses settings designed to primarily minimize false detections, at the cost of

Figure 3.6. Stations and events used in FZHW analysis. (Top) Waveforms from events (all circles)
10 km from the fault recorded at the array and reference stations BCCC and RHIL 3 km southwest
and 2 km northeast of the fault, respectively, are analyzed for FZHW. A total of 24 events (large
green, red and light blue circles) producing FZHW at the array and station BCCC are identified.
Figures 3.7 and 3.8 contain waveforms and analysis from one event (ref, large red black-filled
circle) with no FZHW and two events generating FZHW (hw1 and hw2, large red circles).
(Bottom) A depth section of events projected along the profile A-A' on top. Yellow triangle
represents the location of the array.

60
reducing detection of events with FZHW, and performing this additional search helps to make up
for this shortcoming. The identification steps produce 49 events generating candidate FZHW at all
BB stations and only reference station BCCC for events with large enough propagation distance
along the fault (Fig. 3.7, top and middle panels). FZHW picks for these events are adjusted to
where associated emergent phases begin to rise above the noise level (Fig. 3.7, top and middle
panels).
Next, we apply HPM analysis on the early P waveforms that are de-trended, filtered using a
1-30 Hz one-pass Butterworth filter and integrated to displacement. Similar to previous studies
(e.g., Allam et al., 2014a; Najdahmadi et al., 2016), we examine HPM in displacement
seismograms with consecutive moving time windows of length 0.1 s (20 samples) that overlap by
1 sample. For each window, all three components of motion are combined in a 20×3 matrix, the
covariance of the matrix is computed (Bulut et al., 2012) and the largest eigenvalue and
eigenvector of the covariance matrix (major axis of the polarization ellipse) are obtained. We then
test to see if the azimuth of the largest eigenvector for windows starting at the FZHW and direct P
picks point, respectively, towards the fault and the epicenter direction. The results indicate that for
the 49 candidate events there is considerable variability in the particle motion directions. The
azimuths calculated from windows containing direct P waves do not consistently point to the
epicenters and azimuths calculated for the same event often vary up to ~60° between BB stations.
Similar variations are observed for the head waves. The variability in HPM is likely caused by the
complex structure beneath the array and changes in topography (Jepsen and Kennett, 1990;
Neuberg and Pointer, 2000).
Instead of determining the onset of direct P waves from characteristics of the polarization
ellipse a different approach is used. HPM of P waveforms from events generating FZHW are
visually compared with those from reference events without FZHW. The onset of direct P waves
in waveforms with FZHW are chosen to be the times when HPM of those waveforms becomes
most coherent with HPM of reference P waves. The reference events are identified with criteria
opposite to those used to identify FZHW, namely: 1) FZHW are not picked for any station and 2)
first arrivals are impulsive and highly coherent between stations (Fig. 3.7, bottom panel). The
identified reference events are located closer to the recording station and/or more off fault
compared to events generating FZHW. For an event generating FZHW, the reference event with
the closest hypocenter and a difference in back azimuth less than 20° is used.
61

Figure 3.7. Comparison of displacement waveforms (vertical component) from events hw1 (top)
and hw2 (middle) generating FZHW at the BB array and station BCCC and event ref (bottom) that
does not generate FZHW. No FZHW are observed at station RHIL for these events. Waveforms
are aligned on direct P phases and BB traces are normalized relative to BB01 while waveforms
recorded by reference stations are normalized by themselves. Left panels contain 2 s windows
centered on direct P waves and right panels are 0.55 s zooms of early P waveforms.
62
Fig. 3.8 shows a comparison based on this approach of the first 0.5 s of waveforms with
FZHW to the first P wave wiggle of reference waveforms. HPM of traces with FZHW contain
parts that are uncorrelated (red HPM, Fig. 3.8 top and middle panels) and correlated (blue HPM,
Fig. 3.8 top and middle panels) with the HPM of reference traces (blue HPM, Fig. 3.8 bottom
panel). Uncorrelated and correlated HPM correspond to FZHW and direct P waves, respectively.
As can be seen, uncorrelated and correlated HPM do not necessarily point in the fault and epicenter
directions, respectively. This is probably associated with the complex structure and is the case
even for reference stations. Similar results are obtained for a total of 24 out the 49 candidate events
(locations in Fig. 3.6). The direct P picks for those events are adjusted accordingly. Differential
times ∆𝑡 computed from adjusted FZHW and direct P picks are greatest for BB07 and decrease
towards BCCC (Fig. 3.7 top and middle panels) for all 24 events.
Additional support for FZHW is given by the ratio of largest eigenvalues calculated for noise,
FZHW and direct P waves. Ideally, the largest eigenvalue of a window containing a FZHW will
be larger than noise, and the largest eigenvalue corresponding to a direct P wave would be larger
still (Bulut et al., 2012; Allam et al., 2014a). Eigenvalues are computed for phases generated by
the 24 events and recorded at station BB07. For each event, calculations are made for two non-
overlapping time windows of length ∆𝑡 (for that event), where the first sample of the second
window was firstly aligned with the FZHW pick (to compare noise and FZHW) and then shifted
to align with the direct P pick (to compare FZHW and direct P wave). On average, the eigenvalue
ratio between windows containing FZHW and noise is 26.14 (minimum of 0.21 and maximum of
514.44), and between direct P wave and FZHW windows it is 27.28 (minimum of 1.99 and
maximum of 85.61).
Figure 3.9 shows the moveout, ∆𝑡, between the FZHW and direct P waves versus along-fault
distance for the 24 events generating FZHW. The observed moveout is used to estimate an average
velocity contrast across the Clark Fault (Equation 3.1) in the study area. The moveout is similar
for events located both northwest and southeast of the array. The obtained average velocity contrast
over the propagation paths associated with the used events is ~3.2%. This estimate assumes an
average P wave velocity of 6.5 km/s, based on the P velocity of the regional 1D model at the 16
km median depth of the 24 events. The moveout becomes constant for events located >40 km
southeast of the array (Fig. 3.9). This can be explained by these events being located more off fault
at depth compared to closer events (their epicenters have largest fault normal distances, Fig. 3.6).
63
Alternatively, if these events are located close to the fault the constant moveout indicates the
imaged bimaterial interface is limited laterally and in depth beyond 40 km epicentral distance.

3.3.4 Fault zone trapped waves
3.3.4.1 Methodology
Relatively uniform low velocity fault damage zones can act as waveguides and generate
constructive interference of S, P and noise phases giving rise to FZTW (e.g., Ben-Zion and Aki,
Figure 3.8. Horizontal particle motion (HPM) are plotted for two events, hw1 and hw2, that
generated FZHW (rows 1 and 2, respectively) and a reference event, ref, that did not generate
FZHW (row 3). Column 1 shows displacement P waveforms (vertical component) from events
hw1, hw2 and ref recorded at stations RHIL, BB07 and BCCC. Direct P waves are highlighted in
blue and FZHW are highlighted in red. Columns 2, 3 and 4 show corresponding HPM for the three
events of direct P waves (blue dots) and FZHW (red dots) recorded at stations RHIL, BB07 and
BCCC, respectively.

64
1990; Igel et al., 1997, Jahnke et al., 2002; Hillers et al., 2014). These phases have been observed
in various fault and geologic settings in California (Li et al., 1994; Peng et al., 2003; Lewis and
Ben-Zion, 2010) including the SJFZ (Li and Vernon, 2001; Lewis et al., 2005; Qiu et al., 2017),
Turkey (Ben-Zion et al., 2003), Italy (Rovelli et al., 2002; Calderoni et al., 2012; Avallone et al.,
2014), Japan (Mizuno and Nishigami, 2006) and New Zealand (Eccles et al., 2015).
Figure 3.9. Displacement waveforms (vertical component) from 24 events generating FZHW at
BB07 sorted by propagation distance along the fault. The linear moveout of FZHW (green lines
and squares) relative to direct P (red triangles) arrivals from events less than ~40 km away
corresponds to a 3.2% average velocity contrast across the fault. The moveout is constant for
events >40 km southeast of the array (cyan squares and blue line).

65
FZTW appear on seismograms as high amplitude, long duration, low frequency phases that
follow direct arrivals and are observed only at stations that are within or very close to the trapping
structure (e.g., Li and Leary, 1990; Ben-Zion et al., 2003; Lewis and Ben-Zion, 2010). In the
present study the focus is on identifying and analyzing Love-type FZTW that follow the direct S
wave (Ben-Zion, 1998). The first step in identifying events generating candidate FZTW is
automatic detection (Ross and Ben-Zion, 2015). The detection algorithm is based on the dominant
period, wave energy, ratio between absolute peak amplitude and average amplitude, and delay
between the absolute peak and S pick within a 1 s window starting at the S pick for each station.
In order to minimize false detections due to site amplification, the energy in a longer 6 s window
is also computed. The computations are done on vertical and fault-parallel component velocity
seismograms. An outlier detection is used to flag station(s) with calculated values for the short
time windows after the direct S wave significantly larger than the median values of all stations.
This detection method works best if the number of stations with no FZTW phases outnumber the
ones with FZTW phases.
After candidate FZTW are detected they are visually inspected. Any anomalous phase flagged
for a given event at only one station is discarded as a possible FZTW. This is because trapping
structures are typically ~100 m wide (e.g., Li and Vernon, 2001; Lewis et al., 2005; Qiu et al.,
2017) so FZTW should be observed at multiple stations of the dense array. Noise components
observed at a single or several stations are also sometimes flagged as possible FZTW and are
discarded during visual inspection. Waveforms generated by events similar in size and within 20
km from those generating the remaining FZTW are inspected to identify additional candidates.
Waveforms from events producing clear FZTW are inverted for parameters of the trapping
structure, using the genetic inversion algorithm of Michael and Ben-Zion (1998) with a forward
kernel based on the analytical solution of Ben-Zion and Aki (1990) and Ben-Zion (1998). This
inversion process explores systematically the significant trade-offs between the key parameters
governing properties of FZTW (e.g., Peng et al., 2003; Qiu et al., 2017).

3.3.4.2 Results
The automatic detection algorithm of Ross and Ben-Zion (2015) flagged potential FZTW
within waveforms from 624 events, with 94% of all detections shared between stations BB04 to
BB07. The flagged waveforms are visually inspected, erroneous picks are discarded and additional
66
FZTW phases are identified. Newly identified FZTW waveforms are only observed for stations
BB04 to BB07. This procedure leads to identification of 16 events that produce high quality
waveforms with FZTW (Fig. 3.10). All but one of the events are located north-northeast of the
array and most are at considerable distance from the Clark fault (Fig. 3.10 left). The generation of
FZTW by events at considerable distance from the fault indicates that the trapping structure
extends primarily over the top few km of the crust (Ben-Zion et al., 2003; Fohrmann et al., 2004).
The velocity waveforms generated by two example events (tw1 and tw2, Fig. 3.10) with clear
FZTW phases are pre-processed for inversion. The waveforms are corrected for the instrument
response, rotated to the fault-parallel component, bandpass filtered at 2 – 20 Hz and integrated to
displacement (Figs. 3.11a and 3.12a). As a final step, the seismograms are convolved with 1/t
1/2
to
convert a point source response to that of an equivalent line dislocation source (e.g., Igel et al.,
2002; Ben-Zion et al., 2003). The inverted model parameters are: (1-3) S velocities of the two
quarter spaces (assumed different based on section 3.3.3.2) and the fault zone layer, (4-5) width
and Q value of the fault zone layer, (6) location of contact between the fault and left quarter space,
and (7) propagation distance within the fault zone layer. Estimates of the location where energy
enters the low velocity layer (virtual source) and the travel time outside this layer are derived from
the 7 parameters. The allowable bounds for the first six parameters and incremental changes
allowed in each are shown in Table 3.2.
The genetic inversion algorithm maximizes the correlation between sets of observed
waveforms (7 each in this study) and synthetic seismograms generated with the solution of Ben-
Zion and Aki (1990) and Ben-Zion (1998), while exploring systematically a large parameter-space.
This is accomplished by calculating fitness values associated with different sets of model
parameters and migrating in the parameter-space overall in the direction of larger fitness values.
The fitness is defined as (1+C)/2 where C is the cross-correlation coefficient between observed
and synthetic waveforms. When C varies over the range –1 (perfect anti-correlation) to 1 (perfect
correlation), the fitness value changes from 0 to 1. Figure 3.11b shows synthetic (dark blue lines)
waveform fits produced during 10,000 inversion iterations (testing 10,000 sets of model
parameters). Figure 3.11c displays the fitness values (dots) calculated by the inversion algorithm
for the final 2,000 iterations. The curves in Fig. 3.11c give probability density functions for the
various model parameters, calculated by summing the fitness values of the final 2,000 inversion
iterations and normalizing the results to have unit sums. The model parameters associated with the
67
highest fitness values (solid circles in Fig. 3.11c) are used to produce the synthetic waveform fits
of Fig. 3.11b.
Figure 3.12 presents corresponding inversion results for the second example event. The best
fitting and most likely parameters of the trapping structure produced by inversions of waveforms
generated by different events should be similar. This is the case for the results in Figs. 3.11 and
3.12 and inversion results of several other high-quality waveforms with FZTW. Based on the
inversion results, the fault/damage zone is estimated to start beneath station BB04 (the local
coordinate system is centered on BB04), extend about 130-200 m to the NE, have a Q value of 10-
20, an S velocity reduction of 30-40% relative to the neighboring rock and a depth extent of 3.3-4
km. The latter range is estimated by dividing the most likely total propagation distance within the
fault zone by √2 to account for a horizontal propagation component.
Figure 3.10. Locations and waveforms from events analyzed during the FZTW study. (Top left)
BB array (yellow triangle), all events (circles) analyzed for the presence of FZTW and the 16
events (large circles) for which clear FZTW are observed. (Bottom left) A depth section of events
projected along the profile A-A' on top. (Right) Velocity waveforms (fault-parallel component)
produced by the 16 events and recorded at stations BB01 and BB05. The two traces for each event
are normalized by BB01. Highlighted are waveforms from two example events tw1 and tw2 (large
red circles on left) used during inversion for fault zone structure.

68

3.4 Discussion
The different types of analysis presented in section 3.3 can be combined to produce a detailed
model for the internal structure of the SJFZ in the study area (Fig. 3.13). Both the local and
teleseismic delay time analyses show larger slowness beneath BB07 compared to BB01. The
change in slowness observed in the teleseismic data is gradual compared to the more abrupt change
based on the local earthquake seismograms (compare Fig. 3.2d with Figs. 3.4 and 3.5a). The
difference can be explained by the fact that teleseismic arrivals are associated with longer
wavelengths leading to smoother results. The small-scale variations of slowness based on the local
P waves likely reflect local structural variations such as near surface sediments and small-scale
topography.
Stations BB04-BB07 with the lowest P wave velocities record also fault zone trapped S waves
so they are within the core damage zone of the fault. The broad distribution of events generating
FZTW (Fig. 3.10) implies that the trapping structure is relatively shallow (Fohrmann et al., 2004).
Inversions of waveforms including FZTW indicate (Figs. 3.11c and 3.12c) that the trapping
structure extends to a depth of ~3.5 km and has width of ~200 m, Q value of 10-20 and S velocity
reduction of 30-40%. These parameters are similar to properties of trapping structures at other
sections of the SJFZ (Lewis et al., 2005; Qiu et al., 2017), San Andreas fault at Parkfield (Lewis
and Ben-Zion, 2010), Karadere branch of the North Anatolian fault (Ben-Zion et al., 2003) and
Table 3.2. Upper and lower bounds placed on, and incremental change allowed in, the parameter
space during inversions of events tw1 and tw2 (Figs. 3.11c and 3.12c). Density is fixed at 2.5 g/cm
3

during inversion. FZ – fault zone, QS – quarter space, S – shear, Q – quality factor and FZ centre
is the contact between the left QS and FZ layer.
Parameter Lower bound Upper bound Increment
FZ S velocity 0.7 (km/s) 4.5 (km/s) 0.1 (km/s)
Left QS S velocity 3.5 (km/s) 5.0 (km/s) 0.2 (km/s)
Right QS S velocity 3.15 (km/s) 5.5 (km/s) 0.2 (km/s)
FZ Q 1 35 1
Left QS Q 200 200 Fixed
Right QS Q 200 200 Fixed
FZ center -100 m 100 m 10 m
FZ width 100 m 300 m 10 m

69

Figure 3.11. Fault model inversion results for event tw1. a) Prior to inversion waveforms are
deconvolved with the instrument response, band-pass filtered (2-20 Hz) and integrated to
displacement. b) Waveforms convolved with 1/t
1/2
(black traces) are compared to forward modeled
waveforms (blue traces) using parameters corresponding to the best-fit solution. c) Parameter
space for the final 10 generations (2,000 iterations). Green dots represent all solutions, black lines
are cumulative solution density curves and black dots correspond to the best-fit solution.
70

Figure 3.12. Fault model inversion results for event tw2. The layout and steps are the same as Fig.
3.11.

71
other active strike-slip faults and rupture zones. Suggestions of trapping structures at the SJFZ and
other locations that extend to the bottom of the seismogenic zone (e.g., Li et al., 2004; 2007; Li
and Vernon, 2001) were not supported by more quantitative subsequent analyses using larger data
sets (e.g., Peng et al., 2003; Yang and Zhu, 2010).
The FZHW observed at the BB array and reference station BCCC (Fig. 3.7) reveal a fault
bimaterial interface that extends at least ~40 km to the southeast and to a depth of ~20 km (depth
of deepest event within 40 km generating FZHW, Fig. 3.6). The bimaterial interface extents also
to the northwest but is more limited in space, because no events beyond ~20 km to the northwest
produce FZHW. An average P velocity contrast of 3.2% across the interface is calculated from the
moveout between the FZHW and direct P waves with increasing along-fault distance (Fig. 3.9).
This contrast is not as large as across the San Andreas fault south of Hollister (McGuire and Ben-
Zion, 2005) but is comparable to values obtained for the San Andreas fault around San Gorgonio
Figure 3.13. Conceptual model for the Clark fault at Blackburn Saddle based on results presented
here.

Slow Fast
BB01 BB07
200 m
Low
velocity
zone
BB04
SW NE
depth
~ 3.5 km
72
Pass (Share and Ben-Zion, 2016), Hayward fault (Allam et al., 2014a) and North Anatolian fault
(Najdahmadi et al., 2016). The obtained value reflects an average velocity contrast over the top 20
km of the crust. The velocity contrast typically decreases with depth (Ben-Zion et al., 1992; Lewis
et al., 2007), so the contrast may be double in the uppermost 7.5 km or so of the crust.
The existence of FZHW at reference station BCCC and not at station RHIL implies slower
regional structure southwest of BB07 than northeast of it. This is consistent with tomographic
results for the SJFZ region based on local earthquakes and ambient seismic noise (Allam and Ben-
Zion, 2012; Zigone et al., 2015; Fang et al., 2016). The inferred contrast is also consistent with the
surface geology, showing pre mid-Cretaceous banded gneisses on the southwest side of the fault
juxtaposed against mid-Cretaceous tonalitic rocks on the northeast (Sharp, 1967). The tomographic
results show a reversal in the sense of velocity contrast across the Clark Fault to the northwest of
the array, which explains the lack of FZHW from events farther than ~20 km in that direction. The
bimaterial interface is closest to BB07 because the moveout between the head and direct P waves
decreases from that station. The region beneath BB04-BB07 has low P velocities based on the
teleseismic and local delay times (Figs. 3.2d, 3.4 and 3.5a) and acts as a trapping structure for S
waves (Figs. 3.11 and 3.12). Based on the results from the different data sets and analyses, the
head waves propagate along a bimaterial interface that is at the edge of the core damage zone in
the top few km and merges with the main Clark fault at depth (Fig. 3.13). The best available
geological data places the main Clark fault trace directly beneath BB04. The trapping structure
and zone with largest delay times exist primarily in the crustal block with faster seismic velocity
at depth.
The sense of velocity contrast across the Clark fault at depth and theoretical results on
bimaterial ruptures (e.g., Weertman, 1980; Ben-Zion and Andrews, 1998; Ampuero and Ben-Zion,
2008; Brietzke et al., 2009) suggest that earthquakes in the area tend to propagate to the northwest.
This is consistent with observed directivities of small to moderate events on the section southeast
of the array (Kurzon et al., 2014; Ross and Ben-Zion, 2016), along-strike asymmetry of aftershocks
in the area (Zaliapin and Ben-Zion, 2011) and small reversed-polarity deformation structures in
the Hemet stepover region to the northwest (Ben-Zion et al., 2012). Persistent occurrence of
bimaterial ruptures with preferred propagation direction is expected to produce more damage on
the side with faster velocity at depth (Ben-Zion and Shi, 2005). This is in agreement with the
73
observations summarized in Fig. 3.13 and geological mapping near Hog Lake southeast of the
array (Dor et al., 2006b).
Stations at larger distance from the fault may record, in addition to the phases analyzed in this
work, also P and S body waves reflected within a low velocity fault zone layer (Yang et al., 2014)
and waves reflected from bimaterial fault interfaces to off-fault stations (Najdahmadi et al., 2016).
A recent deployment of a longer aperture array across the Clark fault at the same location of the
BB array (Lin et al., 2016) provides opportunities for analyzing these and other signals indicative
of the inner structure of the fault. This will be done in a follow up work.

74
4. Structural properties of the San Jacinto fault zone at Blackburn Saddle
from seismic data of a dense linear array
(Share et al., 2018b)

4.1 Introduction
Clarifying the internal velocity structure of fault zones is important for many seismological
and fault mechanics studies. A velocity contrast across a fault can affect the directivity, velocity
and crack versus pulse style of earthquake ruptures (e.g., Weertman, 1980; Andrews and Ben-
Zion, 1997; Shlomai and Fineberg, 2016), as well as derived earthquake locations and focal
mechanisms (e.g., Oppenheimer et al., 1988; McGuire and Ben-Zion, 2005). Properties of the core
damage zone around the fault contain information on processes and stress conditions operating
during earthquake ruptures (e.g., Sibson, 1989; Xu et al., 2012). Asymmetric damage zone with
respect to the main fault may indicate a preferred directivity of large earthquake ruptures on that
fault section (e.g., Ben-Zion and Shi, 2005; Dor et al., 2006b; DeDontney et al., 2011). Along-
strike (dis)continuity of bimaterial fault interfaces can be relevant for the likely extent of
earthquake ruptures (e.g., Dor et al., 2008; Share and Ben-Zion, 2016).
A set of seismic arrays that cross the San Jacinto fault zone (SJFZ) in southern California at
various locations (Fig. 4.1a) provides unique opportunities for detailed imaging of bimaterial fault
interfaces and damage zones, in relation to geological (e.g., Sharp, 1967; Rockwell et al., 2015)
and regional tomographic (e.g., Allam and Ben-Zion, 2012; Barak et al., 2015; Fang et al., 2016)
results. Previous studies analyzed data recorded by different linear arrays with apertures of ~200-
450 m (e.g., Yang et al., 2014; Qiu et al., 2017) at various locations, and a dense rectangular array
with a linear dimension of ~600 m (Ben-Zion et al., 2015; Hillers et al., 2016; Qin et al., 2018) at
the SGB site. In particular, Share et al. (2017) analyzed data from local earthquakes and teleseismic
events recorded by the BB linear array across the Clark branch (CF) of the SJFZ at Blackburn
Saddle northwest of Anza, consisting of 7 sensors with ~200 m aperture. The site ruptured during
two large earthquakes in the last 250 years, the M 7.2-7.5 1800 (Salisbury et al., 2012) and the M
6.8 1918 (Sanders and Kanamori, 1984) events. The results indicate the presence of a >100 m wide
asymmetric damage zone (most damage on northeast side of the CF), which terminates at a local
bimaterial interface northeast of array that is connected to a regional bimaterial interface at depth.

75

Figure 4.1. Dense linear array studies along the San Jacinto fault zone (SJFZ). a) Locations of the
Blackburn Saddle array analyzed in the present study (BS, large yellow triangle) and other arrays
(BB, RA, SGB, DW, JF and TR, small red triangles) installed along the SJFZ. The towns of Hemet,
Anza and Palm Springs (black squares) are shown for reference. b) 108 nodes of the BS array
(yellow icons) crossing the Clark fault (CF) surface trace. For reference, locations of the 7 BB
stations are also shown (red icons).

76
Due to the short aperture of the array, the spatial extent of the damage zone, which can be wider
than 1 km (e.g., Cochran et al., 2009), was not well constrained. Also, the location of the bimaterial
interface at the surface was not observed but inferred from comparison with data recorded by
stations of the regional seismic network.
To provide a more complete information on the inner structure of the SJFZ at Blackburn
Saddle over a much larger area, a new array (BS) extending ~1 km on each side of the CF was
deployed in late 2015. The BS array had 108 densely-spaced 3-component Zland nodes covering
and extending well beyond the locations of the previous short BB array (Fig. 4.1b). In the present
study we analyze data recorded by this first-of-its-kind 3-component nodal array to derive a high-
resolution structural model of the fault zone over a spatial extent that can be merged with the
tomographic images in the region. The analyzed data include P and S waveforms generated by
local earthquakes and P waveforms of teleseismic events. The employed seismic phases are
sensitive to bimaterial interfaces, damage zone structure and the surrounding rocks, and can better
be distinguished and analyzed using the large aperture BS array spanning all these structures.
Variations in teleseismic arrivals are used to determine large-scale structural changes across the
fault zone including the extent of a broad damage zone at the site. Local arrivals are analyzed for
the presence of waveform changes across the fault, fault zone head waves (FZHW) and fault zone
trapped waves (FZTW). The fault zone phases are used to image properties of the fault bimaterial
interface and the core inner damage zone that acts as a seismic trapping structure.
In the next section (4.2) we give more detail on the array stations and data. The analysis
techniques and results are described in sections 4.3 and 4.4. Section 4.3 examines spatial variations
of polarizations, amplitudes and delay times of teleseismic data across the array. Section 4.4
includes analysis of waveform changes across the array and analyses of FZHW and FZTW based
on local earthquake waveforms. The results are summarized and discussed in relation to other
studies of fault zones and earthquake physics in section 4.5.

4.2 Instrumentation and data
The BS array comprises 108 three-component 5 Hz geophones installed along an
approximately fault-perpendicular line crossing the CF surface trace with 26 instruments placed
off the main line. Station BS55 is nearest to the CF trace and stations BS1 and BS108 are at the
northeastern and southwestern ends, respectively (Fig. 4.1b). Along the line, stations are 10 m
77
apart in a 400 m wide zone centered on the surface trace and spaced ~30 m to the northeast and
southwest of that zone. Only data from stations BS1 to BS108 are used in this study given our
focus on across-fault structural variations. The stations were installed on 11/21/15 and recorded
continuously at 1000 Hz sampling for 35 days.
The earthquake catalog of the Southern California Earthquake Data Center (SCEDC, 2013)
was used to extract P waveforms for all M>5 teleseismic earthquakes (within 30°-100°) deeper
than 500 km that occurred during the installation period. Four earthquakes matched these criteria
but two of them occurred within 60 km and 5 minutes of each other, so the second of these was
not considered. Waveforms for the 3 remaining events (Fig. 4.2a and Table 4.1) were extracted
from continuous BS recordings using the P wave arrival time at the BS55 location at sea level
predicted with TauP (Crotwell et al., 1999). After extraction the mean and trend were removed
and a low-pass Butterworth filter with corner frequency 2 Hz was applied.
The Hauksson et al. (2012) relocated catalog for Southern California (extended to later years)
was used to extract waveforms from local events. In total, 143 MW>1 earthquakes located within
a 110 km by 100 km box centered on the BS array and aligned with the CF trace occurred during
the installation period (Fig. 4.2b). For these events we use waveforms 5 s before and 30 s after the
origin times reported in the catalog. A band-pass Butterworth filter between 2 Hz and 20 Hz was
applied to these waveforms after the initial removal of any trend from the data. As illustrated in
Fig. 4.3a and later plots of waveforms, the long BS array provides a far better coverage of the fault
zone than the previously analyzed data recorded by the short BB array.

4.3 Teleseismic earthquake analyses
The early P waves from teleseismic events are analyzed for changes in polarization, amplitude
and arrival times. Polarization and amplitude will change in response to topographic and structural
changes that have length scales comparable to the long wavelengths (>1 km) of the teleseismic
energy (Jepsen and Kennett, 1990; Neuberg and Pointer, 2000). Arrival times corrected for
variations due to different propagation paths can be used to determine slowness changes at similar
large length scales (e.g., Schmandt and Clayton, 2013; Qiu et al., 2017). Here such analyses are
done to identify large-scale vertical (or near vertical) fault interfaces and regions of increased
slowness potentially corresponding to fault damage zones.

78

Table 4.1. Catalog of the teleseismic earthquakes analyzed.
Event SCEDC ID Origin time UTC Latitude Longitude Depth (km) Magnitude
TS1 37268101
2015/11/24,
13:21:35.800
18.7906 145.3115 586.20 6.00
TS2 37268269
2015/12/06,
17:09:28.800
-18.1903 -178.6798 544.30 5.80
TS3 37268157
2015/11/24,
22:45:38.000
-10.5484 -70.9038 600.60 7.60

4.3.1 Polarization and amplitude variations
For each of the 3 teleseismic events we use the vertical component to manually pick the first
coherent peaks (positive or negative) across the array and designate them P arrivals (P picks in
Figure 4.2. Data employed in the present study. a) Locations of the 3 M>5 teleseismic earthquakes
used (TS1-TS3). Results of data analysis from an example event (TS1) appear in Fig. 4.3.
Waveforms from events TS2 and TS3 are shown in Fig. A4.1. b) Locations of MW>1 earthquakes
(143 in total, circles) within a 110 km by 100 km region (big black box) centered on the BS array
(yellow triangle) and aligned with the CF surface trace. The small black rectangle highlights
earthquakes within 10 km of the CF. Waveforms from an example event (EVE1, red circle) are
shown in Fig. 4.5a. H=Hemet; A=Anza; PS=Palm Springs; SAF=San Andreas fault;
ECSZ=Eastern California Shear Zone. A depth section of events projected along the profile A-A'
is plotted at the bottom.

79
Fig. 4.3a). We then visually inspect the waveforms 2 s after these picks for abrupt changes in
character. For an example event TS1 (location in Fig. 4.2a and Table 1), most rapid changes in
arrival time and waveform are observed around station BS34 (270 m northeast of the CF trace,
Fig. 4.3a). Similar changes in arrival time and waveform are found near the same station for the
other 2 events (Fig. A4.1). A sharp change in arrival time and waveform character occurs near
station BS55 (location of CF trace) for event TS3 (Fig. A4.1b).
Next, we apply polarization analysis (Jurkevics, 1988) to all three components to quantify
waveform changes recorded across the array. The analysis employs a moving time window of
length 0.75 s (750 samples). For each window, the three components are combined in a 750×3
matrix and the covariance of that matrix is computed. The eigenvalues (𝜆
3
, 𝜆
V
, 𝜆
1
) and
eigenvectors (𝒖
3
, 𝒖
V
, 𝒖
1
) of the covariance matrix give, respectively, the amplitudes and directions
of the axes of the polarization ellipse. The largest eigenvalue 𝜆
3
corresponds to the amplitude of
maximum polarization with direction 𝒖
3
. For a P wave, direction 𝒖
3
gives the inclination angle as
Figure 4.3. Analysis of waveforms recorded by the BS array from an example teleseismic event.
a) Early P waveforms (velocity seismograms) generated by event TS1. All seismograms are
normalized by the same value. Noisy traces are removed. Red triangles denote manual P picks and
the black arrow points to an abrupt change in waveform character across traces recorded near
BS34. Waveforms recorded near BB array locations are colored purple with the southwestern most
and northeastern most ones denoted by BB01 and BB07. b)-c) Inclination of the largest eigenvector
and linearity calculated using a 0.75 s window. White arrows highlight abrupt changes in these
parameters.

80
𝐼𝑛𝑐 = cos
U3
(𝑢
33
). (4.1)
From the eigenvalues, the degree of polarization/linearity can be calculated using
𝑞 = 1−q(𝜆
V
+𝜆
1
)/2𝜆
3
x, (4.2)
with 𝑞 being 1 when the P wave is linearly polarized and 0 for perfectly uncorrelated motion. The
inclination and linearity are calculated for each 0.75 s window using Equations 4.1 and 4.2 and are
assigned to the time of the final sample in the moving window (time windows are trailing).
Application to the example event TS1 indicates large changes in inclination (+/- 30°, Fig. 4.3b)
and linearity (+/- 0.1, Fig. 4.3c) around station BS34. Large changes in inclination and linearity
are also observed near BS55. Similar changes in inclination and linearity are detected at BS34 and
BS55 for the other 2 events. Less prominent changes in inclination and linearity across the array
(for example near BS20 and BS80) correspond to smaller scale structures in the area that are left
unresolved.
The changes in arrival time and waveform character near BS34 correspond to lateral variations
in large-scale subsurface structure rather than sudden changes in topography, since similar
topographic gradient exists for stations BS1-55 (Fig. 4.4). The abrupt waveform change near
station BS55 is correlated with the mapped surface trace of the CF at the site (Sharp, 1967;
Rockwell et al., 2015). Figure 4.4a displays the maximum amplitude within a 1.5 s window starting
at each P pick for the 3 teleseismic events. A 1.5 s window is approximately equal to the largest
period observed for the early P waves. The results show that stations BS40-44 (110-150 m
northeast of the CF trace) consistently record the largest amplitude ground motion. The amplified
motion may be caused by a core fault damage zone beneath stations BS40-44 (e.g., Ben-Zion and
Aki, 1990; Lewis et al., 2005). The smaller scale changes in the results of this section are likely
related to secondary lateral variations of the fault zone structure and topography.

4.3.2 Delay time analysis
To focus on changes associated with the fault zone structure, arrival times are corrected in
two steps. For a given event, the travel times from source to stations without taking elevation into
account are predicted. Following this, travel times from sea level to the surface are computed using
a homogeneous reference velocity. The total predicted arrival times are then subtracted from the
picked times and the residual variations (delay times) between stations are used to infer subsurface
slowness. Stations with larger delay after correction overlie regions of greater slowness.
81
Figure 4.4. Amplitude and arrival time variations of teleseismic P waves across the array. a)
Maximum amplitudes within 1.5 s windows starting at the P picks for the 3 events (TS1 – solid
line; TS2 – dotted line; TS3 – dashed line, information in Table 4.1). Vertical black dot-dashed
lines represent the locations of stations BS55 (left) and BS34 (right) and the horizontal black dot-
dashed line depicts the median of the delay times shown. Gray line depicts changes in topography
across the array. b) Delay times for 3 events (same line styles as in (a)) and reference velocities of
2 km/s (red), 3 km/s (green), 4 km/s (blue) and 5 km/s (black) used during correction for
topography.

Arrival times at sea level for different stations are approximated using TauP (Crotwell et al.,
1999) and a 1D velocity model that is a combination of the IASP91 model for the mantle (Kennett,
1991) and the model of Hadley and Kanamori (1977) for the crust. Travel times from sea level to
the surface are calculated using the ray parameter for each event obtained from TauP and reference
velocities of 2 km/s, 3 km/s, 4 km/s and 5 km/s (Fig. 4.4b). After subtracting the calculated times
from the picked times, each resulting delay time profile is normalized by subtracting the current
maximum delay. This allows comparisons of profiles for different events and reference velocities.
82
There is generally good agreement among the normalized profiles, although the large topographic
gradient northeast of the CF produces a spread in delay time profiles for the different reference
velocity corrections (Fig. 4.4b). Irrespective of the spread between the curves, maximum delay
times are consistently observed for stations BS41-47 (90-140 m northeast of the CF trace). This
suggests greater slowness beneath these stations, which agrees overall with the maximum
amplitudes plotted in Fig. 4.4a and subsequent analysis of FZHW and FZTW in sections 4.4.2 and
4.4.3. Less prominent maxima in amplitude and delay time curves (for example 600-700 m
northeast of CF trace, Fig. 4.4) again suggest minor damage structures or other heterogeneities.
These features should be considered as potential targets for future geological/geophysical studies
around the Blackburn Saddle site.

4.4 Local earthquake analyses
4.4.1 Waveform visual inspection
We perform a similar visual inspection of waveforms from local events for evidence of lateral
changes in the subsurface structure beneath the array. The shorter wavelengths (higher
frequencies) of local arrivals allow observation of large-scale variations across the fault at higher
resolution compared to teleseismic arrivals.
Waveforms from local earthquakes (Fig. 4.2b) are first processed with an automatic algorithm
(Ross and Ben-Zion, 2014; Ross et al., 2016) to identify the onset times of P and S waves. We
then visually inspect early P and S waves 0.5 s after these automatic picks for anomalous changes
in character. Figure 4.5a shows vertical component waveforms from an example event EVE1
(location in Fig. 4.2b) and associated P and S picks. The P waveforms recorded southwest of
station BS34 show a 4-5 factor increase in maximum amplitude compared to P waveforms
recorded northeast of BS34. Similar to section 4.3.1, this abrupt change in waveform character
implies a lateral change in subsurface structure near BS34. Moreover, P peak amplitudes recorded
southwest of BS34 are associated with phases (right orange line in Fig. 4.5b) arriving ~0.2 s after
the P first arrivals (left orange line in Fig. 4.5b). The P first arrivals recorded at these stations are
potential FZHW while the trailing large amplitude phases are direct P waves. This suggests that
the structural change near BS34 is associated with a bimaterial interface. Following the S arrivals,
stations BS41-45 record an anomalous large amplitude wave packet (orange box in Fig. 4.5c).
These are potential FZTW associated with the core damage zone along the CF through Blackburn
83
Figure 4.5. Waveforms (velocity seismograms) generated by local event EVE1 (MW 1.9). a)
Waveforms recorded across the array with automatic P (red triangles) and S (blue triangles) picks
highlighting the arrivals of these respective phases. All seismograms are normalized by the same
value. Noisy traces are removed. The black dot-dashed lines show locations of the CF trace
(bottom line) and another large-scale structure to the northeast of the CF (top line, see section
4.3.1). Waveforms recorded near BB array locations are colored purple with the southwestern most
and northeastern most ones denoted by BB01 and BB07. b) Zoom in of early P arrivals at stations
BS20-50. The leftmost orange line represents manually picked P first arrivals across these stations
whereas the orange line to the right of it indicates the later arrival of large amplitude P phases at
stations southwest of ~BS34 only. c) Zoom in of S arrivals at stations BS29-59. The orange square
highlights an anomalous large amplitude wave packet recorded at stations BS41-45.
84
Saddle. Similar FZHW and FZTW are observed in waveforms from several other events and we
analyze them in greater detail below.

4.4.2 Fault zone head waves
4.4.2.1 Methodology
FZHW are critically refracted emergent phases that travel along a fault bimaterial interface
with the velocity of the faster medium (Ben-Zion, 1989). They arrive at stations in the slower
medium before the impulsive direct P waves and are observed up to a fault-normal critical distance
𝑥
l
given by
𝑥
l
= 𝑟∙tanqcos
U3
q𝛼
`
/𝛼
w
xx, (4.3)
where 𝑟, 𝛼
`
and 𝛼
w
denote the propagation distance along the fault (both along-strike and up-dip
direction) and average P wave velocities of the slower and faster media, respectively. The
differential time ∆𝑡 between FZHW and direct P wave for a station near the fault is given by
∆𝑡 = 𝑟/𝛼
w
∙𝜂/(1−𝜂)
, (4.4)

where 𝜂 = q𝛼
w
−𝛼
`
x/𝛼
w
is the fractional velocity contrast across the fault (Ben-Zion, 1989; Qiu
et al., 2017). Because FZHW radiate from the fault and direct P waves propagate from the
epicenter, FZHW have horizontal particle motion with a significant fault-normal component
(Bulut et al., 2012; Allam et al., 2014a; Share and Ben-Zion, 2016) compared to that of direct P
waves. For on-fault events, the FZHW and trailing direct P waves on the slow side of the fault
have opposite first motion polarities (Ben-Zion, 1989; 1990).

4.4.2.2 Results
We identify FZHW with an automatic detector (Ross and Ben-Zion, 2014) in conjunction with
visual inspection and analysis using the aforementioned criteria. The detector flags P waveforms
with an emergent phase followed by an impulsive arrival with a time separation between a
minimum value (0.065 s representing the width of a narrow P wave wiggle) and a maximum value
that depends on hypocentral distance. As an example, a maximum time separation of 0.8 s is
allowed over a distance of 40 km using the default parameters of the Ross and Ben-Zion (2014)
algorithm. The detector is applied similarly to previous linear SJFZ array studies (Qiu et al., 2017;
Share et al., 2017) using a slightly lower STA/LTA trigger threshold due to the lower SNR of BS
array waveforms compared to the other arrays.
85
Focusing first on the 75 events within 10 km of the CF trace (Fig. 4.6a), which have
magnitudes MW 1-3, the automatic detector flags 316 potential FZHW for the 108 stations
analyzed. The picked FZHW are from events located both northwest and southeast of the array
(Fig. 4.6a). Though all stations have potential detections, those southwest of BS34 have more than
double detections compared to stations northeast of BS34 (Fig. 4.6b). Some of the automatic picks
are expected to be false (Ross and Ben-Zion, 2014), and some may be associated with the edge of
the damage zone (Najdahmadi et al., 2016) or a basin (Qiu et al., 2017), rather than a deep fault
bimaterial interface that is the main imaging target here.
Based on visual inspection of the 75 events generating candidate FZHW, we find that events
with epicenters near the CF surface trace within the trifurcation area (having large along-fault
propagation distance) produce the clearest head waves. The locations of two example events HW1
and HW2 are shown in the inset of Fig. 4.6a. These events generate emergent (lower frequency,
Fig. A4.2a) low amplitude first P arrivals (candidate FZHW) at stations southwest of BS34 (Fig.
4.7a 1
st
and 2
nd
row), followed by larger amplitude impulsive arrivals (direct P waves) 0.1-0.3 s
later. The differential time between the identified FZHW and direct P arrivals is largest near BS34
and decreases towards BS108 in the southwest (Equation 4.3). Only impulsive P arrivals are
recorded at stations northeast of BS34. These observations imply that a deep fault bimaterial
interface is present below the surface location of station BS34.
The candidate FZHW and trailing direct P waves from near-fault events are substantiated by
comparing their moveouts across the array with those of nearby reference events that do not
produce FZHW propagating at depth. In the presence of a bimaterial fault, a reference event
located in the slower medium produces a direct P wavefront at locations on the slow side with
azimuth pointing to the epicenter. In contrast, direct P waves from a fast side reference event and
FZHW from an event near the interface are refracted along the fault so their arrivals on the slow
side have azimuth with a larger fault-normal component compared to the source-receiver azimuth.
If the fast side reference event is relatively close to a near-fault event, the moveout of the wavefront
across locations on the slow side from the former will be similar to that of critically refracted
FZHW generated by the latter. Likewise, moveouts of direct P wavefronts from slow side reference
and near-fault events will be similar. We therefore identify FZHW and trailing direct P waves as
arrivals with most similar moveouts to that of first P arrivals from fast side and slow side reference
events, respectively.
86

We choose events near the Coyote Creek (REF1, Figs. 4.6a inset and 4.7a 3
rd
row) and Buck
Ridge (REF2, Figs. 4.6a inset and 4.7a 4
th
row) faults as reference events on nominally slow and
fast sides of the CF, respectively (Allam and Ben-Zion, 2012). First arrival P waves from the
reference events are impulsive and large in amplitude at all stations (Fig. A4.2b). Compared to
REF1 (magenta line in Fig. 4.7a 3
rd
row), REF2 has P arrival times that increase more with fault
normal distance from BS34 towards stations in the southwest (cyan line in Fig. 4.7a 4
th
row). This
implies the wavefront from REF2 has the azimuth with larger fault normal component on the
nominally slow side of the fault. FZHW moveouts across stations southwest of BS34 from events
HW1 and HW2 are highly correlated with the moveout of P arrivals from REF2 (cyan lines Fig.
4.7a 1
st
and 2
nd
row). At the same stations, direct P wave moveouts from events HW1 and HW2
are approximately equal to the moveout observed for REF1 (magenta lines Fig. 4.7a 1
st
and 2
nd

row). These comparisons show that the observed FZHW and direct P waves are radiated from the
Figure 4.6. Identifying FZHW using an automatic detection algorithm. a) Locations of events
analyzed (circles inside black rectangle). Circle colors correspond to the number of stations with
FZHW detections (blue=no detections, pink=most detections). The inset shows a zoom in of
selected events located within the trifurcation area (dashed black box). FZHW are confirmed for
events HW1 and HW2 near the CF (large red circles) using automatic detection and visual
inspection (Fig. 4.7 1
st
and 2
nd
rows). Reference events near the Coyote Creek (CCF) (REF1,
magenta circle) and Buck Ridge (BRF) (REF2, cyan circle) faults that do not generate FZHW (Fig.
4.7 3
rd
and 4
th
rows). A depth section of events projected along the profile A-A' is plotted at the
bottom. b) Histogram of number of FZHW detections per station for events in (a). Dot-dashed
lines represent the locations of stations BS55 (left) and BS34 (right).

87
Figure 4.7. Visual inspection and polarization analysis of waveforms containing FZHW. (a) P
waveforms (velocity seismograms) generated by events HW1 (1
st
row), HW2 (2
nd
row), REF1 (3
rd

row) and REF2 (4
th
row). Seismograms are self-normalized and noisy traces are removed. Green
squares and red triangles depict automatic FZHW and P picks, respectively. Waveforms recorded
at locations southwest of BS34 (black arrows) from HW1 and HW2 contain FZHW while
waveforms from REF1 and REF2 do not. Magenta line in row 3 and cyan line in row 4 represent
manual picks for P first arrivals from events REF1 and REF2, respectively. Cyan and magenta
lines in rows 1 and 2 are the same as in rows 3 and 4 but shifted in time to align with the P first
arrival (candidate FZHW) and onset of the impulsive direct P wave at station BS55, respectively.
(b-d) The ratio of the largest eigenvalue between neighboring windows, inclination of the largest
eigenvector and linearity calculated within the leading window for events HW1 (1
st
row), HW2
(2
nd
row), REF1 (3
rd
row) and REF2 (4
th
row). Symbols and lines have the same meaning as in (a).
Dot-dashed box in 4
th
row depict waveforms with candidate local FZHW (see text for details).

88
fault and the source, respectively. In addition, the comparisons reveal that the differential time ∆𝑡
between FZHW and direct P wavefronts is larger for event HW2 compared to HW1 (colored lines
are farther apart for HW2 versus HW1). This observation and Equation 4.4 imply a continuous
deep bimaterial interface associated with the CF, since HW2 is located farther from the array than
HW1 (38 km versus 36 km).
Waveform characteristics of FZHW are quantified using polarization analysis as described in
section 4.3.1. Eigenvalues are computed using consecutive sliding windows each 0.07 s long that
overlap by 50%. In general, the largest eigenvalue of a window containing a FZHW will be larger
than a signal containing only noise, and the largest eigenvalue corresponding to a direct P wave
would be larger still (Allam et al., 2014a; Share et al., 2017). For example, eigenvalue ratios
computed between neighboring windows are >5 at all stations for first arrivals from events HW1
and HW2 (Fig. 4.7b 1
st
and 2
nd
row). Additionally, eigenvalue ratios >5 are observed at stations
southwest of BS34 upon later arrivals of direct P waves. Inclination of the largest eigenvector (Fig.
4.7c) and linearity (Fig. 4.7d) are also computed for the leading time window. First arrivals appear
as coherent <15° anomalies in inclination that persist for 0.1-0.15 s at stations northeast of BS34
for events HW1 and HW2 (Fig. 4.7c 1
st
and 2
nd
row). For stations southwest of BS34 and same
events these anomalies persist for up to 0.3-0.4 s as they represent both FZHW and direct P waves.
Coherent anomalies in linearity approximately equal to 1 similarly persist for longer times at
stations southwest of BS34 (Fig. 4.7d 1
st
and 2
nd
row). Similar anomalous eigenvalue ratios,
inclinations and linearity values are not observed for reference events REF1 and REF2 (Fig. 4.7b-
d 3
rd
and 4
th
row). The polarization analysis indicates that FZHW have clear arrivals that stand out
from the noise while the trailing direct P waves contain more energy than FZHW. The results are
consistent with theoretical expectations (Ben-Zion, 1989; 1990) and previous observational
studies. The analysis also shows, as expected, that both phases have near-vertical incidence and
are linearly polarized.
Following the discussed analysis steps, 14 events near the trifurcation area generating clear
FZHW propagating at depth are identified. In the case where events cluster only one event per
cluster is selected for further analysis. The selection process produces 5 events (Fig. 4.8a) with
distinctly different hypocentral distances, and waveforms from these events are integrated to
displacement prior to manual picking of FZHW and direct P arrivals. Using the picks and Equation
4.4 we estimate the average velocity contrast across the CF between the BS array and generating
89
events. FZHW picks are made on waveforms from stations southwest of BS34 where associated
emergent phases begin to rise above the noise level (green squares in Fig. 4.8b). Direct P waves
are picked (red triangles in Fig. 4.8b) using both a comparison of horizontal particle motion of
early P waves from reference event REF1 and waveforms containing FZHW (similar to Share et
al., 2017) and the moveout comparison discussed earlier using REF1 (Fig. 4.7). The obtained
average velocity contrast over the propagation paths associated with the 5 events and
representative station BS45 is 11.2% (Fig. 4.8c). This estimate assumes an average 𝛼
w
(Equation
4.4) of 6.1 km/s based on the tomographic results of Allam and Ben-Zion (2012). Using other
stations lead to similar average velocity contrasts. From the event locations in Fig. 4.8a, the 11.2%
velocity contrast represents an average value over the propagation path from the trifurcation area
and the top 10 km of the crust. The average velocity contrast in the shallow crust is likely to be
larger, since the velocity contrast typically decreases with depth (Ben-Zion et al., 1992; Lewis et
al., 2007). On the other hand, the average velocity contrast at depth closer to the array is about one
third of that value, as found by Share et al. (2017) using events closer to the (short BB) array and
events near the trifurcation area that are more off-fault.
Figure 4.8. Average velocity contrast across the CF from FZHW moveout analysis. a) Locations
of 5 events (large red circles) used during moveout analysis. The event closest to the array that
generates FZHW is highlighted (HW3) and also shown is the location of a reference event (large
red black-filled circle). A depth section of events projected along the profile A-A' is plotted at the
bottom. b) P displacement seismograms from event HW3 and associated manual direct P (red
triangles) and FZHW (green squares) picks (see text for picking procedure). Seismograms are self-
normalized and noisy traces are removed. c) Plot of displacement waveforms showing the linear
moveout of FZHW (green line and squares) relative to direct P waves (red triangles) recorded at
station BS45. The moveout corresponds to a 11.2% average velocity contrast across the fault. For
comparison, the waveform from the reference event in (a) is also plotted (y≈20 km).

90
The analysis of near-fault events also shows evidence for local FZHW propagating
exclusively along the edge of the fault damage zone (e.g., Najdahmadi et al., 2016). Stations
located within the low-velocity damage zone (Figs. 4.4 and 4.5c) have the most automatic
detections (Fig. 4.6b) with some detections associated with events that do not generate FZHW
propagating at depth. For example, event REF2 (Fig. 4.7 4
th
row) does not generate FZHW
traveling at depth, but stations BS34-65 record emergent P first arrivals followed by large
amplitude impulsive arrivals ~0.1 s later (dot-dashed rectangle in Fig. 4.7 4
th
row). The emergent
arrivals are potential local FZHW refracting along the contact between the damage zone and host
rock (at the faster speed of the latter), while the trailing impulsive phases are direct P waves
propagating within the damage zone.
To explore this further, we apply the automated head wave picker to the 68 events >10 km
from the CF and combine the results of all events to search for local FZHW. Only near-fault events
will generate FZHW propagating along a fault at depth. On the other hand, both near-fault and off-
fault events can produce local FZHW since the waves from all events propagate through the
damage zone before being recorded at surface stations above this zone. The automatic detector
flags 357 potential local FZHW for the 68 off-fault events analyzed. As in Fig. 4.6b, stations within
the damage zone have the most picks (Fig. 4.9a) that correspond to potential local FZHW. Stations
northwest and southeast of the damage zone have similarly low numbers of detections (compare
Figs. 4.6b and 4.9a), indicating (as expected) that none of the off-fault events generate FZHW
propagating along the CF at depth.
Visual inspection and further analysis establish several clear examples of local FZHW at
stations BS34-65 only. The locations of 15 such events and their respective waveforms recorded
at representative stations within (BS43) and outside (BS25 and BS81) the damage zone are shown
in Figs. 4.9b and c. The local FZHW are emergent phases (green line and squares) arriving before
the direct P waves (red triangles) recorded at station BS43 (Fig. 4.9c top). These phases are clearly
not FZHW propagating along the CF at depth because an approximately constant ∆𝑡 (~0.12 s) is
observed with increasing hypocentral distance. If we assume that the local FZHW propagate near-
vertically along the edge of the damage zone, ∆𝑡 can be used to approximate the average P velocity
contrast between damage zone and surrounding rock. Using Equation 4.4, a depth extent of 2 km
for the damage zone (see section 4.4.3.2) and velocities of 2-5 km/s for the surrounding rock
(similar to section 4.3.2) give estimated velocity contrasts of ~11-23%. This range is higher on
91

Figure 4.9. Analysis of local FZHW from off-fault events a) Histogram of number of FZHW
detections per station for events >10 km from CF (locations in Fig. 4.2b). Dot-dashed lines
represent the locations of stations BS55 (left) and BS34 (right). b) Locations of 15 example events
(large red circles) that produce FZHW propagating exclusively along the damage zone edge. A
depth section of events projected along the profile A-A' is plotted at the bottom. c) Self-normalized
P velocity seismograms from events in (b) and recorded by stations within (BS43, top) and
northeast (BS25, middle) and southwest (BS81, bottom) of the damage zone. Red triangles and
green squares show manual direct P and local FZHW picks and the green line depicts the average
early arrival time of the latter showing no apparent moveout.

92
average than the estimated contrast across the CF at depth as expected for a contrast between the
damage zone and host rock.

4.4.3 Fault zone trapped waves
4.4.3.1 Methodology
Spatially continuous low velocity fault damage zones act as waveguides and generate
constructive interference of S, P and noise phases that give rise to FZTW (e.g., Ben-Zion and Aki,
1990; Hillers et al., 2014; Qin et al., 2018). These phases have been observed in various fault and
geologic settings in California (Li et al., 1994; Lewis and Ben-Zion, 2010) including the SJFZ
using similar dense deployments (e.g., Qiu et al., 2017; Share et al., 2017) and elsewhere in the
world (e.g., Ben-Zion et al., 2003; Mizuno and Nishigami, 2006; Eccles et al., 2015). FZTW appear
on seismograms as high amplitude, long duration, anomalous frequency phases that follow direct
arrivals and are observed only at stations that are within or near the trapping structure (e.g., Ben-
Zion and Aki, 1990; Li and Leary, 1990; Lewis et al., 2005). Here we focus on identifying and
analyzing Love-type FZTW that follow the direct S wave (Ben-Zion, 1998).
Similar to previous studies (Qiu et al., 2017; Share et al., 2017) we use an automated detection
algorithm (Ross and Ben-Zion, 2015) to flag candidate FZTW and visual inspection to confirm
high quality candidates. The detection algorithm flags a station(s) if the recorded S waveform(s)
has dominant period, wave energy, absolute peak amplitude to average amplitude ratio, and
absolute peak delay (relative to S arrival) significantly larger than the median values of all stations
analyzed. The computations are done on vertical and fault-parallel component velocity
seismograms. After visual confirmation, waveforms from selected stations are inverted for
parameters of the trapping structure using a genetic inversion algorithm with a forward kernel
based on the analytical solution of Ben-Zion and Aki (1990) and Ben-Zion (1998). The
topographic gradient to the northeast of the CF (Fig. 4.4) is not accounted for by the analytical
solution, so the inversion results are likely to be less good than in previous applications at other
SJFZ arrays along flatter profiles (Qiu et al., 2017; Share et al., 2017; Qin et al., 2018).

4.4.3.2 Results
The automatic detection algorithm of Ross and Ben-Zion (2015) is applied on waveforms
from local 143 earthquakes within a broad region that includes the SJFZ, San Andreas and Elsinore
93
faults (Fig. 4.10a). A search window for detecting FZTW is defined using the S picks described in
section 4.4.1. The automatic detector flags 194 potential FZTW from events throughout the region
(Fig. 4.10a), with a maximum number of 4 picks made per event. The latter suggests FZTW are
most pronounced in waveforms recorded by ≥ 4 stations (as in Figs. 4.4 and 4.5c). Approximately
80% of all detections are made for stations northeast of the CF trace (Fig. 4.10b).
We next visually inspect flagged FZTW. Seismograms are integrated to displacement prior to
inspection and in preparation for waveform inversion. Subsequent to integration, the waveforms
are convolved with 1/t
1/2
to convert a point source response to that of an equivalent line dislocation
source (e.g., Igel et al., 2002; Ben-Zion et al., 2003). Several events producing FZTW are
confirmed through visual inspection and waveforms of two events (TW1 and TW2, Fig. 4.10a) are
selected for inversion of fault zone properties. FZTW from these events appear most clearly (Figs.
4.11a and A4.3a) on fault-parallel rotated waveforms at stations around BS45 (106 m northeast of
CF trace).
The forward model for the inversion consists of a fault zone layer sandwiched between two
quarter spaces (Fig. 2 in Ben-Zion and Aki, 1990). This basic model provides a useful modelling
approach because it captures the key average geometrical and material properties affecting FZTW
while accounting analytically for the strong trade-offs that exist between these parameters (Ben-
Zion, 1998). Various likely velocity gradients, internal scatterers and other small-scale
heterogeneities in the trapping structure have small effects on FZTW that average out these
heterogeneities (e.g., Igel et al., 1997; Jahnke et al., 2002). The inverted fault zone parameters are:
(1-3) S velocities of the two quarter spaces (assumed different based on section 4.4.2.2) and the
fault zone layer, (4-5) width and Q value of the fault zone layer, (6) location of contact between
the fault and left quarter space, and (7) propagation distance within the fault zone layer.
The genetic inversion algorithm maximizes the correlation between sets of synthetic
seismograms calculated with the forward model and observed waveforms, while exploring
systematically a large parameter-space. This is accomplished by calculating fitness values
associated with different sets of model parameters and migrating in the parameter-space overall in
the direction of larger fitness values. The fitness is defined as (1+C)/2 where C is the correlation
coefficient between synthetic and observed data. When C varies over the range –1 (perfect anti-
correlation) to 1 (perfect correlation), the fitness value changes from 0 to 1. Only waveforms from
stations BS34-58 (with noisy traces excluded) are inverted; that subset includes stations where
94
FZTW are recorded and a sufficient number of stations to the northeast and southwest with no
observed FZTW.
Synthetic waveforms (light blue lines in Fig. 4.11b) associated with the best-fit solution after
10,000 inversion iterations (black circles in Fig. 4.11c) are compared with recorded waveforms
(black lines in Fig. 4.11b) from TW1. Summing the fitness values of the final 2,000 inversion
iterations (green dots in Fig. 4.11c) and normalizing the results to have unit sums give probability
density functions for the various model parameters (curves in Fig. 4.11c). Corresponding inversion
results for event TW2 are shown in Fig. A4.3. Because FZTW are generated by the same
waveguide structure, inversions of waveforms associated with different high-quality candidates
provide similar values for the most likely fault zone parameters. However, the parameters are
subjected to significant trade-offs as discussed in previous studies (e.g., Ben-Zion, 1998; Peng et
al., 2003) and illustrated by the broad regions of high fitness values in Figs. 4.11c and A4.3c.
Based on the best-fit inversion results, the core damage zone is estimated to start at BS55
(main CF, Fig. 4.3b) and have width of ~150 m, Q value of ~40 and S velocity reduction of ~55%
Figure 4.10. Identifying FZTW using an automatic detection algorithm. a) Locations of the 143
events analyzed (circles inside black box). Circle colors correspond to the number of stations with
FZTW detections (blue=no detections, pink=most detections). Waveforms from events TW1 (Fig.
4.11) and TW2 (Fig. A4.3) are inverted using a genetic algorithm approach to determine fault zone
properties. A depth section of events projected along the profile A-A' is plotted at the bottom. b)
Histogram of number of FZTW detections per station for events in (a). Dot-dashed lines represent
the locations of stations BS55 (left) and BS34 (right).

95
Figure 4.11. Fault model inversion results for event TW1. a) Processed S waveforms (see text for
processing steps) for the array with the dot-dashed box highlighting waveform analyzed during
inversion. b) Waveforms for stations BS34-58 (black traces) compared to synthetic waveforms
(light blue traces) using fault parameters associated with the best-fit solution. The box shows the
most pronounced FZTW. c) Parameter space for the final 10 generations (2,000 iterations). Green
dots indicate fitness values for all solutions, black dots correspond to the best-fit solution and black
lines are probability density curves of the inverted parameters.
96
relative to the bounding rocks. Event TW1 is beneath the array (Fig. 4.10a), so energy from the
event propagates almost exclusively in the vertical direction and the best-fitting propagation
distance of ~2 km (Fig. 4.11c) reflects the depth extent of the core damage zone. On the other
hand, event TW2 is located in the trifurcation area and FZTW generated by this event propagates
a significant distance along-strike. Using for this event a propagation distance of 6 km (Fig. A4.3c)
and a trapping structure depth of 2 km (Fig. 4.11c), the horizontal propagation distance within the
waveguide from event TW2 is less than 6 km, considerably smaller than the epicentral distance of
TW2 from the array (40 km, Fig. 4.10a). This indicates that the trapping structure does not extend
continuously along-strike more than a few km, in agreement with inferences from other studies
(e.g., Peng et al., 2003; Lewis and Ben-Zion, 2010). For both TW1 and TW2, the inversion yields
an asymmetric damage zone centered northeast of the fault trace with near-zero probability of
asymmetry in the other direction. Though the reduction in shear velocity is similar, the average
velocity overall is higher for TW2, which is likely due to the difference in propagation distance
along-fault.

4.5 Discussion
The performed analyses provide high-resolution imaging results on the damage zone and
bimaterial interfaces in the structure of the Clark fault near Blackburn Saddle (BS) using a dense
long-aperture linear array across the fault. The BS array has an order of magnitude larger aperture
and number of stations than the earlier installed BB array (Share et al., 2017). Unlike other across-
fault arrays (e.g., Schmandt and Clayton, 2013; Ben-Zion et al., 2015), the BS array has three-
component nodes recording at 1000 Hz sampling. This allows applications of high-resolution
polarization analysis, accurate detection and modelling of S-type FZTW, and other tools that
require dense arrays with three components of motion.
The denser spatiotemporal sampling over a significantly longer transect across the fault has
several consequences for imaging the CF structure (Fig. 4.12). First, unlike the short BB array that
covers only part of the damage zone, the longer BS array provides data that can be used to recover
with teleseismic data (Fig. 4.4) the expected slowness profile across the entire damage zone
(compare red and black profiles in Fig. 4.12a top). Furthermore, comparisons between local P
earthquake waveforms recorded within and outside the damage zone allows recognition of two
different types of FZHW: head waves refracting along a deep bimaterial interface separating
97

Figure 4.12. Schematic representation of arrival time variations of early P waves across the fault
zone. a) Expected variations in P first arrivals from teleseismic earthquakes (top), direct P and
FZHW from on-fault local earthquakes (middle), and off-fault local earthquakes (bottom) given
the fault model in (b). The black lines represent profiles recorded by the BS array whereas the red
profiles denote the limited profiles observed with the shorter BB array. The vertical gray dot-
dashed line indicates abrupt changes in arrivals associated with a major bimaterial interface at the
site and detected with the BS array only. b) Simplified fault zone model suggested by analysis
done in this work (not to scale) with nominally slow block on the southwest (green) separated from
a fast block to the northeast (blue) by the main Clark fault (thick black line) and a low-velocity
damage zone near the surface (warm colors with warmer representing greater damage). Fault zone
parameters and example ray paths of several fault zone phases (white, gray and cyan lines) are
indicated. Triangles on the surface show the BS array spanning the entire fault zone and red
triangles mark the BB array covering only the southwestern side of the damage zone.
BS55 BS48 BS61
150 m
Slow Fast
BS34
FZHW/Local FZHW
FZTW
SW NE
~2 km
BS20 BS90
310 m
depth
time time time
Local earthquake
w/ FZHW
Teleseismic earthquake
Local earthquake
w/ local FZHW
FZHW
FZHW
local
Direct P
Direct P
Direct P
Direct P
Direct P
a
b
Clark fault
FZHW
Low velocity
zone
98
different crustal blocks across the CF (Fig. 4.12a middle) and local FZHW propagating along the
edge of the damage zone (Fig. 4.12a bottom). The local FZHW were not analyzed in the earlier
study using the BB array (Share et al., 2017) since that short array did not include stations outside
the damage zone (Fig. 4.12b).
The results indicate that the main seismogenic CF at depth is close to its mapped surface trace
(station BS55), and it manifests as abrupt changes in waveform character, inclination and linearity
of teleseismic P waveforms (Figs. 4.3b, c and A4.1b). The large amplitude and delay time
anomalies in teleseismic P arrivals (Fig. 4.4) and the presence of FZTW from local events (Figs.
4.5c, 4.11a and A4.3a) indicate an asymmetric damage zone (Fig. 4.12b) with a center 90-150 m
(BS40-47) from the surface trace of the CF. The large-scale structural variation inferred from
changes in waveforms recorded ~270 m northeast of the CF (BS34, Figs. 4.3, 4.5b and A4.1) is
most likely the northeastern edge of this damage zone (Fig. 4.12b). The analysis of teleseismic P
arrivals reveal potential minor faults and associated damage outside (>300 m from CF) the broader
damage zone. These features are unresolved in our analyses and could be targets for a future study
in the area.
A combination of high-resolution polarization, across- and along-fault moveout analyses
applied to waveforms of local events reveals FZHW at locations southwest of BS34 and not
northeast of that station (Fig. 4.7). These results suggest the broader damage zone is bounded to
the northeast by a sharp bimaterial interface separating structure to the northeast that is on average
faster than the structure to the southwest (Fig. 4.12b). The existence and sense of velocity contrast
are supported by tomographic models (Allam and Ben-Zion, 2012) and regional geology (e.g.,
Sharp, 1967; Gutierrez et al., 2010; Salisbury et al., 2012). The results also suggest that the
bimaterial interface merges with the CF at depth and is continuous for at least 50 km to the
southeast, as the observed FZHW are generated by events located within and beyond the
trifurcation area (Fig. 4.8a). This implies the CF is continuous at depth through the Anza seismic
gap and over a larger distance than previously estimated (Share et al., 2017), highlighting the
potential for large earthquake rupture through the region as suggested by paleoseismic studies
(Rockwell et al., 2015).
A P-wave velocity contrast in the range ~11-23% across the damage zone is calculated from
the differential time between local FZHW propagating at the edge of the damage zone and direct
P waves (Fig. 4.9c). This range of P velocity contrast corresponds to the S velocity contrast across
99
the trapping structure derived from analysis of FZTW, assuming Vp/Vs ratios in the range 2.95-
3.45 that may represent shallow damage rocks (e.g., O’Connell and Budiansky, 1974; Mavko et
al., 1998). Similar near-surface contrasts across the SJFZ are observed at a site ~50 km to the
southeast (Qiu et al., 2017). A smaller average velocity contrast of 11.2% across the CF at depth
is calculated from the moveout between FZHW and direct P waves with along-fault distance (Fig.
4.8c). These values are comparable to the contrast across the San Andreas fault south of Hollister
(McGuire and Ben-Zion, 2005) and sections of the Hayward fault (Allam et al., 2014a). A similar
contrast of 9.2% is derived from velocities 1.5 km on either side of the CF up to 10 km depth
(median depth of events generating FZHW) in the VP model of Allam and Ben-Zion (2012). The
smaller velocity contrast observed for the CF at depth closer to the array is, in turn, comparable to
results for the San Andreas fault around San Gorgonio Pass (Share and Ben-Zion, 2016) and North
Anatolian fault (Najdahmadi et al., 2016). The observed FZHW indicate the existence of a
continuous sharp bimaterial interface in the core structure of the CF, between the generating events
and the BS array (Fig. 4.8a), despite structural discontinuities at the surface (Sanders and
Magistrale, 1997) and irrespective of (lateral and/or vertical) changes in the velocity structure.
Properties of the fault damage zone are obtained from delay time analysis and inversions of
FZTW following S arrivals. The extent of the broader damage zone is inferred from the number of
stations that record teleseismic P arrivals with above average delay and local FZHW (sections
4.3.2 and 4.4.2.2). Relative to the median delay (horizontal dot-dashed line Fig. 4.4b), stations
BS35-61 consistently have larger delay times for all events and reference velocities considered.
These stations therefore overlie a region of above average slowness. The same set of stations also
record local FZHW (Fig. 4.7 4
th
row). This implies the damage zone edge(s) along which local
FZHW propagate locate on either side of these stations. We therefore estimate the width of the
broader damage zone to be comparable to the 310 m wide zone spanning stations BS35-61 (Fig.
4.12b). Inversions of waveforms with FZTW provide information on the fault zone trapping
structure within this broader damage zone. The results indicate (Figs. 4.11c and A4.3c) a trapping
structure that extends <6 km along-strike and to a depth of 2 km based on the best-fit propagation
distances from the FZTW inversions. The trapping structure is about 150 m wide and has S velocity
reduction of 55% and Q value of about 40. These parameters are similar to properties of trapping
structures along other sections of the SJFZ (Lewis et al., 2005; Qiu et al., 2017; Qin et al., 2018),
San Andreas fault at Parkfield (Lewis and Ben-Zion, 2010), Karadere branch of the North
100
Anatolian fault (Ben-Zion et al., 2003) and other active strike-slip faults and rupture zones. The
larger aperture BS array (compared to BB, Figs. 4.3a and 4.5a) allows for better resolution of the
trapping structure within the broader damage zone, and the properties estimated here (Figs. 4.11c
and A4.3c) are therefore more accurate than those reported in Share et al. (2017).
The sense of velocity contrast across the CF at depth towards the southeast (Fig. 4.8) and
results on bimaterial ruptures (e.g., Weertman, 1980; Andrews and Ben-Zion, 1997; Brietzke et
al., 2009; Shlomai and Fineberg, 2016) suggest that earthquakes in the area tend to propagate to
the northwest. This is consistent with studies on directivity of small to moderate events along the
CF between the trifurcation area and the BS array (Kurzon et al., 2014; Ross and Ben-Zion, 2016).
Persistent occurrence of bimaterial ruptures with preferred propagation direction is expected to
produce more damage on the side with faster velocity at depth (Ben-Zion and Shi, 2005). This is
in agreement with geological mapping of Dor et al. (2006b) near Anza, and the observed damage
asymmetry across the CF (Figs. 4.4, 4.5, 4.9 and 4.11) summarized in Fig. 4.12b. A recent example
of both rupture directivity and damage asymmetry across the CF is provided by the June 2016 MW
5.2 Borrego Springs earthquake, where rupture initiated in the trifurcation area, propagated to the
northwest and generated significantly more aftershocks (damage) northeast of the fault (Ross et
al., 2017a; 2017b).
The results of this work are obtained by a set of separate analyses that examine different
phases with different techniques and largely ignore the topography (other than travel time
corrections). A more complete analysis that merges the detailed fault zone structure within a larger-
scale tomographic model and accounts for topography may be done with 3D numerical simulations
of the type explored by Allam et al. (2015). This approach is computationally very demanding but
is currently feasible. Additional information about the fault zone structure may be obtained by
analyzing the ambient seismic noise recorded by the dense linear array (e.g., Hillers et al., 2014).
These studies may be the subject of future work.

101
Discussion
I use several seismic signals and techniques to reveal multi-scale geometric and material
properties of the San Jacinto (SJF) and San Andreas (SAF) faults in Southern California. In
addition to structural properties, the results provide important information on the likely size,
propagation direction and location of large earthquakes along the SAF and SJF.
Double-difference tomography (Chapter 1: Share et al., 2018a) applied at a regional scale (1-
100 km) shows spatial variations in damage zones between several fault segments in Southern
California and prominent bimaterial structures associated with the northern SJF and southern SAF.
Depending on location and depth, the imaged damage zones have characteristically low or high
VP/VS. This attests to heterogeneity in the region of factors that control VP/VS, namely; crack
geometry, fluid content, lithology, temperature, etc. (e.g., O’Connell and Budiansky, 1974;
Christensen, 1996; Bocher, 2005; Shinevar et al., 2017). Based on the imaged damage zones and
bimaterial interfaces, I infer that the northern SJF and SAF near San Bernardino are near vertical,
whereas the SAF segment near Coachella Valley dips to the northeast. My tomography results also
reveal significant changes in elastic properties at various depths across San Gorgonio Pass (SGP).
These results suggest that structural complexity of the SAF at the surface around SGP continues
to great depth.
Head wave (FZHW) analysis of earthquakes around SGP (Chapter 2: Share and Ben-Zion,
2016) corroborates structural complexity of the SAF system at depth. The analysis reveals two
separate bimaterial interfaces, one northwest and the other southeast of SGP, with opposite
velocity contrast polarities across the SAF. This polarity flip combined with expected behavior of
bimaterial ruptures (e.g., Weertman, 1980; Ben-Zion and Andrews, 1998; Ampuero and Ben-Zion,
2008; Shlomai and Fineberg, 2016) show that typical subshear earthquakes along the SAF through
SGP are unlikely. Specifically, the results reveal that ruptures that enter SGP from the northwest
or southeast would likely be arrested by the reversal of the velocity contrast they would encounter
with continuing propagation. The SGP region is therefore not only a "geometrical knot" (Yule,
2009) but also a "dynamical knot" for earthquake ruptures. Furthermore, the FZHW results also
support the proposed northeast dip of the SAF near Coachella Valley. The earthquakes that
generate FZHW along the SAF in the area are located northeast of the mapped surface fault trace
at depth. This observation, in combination with the sensitivity of FZHW with increasing fault-
normal distance of the generating earthquake (Chapter A3), shows these earthquakes are most
102
probably near the SAF at depth. This supports the northeast dip of the SAF in the area (e.g., Fialko,
2006; Fuis et al., 2017; Share et al., 2018a).
Properties of the SJF are imaged at high-resolution (~100 m-1 km) using analyses of data from
dense linear arrays crossing the fault (Chapters 3-4: Share et al., 2017; Share et al., 2018b). FZHW
studies reveal a long (~60 km), continuous segment of the central SJF with across-fault VP
contrasts as large as 10% (Share et al., 2018b). This is consistent with historical records of large
earthquakes along this fault (Rockwell et al., 2015). The analysis also shows structures to the
southwest of this segment generally have lower velocities, which, based on expected properties of
bimaterial ruptures, predicts earthquakes have preferred propagation direction to the northwest.
This northwest propagation will produce significantly larger ground motion in that direction
compared to the opposite direction (e.g., Meng et al., 2017). Travel-time changes from earthquakes
recorded across these arrays and analysis of local FZHW show the broader damage zone at the site
is at least 300 m wide (Share et al., 2018b). Moreover, modeling of trapped waves (FZTW) reveals
a 100-200 m wide and <5 km deep core damage zone within this broader damage zone (Share et
al., 2017; Share et al., 2018b). Most of this damage is located northeast of the seismogenic SJF.
Similar sized asymmetric core damage zones are observed at several other locations along the SJF
(Qiu et al., 2017; Qin et al., 2018). These fault properties are consistent with preferred propagation
direction of earthquake ruptures in the area to the northwest (Ben-Zion and Shi, 2005).
The imaged along-fault heterogeneities, fault geometries, bimaterial structures and damage
asymmetry should be incorporated in future seismic hazard assessments. This will facilitate more
accurate predictions on the potential and amount of ground motion expected from large
earthquakes along the SAF and SJF.
Some potential topics for future research on the Southern California plate boundary are: 1) A
FZHW study focused on the imaged bimaterial structures along the northern SJF. This will inform
whether the SJFZ and SAF are continuous at depth and the potential for large earthquake ruptures
involving both faults (e.g., Lozos, 2016). 2) Joint inversion of geophysical data recorded in the
area. The seismic imaging results should be combined with other geophysical results in a
quantitative manner to better constrain, for example, the competing factors that influence VP/VS
near faults. Two suitable geophysical datasets to consider are electrical conductivity measurements
from electrical and electromagnetic techniques (e.g., Becken and Ritter, 2012), and, measurements
of surface displacements from geodesy (e.g., Smith-Konter et al., 2011; Lindsey and Fialko, 2013;
103
Donnellan et al., 2015). Both these datasets are sensitive to some or all of the factors influencing
elastic properties but will complement the seismic results, because of inherently different structural
resolution. 3) Machine learning applied to seismic phase detection. The combination of automatic
detection and visual inspection to accurately detect FZHW and FZTW (Share and Ben-Zion, 2016;
Share et al., 2017; Share et al., 2018b), allows for the rapid accumulation of large sets of
waveforms containing these unique phases. These large datasets should in future be incorporated
in machine learning applications (e.g., convolutional neural networks, Mnih et al., 2015) aimed at
further enhancing detection capabilities of these and other seismic phases. This will dramatically
improve our ability to analyze and image fault structures in a rapid and robust manner, and, further
our understanding of the role faults play in the plate tectonic framework.

104
A1. Chapter 1 supplementary figures

Figure A1.1. Discretization of the model space horizontally (left) and vertically (right). Grid nodes
are represented by green squares/rectangles and are spaced 1 km apart within the region centered
on SGP (X=-66 km to 76 km and Y=-30 km to 42 km) and down to 15 km depth and as much as
2 km outside that region.

105

Figure A1.2. Checkerboard recovery results at 10 km depth (VP/VS). a) Checkerboard model with
a constant VP/VS=1.73. Recovered models for the case where VP and VS checkerboard anomalies
span 6 grid nodes (Figs. 1.2b and 1.3b) and noise-free (b) and noisy (c) synthetic data. d) Resultant
model after 3 by 3 median filter is applied to (c). Dark blue contours depict DWS=20 and magenta
contours show 𝑟
`
=0.008.

106

Figure A1.3. Checkerboard recovery results in cross-section at X=8 km (VP/VS). a) Checkerboard
model with a constant VP/VS=1.73. Recovered models for the case where VP and VS checkerboard
anomalies span 6 grid nodes (Figs. 1.4b and 1.5b) and using noise-free (b) and noisy (c) synthetic
data. d) Resultant model after 3 by 3 median filter is applied to (c). Dark blue contours depict
DWS=20 and magenta contours show 𝑟
`
=0.008.
107

Figure A1.4. Harvard (a), SCEC (b) and Fang2016 (c) VP (left) and VS (right) models at 13 km
depth.
108

Figure A1.5. Same as Fig. 1.8 but now instead showing 𝑟
`
=0.008 contours (magenta lines) and
regions with 𝑟
`
<0.008 shaded.
109

Figure A1.6. Same as Fig. 1.10 but now instead showing 𝑟
`
=0.008 contours (magenta lines) and
regions with 𝑟
`
<0.008 shaded.

110

Figure A1.7. Same as Fig. 1.11 but now instead showing 𝑟
`
=0.008 contours (magenta lines) and
regions with 𝑟
`
<0.008 shaded.

111

Figure A1.8. Same as Fig. 1.12 but now instead showing 𝑟
`
=0.008 contours (magenta lines) and
regions with 𝑟
`
<0.008 shaded.
112
A2. Chapter 2 supplementary figures

Figure A2.1. Vertical component velocity seismograms for events and the five westernmost
stations highlighted in Fig. 2.2 (top) of the main text.
−0.5 0 0.5
−50
−40
−30
−20
−10
0
10
20
30
P body wave pick
−0.5 0 0.5
−60
−50
−40
−30
−20
−10
P body wave pick
−0.5 0 0.5
−60
−50
−40
−30
−20
−10
0
10
20
P body wave pick
−0.5 0 0.5
−90
−80
−70
−60
−50
−40
−30
P body wave pick
−0.5 0 0.5
−80
−70
−60
−50
−40
−30
−20
−10
0
SNO
Time relative to aligned P phase (sec)
Hypocentral distance W (−) and E (+) (km) Hypocentral distance W (−) and E (+) (km)
P body wave pick
FZHW pick
Hypocentral distance W (−) and E (+) (km)
MSC
5442
FHO SVD
Time relative to aligned P phase (sec)
113

Figure A2.2. Vertical component velocity seismograms for events and the five easternmost
stations highlighted in Fig. 2.3 (top) of the main text.
−0.5 0 0.5
30
40
50
60
70
80
90
100
P body wave pick
−0.5 0 0.5
20
30
40
50
60
70
80
P body wave pick
−0.5 0 0.5
10
15
20
25
30
35
40
45
50
P body wave pick
−0.5 0 0.5
−30
−20
−10
0
10
20
30
40
P body wave pick
−0.5 0 0.5
0
10
20
30
40
50
60
70
Time relative to aligned P phase (sec)
Hypocentral distance W (−) and E (+) (km)
P body wave pick
FZHW pick
Hypocentral distance W (−) and E (+) (km) Hypocentral distance W (−) and E (+) (km)
MSC
WWC DEV
SNO FHO
Time relative to aligned P phase (sec)
114
A3. Sensitivity of fault zone head waves to the fault-normal distance of
generating events
In the majority of studies using fault zone head waves (FZHW, e.g., McGuire and Ben-Zion,
2005; Bulut et al., 2012; Najdahmadi et al., 2016; Share et al., 2017), the research focus is phase
detection and validation, followed by inferences about the continuity of the fault under
investigation and the velocity contrast across it. FZHW are highly sensitive to these fault properties
(Allam et al., 2015), and the extraction of such information is more than sufficient for most
research purposes. However, additional information can be revealed through FZHW analysis. One
example is the geometry of a bimaterial fault. FZHW, and the direct and reflected phases (for
example) that accompany them, show characteristic changes in amplitude, arrival time, polarity,
frequency, etc., depending on the distance of the generating event from the fault interface.
Therefore, it should in principal be possible to determine, for example, the location of a bimaterial
fault patch at depth by examining the changes in these recorded waveform characteristics from
clustered earthquakes surrounding the fault. Determining the location of fault patches is especially
useful in cases where complex fault geometries are suspected, such as the variable dip of the
southern San Andreas Fault (SAF) through San Gorgonio Pass (Fuis et al., 2012; Fuis et al., 2017).
Share and Ben-Zion (2016) note the predominant locations of earthquakes northeast of the
SAF generating FZHW suggest the SAF is located northeast of its mapped surface trace at depth
(Fig. 2.3). This implies the SAF dips in that direction. Here, I explore this finding with a brief
study on the sensitivity of FZHW to the relative locations of generating earthquakes to a bimaterial
fault.
Using the analytical solution of Ben-Zion (1998), I generate synthetic velocity waveforms for
earthquakes near a bimaterial fault using the following experimental setup. An event at 10 km
depth generates SH-type waves that propagate along a fault with a 10% velocity contrast and are
recorded by a linear array of stations at the surface. The fault-normal distance of the event is
changed gradually from 10 km on the fast side (negative direction in employed coordinate system)
of the fault to 10 km on the slow side (positive direction in employed coordinate system) of the
fault. The linear array has 20 km aperture with 100 m spacing in a 1 km zone centered on the fault
and 500 m spacing outside that zone. Both the fast and slow side quarter spaces have density= 2.7
g/cm
3
and Q=1000. Waveforms generated from example events located at -500 m, 0 m and 500 m
are shown in Figs. A3.1, A3.2 and A3.3, respectively. The first maxima or minima in amplitude
115
following the predicted arrival times of the respective phases (colored asterisks in Figs. A3.1-A3.3)
are recorded for all event and station locations and are summarized in Fig. A3.4. In terms of
amplitudes and phase types in the results, the following characteristics stand out:
1) Events on the fast side of the fault at depth generate in addition to direct arrivals (both
sides) also reflected arrivals (blue asterisks in Fig. A3.1) at locations on the fast side.
2) Events on the fast side and very near the fault at depth generate an anomalously low
amplitude direct arrival (red asterisks in Fig. A3.1) at near-fault stations atop the slow side
of the fault. These amplitudes are 50-90% less than those observed at the same stations but
produced by near-fault events on the slow side of the fault (Fig. A3.4).
3) As expected, an on-fault event generates FZHW (green asterisks in Fig. A3.2) that are only
observed at locations atop the slow side and within a critical distance from the fault that
depends on the velocity contrast and along-fault propagation distance (Ben-Zion, 1989).
This event also generates direct arrivals within the respective fast and slow blocks.
4) Off-fault events on the slow side at depth generate in addition to direct arrivals (both sides)
also reflected arrivals (blue asterisks in Fig. A3.3) at locations on the slow side.
Figure A3.1. Waveforms (velocity) generated by an earthquake located at 10 km depth and 500
m from the fault on the fast side. Asterisks depict, respectively, the first maxima or minima
following the predicted arrival times of direct (red) and reflected (blue) P phases.

116

5) The same events also generate FZHW at slow side locations. However, there is a dramatic
decrease in FZHW amplitudes with increasing event fault-normal distance. Events 100 m
from the fault generate FZHW with amplitudes reduced by ~75% compared to FZHW
produced by on-fault events (Fig. A3.4). Events located beyond 100 m have further reduced
FZHW amplitudes.
6) Events on the slow side at and beyond the critical distance (Ben-Zion, 1989) do not produce
FZHW at any stations.
Tests are made with several other combinations of event depth and velocity contrast across
the fault. Using longer propagation distances and/or larger velocity contrasts, for example,
increases the duration of the FZHW and critical distance from the fault. However, the rapid
decrease in FZHW amplitude with increasing event distance from the fault is observed for all
combinations of parameters tested. The same holds for the low amplitude direct arrivals observed
at near-fault locations on the slow side produced by near-fault events on the fast side at depth. It
is worth noting that the rapid decrease in FZHW amplitude with increasing event distance from
the fault is achieved using an unrealistically high Q=1000 for a near-fault region. Lowering this to
Figure A3.2. Waveforms (velocity) generated by an on-fault earthquake at 10 km depth. Asterisks
depict, respectively, the first maxima or minima following the predicted arrival times of direct P
(red) and FZHW (green) phases.

117
a more appropriate value (e.g., Q<200) will lead to an even more dramatic decrease in FZHW
amplitude with fault-normal distance. Given the noise levels in recorded waveforms, this suggests
a low likelihood that FZHW generated by events far from a bimaterial fault on the slow side will
be observed at a surface station atop the slow block. Therefore, FZHW detected in the recorded
data are most probably only produced by events near bimaterial faults along which they propagate.
This supports the northeast dip of the southern SAF near Coachella Valley as suggested by Share
and Ben-Zion (2016) and others (e.g., Fialko, 2006; Fuis et al., 2012; Lindsey and Fialko, 2013;
Fuis et al., 2017; Share et al., 2018a).

Figure A3.3. Waveforms (velocity) generated by an earthquake located at 10 km depth and 500
m from the fault on the slow side. Asterisks depict, respectively, the first maxima or minima
following the predicted arrival times of direct (red) and reflected (blue) P phases, and FZHW
(green).

118
Figure A3.4. Amplitude variations of FZHW (black solid lines), direct (colored solid lines) and
reflected (colored dotted lines) arrivals associated with generating earthquakes (sources, y-axis)
and recording stations (x-axis) different normal distances from the fault. Figs. A3.1-A3.3 show the
extraction of these amplitudes from waveforms generated by fast-side, on-fault and slow-side
events, respectively. All amplitudes are normalized by the same value, however, normalized
FZHW values are multiplied by 30 to make their associated low-amplitudes comparable to the
amplitudes of other phases.

-5 0 5
Station distance from fault on fast(-) and slow(+) side (km)
-5
-4
-3
-2
-1
0
1
2
3
4
5
Source distance from fault on fast(-) and slow(+) side (km)
Velocity contrast=10% Propagation distance=10 km
fast side direct wave
fast side reflected wave
slow side direct wave
slow side reflected wave
FZHW
119
A4. Chapter 4 supplementary figures
Figure A4.1. Early P waveforms (velocity seismograms) generated by events TS2 (a) and TS3
(b). All seismograms are normalized by the same value and noisy traces are removed. Red triangles
denote manual P picks and black arrows point to abrupt changes in waveform character.

120
Figure A4.2. Normalized amplitude spectra of P waveforms from events HW1 (a) and REF1 (b).
Highlighted are frequencies associated with impulsive direct P waves recorded across the array (a
and b, gray dot-dashed box) and anomalous low frequencies associated with emergent FZHW
observed at stations in the southwest only (a, magenta dot-dashed box). FZHW are most easily
distinguished from direct P waves at stations BS61-108, which locate outside the broader damage
zone (BS35-61), as waveforms recorded there are not contaminated as much by other fault zone
phases.

121
Figure A4.3. Fault model inversion results for event TW2. The layout is the same as in Fig. 4.11.

122
References
Allam, A. A., Ben-Zion, Y., and Peng, Z., 2014a, Seismic imaging of a bimaterial interface along
the Hayward fault, CA, with fault zone head waves and Direct P arrivals: Pure and Applied
Geophysics, v. 171(11), p. 2993-3011.
Allam, A. A., Ben-Zion, Y., Kurzon, I., and Vernon, F., 2014b, Seismic velocity structure in the
Hot Springs and Trifurcation areas of the San Jacinto fault zone, California, from double-
difference tomography: Geophysical Journal International, v. 198, p. 978-999.
Allam, A. A., Tape, C., and Ben-Zion, Y., 2015, Finite-frequency sensitivity kernels of seismic
waves to fault zone structures: Geophysical Journal International, v. 203, p. 2032-2048.
Allam, A., and Ben-Zion, Y., 2012, Seismic velocity structures in the southern California plate
boundary environment from double difference tomography: Geophysical Journal
International, v. 190, p. 1181-1196.
Allen, C., 1957, San Andreas fault zone in San Gorgonio Pass, southern California: Geological
Society of America Bulletin, v. 68, p. 315-350.
Ampuero, J. P., and Ben-Zion, Y., 2008, Cracks, pulses and macroscopic asymmetry of dynamic
rupture on a bimaterial interface with velocity-weakening friction: Geophysical Journal
International, v. 173(2), p. 674-692.
An, M., 2012, A simple method for determining the spatial resolution of a general inverse problem:
Geophysical Journal International, v. 191, p. 849-864.
Anderson, M., Matti, J., and Jachens, R., 2004, Structural model of the San Bernardino basin,
California, from analysis of gravity, aeromagnetic and seismicity data: Journal of
Geophysical Research, v. 109, p. B04404.
Andrews, D. J., and Ben-Zion, Y., 1997, Wrinkle-like slip pulse on a fault between different
materials: Journal of Geophysical Research, v. 102, p. 553-571.
Avallone, A., Rovelli, A., Di Giulio, G., Improta, L., Ben-Zion, Y., Milana, G., and Cara, F., 2014,
Waveguide effects in very high rate GPS record of the 6 April 2009, Mw 6.1 L’Aquila,
central Italy earthquake: Journal of Geophysical Research, v. 119, p. 490-501.
Axen, G. J., and Fletcher, J. M., 1998, Late Miocene-Pleistocene extensional faulting, northern
Gulf of California, Mexico and Salton Trough, California: International Geology Review,
v. 40(3), p. 217-244.
123
Bailey, I. W., Becker, T. W., and Ben-Zion, Y., 2009, Patterns of Co-seismic Strain Computed
from Southern California Focal Mechanisms: Geophysical Journal International, v. 177, p.
1015-1036, doi: 10.1111/j.1365-246X.2009.04090.x.
Bailey, I. W., Ben-Zion, Y., Becker, T. W., and Holschneider, M., 2010, Quantifying focal
mechanism heterogeneity for fault zones in central and southern California: Geophysical
Journal International, v. 183(1), p. 433-450.
Bankey, V., Cuevas, A., Daniels, D., Finn, C. A., Hernandez, I., Hill, P., Kucks, R., Miles, W.,
Pilkington, M., Roberts, C., Roest, W., Rystrom, V., Shearer, S., Snyder, S., Sweeney, R.,
Velez, L., Phillips, J. D., and Ravat, D., 2002, Digital data grids for the magnetic anomaly
map of North America: U.S. Geological Survey Open-File Report 02-414.
Barak, S., Klemperer, S. L., and Lawrence, J. F., 2015, San Andreas Fault dip, Peninsular Ranges
mafic lower crust and partial melt in the Salton Trough, Southern California, from ambient-
noise tomography: Geochemistry Geophysics Geosystems, v. 16, p. 3946-3972.
Becken, M., and Ritter, O., 2012, Magnetotelluric studies at the San Andreas Fault Zone:
Implications for the role of fluids: Surveys in Geophysics, v. 33(1), p. 65-105.
Becker, T. W., Hardebeck, J. L., and Anderson, G., 2005, Constraints on fault slip rates of the
Southern California plate boundary from GPS velocity and stress inversions: Geophysical
Journal International, v. 160, p. 634-650.
Bekas, C., Kokiopoulou, E., and Saad, Y., 2007, An estimator for the diagonal of a matrix: Applied
Numerical Mathematics, v. 57(11-12), p. 1214-1229.
Ben-Zion, Y., 1989, The response of two joined quarter spaces to SH line sources located at the
material discontinuity interface: Geophysical Journal International, v. 98, p. 213-222.
Ben-Zion, Y., 1990, The response of two half spaces to point dislocations at the material interface:
Geophysical Journal International, v. 101, p. 507-528.
Ben-Zion, Y., 1998, Properties of seismic fault zone waves and their utility for imaging low-
velocity structures: Journal of Geophysical Research, v. 103, p. 12567-12585.
Ben-Zion, Y., 2001, Dynamic Rupture in Recent Models of Earthquake Faults: Journal of the
Mechanics and Physics of Solids, v. 49, p. 2209-2244.
Ben-Zion, Y., and Aki, K., 1990, Seismic radiation from an SH line source in a laterally
heterogeneous planar fault zone: Bulletin of the Seismological Society of America, v. 80,
p. 971-994.
124
Ben-Zion, Y., and Andrews, D. J., 1998, Properties and implications of dynamic rupture along a
material interface: Bulletin of the Seismological Society of America, v. 88(4), p. 1085-
1094.
Ben-Zion, Y., and Huang, Y., 2002, Dynamic rupture on an interface between a compliant fault
zone layer and a stiffer surrounding solid: Journal of Geophysical Research, v. 107(B2), p.
2042.
Ben-Zion, Y., and Malin, P., 1991, San Andreas fault zone head waves near Parkfield, California:
Science, v. 251, p. 1592-1594.
Ben-Zion, Y., and Sammis, C. G., 2003, Characterization of Fault Zones: Pure and Applied
Geophysics, v. 160, p. 677-715.
Ben-Zion, Y., and Shi, Z., 2005, Dynamic rupture on a material interface with spontaneous
generation of plastic strain in the bulk: Earth and Planetary Science Letters, v. 236(1), p.
486-496.
Ben-Zion, Y., Katz, S., and Leary, P., 1992, Joint inversion of fault zone head waves and direct P
arrivals for crustal structure near major faults: Journal of Geophysical Research, v. 97, p.
1943-1951.
Ben-Zion, Y., Peng, Z., Okaya, D., Seeber, L., Armbruster, J. G., Ozer, N., Michael, A. J., Baris,
S., and Aktar, M., 2003, A shallow fault-zone structure illuminated by trapped waves in
the Karadere–Duzce branch of the North Anatolian Fault, western Turkey: Geophysical
Journal International, v. 152(3), p. 699-717.
Ben-Zion, Y., Rockwell, T., Shi, Z., and Xu, S., 2012, Reversed-polarity secondary deformation
structures near fault stepovers: Journal of Applied Mechanics, v. 79, p. 031025.
Ben-Zion, Y., Vernon, F., Ozakin, Y., Zigone, D., Ross, Z., Meng, H., White, M., Reyes, J., Hollis,
D., and Barklage, M., 2015, Basic data features and results from a spatially dense seismic
array on the San Jacinto fault zone: Geophysical Journal International, v. 202, p. 370-380.
Bennington, N. L., Thurber, C., Peng, Z., Zhang, H., and Peng, Z., 2013, Incorporating fault zone
head wave and direct wave secondary arrival times into seismic tomography: Application
at Parkfield, California: Journal of Geophysical Research, v. 118, p. 1008-1014.
Blisniuk, K., Oskin, M., Me Riaux, A.-S., Rockwell, T., Finkel, R. C., and Ryerson, F. J., 2013,
Stable, rapid rate of slip since inception of the San Jacinto fault, California: Geophysical
Research Letters, v. 40, p. 4209-4213.
125
Bocher, T. M., 2005, Empirical Relations between Elastic Wavespeeds and Density in the Earth’s
Crust: Bulletin of the Seismological Society of America, v. 95(6), p. 2081-2092.
Bogiatzis, P., Ishii, M., and Davis, T. A., 2016, Toward using direct methods in seismic
tomography: Computation of the full resolution matrix using high performance computing
and sparse QR factorization: Geophysical Journal International, v. 205(2), p. 830-836.
Brietzke, G. B., and Ben-Zion, Y., 2006, Examining tendencies of in-plane rupture to migrate to
material interfaces: Geophysical Journal International, v. 167, p. 807-819.
Brietzke, G. B., Cochard, A., and Igel, H., 2009, Importance of bimaterial interfaces for earthquake
dynamics and strong ground motion: Geophysical Journal International, v. 178, p. 921–
938.
Brocher, T. M., 2005, Empirical relations between elastic wavespeeds and density in the Earth’s
crust: Bulletin of the Seismological Society of America, v. 95(6), p. 2081-2092.
Bruhn, R., Parry, W., Yonkee, W., and Thompson, T., 1994, Fracturing and hydrothermal thermal
alteration in normal fault zones: Pure and Applied Geophysics, v. 142, p. 609-644.
Bulut, F., Ben-Zion, Y., and Bonhoff, M., 2012, Evidence for a bimaterial interface along the
Mudurnu segment of the North Anatolian Fault Zone from polarization analysis of P
waves: Earth and Planetary Science Letters, v. 327-328, p. 17-22.
Calderoni, G., Di Giovambattista, R., Vannoli, P., Pucillo, S., and Rovelli, A., 2012, Fault-trapped
waves depict continuity of the fault system responsible for the 6 April 2009 MW 6.3
L’Aquila earthquake, central Italy: Earth and Planetary Science Letters, v. 323-324, p. 1-
8.
Carena, S., Suppe, J., and Kao, H., 2004, Lack of continuity of the San Andreas Fault in southern
California: Three-dimensional fault models and earthquake scenarios: Journal of
Geophysical Research, v. 109, p. B04313.
Christensen, N. I., 1996, Poisson’s ratio and crustal seismology: Journal of Geophysical Research,
v. 101(B2), p. 3139-3156.
Cochran, E. S., Li, Y.-G., Shearer, P. M., Barbot, S., Fialko, Y., and Vidale, J. E., 2009, Seismic
and geodetic evidence for extensive, long-lived fault damage zones: Geology, v. 37(4), p.
315-318.
126
Cooke, M. L., and Dair, L., 2011, Simulating the recent evolution of the southern big bend of the
San Andreas fault, Southern California: Journal of Geophysical Research, v. 116, p.
B04405.
Crotwell, H. P., Owens, T. J., and Ritsema, J., 1999, The TauP Toolkit: Flexible seismic travel-
time and ray-path utilities: Seismological Research Letters, v. 70(2), p. 154-160.
Dair, L., and Cooke, M. L., 2009, San Andreas fault geometry through the San Gorgonio Pass,
California: Geology, v. 37(2), p. 119-122.
DeDontney, N., Templeton-Barrett, E. L., Rice, J. R., and Dmowska, R., 2011, Influence of plastic
deformation on bimaterial fault rupture directivity: Journal of Geophysical Research, v.
116, p. B10312, doi:10.1029/2011JB008417.
Donnellan, A., Grant Ludwig, L., Parker, J. W., Rundle, J. B., Wang, J., Pierce, M., Blewitt, G.,
and Hensley, S., 2015, Potential for a large earthquake near Los Angeles inferred from the
2014 La Habra earthquake: Earth and Space Science, v. 2, doi:10.1002/2015EA000113.
Dor, O., Ben-Zion, Y., Rockwell, T. K., and Brune, J., 2006a, Pulverized Rocks in the Mojave
section of the San Andreas Fault Zone: Earth and Planetary Science Letters, v. 245, p. 642-
654.
Dor, O., Rockwell, T. K., and Ben-Zion, Y., 2006b, Geological Observations of Damage
Asymmetry in the Structure of the San Jacinto, San Andreas and Punchbowl Faults in
Southern California: A Possible Indicator for Preferred Rupture Propagation Direction:
Pure and Applied Geophysics, v. 163, p. 301-349.
Dor, O., Yildirim, C., Rockwell, T. K., Ben-Zion, Y., Emre, O., Sisk, M., and Duman, T. Y., 2008,
Geological and geomorphologic asymmetry across the rupture zones of the 1943 and 1944
earthquakes on the North Anatolian fault: possible signals for preferred earthquake
propagation direction: Geophysical Journal International, v. 173, p. 483–504.
Eccles, J. D., Gulley, A. K., Malin, P. E., Boese, C. M., Townend, J., and Sutherland, R., 2015,
Fault Zone Guided Wave generation on the locked, late interseismic Alpine Fault, New
Zealand: Geophysical Research Letters, v. 42(14), p. 5736-5743.
Ellsworth, W. L., and Malin, P. E., 2011, Deep rock damage in the San Andreas Fault revealed by
P-and S-type fault-zone-guided waves: Geological Society, London, Special Publications,
v. 359(1), p. 39-53.
127
Fang, H., Zhang, H., Yao, H., Allam, A., Zigone, D., Ben-Zion, Y., Thurber, C., and van der Hilst,
R., 2016, A new three-dimensional joint inversion algorithm of body-wave and surface-
wave data and its application to the Southern California Plate Boundary Region: Journal
of Geophysical Research, v. 121(5), p. 3557-3569.
Faulkner, D. R., Jackson, C. A. L., Lunn, R. J., Schlische, R. W., Shipton, Z. K., Wibberley , C.
A. J., and Withjack, M. O., 2010, A review of recent developments concerning the
structure, mechanics and fluid flow properties of fault zones: Journal of Structural
Geology, v. 32, p. 1557-1575.
Fialko, Y., 2006, Interseismic strain accumulation and the earthquake potential on the southern
San Andreas fault system: Nature, v. 441, p. 968–971.
Fichtner, A., and van Leeuwen, T., 2015, Resolution analysis by random probing: Journal of
Geophysical Research, v. 120, p. 5549-5573.
Fohrmann, M., Igel, H., Jahnke, G., and Ben-Zion, Y., 2004, Guided waves from sources outside
faults: an indication for shallow fault zone structure: Pure and Applied Geophysics, v. 161,
p. 2125-2137.
Fuis, G. S., Bauer, K., Goldman, M. R., Ryberg, T., Langenheim, V. E., Scheirer, D. S., Rymer,
M. J., Stock, J. M., Hole, J. A., Catchings, R. D., Graves, R. W., and Aagaard, B., 2017,
Subsurface geometry of the San Andreas Fault in Southern California: Results from the
Salton Seismic Imaging Project (SSIP) and strong ground motion expectations: Bulletin of
the Seismological Society of America, v. 107(3), p. 1642-1662.
Fuis, G. S., Scheirer, D. S., Langenheim, V. E., and Kohler, M. D., 2012, A new perspective on
the geometry of the San Andreas fault in southern California and its relationship to
lithospheric structure: Bulletin of the Seismological Society of America, v. 102(1), p. 236-
251.
Fukao, Y., Hori, S., and Ukawa, M., 1983, A seismological constraint on the depth of basalt-
eclogite transition in a subducting oceanic crust: Nature, v. 303, p. 413-415.
Gentili, S., and Michelini, A., 2006, Automatic picking of P and S phases using a neural tree:
Journal of Seismology, v. 10, p. 39-63.
Goebel, T. H. W., Hauksson, E., Shearer, P. M., and Ampuero, J. P., 2015, Stress-drop
heterogeneity within tectonically complex regions: a case study of San Gorgonio Pass,
southern California: Geophysical Journal International, v. 202(1), p. 514-528.
128
Gold, P. O., Behr, W. M., Rood, D., Sharp, W. D., Rockwell, T. K., Kendrick, K., and Salin, A.,
2015, Holocene geologic slip rate for the Banning strand of the southern San Andreas Fault,
southern California: Journal of Geophysical Research, v. 120(8), p. 5639-5663.
Guo, H., and Zhang, H., 2017, Development of double-pair double difference earthquake location
algorithm for improving earthquake locations: Geophysical Journal International, v. 208,
p. 333-348.
Gutierrez, C., Bryant, W., Saucedo, G., and Wills, C., 2010, Geologic map of California, California
Geological Survey.
Hadley, D., and Kanamori, H., 1977, Seismic structure of the Transverse Ranges, California:
Geological Society of America Bulletin, v. 88, p. 1469-1478.
Hamiel, Y., Liu, Y., Lyakhovsky, V., Ben-Zion, Y., and Lockner, D., 2004, A Visco-Elastic
Damage Model with Applications to Stable and Unstable fracturing: Geophysical Journal
International, v. 159, p. 1155-1165, doi: 10.1111/j.1365-246X.2004.02452.x.
Hardebeck, J. L., and Hauksson, E., 1999, Role of fluids in faulting inferred from stress field
signatures: Science, v. 285, p. 236-239.
Harden, J. W., and Matti, J. C., 1989, Holocene and late Pleistocene slip rates on the San Andreas
fault in Yucaipa, California, using displaced alluvial-fan deposits and soil chronology:
Geological Society of America Bulletin, v. 101, p. 1107-1117.
Hauksson, E., Yang, W., and Shearer, P. M., 2012, Waveform relocated earthquake catalog for
southern California (1981 to June 2011): Bulletin of the Seismological Society of America,
v. 102(5), p. 2239-2244.
Hearn, T. M., and Clayton, R. W., 1986, Lateral velocity variations in southern California. I.
Results for the upper crust from Pg waves: Bulletin of the Seismological Society of
America, v. 76(2), p. 495-509.
Hillers, G., Campillo, M., Ben-Zion, Y., and Roux, P., 2014, Seismic fault zone trapped noise:
Journal of Geophysical Research, v. 119(7), p. 5786-5799.
Hillers, G., Roux, P., Campillo, M., and Ben-Zion, Y., 2016, Focal spot imaging based on zero lag
cross correlation amplitude fields: Application to dense array data at the San Jacinto fault
zone: Journal of Geophysical Research, v. 121, doi:10.1002/2016JB013014.
129
Hough, S. E., Ben-Zion, Y., and Leary, P., 1994, Fault zone waves observed at the southern Joshua
Tree earthquake rupture zone: Bulletin of the Seismological Society of America, v. 84, p.
761-767.
Huang, Y., and Ampuero, J.-P., 2011, Pulse-like ruptures induced by low-velocity fault zones:
Journal of Geophysical Research, v. 116, p. B12307.
Igel, H., Ben-Zion, Y., and Leary, P., 1997, Simulation of SH and P SV wave propagation in fault
zones: Geophysical Journal International, v. 128, p. 533-546.
Igel, H., Jahnke, G., and Ben-Zion, Y., 2002, Numerical simulation of fault zone guided waves:
accuracy and 3-D effects: Pure and Applied Geophysics, v. 159, p. 2967-2083.
Jahnke, G., Igel, H., and Ben-Zion, Y., 2002, Three-dimensional calculations of fault zone guided
waves in various irregular structures: Geophysical Journal International, v. 151, p. 416-
426.
Jepsen, D. C., and Kennett, B. L. N., 1990, Three component analysis of regional seismograms:
Bulletin of the Seismological Society of America, v. 80, p. 2032-2052.
Johnson, H. O., Agnew, D. C., and Wyatt, F. K., 1994, Present-day crustal deformation in southern
California: Journal of Geophysical Research, v. 99(B12), p. 23951-23974.
Jones, L. M., Hutton, L. K., Given, D. D., and Allen, C. R., 1986, The North Palm Springs,
California, earthquake sequence of July 1986: Bulletin of the Seismological Society of
America, v. 76, p. 1830-1837.
Jurkevics, A., 1988, Polarization Analysis of Three-component Array Data: Bulletin of the
Seismological Society of America, v. 78(5), p. 1725-1743.
Kendrick, K. J., Matti, J. C., and Mahan, S. A., 2015, Late Quaternary slip history of the Mill
Creek strand of the San Andreas fault in San Gorgonio Pass, southern California: The role
of a subsidiary left-lateral fault in strand switching: Geological Society of America
Bulletin, v. 127(5-6), p. 825-849.
Kennett, L. N., 1991, IASPEI 1991 Seismological Tables: Bibliotech, v. 167.
Kurzon, I., Vernon, F. L., Ben-Zion, Y., and Atkinson, G., 2014, Ground Motion Prediction
Equations in the San Jacinto Fault Zone – Significant Effects of Rupture Directivity and
Fault Zone Amplification: Pure and Applied Geophysics, v. 171, p. 3045–3081.
Langenheim, V. E., Jachens, R. C., Matti, J. C., Hauksson, E., Morton, D. M., and Christensen,
A., 2005, Geophysical evidence for wedging in the San Gorgonio Pass structural knot,
130
southern San Andreas fault zone, southern California: Geological Society of America
Bulletin, v. 117, p. 1554-1572.
Langenheim, V. E., Jachens, R. C., Morton, D. M., Kistler, R. W., and Matti, J. C., 2004,
Geophysical and isotopic mapping of preexisting crustal structures that influenced the
location and development of the San Jacinto fault zone, southern California: Geological
Society of America Bulletin, v. 116, p. 1143-1157.
Lee, E.-J., Chen, P., Jordan, T. H., Maechling, P. B., Denolle, M. A. M., Beroza, G., 2014, Full-3-
D tomography for crustal structure in Southern California based on the scattering-integral
and the adjoint-wavefield methods: Journal of Geophysical Research, v. 119(8), p. 6421-
6451.
Lévêque, J. J., Rivera, L., and Wittlinger, G., 1993, On the use of the checkerboard test to assess
the resolution of tomographic inversions: Geophysical Journal International, v. 115, p. 313-
318.
Lewis, M. A., and Ben-Zion, Y., 2010, Diversity of fault zone damage and trapping structures in
the Parkfield section of the San Andreas Fault from comprehensive analysis of near fault
seismograms: Geophysical Journal International, v. 183(3), p. 1579-1595.
Lewis, M. A., Ben-Zion, Y., and McGuire, J., 2007, Imaging the deep structure of the San Andreas
Fault south of Hollister with joint analysis of fault-zone head and Direct P arrivals:
Geophysical Journal International, v. 169, p. 1028-1042.
Lewis, M. A., Peng, Z., Ben-Zion, Y., and Vernon, F. L., 2005, Shallow seismic trapping structure
in the San Jacinto fault zone near Anza, California: Geophysical Journal International, v.
162, p. 867-881.
Li, Y.-G., Aki, K., Adams, D., Hasemi, A., and Lee, W. H. K., 1994, Seismic guided waves trapped
in the fault zone of the Landers, California, earthquake of 1992: Journal of Geophysical
Research, v. 99, p. 11705–11722.
Li, Y.-G., and Leary, P. C., 1990, Fault zone trapped seismic waves: Bulletin of the Seismological
Society of America, v. 80(5), p. 1245-1271.
Li, Y.-G., and Vernon, F. L., 2001, Characterization of the San Jacinto fault zone near Anza,
California, by fault zone trapped waves: Journal of Geophysical Research, v. 106, p.
30671-30688.
131
Li, Y.-G., Chen, P., Cochran, E. S., and Vidale, J. E., 2007, Seismic velocity variations on the San
Andreas fault caused by the 2004 M6 Parkfield Earthquake and their implications: Earth
Planets and Space, v. 59(1), p. 11705-11722.
Li, Y.-G., Vidale, J. E., and Cochran, E. S., 2004, Low-velocity damaged structure of the San
Andreas Fault at Parkfield from fault zone trapped waves: Geophysical Research Letters,
v. 31, p. L12S06.
Li, Z., Peng, Z., Ben-Zion, Y., and Vernon, F. L., 2015, Spatial variations of Shear-wave
Anisotropy Near the San Jacinto Fault Zone in Southern California: Journal of Geophysical
Research, v. 120, p. 8334-8347, doi: 10.1002/2015JB012483.
Lin, F.-C., Allam, A. A., Share, P.-E., Ben-Zion, Y., Vernon, F., Schuster, G. T., and Karplus, M.
S., 2016, Constraints on Fault Damage Zone Properties and Normal Modes from a Dense
Linear Array Deployment along the San Jacinto Fault Zone: Abstract S43B-2849 presented
at 2016 Fall Meeting, AGU, San Francisco, Calif., 12-16 Dec.
Lin, G., and Shearer, P. M., 2009, Evidence for water-filled cracks in earthquake source regions:
Geophysical Research Letters, v. 36, p. L17315.
Lindsey, E. O., and Fialko, Y., 2013, Geodetic slip rates in the southern San Andreas Fault system:
Effects of elastic heterogeneity and fault geometry: Journal of Geophysical Research, v.
118, p. 689-697.
Lindsey, E. O., Sahakian, V. J., Fialko, Y., Bock, Y., Barbot, S., and Rockwell, T. K., 2014,
Interseismic Strain Location in the San Jacinto Fault Zone: Pure and Applied Geophysics,
v. 171, p. 2937-2954.
Lockner, D. A., Byerlee, J. D., Kuksenko, V., Ponomarev, A., and Sidrin, A., 1992, Observations
of Quasistatic Fault Growth from Acoustic Emissions in Fault Mechanics and Transport
Properties of Rocks, p. 3-31.
Lozos, J. C., 2016, A case for historic joint rupture of the San Andreas and San Jacinto faults:
Science Advances, v. 2(3), p. e1500621.
MacCarthy, J. K., Brochers, B., and Aster, R. C., 2011, Efficient stochastic estimation of the model
resolution matrix diagonal and generalised cross-validation for large geophysical inverse
problems: Journal of Geophysical Research, v. 116, p. B10304.
132
Magistrale, H., and Sanders, C., 1996, Evidence from precise earthquake hypocenters for
segmentation of the San Andreas fault in San Gorgonio Pass: Journal of Geophysical
Research, v. 101, p. 3031–3041.
Marliyani, G. I., Rockwell, T. K., Onderdonk, N. W., and McGill, S. F., 2013, Straightening of the
northern San Jacinto Fault, California as seen in the fault-structure evolution of the San
Jacinto Valley step-over: Bulletin of the Seismological Society of America, v. 103(1), p.
519-541.
Martel, S. J., Pollard, D. D., and Segall, P., 1988, Development of simple fault zones in granitic
rock, Mount Abbot Quadrangle, Sierra Nevada, California: Geological Society of America
Bulletin, v. 100, p. 1451-1465.
Matti, J. C., Morton, D. M., and Cox, B. F., 1992, The San Andreas fault system in the vicinity of
the central Transverse Ranges province, southern California: U.S. Geological Survey
Open-File Report 92-354.
Matti, J., and D. Morton, 1993, Paleogeographic evolution of the San Andreas fault in southern
California: A reconstruction based on a new cross-fault correlation (R. Powell, R. J.
Weldon, and J. Matti, Eds.): The San Andreas fault system: Displacement, palinspastic
reconstruction, and geologic evolution: Geological Society of America Memoirs, v. 178,
p. 107-159.
Mavko, G., Mukerji, T., and Dvorkin, J., 1998, The Rock Physics Handbook: Tools for Seismic
Analysis in Porous Media, Cambridge University Press, Cambridge, U.K., 329 pp.
McGill, S. F., Spinler, J. C., McGill, J. D., Bennett, R. A., Floyd, M. A., Fryxell, J. E., and Funning,
G. J., 2015, Kinematic modeling of fault slip rates using new geodetic velocities from a
transect across the Pacific-North America plate boundary through the San Bernardino
Mountains, California: Journal of Geophysical Research, v. 120(4), p. 2772-2793.
McGill, S., Owen, L., Weldon, R. J., and Kendrick, K., 2013, Latest Pleistocene and Holocene slip
rate for the San Bernardino strand of the San Andreas Fault, Plunge Creek, Southern
California: Implications for strain partitioning within the southern San Andreas Fault
system for the last ~35 k.y.: Geological Society of America Bulletin, v. 125(1/2), p. 48-72.
McGuire, J., and Ben-Zion, Y., 2005, High-resolution imaging of the Bear Valley section of the
San Andreas Fault at seismogenic depths with fault-zone head waves and relocated
seismicity: Geophysical Journal International, v. 163, p. 152-164.
133
McNally, K.C., and McEvilly, T.V., 1977, Velocity contrast across the San Andreas fault in central
California: small-scale variations from P-wave nodal plane distortion: Bulletin of the
Seismological Society of America, v. 67, p. 1565-1576.
Meade, B. J., and Hager, B. H., 2005, Block models of crustal motion in southern California
constrained by GPS measurements: Journal of Geophysical Research, v. 110, p. B03403.
Meng, H., Ben-Zion, Y., and McGuire, 2017, Towards automated estimates of directivity and
related source properties of small to moderate earthquakes with second seismic moments:
Abstract presented at 2017 SSA Meeting, Denver, Co., 18-20 Apr. 2017.
Michael, A. J., and Ben-Zion, Y., 1998, Inverting fault zone trapped waves with genetic algorithm:
EOS Transactions of the American Geophysical Union, v. 79, p. F584.
Mitchell, T. M., and Faulkner, D. R., 2009, The nature and origin of off-fault damage surrounding
strike-slip fault zones with a wide range of displacements: A field study from the Atacama
fault system, northern Chile: Journal of Structural Geology, v. 31, p. 802-816.
Mizuno, T., and Nishigami, K., 2006, Deep structure of the Nojima fault, southwest Japan,
estimated from borehole observation of fault-zone trapped waves: Tectonophysics, v. 417,
p. 231-247.
Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A.,
Riedmiller, M., Fidjeland, M. K., Ostrovski, G., Petersen, S., Beattie, C., Sadik, A.,
Antonoglou, I., King, H., Kumaran, D., Wierstra, D., Legg, S., and Hassabis, D., 2015,
Human-level control through deep reinforcement learning: Nature, v. 518, p. 529-533.
Najdahmadi, B., Bohnhoff, M., and Ben-Zion, Y., 2016, Bimaterial interfaces at the Karadere
segment of the North Anatolian Fault, northwestern Turkey: Journal of Geophysical
Research, v. 121, p. 931-950.
Neuberg, J., and Pointer, T., 2000, Effects of volcano topography on seismic broad-band
waveforms: Geophysical Journal International, v. 143, p. 239-248.
Nicholson, C., 1996, Seismic behavior of the southern San Andreas Fault Zone in the Northern
Coachella Valley, California: Comparison of the 1948 and 1986 earthquake sequences:
Bulletin of the Seismological Society of America, v. 86, p. 1331-1349.
Nippress, S. E. J., Rietbrock, A., and Heath, A. E., 2010, Optimized automatic pickers: application
to the ANCORP data set: Geophysical Journal International, v. 181, p. 911-925.
134
O’Connell, R. J., and Budiansky, B., 1974, Seismic velocities in dry and saturated cracked solids:
Journal of Geophysical Research, v. 79(35), p. 5412-5426.
Onderdonk, N. W., McGill, S. F., and Rockwell, T. K., 2015, Short-term variations in slip rate and
size of prehistoric earthquakes during the past 2000 years on the northern San Jacinto fault
zone, a major plate-boundary structure in southern California: Lithosphere, v. 7(3), p. 211-
234.
Oppenheimer, D. H., Reasenberg, P. A., and Simpson, R. W., 1988, Fault plane solutions for the,
1984 Morgan Hill, California, earthquake sequence: evidence for the state of stress on the
Calaveras fault: Journal of Geophysical Research, v. 143, p. 239-248.
Oskin, M., Perg, L., Shelef, E., Strane, M., Gurney, E., Singer, B., and Zhang, X., 2008, Elevated
shear zone loading rate during an earthquake cluster in eastern California: Geology, v.
36(6), p. 507-510.
Paige, C. C., and Saunders, M. A., 1982, Algorithm 583; LSQR: Sparse linear equations and least
squares problems: ACM Transactions on Mathematical Software, v. 8(2), p. 195-209.
Peng, Z., Ben-Zion, Y., Michael, A. J., and Zhu, L. P., 2003, Quantitative analysis of seismic fault
zone waves in the rupture zone of the Landers, 1992, California earthquake: evidence for
a shallow trapping structure: Geophysical Journal International, v. 155, p. 1021-1041.
Perrin, C., Manighetti, I., Ampuero, J.-P., Cappa, F., and Gaudemer, Y., 2016, Location of largest
earthquake slip and fast rupture controlled by along-strike change in fault structural
maturity due to fault growth: Journal of Geophysical Research, v. 121, p. 3666-3685.
Platt, J. P., and Becker, T. W., 2010, Where is the real transform boundary in California:
Geochemistry Geophysics Geosystems, v. 11, p. Q06012.
Popp, T., and Kern, H., 1994, The influence of dry and water saturated cracks on seismic velocities
of crustal rocks – A comparison of experimental data with theoretical model: Surveys in
Geophysics, v. 15, p. 443-465.
Porter, K., Jones, L., Cox, D., Goltz, J., Hudnut, K., Mileti, D., Perry, S., Ponti, D., Reichle, M.,
Rose, A. Z., Scawthorn, C. R., Seligson, H. A., Shoaf, K. I., Treiman, J., and Wein, A.,
2011, The ShakeOut Scenario: A hypothetical Mw7.8 earthquake on the Southern San
Andreas fault: Earthquake Spectra, v. 27(2), p. 239-261.
Qin, L., Ben-Zion, Y., Qiu, H., Share, P.-E., Ross, Z. E., and Vernon, F. L., 2018, Internal structure
of the San Jacinto fault zone in the trifurcation area southeast of Anza, California, from
135
data of dense seismic arrays: Geophysical Journal International, v. 213, p. 98-114, doi:
10.1093/gji/ggx540.
Qiu, H., Ben-Zion, Y., Ross, Z. E., Share, P.-E., and Vernon, F., 2017, Internal structure of the
San Jacinto fault zone at Jackass Flat from data recorded by a dense linear array:
Geophysical Journal International, v. 209(3), p. 1369-1388.
Ranjith, K., and J. Rice, 2001, Slip dynamics at an interface between dissimilar materials: Journal
of the Mechanics and Physics of Solids, v. 49, p. 341-361.
Rawlinson, N., Fichtner, A., Sambridge, M., and Young, M. K., 2014, Seismic tomography and
the assessment of uncertainty: Advances in Geophysics, v. 55, p. 1-76.
Rempe, M., Mitchell, T. M., Renner, J., Nippress, S., Ben-Zion, Y., and Rockwell, T. K., 2013,
Damage and seismic velocity structure of pulverized rocks near the San Andreas Fault:
Journal of Geophysical Research, v. 118, p. 2813-2831.
Rockwell, T. K., Dawson, T. E., Ben-Horin, J. Y., and Seitz, G., 2015, A 21-event, 4,000-year
history of surface ruptures in the Anza seismic gap, San Jacinto Fault, and implications for
long-term earthquake production on a major plate boundary fault: Pure and Applied
Geophysics, v. 172(5), p. 1143-1165.
Ron, H., Freund, R., Garfunkel, Z., and Nur, A., 1984, Block rotations by strike-slip faulting:
Structural and paleomagnetic evidence: Journal of Geophysical Research, v. 89(B7), p.
6256-6270.
Ross, Z. E., and Ben-Zion, Y., 2014, Automatic picking of direct P, S seismic phases and fault
zone head waves: Geophysical Journal International, v. 199, p. 368-381.
Ross, Z. E., and Ben-Zion, Y., 2015, An algorithm for automated identification of fault zone
trapped waves: Geophysical Journal International, v. 202, p. 933-942.
Ross, Z. E., and Ben-Zion, Y., 2016, Toward reliable automated estimates of earthquake source
properties from body wave spectra: Journal of Geophysical Research, v. 121,
doi:10.1002/2016JB013003.
Ross, Z. E., Hauksson, E., and Ben-Zion, Y., 2017a, Abundant off-fault seismicity and orthogonal
structures in the San Jacinto fault zone: Science Advances, v. 3, p. e1601946.
Ross, Z. E., Kanamori, H., and Hauksson, E., 2017b, Anomalously large complete stress drop
during the 2016 MW 5.2 Borrego Springs earthquake inferred by waveform modeling and
near-source aftershock deficit: Geophysical Research Letters, v. 44, p. 5994-6001.
136
Ross, Z. E., White, M. C., Vernon, F. L., and Ben-Zion, Y., 2016, An improved algorithm for real-
time S-wave picking with application to the (augmented) ANZA network in Southern
California: Bulletin of the Seismological Society of America, v. 106(5), p. 2013-2022.
Roux, P., Moreau, L., Lecointre, A., Hillers, G., Campillo, M., Ben-Zion, Y., Zigone, D., and
Vernon, F., 2016, A methodological approach towards high-resolution surface wave
imaging of the San Jacinto Fault Zone using ambient-noise recordings at a spatially dense
array: Geophysical Journal International, v. 206(2), p. 980-992.
Rovelli, A., Caserta, A., Marra, F., and Ruggiero, V., 2002, Can seismic waves be trapped inside
an inactive fault zone? The case study of Nocera Umbra, Central Italy: Bulletin of the
Seismological Society of America, v. 92, p. 2217-2232.
Rubin, A.M., and J.-P. Ampuero, 2007, Aftershock asymmetry on a bimaterial interface: Journal
of Geophysical Research, v. 112, p. B05307.
Salisbury, J. B., Rockwell, T. K., Middleton, T. J., and Hudnut, K. W., 2012, LiDAR and field
observations of slip distribution for the most recent surface ruptures along the central San
Jacinto Fault: Bulletin of the Seismological Society of America, v. 102(2), p. 598-619.
Sanders, C. O., and Kanamori, H., 1984, A seismotectonic analysis of the Anza seismic gap, San
Jacinto fault zone, southern California: Journal of Geophysical Research, v. 89(B7), p.
5873-5890.
Sanders, C., and Magistrale, H., 1997, Segmentation of the northern San Jacinto fault zone,
southern California: Journal of Geophysical Research, v. 102(B12), p. 27453-27467.
SCEDC, 2013, Southern California Earthquake Data Center: Caltech.Dataset,
doi:10.7909/C3WD3xH.
Schmandt, B., and Clayton, R. W., 2013, Analysis of teleseismic P waves with a 5200-station array
in Long Beach, California: Evidence for an abrupt boundary to Inner Borderland rifting:
Journal of Geophysical Research, v. 118, p. 1-19.
Share, P.-E., and Ben-Zion, Y., 2016, Bimaterial interfaces in the South San Andreas Fault with
opposite velocity contrasts NW and SE from San Gorgonio Pass: Geophysical Research
Letters, 43(20), p. 10680-10687.
Share, P.-E., Ben-Zion, Y., Ross, Z. E., Qiu, H., and Vernon, F., 2017, Internal structure of the
San Jacinto fault zone at Blackburn Saddle from seismic data of a dense linear array:
Geophysical Journal International, v. 210(2), p. 819-832.
137
Share, P.-E., Guo, H., Thurber, C. H., Zhang, H., and Ben-Zion, Y., 2018a, Seismic imaging of
the Southern California plate-boundary around the South-Central Transverse Ranges using
double-difference tomography: Pure and Applied Geophysics, in review.
Share, P.-E., Allam, A. A., Lin, F.-C., and Vernon, F., 2018b, Structural properties of the San
Jacinto fault zone at Blackburn Saddle from seismic data of a dense linear array: Pure and
Applied Geophysics, in review.
Sharp, R. V., 1967, San Jacinto Fault Zone in the Peninsular Ranges of Southern California:
Geological Society of America Bulletin, v. 78(6), p. 705-730.
Shaw, J. H., Plesch, A., Tape, C., Suess, M. P., Jordan, T. H., Ely, G., Hauksson, E., Tromp, J.,
Tanimoto, T., Graves, R., Olsen, K., Nicholson, C., Maechling, P. J., Rivero, C., Lovely,
P., Brankman, C. M., and Munster, J., 2015, Unified structural representation of the
southern California crust and upper mantle: Earth and Planetary Science Letters, v. 415, p.
1-15.
Shi, Z., and Ben-Zion, Y., 2006, Dynamic Rupture on a Bimaterial Interface Governed by Slip-
Weakening Friction: Geophysical Journal International, v. 165, p. 469-484.
Shinevar, W., Behn, M., Hirth, G., and Jagoutz, O., 2017, Inferring Crustal Viscosity from Seismic
Wavespeeds: Applications to the Rheologic Structure of the Himalayas: Abstract of the
Fall meeting of the American Geophysical Meeting.
Shlomai, H., and Fineberg, J., 2016, The structure of slip-pulses and supershear ruptures driving
slip in bimaterial friction: Nature Communications, v. 7:11787.
Sibson, R. H., 2003, Thickness of the seismic slip zone: Bulletin of the Seismological Society of
America, v. 93, p. 1169-1178.
Sibson, R. H.,1989, Earthquake faulting as a structural process: Journal of Structural Geology, v.
11, p. 1-14.
Silver, L. T., and Chappell, B. W., 1987, The Peninsular Ranges batholith: An insight into the
evolution of the Cordilleran batholiths of southwestern North America: Transactions of
Royal Society of Edinburgh: Earth Sciences, v. 79, p. 105-121.
Smith-Konter, B. R., Sandwell, D. T., and Shearer, P., 2011, Locking depths estimated from
geodesy and seismology along the San Andreas Fault System: Implications for seismic
moment release: Journal of Geophysical Research, v. 116, p. B06401.
138
Spinler, J. C., Bennett, R. A., Anderson, M. L., McGill, S. F., Hreinsdottir, S., and McCallister,
A., 2010, Present-day strain accumulation and slip rates associated with southern San
Andreas and Eastern California shear zone faults: Journal of Geophysical Research, v. 115,
p. B11407.
Spotila, J. A., Farley, K. A., Yule, J. D., and Reiners, P. W., 2001, Near-field transpressive
deformation along the San Andreas fault zone in southern California, based on exhumation
constrained by (U-Th)/He dating: Journal of Geophysical Research, v. 106, p. 30909-
30922.
Tape, C., Liu, Q., Maggi, A., and Tromp, J., 2010, Seismic tomography of the southern California
crust based on spectral-element and adjoint methods: Geophysical Journal International, v.
180, p. 433-462.
Tapponnier, P., Armijo, R., Manighetti, I., and Courtillot, V., 1990, Bookshelf faulting and
horizontal block rotations between overlapping rifts in southern Afar: Geophysical
Research Letters, v. 17(1), p. 1-4.
Thurber, C., and Eberhart-Phillips, D., 1999, Local earthquake tomography with flexible gridding:
Computers and Geosciences, v. 25(7), p. 809-818.
Thurber, C., Zhang, H., Waldhauser, F., Hardebeck, J., Michael, A., and Eberhart-Phillips, D.,
2006, Three-dimensional compressional wavespeed model, earthquake relocations, and
focal mechanisms for the Parkfield, California, region: Bulletin of the Seismological
Society of America, v. 96, p. S38-S49.
Trampert, J., Fichtner, A., and Ritsema, J., 2013, Resolution tests revisited: The power of random
numbers: Geophysical Journal International, v. 192, p. 676-680.
Vernon, F., and Ben-Zion, Y., 2010, San Jacinto Fault Zone Experiment. International Federation
of Digital Seismograph Networks: Other/Seismic Network, doi:10.7914/SN/YN_2010.
Vernon, F., UC San Diego, 1982, ANZA Regional Network. International Federation of Digital
Seismograph Networks: Other/Seismic Network, doi:10.7914/SN/AZ.
Waldhauser, F., and Ellsworth, W. L., 2000, A double-difference earthquake location algorithm:
Method and application to the northern Hayward fault: Bulletin of the Seismological
Society of America, v. 90, p. 1353-1368.
139
Wechsler, N., Rockwell, T. K., and Ben-Zion, Y., 2009, Application of high resolution DEM data
to detect rock damage from geomorphic signals along the central San Jacinto Fault:
Geomorphology, v. 113, p. 82-96.
Weertman, J., 1980, Unstable slippage across a fault that separates elastic media of different elastic
constants: Journal of Geophysical Research, v. 85, p. 1455-1461.
Weertman, J., 2002, Subsonic type earthquake dislocation moving at approximately √2× shear
wave velocity on interface between half spaces of slightly different elastic constants:
Geophysical Research Letters, v. 29(10), p. 109-1-109-4.
Wesnousky, S.G., 2006, Predicting the endpoints of earthquake ruptures: Nature, v. 444, p. 358-
360.
White, M. C. A., Ross, Z. E., Vernon, F. L., and Ben-Zion, Y., 2016, A detailed automatic 1998-
2015 earthquake catalog of the San Jacinto fault zone region: Abstract of the annual
meeting of the Southern California Earthquake Center.
Xu, S., Ben-Zion, Y., and Ampuero, J. P., 2012, Properties of inelastic yielding zones generated
by in-plane dynamic ruptures-II. Detailed parameter-space study: Geophysical Journal
International, v. 191(3), p. 1343-1360.
Yang, H., and Zhu, L., 2010, Shallow low-velocity zone of the San Jacinto Fault from local
earthquake waveform modelling: Geophysical Journal International, v. 183, p. 421–432.
Yang, H., Li, Z., Peng, Z., Ben-Zion, Y., and Vernon, F., 2014, Low velocity zones along the San
Jacinto Fault, Southern California, from body waves recorded in dense linear arrays:
Journal of Geophysical Research, v. 119, p. 8976–8990.
Yang, W., Hauksson, E., and Shearer, P., 2012, Computing a large refined catalog of focal
mechanisms for Southern California (1981–2010): Temporal stability of the style of
faulting: Bulletin of the Seismological Society of America, v. 102(3), p. 1179-1194.
Yang, W., Peng, Z., Wang, B., Li, Z., and Yuan, S., 2015, Velocity contrast along the rupture zone
of the 2010 Mw6.9 Yushu, China earthquake from systematic analysis of fault zone head
waves: Earth and Planetary Science Letters, v. 416, p. 91–97.
Yule, D., 2009, The enigmatic San Gorgonia Pass: Geology, v. 37(2), p. 191-192.
Yule, D., and Sieh, K., 2003, Complexities of the San Andreas fault near San Gorgonio Pass:
Implications for large earthquakes: Journal of Geophysical Research, v. 108, p. B11.
140
Zaliapin, I., and Ben-Zion, Y., 2011, Asymmetric distribution of aftershocks on large faults in
California: Geophysical Journal International, v. 185, p. 1288-1304.
Zhang, H., and Thurber, C. H., 2003, Double-difference tomography: The method and its
application to the Hayward Fault, California: Bulletin of the Seismological Society of
America, v. 93(5), p. 1875-1889.
Zhang, H., and Thurber, C. H., 2007, Estimating the model resolution matrix for large seismic
tomography problems based on Lanczos bidiagonalization with partial
reorthogonalization: Geophysical Journal International, v. 170, p. 337-345.
Zhang, H., and Thurber, C., 2006, Development and applications of double-difference seismic
tomography: Pure and Applied Geophysics, v. 163(2-3), p. 373-403.
Zhang, H., Nadeau, R. M., and Guo, H., 2017, Imaging the nonvolcanic tremor zone beneath the
San Andreas fault at Cholame, California using station-pair double-difference tomography:
Earth and Planetary Science Letters, v. 460, p. 76-85.
Zigone, D., Ben-Zion, Y., Campillo, M., and Roux, P., 2015, Seismic tomography of the Southern
California plate boundary region from noise-based Rayleigh and Love Waves: Pure and
Applied Geophysics, v. 172, p. 1002-1032.

Abstract (if available)

Abstract Knowledge of material and geometric properties along major faults in Southern California is critical for understanding the large earthquake potential and seismic hazard in the region. Using different signals and techniques, I image these properties at multiple length-scales (100 km to 100 m) with a focus on the two most seismically active faults in the area, the San Andreas (SAF) and San Jacinto (SJF) faults. ❧ Analysis of fault zone head waves reveals long (~60 km), continuous segments of the central SJF, and, the SAF northwest and southeast of San Gorgonio Pass (SGP) with across-fault VP contrasts of 3-10%. High-resolution travel-time tomography shows distinct along-fault changes in large-scale (>4 km) structure through SGP, potentially impeding large earthquakes propagating through the region. It also reveals a northeast-dipping segment of the southern SAF around Coachella Valley. Broad damage and deformation associated with the SJF and SAF manifest as high VP/VS anomalies at shallow depth (<5km) and in some areas at greater depth (>10 km) as low VP/VS anomalies. These variations give insight into the complex interplay between fault-associated damage, crack geometry, fluid content and lithology near these faults. At smaller scale (~100 m-1 km), the internal structure of the SJF is imaged using characteristics of earthquake waveforms, including arrival time variations and fault zone trapped waves recorded by dense linear arrays crossing the fault. These analyses reveal 100-200 m wide and <5 km deep core damage zones located predominantly northeast of the seismogenic SJF. ❧ The determined fault geometries, bimaterial structures, damage asymmetry and deformation can influence for example the rupture propagation direction and length (magnitude) of earthquakes along the SAF and SJF. These results should be incorporated in future seismic hazard analysis to improve predictions of ground motion from large earthquakes.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Multi-scale imaging and monitoring of crustal and fault zone structures in southern California

PDF

Heterogeneity of earthquake stress drops, focal mechanisms and active fault zones

PDF

High-resolution imaging and monitoring of fault zones and shallow structures: case studies in southern California and on Mars

PDF

Multi-scale imaging of the fault zone velocity structure: double-difference tomography, inversion of fault zone headwaves, and fault zone sensitivity kernels

PDF

Spatiotemporal variations of stress field in the San Jacinto Fault Zone and South Central Transverse Ranges

PDF

Multi-scale damage signatures across major strike-slip faults

PDF

Determining fault zone structure and examining earthquake early warning signals using large datasets of seismograms

PDF

Volumetric interactions between major ruptures and fault zones illuminated by small earthquake properties

PDF

Symmetry properties, pulverized rocks and damage architecture as signatures of earthquake ruptures

PDF

Seismic velocity structure and spatiotemporal patterns of microseismicity in active fault zones of Southern California

PDF

Detection and modeling of slow slip events as creep instabilities beneath major fault zones

PDF

Analysis of waveform and catalog data of aftershocks for properties of earthquakes and faults

PDF

Dynamic rupture processes and seismic radiation in models of earthquake faults separating similar and dissimilar solids

PDF

Elements of seismic structures near major faults from the surface to the Moho

PDF

Direct observation of fault zone structure and mechanics in three dimensions: a study of the SEMP fault system, Austria

PDF

Stress-strain characterization of complex seismic sources

PDF

Seismicity distribution near strike-slip faults in California

PDF

Paleoseismologic and slip rate studies of three major faults in southern California: understanding the complex behavior of plate boundary fault systems over millenial timescales

PDF

Microseismicity, fault structure, & the seismic cycle: insights from laboratory stick-slip experiments

PDF

Integration and validation of deterministic earthquake simulations in probabilistic seismic hazard analysis

Asset Metadata

Creator Share, Pieter-Ewald (author)

Core Title Multi-scale imaging of major fault zones in Southern California

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Geological Sciences

Publication Date 06/29/2018

Defense Date 01/19/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag earthquake physics,fault zone structure,OAI-PMH Harvest,San Andreas fault,San Jacinto fault zone,seismic hazard,seismic imaging

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Ben-Zion, Yehuda (committee chair), Haldar, Justin (committee member), Okaya, David (committee member), Sammis, Charles (committee member)

Creator Email pieter.share@gmail.com,pshare@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-14116

Unique identifier UC11671820

Identifier etd-SharePiete-6363.pdf (filename),usctheses-c89-14116 (legacy record id)

Legacy Identifier etd-SharePiete-6363.pdf

Dmrecord 14116

Document Type Dissertation

Format application/pdf (imt)

Rights Share, Pieter-Ewald

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA