by
Frederick D. Pearce
M.S., Geosystems, Massachusetts Institute of Technology, 2003
B.S., Geological Engineering, University of Idaho, 2001
Submitted to the Department of Earth, Atmospheric and Planetary Sciences in partial fulfillment of the requirements for the degree of
Doctor of Philosophy at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
February 2015
© Massachusetts Institute of Technology 2015. All rights reserved.
Signature of Author…………………………………………………………………………
Department of Earth, Atmospheric and Planetary Sciences
December 12, 2014
Certified by…………………………………………………………………………………
Stéphane Rondenay
Professor of Seismology
Thesis Co-Supervisor
Certified by…………………………………………………………………………………
B. Clark Burchfiel
Schlumberger Professor of Geology
Thesis Co-Supervisor
Accepted by…………...……………………………………………………………………
Robert van der Hilst
Schlumberger Professor of Earth and Planetary Sciences
Head, Department of Earth, Atmospheric and Planetary Sciences

2

by
Frederick D. Pearce
Submitted to the Department of Earth, Atmospheric and Planetary Sciences on December 12, 2014, in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
In this dissertation, I investigate the structure and dynamics of the Western Hellenic
Subduction Zone (WHSZ) by using two complementary seismic imaging methods and interpreting the resulting images with models that describe the dynamics of retreating subduction. First, I produce high-resolution seismic images across northern and southern
Greece using a two-dimensional teleseismic migration method. These images show subducted oceanic crust beneath southern Greece and subducted continental crust beneath northern Greece, with the relative position of the two crusts indicating ~70 km of additional slab retreat in the south relative to the north, a result consistent with the predicted relationship between slab buoyancy and retreat rates in recent geodynamic models. Second, I develop a three-dimensional receiver function imaging method, test it with synthetic data, and use it to constrain along strike variations in lithospheric structure. I find a continuous slab Moho across northern and southern Greece between
~40 and 80 km depth, with a gentle, trench-parallel component of dip accommodating the observed differential slab retreat. The overriding Moho is deepest beneath the northern
Hellenides (35-40 km) and shallowest beneath the Aegean Sea (25-30 km). It also exhibits several characteristics consistent with a retreating subduction model: (1) it is asymmetric when viewed perpendicular to the trench, not symmetric as has been found in previous studies, (2) the location of its leading edge closely tracks the 70 km depth contour of the slab Moho, (3) a well-developed Moho is not observed below the peak topography of the Hellenides, and (4) it exhibits Moho depth fluctuations that are much larger than those predicted assuming surface topography is locally compensated by Airy-
Heiskanen isostasy (>+/-4 km). Finally, I combine the seismic-based constraints with those from geologic data and geodynamic models to better understand how the overriding lithosphere is built and deformed during slab retreat. In northern Greece, the overriding crust is found to be predominately built by accretion of slab-derived continental blocks, while in southern Greece the present-day subduction of an oceanic slab domain has caused previously accreted continental blocks to rapidly extend, yielding an asymmetric, valley-shaped pattern in the top of the crystalline basement.
Thesis Co-Supervisor: Stéphane Rondenay Thesis Co-Supervisor: B. Clark Burchfiel
Title: Professor of Seismology Title: Schlumberger Professor of Geology
3

4

*
Reflecting on the twelve years since I first came to MIT for a Master’s degree, I find it all but impossible to do this section justice, and so, I set out with the goal of keeping this short, but alas, brevity is not my specialty. I would like to thank the members of my thesis committee: Stéphane Rondenay, Clark Burchfiel, Leigh (Wiki) Royden, Geoff
Abers, and Rob van der Hilst. Thank you, Stéphane, for supporting me through what was a lengthy thesis writing process, and for all of your advice, which, amongst many other things, was very helpful in improving my writing skills. I am also very grateful for your patience, and for the opportunity to work on such an interesting, multidisciplinary project. Thank you, Clark, for being a co-adviser on my thesis, for your support and encouragement through thick and thin, and for your inspiring thoughts on the tectonics of
Greece. Thank you, Wiki, for leading project MEDUSA, for chairing my thesis committee, and for meeting with me to discuss your papers on the tectonics and geodynamics of Greece. Thank you, Geoff, for thoroughly reviewing both my JGR paper and my thesis as a whole. This work greatly benefited from your insightful comments.
I started working on project MEDUSA towards the end of 2008, long after the project started, and so, I am greatly indebted to all those that made the project possible.
In particular, I would like to thank the people who helped with the seismic data collection and archiving, including Marinos Charalampakis, Giouli Petrou, Maria Sachpazi,
Stéphane Rondenay, Chin-Wu Chen, Jenny Suckale, Aleksandra Hosa, and many others.
I’d like to thank the National Observatory of Athens for their assistence with the seismology portion of project MEDUSA, and for access to the broadband data from a few of their seismic stations. Many thanks to IRIS-PASSCAL for providing the instruments and assisting with the archiving of the data, particularly Eliana Arias and George Slad.
Many thanks also to Chin-Wu, Jenny, and Aleksandra for sharing their experience working with the data, and for sharing their codes, which greatly helped me get up to speed on things when I started the project.
*
This research was carried out as part of project MEDUSA, funded by the NSF Continental
Dynamics Program, grant EAR-0409373
5

I had the opportunity to do field work in Greece twice during my thesis years, both memorable experiences. The first trip was for the final collection of the MEDUSA array data in which we traveled across the beautiful mountains of northern Greece. I am greatly indebted to Marinos, Aleksandra, and Giouli for their tireless efforts on that trip, and especially Marinos, who helped me discover my inner Greek (race car) driver, which
I never new existed, and haven’t seen since. My second trip to Greece was with a geology PhD student, Kyle Bradley, who brought me, a geophysics student, to help him with his final round of data collection in Evia for his dissertation. Thank you for that opportunity, Kyle, and for our many interesting, out-of-the-box discussions on the tectonics and geology of Greece. It was on this second trip that I also witnessed firsthand one of the countless terrible effects of Greece’s financial crisis: all the fuel stations were out of gas for a couple days during our trip, fallout from a fuel trucker strike. And so, it is with a sympathetic heart that I would like to thank the people of Greece for the opportunity to work in their country, with its rich history, both human and, I hope you will agree, geologic. I sincerely hope that a return to richness is well underway.
PhD programs are notorious for being wrought with challenges and obstacles, both scientific and personal, and my experience was certainly no exception. I would like to thank the many people that helped me navigate some of the more challenging times. I benefited from interesting discussions on seismic imaging and data processing with
Stéphane, Chin-Wu, Shane McGary, Tim Seher, Geoff Abers, Doug Miller, Mark Willis,
Gene Humphreys, Meghan Miller, YoungHee Kim, and many others. Thank you, Oli
Jagoutz, for our many stimulating conversations at the intersection of seismology and petrology. I would like to thank Alan Levander for organizing the CIDER 2011 workshop and all the workshop participants for the interesting lectures and discussions on the mountain belts of the world. Thank you, Dale Morgan, for the field adventures to the
Carribean, and for the countless interesting discussions about science and life. I think I may have found my keys.
I would also like to thank Dean Blanche Staton and Ombudsperson Toni
Robinson for their support and advice in navigating some of the more difficult stretches during the writing of this thesis. Thank you again, Chin-Wu, this time for your well written, organized thesis, which provided a much-appreciated guide for laying out this
6

thesis. A special thank you to Alejandra Quintanilla Terminel for her wisdom, advice, unwavering support, energy, encouragment, and kind-heart that helped keep me pushing to finish, even on the bleakest of days. I greatly appreciate your support and encouragement over the years. I would also like to thank Dan Burns for being a great adviser for my Master’s thesis, and for being a good source of honest, kind-hearted advice and support over all of my years at MIT. I couldn’t have done it without your encouragement, support, and friendship. Peace always. I would also like to thank
Hussam Busfar for his friendship, encouragement, and help in maintaining some semblance of work/life balance during some of the most difficult times in the writing of this thesis. Many thanks to Terri Macloon for helping me with many logistical things over the years, and for enjoyable commiserating talks on life that made many a dark day brighter. Thank you also to Susan at the writing center for helpful suggestions on the process of writing.
I spent the first three years in the PhD program working on two general exam projects that are not part of this dissertation. I would like to thank the members of my general exam committee: Nafi Toksöz, Dale Morgan, Brian Evans, and Rob van der
Hilst. Thank you to the Institute for Geophysics and Planetary Physics and Earth
Resources Laboratory (ERL) consortium for supporting my general exam projects.
Thank you, Nafi, for being my adviser during my pre-general’s years, and for introducing me to the interesting topic of focal depth determination. Thank you, Dale, for your willingess to be my adviser on my first general exam project, with little time before the exam. I would like to thank many people from the ERL for their help and support along the way: Mark Willis, Mike Fehler, Sadi Kuleli, Tim Seher, Junlun Lee, Fuxian Song,
Haijiang Zhang, Youshun Sun, Hussam Busfar, Nasruddin Nazerali, Abdulaziz
AlMuhaidib, Pierre Gouedard, Scott Burdick, Darrell Coles, Burke Minsley, Rama Rao,
Sue Turbak, and many others.
A special thank you to Paul Johnson for his mentoring on my first general’s project, for the opportunity to learn about the fascinating world of nonlinear elasticity, and for his advice and support during my years at Los Alamos National Laboratory
(LANL). I’m not sure I would have started, let alone finished, this thesis without your encouragement, support, and friendship. I would also like to thank many collegues at
7

LANL for their help and encouragment over the years, including Mike Fehler, James
TenCate, Carène Larmat, T. J. Ulrich, Scott Phillips, Lianjie Huang, James Rutledge, and many others.
My early years at MIT brought some life long friendships that were an essential source of support, encouragement, and much needed work/life balance. I would like to thank the 647 crew - Joe, Laura, Mauro, Melissa, Alejandro, Eric, and Bonna for all our good times together. Mauro and Melissa, thank you so much for your friendship and encouragement through challenging times. I will always remember our frozen pizza nights together with great fondness. Thank you to Eric and Virginia for their support, encouragement, and hospitality, particularly during the final stretch in writing this thesis.
I would like to thank Kerri for her support, encouragement, and friendship during the early years of my PhD. Thank you, Katerina, for being a good friend during my Master’s thesis, for your support and encouragement to persevere to finish my thesis, and for a warm introduction to the richness and beauty of Greek culture.
Finally, I would like to thank my family and close, old friends for all of their love and support over the many years since I came to MIT. Thank you to my friends from my undergraduate days – Ryan, Jon, Brian, and Jaimos – for all of your support and advice.
Thank you to my many Grant high school friends for all their love and support, including
Dono, Anne, Erica, the Daves (Gr., D., Ga., F.), J.K., Andrew, Matt, Meri, Carissa, Sean,
Tyler, and many others. And, most importantly, I would like to thank Jamie, my mother, father, brother, cousins, aunts, uncles, grandparents, Charlie, Dorothy, as well as, Snake
Boy, and Kineva, for all of your love, support, and encouragement. This would not have been possible without you.
8

1 Introduction.................................................................................................................. 18
1.1 Retreating convergent plate boundaries.................................................................. 18
1.2 The western Hellenic subduction zone and the MEDUSA experiment.................. 21
1.3 Thesis outline .......................................................................................................... 22
1.4 Figures..................................................................................................................... 24
2 Seismic investigation of the transition from continental to oceanic subduction along the western Hellenic subduction zone................................................................. 26
2.1 Introduction............................................................................................................. 27
2.2 Geologic Setting...................................................................................................... 29
2.2.1 Southern segment............................................................................................. 29
2.2.2 Northern segment............................................................................................. 30
2.2.3 Cephalonia transform fault .............................................................................. 30
2.2.4 Overriding plate ............................................................................................... 31
2.2.5 Constraints from previous seismic investigations............................................ 31
2.3 Methods................................................................................................................... 34
2.3.1 Preprocessing ................................................................................................... 35
2.3.2 Teleseismic migration...................................................................................... 36
2.4 Data ......................................................................................................................... 37
2.5 A priori model parameters ...................................................................................... 38
2.6 Results..................................................................................................................... 39
2.6.1 Composite images............................................................................................ 40
2.6.2 Along-strike variations..................................................................................... 42
2.7 Discussion ............................................................................................................... 44
2.7.1 Southern segment............................................................................................. 44
2.7.2 Northern segment............................................................................................. 47
2.7.3 Slab geometry along the WHSZ ...................................................................... 50
2.8 Concluding remarks ................................................................................................ 53
2.9 Tables...................................................................................................................... 55
9

2.10 Figures................................................................................................................... 56
3 Three-dimensional lithospheric structure along the western Hellenic subduction zone from Ps receiver functions..................................................................................... 67
3.1 Introduction............................................................................................................. 68
3.2 Method .................................................................................................................... 70
3.2.1 Preprocessing ................................................................................................... 70
3.2.2 Imaging ............................................................................................................ 72
3.3 Data ......................................................................................................................... 76
3.4 Model domain ......................................................................................................... 77
3.5 Results..................................................................................................................... 78
3.5.1 Southern Greece............................................................................................... 78
3.5.2 Northern Greece............................................................................................... 81
3.5.3 Back-arc ........................................................................................................... 83
3.5.4 Central Greece ................................................................................................. 84
3.6 Discussion ............................................................................................................... 86
3.6.1 Slab Moho........................................................................................................ 86
3.6.2 Overlapping slab segments? ............................................................................ 92
3.6.3 Slab transition from northern to southern Greece............................................ 93
3.6.4 Lithosphere-asthenosphere boundary .............................................................. 93
3.6.5 Overriding Moho ............................................................................................. 94
3.6.6 Where does the overriding lithosphere start?................................................. 102
3.6.7 Do subduction forces help support the Hellenides?....................................... 104
3.6.8 Top of crystalline basement ........................................................................... 107
3.9 Figures................................................................................................................... 113
4 Imaging the architecture of a developing core complex along the western Hellenic subduction zone............................................................................................................. 134
4.1 Introduction........................................................................................................... 134
4.2 Tectonic setting..................................................................................................... 136
4.3 Geodynamic modeling .......................................................................................... 139
10

4.4 Integrated structural interpretation........................................................................ 142
4.4.1 Collapsing crust along southern Greece ........................................................ 143
4.4.2 Building crust along northern Greece ............................................................ 146
4.5 Discussion ............................................................................................................. 148
4.5.1 Three-dimensional variation in the Pindos suture ......................................... 148
4.5.2 How much External crust has accreted to northern Greece? ......................... 149
4.5.3 How does the overriding mantle lithosphere “heal”? .................................... 151
4.6 Figures................................................................................................................... 153
5 Conclusions................................................................................................................. 157
A Chapter 2 supplemental material ............................................................................ 162
A.1 Introduction.......................................................................................................... 162
A.2 Teleseismic events and scattering modes............................................................. 162
A.3 A priori model parameters ................................................................................... 162
A.3.1 2-D regional strike ........................................................................................ 163
A.3.2 Background velocity model .......................................................................... 163
A.4 Differential slab retreat analysis .......................................................................... 164
A.5 Tables ................................................................................................................... 166
A.6 Figures.................................................................................................................. 169
B Chapter 3 supplemental material ............................................................................ 172
B.1 Introduction .......................................................................................................... 172
B.2 Automated multichannel alignment method ........................................................ 172
B.2.1 Multichannel preprocessing steps ................................................................. 173
B.2.2 Identifying misaligned stations ..................................................................... 177
B.2.3 Method description........................................................................................ 178
B.3 Synthetic test ........................................................................................................ 180
B.4 Error analysis........................................................................................................ 182
B.4.1 Conversion point error .................................................................................. 183
B.4.2 Image dip error .............................................................................................. 184
11

B.5 Figures.................................................................................................................. 187
Bibliography .................................................................................................................. 196
12

13

1.1 Map of the western Hellenic subduction zone and MEDUSA stations .................... 24
2.1 Map of the western Hellenic subduction zone including seismic stations and teleseismic events...................................................................................................... 56
2.2 Schematic showing segmented slab along the Hellenic subduction zone ................ 57
2.3 Composite migration images across southern Greece using all events .................... 58
2.4 Migration images across southern Greece using only free-surface-reflected phases 59
2.5 Composite migration images across northern Greece using all events..................... 60
2.6 Migration images across southern Greece using different event back-azimuth bins 61
2.7 Migration images across northern Greece using different event back-azimuth bins 62
2.8 Composite images across northern and southern Greece with local seismicity ....... 63
2.9 Geometry of slab and overriding lithosphere along northern and southern Greece . 64
2.10 Intermediate-depth seismicity along the western Hellenic subduction zone ............ 65
3.1 Map of western Hellenic subduction zone showing stations and teleseismic events
................................................................................................................................. 113
3.2 Schematic illustrating 3-D common conversion point imaging method ................ 114
3.3 Map view of 3-D receiver function model domain................................................. 115
3.4 Receiver function profiles beneath southern Greece filtered from 0.05 Hz to 0.5 Hz
................................................................................................................................. 116
3.5 Receiver function profiles beneath southern Greece filtered from 0.05 Hz to 1.5 Hz
................................................................................................................................. 117
3.6 Receiver function profiles beneath southern Greece with structural interpretation 118
3.7 Receiver function profiles beneath northern Greece .............................................. 119
3.8 Receiver function profiles beneath northern Greece with structural interpretation 120
3.9 Receiver function profiles beneath back-arc........................................................... 121
3.10 Receiver function profiles beneath back-arc with structural interpretation............ 122
3.11 Large-scale version of RF profiles from Figures 3.6, 3.8, and 3.10 ....................... 123
3.12 Receiver function imaging results for station VLS................................................. 124
14

3.13 Receiver function imaging results for station LKD................................................ 125
3.14 Receiver function imaging results for station RLS................................................. 126
3.15 Receiver function imaging results for station EVR ................................................ 127
3.16 Receiver function imaging results for station AGG ............................................... 128
3.17 Receiver function imaging results for stations KEK and JAN ............................... 129
3.18 Maps of Moho depth for the slab and overriding lithosphere................................. 131
3.19 Overriding Moho depths versus topography along the WHSZ .............................. 133
4.1 Geologic map of WHSZ, P-T-t paths of HP rocks, and schematic of subductionexhumation cycle .................................................................................................... 153
4.2 Integrated interpretation of crustal structure along southern Greece...................... 154
4.3 Integrated interpretation of crustal structure along northern Greece ...................... 155
A.1 Composite images for SL and NL using different projection line angles............... 169
A.2 Individual scattering mode images for SL and NL................................................. 170
B.1 P wavefield alignment from standard preprocessing .............................................. 187
B.2 Aligned P wavefield and incident wavefield before and after automated multichannel alignment........................................................................................... 189
B.3 SV receiver functions before and after automated multichannel alignment........... 191
B.4 Deconvolved P wavefield and resolution kernel before and after automated multichannel alignment........................................................................................... 192
B.5 Delay time corrections as a function of iteration .................................................... 193
B.6 Test of the 3-D common conversion point imaging method .................................. 194
15

2.1 Optimal background model used in 2-D GRT of the SL dataset.............................. 55
2.2 Optimal background model used in 2-D GRT of the northern line dataset .............. 55
3.1 1-D background model for receiver function imaging of southern Greece ............ 112
3.2 1-D background model for receiver function imaging of northern Greece ............ 112
3.3 1-D background model for receiver function imaging of back-arc ........................ 112
A.1 Selected events and scattering modes used in 2-D GRT of SL dataset .................. 167
A.2 Selected events and scattering modes used in 2-D GRT of NL dataset.................. 168
16

17

1.1 Retreating convergent plate boundaries
Convergent plate boundaries, where one lithospheric plate plunges beneath another, play a fundamental role in the plate tectonic cycle, shaping in many ways the composition and structure of the Earth. Their inherent asymmetry (e.g. King [2001]), the downgoing plate dips beneath the overriding plate, gives them a broader spectrum of characteristics compared to divergent plate boundaries. For example, the two interacting plates can have different compositions, with the typical scenario involving a denser oceanic plate sliding beneath a more buoyant continental plate (i.e. ocean-continent boundary); however, it is not uncommon to find either ocean-ocean or continent-continent convergent boundaries as well. Convergent boundaries also exhibit a great deal of diversity in their kinematics, as two relative convergence rates are needed to characterize their behavior: (1) the rate of overall plate convergence, measured between two points within the interior of the plates
(i.e. far from where active deformation occurs), and (2) the rate of subduction (or
“retreat” rate), a more local measure taken between the interior of the downgoing plate and the front of the subduction zone (e.g. outer part of trench or foreland basin) [Royden,
1993a]. The relative magnitude of these two rates provides another useful way of classifying convergent boundaries. When the rate of overall convergence exceeds the rate of subduction, a convergent boundary is termed an “advancing” boundary, a scenario that leads to horizontal shortening in the overriding plate. Conversely, a retreating
18

convergent system is one in which the subduction rate exceeds the overall convergence rate, causing the overriding plate to extend horizontally [Royden, 1993a].
Convergent boundaries also play a fundamental role in continent formation, where they display yet another one of their interesting characteristics, which is that overall convergence may be accommodated by several distinct subduction systems distributed over a broad region, rather than a single, well-defined plate “boundary”.
Thus, continental tectonics does not fit neatly within the broader plate tectonic paradigm
(e.g. Molnar [1988]). One notable example of this phenomenon occurs within the
Mediterranean region where overall Africa-Eurasia convergence is accommodated by several different “types” of subduction systems distributed across a width of over 1500 km. For example, the Alps are a continent-continent, advancing system, while the
Hellenic system is a retreating system that has both ocean-continent and continentcontinent parts, as described in Section 1.2.
Convergence within the retreating systems of the Mediterranean is generally oblique to the direction of overall Africa-Eurasia convergence, so it is not clear how each individual subduction system contributes to overall convergence. It has been argued that such a pre-collisional setting evolved naturally due to the irregular shape of the African margin as it began to collide with Eurasia (Royden [1993b], and references therein), the idea being that some buoyant continental parts of the Africa margin came in contact with
Eurasia relatively early (e.g. Arabia), leading to the development of advancing continentcontinent convergent boundaries (i.e. continental collision zones), while other parts of the
African margin still contained dense, oceanic lithosphere that continued to subduct rapidly, but in a retreating mode, due to the slowdown in the rate of overall convergence.
The juxtaposition of diverse convergent systems within the Mediterranean region has also fueled a long-standing debate as to the tectonic forces (i.e. dynamics) driving the distributed deformation in zones of continental convergence (e.g. Royden [1993b] and references therein). Some favor a mechanism by which the excess gravitational potential energy stored in the thickened crust of collisional zones (e.g. Arabia) drives the lateral extrusion of continental blocks (e.g. Anatolia) out and over the remnants of oceanic lithosphere (e.g. Mediterranean seafloor) [McKenzie, 1972; Sengor et al., 1985], while others prefer to invoke the pull of dense, oceanic lithosphere within retreating subduction
19

systems as the primary driver of continental blocks (e.g. Aegean) [Le Pichon and
Angelier, 1979; Royden, 1993a]. While countless geologic, geodynamic, and geophysical investigations have sought to discriminate between these two end member tectonic models, there is, as of yet no consensus regarding the dynamics driving deformation within the Mediterranean region.
The goal of this thesis is to use seismic imaging methods to constrain the structure of the lithosphere along a retreating convergent boundary that is undergoing a transition in the composition of its downgoing lithosphere (hereafter referred to as a “slab”), and to compare this seismic-based structure to the structure predicted in recent geodynamic studies that employ a retreating subduction system model for the WHSZ. Specifically, we focus on the following questions:
(1) What is the composition and geometry of the subducting slab, and how does it transition along strike?
(2) Is there evidence for along strike differences in the rate of slab retreat, and, if so, do these differences correlate with slab composition?
(3) How does the crustal thickness of the overriding lithosphere vary along strike?
(4) Where does the overriding lithosphere start, and how does this starting point relate to the slab, and to topography?
(5) Do we find evidence that the overriding lithosphere is built from crustal material scraped off the slab, and if so, how does this process relate to slab composition?
(6) Can we integrate geologic, geodynamic, and seismic constraints to capture the architecture of a developing core complex, including its accreted continental blocks and oceanic suture zones?
We’ll focus our investigation on the western portion of the Hellenic subduction zone, a well-documented retreating system in the eastern Mediterranean, where we can take advantage of two recently deployed high-density seismic arrays, as discussed in the following section.
20

1.2 The western Hellenic subduction zone and the MEDUSA experiment
The western Hellenic subduction zone (WHSZ) provides a natural laboratory for studying the dynamics of a retreating subduction system, as it exhibits well-documented variations in slab composition, retreat rate, and overriding lithosphere deformation. Oceanic lithosphere is currently subducting along southern Greece, where retreat rates are large
(~35 mm/yr) and the overriding lithosphere is extending rapidly (e.g. ~10 mm/yr across the Gulf of Corinth), while continental lithosphere is currently entering the trench along northern Greece, where retreat rates are slow (5-8 mm/yr) and overriding lithosphere extension is much smaller (Figure 1.1; see also Armijo et al. [1996], Baker et al. [1997],
McClusky et al. [2000], Finetti and Del Ben [2005], and geologic setting description in
Section 2.2). Extensive geologic studies have provided a detailed record of the subduction history along the WHSZ (e.g. Jacobshagen et al. [1978]; Burchfiel [1980]), which has been implemented in recent geodynamic modeling studies that are able to reproduce many of the key observables along the WHSZ, including the difference in retreat rates between the southern and northern segments [Royden and Papanikolaou,
2011], and the detailed deformation structures within the overriding lithosphere [Tirel et al., 2013]. However, such models have been difficult to test, as most seismic studies are unable to resolve geologic features with sufficient detail, and/or have focused on the active, southern segment of the WHSZ, with only limited constraints available on the structure of the subduction boundary along the northern segment.
As part of the Multidisciplinary Experiments for Dynamic Understanding of
Subduction under the Aegean Sea (MEDUSA) project, we deployed two high-density seismic arrays across northern and southern Greece in a fan-shaped geometry (Figure
2.1a). This geometry was ideal as it provided (1) the dense station spacing required for the 2-D teleseismic migration algorithm used in Chapter 2, (2) distributed stations for constraining along-strike variations in lithospheric structure using receiver function imaging methods, as described in Chapter 3, and (3) similar arrays across northern and southern Greece that facilitated a direct comparison of the lithospheric structure beneath these two contrasting tectonic regimes. We also collected data from all the publicly
21

available seismic stations across the WHSZ to further improve our resolution of alongstrike variations in lithospheric structure.
1.3 Thesis outline
This thesis is comprised of three main chapters. The first two main chapters each focus on interpreting the results from a specific seismic imaging method, 2-D teleseismic migration in Chapter 2, and 3-D receiver function imaging in Chapter 3, while Chapter 4 focuses on integrating the available geologic, geodynamic, and seismic constraints into a detailed picture of the overriding lithosphere. In the following, I give a brief description of each chapter, including its salient objectives.
Chapter 2 employs a 2-D Generalized Radon Transform-based teleseismic migration algorithm (2-D GRT) to identify sharp changes in material properties beneath each MEDUSA array. The 2-D GRT provides high-resolution images of perturbations in
P- and S-wave velocity, but requires (1) dense seismic arrays and (2) careful preprocessing of the raw data (total wavefield) to isolate the scattered wavefield. This chapter also includes several improvements to the preprocessing that enhance image resolution. The resulting 2-D GRT images are used to interpret the composition of the slab beneath southern and northern Greece, and to infer the amount of differential slab retreat between the southern and northern parts of the slab.
In Chapter 3, I develop a 3-D receiver function (3-D RF) method to image dipping discontinuities in S-wave velocity. It has the advantage of being able to use data from both station arrays and isolated stations, but has lower resolution than the 2-D GRT method. The method is tested with synthetic data from the RAYSUM package of
Frederiksen and Bostock [2000], and then applied to form common conversion point images across northern Greece, southern Greece, and the back arc. A single station RF analysis is also performed on data from a few isolated stations in central Greece to constrain the structure of the lithosphere beneath this region. These images are used to produce maps of the slab Moho and overriding Moho across the WHSZ, which are then compared to previously published Moho constraints along the WHSZ. Notable topics covered include (1) resolving the differences in previous interpretations of the slab
22

structure beneath northern Greece, (2) identifying the starting point of the overriding lithosphere, and (3) demonstrating that the observed relationship between the geometry of the slab, the overriding Moho, and surface topography provide strong evidence in support of a retreating subduction model for the WHSZ. We also interpret the location of the top of the crystalline basement, and describe several of its notable features and implications.
In Chapter 4, I use the RF images from Chapter 3 to better understand how the overriding lithosphere is built and deformed in a retreating subduction system, such as the
WHSZ. This is done by using previous geologic and geodynamic constraints to refine our interpretation of the overriding lithosphere structure observed in our RF images, separating it into two distinct continental domains (External and Internal Hellenides) separated by a thin, weak suture zone marking the subduction of the intervening oceanic domain (Pindos). Comparing differences in the crustal architecture between southern and northern Greece provides an unprecedented view into how the overriding lithosphere develops as a retreating subduction system undergoes a transition in slab composition.
The key points from these three main chapters are summarized in Chapter 5.
23

1.4 Figures
Figure 1.1: Map of study area along the western Hellenic subduction zone. Yellow lines show the active, southern segment boundaries: thrust front of the Mediterranean ridge
(dashed) and trench (solid). Dashed orange line shows thrust front of northern segment.
Bright red line denotes the Cephalonia transform fault (CTF) and dark red dotted line marks the Apulian Escarpment (i.e. the oceanic-continental transition). Dashed gray line outlines the North Aegean trough (NAT) and the central Hellenic shear zone. GPS velocities (magenta arrows) relative to Eurasia are from McClusky et al. [2000]. Seismic stations deployed across southern and northern Greece are denoted by white squares and circles, respectively. Pel is short for Peloponnesus.
24

25

*
Abstract
The western Hellenic subduction zone (WHSZ) exhibits well-documented along-strike variations in lithosphere density (i.e. oceanic vs. continental), subduction rates, and overriding plate extension. Differences in slab density are believed to drive deformation rates along the WHSZ; however, this hypothesis has been difficult to test given the limited seismic constraints on the structure of the WHSZ, particularly beneath northern
Greece. Here, we present high-resolution seismic images across northern and southern
Greece to constrain the slab composition and mantle wedge geometry along the WHSZ.
Data from two temporary arrays deployed across Greece in a northern line (NL) and southern line (SL) are processed using a 2D teleseismic migration algorithm based on the
Generalized Radon Transform. Images of P- and S-wave velocity perturbations reveal
N60E dipping low-velocity layers beneath both NL and SL. The ~8 km thick layer beneath SL is interpreted as subducted oceanic crust while the ~20 km thick layer beneath NL is interpreted as subducted continental crust. The thickness of subducted
* Most of this chapter has been published as “Pearce, F.D., S. Rondenay, M. Sachpazi, M.
Charalampakis, and L.H. Royden (2012), Seismic investigation of the transition from continental to oceanic subduction along the western Hellenic Subduction Zone. J.
Geophys. Res.
, 117(B07306), http://dx.doi.org/10.1029/2011JB009023”
26

continental crust inferred within the upper mantle suggests that ~10 km of continental crust has accreted to the overriding plate. The relative position of the two subducted crusts implies ~70-85 km of additional slab retreat in the south relative to the north.
Overall, our seismic images are consistent with the hypothesis that faster sinking of the denser, oceanic portion of the slab relative to the continental portion can explain the different rates of slab retreat and deformation in the overriding plate along the WHSZ.
2.1 Introduction
The Hellenic subduction system provides a natural laboratory for exploring the relationship between slab density, subduction rate, and overriding plate deformation.
Extensive GPS studies have documented rapid subduction along southern Greece (~35 mm/yr) and slow subduction (5-8 mm/yr) along northern Greece [Figure 2.1a; McClusky et al., 2000; Hollenstein et al., 2008]. Marine seismic data have shown that the foreland of the northern Hellenides consists of continental lithosphere [Finetti and Del Ben, 2005], whereas the foreland of southern Greece, beneath the Ionian Sea, is of oceanic lithosphere affinity [Finetti et al., 1991; de Voogd et al., 1992]. Thus, the Western
Hellenic Subduction Zone (WHSZ) can be separated into northern and southern segments, each with a different convergence rate and slab composition [Papanikolaou and
Royden, 2007]. These segments are dextrally offset by ~100 km along the Cephalonia transform fault, which has recently (~1-4 Ma) linked with the North Anatolian fault system through a broad region of extensional and strike-slip faults known as the Central
Hellenic Shear Zone [Papanikolaou and Royden, 2007; Reilinger et al., 2010; Vassilakis et al., 2011].
Numerous tectonic models have been proposed to explain the large variations in convergence rate and overriding plate deformation along the western Hellenic subduction zone. Such models generally appeal to two large-scale tectonic forces: (1) the westward extrusion of Anatolia driven by Arabia collision [McKenzie, 1972; Sengor et al., 1985] and (2) the southwestward pull of the subducting Ionian lithosphere along the southern segment [Le Pichon and Angelier, 1979; Royden, 1993]. The importance of rapid trench retreat along the southern segment has been shown through GPS studies (i.e. larger, southerly velocity of the Aegean block compared to Anatolia) and models that explain the pattern of extension in the overriding plate [Meijer and Wortel, 1996; Gautier et al.,
27

1999; McClusky et al., 2000]. In contrast, slow convergence along the northern segment has been attributed to the resistance offered by the positively buoyant continental lithosphere to being subducted [Taymaz et al., 1991], a resistance that may have led to its separation from the rapid subduction regime south of the Cephalonia transform fault
[e.g., Kahle et al., 2000; Royden and Papanikolaou, 2011]. Recent numerical and experimental models have also shown that variations in slab density play an important role in driving slab rollback and overriding plate motion [Brun and Faccenna, 2008;
Husson et al., 2009; Faccenna and Becker, 2010; Chemenda et al., 1996; Bialas et al.,
2011].
The transition between the northern and southern segments of the WHSZ has been the focus of intense study, as it is one of the most seismically active regions in Europe.
At shallow depths, the Cephalonia transform fault defines the northern limit of rapid subduction along the WHSZ; however, the nature of the separation between the two segments at depth is highly debated. Numerous authors have suggested a trench-parallel slab tear may be propagating from north to south along the WHSZ in response to progressive continental collision [Wortel and Spakman, 2000 and references therein].
The Cephalonia transform fault has also been proposed as the surface manifestation of a vertical tear in the lithosphere, termed a Subduction-Transform Edge Propagator, which separates the northern and southern segments at depth [Govers and Wortel, 2005]. More recently, Suckale et al. [2009] suggest that a trench-perpendicular tear in the subducted slab beneath the Central Hellenic Shear Zone may have developed along the boundary between oceanic and continental lithosphere. Based on geodynamic modeling, Royden and Papanikolaou [2011] argue that the recent entry of denser, oceanic lithosphere into the southern segment has produced faster trench retreat, leading to the segmentation of the WHSZ along the Cephalonia transform fault as shown in Figure 2.2. It is not known how the differential slab retreat may be accommodated at greater depth (D > 30 km), perhaps through bending and/or tearing. Such subduction models have been difficult to test, as most seismic studies have focused on the active, southern segment, with only limited constraints available on the structure of the subduction boundary along the northern segment.
28

The objective of this study is to better constrain the transition from the northern to southern segment of the WHSZ through detailed imaging of the subducted slab in both regions. To do so, we deployed two high-density seismic arrays across northern and southern Greece (Figure 2.1a), as a part of the Multidisciplinary Experiments for
Dynamic Understanding of Subduction under the Aegean Sea (MEDUSA) project. Here, we present and discuss high-resolution images obtained by 2-D teleseismic migration of the data recorded by these arrays. The seismic images introduce new, independent evidence that continental crust is subducting below northern Greece, in contrast to the oceanic crust subducting beneath the Peloponnesus in the south. The relative position of the two segments suggests 70-85 km of additional trench retreat may have occurred along the southern segment compared to the northern segment, thereby providing a simple mechanism that can explain the majority of the ~100 km offset along the Cephalonia transform fault. Thus, our results provide important new constraints on the composition and geometry of the subducted lithosphere along the WHSZ.
2.2 Geologic Setting
The Hellenic subduction zone defines the boundary between the slowly converging
African and Eurasian plates within the central Mediterranean region (Figure 2.1a). The western portion of the Hellenic subduction zone has a length of approximately 1000 km trending northwest from the central Adriatic Sea to the west coast of Crete. In this section, we review the geological and geophysical constraints on the southern segment, the northern segment, the Cephalonia transform fault, and the overriding plate.
The oceanic crust beneath the Ionian Sea is probably Triassic or Jurassic in age and consists of approximately 5-8 km of igneous crust overlain by 6-10 km of sedimentary cover [Makris, 1985; de Voogd et al., 1992; Kopf et al., 2003; Finetti and Del Ben,
2005]. Gravity anomalies imply a very dense, negatively buoyant slab within the upper
100–200 km of the southern segment, consistent with the subduction of old oceanic
29

lithosphere [Royden, 1993; Tsokas and Hansen, 1997]. Pliocene–Quaternary arc volcanoes are located ∼ 200 km north and east of the present-day Hellenic trench (Figure
2.1a) [Fytikas et al., 1976; Pe-Piper and Piper, 2007]. Thrust faults and folds occur within the accretionary prism outboard of the trench, over a width of several hundred kilometers to the front of the Mediterranean ridge [Kopf et al., 2003].
The continental crust beneath the southern Adriatic Sea is continental or transitional in character, with a crustal thickness of 25-30 km [e.g. Morelli et al., 1975; Marone et al.,
2003; Cassinis et al., 2003; Finetti and Del Ben, 2005]. The active thrust front of the northern segment lies just west of the Greek coastline, near the island of Corfu and between the island of Paxos and mainland Greece [Monopolis and Bruneton, 1982;
Finetti and Del Ben, 2005; Vassilakis et al., 2011]. There is no expression of a trench in the bathymetry along the northern segment, but gravity data indicate that the basement is flexed downward beneath the thrust front and that the resulting depression is filled with sedimentary foredeep deposits [Moretti and Royden, 1988]. The southern boundary between Apulian continental crust and Ionian oceanic crust occurs along the Apulian escarpment, which is marked by an abrupt change in water depth and Permian-Triassic rift faults (Figure 2.1a) [Finetti and Del Ben, 2005].
The Cephalonia transform fault separates the northern and southern segments of the western Hellenic subduction zone [e.g. Dewey and Sengor, 1979; Finetti, 1982; Kahle and Muller, 1998; Kahle et al., 1995; Hollenstein et al.
, 2008]. Focal solutions from earthquakes located along the Cephalonia transform fault show right-lateral slip on steeply-dipping, southwest-striking fault planes and also thrust faulting on northeaststriking fault planes [Royden and Papanikolaou, 2011]. The WHSZ appears to be dextrally offset by ~100 km across the Cephalonia transform fault; however, the precise offset is difficult to determine given the distributed network of faults and block rotations along the south side of the transform [Vassilakis et al., 2011]. The Cephalonia transform
30

fault extends to the northeast into mainland Greece to merge with the broadly defined zone of dextral and extensional deformation in the Central Hellenic Shear Zone (Figure
2.1a) [Roberts and Jackson, 1991; Armijo et al., 1996; Goldsworthy et al., 2002;
Papanikolaou and Royden, 2007; Vassilakis et al., 2011].
The overriding plate of the WHSZ consists of subduction-related thrust units that have been subjected to variable modes of extension over time. From the Oligocene to the late
Miocene, upper plate extension was generally arc-perpendicular in response to a regime similar to back-arc spreading – though one that encompassed both synarc and forearc domains [Mercier et al, 1989; Papanikolaou, 1993]. Extension began to focus in the overriding plate of the southern segment in the Late Miocene to Early Pliocene
[Papanikolaou and Royden, 2007]. Since the Late Pliocene, extension has been concentrated along a series of normal faults oriented approximately E-W and crosscutting older arc-parallel structures. Currently, E-W faults along the Gulf of Corinth and Gulf of
Evia accommodate most of the extension in central Greece [McClusky et al, 2000].
Extension in northern Greece has remained arc-parallel but has been much slower since the Late Miocene [Meyer et al., 2002]. This contrast in extension regime between the northern and southern segments has been attributed to an acceleration of the rate of motion of the Hellenic trench to the south of Cephalonia in late Miocene time, in response to a transition to negatively buoyant oceanic lithosphere in that region [e.g.,
Papanikolaou and Royden, 2007]. This dichotomy has led to considerable differences in crustal thickness of the overriding plate between the northern and southern segments, which have been documented through various seismic studies described in the next section.
A variety of seismic studies have been undertaken in the past decades to image the
Hellenic subduction zone. These include traveltime tomography [Ligdas et al. 1990;
Spakman et al. 1993; Papazachos and Nolet 1997; Piromallo and Morelli 2003; Schmid
31

et al. 2006], reflection and refraction profiles [Makris 1978; Hirn et al. 1996; Mascle and
Chaumillon 1998; Clément et al. 2000; Bohnhoff et al. 2001; Kopf et al. 2003; Kokinou et al. 2003, 2005], receiver-function analyses [Saunders et al. 1998; Knapmeyer and
Harjes 2000; Tiberi et al. 2001; Li et al.
2003; van der Meijde et al. 2003; Meier et al.
2004; Kreemer et al. 2004; Endrun et al. 2005; Sodoudi et al. 2006; Zhu et al.
2006;
Snopek et al. 2007; Gesret et al. 2010], surface-wave dispersion [Calcagnile et al. 1982;
Martinez et al. 2001; Pasyanos and Walter 2002; Bourova et al. 2005; Karagianni et al.
2005; Di Luccio and Pasyanos 2007; Endrun et al. 2008], and seismic anisotropy measurements [Hearn 1999; Schmid et al. 2004]. Here, we review the results from these previous seismic studies, focusing on the insight they provide on the Moho of the overriding plate and subducting slab.
The Moho of the overriding plate has been the subject of several regional investigations across the Gulf of Corinth [Tiberi et al.
2000, 2001; Clément et al. 2004;
Zelt et al. 2005), central Greece [Sachpazi et al. 2007], and the Aegean region [Makris
1978; Marone et al. 2003; Tirel et al. 2004]. Traveltime tomography and receiver function results from these investigations indicate that the depth of the continental Moho ranges from 20 to 45 km. Depths in excess of 40 km are found beneath the Hellenides in the Peloponnesus and mainland Greece, and are attributed to isostatic compensation of surface relief [Tiberi et al., 2000; Marone et al., 2003]. Shallower Moho depths (<30 km) are found beneath the eastern Gulf of Corinth and the Aegean Sea, with the shallowest depths (<25 km) below the North Aegean Trough and the Sea of Crete [Tirel et al., 2004;
Zelt et al., 2005; Sodoudi et al., 2006]. Sachpazi et al. [2007] document a line striking
SW–NE from the western Gulf of Corinth to the North Aegean Trough as marking the boundary between a domain of thinned crust (~20-35km) to the southeast (eastern
Peloponnesus, Attiki Peninsula, Evia) and one of thicker crust (>35 km) beneath central and northern Greece. In general, the Moho of the overriding plate in northern Greece is more poorly constrained than in the south due to the limited seismic station coverage in this region.
The nature of the foreland slab, as seen prior to reaching the trench, has been constrained primarily by marine reflection and refraction profiles. In the southern segment, such profiles indicate that the crust beneath the Ionian Sea is likely oceanic in
32

origin, with an estimated thickness ranging between 5 km [Finetti et al., 1991; Finetti and
Del Ben, 2005] and 8 km [de Voogd et al., 1992]. The thickness of the overlying sediments varies greatly from 4-6 km west of the deformation front to up to 10 km in the accretionary prism [Hirn et al. 1996; Mascle and Chaumillon 1998; Bohnhoff et al. 2001;
Kopf et al. 2003; Kokinou et al. 2003]. East of the trench, detailed active source studies in the Ionian Islands have identified a landward dipping reflector of variable topography at 13 km depth, which is interpreted as the interplate boundary of the subduction zone
[Hirn et al. 1996; Clement et al. 2000]. In the northern segment, a recent series of marine seismic lines across Apulia give a detailed picture of the foreland crust currently entering the subduction system. In particular, the M-34 seismic line [Finetti and Del Ben, 2005] represents a near-parallel extension of the northern land-based profile presented in this study (see Section 2.7.2 and Figure 2.9), starting ~25 km to the west of the coast. The interpreted M-34 line shows that the undeformed continental crust entering the thrust front has ~8 km of sedimentary rocks, ~12 km of crystalline upper crust, ~7 km of lower crust and a Moho depth of ~28 km (considering a water depth of ~1 km). As we shall see, these marine seismic data provide critical constraints for the interpretation of our seismic images beneath mainland Greece.
To date, the structure of the WHSZ at depth has been constrained primarily by seismic traveltime tomography at global scale [Ligdas et al., 1990; Spakman et al., 1993;
Karason, 2002; Piromallo and Morelli, 2003] and regional scale [Papazachos and Nolet,
1997; Tiberi et al., 2000]. These tomographic models show the subducted slab as a tabular, high-velocity anomaly dipping to the NE beneath the western Hellenic subduction zone. Global models show that the fast anomaly extends well below the transition zone at 660 km, indicating slab penetration into the lower mantle [van der Hilst et al., 1997]. At regional scale, there is evidence for a sudden increase in subduction angle between 70 and 90 km depth beneath the Gulf of Corinth [Papazachos and Nolet,
1997; Tiberi et al., 2000; Sodoudi et al.
, 2006]. The distribution of earthquake hypocentres outlines a diffuse Wadati–Benioff zone that abruptly increases its dip at 90–
100 km [Papazachos and Nolet, 1997]. Some tomographic models show the northeast dipping velocity anomaly appears to be weaker north of the Cephalonia transform fault
[e.g. Wortel and Spakman, 2000] and has a lateral disruption beneath the Cephalonia
33

transform fault [Suckale et al., 2009; Hosa, 2008]. These observations have been interpreted as evidence for possible slab detachment in this area [Spakman et al., 1988;
Carminati et al., 1998; Wortel and Spakman, 2000]. In contrast, P and S receiver functions by Sodoudi et al. [2006] show a continuous Moho of the subducting lithosphere beneath northern and southern Greece between 40 and 220 km depth while Zelt et al.
[2005] observed a reflection from the subducted slab beneath the western Gulf of Corinth at approximately 75 km depth. Thus, the available seismic evidence suggests a continuous slab above ~200 km depth with a possible slab tear beneath the Cephalonia transform fault at greater depths.
A high-resolution seismic profile of the southern segment based on migrated teleseismic scattered waves was recently presented by Suckale et al [2009]. Their image shows the subducted crust as a dipping low-velocity layer extending from 40 km depth beneath the west coast of the Peloponnesus to at least 80 km depth beneath the Isthmus of
Corinth. The observed low-velocities are attributed to hydrated metabasalts in the oceanic crust, and the reduction in signal beyond 80 km depth suggests progressive dehydration-eclogitization of these rocks. The data used by Suckale et al. [2009] are reanalyzed in this paper, in conjunction with the data collected above the northern segment, to produce consistent images of the southern and northern segments that will be readily amenable to comparison.
2.3 Methods
The imaging method used in this study migrates scattered signals in the coda of teleseismic P-waves recorded by dense arrays of broadband seismographs to identify discontinuities in material properties within the subsurface [Bostock et al. 2001; Shragge et al. 2001; Rondenay et al. 2001]. It assumes that the scattered wavefield is generated by volumetric perturbations in P - ( d α / α ) and S -wave velocities ( d β / β ) within a smoothly varying background velocity model. Preprocessing of the raw data (total wavefield) to isolate the scattered wavefield is critical to forming a high-quality image. The following sections briefly describe the data preprocessing algorithm and the imaging method used in this study.
34

A multichannel approach similar to that of Bostock and Rondenay [1999] is employed to isolate the scattered wavefield from the total recorded wavefield. Such multichannel schemes are favored over single-station deconvolution because they provide a more robust estimate of the incident wavefield and thus more stable deconvolved waveforms to higher frequencies. The approach involves the following steps [see Rondenay et al., 2005 for details]: (1) transform recorded wavefield from N-E-Z to upgoing P-SV-SH using the free-surface transfer matrix [Kennett, 1991]; (2) apply multichannel cross-correlation
[VanDecar and Crosson, 1990] to align the wavefield with respect to the incident P wave;
(3) obtain the incident wavefield from the first principal component of the P wavefield;
(4) deconvolve the incident wavefield from each residual component to produce an estimate of the normalized scattered wavefield.
In this study, several improvements have been made to the preprocessing workflow described above. First, the longest possible incident wavefield signals (up to
~180 s) are used to ensure that source-side scattering is included in the estimate of the incident wavefield. Second, several iterations over preprocessing steps 2 to 4 are performed to precisely align the scattered wavefield relative to the incident P-wave.
Lastly, in the deconvolution step, an optimal damping parameter (i.e., water level) is independently determined for each station component – as opposed to using ad-hoc, uniform damping across all components. We chose the smallest damping value such that unstable oscillation are restricted below a prescribed energy threshold (in our case
~0.01% of the undamped energy). If this rather stringent criterion is not met, then the damping is fixed at 5% of the peak in the source wavelet’s power spectrum. This procedure reduces overdamping and excessive low-pass filtering of the scattered wavefield. In summary, these improvements yield an input scattered signal that is better aligned along the incident P-wave and more stable than that obtained by the traditional workflow of Rondenay et al [2005].
35

The teleseismic migration approach used here assumes single scattering from 2-D line scatterers embedded in a smoothly varying 1-D background model [Bostock et al., 2001].
The approach is based on the Generalized Radon Transform (GRT) and its inverse, and is thus commonly referred to as the 2-D GRT inversion [Rondenay et al., 2005]. The inverse problem can be viewed as a weighted diffraction stack over all sources and receivers that yields an estimate of the scattering potential at a given point in the subsurface, with the weights determined by the analogy between the forward-scattering equation and the GRT [see, e.g., Miller et al., 1987]. The scattering potential is then linearly inverted for velocity perturbations at this point [Rondenay et al., 2005]. The inverse problem is solved for all points in model space to obtain a 2D image of velocity perturbations. For a full theoretical derivation of the 2-D GRT inversion, see Bostock et al. [2001].
The 2-D GRT inversion operates on a series of individual forward- and backscattered modes. The contribution of each mode is inverted based on analytical expressions for the traveltimes and amplitudes of the relevant combination of incident and scattered waves [Rondenay et al., 2001]. In particular, we consider the following scattering modes [see, e.g., Rondenay et al., 2010, for the full list of modes]: the incident
P -wave forward scattered as an S -wave (P d s); the free-surface-reflected P -wave backscattered as a P -wave (Pp d p); the free-surface-reflected P -wave backscattered as an
S -wave (Pp d s); and the free-surface-reflected S -wave backscattered as an Sv -wave (Ps d s| v
) and Sh -wave (Ps d s| h
). The various scattering modes are weighted based on the rationale developed in Rondenay et al. [2001].
The theoretical resolution of the 2D-GRT has been extensively studied using synthetic and field data [Bostock et al. 2001; Shragge et al. 2001; Rondenay et al. 2005,
2008; Rondenay, 2009], and has been shown to depend largely on the frequency content of the scattered signal and on the source/receiver distribution. The maximum volume resolution is approximately equivalent to a quarter of the wavelength of the scattered signal (for backscattered waves). Given the high cut-off frequency of 0.5 Hz used in this study, we expect a maximum volume resolution on the order of 2–3 km for structure in the lower crust and upper mantle.
36

The robustness of the resulting images has been shown to depend largely on the degree to which the geometrical assumption of the technique are met in a given study area [Rondenay et al.
, 2005]. For the western Hellenic system, given the rapid transition in lithosphere composition and convergence rates between the southern and northern segments, we expect along-strike variations in subsurface structure that may violate the
2-D assumption. The potential effects of these variations on the resulting image will be tested in Section 2.6.2 by examining contributions from events with different backazimuths.
2.4 Data
The data analyzed in this study were collected using portable, three-component broadband seismometers from the IRIS-PASSCAL instrument pool. The instruments were deployed in two temporary arrays: a first array spanning across southern Greece was operational from June 2006 to October 2007 (SL); and a second array extending across northern Greece was operational from November 2007 to April 2009 (NL). Each temporary array was deployed roughly perpendicular to the strike of the western Hellenic subduction zone in a fan-shaped pattern, as shown in Figure 2.1a. The array geometry was chosen to provide dense station coverage near the trench for maximum resolution at shallow depths and a more diffuse cloud of stations further from the trench to accommodate other types of seismic analyses (e.g., tomography, surface waves dispersion, shear-wave splitting).
As in previous applications of 2-D GRT inversion, the selection of useful events is done on the basis of the following criteria [see, e.g., Rondenay 2001; Rondenay 2005;
Suckale et al., 2009]: (1) epicentral distance from the center of the array between 30° –
90°, to avoid mantle transition zone triplications and the core shadow zone; (2) a magnitude (Ms, Mb, or Mw) greater than 5.5; (3) an incident P-wave arrival that can be identified across the entire array; (4) no overlaps with foreshocks or aftershocks; and (5) minimal contamination of the coda by secondary phases (e.g., PcP, PP). To avoid biases in back-azimuthal coverage that could affect the resulting image, we tighten the selection criteria for regions that are seismically very active and thus return large numbers of
37

events – for example Indonesia. For such regions, we apply the 2-D GRT to each event and include only the events that yield consistent structure across each scattering mode image. The combination of selection criteria and improved preprocessing workflow (c.f.,
Section 2.3.1) allowed the inclusion of events from regions of sparse and relatively weak seismicity, such as central Africa. As a result, we obtained over 50 high-quality events with comprehensive back-azimuthal distribution for both NL and SL arrays (Figure 2.1b;
Table A1, A2 in Appendix A).
A last measure of data selection and preprocessing is implemented to account for the fact that stations from both arrays were installed in small buildings such as churches and monasteries, as described by Suckale et al. [2009]. This issue is addressed by performing tight quality control on the data to assure the removal of site-specific noise.
Specifically, the spectral content of the signal recorded at each station is inspected for every selected event. Based on this information, a unique filter that emphasizes bodywave signal is designed and applied to each station-event pair. Station-event pairs that exhibit excessive noise within the body-wave band are excluded from the analysis. On average, this procedure leads to the removal of 2 to 4 stations for each event, with no significant loss in sampling density or aperture of the array.
2.5 A priori model parameters
The 2-D GRT inversion relies on the a priori knowledge of two sets of model parameters: the 2-D regional strike and the background velocity field. The 2-D regional strike is the dominant azimuth of line-scatterers in the model space. In the case of a subduction zone, it is often considered to be the azimuth perpendicular to the line of the slab’s steepest descent [see, e.g., Rondenay et al., 2010]. For a cloud of stations, the 2-D strike may be found by producing images along profiles of various azimuths and determining the azimuth that yields the most focused signal from the subducted crust. This exercise applied to both SL and NL arrays yields an optimal 2-D strike of N30W ± 10 ° , meaning that the profiles will be computed along an azimuth of N60E (see Appendix A for details,
Figure A1). We note that local estimates of the slab dip-direction along the southern
Peloponnesus derived from multi-azimuth analysis of receiver functions show
38

counterclockwise deviations from N60E by ~20° [Gesret et al., 2011]. However, these local deviations in dip-direction are unlikely to influence the SL image because the analyzed stations are > 50 km to the south of SL (see Section 2.6.2), and show some evidence for a rotation towards N60E from south to north (i.e. towards SL).
An optimal background velocity model is determined by identifying the 1-D P- and S-wave velocity model that produces a consistent location of target structures (e.g., overriding Moho, subducted crust) between different scattering modes. We start with a
1-D P-wave model for SL and NL derived by averaging P-wave velocities from local tomographic models beneath each line [Papazachos et al., 1995; Papazachos et al., 1997].
We then adjust the 1-D S-wave velocities in each model layer using realistic values of
Vp/Vs ratios until we obtain a consistent location of the target structures between each scattering mode image (see Appendix A for details, Figure A2). This procedure yields optimal 1-D velocity models for SL and NL that are described in Tables 1 and 2, respectively.
Uncertainty in the appearance of target structures, beyond that related to the theoretical resolution of the 2-D GRT (c.f., Section 2.3.2), may also be caused by velocity heterogeneity and anisotropy not adequately represented by the optimal background model [Rondenay et al., 2005]. While such uncertainties are difficult to quantify, we can estimate them by applying the 2-D GRT to both real data and synthetic data computed with the RAYSUM package [Frederiksen and Bostock, 2000] for a range of realistic
Vp/Vs and anisotropic parameters. The results of this test suggest that discontinuities in the composite image have a dip uncertainty of <3° and a depth uncertainty of <4 km, and that a low-velocity layer exhibiting typical velocity reduction for subducted oceanic crust
(Hacker and Abers [2004]) has a thickness uncertainty of <4 km.
2.6 Results
In this section, we describe the 2-D GRT images obtained across the southern and northern segments of the western Hellenic subduction zone. First, we show the composite images beneath the SL and the NL arrays, which are produced by simultaneous inversion
39

of all selected events. Then, we present images formed using a subset of events from different back-azimuths to examine along-strike variations in lithospheric structure.
The composite images for SL and NL are shown in Figures 3 and 5, respectively. They are produced by simultaneously inverting all the events selected for each array (Figure
2.1b; Tables A1 and A2 in Appendix A), using the profile orientations shown in Figure
2.1a and the background velocity models from Tables 1 and 2. The images show perturbations in either P- or S-wave velocity with red representing a velocity reduction and blue a velocity increase. The method is sensitive to velocity discontinuities, which are identified by crossovers in the color scale, i.e., red-to-blue indicating slow-to-fast and blue-to-red indicating fast-to-slow velocity perturbations. In the remainder of this paper, we will refer to discontinuities marking an increase/decrease in velocity with depth as positive/negative discontinuities, respectively.
As in previous applications of the 2-D GRT, we will focus our discussion on the
S-velocity ( d β / β ) profile, as d β / β images have been shown to be more robust than the d α / α images [Rondenay et al., 2001; Rondenay, 2005, 2009]. However, the d α / α profiles are also shown as they help support the identification of actual structures as opposed to artifacts from multiples, which are more prevalent in the d β / β images.
SL images
The SL composite images show two prominent features: a low velocity layer (LVL) dipping to the ENE from ~30 to 80 km depth, and a positive, sub-horizontal discontinuity with an average depth of ~35 km (Figure 2.3). These structures are similar to those found by Suckale et al. [2009]; however, the improved resolution in this study allows us to clarify several issues raised in that paper.
The LVL dips at ~17° with an average thickness of ~8 km (Figure 2.3), which is significantly thinner than the 15-20 km thickness reported in Suckale et al. [2009]. It appears to change thickness abruptly at a depth of ~40 km, leading us to separate it into updip and downdip segments. The updip segment (i.e. above 40 km depth) is
40

characterized by a greater average thickness of ~12 km, a sharp upper boundary, and some potential internal structure. Conversely, the downdip segment (i.e. below 40 km depth) thickens gradually from 7 km to 10 km with increasing depth and is marked by a more diffuse upper boundary. The d β / β magnitudes along the top and bottom of the LVL are approximately -10% and +10%, respectively. However, the magnitude of these perturbations should be interpreted with caution because of 1) the rather diffuse color crossovers marking the upper and lower boundaries of the LVL, and 2) the potential for amplification effects due to data sparseness – an effect that has been documented in numerical and field data examples [Rondenay et al., 2005; MacKenzie et al., 2010].
The signal marking the top of the LVL diminishes rather abruptly at a depth of
~80 km on both the d α / α and d β / β composite images, while a faint signal from the base of the LVL appears to continue to ~125 km depth. We can assess the robustness of the signal marking the base of the LVL between 80-125 km by considering the contributions of the individual scattering modes to the final image (Figure A2a in Appendix A). As was already noted in the study of Suckale et al. [2009], this signal is strong on all the individual modes but it does not necessarily stack constructively because of LVL dip discrepancies between the forward and backscattered modes (see Figure A2 and discussion in Appendix A, Section A3.2). We note that when the image is constructed using only the backscattered modes (Figure 2.4), the base of the LVL produces a strong and robust signal down to 125 km depth.
The positive sub-horizontal discontinuity emerges from near the separation between the updip and downdip segments of the LVL (Figure 2.3) and is clearly observed at ~40 km depth beneath the Hellenides (X
SL
= 60 km). It shallows rapidly to ~34 km beneath the eastern Peloponnesus (X
SL
= 150 km) then dips gently beneath eastern
Greece to attain a depth of ~30 km beneath the Aegean coast (X
SL
= 250 km). The discontinuity exhibits an average d β / β of approximately +10% beneath the Hellenides and eastern Greece. Its magnitude is more difficult to estimate in the region adjacent to the LVL and beneath the eastern Peloponnesus. The interpretation and implications of these observations are discussed in Sections 2.7.1 and 2.7.3.
NL images
41

The NL composite images show two features similar to those observed to the south: a low velocity layer (LVL) dipping to the ENE from ~30 to 70 km depth and a positive, sub-horizontal discontinuity with an average depth of 40 km (Figure 2.5).
The LVL dips at approximately 17° and has an apparent thickness that increases from 12 km to 20 km with increasing depth (Figure 2.5). The base of the LVL is continuous and clearly identified down to a depth of nearly 100 km, with a consistent location in both the d α / α and d β / β images. It exhibits an average d β / β of +7% marked by a progressive reduction in magnitude with increasing depth. Conversely, the top boundary of the LVL exhibits different signatures updip and downdip of ~50 km depth, leading us to separate it into two segments as was done for SL. The updip segment (i.e. above 50 km depth) comprises a thinner basal LVL (~12 km) overlain by several closely spaced discontinuities of limited lateral extent, including a positive discontinuity just above the basal LVL. This complex signature makes it difficult to estimate a meaningful d β / β magnitude. The down-dip segment, on the other hand, has a clear negative discontinuity that marks the top of the LVL on both the d α / α and d β / β images with a d β / β of approximately -7%. Thus, the downdip segment comprises a single 20 km-thick LVL that disappears between 70 and 90 km depth, in a more gradual fashion than its southern counterpart.
The NL composite images also show a sub-horizontal positive discontinuity that extends across the eastern part of the profile (Figure 2.5). It appears at a depth of 44 km near the LVL (X
NL
~ 100 km) and shallows gradually to a depth of 35 km at the eastern edge of NL. The positive discontinuity exhibits large lateral variations in d β / β , with three regions having magnitudes averaging +7% centered at X
NL
= 130, 230 and 280 km separated by two regions that lack signal at X
NL
= 180 km and X
NL
= 260 km. The interpretation and implications of these observations are discussed in Sections 2.7.2 and
2.7.3.
Along-strike variations in the subduction system beneath NL and SL are investigated by generating images with events from separate back-azimuthal bins, as shown in Figures
42

2.6 and 2.7. Each back-azimuthal bin contains a significant number of events (>14) that illuminate the subsurface from both up-dip and down-dip directions. The different backazimuthal contributions allow us to isolate scattered waves coming from the north of the profile (bin 1), from below the profile (bin 2), and from the south of the profile (bin 3).
In general, obliquely incident waves such as those used in bins 1 and 3 strike the imaging target at a perpendicular offset from the station array, and this offset becomes larger with increasing target depth [Rondenay et al., 2005; Rondenay et al., 2010]. From the results of Rondenay et al. [2005, their Figure 5], the maximum perpendicular offset of scattered waves contributing to the bin 1 and bin 3 images is approximately 20 km for targets at 30 km depth, and 40 km for targets at 80 km depth. Thus, bin-specific images provide a means of detecting along-strike variations in the imaging target and testing how well the geometric assumption of the 2-D GRT are met.
Figure 2.6 shows the SL bin-specific images for d α / α and d β / β . The thickness and position of the SL LVL are remarkably consistent in both images. We also note that in each bin-specific d β / β image, the LVL exhibits the same up-dip and down-dip attributes as those observed in the composite image (e.g. a thicker layer in the up-dip segment, see
Section 2.6.1 and Figure 2.3). Close examination of the d α / α bin-specific image reveals a secondary LVL beneath the primary LVL described in Section 2.6.1, which increases in amplitude from the NNW to the SSE (i.e., from bin 1 to bin 3). This feature may be an artifact due to cross-mode contamination, or may be related to along-strike variations in the subduction system, something we discuss further in Section 2.7.3.
Figure 2.7 shows the NL bin-specific images of d α / α and d β / β . All of these binspecific images show an up-dip segment of the LVL that is remarkably consistent with the composite image in Figure 2.5. The transition from updip to downdip segments is observed in each bin-specific d β / β image at a depth of ~50 km. However, the seismic signature of the down-dip segment varies significantly along strike with a “braided” texture in the bin 3 image compared to the bin 1 and bin 2 images. Along strike variations in the structure of the down-dip segment explain its weaker amplitude in the composite image and may be indicative of anisotropy within the down-dip segment, something that we discuss further in Section 2.7.2.
43

2.7 Discussion
In this section, we discuss the tectonic and geodynamic implications of our seismic images by focusing on three main topics: the southern segment, the northern segment, and the slab structure between these two domains. To aid the discussion, the composite images have been reproduced in Figure 2.8, along with profiles of local seismicity within
25 km of each line.
Nature of the subducted slab
The SL composite image shows an ~8 km thick LVL with a dip of ~17° extending into the upper mantle below southern Greece (Figure 2.3, Figure 2.8b). To the west of our profile, marine seismic data have shown that the oceanic crust beneath the abyssal plain of the Ionian Sea consists of 5-8 km of igneous crust [de Voogd et al., 1992; Finetti and
Del Ben, 2005] overlain by 4-6 km of sediments [Kopf et al., 2003; Finetti and Del Ben,
2005]. The similarity in thickness between the LVL and the igneous crust entering the trench strongly suggests that the LVL represents the continuation of this subducted crust at depth, provided that most of the sediments are scraped off before the crust reaches the mantle wedge. Thus, the subducted lithosphere imaged beneath the southern segment appears oceanic in nature. This interpretation is in agreement with recent receiver function studies that observe a 7 km-thick LVL with a ~17° dip angle at 60 km depth beneath the southern Peloponnesus [Gesret et al., 2010, 2011].
Extent and duration of subduction
Based on the total length of the imaged subducted crust and a trench location ~100 km to the west of the profile, we estimate that at least 250 km of oceanic lithosphere has subducted beneath the southern segment (Figure 2.3). If we assume that the subduction rate along the southern segment has remained constant at the current rate of ~35 mm/yr, the deepest subducted crust beneath southern Greece would have entered the trench ~7
Ma ago. On the other hand, if we consider a gradual increase in subduction rate from 5-
44

12 mm/yr during the middle Miocene to 35 mm/yr at present time, as suggested by
Royden and Papanikolaou [2011], the deepest subducted crust would have entered the trench ~10 Ma ago. These time constraints are consistent with the proposed age of transition from continental to oceanic subduction along the trench of the southern segment [Royden and Papanikolaou, 2011] and suggest this transition may currently be located near the termination of the SL subducted crust at depth (see Section 2.7.3 for further discussion of the ocean-continent transition).
Subduction interface
As first noted in Section 2.6.1, the seismic properties of the subduction interface vary abruptly at a depth of ~40 km, corresponding approximately to its intersection with the overriding Moho. The updip segment appears to comprise closely spaced discontinuities that run near, and parallel to, the subduction interface, producing a thicker LVL (~12 km) than in the lower segment (see region labeled SI in Fig 3). This internal layering may be interpreted in one of two ways. (1) The top boundary represents an inverted Moho resulting from serpentinization of an overriding sliver of mantle wedge, as was suggested to explain receiver functions in this region by Sodoudi et al. [2006]. (2) The top layer represents a sediment-filled subduction channel separating overlying forearc crust from the subducted oceanic crust. In both these cases, the lower, positive internal boundary would mark the top of the actual subducted crust as illustrated in Figure 2.3. The first option is unlikely given that gravity data [Royden, 1993; Tsokas and Hansen, 1997] and seismic velocities [Papazachos et al., 1997; Karagianni and Papazachos, 2007] show no evidence for mantle rocks extending in this region beneath the western Hellenides [see
Suckale et al., 2009, for further discussion]. The presence of subduction channel sediments is thus the preferred explanation, as it can explain both the observed negative discontinuity at the top of the subducted crust and the sudden change in LVL thickness at
~40 km depth. The imaged velocity contrast (>10 %) is consistent with a transition from forearc crust above to subducted sediments below, with a potential contribution from high pore-fluid pressure, as has been recently suggested for the Cascadia subduction zone
[Abers et al., 2009; Audet et al, 2009]. The sudden change in thickness can be explained by the accretion of some of these sediments underneath the overriding crust [see, e.g.,
45

Waschbusch and Beaumont, 1996; Ellis et al., 1999; Beaumont et al., 1999], a process that would be confined to depths shallower than that of the overriding Moho, and/or by a sudden drop in fluid pressures as the subduction interface transitions from overriding crust into the mantle wedge [see, e.g., Audet et al., 2009].
The downdip segment (50-80 km depth) is marked by a single, negative discontinuity yielding an LVL thickness of 8 to 10 km. The region above the downdip segment has been shown to have relatively low attenuation and heat flow [Hashida et al.,
1988; Fytikas et al., 1979]. Based on a comparison with thermal models, these geophysical observations suggest that a relatively cold (i.e. stagnant) portion of the mantle wedge overlies the segment of the subduction interface extending from 50 to 80 km [Wada and Wang, 2009; Syracuse et al., 2010].
Disappearance of the subducted crust at depth
As observed in Figure 2.3 (see also Figure A2 in Appendix A), the seismic signal marking the top of the subducted crust disappears abruptly at approximately 80 km depth, while a faint signal attributable to the subducted Moho remains visible to depths of ~125 km. The loss of signal from the top boundary is attributed to widespread dehydration of the upper crust’s basaltic layer as it undergoes eclogitization, as has been documented in numerous subduction zones using similar imaging techniques [Rondenay et al., 2008;
Abers et al., 2009; MacKenzie et al., 2010]. A rapid increase in attenuation, conductivity, and heat flow has also been observed above where the top of the subducted crust disappears [Hashida et al., 1988; Fytikas et al., 1979; Galanopoulos et al., 2005], which suggests this region sits within the circulating portion of the mantle wedge [Wada and Wang, 2009; Syracuse et al., 2010]. Thermal and petrologic modeling by Van Keken et al. [2011] also predicts the rapid dehydration of the Ionian slab’s upper crust at ~80 km depth while the lower gabbroic crust and upper slab mantle remain hydrated to significantly greater depths (>200 km). This latter prediction explains why the signal of the subducted Moho may persist (albeit faintly) down to at least 125 km depth.
Moho of the overriding plate
46

The positive, sub-horizontal discontinuity imaged in our SL profile at ~35 km depth is interpreted as the Moho of the overriding plate (Figure 2.3). It exhibits depth fluctuations that are broadly consistent with isostatic compensation of surface topography, in agreement with prior results obtained by Suckale et al. [2009]. The reader is referred to their paper for further discussion regarding the structure of the overriding plate.
Nature of the subducted slab
The NL composite image shows a ~20 km thick LVL dipping at ~17° within the upper mantle beneath northern Greece (Figure 2.5, Figure 2.8a). Marine seismic data have shown that the continental crust within the foreland consists of ~20 km of crystalline crust overlain by ~8 km of sediments [Finetti and Del Ben, 2005]. The LVL is thus interpreted as subducted continental crust, provided that ~8 km of sediments are being scraped from the subducting slab and accreted into the thrust belt of the overriding plate.
Our interpretation is in agreement with recent geodynamic models of continental subduction along the northern segment, which suggest that ~10 km of crustal material must be removed from the continental lithosphere in order to achieve a buoyancy that is sufficiently negative to match present-day subduction rates of 5-8 mm/yr [Royden and
Papanikolaou, 2011].
Extent and duration of subduction
Based on the imaged extent of the subducted continental crust (Figure 2.5) and the trench location ~50 km seaward of the NL line [Moretti and Royden, 1988; Finetti and Del Ben,
2005], we estimate that at least 220 km of continental lithosphere has subducted beneath the northern segment. Royden and Papanikolaou [2011] provide bounds on the subduction rates along northern Greece since Late Eocene to be 5-8 mm/yr from 0-5 Ma ago, 5-12 mm/yr from 5-20 Ma ago, and 25-35 mm/yr from 20-35 Ma ago. From these estimates, the deepest subducted continental crust would have entered the trench between
20-25 Ma ago. This analysis suggests continuous subduction of continental lithosphere
47

along the northern segment since the early Miocene, in general agreement with the record of subduction inferred from the nappes of the external Hellenides along NW Greece [van
Hinsbergen et al., 2005; Papanikolaou, 2009].
Subduction interface
Having made the case for continental subduction along the northern segment, we now look more closely at the structure of the plate interface. First, we examine the updip segment of the subducted crust, which comprises a ~12 km thick LVL overlain by several closely spaced discontinuities (Figure 2.5). Local tomographic models show that the region above the subducted continental crust observed in the d β / β image (Figure 2.5) has relatively low P- and S-velocities consistent with accreted sediments [Papazachos and
Nolet, 1997]. In this case, the subduction interface would not manifest itself as a negative discontinuity (as was interpreted for SL) but instead would be marked by a positive discontinuity resulting from the velocity contrast between the sediments accreted to the overriding plate (above) and the crystalline crust of the subducting continental slab
(below). Thus, we interpret the positive discontinuity above the ~12 km thick LVL
(upper solid black line in the UD segment of Figure 2.5) as the top of the subducted crystalline crust, an interpretation consistent with both the thickness and position of the crystalline crust observed in marine seismic data (see Section 2.7.2 and Figure 2.9). The closely spaced discontinuities above the top of the subducted crust appear to be of limited lateral extent and may reflect internal structure within a broad subduction channel in which subducted sediments, and perhaps some crystalline crust, are being detached from the slab and accreted to the overriding plate. Such broad subduction channels are predicted by thermal-mechanical models of retreating continental subduction boundaries
[e.g. Warren et al., 2008] and may favor the subduction of the entire crystalline crust into the mantle [De Franco et al., 2008].
The subduction interface of the downdip segment is clearly marked by a single negative discontinuity, which we attribute to the transition from mantle peridotites
(above) to crystalline continental crust (below). We note, however, that the velocity contrast observed is weaker than that expected for a peridotite to continental crust transition [Hacker et al., 2003; Hacker, 2008]. This is due, at least in part, to the along-
48

strike variations of the NL subducted crust discussed in Section 2.6.2 (Figure 2.7), which leads to the defocusing of the crust [Rondenay, 2005].
Disappearance of the subducted crust at depth
The seismic signal from the top of the subducted crust disappears at a depth of ~70 km, while the subducted Moho persists to a depth of nearly 100 km. There are several possible explanations for the loss of seismic signal from the subducted crust. First, it may be partially due to a loss in image resolution in this depth range. Indeed, multiples from the continental Moho appear to contaminate the d α / α and d β / β images at ~100 km and
~80 km, respectively (Figure 2.5). In addition, the dip of the subducting continental slab may increase rapidly within this depth range due its slower rate of rollback, as has been suggested by geodynamic models [Royden and Husson, 2006; Royden and Husson,
2009]. This could cause the lower part of the slab to dip too steeply to be resolved by the
2-D GRT [see, e.g., Rondenay, 2005; MacKenzie et al., 2010]. Second, the subducted crust may undergo progressive dehydration via eclogitization, as has been suggested for oceanic subduction zones [e.g. Rondenay et al., 2008]. Progressive dehydration of subducting continental lower crust has also been interpreted beneath the northern
Apennines based on receiver function images of an anisotropic, low S-velocity layer at
45-65 km depth [Piana Agostinetti et al., 2011]. Furthermore, thermal-mechanical models predict a similar conversion of continental lower, and possibly middle, crust to eclogite and/or coesite eclogite [Warren et al., 2008], which have seismic velocities similar to those of mantle peridotites and could thus explain the loss of seismic signal
[Hacker et al., 2004]. If slab dehydration is occurring along the downdip segment, it appears to do so without generating intermediate depth earthquakes (Figure 2.8a), unlike oceanic subduction settings where such earthquakes are commonly attributed to dehydration reactions [Hacker et al., 2003]. Third, the termination of the down-dip segment could be the result of a slab tear – something we discuss in Section 2.7.3.
Moho of the overriding plate
The positive, sub-horizontal discontinuity imaged in our NL profile at ~40 km depth is interpreted as the Moho of the overriding plate (Figure 2.5). It gradually shallows from
49

the eastern edge of the Hellenides to the northern Aegean, in qualitative agreement with previous studies [Marone et al., 2003; Sodoudi et al., 2006]. However, the high-density array used in this study reveals details of its structure that exceed the resolution of previous studies. The strength of its signal varies greatly as a function of location, with three distinct high-amplitude regions separated by two regions where no clear Moho signal exists. These variations may be related to the complex tectonic evolution of northern Greece, which has experienced the subduction and accretion of alternating continental fragments and oceanic basins [van Hinsbergen et al., 2005; Papanikolaou
2009].
Understanding the variations in slab geometry from the northern to the southern seismic profile can yield important insight into how a single subduction system accommodates a transition from continental to oceanic slab subduction. Figure 2.9 shows an along-strike view of the mantle wedge along the WHSZ constructed based on the interpreted structures from Figures 3 and 5, along with the foreland crustal structure inferred from marine seismic data [Finetti et al., 1991; Finetti and Del Ben, 2005]. The outlines depicted in Figure 2.9b exhibit two interesting features: 1) the oceanic subducted crust beneath SL appears to sit deeper than the continental subducted crust beneath NL, and 2) the portions of the slab beneath NL and SL have similar dips despite their differences in crustal thickness. In the following, we discuss the implications of these observations as they relate to differential slab retreat, the flexural strength of the slab, and the connection between the two imaged portions of the slab.
Differential slab retreat between the northern and southern portions of the slab
The geometrical attributes of the WHSZ, as depicted in Figure 2.9b, bear remarkable resemblance to the simple model of differential retreat shown in Figure 2.2. To explore this link further, quantitative estimates of differential retreat are derived from Figure 2.9b and compared to the offset of the WHSZ produced by the Cephalonia transform fault. We have to make some simplifying assumptions to do this analysis, namely that the trench of
50

the WHSZ was originally linear and that the slab maintained a constant dip at depths
<100 km after it segmented. These assumptions are reasonable given that (1) geologic and GPS observations suggest recent (latest Miocene to Pliocene time) segmentation of a previously continuous thrust belt that extended along the WHSZ [Reilinger et al., 2010;
Royden and Papanikolaou, 2011]; and (2) our images show that the two slab segments currently have similar dips despite their differences in crustal thickness (see Section
2.7.3). Using the subducted crust outlines in Figure 2.9b, we calculate an additional 70-
85 km of seaward retreat for the southern portion of the slab compared to the northern portion (see Appendix A for further details). This estimate of differential retreat is similar to the ~100 km offset produced by the Cephalonia transform fault. Thus our seismic images provide evidence in support of segmentation of the WHSZ through differential slab retreat.
Similar geometries for the northern and southern portions of the slab?
The dip of the slab at shallow depths (~30 to 100 km) is remarkably similar beneath the northern and southern portions of the WHSZ (Figure 2.9c), despite their differences in crustal thickness and buoyancy. Such lack of correlation between slab dip and buoyancy has also been observed in global compilations of oceanic subduction zone attributes [e.g.
Lallemand et al., 2005; Cruciani et al., 2005]. A variety of mechanisms have been invoked to explain shallow slab dip including flow associated with plate motions [Hager and O’Connell, 1978]. Lallemand et al. [2005] find that slab dip at shallow depths is globally correlated with the rates of absolute motion of the subducting plate, the trench, and the overriding plate. However, the WHSZ does not show a correlation of these factors with slab dip; the trench and overriding plate motions vary greatly between the northern and southern slab segments, while the shallow slab dip is uniform.
Other factors that may affect slab dip include buoyancy and flexural slab strength
(see subduction dynamics review by Becker and Faccenna [2009] for discussion of processes affecting slab dip and references). Royden and Husson [2006] developed a dynamically consistent subduction model that accounts for a wide variety of subduction processes with boundary conditions appropriate for the WHSZ (see discussion in Royden and Papanikolaou [2011]). Their models show that slab buoyancy strongly influences
51

trench retreat rates but does not greatly influence shallow slab dip. They show instead that several other factors may play a prominent role in determining slab dip at shallow depths. These include the flexural slab strength (elastic or viscous) and the thickness and density of the overriding plate and accretionary prism.
While a detailed analysis of these factors is beyond the scope of this paper, there is independent evidence that the slab’s flexural strength may be an important factor
[Moretti and Royden, 1988; Royden, 1993]. Gravity and flexure data yield an elastic thickness estimate for the southern segment of 70 km [Royden, 1993a] and a lower limit of 20 km for the northern segment, with larger values permissible provided that flexureinducing loads are correspondingly increased [Moretti and Royden, 1988]. Royden and
Husson [2006] show that slabs with elastic thicknesses greater than ~30 km have shallow dips (< ~20°) persisting to at least 70-80 km depth, similar to what is seen in our images
(Figure 2.9c). Moreover, variations in elastic thicknesses over 30 km do not greatly affect the retreat rate, which is instead dominated by slab buoyancy and other factors cited above (see Figure 2.10 in Royden and Husson, 2006). Thus, we propose that the two portions of the WHSZ slab have different retreat rates resulting primarily from their different buoyancies, while they exhibit similar low dips at shallow depths because their flexural rigidities are similar and large (i.e., ~70 km). If this inference is correct, it supports the assumption of constant slab dip over time.
Transition from the northern to the southern portion of the slab
We now turn out attention to the possible links between the northern and southern segments. First we consider the case of a smooth, continuous ramp between the two segments. The trench-parallel distance between the NL and SL profiles is ~270 km and the vertical offset between the Moho of the subducted crusts within the upper mantle is
~10 km (Figure 2.9). Accordingly, a smooth ramp between the Moho of the NL and SL subducted crusts would require a slope of only ~2° dipping to the SSE. Such a model is consistent with the following constraints from previous investigations: 1) the presence of a continuous, high-Q slab between NL and SL [Konstantinou and Melis, 2008]; 2) the continuity of the high-velocity anomaly in local tomographic models beneath central
Greece [Papazachos and Nolet, 1997]; 3) the continuity of the slab Moho in the S
52

receiver functions of Sodoudi et al. [2006] (albeit at much greater depth beneath northern
Greece than our NL image); and 4) the slab depth of ~75 km beneath the western Gulf of
Corinth inferred by Zelt et al. [2005].
Second, we explore the possibility that the northern and southern portions of the slab are separated by a tear. We consider the two possible tear locations that have been proposed in the literature (see Section 2.1): along the Cephalonia transform or along the ocean-continent transition. The Cephalonia transform is a possible candidate because its extrapolation into northern Greece intersects the downdip segment of the NL subducted crust at a depth of ~70 km, very close to where the low-velocity signal of the subducted crust disappears (70-100 km depth; compare Figure 2.8 and 2.10). A tear along the ocean-continent transition is more difficult to reconcile with our seismic images. Indeed, while this structure intersects the SL subducted crust at ~80 km, a signal attributable to the subducted Moho is observed well beyond that depth, which would argue against tearing of the slab in this region.
To summarize, the simplest (and arguably best supported) form of transition between the northern and southern portions of the slab is a smooth ramp – at least for the top ~100 km of the system. Nevertheless, a slab tear along the Cephalonia transform fault cannot be ruled out, though the disappearance of the subducted crust at ~70-100 km depth can be equally explained by eclogitization reactions (see Section 2.7.2). Future studies using additional imaging techniques (e.g. receiver functions, local tomography, etc.) and data from trench/arc-parallel receiver arrays [e.g., ‘THALES WAS RIGHT’,
Sachpazi et al., in prep.] should provide a more detailed picture of the connection between the two segments.
2.8 Concluding remarks
In this study, a 2-D GRT migration algorithm was applied to teleseismic data collected by two arrays of broadband seismographs to generate high-resolution seismic images across the northern and southern segments of the western Hellenic subduction zone. These images were used to draw the following conclusions regarding the structure and dynamics of a system that involves a transition from oceanic to continental subduction:
53

1) Subducted crust is observed beneath both the northern and southern segments of the
WHSZ (Figures 2.3 and 2.5, respectively) and both segments have a similar strike of approximately N30W +/- 10° and dip of approximately 17°.
2) The ~8 km thick subducted crust beneath the southern segment is interpreted as the oceanic crust of the Ionian (Sea) lithosphere, provided that most of the overlying sediments are scraped off before the crust reaches the mantle wedge.
3) The ~20 km thick subducted crust beneath the northern segment is interpreted as the crystalline continental crust of the Apulian lithosphere, which has a thickness and position consistent with constraints on the foreland continental crust from marine seismic data, provided that ~10 km of overlying sediments have been accreted to the overriding plate.
4) The offset between the SL and NL subducted crusts implies 70-85 km of additional slab retreat along the southern segment as compared to the northern segment, which is in general agreement with the offset of ~100 km along the Cephalonia transform fault.
5) A smooth ramp between the two slab segments is the simplest model for their connection in the top ~100 km of the system, as it only requires an arc-parallel slope of approximately 2°. Nevertheless, based on our 2-D profiles we cannot rule out a slab tear along the Cephalonia transform fault or (for depths >100 km) the ocean-continent transition.
54

2.9 Tables
Layer Z (km) € a (km/s) b (km/s) r (kg/m^3)
1
2
20
40
6.2
6.8
3.6
3.8
2.6
2.8
3 60 7.6 4.2 3.0
4 - 8.0 4.5 3.2
Table 2.1: Optimal 1D background velocity model for SL imaging
Layer Z (km) € a (km/s) b (km/s) r (kg/m^3)
1
2
3
20
40
60
6.0
6.8
7.7
3.4
3.8
4.3
2.6
2.8
3.0
4 - 8.0 4.5 3.2
Table 2.2: Optimal 1D background velocity model for NL imaging
€ Z is the depth from the free surface to the bottom of the layer, a is P-wave velocity, b is
S-wave velocity, and r is density.
55

2.10 Figures
Figure 2.1: Map of the study area showing distribution of seismic stations and teleseismic events used in this study. a) Map of the Western Hellenic subduction zone. Yellow lines show the active, southern segment boundaries: thrust front of the Mediterranean ridge
(dashed) and trench (solid). Dashed orange line shows thrust front of northern segment.
Bright red line denotes the Cephalonia transform fault (CTF) and dark red dotted line marks the Apulian Escarpment (i.e. the oceanic-continental transition). Dashed gray line outlines the North Aegean trough (NAT) and the central Hellenic shear zone. GPS velocities (magenta arrows) relative to Eurasia are from McClusky et al. [2000]. Seismic stations deployed across southern and northern Greece are denoted by white squares and circles, respectively. The surface projection of our two 2D-GRT profiles are denoted by thick black lines (SL and NL). Additional acronyms: II, Ionian Islands (Cephalonia in middle); GC, Gulf of Corinth; GE, Gulf of Evia; Pel, Peloponnesus. b) Distribution of events used in teleseismic migration of SL (red circles) and NL (white circles). Map is centered between the two arrays with white concentric circles denoting 10° increments in epicentral distance (innermost circle at 30°).
56

Northern Segment
Southern Segment
Figure 2.2: Schematic diagram showing the hypothesized geometry of a segmented slab in the Hellenic subduction zone, modified from Royden and Papanikolaou [2011]. There may be either a tear, as depicted, or a smooth ramp between the Adriatic and Ionian slabs
(in the region containing horizontal dashed lines).
57

Figure 2.3: Composite images for SL showing d α / α (a, c) and d β / β (b, d) perturbations obtained by 2-D GRT inversion of Pp d p (a, c) and P d s, Pp d s, Ps d s| v
, Ps d s| h
(see explanation of scattering modes in Section 2.3.2). Panels (c, d) show the same images as (a, b) but also include our structural interpretations (black lines): the updip (UD) and downdip
(DD) segments of the subducted crust, the location of the thick subduction channel interface (SI), and the Moho of the overriding plate (SOM). Black triangles denote the station locations along SL with 10:1 vertical exaggeration in elevation (i.e. negative depth). Figure 2.8b contains a larger version of panel (d) that includes earthquake hypocenters.
58

Figure 2.4: Composite images for SL showing d β / β obtained by 2-D GRT inversion of free-surface-reflected phases only (i.e. Pp d s, Ps d s| v
, and Ps d s| h
). Panel b) shows the same image as a) but also includes the structural interpretations from Figure 2.3c,d plus the extension of the subducted Moho to ~125 km depth (dotted line). Black triangles denote the station locations along SL with 10:1 vertical exaggeration in elevation (i.e. negative depth).
59

Figure 2.5: Composite images for NL showing d α / α (a, c) and d β / β (b, d) perturbations obtained by 2-D GRT inversion of Pp d p (a, c) and P d s, Pp d s, Ps d s| v
, Ps d s| h
(see explanation of scattering modes in Section 2.3.2). Panels c-d) show the same images as a-b) but also include our structural interpretations (black lines): the updip (UD) and downdip (DD) segments of the subducted crust, and the Moho of the overriding plate (NOM). Black triangles denote the station locations along NL with 10:1 vertical exaggeration in elevation (i.e. negative depth). Figure 2.8a contains a larger version of panel (d) that includes earthquake hypocenters
60

Figure 2.6: SL images of d α / α (a, c, e) and d β / β (b, d, f) perturbations obtained for subsets of the data divided into back-azimuthal bins: bin 1 uses events from NNW (a, b), bin 2 uses events from WSW and ENE (c, d), and bin 3 uses events from SSE (e, f).
Interpreted structural boundaries (black lines) are the same as in Figure 2.3. The color scale varies between panels to take into account variations in the number and the quality of events used to generate each image, something that can cause significant amplitude variations when illumination is incomplete (see Rondenay et al., 2005).
61

Figure 2.7: NL images of d α / α (a, c, e) and d β / β (b, d, f) perturbations obtained for subsets of the data divided into back-azimuthal bins: bin 1 uses events from NNW (a, b), bin 2 uses events from WSW and ENE (c, d), and bin 3 uses events from SSE (e, f).
Interpreted structural boundaries (black lines) are the same as in Figure 2.5. Note that the velocity perturbations are amplified due to the limited illumination afforded by the subsets of data, but that the structures are still adequately resolved (see, Rondenay et al.,
2005 for a complete discussion of these effects).
62

Figure 2.8: Comparison between the composite d β / β images for (a) NL and (b) SL (same as Figure 2.5d and Figure 2.3d, respectively), with local seismicity within 25 km of each line. Local event hypocenters (black dots) are from the National Observatory of Athens
Bulletin. We plot events that occurred from June 2002 to January 2012 and had the following specifications: (1) M
L
>= 3, (2) located using time picks from at least 8 stations, and (3) located by groups of stations with an azimuthal separation < 150°. Dot size is proportional to magnitude, with the smallest dots corresponding to M
L
=3 and the largest dots to M
L
=5.1. The black arrow above a) shows where the extension of the
Cephalonia transform fault (CTF) intersects NL, as depicted in Figure 2.10.
63

Figure 2.9: Comparison between the northern and southern slab geometry a) Map showing the common coordinate system (X
ST
) used to compare the NL and SL slab geometries, which is oriented along the optimal projection lines of NL and SL (N60E, see black arrow) and has its origin at the intersection of SL and the southern trench (solid black line at X
ST
=0). b) Relative positions of the NL (gold) and SL (purple) subducted crusts and overriding Mohos imaged in this study (solid), with connection to the approximate location of the foreland crust inferred from surface seismic studies (beneath the unconnected black squares in a)) [Finetti and Del Ben, 2005]. Unfilled subducted crust regions show interpolation between two imaged structures. Also shown are the approximate trench locations along SL (purple T) and NL (yellow T). c) Same as previous panel, but with a landward horizontal displacement of 70 km (i.e. to the right, as shown by the translation of the purple T) applied to the SL slab and overriding crust.
Note that the X-axis in c) does not have geographical significance as we have not accounted for the absolute motion of the northern segment.
64

Figure 2.10: Map of Greece showing the location of deep seismicity (>50 km) from the
International Seismological Centre [2001] catalog with reported depth errors less than 10 km. We map events that occurred between January 1964 to May 2006, using the following color code for hypocenter depth: 50 km – 100 km (yellow), 100 km – 150 km
(red), and >150 km (purple). The two potential tear locations discussed in the text are also shown: (1) the approximate location of the Apulian escarpment (AE, ocean-continent transition) beneath southern Greece, as estimated from the paleogeographic reconstruction of Royden and Papanikolaou [2011], is denoted by a dashed brown line, and (2) the Cephalonia transform fault (CTF) beneath northern Greece is denoted by a dashed red line. The surface projection of our two 2D-GRT profiles are denoted by thick black lines (SL and NL).
65

66

Abstract
Obtaining a detailed picture of the three-dimensional lithospheric structure along the western Hellenic subduction zone (WHSZ) is critical for discriminating between different geodynamic models of the region. While our high-resolution seismic images from
Chapter 2 provide evidence for differential retreat between the southern, oceanic and northern, continental portions of the slab, the connection between these two domains remains poorly understood. Here, a receiver function (RF) imaging algorithm that handles 3-D dipping boundaries is applied to broadband seismic data from stations distributed across mainland Greece. The resulting 3-D common conversion point image, along with RF results from individual stations in central Greece, are used to map the location of the slab Moho, overriding Moho, and basement top across the WHSZ. We find a continuous slab extending from northern to southern Greece that is consistent with the ~70 km of differential slab retreat inferred in Chapter 2. The Moho of the overriding lithosphere is characterized by a large-scale decrease in Moho depth from the northern
Hellenides (35-40 km) to the Aegean Sea (25-30 km) and small-scale variations in Moho depth (+/- ~4 km) along a roughly N-S corridor between NW Greece and the Aegean Sea.
The overriding Moho also displays several additional characteristics: (1) it is asymmetric perpendicular to the trench, not “U” shaped as has been found in previous studies, (2) the peak topography of the Hellenides lies above the tip of the mantle wedge, a region that lacks a well developed overriding Moho, (3) the first emergence of a clear overriding
Moho signal occurs only after the slab has subducted well into the mantle (slab Moho
67

depth of ~70 km). In addition, we also find that the observed small-scale variations in overriding Moho depth (+/- ~4 km) cannot be explained by a model of local Airy-
Heiskanen isostasy. The top of the crystalline basement is observed across much of the
WHSZ at an average depth of ~8 km. Its start (i.e. crustal backstop) along northern
Greece is offset landward by ~180 km relative to the backstop along southern Greece, indicating that differential basement retreat far exceeds differential slab retreat (i.e. 180 km vs. 70 km). The basement top also forms several asymmetric valley-like patterns beneath southern Greece that appear to coincide with regions of active extension in the overriding crust. The 3-D lithosphere geometry found in this study provides evidence in support of a geodynamic model for the WHSZ in which slab buoyancy controls the rates of slab retreat and overriding lithosphere deformation.
3.1 Introduction
The western Hellenic subduction zone (WHSZ) provides an ideal setting for studying the structure of a retreating convergent boundary as it undergoes a transition in slab composition. Currently an oceanic slab is entering the trench along southern Greece, with a convergence (equivalent to retreat) rate of ~35 mm/yr, while a continental slab is entering the trench along northern Greece, with a retreat rate of ~5-8 mm/yr (Figure 3.1;
McClusky et al. [2000]; Royden and Papanikolaou [2011]). Tectonic models generally agree that the rapid retreat rate along southern Greece are driven by the subduction of dense, oceanic lithosphere, as has been confirmed by several recent seismic imaging studies [Bohnhoff et al., 2001; Li et al., 2003; Gesret et al., 2010; Pearce et al., 2012].
However, tectonic models dramatically diverge in their view of the tectonic setting along northern Greece, with some studies invoking a continental collision boundary that is separated from the subduction system along the south (e.g. Taymaz et al. [1991]), possibly by a slab tear (e.g. Wortel and Spakman [2000], Govers and Wortel [2005]), while others argue that a retreating subduction model can explain both the southern and northern portions of the WHSZ [Royden and Papanikolaou, 2011].
Several seismic imaging studies have sought to constrain the lithospheric structure along the WHSZ, yet there remains a great deal of uncertainty regarding its geometry.
For example, there have been three different slab models proposed for the WHSZ. Gesret et al. [2011] infer a shallowing of the slab north of the Peloponnesus, with a possible transition to a separate, continental slab segment, while Sodoudi et al. [2006] infer an
68

abrupt deepening of the slab north of the Gulf of Corinth, with depths reaching greater than 200 km beneath northern Greece. Our seismic imaging results from Chapter 2 show a shallow dipping slab at ~40 to 70 km depth beneath northern Greece, with the simplest connection to the south involving a gently dipping ramp (see Figure 2.9), a slab geometry that is also consistent with several other geophysical constraints (e.g. local tomography, seismic attenuation; see additional discussion in Section 2.7.3).
There are also two different models that have been proposed for the geometry of the overriding Moho (seismically defined crust-mantle boundary of the overriding lithosphere). On the one end, there are “symmetric” models in which the overriding
Moho is “U” shaped in a direction normal to the trench, with its deepest part lying roughly below the peak topography of the Hellenides (e.g. Tsokas and Hansen [1997];
Sodoudi et al. [2006]). Conversely, we identify an asymmetric overriding Moho geometry in our 2-D GRT images, with its deepest part emerging from near the slab
(Figure 2.4 and 2.5). These different models for the geometry of the slab and overriding
Moho make it difficult to distinguish between the contrasting tectonic models for the northern portion of the WHSZ and its transition to the retreating subduction system along southern Greece.
The goal of this study is to better constrain the three-dimensional lithospheric structure of the WHSZ using a Ps receiver function imaging method. The method developed here, termed the 3-D RF method, is capable of accurately locating threedimensional, dipping discontinuities in S-wave velocity. The method is first tested using synthetic data, then applied to broadband data from a dense network of distributed seismic stations across the WHSZ. The resulting 3-D RF model is used to identify the location of the slab Moho, overriding Moho, and top of the crystalline basement along the
WHSZ. Our map of slab Moho depth shows a continuous slab that extends across northern and southern Greece, implying that the shallower and deeper slabs found in previous studies may be additional slab segments. Our map of the overriding Moho depth shows that the overriding lithosphere is asymmetric normal to the trench, with a clearly defined overriding Moho signal only appearing after the slab has reached ~70 km depth. The top of the crystalline basement is observed as a ubiquitous feature across the
WHSZ, suggesting the bulk of the overriding crust is comprised of at least two layers:
69

low-velocity sediments overlying high velocity crystalline crust. As we will see, the slab and overriding Moho geometry observed here, and their relationship to surface topography, provide numerous lines of evidence in support of a retreating subduction model for both the southern and northern portions of the WHSZ.
3.2 Method
The imaging method used in this study analyzes P-to-S (henceforth Ps) converted waves in the coda of teleseismic P waves to identify sharp changes in S-wave velocity. It assumes Ps conversions are generated at planar discontinuities embedded within a background velocity model as illustrated in Figure 3.2. Under these assumptions, the position and magnitude of the discontinuity can be calculated from the delay time and relative amplitude between Ps and P, respectively. This basic idea was first exploited as an imaging method by Vinnik [1977] and Langston [1979] and has since been extensively used to characterize crustal and upper mantle discontinuities [e.g. Bostock,
1998; Rondenay et al., 2000]. Such methods are often referred to as receiver function
(RF) imaging methods [Langston, 1979] and generally consist of a preprocessing step and an imaging step, which we briefly describe in the following two sections.
Preprocessing is done to isolate the portion of the recorded wavefield generated by the Swave velocity structure beneath the station. It involves first partitioning the recorded wavefield into estimates of the incident and scattered wavefields, then normalizing the later by the former to produce so-called receiver functions [Langston, 1979]. The preprocessing is important as it determines the fidelity of the receiver functions and thus their coherence when stacked during the imaging step [Rondenay, 2009]. Here, a multichannel preprocessing method is used to compute the RFs. It is similar to the original approach of Bostock and Rondenay [1999], but incorporates a number of modifications, including a new method for accurately aligning the RFs relative to the incident wavefield.
70

The multichannel preprocessing method used here is comprised of the following steps [see Rondenay et al., 2005 for details]: (1) transform the recorded wavefield from
N-E-Z to upgoing P-SV-SH using the free-surface transfer matrix [Kennett, 1991], such that the incident wavefield, I, is concentrated in the P wavefield; (2) apply multichannel cross-correlation [VanDecar and Crosson, 1990] to align the P wavefield with respect to the incident wave; (3) perform principal component analysis on the P wavefield and extract the first component as an estimate of the incident wavefield; (4) deconvolve the incident wavefield from the SV and SH components to produce SV and SH RFs, respectively. This procedure has been further improved by including the longest possible incident wavefield signals to capture source side scattering [Dueker and Sheehan, 1997], by iterating over steps (2) to (4) to improve the RF alignment, and by using independent damping parameters for each station component during step (4) (see Section 2.3.1 for additional details).
An additional improvement has been developed to ensure accurate alignment of the incident wavefield, particularly for station/event pairs with low signal-to-noise ratios.
Errors in alignment time arise from noise within the P time window used in the multichannel cross-correlation [VanDecar and Crosson, 1990]. Fortunately, we can evaluate such errors following the multichannel preprocessing by examining the deconvolved P wavefield, P D (t),
P D (t) = FFT − 1
P(
)I(
I(
)I(
) *
)
+
*
[1] where P( ω ) and I( ω ) are the Fourier transforms of the P wavefield and incident
ε
is the damping parameter, and FFT -1 is the inverse Fourier transform. If the P wavefield is accurately aligned (and scattering is weak), P D (t) will be dominated by a single (band-limited) delta function centered at a delay time of zero (or some other arbitrary constant) for all stations. Our original approach to improve the alignment was to test the multichannel cross-correlation on different P time windows until minimal time differences are
71

observed in the P D (t) of each station. However, this approach is time consuming and is often unable to accurately align the P wavefields recorded at all stations, particularly when the signal-to-noise ratio is low.
As an alternative, we have developed an automated procedure that yields an optimal P wavefield alignment. It solves for optimal alignment time corrections by iteratively applying multichannel cross-correlation to P D (t) (see Appendix B for details).
The result is an aligned P wavefield with time errors smaller than the sampling rate (0.1 sec in this study), which is critical given the small differential times expected from dipping interfaces [Langston, 1977; Cassidy, 1992]. It also stabilizes the RF amplitudes and allows us to expand data coverage through the inclusion of events with relatively low signal-to-noise ratio.
Next, the RFs from different events are combined to construct an image of subsurface discontinuities in S-wave velocity. This is done by first mapping the RF amplitudes to their subsurface conversion points, and then stacking them to form an image. The stacking is done in two different ways depending on the station density. In the following, we briefly describe each of these steps.
First, we describe our approach for mapping the RF amplitudes to their subsurface conversion points. This is done using analytical expressions that relate Ps-P delay time to subsurface conversion point for a single station recording multiple teleseismic events as shown in Figure 3.2. To do this, assumptions must be made regarding the incident wave, the background velocities, and the discontinuity geometry. The incident wave is assumed to be planar with a known slowness vector such that we may define a generic receiver function, (SV- or SH-) RF,
RF = A m , n
(
t m
,
n
, s n
)
[2].
A m,n
is a matrix of receiver function amplitudes with rows representing M discrete Ps-P delay times,
t m
, and columns denoting N events (i.e. incident plane waves), each having
72

a particular backazimuth, φ n
, and slowness magnitude, | s n
|. For simplicity, we assume a homogeneous background model and allow the discontinuity to have variable strike and dip.
Under these assumptions, analytical expressions relating the Ps-P delay time (
Δ t, subscript m omitted from here on) to the Ps conversion point (e.g. C
1
in Figure 3.2) are derived as follows. First, we use the assumed strike of the discontinuity to define a horizontal coordinate system (i.e. parallel to the Earth’s surface), with its X and Y axes pointing in the discontinuity’s down-dip and strike directions, respectively (Figure 3.2).
The incident slowness vector is then decomposed into its X and Y components
[ p x
P p y
P ] = s * [sin(
Y ) cos(
Y )]
[3] where | s | is the incident slowness magnitude of a given event (event index n omitted from here on);
φ
Y is the backazimuth angle measured clockwise (in radians) from the strike of
the discontinuity, as opposed to
φ
, which is measured clockwise from geographic North
(Figure 3.2); and p x
P and p y
P are the incident P-wave’s downdip and along-strike slowness components, respectively. This choice of coordinate system has a couple advantages, one can model different discontinuity strikes by simply rotating the X and Y axes so they are aligned with each discontinuity’s dip-direction and strike (i.e. changing the mapping of φ to φ
Y ), and it also simplifies the derivation of analytical expressions relating
Δ t and C , as described next.
The expression relating Ps-P delay time ( Δ t) and conversion depth (C z
) is derived by solving for the difference in arrival time between the incident P-wave and the Ps conversion (e.g. blue (P) and blue-to-red (P-to-Ps) ray segments from the wavefront labeled t=t
0
to the black triangle in Figure 3.2). We do this calculation separately in the
X-Z and Y-Z planes, and then combine the results. In the Y-Z plane, the discontinuity has no dip (it is horizontal), and so, Snell’s law dictates that the Y-component of converted wave’s slowness is equal to the Y-component of the incident wave’s slowness
(i.e. p y
P = p y
S ). This is not the case in the X-Z plane, as Snell’s law dictates that the slowness component parallel to the discontinuity’s dip-direction is constant (i.e. p xd
P =
73

p xd
S ), and so, incident rays that travel down the discontinuity dip (+X
D
direction) are bent less upon conversion to Ps, compared to incident rays that travel up the discontinuity dip
(-X
D
), as illustrated by the two example events in Figure 3.2. It can be shown that solving the geometric problem described above leads to the following relationship between Δ t and C z
:
t = C z
(
S
P )
[4] where Η
S is an effective S-wave vertical slowness,
S =
( ) − 2
p x
S p x
P
S
( )
2
[5];
η
S is the Ps converted wave’s vertical (i.e. Z) slowness,
S
=
− 2
( ) 2
( )
2
[6];
η
P is the P-wave vertical (i.e. Z) slowness,
P =
− 2
( )
2
( )
2
[7];
α and β are the P- and S-wave velocities of the half-space. The Ps converted wave’s X-
x
S , is calculated by (1) rotating the incident wave’s slowness components from the (X Y Z) coordinate system (i.e. [p x
P p y
P
η
P ]) to the (X
D
Y Z
D
) coordinate system (i.e. [p xd
P p y
P
η d
P ]; see Figure 3.2), (2) equating the X
D
slowness component of the incident and converted waves, as dictated by Snell’s law (p xd
P = p xd
S ),
(3) calculating the Z
D
slowness component for the converted wave ( η d
S ) by substituting
74

p xd
S for p x
S in equation [6], and finally (4) rotating the converted wave’s slowness components from the (X
D
Y Z
D
) coordinate system (i.e. [p xd
S p y
P
η d
S ]) to the (X Y Z) coordinate system (i.e. [p x
S p y
P
η
S ]). Note that if the discontinuity dip is zero, then p x
S = p x
P , which means that Η
S = η
S , and so the term in parentheses in equation [3] reduces to the difference in vertical slownesses between the Ps conversion and incident P wave ( η
S -
η
P ), the widely used Δ t versus C z
relationship for horizontal discontinuities (equation [8] in Rondenay [2009]).
Finally, the downdip (C x
) and along-strike (C y
) components of the conversion point relative to the station are computed from the slope of the converted wave’s raypaths in the X and Y directions, respectively:
C x
=
⎛ p
⎜
⎝
η
S
S x
⎞
⎟
⎠
C z [8],
C y
=
⎛ p y
I
⎜
⎝
S
⎞
⎟
⎠
C z [9].
€
There are a couple points worth mentioning with regard to the generality of these
€ background model having any number of layers by computing the necessary incident and converted wave slowness vectors for each layer, and then integrating the increments of
Δ t, C x
, and C y
for each layer as a function of C z
. Note that the total number of converted wave slownesses one must compute is equal to the cumulative sum of the number of layers being modeled, including the half-space (e.g. a layer over a half-space model requires a total of three unique converted wave slowness vectors, [p x
S p y
P
η
S ], but only two unique values of p x
S ). Second, the assumption of planar discontinuities is only made locally about the conversion point such that some degree of velocity heterogeneity beneath a station can also be modeled, if known a priori. In this study, we take advantage of point one above and use a background model having a single, horizontal layer over a
75

half-space, as depicted in Figure 3.2, leaving the latter point for future method development.
Next, we describe how we use equations [2-9] to form RF images. This is done in two different ways depending on the station distribution. When station density is sparse
(i.e. > ~30 km), the conversion points from different stations do not significantly overlap at the depths of interest (C
Z
>~30-100 km), so RF images are formed for each individual station [Rondenay, 2009]. The RF amplitudes for each event are mapped to their conversion depth (C
Z
) using [3] to account for their different slownesses. Then the RFs are stacked to form a single RF trace and used to interpret the S-wave discontinuity structure. However, it is also necessary to inspect the individual RFs prior to stacking to check for signs of anisotropy or velocity heterogeneity [Rondenay, 2009]. In this case, the horizontal conversion points may help distinguish between these sources of complexity as one can compare them with the location of known geologic features (see examples of individual station RF images in Figures 3.12-3.17).
When densely spaced stations are available, RF amplitudes are first mapped to their conversion points using [3-9]. Then the RF amplitudes that fall within a prescribed subsurface volume are stacked in an approach referred to as Common-Conversion Point
(CCP) imaging [Dueker and Sheehan, 1997]. Here we employ such an approach using 3-
D CCP bins to account for incident waves with any slowness vector and Ps conversions from 3-D dipping discontinuities. Thus we obtain a 3-D RF model that can be sliced in different directions to form 2-D images that are then used to interpret the S-wave discontinuity structure (see example of 2-D image in Figure 3.4). It is quite straightforward to use different discontinuity dips for constructing different regions of the
3-D RF model (e.g. horizontal at Moho depths and dipping at subducting slab depths) because all conversion points are mapped into a single coordinate system that is paraellel to the assumed discontinuity strike (X Y Z). With these imaging methods in hand, we now describe the data used to construct our 3-D RF model of the WHSZ.
3.3 Data
76

The data used in this study consist of teleseismic P waveforms recorded at permanent and temporary broadband seismic stations along the WHSZ (Figure 3.1). Station data are collected from three different sources. The first dataset was derived from 40 broadband seismometers deployed in two arrays across northern and southern Greece as part of the
Multidisciplinary Experiment for Dynamic Understanding of Subduction under the
Aegean (MEDUSA) project. Each array contained 40 stations laid out in a fan-shape roughly perpendicular to the WHSZ and collected data for ~1.5 years (Figure 3.1; see section 2.4 for additional details). The second dataset is comprised of 15 broadband stations along the WHSZ that are publicly available through the WebDC data portal
( http://www.webdc.eu
), while the third dataset is comprised of four stations obtained from the National Observatory of Athens (i.e. VLS, RLS, JAN, EVR). Thus, we obtain high-density station coverage along northern and southern Greece primarily from the first dataset, and several sparsely distributed stations in central Greece from the other two datasets.
Waveform data from each station were collected for teleseismic events that occurred during the deployment of the MEDUSA arrays (Figure 3.1b). We selected events with epicentral distances from 30-deg to 90-deg and magnitudes greater than 5.5 based on earthquake catalog information from the Preliminary Determination of
Epicenters (PDE) Bulletin ( ftp://hazards.cr.usgs.gov/pde/ ). Our selection criteria yielded numerous events from eastern Asia but relatively few events from other regions, such as
Africa or the Atlantic spreading ridge (Figure 3.1b). For each event, the back-azimuth
(
φ
) and incident slowness magnitude (| s |) at each station were computed with the TauP
Toolkit, [Crotwell et al., 1999] using IASPEI91 as the reference Earth model [Kennett and Engdahl, 1991].
3.4 Model domain
A single model domain is used to generate and display all RF imaging results. It is oriented perpendicular (N60E, X) and parallel (N30W, Y) to the WHSZ based on the slab dip direction determined in Section 2.5 (Figure 3.3). Station latitudes and longitudes are mapped into the model domain using an oblique Mercator projection. The model domain
77

is discretized into three-dimensional Common Conversion Point (CCP) bins with a height
(Z) of 2 km and a width (X and Y) of 20 km, the approximate width of the first Fresnel zone for a PdS conversion at ~40 km. The amplitude of each CCP bin is obtained by averaging the RF amplitudes whose conversion points fall within it. CCP bins are overlapped in the X and Y directions by 50% (i.e. 10 km), and then the entire domain is filtered using a three-dimensional, 3-point Gaussian (i.e. triangular) function with unit standard deviation. This last step reinforces image smoothness within the first Fresnel zone.
3.5 Results
In this section, we present results from our 3-D receiver function model for four different regions along the WHSZ. Three regions contain sufficiently dense station coverage to produce robust CCP images: two regions perpendicular to the WHSZ along southern and northern Greece, and one region parallel to the WHSZ along the back-arc (Figure 3.3).
The fourth region in central Greece only contains sparsely distributed stations, so we present both the individual RF traces and CCP images beneath each station.
Figure 3.4 shows the SV and SH images for three cross-sections across southern Greece.
The SV RF images contain three prominent features: (1) a pair of negative and positive peaks dipping in the +X direction (i.e. N60°E) at ~17°, indicative of a low-velocity layer
(LVL), (2) a sub-horizontal positive discontinuity at an average depth of ~32 km, termed the deeper sub-horizontal discontinuity, and (3) a sub-horizontal positive discontinuity at an average depth of ~7 km, termed the shallower sub-horizontal discontinuity (Figure
3.4). The first two features are similar to those observed in previous receiver function
[Sodoudi et al., 2006; Gesret et al., 2010; Gesret et al., 2011] and teleseismic migration
[Suckale et al., 2009; Pearce et al., 2012] images along southern Greece, while the last feature is broadly consistent with the reflection profiles of Makris [1978]. However, our
78

dense station coverage and the 3-D nature of our RF imaging method allow us to identify variations in these features not resolvable in previous studies.
Dipping Low-velocity Layer
The dipping LVL is observed between 45 km and 85 km depth on all three RF profiles, though it is clearest on the SL2 profile, as this profile has the highest conversion point density (Figure 3.4). Its basal positive discontinuity is continuous and planar from X=80 km to X=250 km on the SL2 and SL3 profiles, but is quite irregular on the SL1 profile with some evidence of two overlapping positive discontinuities. The negative discontinuity defining the top of the LVL appears to be disrupted at X=~150 km, consistent with the abrupt change in LVL thickness observed by Pearce et al. [2012] (see additional discussion in Section 2.6.1).
The downdip portion of the LVL (X>150 km) is thicker than the one imaged by
Pearce et al. [2012] (12-15 km in Figure 3.4 vs. 8-10 km in Figure 2.3). Both images are formed using the same frequency band (~0.05 Hz < f < 0.5 Hz), so this discrepancy is most likely due to the higher resolution of the 2-D GRT method [Rondenay et al., 2005;
Rondenay, 2009]. When we include frequencies up to 1.5 Hz, the thickness of the LVL drops to ~9 km (Figure 3.5), which is in better agreement with Pearce et al. [2012], as well as the RF analysis of Gesret et al. [2010, 2011]. However, including higher frequencies also reduces the along-dip coherence of the LVL, giving way to more discrete segments that resemble stair steps (see SL2 profile in Figure 3.5), an effect possibly related to anisotropy and/or scattering from local heterogeneities. The images formed with a low-frequency cutoff of 0.5 Hz preserve LVL resolution while minimizing distortion from anisotropy/scattering, and so, are used for subsequent interpretation of the
LVL’s basal discontinuity in Section 3.6.1.
The SH images also show a dipping LVL signature with large amplitudes (+/-
~8%) on the SL1 and SL3 profiles but not on the SL2 profile (Figure 3.4). For SL3, the
LVL has the same depth and polarity as seen in the SV image (Figure 3.4) while for SL1 it appears to have the opposite polarity on the SH and SV images. This observation is consistent with the SH RF polarities expected from a layer dipping in the +X direction
(see synthetic tests in Appendix B3; Langston, 1977; Frederiksen and Bostock, 2000).
79

Deeper sub-horizontal discontinuity
The sub-horizontal positive discontinuity at an average depth of ~32 km is observed on the SV images from X=~160 km to ~400 km (Figure 3.4). Its depth decreases from ~35 km to ~30 km as X increases, in general agreement with the results of previous studies
[Sodoudi et al., 2006; Suckale et al., 2009; Pearce et al., 2012], though our images show a few additional features not resolved in previous studies. For example, the SL1 profile has overlapping positive discontinuities in two regions, a subtle “double” discontinuity at
X=250 km and a pronounced one between X=340 km and X=380 km. The positive discontinuity is also slightly disrupted on the SL2 profile at X=275 km (it is more clearly seen in the apparent negative discontinuity that overlies it). As the positive discontinuity approaches the LVL (X < ~160 km), it transitions to a negative discontinuity on the SL2 profile and a closely spaced sequence of negative and positive discontinuities on the SL3 profile. In general, the sub-horizontal discontinuity exhibits relatively weak amplitudes
(2-3%) in the SH images, except for the SL1 profile, which has large amplitudes (+/- 8%) between X=180 km and X=230 km (Figure 3.6c).
Shallower sub-horizontal discontinuity
The sub-horizontal positive discontinuity at an average depth of ~7 km is observed from
X = ~90 km to ~410 km (Figure 3.4). It exhibits a gentle “V” shape on the SL2 profile from X = 90 km to X = 200 km with its maximum depth of ~12 km centered at X = ~110 km and its lateral boundary at X = ~200 km clearly defined by an apparent surface
“breach” (Figure 3.4). More subtle “V” shaped patterns are also observed in three other locations, one centered at X = 260 km on the SL2 profile, as well as two others centered at X=~230 km and X = ~310 km on the SL1 profile. There is also a sharp transition in its signal near the trench along the SL3 profile (X = ~110 km in Figure 3.4e), an observation consistent with the sharp lateral transition in P-wave velocity found beneath this region by Makris [1978].
80

Figure 3.7 shows the SV and SH images for three RF profiles across northern Greece.
The filtering strategy used to form these images differs from the one used for southern
Greece (see Section 3.5.1) because the northern RFs are depleted in high frequencies and enriched in low frequencies relative to the southern RFs. The high-frequency cutoff has minimal effect on the images down to ~0.5 Hz, so we keep as much of this portion of the signal as possible by setting this cutoff at 1.5 Hz. The low-frequency cutoff was set at
0.15 Hz to provide better resolution of closely spaced discontinuities, with minimal loss of bandwidth. This difference in RF frequency content may be due a relative increase in the attenuation of Ps conversions beneath northern Greece relative to those beneath southern Greece, an interesting topic for future study.
The RF profiles show three prominent features that are similar to those observed to the south: (1) a low-velocity layer (LVL) dipping in the +X direction at ~17°, (2) a
(deeper) sub-horizontal positive discontinuity at an average depth of ~36 km, and (3) a
(shallower) sub-horizontal positive discontinuity at an average depth of ~9 km (Figure
3.7). These first two features are generally consistent with the discontinuities found in the teleseismic migration images of Pearce et al. [2012] (Figure 2.5; see also Section 2.6), while the third feature is imaged for the first time in this study.
Dipping Low-velocity Layer
The dipping LVL is observed on all three RF profiles (Figure 3.7). It has a similar dip to the LVL observed beneath southern Greece, but it is significantly thicker (~20 km), and appears more segmented. The negative discontinuity marking its top is best resolved on the NL2 profile (Figure 3.7b), while the positive discontinuity marking the base of the
LVL is clearly observed as far west as X=~70 km beneath NL1 (i.e. beneath Corfu) and as far east as X=~280 km beneath NL3 (Figure 3.7). The basal positive discontinuity is difficult to follow to depths greater than ~70 km as it appears to be disrupted by crustal multiples and/or crosscutting negative discontinuities.
The LVL exhibits different seismic signatures above and below a depth of ~50 km (i.e. either side of X=~180km), consistent with its separation into updip and downdip segments by Pearce et al. [2012] (see Figure 2.5). The updip segment of the LVL is
81

overlain by a low-amplitude positive discontinuity dipping towards +X while the downdip segment is overlain by a sub-horizontal negative discontinuity (see the NL2 profile in Figure 3.7), which is also consistent with the 2-D GRT image in Figure 2.5.
The SH images also contain large amplitude (+/- 6-8%) signals that are roughly collocated with the LVL observed in the SV images. The SH amplitudes found along the
LVL are consistent with a discontinuity dipping in the +X direction, as they have the same polarity as the SV amplitudes in the NL3 profile, but the opposite polarity in the
NL1 profile (see discussion in Section 3.5.1 and synthetic test in Appendix B3), though our resolution of this latter feature is very limited (Figure 3.8b).
Deeper sub-horizontal discontinuity
The sub-horizontal positive discontinuity at an average depth of ~36 km is observed on the SV images from approximately X=200 km to X=450 km (Figure 3.7). It exhibits significant topography (~10 km), particularly along NL1 and NL2. There are two regions with overlapping positive discontinuities on the NL1 profile at X=290 km and X=380 km, while the discontinuity on the NL2 profile is disrupted at X=290 km and exhibits a
“dome-like” shape centered at X=400 km. The discontinuity on the NL3 profile is rather flat and continuous, except for a slight disruption at X=300 km.
Shallower sub-horizontal discontinuity
The sub-horizontal positive discontinuity at an average depth of ~9 km is observed across all three profiles from X = ~100 km to X = ~450 km. It exhibits a broad, asymmetric
“V” shape with a maximum depth of ~18 km centered at X = ~300 km (see NL2 profile in Figure 3.7). This pattern is similar in shape to the one observed beneath southern
Greece (Figure 3.4), but it is located much further inland and appears to reach greater depths. There are three more subtle “V” shaped patterns centered on X = ~100 km, ~210 km, ~340 km, as well as a broad dome-like feature from X = 360 km to X = 420 km. The apex of the dome marks the shallowest depths of this feature beneath northern Greece (~5 km), with no evidence for a surface breach like the one found beneath southern Greece
(see Section 3.5.1).
82

Figure 3.9 shows the SV and SH images for three RF profiles parallel to the WHSZ along the back-arc. The SV images show two sub-horizontal positive discontinuities, one at an average depth of ~34 km (i.e. deeper) and the other at an average depth of ~10 km (i.e. shallower). The LVL is not observed beneath the back-arc, but there appear to be largeamplitude discontinuities at depths of ~100 km (e.g. 100 km < X < 200 km in profile
BA1), which are most likely dominated by crustal multiples.
Deeper sub-horizontal discontinuity
The sub-horizontal positive discontinuity gradually increases from a depth of 30 km beneath southern Greece to ~35 km depth beneath northern Greece, in general agreement with results from previous studies [Makris, 1978; Sodoudi et al., 2006; Pearce et al.,
2012]. Its signal is relatively sharp beneath southern Greece (Y<100 km) but is more irregular beneath northern Greece (Y>220 km), consistent with what we find in the SL and NL profiles shown in Figures 3.6 and 3.8, respectively. There is some indication of overlapping positive discontinuities beneath central Greece (i.e. on NL2 from Y=100 km to Y=200 km), though this region has limited resolution (Figure 3.8b). The SH images also show fairly continuous signals that parallel the overriding Moho beneath southern
Greece (i.e. from X=-100 km to X=100 km), but are more incoherent beneath central and northern Greece (X>100 km).
Shallower sub-horizontal discontinuity
The sub-horizontal positive discontinuity at an average depth of ~10 km is well resolved beneath both southern (Y < 100 km) and northern Greece (Y > 200 km), particularly on the BA2 profile, but is poorly resolved beneath central Greece (i.e. 100 km < Y < 200 km). It exhibits an asymmetric “V” shape in two places, similar to those found beneath southern and northern Greece, with a pronounced one reaching its greatest depths at X =
~210 km and a more subtle one at X = 0 km (see the BA2 profile in Figure 3.9). It rises close to the surface (~6 km depth) at X = ~50 km and ~290 km, but does not appear to breach the surface.
83

In this section, we present RF results for the sparsely distributed stations in central
Greece, starting with stations VLS, LKD, and RLS within the Ionian islands region and followed by stations EVR and AGG within the central Hellenides (see station locations in
Figure 3.3).
At station VLS, the SV RFs exhibit a large amplitude negative peaks (-12%) at
~20 km depth (delay time of ~2.2 sec) surrounded by positive peaks, a shallower one at
~9 km depth (~1.1 sec) and a deeper one with two apparent peaks between 30-40 km depth (3.3 to 5.0 sec; Figure 3.12b). The SV RF stack agrees quite well with the one produced by Sodoudi et al. [2006]. The RF waveforms are similar for backazimuths from
20° to 190° but vary rapidly from 190° to 20°, though there are very few events in this backazimuth range (Figure 3.12a). The CCP image shows that the top three peaks are approximately horizontal while the deepest one at ~40 km depth dips slightly towards -X
(Figure 3.12e). Features at depths greater than 40 km are likely free-surface multiples
(e.g. large amplitude negative peak at 20 km also seen at 60 km).
At station LKD, the SV RFs contain a large-amplitude (>20%) positive peak at a depth of ~20 km (~2.5 sec), followed by two positive peaks at ~40 km and ~60 km depth
(~5.0 and 7.0 sec; Figure 3.13b). The individual RF waveforms are quite similar for backazimuths between 0° and 190° but vary rapidly between 190° and 330°. This variability leads to a CCP image with coherent, high-amplitude peaks at X>100 km (i.e. landward of LKD) and low-amplitude, blurry peaks at X<100 km (Figure 3.13e). The one at ~60 km depth is an exception as it is clearly observed beneath both sides of LKD, but this could be because it is a multiple from one of the shallower discontinuities (e.g. the one at 20 km depth).
At station RLS, the SV RFs contain two large amplitude (> 13%) positive peaks at depths of ~8 km and 17 km (~0.8 and 2.1 sec, respectively), which overlie a large amplitude negative peak (< -18%) at ~25 km depth (~3.0 sec; Figure 3.14b). There are also positive peaks distributed over a depth range from 40 km to 70 km (~5.1 to 8.1 sec) and a negative peak at ~85 km depth (~9.3 sec). The SV RF stack is quite similar to the one produced by Sodoudi et al. [2006]. The RF waveforms exhibit a greater amount of variability with backazimuth as compared to VLS and LKD. The CCP image shows that
84

the peaks above ~45 km dip towards –X, while the distributed positive peaks from 45 km to 65 km depth form a coherent feature dipping at ~23° towards +X (Figure 3.14e).
Deeper features are clearly contaminated by free-surface multiples.
At station EVR, the SV RFs contain a sequence of positive and negative peaks that generally increase in amplitude with depth (Figure 3.15b). Lower amplitude (~9%) positive peaks are observed at ~13 km and ~40 km depth (~1.7 and 4.5 sec, respectively) while larger amplitude positive peaks are present at depths of ~67 km (11% at ~8 sec) and ~130 km (17% at ~15 sec). The pair of positive and negative peaks between ~30 and
~40 km exhibit an apparent flip in polarity versus backazimuth, similar to the polarity flip observed at stations VLS and RLS. EVR has limited data, as only ~30% of the available events yielded RFs with stable amplitudes following preprocessing (Figure 3.15a).
However, our RF stack is very similar to the one generated by Sodoudi et al. [2006] for
EVR using 58 events (see their Figure 4), suggesting our data is sufficient to constrain the subsurface structure. The CCP image shows (1) the polarity flip of the peaks at 30-40 km depth leads to an apparent decrease in their depth of ~6 km as X increases, (2) the positive discontinuity at ~67 km dips slightly towards +X while the underlying sequence of positive and negative peaks dips towards –X (Figure 3.15e).
At station AGG, the SV RFs contain several positive peaks with amplitudes of
~5% at depths of ~10 km (~0.9 sec), ~18 km (~2.2 sec), ~58 km (~6.8 sec), and ~88 km
(~10 sec) with the latter two surrounding a large amplitude (10%) negative peak at ~73 km depth (~8.2 sec; Figure 3.16b). There is also a peak at ~38 km depth (~4.5 sec) that is positive for backazimuths from ~60° to 250° and negative from 250° to 60° (Figure
3.16a,b). Our RF stack is quite similar to the one reported by Sodoudi et al. [2006] for station AGGI (collocated with AGG), except the positive peak at ~38 km depth is smaller, probably because their events mostly come from backazimuths where it is positive (i.e. ~60° to 250°). The CCP image shows the peaks beneath AGG are approximately horizontal (X > ~240 km in Figure 3.16e), and appear disconnected from the ones observed beneath EVR (X < ~ 240 km in Figure 3.15e), at least below ~40 km depth.
The RF results from these stations are used to interpret the overriding and slab
Moho beneath the WHSZ as described in the following section.
85

3.6 Discussion
In this section, our RF imaging results are used to interpret the location of the slab Moho, overriding Moho, and top of the basement beneath southern, northern, and central
Greece. These constraints on the 3-D lithospheric structure along the WHSZ are used to produce maps of the slab and overriding Moho, which we discuss in relation to Moho maps from previous studies and tectonic models of the region.
Southern Greece
The location of the slab Moho beneath southern Greece is interpreted from the RF results presented in Section 3.5.1. The SV CCP images show a LVL dipping at ~17° from 50 km to 100 km depth beneath the Peloponnesus (Figure 3.4). Several previous seismic imaging studies have identified a similar LVL beneath this region using receiver functions [Gesret et al., 2010; Gesret et al., 2011] and teleseismic migration imaging
[Suckale et al., 2009; Pearce et al., 2012]. Following these studies, the LVL is interpreted as subducted oceanic crust with its basal positive discontinuity marking the
Moho of the subducted oceanic slab (i.e. oceanic slab Moho). The location of the oceanic slab Moho is identified by following the maximum amplitude of the dipping positive discontinuity in each SV image from X=~100 km to X=~250 km (Figure 3.6). It was difficult to follow along the SL1 profile due to an apparent “double Moho” signal (i.e. two overlapping positive discontinuities) from X=200 km to X=250 km. In this case, we chose the slab Moho that is most consistent with the interpretation from the teleseismic images of Pearce et al. [2012]. We note that this double Moho signature coincides with the extension of the ocean-continent transition beneath this region (see Figure 2.10;
Royden and Papanikolaou, 2011) and is overlain by active normal faults bounding the southern edge of the Gulf of Corinth [Armijo et al., 1996]. Thus our location of the oceanic slab Moho beneath southern Greece is in overall agreement with previous
86

studies, though our northernmost profile shows some evidence for two overlapping slab
Moho signals.
Northern Greece
The location of the slab Moho beneath northern Greece is interpreted from the RF results presented in Section 3.5.2. The SV CCP images show a ~20 km thick LVL dipping at
~17° from 40 to 100 km depth (Figure 3.7), consistent with the teleseismic migration results of Pearce et al. [2012] (Figure 3.8; see section 2.7.2, Figure 2.5). Following their interpretation, the LVL represents subducted continental crust with the positive discontinuity marking its base representing the continental slab Moho. The location of the slab Moho is identified by following the maximum amplitude of the dipping positive discontinuity in the SV profiles, as was done in Section 3.6.1. The slab Moho is clearly observed on all three RF profiles extending to a depth of ~70 km (X=~230 km) with a peak amplitude of ~8%. It loses coherence rather abruptly beyond X > 230 km where it appears to be cross cut by a negative discontinuity dipping towards –X. There is some indication of weaker slab Moho signals (<%5) continuing to greater depths, but these features are not considered robust due to potential contamination from crustal multiples.
Our interpretation of a shallow slab beneath northern Greece disagrees with the structural interpretation recently proposed by Sodoudi et al. [2006]. They do not identify a shallow slab beneath northern Greece, opting instead to associate positive discontinuities in their Ps receiver functions with a relatively shallow (~28 km) Moho of the Aegean (overriding) plate (see their Figure 16). Their slab Moho is interpreted to be at much greater depth (~200 km) based on discontinuities observed in their Sp receiver functions. We focus here on the shallow slab (Z < 100 km), leaving the discussion of a deeper slab to Section 3.6.2 and the overriding Moho to Section 3.6.5. The difference in structural interpretation at shallow depths largely arises because they only have data from stations KEK and JAN in northwestern Greece while we have the benefit of a highdensity station array (including stations KEK and JAN; see Figure 3.3). Our interpretation is based on robust CCP images, as well as the 2-D GRT images from
Pearce et al. [2012], both of which provide clear evidence of a shallow slab extending into the upper mantle beneath northern Greece (Figure 3.7, Figure 2.5). Furthermore,
87

both studies produce Ps receiver function stacks for KEK and JAN that are similar and exhibit clear positive discontinuities consistent with our shallow slab interpretation
(compare our Figure 3.17 to their Figure 5). As a side note, our RF stacks appear to be enriched in high frequencies compared to the ones from Sodoudi et al. [2006], which we attribute to the higher-fidelity RFs produced by the multichannel preprocessing approach used here (see Section 3.2.1).
Thus, the RF results from both this study and Sodoudi et al. [2006] are consistent with the presence of a shallow slab beneath northern Greece, an interpretation made possible by the high-density array used in this study.
Central Greece
The location of the slab Moho beneath central Greece is interpreted from the RF results for stations VLS, LKD, RLS, EVR, and AGG described in Section 3.5.3 (see Figure 3.3 for station locations). For VLS, the two positive peaks in the SV stack at delay times of
3.3 sec and 5.0 sec are considered slab Moho candidates (Figure 3.12b), given their corresponding depths of ~26 km and ~40 km in the CCP image, respectively (Figure
3.12e). Sodoudi et al. [2006] also identified two similar positive discontinuities beneath
VLS at similar depths of 28 km and 37 km, which they interpreted as the overriding
Moho and slab Moho, respectively. Following their interpretation, we interpret the positive discontinuity at ~40 km depth as the slab Moho (Figure 3.12a). RF traces with conversion points closer to the CTF have higher amplitudes and map to slightly greater depths (~43 km) (Figure 3.12), an effect possibly related to the ocean-continent transition observed beneath this region in marine seismic data [Clement et al., 2000]. Thus, we interpret the slab Moho at ~40 km depth beneath station VLS, with variations in the amplitude of its signal versus backazimuth possibly related to the ocean-continent transition.
For LKD, the positive discontinuity at a delay time of 5.0 sec (~42 km depth) in the SV stack is interpreted as the slab Moho for three reasons (Figure 3.13b). First, it implies minimal slab Moho topography across the ~60 km distance between stations
LKD and VLS. Second, the foreland lithosphere on the other side of the CTF has a
Moho depth of ~30 km [Finetti and Del Ben, 2005], so this interpretation is consistent
88

with a subducting slab south of the CTF (as opposed to an overthrusting slab if we chose a shallower peak as the slab Moho). Third, its amplitude in the SV RF stack is ~10% of the incident P wave amplitude, a typical value for the velocity contrast at the Moho
[Frederiksen and Bostock, 2000]. Sodoudi et al. [2006] interpret a deeper slab Moho at
~115 km depth just south of LKD based on Sp receiver functions. Our Ps RFs are contaminated by crustal multiples at these depths, so we cannot weigh in on the presence of a deeper slab in this region. Thus, we interpret the slab Moho at ~42 km beneath station LKD, but cannot rule out the deeper one at ~115 km identified by Sodoudi et al.
[2006].
For RLS, the positive peaks at delay times of 6.4 sec and 8.4 sec in the SV stack are considered slab Moho candidates (Figure 3.14b). These peaks are clearly contaminated by multiples from the large-amplitude discontinuities at shallower depths
(e.g. ~13% discontinuity at 8 km depth), making our slab Moho interpretation more difficult. So, we rely on the CCP image in Figure 3.14e and slab Moho constraints from previous studies to assist in our interpretation. The CCP image shows that the peak at 6.4 sec maps to a discontinuity dipping in the +X direction at ~17
°
between ~50 km and ~60 km depth (Figure 3.14e). This discontinuity is also consistent with (1) the slab Moho depth observed along southern and northern Greece at this distance from the southern trench (X=~125 km) in both receiver function (Figure 3.6, Figure 3.8) and teleseismic migration (Figure 2.3, Figure 2.5) images, (2) the top of the subducted slab beneath RLS in the local tomography model of Papazachos and Nolet [1997] (see their Plate 2a), and
(3) a reflection from the slab at ~70-75 km depth interpreted from surface seismic data by
Zelt et al. [2005] (see location in Figure 3.14d). Thus, we interpret the slab Moho beneath
RLS to be dipping at ~17° between ~50 km to 60 km depth.
Sodoudi et al. [2006] interpret the slab Moho to be at ~70 km depth beneath RLS based on the maximum in their Ps RF stack. This discrepancy in Moho depth is likely the result of interference from crustal multiples, but detailed modeling of the RF signals is needed to address this issue further.
For EVR, the two positive peaks at delay times of 8.0 sec (~65 km depth) and
14.5 sec (~130 km depth) in the SV stack are considered slab Moho candidates (Figure
3.15). Our dataset is limited (preprocessing yielded only 12 events with stable RF
89

amplitudes), but provides an RF stack that is very similar to the one produced by Sodoudi et al. [2006] using 58 events (see their Figure 4), so appears sufficient to interpret a slab
Moho depth. Following a similar rational as that used for RLS, we interpret the positive discontinuity at ~65 km as the slab Moho because (1) it appears to dip towards +X while the deeper peak appears to dip towards –X (Figure 3.15e), (2) it is consistent with the slab
Moho position along southern and northern Greece (Figure 2.3, Figure 2.5, Figure 3.6,
Figure 3.8) as well as the slab top observed by Papazachos and Nolet [1997] (see their
Plate 2c) (3) there is intermediate depth seismicity (Z = ~80 km to ~110 km) just ~30 km to the east of EVR (Figure 3.15d). We note that the positive peak at ~130 km depth has the largest amplitude (~20%) and is consistent with the slab Moho interpreted by Sodoudi et al. [2006] using Sp receiver functions (see their Figure 8); however, modeling of crustal multiples in our Ps RFs would be required to further support their deeper slab
Moho interpretation. Thus, we interpret the slab Moho at ~65 km depth beneath EVR, though there is also some evidence for the slab Moho at ~130 km interpreted by Sodoudi et al. [2006].
For AGG, we interpret the moderate-amplitude positive peak (~8%) at a delay time of 10.3 sec in the SV stack as the slab Moho (Figure 3.16b), which corresponds to a depth of ~86 km as shown in the CCP image (Figure 3.16e). This interpretation is based on a similar argument as was used for EVR, as it is consistent with (1) the slab Moho observed beneath northern and southern Greece (Figure 2.3, Figure 2.5, Figure 3.6,
Figure 3.8), (2) the slab top observed by Papazachos and Nolet [1997] (see their Plate
2c), and (3) the location of intermediate depth seismicity (Figure 3.16d). Furthermore, our images along northern and southern Greece also show a relatively low amplitude slab
Moho signal at depths greater than ~80 km (see references in (1)), with the amplitude reduction attributed to slab dehydration, an abrupt increase in slab dip, and/or slab tearing
(see “Disappearance of the subducted crust at depth” discussion in Section 2.7.1 and
2.7.2). It is possible these processes are also responsible for the large-amplitude negative peak (-12%) we observe at ~71 km depth, a possibility we leave for a future study. We also note that the CCP image shows an apparent offset between the slab Moho beneath stations EVR and AGG (Figure 3.16e), which roughly coincides with the seismicity in
(3), though this apparent disconnect may simply result from differences in crustal
90

multiples. Thus, we tentatively interpret the slab Moho at ~86 km depth beneath AGG.
Improved event coverage and detailed modeling of the RF waveforms is needed to better image the slab Moho structure between stations EVR and AGG.
In conclusion, the slab Moho is observed beneath each of our five stations in central Greece, starting at ~40 km beneath the island of Cephalonia (i.e. station VLS) and extending to ~86 km beneath the central Hellenides (i.e. station AGG).
Slab Moho map
A map of slab Moho depth along the WHSZ is produced using our interpretation of the
RF images described in the previous three sections. It is constructed by interpolating our picks of the slab Moho location (see Figures 3.5, 3.7, 3.10, 3.11), along with the ~70 km slab depth constraint beneath the westernmost Gulf of Corinth from Zelt et al. [2005], to a 2-D grid using a nearest neighbor algorithm. The gridded depth values are contoured at an interval of 10 km, a conservative value given the depth uncertainties from the velocity model (~4 km).
Figure 3.18a shows the resulting slab Moho map, which is characterized by (1) a broad ramp from northern to southern Greece and (2) locally variable Moho topography near the Gulf of Corinth (20 km < Y <80 km). The slab ramp deepens from northern to southern Greece with a slope of only ~2°, consistent with the ~70 km of differential slab retreat inferred by Pearce et al. [2012] (see Figure 2.9). It is also consistent with the results of gravity, local tomography, and attenuation studies [Tsokas and Hansen, 1997;
Tiberi et al., 2001; Papazachos and Nolet, 1997; Hashida et al., 1988; Konstantinou and
Melis, 2008 ].
Local deviations from the smooth slab ramp are observed beneath central and southern Greece (20 km < Y < 100 km), with an apparent “valley” beneath the Gulf of
Corinth and an apparent “ridge” along our SL1 profile (Figure 3.6). Several previous studies also show evidence for local, abrupt changes in slab topography. Zelt et al.
[2005] infer a sudden change in slab dip to explain why reflections from the slab beneath the western Gulf of Corinth suddenly disappear beneath the central Gulf of Corinth (see their Figure 4). The local tomographic model of Papazachos and Nolet [1997] shows a high-velocity slab that appears segmented (~100-150 km wide blocks) with both flat-
91

lying (e.g. beneath Attica and Evia at X=325 km on their Plate2b) and steeply dipping regions (beneath the eastern Peloponnesus at X= 225 km on their Plate 2a). Papazachos et al. [2000] also infer abrupt changes in slab topography beneath the eastern
Peloponnesus based on the location of Wadati-Benioff seismicity. Thus, our results, combined with those from previous studies, suggest the slab may be locally segmented in some way, at least near its top. Denser station coverage, particularly beneath central
Greece, and detailed modeling of the RF signals are needed to better resolve small-scale segmentation of the slab along the WHSZ.
Overall, our slab Moho map shows a continuous slab at shallow depths (<100 km) extending from southern to northern Greece, but there is also some evidence for local variations in slab topography around the Gulf of Corinth.
Comparing our slab Moho map with the map of Sodoudi et al. [2006] suggests there may be overlapping slab segments beneath northern Greece. Their slab Moho depths are similar to ours along southern Greece (i.e. beneath the Peloponnesus), but are ~130 km deeper than ours beneath northern Greece (see also Section 3.6.2). There are two possible explanations for this discrepancy: (1) the positive discontinuity at ~200 km depth observed in their Sp receiver functions beneath northern Greece does not represent a slab Moho or (2) two overlapping slab segments exist beneath northern Greece. Our study is limited to Ps receiver functions, so we cannot evaluate this first explanation.
However, it is likely that the ~200 km deep discontinuity they observe is related to subduction processes in some way given the geologic evidence for long-lived subduction in the Aegean (since Pre-late Jurassic) [Papanikolaou, 2009 and references therein] and the large volume of high velocity material (i.e. subducted slabs) imaged down to the transition zone beneath the Aegean sea [van der Hilst et al., 1997]. Thus, there may be two overlapping slabs beneath northern Greece, one dipping at 17° towards N60E from
30 km to 70 km depth (Figure 3.13a) and the other sitting horizontally at ~200 km depth as interpreted by Sodoudi et al. [2006].
92

The transition between the southern and northern portions of the slab may be accommodated in one of three ways: (1) a continuous shallow slab as depicted in Figure
3.18a, with a separate deeper slab beneath northern Greece, (2) a continuous slab that ramps from shallow depths beneath southern Greece to ~200 km depth beneath northern
Greece as proposed by Sodoudi et al. [2006], with a separate shallow slab beneath northern Greece, or (3) a set of three distinct/decoupled/individual subducted slabs - two shallow slabs beneath southern and northern Greece, and one deep slab beneath northern
Greece. The available evidence suggests that the connection described in (1) is most likely, as the gravity, attenuation, and tomography studies noted above do not show evidence for a shallow slab tear (see Section 3.6.1). Our RFs from stations EVR and
AGG provide tentative evidence for both a shallow slab Moho at 75-90 km and a deeper one at ~130 km depth (see Section 3.6.1; Figures 3.15e and 3.16e), but this deeper signal is contaminated by crustal multiples and cannot be explicitly connected to the deeper slab imaged by Sodoudi et al. [2006]. Thus, the shallow slab is likely continuous across northern and southern Greece, with the deeper slab detached in some way. Additional data from central Greece are needed to resolve the nature of any tear between the shallow and deep slabs.
Sodoudi et al. [2006] observe a negative discontinuity at ~170 km depth beneath northern
Greece in their Sp receiver functions, which they attribute to the Lithosphere-
Asthenosphere Boundary (LAB) of the overriding (Aegean) lithosphere. This interpretation is clearly at odds with the shallow slab seen in our images, as it implies the shallow slab imaged here is embedded within the overriding lithosphere. An alternative interpretation is that the negative discontinuity represents the LAB of the shallow slab imaged beneath northern Greece in this study (Figure 3.8) and Pearce et al. [2012] (see
Section 2.7.2). This interpretation yields a slab thickness of ~100-130 km beneath northern Greece, which is similar to the slab thickness of ~120 km inferred beneath southern Greece using Sp receiver functions [Sodoudi et al., 2006] and magnetotelluric
93

data [Galanopoulos et al., 2005]. It is also consistent with the hypothesis that the slabs along northern and southern Greece have similar elastic thicknesses, an idea proposed by
Pearce et al. [2012] to explain the similar slab dips along both slab segments (see section
2.7.3). Thus, we advocate for an alternative interpretation of the LAB beneath northern
Greece, in which the negative discontinuity at ~170 km depth observed by Sodoudi et al.
[2006] represents the LAB of the shallow slab observed in this study (Figure 3.18a), not the LAB of the overriding lithosphere.
Southern Greece
The positive, sub-horizontal discontinuity at ~30 km to 35 km depth beneath southern
Greece is interpreted as the Moho of the overriding lithosphere, hereafter referred to as the overriding Moho (Figure 3.6), an interpretation consistent with previous teleseismic imaging studies [Sodoudi et al., 2006; Suckale et al., 2009; Pearce et al., 2012], as well as studies using gravity [Tsokas and Hansen, 1997], body-wave [Papazachos and Nolet,
1997] and surface-wave tomography [Karagianni and Papazachos, 2007], as well as reflection/refraction data [Makris, 1978; Zelt et al., 2005; Sachpazi et al., 2007].
The location of the overriding Moho is identified by following the maximum amplitude of the positive sub-horizontal discontinuity in each CCP image, similar to what was done for the slab Moho (see Section 3.5.1), except that we include a couple additional measures to assure our picks are robust. First, we pick from CCP images formed with frequencies up to 1.5 Hz (Figure 3.5) to improve resolution, as opposed to the maximum frequency of 0.5 Hz used when picking the slab Moho (see further discussion in Section 3.6.1). Second, we define a set of criteria to separate robust picks of the overriding Moho from more speculative ones. Robust picks identify where the overriding Moho adheres to the assumptions of the RF imaging method used here (i.e. planar, first-order discontinuity). They are defined based on the following criteria: (1) maximum amplitude greater than ~5%, (2) no other positive peaks within +/- ~10 km, (3) no rapid depth variations (i.e. less than ~10 km over a horizontal distance of 30 km). In
94

general, robust picks of the overriding Moho are obtained from X = ~160 km to X =
~400 km, while more speculative (i.e. poorly resolved) picks are found in a few places, most notably nearest the subducted Moho (X<~180 km) and on the SL1 profile between
X=340 km and X=380 km (Figure 3.6).
The positive discontinuity marking the overriding Moho transitions to a negative discontinuity as it approaches the NW coast of the Peloponnesus (i.e. X < ~160 km, see dotted line in Figure 3.6d), consistent with the results of Pearce et al. [2012] and Sodoudi et al. [2006]. This negative discontinuity has been interpreted as marking the top of (1) a serpentinized forearc mantle [Sodoudi et al., 2006] or (2) subduction channel sediments
[Pearce et al., 2012]. We prefer the latter explanation for two reasons (see the
“Subduction interface” portion of Section 2.7.1 for further discussion). First, the oceanic slab is old (~200-250 Ma), and therefore cold, so its basaltic crust is not expected to undergo significant dehydration until ~80 km depth [van Keken et al., 2011], which argues against forearc mantle serpentinization, as proposed by Bostock et al. [2002].
Second, a sliver of high-velocity material with mantle-like velocities (Vp = ~7.8 km/s) has been observed beneath the SW Peloponnesus [Makris, 1978], but it is horizontally separated by ~60 km from the coherent overriding Moho we observe at X > 160 km
(Figure 3.6). This suggests that even if this sliver represents serpentinized mantle, it is not (yet?) a part of the overriding lithosphere, but is instead a block of high-velocity material within the subduction channel (see further discussion in Section 3.6.6).
In conclusion, the overriding Moho is observed as a sharp, positive discontinuity beneath southern Greece from 160 km < X < 400 km.
Northern Greece
The positive, sub-horizontal discontinuity at ~32 km to ~38 km depth beneath northern
Greece is interpreted as the overriding Moho (Figure 3.8), in general agreement with the
Moho depths obtained from teleseismic migration imaging [Pearce et al., 2012], receiver functions [Sodoudi et al., 2006], local surface-wave tomography [Karagianni and
Papazachos, 2007], and gravity data [Tsokas and Hansen, 1997].
The location of the overriding Moho beneath northern Greece is identified by following the sub-horizontal discontinuity using the same picking approach as the one
95

used for southern Greece. In general, the overriding Moho beneath northern Greece is more discontinuous, or “patchy”, than the one found beneath southern Greece, as observed in the 2-D GRT images of Pearce et al. [2012] (i.e. Figures 3.6 versus 3.8;
Figures 2.3 versus 2.5). It is observed as a continuous, first-order discontinuity within three robust regions centered at X = 250 km, 330 km, and 400 km, which are separated by three speculative regions where the overriding Moho is difficult to identify, one nearest the slab Moho (X < ~220 km) and two others centered at X = ~300 km and ~380 km.
Our overriding Moho depths of 38 to 40 km beneath the northern Hellenides (X <
~280 km) are deeper than the 33 to 36 km depths obtained from the receiver function study of Sodoudi et al. [2006], but shallower than the depths of 40 to 45 km found in previous gravity- [Tsokas and Hansen, 1997] and surface wave-based [Karagianni and
Papazachos, 2007] overriding Moho maps. We attribute the shallower depths found by
Sodoudi et al. [2006] to (1) their lack of station coverage where all other studies observe the deepest overriding Moho depths (i.e. X = ~220-280 km) and (2) differences in the background velocity models used in the two studies, which can explain a discrepancy of
~2 km. The larger depths obtained by Tsokas and Hansen [1997] and Karagianni and
Papazachos [2007] generally occur to the west of where we observe the start of the overriding Moho (i.e. X < ~220 km in Figure 3.8, Section 3.6.6), so are likely reflect crustal material within the subduction channel. Karagianni and Papazachos [2007] extend these larger depths into the region where we find a well-developed Moho, but this may result from smearing of the low-velocities within the subduction channel, given the limited lateral resolution afforded by surface wave inversions (see their Figure 8 for resolution tests). Thus, we favor our overriding Moho depths of 38 to 40 km beneath the northern Hellenides, as they are based on CCP images formed with high-density array data (Figure 3.8) that allow us to distinguish the signature of a well-developed overriding
Moho from crustal material within the subduction channel (see further discussion in
Section 3.6.6).
In conclusion, the overriding Moho beneath northern Greece is observed as a
“patchy” discontinuity between ~40 and 30 km depth from X = ~200 km to X = ~450 km. Ascertaining the nature of the disruptions in its signal requires detailed modeling of
96

the RF signals, and careful consideration of the tectonic history along northern Greece, the latter of which is the focus of Chapter 4.
Back-Arc
The positive, sub-horizontal discontinuity at ~25 km to ~37 km depth beneath the backarc region is interpreted as the overriding Moho (Figure 3.10), an interpretation that is broadly consistent with results from local surface-wave tomography [Karagianni and
Papazachos, 2007], gravity data [Tsokas and Hansen, 1997], and seismic refraction imaging [Makris, 1978]. The location of the overriding Moho is identified using the same picking approach as the one used for southern Greece. It is observed as a sharp, coherent discontinuity beneath southern Greece, and transitions to a more irregular,
“patchy” set of discontinuities beneath northern Greece, consistent with our previous interpretations beneath these two regions (see “Southern Greece” and “Northern Greece” sections above). The transition from a sharp Moho to a “patchy” one occurs at approximately Y = 130 km (Figure 3.10), which coincides with the northernmost limit of the Central Hellenic Shear zone, a broad zone of active right-lateral and extensional deformation [Papanikolaou and Royden, 2007]. This transition lies within a broader pattern of overriding Moho depth fluctuations occurring along the NW Aegean coast, which we discuss further in the “Overriding Moho map” part of this section.
Central Greece
The overriding Moho beneath central Greece is interpreted based on the RF results presented in Section 3.5.4. We break our discussion down into two parts, with the first part addressing stations VLS, LKD, and RLS in the Ionian Islands, and the second part focusing on stations EVR and AGG in the central Hellenides. We do not interpret an overriding Moho beneath any of the stations in the Ionian Islands region, because (1) we do not observe coherent, positive discontinuities that are clearly attributable to the overriding Moho and (2) our RF profiles along southern and northern Greece show that the slab is too shallow within this portion of the forearc (~40 to 60 km depth) to allow for a well-developed overriding Moho (Figures 3.6 and 3.8; see Section 3.6.6).
97

For the central Hellenides region, the positive discontinuity observed at ~37 km depth beneath stations EVR and AGG is interpreted as the overriding Moho (Figure 3.15 and 3.16), consistent with the results from local surface-wave tomography [Karagianni and Papazachos, 2007] and gravity data [Tsokas and Hansen, 1997; Tiberi et al., 2001].
The location of the overriding Moho is interpreted using the same picking approach as the one used for southern Greece. Robust picks of the overriding Moho are only obtained along the sub-horizontal portion of the positive discontinuity from X = 200 km to X =
260 km (Figure 3.15 and 3.16).
The positive discontinuity to the SW of station EVR (X < 200 km) appears to abruptly deepen to ~40 km in the CCP image (Figure 3.15e). This feature should be interpreted with caution because it is produced by only 3 events (backazimuths from 150 ° to 330 ° in Figure 3.15a), and the observed variations in RF amplitude between ~4 and 4.5 sec (corresponding to ~40 km depth) in Figure 3.15b may be influenced by anisotropy
[Frederiksen and Bostock, 2000]. That said, we favor an interpretation in which the overriding Moho deepens SW of EVR because (1) it is supported by the gravity and surface-wave tomography results mentioned above, and (2) a similar deepening of the overriding Moho is observed beneath southern and northern Greece close to the slab (e.g.
X = ~200 km in the SL2 profile of Figures 3.6d).
In contrast, Sodoudi et al. [2006] interpret a shallowing of the overriding Moho
SW of station EVR based on their observation of positive discontinuities at depths of 33 km and 36 km beneath stations EVR and AGGI (collocated with AGG), respectively.
We find a similar overriding Moho depth beneath station AGG (~38 km), but there is significant disagreement regarding the overriding Moho depth beneath station EVR (our depth of 37 to 40 km vs. their depth of 33 km). This discrepancy likely arises because their interpretation is based exclusively on a single-station RF stack for EVR, while ours is also based on a CCP image beneath EVR (Figure 3.15e). Single-station RF stacks in situations where there are large variations in amplitude versus backazimuth, as is the case at station EVR (Figure 3.15a,b), can lead to significant uncertainties in the discontinuity amplitude and delay time (i.e. depth). Indeed, the single-station RF stack for EVR produced by Sodoudi et al. [2006] shows a weak amplitude (only ~25% of their slab
Moho’s amplitude) positive discontinuity at ~4 sec (see their Figure 5), an indication of
98

destructive interference during stacking. Thus, there is significant uncertainty in the location of the overriding Moho SW of station EVR, but we favor a deepening of the overriding Moho given the arguments presented in the previous paragraph.
In conclusion, the overriding Moho beneath central Greece does not extend beneath stations in the Ionian Islands region, but it is observed to the NE of the
Hellenides (X > 200 km) beneath stations AGG and EVR at ~37 km depth. There is also some evidence that supports a deepening of the overriding Moho SW of station EVR, but additional data and detailed modeling of the RF waveforms are needed to better resolve the overriding Moho structure beneath central Greece.
Overriding Moho map
A map of the overriding Moho is produced from our interpretation of the RF results for each region along the WHSZ (see preceding paragraphs of this section). The map is built by first interpolating our robust Moho picks to a 2-D grid using a nearest neighbor algorithm, and then contouring the gridded values at an interval of 4 km.
Figure 3.18b shows the resulting overriding Moho map, which exhibits two pronounced features: (1) a large-scale decrease in Moho depth from ~38 km beneath the northern Hellenides to ~28 km beneath the Aegean Sea, and (2) small-scale variations in
Moho depth within a N-S oriented corridor located between the northern Hellenides and
Aegean Sea.
The large-scale feature is characterized by an approximately E-W oriented decrease in Moho depth from ~35-40 km below the northern Hellenides to 25-30 km beneath the Aegean Sea over a horizontal distance of 200-250 km. This feature is broadly consistent with previous maps of the overriding Moho derived from receiver functions [Sodoudi et al., 2006], surface wave tomography [Karagianni and Papazachos,
2007], and gravity data [Tsokas and Hansen, 1997]. The relatively thin crust beneath the
Aegean Sea has been attributed to widespread crustal extension over at least the last 30
Ma due to rapid retreat of the oceanic slab along southern Greece [Jolivet et al., 2010 and references therein]. The origin of the thick crust beneath NW Greece is more controversial. Previous studies suggest the thick crust developed in a collisional setting and is now supported by isostatic forces [e.g. Sodoudi et al., 2006], while our images
99

suggest the crust was thickened by material scraped off the top of a subducting continental slab (see further discussion in Section 2.7.2) with subduction-related forces also playing a prominent role [i.e. Royden and Papanikolaou, 2011], as discussed in
Section 3.6.7.
The small-scale feature is characterized by large-amplitude (~3-5 km) fluctuations in Moho depth with a wavelength of 120-160 km occurring along a N-S oriented corridor from the volcanic arc to the northern tip of the Thermaikos Gulf ([X,Y] = [250, -40] to
[310, 320] in Figure 3.18b). This feature exhibits a smaller wavelength (~120 km), larger amplitude (~5 km) and deeper average Moho depth (~35 km) in the north, as compared to the values of ~160 km, ~3 km, and ~30 km in the south, respectively. Note that its peaks and valleys are well constrained by robust picks from our CCP images, while transition regions between peaks and valleys are poorly constrained, so should be interpreted with caution.
The Moho depths along the N-S corridor found in previous studies are broadly consistent with the small-scale feature observed in our overriding Moho map. Karagianni and Papazachos [2007] observe fluctuations in Moho depth along the N-S corridor that decrease in both wavelength (~120 km to 160 km) and amplitude (3 km to 2 km) from central to southern Greece, in good agreement with our small-scale feature. Tsokas and
Hansen [1997] also identify fluctuations in Moho depth along the N-S corridor, though their fluctuations are smoother than ours (amplitudes of 1-2 km vs. 3-5 km). They also obtain large gravity residuals (>60 mGal) along the N-S corridor (see Figure 12 in Tsokas and Hansen [1997]), so this difference in smoothness may be due, at least in part, to the regularization used in their gravity inversion. Sodoudi et al. [2006] find fluctuations in
Moho depth along the N-S corridor that are both smoother and less regularly spaced than the depth fluctuations in our map. This difference is likely due to (1) their limited station coverage along the N-S corridor (~12 stations), (2) the fact that several of their stations are located where we observe speculative picks of the overriding Moho (see further discussion in following paragraph), (3) their exclusive use of single-station RF stacks to define their Moho depths while we also use CCP images. We also note that an even smaller-scale of Moho depth fluctuations, with wavelengths of ~60 km or less, has been observed in local studies centered on the Gulfs of Corinth and Evia [Tiberi et al., 2001;
100

Sachpazi et al., 2007]. Thus, previous studies generally support the small-scale fluctuations in Moho depth we observe along the N-S corridor, with one notable exception.
The overriding Moho depths of ~40 km that we find beneath the northern
Thermaikos Gulf ([X,Y] = [350,240] in Figure 3.18b) are markedly different than the 30 to 32 km depths observed in previous studies [Tsokas and Hansen, 1997; Sodoudi et al.,
2006]. Our CCP images suggest this depth discrepancy may result from crustal heterogeneity beneath this region, as indicated by (1) the abrupt change in Moho depth from 40 km depth beneath the Thermaikos Gulf to ~32 km depth just 40 km further to the
NE ([X,Y]=[370,270] in Figure 3.8 and Figure 3.18b) and (2) evidence for two overlapping positive discontinuities, or a “double” Moho signature, in the transition zone between the deep and shallow depths described in (1). Crustal heterogeneity can bias the
Moho depths from each study in different ways. The only station Sodoudi et al. [2006] have within this region samples the complex transition zone described in (2), so it makes sense they obtain depths closer to ~30 km. The gravity-based estimate of Tsokas and
Hansen [1997] may be biased because they assume a constant crustal density, while our images suggest there may be significant vertical and lateral variations in crustal density beneath this region. Of course, our depths may also be biased from the constant crustal velocity used in our background model; however, the 3-D CCP imaging approach used here allows us to identify, and therefore avoid, picking in regions with complex Moho variations, so our overriding Moho map should be less sensitive to bias from crustal heterogeneity.
In conclusion, our overriding Moho map contains two robust features: (1) a largescale decrease in Moho depth from ~38 km beneath the northern Hellenides to ~28 km beneath the Aegean Sea, and (2) small-scale variations in Moho depth within a N-S oriented corridor between the northern Hellenides and Aegean Sea. Our overriding
Moho map raises important questions regarding the start of the overriding lithosphere, and the role of slab dynamics and strain localization in shaping the overriding lithosphere, topics we explore further in the following sections.
101

The overriding Moho first emerges from near the slab Moho as a large-amplitude, firstorder discontinuity in our RF profiles beneath southern Greece at X = ~160 km (Figure
3.6), beneath northern Greece at X = ~220 km (Figure 3.8), and more tentatively beneath station EVR in central Greece at X = ~190 km (Figure 3.15). We interpret this first
Moho emergence as the start of the overriding lithosphere, a rather “liberal” definition, if one considers a well-developed Moho to be a necessary, but insufficient, condition for a
“plate-like” overriding lithosphere (see continental tectonics discussion in Molnar
[1988]). That said, our definition is actually quite conservative compared to the one used in previous studies, as discussed next.
Previous overriding Moho maps extend well into the forearc region, up to, and even beyond the NW coast of Greece [Tsokas and Hansen, 1997; Sodoudi et al., 2006;
Karagianni and Papazachos, 2007], implying the overriding lithosphere starts much farther into the forearc, possibly as far as the first emergence of forearc crystalline crust
(i.e. crustal backstop), which occurs ~50 km seaward of the bathymetric trench along southern Greece [Truffert et al., 1993; Lallemant et al., 1994]. This alternative starting point is up to 220 km further into the forearc from where we observe the start of the overriding lithosphere (Figure 3.6), which is certainly evidence for a vast expanse of forearc crystalline crust, but not necessarily an overriding Moho.
The forearc crystalline crust seaward of our overriding lithosphere start (X < ~170 km) is underlain by a low-velocity layer containing subducted sediments/crust (i.e. subduction channel), followed by mantle lithosphere of the slab, as indicated by our RF profiles along southern Greece (Figure 3.6), as well as results from previous studies (e.g.
Makris [1978]; Suckale et al. [2009]; Gesret et al. [2011]). This indicates there are two sharp increases in velocity (or density) with depth that could be misinterpreted as overriding Moho signals, a shallower one marking the top of the overriding crystalline crust, and a deeper one marking the slab Moho (i.e. base of LVL). The forearc Moho depths from gravity [Tsokas and Hansen, 1997] and surface-wave data [Karagianni and
Papazachos, 2007] are actually quite similar to our slab Moho depths (Figure 3.18a,
Section 3.6.1), which explains their apparent overriding Moho signal (i.e. it’s actually the slab Moho). Sodoudi et al. [2006] identify two Moho signals within the forearc, a deeper
102

one interpreted as the slab Moho, and a shallower one they attribute to the overriding
Moho. Given the velocity structure described above and the segmented nature of the discontinuities observed at comparable depths (~30 km) within the forearc (e.g. Makris
[1978]; Figure 3.6; Section 3.6.5), their apparent shallower Moho is likely from a boundary either on top or within the crystalline crust.
Our interpreted start of the overriding lithosphere displays several characteristics that provide insight into how the overriding Moho is generated so close to the slab
(Figure 3.18b): (1) the first emergence of the Moho is roughly linear across the WHSZ, with a N15W orientation that is similar to the long-axis of the short-wavelength feature in our overriding Moho map (see Section 3.6.5), (2) it closely follows the ~70 km depth contour of the slab Moho in Figure 3.18a, and so appears to retreat with the slab, and (3) it is located ~50 km trenchward of the volcanic arc beneath southern Greece. The strong correlation between the structure of the slab and overriding lithosphere described in (1) and (2) suggests they are related to one another somehow (e.g. Royden and Papanikolaou,
[2011]) (see further discussion in Section 3.6.7). Furthermore, it’s unlikely that arc magmatism is primarily responsible for generating the start of the overriding lithosphere, given characteristics (2) and (3) above, along with the ~80 km depth predicted for significant slab dehydration beneath southern Greece [van Keken et al., 2011], and the lack of arc volcanism across northern Greece. One possibility recently proposed by
Burchfiel et al. (in preparation), is that the overriding crust is produced from fragments of continental crust detached from the slab during slab retreat, an idea that we explore further in Chapter 4.
In conclusion, we define the start of the overriding lithosphere based on a welldeveloped Moho signal that emerges only after the slab has subducted into the mantle, a more conservative definition than assumed in previous studies, suggesting slab dynamics may play an important role in shaping the overriding lithosphere, as discussed in the following section.
103

The geometry of the overriding Moho, and its relationship to topography, provide valuable information regarding the forces supporting the Hellenides thrust belt. Different definitions for the start of the overriding lithosphere, as described in Section 3.6.6, have led to very different lithosphere geometries for the WHSZ, with previous studies showing a symmetric, shallow-deep-shallow (i.e. “U” shaped) Moho geometry perpendicular to the trench [Tsokas and Hansen, 1997; Sodoudi et al., 2006; Karagianni and Papazachos,
2007], while we obtain an asymmetric, deep-shallow geometry consistent with only the landward limb of their “U” (Figure 3.18b, see also Figure 2.9).
These two contrasting geometries lead to very different conclusions regarding the forces supporting the Hellenides. The “U” shaped Moho found in previous studies has its largest depths beneath the highest surface topography, leading some to suggest that the
Hellenides topography is compensated by crustal thickening, consistent with an Airy-
Heiskanen model of isostasy (e.g. Sodoudi et al. [2006]; Suckale et al. [2009]). However, we observe a relationship between our overriding Moho depths and surface topography that clearly deviates from that predicted by Airy-Heiskanen isostasy in a couple of ways.
First, let’s examine the Moho structure beneath the peak topography of the
Hellenides, where previous studies find the largest Moho depths (i.e. the bottom of their
“U”). Our RF profiles show that the peak topography along southern and northern
Greece occurs above the tip of the mantle wedge, where the overriding Moho has yet to develop (see X = 140 km and X = 200 km in Figures 3.6 and 3.8, respectively). This is also where our 2-D GRT images show an abrupt transition in the character of the subduction interface (Figures 2.3 and 2.5), which has been interpreted as possible evidence for the accretion of slab-derived material near the base of the overriding crust
[Pearce et al., 2012] (see “Subduction interface” discussion in Sections 2.7.1 and 2.7.2).
This suggests that the peak topography of the Hellenides is supported, at least in part, by slab-derived forces (e.g. its flexural strength, accretion of slab-derived material, etc.), but what about the topography further away from the slab, where a well-developed Moho is present?
To address this issue, we compare our receiver function-based overriding Moho depths (Figure 3.18b) to the depth of the crust-mantle boundary calculated assuming
104

surface topography is locally compensated (i.e. supported) by changes in the thickness of a constant density crust, as predicted by the Airy-Heiskanen model of isostasy (Figure
3.19). Isostasy-based Moho depths (D
AI
) are computed from the surface elevation at a given location (S) using the following expression [Lowrie, 2007]:
D
AI
=
⎧
⎪
⎪
⎨
⎛
⎜
⎝
⎪
⎪
⎩
⎛
⎜
⎝
( ρ
ρ
ρ m c m
ρ
− c
−
−
ρ
ρ
ρ c w c
)
( )
( )
+ 1
⎞
⎟
⎠
S + D
S = 0
+ 1
⎞
⎟
⎠
S + D
S = 0 for S >= 0 for S < 0 [10], where ρ w
, ρ c
, and ρ m
are the densities of water, crust, and mantle, respectively; D
S=0
is the
€ assumed to be ~1000 kg/m^3 and 2850 kg/m^3, respectively. The other two parameters are obtained by inverting all the surface elevation values that coincide with the location of robust picks of the overriding Moho depth (i.e. black astericks in Figure 3.18b) using a grid search across a range of values for
ρ m
(3000 to 3600 kg/m^3) and D
S=0
(28 to 36 km). This yields optimal values of
ρ m
= 3400 kg/m^3 and D
S=0
= 32 km, with a minimum RMS error of ~3.6 km. The relative (i.e. ρ m
– ρ c
), not absolute, density value is well-resolved, such that reasonable changes in the assumed crustal density (2750 to 2950 kg/m^3) produce virtually identical fits to the data, so long as (
ρ m
–
ρ c
) is held constant at
550 kg/m^3.
Figure 3.19a shows the resulting map of isostasy-based overriding Moho depth, produced by (1) collecting the surface elevation value (S) at each point where there is a robust pick of the overriding Moho depth (i.e. black astericks in Figure 3.18b), (2) converting each S value to isostasy-based Moho depth (D
AI
) using equation [10] and the parameter values described above, and (3) contouring the resulting depth values using the same nearest neighbor algorithm used to produce Figure 3.18b. The map of D
AI
contains a large-scale decrease in depth, with the largest depths beneath the high-topography northern Hellenides and the shallowest depths beneath the low-topography Aegean Sea
105

(Figure 3.19a), much like the large-scale feature observed in our RF-based Moho map
(Figure 3.18b; Section 3.6.5). This similarity is clearly observed in our map of residual
Moho depth - the difference between the RF-based and isostasy-based depths (Figure
3.19b) - as most of the residual Moho depths observed across the WHSZ are smaller than
+/- 4 km. This suggests that the surface topography of the overriding lithosphere may be regionally (> ~150 km) compensated by crustal thickness variations.
There are also local zones within the overriding lithosphere that contain large
(>+/- 4 km) residual Moho depths (Figure 3.19b). These anomalous zones are found within the N-S corridor where we observe short-wavelength fluctuations in overriding
Moho depth (black box in Figure 3.19b; see Section 3.6.5), and near the start of the overriding Moho along southern Greece (seaward of the Gulf of Corinth). Our sign convention is such that zones with positive residual depths are overcompensated – the
RF-based Moho depth is larger (i.e. the crust is thicker) than that predicted by isostasy. It is also interesting to note that zones of anomalous residual Moho depth do not form isolated peaks, but instead occur as closely-spaced (~50 km) pairs of overcompensated and undercompensated zones (e.g. the positive-negative pair centered beneath
Thessaloniki at [X,Y]=[350,250] in Figure 3.19b).
The presense of zones containing large residual depths is a clear indication that
Airy-Heiskanen isostasy is an insufficient model to explain how the surface topography is locally supported within the overriding lithosphere. There are a number of possible explanations for such anomalous zones, including (1) variations in crustal density (Pratt-
Hayford model), (2) variations in the strength of the overriding lithosphere (e.g. elastic plate model of Vening Meinesz) [Lowrie, 2007], and (3) dynamic stresses associated with retreating subduction (e.g. Hager [1984]), and its requisite deformation of the overriding lithosphere (see Chapter 4 for further discussion of overriding lithosphere deformation).
The joint inversion of both seismic and gravity data would help us better understand the nature of these local anomalous zones.
In conclusion, the overriding Moho found in this study is (1) asymmetric normal to the trench, (2) not present beneath the peak topography of the Hellenides, and (3) broadly consistent with regional, but not local, Airy-Heiskanen isostasy. Taken together,
106

these results suggest that subduction-related forces may help to support the surface topography of the Hellenides thrust belt.
The shallower sub-horizontal discontinuity observed at an average depth of ~8 km across most of the WHSZ is interpreted as the top of crystalline basement (e.g. Figures 3.6 and
3.8). This interpretation is supported by several geologic observations, particularly along southern Greece where the crystalline basement is exposed in a few places. First, our depths to the basement top are consistent with geologic mapping-based estimates for the approximate thickness (~7 km) of the unmetamorphosed sediments that overly the crystalline basement along southern Greece (e.g. Papanikolaou and Royden [2007]).
Furthermore, the basement top, represented by the Arna unit, is exposed at the surface in the footwall of the East Peloponnesus detachment fault, in precisely the same location where we observe a surface “breach” in the shallower subhorizontial discontinuity along southern Greece (compare cross-section B in Figure 6 of Papanikolaou and Royden
[2007] to X = 200 on profile SL2 in Figure 3.6). We do not observe a surface breach of the shallower discontinuity along the northern Hellenides (X<300 km in Figure 3.8), consistent with the lack of basement exposure that is typical of thin-skinned thrust belts during continental subduction [Royden, 1993a]. Further to the NE (i.e. X>300 km) along northern Greece, the shallower discontinuity forms a dome-like shape with an apex that nears the surface, and reduces its amplitude, beneath the exposed crystalline basement of the Rhodope metamorphic core complex and related units [Dinter and Royden, 1993;
Papanikolaou, 2009]. Thus, field-based observations are generally consistent with our basement top interpretation, and so we now turn to a closer examination of its structure.
Our interpreted basement top places a fundamental constraint on the structure of the overriding lithosphere, which we briefly discuss here, leaving a more in depth discussion of how the overriding lithosphere is built and deformed to Chapter 4. We focus on four features of the overriding crust: (1) its overall stratification, (2) where the basement top starts along northern Greece, (3) the different amounts of basement retreat along northern and southern Greece, and (4) the asymmetric “V” patterns along its top.
107

Crustal stratification
Our basement top interpretation indicates that the overriding crust is broadly stratified into (at least) two layers. The shallower layer is comprised of unmetamorphosed (i.e. lower-velocity) sediments of the thin-skinned thrust belt, with an average thickness of ~8 km, while the deeper layer, between the basement top and overriding Moho, is presumably comprised of crystalline (i.e. higher-velocity) crust, with a thickness of ~25 km. This indicates that, on average, a vertical crustal column contains approximately one part unmetamorphosed sediments to three parts crystalline crust. However, these average values should be interpreted with caution given the thickness variations observed in both the shallower and deeper layers, as well as the evidence for crustal heterogeneity in several locations along the WHSZ (e.g. overlapping Moho signals discussed in Section
3.6.5). A more detailed analysis is needed to better resolve how the crust is layered along the WHSZ, including detailed modeling of the RF waveforms to account for potential multiple contamination from the basement top discontinuity.
Start of crystalline crustal along northern Greece
The shallower positive discontinuity extends well into the forearc along northern Greece, reaching up to the west coast of Corfu (i.e. X = ~100 km in Figure 3.8), though with reduced amplitude (~4%) compared to that observed further inland (~>6% at X = 150 km). Its continuity appears to be a slightly disrupted at X = ~130 km (Figure 3.8), leading us to interpret this point as the start of the basement top along northern Greece, with an uncertainty of +/- 20 km or so. This starting point occurs ~80 km landward of the thrust front (dashed orange line in Figure 3.8a), suggesting that the crystalline portion of the overriding crust develops rapidly during continental slab subduction. One possible explanation for this rapid development, along with the overall continuity of the basement top along northern Greece, is that the crystalline crust is formed from blocks of crystalline crust detached from the continental slab and accreted to what eventually becomes the overriding lithosphere (see start of the overriding lithosphere discussion in
Section 3.6.6), as has been proposed in several recent studies [Royden and Papanikolaou,
2011; Tirel et al., 2013; Burchfiel et al., in prep.].
108

Differential basement retreat between northern and southern Greece
The start of the crystalline basement along northern Greece (X = ~130 km) is offset by
~180 km from the start found beneath southern Greece (X = -50 km) using marine seismic data [Truffert et al., 1993; Lallemant et al., 1994]. Following the rationale laid out in Section 2.7.3, we interpret this offset to be a basement-level estimate for the amount of differential retreat between the northern and southern portions of the WHSZ.
The large difference between the basement- (~180 km) and slab-derived (~70 km in
Figure 2.9) amounts of differential retreat indicates the shallow overriding crust along southern Greece experienced a sudden increase in (1) retreat rate relative to the slab and/or (2) the rate that new crystalline crust is produced from slab-derived material (e.g.
Burchfiel et al., in prep). The large discrepancy in differential retreat rate estimates between the basement and slab raises some interesting questions regarding how the overriding lithosphere is built and deformed in a retreating subduction system such as the
WHSZ (i.e. Burchfiel et al., in prep.), which we address in greater detail in Chapter 4.
Asymmetric “V” patterns
The shallower positive discontinuity forms an asymmetric “V” shaped pattern in a number of places along the WHSZ (see “Shallower sub-horizontal discontinuity” portion of Section 3.5). These are generally subtle features, with dips of 10 ° to 15 ° on their shorter, SW side, and shallower dips of 5 ° to 10 ° on their longer, NE side.
Many of the “V” shaped patterns coincide with extensional systems in the overriding lithosphere. For example, the SW side of the “V” shaped pattern between X =
200 and X= 280 km on our SL1 profile (Figure 3.6) coincides with the rapidly extending
(~10 mm/yr) normal faults bounding the SW edge of the Gulf of Corinth [Armijo et al.,
1996]. The prominent “V” on the SL2 profile from X = 80 to 200 km (Figure 3.6) also traverses a region containing active normal faulting [Papanikolaou and Royden, 2007;
Floyd et al., 2010]. Furthermore, the asymmetric nature of these features is consistent with the geometry of detachment systems in other parts of the WHSZ, such as the
Strymon Valley detachment system of northern Greece [Dinter and Royden, 1993]. Thus, the “V” patterns appear to be related to extensional deformation in the overriding
109

lithosphere, both present and possibly past, but it’s not yet clear how they tie into the broader dynamics of slab retreat along the WHSZ, a topic we will examine further in
Chapter 4 using an integrated interpretation of geologic, geodynamic, and seismic data.
3.7 Concluding Remarks
In this study, a 3-D receiver function algorithm was applied to teleseismic data from broadband seismic stations distributed across the WHSZ. The resulting images were used to interpret the location of the slab Moho, overriding Moho, and basement top along southern, central, and northern Greece, leading to the following conclusions regarding the 3-D lithospheric structure along the WHSZ and its geodynamic implications:
1) The slab Moho dips at 17° in an approximate N60E azimuth beneath southern Greece
(Figure 3.6), northern Greece (Figure 3.8) and, more tentatively, beneath central Greece
(Figures 3.11-3.15). This observation suggests that a continuous slab extends from northern to southern Greece (Figure 3.18a), consistent with the ~70 km of differential retreat inferred by Pearce et al. [2012], though there is some evidence for local variations in slab topography around the Gulf of Corinth (Figure 3.6).
2) Our overriding Moho map exhibits two robust features:
* a long-wavelength decrease in Moho depth from the thick (35-40 km) crust of
NW Greece to the thin crust (25-30 km) beneath the Aegean Sea, and
* short-wavelength fluctuations in Moho depth (+/- ~4 km) along a roughly N-S corridor between NW Greece and the Aegean Sea (Figure 3.18b).
3) The start of the overriding lithosphere is defined based on the first emergence of a well-developed overriding Moho signal, which occurs at X = ~160 km beneath southern
Greece (Figure 3.6) and X = ~220 km beneath northern Greece (Figure 3.8).
4) Our overriding Moho map and its relationship to topography provide several lines of evidence that support a slab retreat model for the WHSZ:
110

* The start of the overriding lithosphere occurs only after the slab Moho has subducted well into the mantle (Figure 3.18);
* The overriding Moho is deepest where it starts, meaning the overriding lithosphere is asymmetric when viewed along the dip direction of the slab (Figure
3.18b), not “U” shaped as has been found in previous studies;
* The peak topography of the Hellenides lies above the tip of the mantle wedge, a region that lacks a well-developed overriding Moho (Figures 3.6 and 3.8);
* Our overriding Moho depths are generally deeper than those expected for isostasy (Figure 3.19b), a possible indication of dynamic topography.
5) A shallow positive discontinuity observed at an average depth of ~8 km across much of the WHSZ is interpreted as the top of the crystalline basement (Figures 3.6, 3.8, 3.10,
3.11-3.15), which indicates the overriding crust contains (at least) two layers: a shallow layer comprised of unmetamorphosed (i.e. low-velocity) sedimentary rocks and a deeper layer comprised of crystalline (i.e. high-velocity) basement rocks.
6) The start of the crystalline basement (i.e. crustal backstop) along northern Greece (see
X = ~130 km in Figure 3.8) is ~180 km landward of the crustal backstop found beneath southern Greece in previous studies, yielding a difference in basement retreat between southern and northern Greece that is much larger than the difference in slab retreat (~180 km vs. ~70 km).
7) The top of the crystalline basement forms several asymmetric “V” shaped patterns along southern Greece (Figure 3.6), which coincide with regions of active crustal extension.
111

3.8 Tables
Layer Z (km) € α
(km/s)
β
(km/s)
ρ (kg/m^3)
1 40 6.5 3.7 2.8
2 - 7.8 4.35
Table 1.1: 1D background velocity model for SL imaging
3.2
Layer Z (km) € α (km/s) β (km/s) ρ (kg/m^3)
1 40 6.4 3.6 2.7
2 - 7.85 4.4
Table 1.2: 1D background velocity model for NL imaging
3.1
Layer Z (km) € α
(km/s)
β
(km/s)
ρ (kg/m^3)
1 32 6.45 3.65 2.7
2 - 7.83 4.38
Table 1.3: 1D background velocity model for back-arc imaging
3.1
€ Z the depth from the free surface to the bottom of the layer, α is P-wave velocity, β is
S-wave velocity, and ρ is density.
112

3.9 Figures
Figure 3.1: Map of the WHSZ showing the distribution of broadband stations and teleseismic events used in this study. a) Stations deployed as part of project MEDUSA along southern (SL, 40 stations) and northern (NL, 40 stations) Greece are denoted by white circles and squares, respectively. Also shown are stations from GEOFON (15 upward-pointing black triangles) and the National Observatory of Athens (6 downwardpointing black triangles), as well as the location of major tectonic boundaries (see description in Figure 2.1). b) Distribution of teleseismic events used in RF imaging for stations from NL (black circles), SL (red circles), GEOFON (white stars), and GEOFON plus NOA (white diamonds). Map is centered on central Greece with white concentric circles denoting 10° increments in epicentral distance (innermost circle at 30°).
113

Figure 3.2: Schematic illustrating the RF imaging method used in this study. Incident planar wave fronts (thick dashed blue lines) from two different events, one propagating down-dip (E
(E
2
1
, shown at an example reference time, t
0
) and the other propagating up-dip
), strike a 3-D, dipping discontinuity in S-wave velocity whose surface is depicted by the dashed black rectangle. The seismic station (black triangle) is located on the Earth’s surface (thick black lines). It records the waves traveling along rays for the incident Pwave (thin blue lines) and the Ps conversion from the dipping discontinuity (thin red lines). Rays travel in a background velocity model with a single layer (thin black lines) over a half-space. The model domain is defined using a horizontal coordinate system oriented along the discontinuity’s down-dip (X) and along-strike (Y) directions with depth (Z) measured from the surface. Thin lines with arrows labeled φ and φ
Y illustrate how the two backazimuth angles discussed in the text are defined. The X- and Z- components of the Ps slowness vector (p x
S and η
S , respectively) are computed by applying Snell’s law in a coordinate system that is aligned with the strike and dip, θ , of the discontinuity (i.e. X
D
, Y, Z
D
). C
1
and C
2
(black dots) identify 3-D conversion points along the dipping discontinuity for events E
1
and E
2
, respectively (i.e. C
1
=[C
X,1
C
Y,1
C
Z,1
]). In this case, they fall within different 3-D common conversion point bins, as indicated by the thin black cubes.
114

Figure 3.3: Map view of the 3-D RF model domain oriented perpendicular (X, N60°E) and parallel (Y, N30°W) to the WHSZ. Station locations are identified by triangles.
Black lines show the location of RF profiles presented in this study (see Figures 3.4-
3.11). RF results are also presented for selected stations in central and northern Greece
(labeled, white triangles; see Figures 3.12-3.17). No results are shown from the station immediately adjacent to LKD (i.e. LKD2) because it only had two events available.
Geologic features are the same as in Figure 3.1.
115

Figure 3.4: RF profiles beneath southern Greece formed with RFs filtered from 0.05 Hz to 0.5 Hz and mapped to depth using the 1-D velocity model in Table 3.1.
Discontinuities are assumed to be horizontal down to 40 km depth and then dip at 17° below 40 km depth, in accordance with the discontinuity dip observed in teleseismic migration images beneath this region [Pearce et al., 2012]. a) Location of RF profiles beneath Y=20 km (SL1), Y=-10 km (SL2), and Y=-40 km (SL3). Black triangles denote station locations. Additional symbols show the location of geologic features described in
Figure 3.1. b) Hit count profiles along SL1, SL2, and SL3 (see profile locations in (a)) showing the # of conversion points in each CCP bin, as denoted by the colorbar, with blue indicating zero conversion points. Note that no smoothing is applied to the hit count profiles in (b). RF profiles extracted from our 3D CCP image along SL1 (c), SL2 (d), and SL3 (e) for SV (left) and SH (right). Colorbars in (c), (d), and (e) indicate the average RF amplitude in each bin as a % of the incident P-wave amplitude.
116

Figure 3.5: RF profiles beneath southern Greece with all panels built the same way as in
Figure 3.4, except that the images in (c), (d), and (e) are formed with RFs filtered from
0.05 Hz to 1.5 Hz.
117

Figure 3.6: RF profiles beneath southern Greece with all panels identical to those in
Figure 3.4, except they also include our structural interpretations from this study. Thick solid black lines with black dots denote the interpreted locations of the basement top
(Z=~10 km), the overriding Moho (Z=~35 km), and the dipping oceanic slab Moho from this study. Black dots identify robust Moho picks used to construct the Moho maps in
Figure 3.18. The SL1 profile also includes the location of the slab Moho (dashed line) interpreted from the SL 2-D GRT images of Pearce et al. [2012] (see Figure 2.3 and
Section 2.7.1). Note the clear presence of free surface multiples from the overriding
Moho in the SV profiles (thin black line with “X” marks).
118

Figure 3.7: RF profiles beneath northern Greece following the same layout as Figure 3.4.
RFs are filtered from 0.15 Hz to 1.5 Hz, as describe in Section 3.5.2, and mapped to depth using the 1-D velocity model in Table 3.2. Discontinuities are assumed to be horizontal down to 40 km depth and then dip at 17° below 40 km, in accordance with the
NL 2-D GRT images of Pearce et al. [2012]. Profile locations are at Y=300 km (NL1),
Y=270 km (NL2), and Y=240 km (NL3).
119

Figure 3.8: RF profiles beneath northern Greece with all panels identical to those in
Figure 3.7, except they also include the structural interpretations from this study. Thick solid black lines with black dots denote the interpreted locations of the basement top
(Z=~10 km), the overriding Moho (Z=~32 km), and the dipping oceanic slab Moho from this study. Black dots identify the discrete picks used to construct the Moho maps in
Figure 3.18. Free surface multiples from the overriding Moho (not shown) are less coherent than the ones observed in the RF profiles beneath southern Greece (Figure 3.6), additional evidence of an irregular overriding Moho structure beneath northern Greece.
120

Figure 3.9: RF profiles beneath the back-arc (BA) following the same layout as Figure
3.4. RFs are filtered from 0.05 Hz to 1.5 Hz, and mapped to depth using the 1-D velocity model in Table 3.3. Discontinuities are assumed to be horizontal at all depths. a) Location of RF profiles beneath X=280 km (BA1), X=310 km (BA2), and X=340 km (BA3) as depicted by solid black lines in a) (additional symbols are the same as Figure 3.4). The view in a) is from the Aegean Sea looking towards the trench of the WHSZ. Dashed black lines denote location of SL2 and NL2 profiles as shown in Figure 3.4 and Figure
3.7, respectively.
121

Figure 3.10: RF imaging results beneath the back-arc with all panels identical to those in
Figures 3.8 except they also include our structural interpretation from this study. Thick solid black lines with black dots denote the interpreted locations of the basement top
(Z=~8 km) and overriding Moho (Z=~32 km), with black dots marking discrete picks of these features, though these picks are not used to build the contour map in Figure 3.18
(only picks from arc-perpendicular profiles are used).
122

Figure 3.11: Large-scale version of interpreted SV RF profiles beneath SL2, NL2, and
BA2, as in Figure 3.6d, 3.8d and 3.10d, respectively. Panels (a, b) also include the location of the interpreted structures from the SL (Figure 2.3) and NL (Figure 2.5) 2-D
GRT images (see Sections 2.7.1 and 2.7.2, respectively). Dashed black lines denote positive discontinuities marking the slab Moho, overriding Moho, and top of subducted continental crust beneath northern Greece (b). Dotted black lines identify negative discontinuities marking the top of subducted oceanic crust and subduction channel interface beneath southern Greece (a).
123

Figure 3.12: RF imaging results for station VLS. a) Backazimuth (black line labeled
φ
) and slowness magnitude (red line labeled p) for each RF trace in b) and c). Traces are sorted by backazimuth, starting with events that propagate in the updip, or –X, direction
(i.e. φ = 60°). b) SV RF traces with red and blue amplitudes corresponding to positive
(downward slow-to-fast) and negative (downward fast-to-slow) discontinuities in S-wave velocity, respectively. All SV traces have the same amplitude scale. The single trace to the right shows the stacked RF trace normalized by the number of traces. The black arrows labeled BT and SM show our interpreted basement top at 9 km and slab Moho at
40 km, respectively. c) SH RF traces with the same layout as the SV traces in b), but with amplitudes saturated at ~70% of those for SV. All traces in b) and c) are filtered from
0.05 Hz to 1.5 Hz. d) Topographic map centered on station VLS (black triangle, see VLS location in Figure 3.3) along with the conversion points for each RF trace at our interpreted slab Moho depth of 40 km (black dots). The white horizontal line marks the location of the CCP image in e). Geologic boundaries are the same as in Figure 2.1. e)
CCP image beneath station VLS extracted from the 3-D RF model. The asterisks identify basement top and slab Moho beneath station VLS, as in b). Conversion points (d) and
CCP image (e) were formed with the velocity model in Table 3.1 assuming horizontal discontinuities for all depths.
124

Figure 3.13: RF imaging results for station LKD, with the same layout as Figure 3.12 (see
LKD location in Figure 3.3). Conversion points in d) correspond to a depth of 42 km.
Black arrows denote our interpretations of the basement top (tentative) and slab Moho.
125

Figure 3.14: RF imaging results for station RLS, with the same layout as Figure 3.11 (see
RLS location in Figure 3.3). The conversion points in d) and CCP image in e) were formed with the velocity model in Table 3.1 assuming discontinuities are horizontal down to 40 km depth and dip at 17° below 40 km depth. Conversion points in d) correspond to depths of 45 km (black dots) and 70 km (plus signs), values that bound our interpreted slab Moho depth of 50 to 60 km depth (solid black line with asterisks in e)).
The black arrow label BT identifies our tentative interpretation of the basement top. Two other features are also shown: 1) the approximate fault trace of the M w
6.3 Movri
Mountain right-lateral earthquake (dashed red line in d)) with a dip of ~84
°
and hypocenter location shown by the black star in e) [Gallovic et al., 2009] and 2) the location of the sub horizontal reflector from the subducting slab (dashed black line) observed by Zelt et al. [2005]
.
126

Figure 3.15: RF imaging results for station EVR, with the same layout as Figure 3.11 (see
EVR location in Figure 3.3). The conversion points in d) and CCP image in e) were computed with the velocity model in Table 3.1, assuming horizontal discontinuities.
Conversion points in d) correspond to depths of 35 km (black dots) and 65 km (plus signs), the latter being our interpreted slab Moho depth (labeled SM1 in b and marked by a black asterisk in e)). The black arrow labeled BT denotes the basement top at ~13 km depth. The stars in d) show the hypocenters for an “anomalously” deep set of earthquakes as reported by the International Seismological Center (ISC) (see Figure 2.10 for more complete map of ISC hypocenters). Star color corresponds to depth interval: magenta for
80-90 km, cyan for 90-100 km, and blue for 100-110 km. The catalog depth uncertainties are less than +/-10 km.
127

Figure 3.16: RF imaging results for station AGG, with the same layout as Figure 3.11
(see AGG location in Figure 3.3). The conversion points in d) and CCP image in e) were computed with the velocity model in Table 3.1, assuming discontinuities are horizontal down to 40 km depth and dip at 17° below 40 km depth. Conversion points in d) correspond to our tentative interpretation of the overriding (OM?) and slab Moho (SM?) depths at 38 km (black dots) and 88 km (plus signs), respectively. These depths are also identified by black asterisks in e). The stars in e) show earthquake hypocenters for different depth intervals (see Figure 3.15 for details): 80-90 km (magenta), 90-100 km
(cyan), and 100-110 km (blue).
128

Figure 3.17: RF traces for stations KEK (a) and JAN (b) in northern Greece (see location of KEK and JAN in Figure 3.3). The top panel shows a topographic map centered on each station (black triangle) and the conversion points corresponding to our interpreted slab Moho depths (black dots) of ~27 km for KEK (a) and ~53 km for JAN (b).
Conversion points are computed using the velocity model in Table 3.2 assuming horizontal discontinuities. Geologic boundaries are the same as in Figure 2.1. The lower three panels follow the same layout as the panels labeled “BAZ + slowness”, “SV RF”, and “SH RF” in Figure 3.11. The RF signals are filtered from 0.05 Hz to 1 Hz, for comparison with the RF results of Sodoudi et al. [2006].
129

130

Figure 3.18: Contour maps of (a) the slab Moho depth (D
SM
), and (b) the overriding
Moho depth (D
OM
) along the WHSZ derived from our 3-D RF model. a) Slab Moho depths contoured from 30 km to 100 km in 10 km increments as indicated by the colorbar. The dashed black line shows the location of the flat-lying slab at ~70 km depth from Zelt et al. [2005]. b) Overriding Moho depths contoured from 20 km to 40 km in 4 km increments. The black box identifies the short-wavelength feature and the thick dashed black line defines our interpreted start of the overriding lithosphere, as described in Sections 3.6.5 and 3.6.6, respectively. In both panels, the symbols (black asterisks within white circles) show our Moho depth picks used to construct the map while in panel b) we also show the locations where the Moho signal was too incoherent to pick
(black “X” marks). Geologic features are the same as those in Figure 3.1. The solid grey lines outline the coastline of Greece.
131

132

Figure 3.19: Contour maps of (a) crustal thickness (i.e. overriding Moho depth) predicted assuming local Airy-Heiskanen isostasy (D
AI
) and (b) residual Moho depth computed by subtracting the isostasy-based Moho depths (i.e. D
AI based depths in Figure 3.18b (i.e. D
OM
- D
AI
in (a)) from our receiver function-
). (a) Isostasy-based Moho depth computed from surface topography assuming
ρ w
= 1000 kg/m^3,
ρ c
= 2850 kg/m^3,
ρ m
= 3400 kg/m^3, and D
S=0
= 32 km (see text for details). Surface topography (S) is filtered prior to computing D
AI
, using a 2-D Gaussian function with unit standard deviation and [X, Y] dimensions of 20 km by 20 km. The contour maps in (a) and (b) are not significantly altered by changes in the smoothing of the surface topography (e.g. no smoothing, [X, Y] dimensions of 120 km by 120 km). Contour levels are from 26 km to 38 km in 2 km increments as indicated by the colorbar. (b) Residual Moho depth computed by subtracting, point-by-point, the RF-based Moho depths in Figure 3.18b from isostasybased depths in Figure 3.19b, then contouring the resulting residual depth values.
Contour levels are from -12 km to 8 km in 4 km increments as indicated by the colorbar.
Other features are the same as in Figure 3.18b.
133

4.1 Introduction
The western Hellenic subduction zone (WHSZ) is an ideal setting for studying how the overriding lithosphere is built in a retreating subduction system. Some type of building mechanism is required because slab retreat creates space that the overriding lithosphere must fill in some how [Burchfiel et al., in prep.]. The WHSZ has been extensively studied (e.g. geologic mapping, petrologic, geophysical, geodetic, etc.; see review in
Section 2.2) making it possible to develop detailed tectonic and geodynamic models to explain overriding lithosphere development. The prevailing view is that the alternating subduction of oceanic and continental slab domains plays an important role in determining slab retreat rates and the amount of overriding lithosphere extension [Brun and Faccenna, 2008; Jolivet and Brun, 2010; Royden and Papanikolaou, 2011; Jolivet et al., 2013 and references therein]. Geodynamic models have had great success in reproducing the observed relationship between slab buoyancy and retreat rates [Royden and Papanikolaou, 2011], and the detailed structural pattern associated with crustal
134

blocks accreted to the overriding lithosphere, including the pressure-temperature-time paths of high-pressure metamorphic (HP) rocks [Tirel et al., 2013].
The lithospheric structure gleaned from the seismic images presented in Chapters
2 and 3 is consistent with many elements of a slab retreat-based model for the WHSZ, including the faster retreat of the southern, oceanic slab relative to the northern, continental slab (Figure 2.9), evidence for the accretion of slab-derived crustal material into the overriding crust (see Sections 2.7.1 and 2.7.2), and the correlation between the geometry of the slab, overriding lithosphere, and surface topography (Figure 3.18 and
3.18). While our seismic images provide substantial evidence that slab dynamics play an important role in shaping the overriding lithosphere along the WHSZ, as discussed in
Section 3.6.7, it remains to be seen what the detailed structure of the overriding crust observed in our RF profiles, such as the geometry of the basement top discussed in
Section 3.6.8, can tell us about the geologic structures that accommodate overriding lithosphere deformation during slab retreat.
The goal of this study is to obtain a detailed picture of the crustal structure along the WHSZ by using geologic data and geodynamic models that describe how the overriding crust is built during slab retreat to refine our seismic-based structural interpretation of the overriding lithosphere from Chapter 3. First, we review the geologic data that constrain the subduction history of each different oceanic and continental slab domain along the WHSZ. Then, we describe recent geodynamic models of the WHSZ, including their predicted relationship between slab buoyancy, retreat rate, and the style of overriding deformation, particularly the three-step, subduction-exhumation cycle proposed by Tirel et al., [2013]. With these tools in hand, we perform an integrated interpretation of the overriding crust, starting with our RF profile along southern Greece where the overriding crust is presently extending, then our RF profile along northern
Greece where continental blocks are currently detaching from the slab to build new crust.
Along southern Greece, we observe a pronounced asymmetric “V” shaped pattern in the basement top, with its shorter trenchward side underlain by several weak amplitude discontinuities and its landward side forming a “dome-like” structure that breaches the surface at its apex. The model of Tirel et al. [2013] predicts a remarkably similar structure, leading us to interpret the asymmetric “V” as the structural fingerprint of a
135

collapsing overriding crust, with upper crustal slices piled up on its trenchward side, while the dome on the seaward side exposes mid to lower crustal rocks at its apex, which is bounded on its landward side by the reactivated Pindos oceanic suture.
4.2 Tectonic setting
Here, we briefly review the tectonic setting along the WHSZ as recorded in the Hellenic thrust sheets (i.e. nappes), packages of sedimentary rocks that were scrapped off the top of the slab as it subducted to the northeast (e.g. Royden and Papanikolaou [2011], and references therein). One can tell whether the originating slab domain was oceanic or continental based on the inferred water depth that the sediments were deposited in: deepwater sediments (and, of course, ophiolites) derive from an oceanic domain while shallow-water sediments derive from a continental domain [Royden and Papanikolaou,
2011]. Figure 4.1a shows that the thrust sheets alternate between deep-water (e.g.
Pindos) and shallow-water (e.g. Pelagonian) sedimentary rocks, and so, the subducted slab must have alternated between oceanic and continental domains.
The oceanic domains of the Pindos and Vardar are generally thin (~2 km thick or less), and so, they are often considered to be a single boundary, termed a suture zone
(e.g., Papanikolaou [2009], Royden and Papanikolaou [2011]). Conversely, a single continental domain may be derived from several morphologically distinct continental
“blocks”, accreted one after another, as is the case for the External Hellenides (e.g.
Paxos, Ionian, Gavrovo Tripolis, etc. in Figure 4.1a). Thus, most of the thrust belt is comprised of continental derived material, with oceanic suture zones marking the boundaries between different continental domains.
We focus here on the timing and deformation history of the four most recent domains, which are, from oldest to youngest, the Internal Hellenides, Pindos suture,
External Hellenides, and Mediterranean (Figure 4.1a), leaving discussion of the more internal Vardar and Rhodope domains to a subsequent study. The basal (i.e. last) thrust sheet of the Internal Hellenides domain was accreted to the overriding lithosphere at approximately 60-55 Ma, marking the end of this phase of continental subduction
[Burchfiel et al., 2008; Tirel et al., 2013]. This was followed by the subduction of the
136

Pindos oceanic domain until the middle Eocene in the Cyclades, and a bit longer
(Oligocene) in the Peloponnesus [Jolivet et al., 2010; Tirel et al., 2013]. A total of approximately 500 km of Pindos oceanic lithosphere is believed to have been consumed during this time [Stampfli and Borel, 2004; Royden and Papanikolaou, 2011].
The earliest (frontal) Pindos thrust sheets, along with the continental-derived thrust sheets of the Internal Hellenides, started their exhumation at ~45 Ma from mantle depths of ~60 km or more, as recorded by the Tinos HP rocks (Figure 4.1b). The late
Eocene (34-37 Ma) closure of the Cycladic portion of the Pindos oceanic domain coincided with an isobaric heating event recorded in the Tinos HP rocks at lower crustal depths (~34 km) (Figure 4.1b; Parra et al., [2002]).
The Pindos domain was followed by the subduction of a series of continental lithosphere “blocks”, collectively referred to as the External Hellenides, beginning in approximately Oligocene time. We can break this event into two stages, based on a change in slab composition, and in the nature of overriding lithosphere extension. The first stage involved Oligocene to late Miocene subduction of continental blocks covered by moderate to shallow water sedimentary rocks (i.e. Ionian and Gavrovo Tripolis).
During this time, subduction rates slowed from 25-35 mm/yr to 5-12 mm/yr, while extension in the overriding crust was generally arc-perpendicular and distributed across what is now northern and southern Greece [Royden and Papanikolaou, 2011]. This stage coincided with a second phase of exhumation for the HP rocks of Tinos, from the lower crust to near the surface (Figure 4.1b), in a back-arc spreading type of tectonic setting
[Papanikolaou, 1993; Parra et al., 2002].
The second, most recent stage of the External Hellenides event is defined by the subduction of the Paxos continental block, and a transition to the subduction of the Ionian oceanic domain, which we will term the “Mediterranean” oceanic domain, following
Tirel et al. [2013], so as not to confuse it with the continental Ionian “block” of the
External Hellenides. The transition from continental to oceanic subduction has progressively marched from south to north, beginning in the mid to late Miocene into the present (see Figure 15 in Royden and Papanikolaou [2011]). In the mid Miocene, most of the WHSZ was consuming Paxos continental lithosphere, perhaps as far south as the
NW tip of Crete, while Mediterranean oceanic lithosphere entered the subduction system
137

further to the south, where Crete is presently located [Le Pichon and Angelier, 1979;
Kopf et al., 2003]. This led to the segmentation of the subduction system, with a southern portion characterized by fast subduction rates (~30 mm/yr) and intense, localized extension within the Cretan sea and East Peloponnesus fault [Papanikolaou and
Royden, 2007], and a northern portion where subduction rates remained slow and extension in the overriding crust was much smaller [Meyer et al., 2002; Royden and
Papanikolaou, 2011]. The boundary separating the “slow” Paxos subduction and “fast”
Mediterranean oceanic subduction has steadily migrated northward, reaching its current location at Cephalonia transform fault at ~5 to 7 Ma, such that Paxos continental lithosphere is still subducting north of here to this day (see Figure 15 in Royden and
Papanikolaou [2011]). Thus, as Mediterranean oceanic lithosphere progressively entered the subduction system, both slab retreat rates and the magnitude of overriding lithosphere extension progressively increased.
The metamorphic grade of the thrust sheets exposed at the surface also progressively increases from northern to southern Greece, tracking the increase in the magnitude of overriding lithosphere extension. In northern Greece, where continental subduction is still occurring and Miocene to present extension has been the smallest
[Meyer et al., 2002; Royden and Papanikolaou, 2011], the External Hellenides thrust sheets are mostly comprised of unmetamorphosed sediments, with little to no exposure of the underlying crystalline basement, a typical characteristic of thin-skinned thrust belts formed in retreating continental subduction settings [Royden, 1993], though notable exceptions do occur within the structurally complex Olympos and Ossa windows [Jolivet and Brun, 2010]. Conversely, furthest to the south, where the most Mediterranean oceanic lithosphere has subducted and the magnitude of extension has been the greatest, there is pervasive exposure of the high-pressure metamorphic rocks that form the crystalline basement of the overriding crust, rocks that have risen from depths of 60 km or more, within “cold” (e.g. Crete) or “hot” (e.g. Tinos in Figure 4.1) metamorphic core complexes [Jolivet and Brun, 2010; Tirel et al., 2013].
Across the Peloponnesus, Attica, and Evia (e.g. southern dashed line in Figure
4.1a), where Mediterranean oceanic lithosphere has subducted more recently, there is a mixture of metamorphic grades, with some thrust sheets largely unmetamorphosed (e.g.
138

offshore Ionian unit), while others, such as the Arna and Mani units, have undergone HP metamorphism and exhumation in “cold” metamorphic core complexes, but there is not
(yet) any exposure of Cycladic-like units that exhibit both “cold” and “hot” core complexes (Figure 4.1) [Jolivet and Brun, 2010; Tirel et al., 2013]. Thus, the overriding lithosphere along the WHSZ is presently in three different stages of development: (1) a presently building, relatively intact thin-skinned thrust belt to the north, (2) a partially collapsed thrust belt across the Peloponnesus, and (3) a completely collapsed thrust belt furthest to the south.
The exposed crystalline basement rocks along southern Greece provide important clues as to how the overriding lithosphere is built during slab retreat. They are generally comprised of units equivalent to the unmetamorphosed thrust sheets, only they have been buried to mantle depths (e.g. Hacker et al. [2011], and references therein), as indicated by their peak pressures (>15 kbars as shown in Figure 4.1b; see summary of P-T-t data for different HP rocks in Jolivet and Brun [2010]), then exhumed, and finally exposed at the surface in detachment faults of core complexes. For example, the highly metamorphosed
Cycladic unit is a paleogeographic equivalent of the weakly metamorphosed Pindos nappe found further to the north [Jolivet and Brun, 2010], while the Arna and Mani units are paleogeographic equivalents of the largely unmetamorphosed Ionian continental unit
[Royden and Papanikolaou, 2011]. This suggests that the overriding crust is built from material scrapped off the top of the subducted slab at two (or more) structural levels, a shallower level comprised of unmetamorphosed sediments of the thin-skinned thrust belt, and a deeper level comprised of highly metamorphosed sediments, and possibly crystalline crust from continental slab domains [Burchfiel et al., in prep.]. However, it is not clear from the geology alone how this building process works exactly, a knowledge gap that recent geodynamic models have sought to fill.
4.3 Geodynamic modeling
Let’s now look at what recent geodynamic models have to say about how the overriding lithosphere is built in a retreating subduction system. We focus on the recent studies by
Royden and Papanikolaou [2011] and Tirel et al. [2013], as they both employ the tectonic
139

scenario described in the previous section, and, as we will see, make several predictions that are supported by our seismic constraints on the lithospheric structure along the
WHSZ. We will start at the largest-scale, with the behavior of the slab, as its buoyancy is the main driver of the subduction system in both models. Then we’ll describe how the overriding crust is built and deformed as the slab retreats, with emphasis on the modeling of Tirel et al. [2013], as they make specific predictions regarding the deformation structures along which continental crust detaches from the slab, and how these structures relate to the burial and exhumation of HP rocks.
Both the geodynamic modeling studies of Royden and Papanikolaou [2011] and
Tirel et al. [2013] invoke the subduction of a single slab comprised of alternating oceanic and continental domains, with the length of each domain derived from geologic considerations (see Royden and Papanikolaou [2011]). Their models show that slab buoyancy primarily controls trench retreat rates and the dip of the slab, with the subduction of oceanic portions of the slab producing faster rates and shallower dips compared to continental portions of the slab (i.e. Royden and Husson [2006]). This dependence between slab buoyancy and retreat rate is consistent with our seismic imaging results from Chapter 2, in that we find evidence for faster slab retreat along southern Greece, where we observe an oceanic slab subducted well into the mantle, relative to the continental slab subducting to the north (see Figure 2.9; discussion in
Section 2.7.3 and Appendix A).
The detachment of crustal material from the top of the slab, particularly from continental domains, strongly influences both the rate of subduction and the style of overriding lithosphere deformation. Royden and Papanikolaou [2011] were the first to demonstrate the importance of crustal detachment through its effect on subduction rates, as their models require the removal of ~10 km of sediments and continental crust from the top of the External Hellenides portion of the slab in order to match the present-day subduction rates along northern Greece. This prediction is consistent with our seismic images along northern Greece, which show that ~10 km of material has been scrapped off the subducting continental slab, prior to its subduction into the mantle (see Figure 2.5 and
Section 2.7.2).
140

The thermomechanical modeling of Tirel et al. [2013] captures the process of crustal detachment and exhumation, by explicitly incorporating an upper and lower crust for continental domains, and a deep-water sedimentary cover and thin crust for oceanic domains, both of which are then tracked in great detail as the subduction system evolves through time. They find continental blocks detach in the subduction channel and accrete to form the bulk of the overriding crust, while oceanic sediments also detach to form thin, suture zones sandwiched between continental blocks, a deformation pattern that closely matches the tectonic history of the WHSZ described in Section 4.2. Here, we focus on two details of their model, the first being how a single continental block is accreted and exhumed to produce new overriding crust, and the second being how the alternating subduction of multiple continental and oceanic domains influences the overall pattern of overriding lithosphere deformation.
As the first continental block, the one analogous to the Internal Hellenides domain, enters the trench, it proceeds through a three-step, subduction-exhumation cycle that mimics a “caterpillar-walk” [Tirel et al., 2013], with each step marking a different stage in the decoupling of the block from the slab (Figure 4.1c). The first step involves the front edge of the block (i.e. the side that is last to subduct) thrusting beneath the rear of the block (i.e. the side that subducted first), leading to a progressive decoupling of the rear of the block from the slab (see “1” in Figure 4.1c). The next step begins once the rear of the block has been completely decoupled from the slab, causing it to bulge upward and extend by reactivating the oceanic suture zone formed from the subduction of the preceding oceanic domain (see “2” in Figure 4.1c).
The final step begins once the entire block has decoupled from the slab, having had sediments from the next oceanic domain completely thrust beneath it. At this point, the block has been accreted to form new overriding crust, which is now free to extend, or
“collapse”, to fill in the space created as the slab rapidly retreats to consume the next oceanic domain (see “3” in Figure 4.1c). The resulting, post-collapse structure exhibits an asymmetric “V” shape, with most of the upper crust piled up on the trenchward side, while the seaward side forms a “dome-like” structure that exposes the lower crust at its apex.
141

When the second continental block enters the subduction system, the one analogous to the External Hellenides domain, it undergoes the same type of “caterpillarwalk”, subduction-exhumation cycle that the first continental block experienced [Tirel et al., 2013]. The increase in slab buoyancy causes the slab to steepen again, which triggers an influx of asthenospheric material that heats up the now accreted first block, thermally weakening it. This process leads to further extension of the accreted first block (see
Internal Hellenides in Figure 4.3d), thereby producing a “hot” metamorphic core complex
[Tirel et al., 2013]. Embedded markers within the second continental block yield pressure-temperature-time (P-T-t) paths that are remarkably similar to those found for the
HP rocks exposed along southern Greece (see Figure 2c in Tirel et al. [2013]), with the heating event described above corresponding to the isobaric heating of the Tinos HP rocks (Figure 4.1b).
The detailed crustal structures predicted by the modeling of Tirel et al. [2013] bear striking resemblance to the geometry of the discontinuities in our RF images from
Chapter 3, including the asymmetric “V” shaped patterns and the dome-like structures.
Thus, we next set out to further refine our structural interpretation of the RF images, integrating the available geologic, geodynamic, and seismic constraints into a detailed picture of how the overriding crust is built and deformed in a retreating subduction setting, such as the WHSZ.
4.4 Integrated structural interpretation
Our RF profiles across southern and northern Greece sample two different stages of corecomplex development, that of a partially collapsed overriding crust along southern
Greece, where oceanic lithosphere is currently subducting, and that of a building overriding crust along northern Greece, where continental lithosphere is subducting (see location of seismic profiles in Figure 4.1a). So, in essence, looking from south to north at our two profiles across the WHSZ is like looking back in time at a core complex in the making, while looking from the trench to the backarc along a single RF profile is like looking back in time at up to four of the tectonic events that have occurred along the
WHSZ: (1) the Mediterannean oceanic suture (presently forming in the south only), (2)
142

the collapsing (south) or accreting (north) External Hellenides continental blocks, (3) the
Pindos oceanic suture, and (4) the collapsed Internal Hellenides continental block.
Our goal in this section is to interpret the structure of the overriding crust along the WHSZ in terms of the four tectonic events described above. To do this, we focus on identifying the location of the Pindos suture zone within the upper crust. Once we know where the Pindos is located, we can separate the upper crust into External and Internal domains, and then look for the deeper reaches of the Pindos suture zone. The RF profile along southern Greece is interpreted first, as it includes the structure of the presently subducting Mediterranean oceanic domain, a valuable analogue for the Pindos domain, plus it samples a more advanced stage of core complex development compared to the north. Along the way, we use the model-based structural interpretation from Tirel et al.
[2013] to help guide our seismic-based interpretation.
Figure 4.2 shows the RF profile across southern Greece along with our structural interpretation and the corresponding model-based schematic of a collapsing overriding crust from Tirel et al. [2013]. The most recent tectonic event, the rapid subduction of the
Mediterranean oceanic slab, is identified by a shallow dipping (~17°), positive discontinuity, or sharp increase in velocity with depth, marking the Moho of the
Mediterranean oceanic slab (Figure 4.2c; also see high-resolution 2-D GRT images in
Figure 2.3, and discussion in Sections 2.7.1 and 3.6.1). The slab Moho is paired with an overlying, sub-parallel negative discontinuity (i.e. sharp decreases in velocity with depth) in several places, producing a thin (8-10 km), apparently segmented low-velocity layer
(LVL). This thin, LVL signal corresponds to the subduction channel of the presently subducting Mediterranean oceanic domain, with its deeper part (~60 km depth in Figure
4.2c) reaching typical peak pressures (~18 kbar) experienced by the HP rocks now exposed at the surface in the Cycladic unit (Figure 4.1). Our structural interpretation of the subducting Mediterranean oceanic domain provides an important reference structure for subsequent interpretation of the older, Pindos oceanic suture, plus it gives us a
143

common “reference” point to align our seismic-based structures to those of Tirel et al.
[2013] (Figure 4.2c,d).
Both the tectonic history and geodynamic modeling of the WHSZ (Sections 4.2 and 4.3, respectively) indicate that accreted continental blocks of the External Hellenides form the overriding crust above the presently subducting Mediterannean oceanic domain.
The question is how far landward of the trench do continental blocks of the External domain extend? The rear boundary of the External Hellenides is defined by the Pindos oceanic suture, a feature that is observed at the surface along our RF profile at X = ~200 km (Figure 4.2a). The extension of the Pindos suture from the surface to upper crustal depths (<~10 km) is interpreted to occur where there is a clear surface “breach” in the basement top (i.e. X = ~215 km in Figure 4.2c), as defined by (1) the apex of a broad dome-like structure in the basement top, (2) an abrupt vertical offset of ~4 km in the basement top, and (3) a sudden reduction in the amplitude of the basement top signal from greater than 6% away from the “breach” to less than 2% within it. Remarkably, the
Pindos suture zone geometry predicted in the geodynamic model of Tirel et al. [2013] exhibits these same characteristics (compare region labeled “P” in Figure 4.2c,d). With the location of the Pindos suture zone in hand, we can now divide the upper crust into either the External Hellenides domain or Internal Hellenides domain, depending on whether the crust is either trenchward or landward of the Pindos suture zone, respectively
(i.e. “E” versus “I” in Figure 4.2c).
To get a better handle on the depth extent of the Pindos suture, we first examine the structural pattern resulting from the External Hellenides domain collapsing over the
Mediterannean oceanic domain, and then we look for this same pattern within the collapsed Internal Hellenides domain. The basement top along the External Hellenides domain forms an asymmetric “V” shaped pattern from X = 80 km to X = 200 km (Figure
4.2c). The shorter, trenchward side of the “V” is underlain by several weaker-amplitude
(< 6%) discontinuities between a depth of 15 km and 30 km, followed by a higheramplitude (>6%) LVL signal below ~35 km depth. A similar structural pattern is produced in the geodynamic model of Tirel et al. [2013], following the collapse of the
External Hellenides domain over the Mediterranean oceanic domain (compare Figure
4.2c with 4.2d). Thus, we interpret the weak-amplitude discontinuities in our RF profile
144

as the signature of upper crustal slices that are “piled up” towards the trench, following their accretion, and subsequent collapse, over what is now a deeply (>35 km) subducted portion of the Mediterranean oceanic domain.
Next, we look for a similar structural “fingerprint” along the trenchward side of the Internal Hellenides domain, the idea being that its accreted continental block(s) collapsed over the Pindos oceanic domain in much the same way as the External
Hellenides continental blocks have collapsed over the Mediterranean oceanic domain.
Our RF profile supports this assertion, as we find a structural pattern for the Internal
Hellenides that is remarkably similar to that of the External Hellenides. Specifically, the
Internal Hellenides exhibits the following characteristics: (1) the basement top forms an asymmetric “V” shape from X = 220 km to X = 320 km, (2) the trenchward side of the
“V” is underlain be weaker-amplitude discontinuities between ~10 km and 20 km depth, and (3) there is a deeper, high-amplitude (>6%) negative discontinuity at 20-25 km depth that appears to gently dip (~20°) in the same approximate direction as the presently subducting Mediterranean slab (Figure 4.2c). This similarity in structural pattern between the External and Internal domains is even present when one compares the receiver function signals from individual seismic stations located above the collapsed crust of each domain (see RF stack for stations S002 versus station S018 in Figure 4.2c).
The geodynamic model of Tirel et al. [2013] also predicts a similar structural pattern for the collapsed crust of the External and Internal Hellenides (Figure 4.2c,d), leading us once again to interpret the weaker-amplitude discontinuities as upper crustal slices piled up on the trenchward side of the Internal Hellenides domain, following their accretion, and subsequent collapse, over a deeper portion of the Pindos oceanic domain. The signal from the Moho of the Pindos slab is no longer present (it being subducted away in the
Eocene), leaving behind subduction channel material in the form of an oceanic suture zone with its top defined by the negative discontinuity at 20-25 km depth, as shown in
Figure 4.2c.
The Pindos suture does not appear to continue into the upper mantle, given our
RF profile shows a continuous, albeit “warped”, overriding Moho beneath where it would extrapolate to (Figure 4.2c). However, it must have extended to mantle depths in the past
(pre-Oligocene) in order to produce HP rocks with peak-pressures of 18 kbar (~60 km
145

depth), such as those of the Cycladic unit (Figure 4.1b). This raises some interesting questions as to how the overriding lithosphere, and its Moho, develops in a retreating subduction system, something we touch on further in Section 4.5.3.
Figure 4.3 shows the RF profile along northern Greece along with our seismic-based structural interpretation and the corresponding model-based schematic of a building overriding crust from Tirel et al. [2013]. The first tectonic event observed here is the subduction of the External Hellenides continental domain, which is identified by the shallow dipping (17°), positive discontinuity marking the Moho of the subducting continental slab (Figure 4.3c; also see high-resolution 2-D GRT images in Figure 2.5, and discussion in Sections 2.7.2 and 3.6.1). The top of the subducting crust is associated with a weak-amplitude positive discontinuity ~20 km above the slab Moho, yielding a position and thickness for the subducting crust that connects smoothly with the foreland crust observed in marine seismic data [Finetti and Del Ben, 2005] (see Figure 2.9). The top of the subducting crust is indistinguishable from the top of the overriding crust for X < ~130 km (Figure 4.3c). Just landward of here (i.e. X = 130 km), the basement top reduces its amplitude and abruptly “kinks” upward, leading us to interpret this as the approximate location of the crustal backstop (Figure 4.3c; see further discussion in Section 3.6.8). The structural pattern observed in this portion of the RF profile (X < 220 km in Figure 4.3c) is quite similar to the model-predicted structure that is produced when a continental block
“buckles” as it undergoes the first two steps in the “caterpillar walk” subductionexhumation cycle proposed by Tirel et al. [2013] (Figure 4.3d; “1” and “2” in Figure
4.1c; see details in Section 4.3).
Next, we look for the Pindos suture zone that defines the rear boundary of the
External Hellenides domain. Its surface location along our RF profile is at X = ~230 km
(Figure 4.3a), but its extension to upper crustal depths is more difficult to identify compared to southern Greece, because the External Hellenides domain has yet to collapse here, leaving the basement top relatively intact and flat-lying from the crustal backstop to the surface expression of the Pindos suture (i.e. X = 130 km to 230 km in Figure 4.3c).
146

So, we take a more conservative approach by interpreting the upper crustal location of the
Pindos suture zone to be somewhere between X = 240 km and 260 km, as this region along the basement top (1) is landward of the surface expression of the Pindos suture (X
> 230 km), (2) reduces its amplitude to ~2% at X = ~260 km, though this disruption in signal is much less pronounced than along southern Greece (see X = 210 km in Figure
4.2c), and (3) is just trenchward of the asymmetric “V” shape associated with the collapsed Internal Hellenides block (X = 260 km to 420 km in Figure 4.3c). The geodynamic model of Tirel et al. [2013] predicts that similar structural characteristics develop around the Pindos suture zone during the accretion of the External Hellenides continental blocks (Figure 4.3d). In particular, the relatively weak amplitude of the basement top signal at X = 260 km (Figure 4.3c) may be due to the fact that there has been minimal reactivation of the Pindos suture zone along northern Greece, such that the
External and Internal blocks are still “stuck together”, without exposing the deeper portions of the External block (Figure 4.3d).
It is more difficult to identify the detailed structure of the collapsed Internal domain along northern Greece, as compared to southern Greece, particularly the weakamplitude discontinuities that we attribute to upper crustal slices piled up towards the trench (compare X = 290 km in Figure 4.3c versus X = 260 km in Figure 4.2c).
However, this may not be related to differences in the actual structure of the Internal
Hellenides beneath these two regions, but instead, may result from a reduction in the high-frequency content of the RF data along northern Greece relative to southern Greece.
It is likely that overlapping discontinuities are also present near the deepest portion of the
Internal Hellenides “V”, but they have been “blurred” together to produce the very thick
(~15 km) positive peak observed at X = ~290 km in Figure 4.3c. Detailed modeling of the RF waveforms is needed to further address this issue.
The Pindos suture zone extends from the upper crust through the overriding
Moho, as indicated by the coherent negative discontinuity that disrupts the overriding
Moho signal from X = ~260 km to X = ~300 km (Figure 4.3c). This suggests that the separation between the continental blocks of the External and Internal domains extends into the upper mantle. However, it’s not clear what this disruption in the overriding
147

Moho signifies, or how deep it extends into the upper mantle, as discussed further in
Section 4.5.3.
On a final note, there is evidence that the Vardar suture zone is present along the landward edge of the northern RF profile, near the apex of a broad dome-like feature in the basement top (X = ~425 in Figure 4.3c), a structural characteristic that is similar to what is observed for the reactivated Pindos suture along southern Greece (X = ~210 km in Figure 4.2c), and for the model-based Vardar suture zone predicted by Tirel et al
[2013] (Figure 4.3d).
4.5 Discussion
Here, we briefly discuss the implications of our integrated structural interpretation of the crustal structure along the WHSZ by highlighting a few examples in which our combined geologic, geodynamic, and seismic constraints don’t seem to fit together as well.
Specifically, we address the three-dimensional variations in the location of the Pindos suture along southern Greece, the width of the External domain along northern Greece, and how the overriding lithosphere deforms during slab retreat.
Our RF profile across southern Greece intersects the surface expression of the Pindos suture zone twice, once at X = ~200 km where the Pindos is approximately parallel to the southern trench, and a second time at X = ~320 km where the Pindos is approximately perpendicular to the southern trench (Figure 4.2c). This indicates that the location of the
Pindos suture zone varies substantially in three-dimensions. The first intersection between our profile and the Pindos displays a structural pattern that is remarkably consistent with geologic, geodynamic, and seismic observations, as described in Section
4.4.1.
The second crossing of the Pindos suture at X = ~320 km is more difficult to interpret, as it is observable in the surface geology (Figure 4.1) and our RF profile
(Figure 4.3c), but it is not predicted by the geodynamic modeling (Figure 4.3d). One possible explanation for this discrepancy is that this second Pindos crossing is related to a
148

trench-perpendicular boundary of an accreted continental block. This could explain why it coincides with (1) a sharp transition in the metamorphic grade of the exposed thrust sheets, with lower-grade Pelagonian rocks to the northwest and higher-grade Cycladic rocks to the southeast (see dotted line in Figure 4.1a) and (2) an abrupt shallowing of the basement top at X = ~325 km, as it transitions from lower- to higher-grade rocks (Figure
4.3c). Our trench-parallel RF profiles presented in Chapter 3 (Figure 3.10; Section 3.5.3) also support this hypothesis, as they contain the same structural “fingerprint” of accreted crustal blocks that we observe in the trench-perpendicular profiles (e.g. Figure 4.3c), including trench-parallel asymmetric “V” shaped patterns (e.g. -100 < X < 50 in Figure
3.10d,e) and weaker amplitude, overlapping discontinuities in the mid to lower crust (e.g.
Y < -20 in Figure 3.10d,e). Thus, there is evidence in the geologic and seismic observations that the location of the Pindos oceanic suture, and possibly other prominent geologic boundaries, varies substantially in three-dimensions, which may, at least in some cases, be related to the trench-perpendicular edges of accreted blocks, an impetus for the development of three-dimensional geodynamic models of the WHSZ.
Accreted continental blocks of the External Hellenides domain extend from the crustal backstop to the Pindos suture in our RF profile along northern Greece (X = 130 km to X
= 250 km in Figure 4.3c), a width of ~120 km. There also appears to be a disruption in the structural pattern of the overriding crust at X = ~200 km, separating the External domain into two distinct accreted blocks, each having a similar sequence of positivenegative-positive discontinuities (i.e. left and right of thin dotted line in Figure 4.3c).
Thus, there appear to be three distinct parts of the overriding crust in our RF profile that may each be related to a different continental block of the External domain: (1) the Paxos block is associated with the presently subducting continental crust dipping at ~17° from
X = 100 to X = 180 km (i.e. from seaward of the crustal backstop to the region between the thick and thin solid lines in Figure 4.3c), (2) the Ionian block occupies the region landward of the crustal backstop (X = 130 km) up to the disruption at X = 200 km (below the thin dotted line in Figure 4.3c), and finally (3) the Gavrovo Tripolis block occupies
149

the region landward of the disruption at X = 200 km, up to the Pindos suture X = ~250 km (below the dashed-dotted line labeled “P” in Figure 4.3c), though its not clear exactly where this last block stops and the Pindos suture starts.
There are two notable differences between our seismic-based structural interpretation of the External domain, and the model-based interpretation of Tirel et al.
[2013]: (1) the External accreted blocks extend over a distance of ~120 km in our RF profile (not counting the presently subducting Paxos block), but they are only approximately 80 km wide in the model of Tirel et al. [2013] (compare region between arrows in Figure 4.3c,d), and (2) there is evidence of two previously accreted continental blocks in our RF profile, while the model of Tirel et al. [2013] only contains one accreted block (again, excluding the Paxos block; see Figure 4.3c,d). Both of these differences may be explained by an External domain that is wider than the ~400 km wide domain employed by Tirel et al. [2013]. This explanation is also consistent with the conclusion of Royden and Papanikolaou [2011] that the External domain must have been at least
~500 km wide in order to match the available data along northern Greece (e.g. subduction rate, timing; see further discussion in their Section 3.2.2.1). A wider domain may also require more than one “buckling” event, where one part of the continental domain thrusts beneath another, to completely decouple it from the slab (e.g. Figure 4.1c; see Section
4.3), which could explain the extra accreted (Gavrovo Tripolis) block observed in our RF profile. The occurrence of a second crustal buckling event may also be influenced by inherited variations in the crustal structure of the External domain (e.g. crustal thickness, weak zones), particularly given the Ionian block is believed to have been comprised of thinner crust (i.e. its sediment cover was deposited in more moderate water depths) relative to the Paxos and Gavrovo Tripolis units (Figure 4.1; Royden and Papanikolaou,
[2011]; Burchfiel, personal communication).
We caution that there are a number of issues that would need to be sorted out prior to performing a quantitative comparison between our seismic-based structural interpretation and the model-based interpretation of Tirel et al. [2013]. For one, we would need to get a better handle on the size of each block, particularly its thickness, through modeling of the RF waveforms, along with other geophysical methods. Block size is important for several reasons. One reason is that there is an inherent tradeoff
150

between domain width and crustal thickness [Royden and Papanikolaou, 2011]. For example, the two geodynamic models considered here use markedly different External domain widths (400 km vs. 700 km, a difference of ~40%), but they also use very different crustal thicknesses (44 km in Tirel et al. [2013] vs. ~30 km in Royden and
Papanikolaou [2011]), such that the two models are much closer to each other in terms of the total crustal volume (~15%). Obtaining the thickness, as well as the width in both the trench-parallel and perpendicular dimensions (see Section 4.5.1) would give us a much better tool for quantitatively reconstructing the pre-subduction External domain via a rigorous mass-balance approach (i.e. Burchfiel et al., in prep.). This would also tell us a great deal about the rheology of the continental lithosphere, another vital, but difficult to constrain, aspect of the geodynamic models.
The overriding Moho is sharp and continuous where it intersects the extension of the
Pindos suture beneath southern Greece (270 km < X < 290 km in Figure 4.2c), while it appears to be “broken” where it intersects the Pindos suture beneath northern Greece
(280 km < X < 300 km in Figure 4.3c). We can gain some insight into how the northern, broken Moho “heals” to form the southern, continuous Moho by considering the presently subducting Mediterranean oceanic domain (Figure 4.2) as an analogue for what the Pindos suture zone looked like during the Eocene. Following this idea, the present start of the overriding Moho above the Mediterranean slab (X = 190 km in Figure 4.2c; see discussion in Section 3.6.6) may be analogous to the start of the “break” along northern Greece (X = ~310 km in Figure 4.3c), which would imply that this break has persisted since the Eocene. This explanation is also supported by the observation that some HP rocks (e.g. Crete, Peloponnesus) have exhumed from mantle depths along the
Pindos suture concurrent with ongoing continental subduction (see Figure 4.1b; Section
4.2; Jolivet and Brun [2010]). The fact that the overriding Moho is now continuous along southern Greece suggests that Moho “healing” transpired after Mediterranean oceanic lithosphere began to enter the subduction system; thus implicating a sudden increase in overriding lithosphere extension and/or arc magmatism as the most likely “healing”
151

mechanism(s). Additional insight into the mechanism of healing may be gained through
(1) detailed study of the velocity structure around the overriding Moho “break” beneath northern Greece, and how this structure compares to the “healed” Moho along southern
Greece, and (2) comparing this refined structure to three-dimensional geodynamic models that capture how the healing process unfolds.
152

4.6 Figures
Figure 4.1: Geologic map, Pressure-Temperature-time (P-T-t) path for Tinos High-
Pressure (HP) rocks, and schematic illustrating “Caterpillar” walk of subductionexhumation cycle from Tirel et al. [2013]. a) Major geologic units along the WHSZ, modified after Burchfiel et al. (in prep.). Thick black lines with barbs on upthrown side delineate the Pindos and Vardar suture zones (i.e. their contact with the basal thrust of the
Pelagonian and Rhodope units, respectively). All units south of the Pindos suture are part of the External Hellenides (e.g. Gavrovo-Tripolis), while units between the Pindos and
Vardar sutures (i.e. Pelagonian) form the Internal Hellenides. Dotted line extending from the Pindos suture marks a sharp transition in metamorphic grade, coincident with normal faulting across Attica and Evia, following Royden and Papanikolaou [2011] (see their
Figure 6). Tinos HP rocks (cyan star labeled T) are part of the Cycladic unit (dark purple), which is predominately composed of blueschist-grade metamorphic rocks.
Dashed black lines identify RF profile locations along northern and southern Greece.
Geographic regions of Attica and Evia identified by arrows labeled At and Ev, respectively. b) P-T-t paths for Tinos HP rocks modified after Parra et al. [2002] and
Tirel et al. [2013]. Blue portion of path follows “colder” geotherm representing exhumation in forearc subduction channel, red portion of path denotes isobaric heating event, and orange portion of path follows “hot” geotherm representing exhumation via reactivation of subduction channel in back-arc setting (see Figure 15 of Parra et al.
[2002]). c) Schematic illustrating three stages of continental block subductionexhumation cycle from Figure 3 in Tirel et al. [2013] (see detailed description in Section
4.2).
153

Figure 4.2: Integrated structural interpretation of overriding lithosphere during Hellenides
“collapse” along southern Greece. a) Topographic map showing location of RF profile
(dashed black line at Y=-10 km), Pindos suture (solid black line with barbs on upthrown side), seismic stations (black triangles), and Plio-Quaternary arc volcanism (red triangles). b) Profile resolution with the colorbar denoting the # of conversion points within each CCP bin (blue is zero). c) Interpreted RF profile, as in Figure 3.6, with structural features defined as follows: slab Moho (solid black line), overriding Moho
(dashed black line), basement top (dotted black line), and subsurface extension of the
Pindos suture (dashed-dotted black line labeled “P”). Horizontal bars separate the top of crystalline basement based into either the External Hellenides (E) or the Internal
Hellenides (I) domain, separated by the Pindos suture (P). The colorscale represents the average RF amplitude as a % of the incident P-wave amplitude, with red and blue saturated at +6% and -6%, respectively. RF stacks for stations S002 and S018 are shown to the right, with the same y-axis as the RF profile, “M” identifying the positive discontinuity from the Mediterranean slab Moho, and “P” identifying the negative discontinuity from the Pindos suture. d) Structural interpretation of Hellenides “collapse” from the geodynamic model of Tirel et al. [2013]. Color denotes “rock” type: dark yellow for Ionian oceanic metasediments; dark green overlain by gray-green for External
Hellenides lower and upper crust, respectively; light yellow for Pindos oceanic
154

metasediments; dark purple overlain by gray-purple for Internal Hellenides lower and upper crust, respectively. Black lines denote zones of localized deformation (e.g. faults), with arrows showing slip directions (all normal in this case). Thick dashed line has been added to highlight the top of the overriding lithosphere (i.e. overriding Moho). Note that the vertical axis has been reduced by 70% (i.e. flattened) relative to the original panel in
Figure 3 of Tirel et al. [2013] in order to match the slab dips in (c) and (d).
Figure 4.3: Integrated interpretation of overriding lithosphere structure during Hellenides
“building” along northern Greece, with panels following the same layout as Figure 4.2.
The arrow labeled “CB” shows the approximate location of the crustal backstop, the thin, solid black line denotes the top of the presently subducting Paxos continental block, and the thin dotted line denotes inferred disruption in the overriding crust between the Ionian and Gavrovo Tripolis blocks. Panel (c) includes our tentative interpretation of the more internal Vardar oceanic domain (“V?”), while panel (d) shows the model-predicted
Vardar oceanic (yellow labeled V in d)) and Rhodope continental domains (lime green labeled R in d)). Note that reverse slip (i.e. compressive) predominates in the External
Hellenides (green), while normal slip (extensional) occurs in the Internal Hellenides
(purple).
155

156

In this dissertation, I investigated the lithospheric structure of the western Hellenic subduction zone (WHSZ) using two complimentary seismic imaging techniques, integrating the resulting structural constraints with geologic data and geodynamic models to characterize the dynamic behavior of this retreating subduction system. The twodimensional, high-resolution profiles obtained from the 2-D GRT method were used to constrain the slab composition beneath southern and northern Greece, and to estimate the amount of differential slab retreat between the southern and northern portions of the slab.
The three-dimensional receiver function model derived from the 3-D CCP method was used to construct maps of the slab and overriding Moho, and to identify the top of the crystalline basement across northern and southern Greece. We then performed an integrated interpretation of two receiver function profiles across northern and southern
Greece, combining the available geologic, geodynamic, and seismic constraints into a detailed picture of a developing core-complex along the WHSZ. In the following we briefly summarize the key findings from each chapter.
Chapter 2 presented high-resolution, 2-D GRT images across the southern and northern portions of the western Hellenic subduction zone, leading to several new constraints on the slab composition, mantle wedge geometry, and dynamics of slab retreat. These constraints are of particular importance for the northern portion of the
WHSZ, where previous studies have been sparse, at best. We found subducted crust extending well into the mantle (> 70 km depth) beneath both southern and northern
Greece, with each crust having a similar strike (approximately N30W +/- 10°) and dip
(~17°). The thin (~8 km thick) subducted crust imaged beneath southern Greece marks
157

the top of the Ionian (Sea) oceanic slab. Its thickness indicates that most of the overlying sedimentary cover observed within the foreland has been accreted to the overriding lithosphere prior to its subduction into the mantle (depths < ~50 km). The relatively thick
(~20 km) crust imaged beneath northern Greece marks the top of the Apulian continental slab. Its crustal thickness and position are consistent with constraints on the foreland continental crust from marine seismic data, provided that the top ~10 km of the subducting continental crust has been accreted to the overriding lithosphere. A comparison of the mantle wedge geometries beneath southern and northern Greece indicated that the southern, oceanic slab has retreated an additional 70-85 km relative to the northern, continental slab, an estimate of differential slab retreat that is consistent with the offset across the Cephalonia transform fault, and with the relationship between slab buoyancy and retreat rate predicted in the recent geodynamic model of Royden and
Papanikolaou [2011].
In Chapter 3, we developed a three-dimensional receiver function imaging method, applied it to teleseismic data from seismic stations distributed across the WHSZ, and then used the resulting model to map the location of the slab Moho, the overriding
Moho, and the top of the crystalline basement. Our map of slab Moho depth showed a continuous slab smoothly ramping from northern to southern Greece, though there is evidence of significant depth fluctuations around the Gulf of Corinth. Our map of overriding Moho depth is broadly consistent with previous studies, with the following exceptions. We observed short-wavelength fluctuations in Moho depth within an approximately N-S oriented corridor from the volcanic arc to the Gulf of Thermaikos, a feature that roughly tracks disruptions in the signal of the overriding Moho. The first emergence of a well-developed overriding Moho signal only occurs after the slab has reached mantle depths (~70 km). This yields an asymmetric geometry for the overriding
Moho, with its starting point occurring much further landward than found in previous studies. The overriding Moho is deepest below the peak topography of the Hellenides, and is generally deeper than would be expected from isostasy alone, further evidence that slab dynamics play a prominent role in shaping the overriding lithosphere.
The top of the crystalline basement is present across much of the WHSZ at an average depth of ~8 km, indicating that the bulk of the overriding crust is comprised of
158

(at least) two layers, with low-velocity sediments overlying high-velocity crystalline rocks. The offset in the basement top between southern and northern Greece suggests that there has been a much larger amount of differential basement retreat (~180 km) relative to the amount of differential slab retreat (~70 km) found in Chapter 2. The basement top also exhibits several asymmetric “V” shaped patterns, with the most prominent ones observed along southern Greece, a feature of particular importance for
Chapter 4.
Chapter 4 focused on refining our structural interpretation of the overriding crust beneath southern and northern Greece, integrating our seismic-based constraints from the previous two chapters with geologic and geodynamic-based constraints that describe how the overriding crust is built and deformed during slab retreat. Along southern Greece, the asymmetric “V” shaped pattern in the basement top above the presently subducting oceanic slab is interpreted as the structural fingerprint of the rapidly extending, or collapsing, External Hellenides continental blocks, with the weak amplitude, positive discontinuities on its trenchward side marking piled up slices of upper crust, and the surface breach in the basement top on its landward side associated with the reactivated
Pindos oceanic suture zone, structural features that are remarkable similar to those predicted in the recent geodynamic model of Tirel et al. [2013]. Landward of the Pindos suture, we observe another asymmetric “V” in the basement top, which is interpreted, in a similar manner, as the signature of the External Hellenides continental block that collapsed during the Eocene over a deeper extension of the Pindos oceanic suture zone.
Thus, Pindos oceanic suture zone is found to extend from the surface to lower crustal depths, splitting the overriding crust into two distinct continental domains, the External
Hellenides on its trenchward side and the Internal Hellenides on its landward side.
Along northern Greece, our RF profile captures the active building of the overriding crust, with the presently subducting Apulian continental slab underthrusting previously accreted continental blocks of the External Hellenides domain. The basement top above the slab is relatively flat and continuous, with no indication of a surface breach, as expected given the overriding crust here has not yet begun to rapidly extend (i.e. collapse). Just landward of the surface expression of the Pindos oceanic suture, there is another pronounced asymmetric “V” shaped pattern that we have interpreted as marking
159

the top of the collapsed Internal Hellenides continental block. By analogy with the structure interpreted along southern Greece, the trenchward edge of the Internal
Hellenides block is interpreted as the location of the Pindos suture, which appears to extend all the way into the upper mantle, following a pronounced negative discontinuity that clearly disrupts the overriding Moho. Our integrated interpretation of the available geologic, geodynamic, and seismic constraints provided a detailed picture of the overriding crust in two different stages of core complex development along the WHSZ, with the crust along southern Greece actively collapsing in response to the recent transition from continental to oceanic slab subduction, while the crust along northern
Greece is still in a building mode, accreting continental blocks of the External Hellenides domain.
The multidisciplinary study described in this dissertation demonstrates that integrating structural constraints from teleseismic imaging methods, geologic observations, and geodynamic modeling provides a powerful tool for capturing the dynamic behavior of a retreating subduction system. It is our belief that having the ability to test geodynamic models, not just with the geologic observations from the surface, but also with detailed structural constraints down to lithospheric depths, will allow us to answer many outstanding questions regarding the dynamic behavior of plate boundaries, both those active today, and those recorded in the geologic units that form continents.
We also note that the multidisciplinary framework used here should be expanded in the future to include constraints from additional disciplines. Two such disciplines jump to mind: petrology and rheology. For example, thermobarometric models that constrain the pressure-temperature-time paths of high-pressure rocks would be a valuable addition, as these rocks must be present at depth within our images, and so, if their location can be accurately identified, they could be used to constrain temperature, thereby providing a means of testing thermal models. Rheological models for the continental lithosphere are also of paramount importance, as they must strongly influence the dynamics of retreating subduction systems (e.g. the size of the continental blocks, the depth they detach from the slab), and many other systems that are critical for informing our understanding of the dynamic Earth we all live on.
160

161

A.1 Introduction
This appendix provides additional details regarding the data and a priori model parameters used to form the SL and NL composite images. It also describes the procedure used to estimate the differential slab retreat between SL and NL.
A.2 Teleseismic events and scattering modes
The teleseismic events and selected scattering modes used to form the SL and NL composite images are described in Table A.1 and Table A.2, respectively. Event selection was based on the criteria described in Section 2.4 using catalog information from the Preliminary Determination of Epicenters (PDE) Bulletin (downloaded from ftp://hazards.cr.usgs.gov/pde/ ). Individual scattering modes were not included in the simultaneous inversion if they had 1) anomalously high peak amplitudes (a factor of two or more greater than other events), 2) strong migration artifacts such as "migration smiles" or operator aliasing [Rondenay et al., 2005], or 3) strong cross-mode contamination.
A.3 A priori model parameters
This section describes the procedure used to determine optimal values of the 2-D regional strike and the background velocity model for the migration of the SL and NL datasets.
162

The regional strike of the dipping low velocity layer (LVL) is determined by applying the
2-D GRT inversion using different profile orientations, or projection lines. The optimal projection line produces the most coherent, high-amplitude image of the subducted crust
[Rondenay, 2010]. The SL and NL composite images formed using a range of projection lines are shown in Figure A.1. For both profiles, the most coherent seismic signal from the subducted crust is obtained for a projection line of N60E to within +/- 10°, which is the optimal value used for all composite images produced in this study.
The background velocity model is determined using a two-step process. First, a 1-D Pwave velocity model for NL and SL is determined by averaging the P-wave velocities from local tomographic models at 20 km depth intervals beneath each line [Papazachos et al., 1995; Papazachos and Nolet, 1997]. This averaging procedure yields the 1-D P-wave velocity models for SL and NL described in Tables 2.1 and 2.2, respectively. Then, we estimate the optimal S-wave velocities, as these are not as well constrained in previous studies, particularly for northern Greece. An initial S-wave model is computed by assuming a Vp/Vs ratio of 1.77 for each layer. The individual scattering mode images formed using this initial model show differences in the location of the dipping LVL and overriding Moho. These differences are minimized when using a Vp/Vs ratio of approximately 1.8 from 20 to 60 km so this value is used to define the optimal S-wave velocities in this depth range. The individual scattering mode images for SL and NL formed using their respective optimal velocity models in Tables 2.1 and 2.2, are shown in
Figure A.2. For both NL and SL, the LVL in each scattering mode image is mapped to a consistent position except for the Pds mode for SL, which appears to be deeper and have a steeper dip as discussed in detail by Suckale et al. [2009]. Such deviations are likely related to velocity heterogeneity and/or anisotropy beneath the SL array that is not adequately represented by the optimal background velocity model in Table 2.1. PdS scattered waves only travel once through the velocity structure so will not average out
163

local velocity structure as much as the reverberated modes. Furthermore, only PdS modes generated by events traveling up the dip of the subducted crust are included in the composite image (Table A.1) so they may be more sensitive to lower velocities and/or a particular anisotropic fabric beneath the Peloponnesus and the Gulf of Corinth.
A.4 Differential slab retreat analysis
The amount of differential slab retreat between the NL and SL profiles is derived from the subducted crust positions shown in Figure 2.9b. First, we measure the difference in vertical displacement between the two profiles to estimate how much more the southern portion of the slab may have sunk relative to the northern portion. We define the difference in vertical displacement between the southern and northern line profiles, dW(SL-NL), as dW(SL-NL) = [Z(MW,SL) - Z(FL,SL)] - [Z(MW,NL) - Z(FL,NL)] (1) where Z is the vertical position of the slab Moho as measured from the surface in km
(positive downward), and abbreviations in parentheses refer to values of Z in the mantle wedge (MW) versus the foreland (FL) and in the southern line (SL) versus the northern line (NL). Technically, the foreland vertical positions in (1) are for the portion of the slab imaged in the mantle wedge prior to its subduction and not its present-day foreland position. Next, we rearrange (1) so that it contains differences in vertical positions within the foreland and mantle wedge: dW(SL-NL) = [Z(FL,NL) - Z(FL,SL)] - [Z(MW,NL) - Z(MW,SL)] (2) where the first and second terms correspond to the vertical differences in Moho position between the two profiles in the foreland, W(FL), and the upper mantle, W(MW), respectively. The formulation in (2) is preferred as it explicitly separates the vertical positions we can measure directly from our images (i.e. W(MW)) from the vertical positions we must infer based on the thickness of the subducted crusts (i.e. W(FL)).
164

From Figure 2.9b, W(MW) is approximately constant at -10 km owing to the similar dips of the two subducted slabs. We assume the present-day difference in vertical position within the foreland is a reasonable estimate of W(FL) because each image shows a subducted crust that has a thickness similar to the foreland crust (see Section 2.7.1). The
Moho depth of the oceanic lithosphere is approximately 15 km [Finetti et al., 1991; de
Voogd et al., 1992] while that of the continental lithosphere is 25-30 km [Finetti and Del
Ben, 2005], which yields a W(FL) of approximately +10-15 km. Substituting the above values into (2) yields a range for dW(SL-NL) of 20-25 km with positive values denoting larger vertical displacement of the subducted slab beneath SL relative to NL. The amount of differential slab retreat between SL and NL is estimated by calculating the difference in landward horizontal displacement of the slab beneath SL required to undo the 20-25 km difference in vertical displacement. To do so, we make two simplifying assumptions: 1) the portions of the slab were originally aligned and 2) the slab dip has remained constant at approximately 17°. Under these assumptions, the landward horizontal displacement of the slab beneath SL, dU(SL-NL), is dU(SL-NL) = dW(SL-NL)/atan(theta) (3) where theta is the slab dip in radians and atan is the inverse tangent function.
Substituting the previously derived range for dW(SL-NL) into (3) yields a dU(SL-NL) between 70-85 km. As shown in Figure 2.9c of the main text, applying the lower limit of
70 km aligns the two subducted slabs such that their difference in vertical displacement is approximately zero. This simple model allows us to explore the effects of uncertainty in our estimates of the vertical displacements. For example, transitional rather than oceanic subducted crust beneath SL would reduce the estimate of W(FL) leading to a reduction in the amount of differential slab retreat. Conversely, thicker subducted crust beneath NL compared to the present-day foreland crust would increase the amount of differential slab retreat.
165

A.5 Tables
Date Time Lat. Lon.
Bi n
Z M
M t
Pds
Pp dp
Pp ds
Psd s|v
Psd s|h
06/18/06 18:28:02 33.03 -39.7 1 9 6.0 W 0 0 0 1 0
06/22/06 10:53:12 45.42 149.34 2 95 6.1 B 1 1 0 0 0
07/08/06 20:40:01 51.21 -179 31 1 22 6.6 W 1 1 1 1 1
07/27/06 11:16:40 1.71 97.15 3 20 6.3 W 1 1 1 1 1
07/29/06 19:53:43 23.59 -63.92 1 10 5.8 W 0 0 1 1 0
08/17/06 15:20:35 46.54 141.91 2 14 6.0 B 1 1 0 0 0
08/20/06 3:01:02 49.82 156.42 1 26 6.0 W 1 1 1 1 1
08/24/06 21:50:37 51.15 157.52 1 43 6.5 W 1 1 1 1 1
09/24/06 22:56:21 -17.74 41.81 3 6 5.7 W 0 0 1 0 1
09/29/06 13:08:26 10.86 -61.67 1 53 6.1 W 0 1 1 1 0
09/30/06 12:47:23 7.28 -34.67 2 10 5.6 B 0 0 0 1 0
09/30/06 17:50:23 46.35 153.17 2 11 6.1 W 1 1 1 1 1
10/09/06 10:01:47 20.65 120.02 2 14 6.3 W 1 1 0 1 1
11/17/06 18:03:12 28.59 129.9 2 22 6.2 W 1 1 0 1 0
12/01/06 3:58:22 3.39 99.08 3 204 6.3 W 1 1 0 1 1
12/07/06 19:10:22 46.15 154.39 2 16 6.4 W 1 1 1 1 1
12/17/06 21:10:22 4.82 95.02 3 36 5.8 W 1 0 0 1 1
12/26/06 12:26:21 21.8 120.55 2 10 7.3 S 1 1 0 1 1
12/30/06 8:30:50 13.31 51.37 3 15 6.6 W 0 0 1 0 1
01/08/07 12:48:41 8.01 92.44 3 11 6.2 S 1 1 1 1 1
01/08/07 17:21:50 39.8 70.31 2 16 6.1 S 1 1 0 1 1
01/09/07 15:49:33 59.42 -137.12 1 10 5.8 B 0 1 1 0 1
01/17/07 23:18:50 10.13 58.71 3 8 6.2 W 0 0 1 0 1
03/06/07 5:49:25 -0.49 100.53 3 11 6.3 W 1 1 0 0 1
03/09/07 3:22:43 43.22 133.53 2 441 6.1 B 1 1 0 1 1
03/25/07 0:41:58 37.34 136.59 2 8 6.8 S 1 1 0 1 1
03/28/07 21:17:11 -6.27 29.67 3 8 5.8 W 0 0 1 0 1
04/03/07 3:35:07 36.45 70.69 2 222 6.2 W 1 0 0 0 0
04/05/07 3:56:50 37.31 -24.62 1 14 6.3 W 0 0 1 1 0
04/07/07 7:09:25 37.31 -24.49 1 8 6.1 W 0 1 0 1 0
04/20/07 1:45:56 25.71 125.11 2 9 6.3 W 1 1 0 0 0
04/20/07 19:37:58 27.47 128.38 2 42 5.9 B 1 0 0 0 0
04/29/07 12:41:57 52.01 -179.97 1 117 6.2 W 0 1 1 1 1
05/04/07 12:06:52 -1.41 -14.92 2 7 6.2 W 0 0 0 1 0
05/05/07 8:51:39 34.25 81.97 2 9 6.1 W 1 1 0 0 0
05/23/07 4:41:47 52.35 -31.81 1 10 5.7 W 0 0 1 1 0
05/30/07 20:22:13 52.14 157.29 1 116 6.4 W 1 1 1 1 1
06/02/07 21:34:58 23.03 101.05 2 5 6.3 S 1 1 0 1 1
06/15/07 18:49:53 1.72 30.83 3 24 5.9 W 0 0 1 0 1
07/03/07 8:26:01 0.72 -30.27 2 10 6.3 W 0 0 0 1 0
07/16/07 1:13:22 37.53 138.45 2 12 6.6 W 1 1 0 1 1
07/17/07 14:10:42 -2.73 36.36 3 8 5.9 W 0 0 1 0 1
07/29/07 4:54:37 53.64 169.7 1 26 6.0 B 1 1 1 1 1
166

07/30/07 22:42:06 19.31 95.61 2 14 6.0 W 1 1 0 1 1
07/31/07 22:55:31 -0.16 -177.95 2 11 6.2 W 0 0 0 1 0
08/08/07 17:05:05 -5.86 107.4 3 280 7.5 W 1 1 0 1 1
08/20/07 12:37:07 -0.26 -18.18 2 10 5.7 W 0 0 0 1 0
08/20/07 22:42:29 8.04 -39.25 2 6 6.5 W 0 0 1 1 0
09/14/07 6:01:32 -4.08 101.17 3 23 6.4 W 1 1 1 1 1
09/20/07 8:31:14 -2 100.14 3 30 6.7 W 1 1 1 1 1
10/02/07 18:00:07 54.51 -161.71 1 32 6.3 W 1 1 1 1 1
Table A.1: Selected events and scattering modes used in 2-D GRT inversion of SL array.
Date Time Lat. Lon.
Bi n
Z M
M t
Pds
Pp dp
Pp ds
Psd s|v
Psd s|h
11/22/07 23:02:13 4.74 95.06 3 49 5.9 B 1 1 1 0 1
11/29/07 19:00:20 14.94 -61.27 1 156 7.4 W 0 1 1 1 1
12/08/07 19:55:19 -7.56 37.65 3 6 5.6 W 1 1 0 0 1
12/19/07 9:30:28 51.36 -179.51 1 34 7.2 W 1 1 1 1 1
12/25/07 14:04:35 38.5 142.03 2 48 6.1 W 1 1 0 0 0
12/26/07 22:04:55 52.56 -168.22 1 25 6.4 W 1 1 1 1 1
01/04/08 7:29:18 -2.78 101.03 3 35 6.0 W 1 1 0 0 0
01/09/08 8:26:45 32.29 85.17 2 10 6.4 W 1 1 1 1 1
01/09/08 14:40:01 51.65 -131.18 1 10 6.1 W 1 1 1 0 0
01/16/08 11:54:44 32.33 85.16 2 9 6.0 B 1 1 1 1 1
02/03/08 7:34:12 -2.3 28.9 3 10 6.1 B 1 1 1 0 0
02/08/08 9:38:14 10.67 -41.9
02/20/08 8:08:31 2.77 95.96
2
3
9
26
6.9 W
7.5 S
0
1
0
1
1
1
1
1
1
1
02/21/08 2:46:18 77.08 18.57 1 12 6.1 W 1 1 1 1 1
02/21/08 14:16:03 41.15 -114.87 1 7 6.0 W 1 1 1 0 0
02/24/08 14:46:21 -2.41 99.93 3 22 6.5 W 1 1 0 1 1
02/25/08 8:36:33 -2.49 99.97 3 25 7.3 S 1 1 1 1 1
03/03/08 9:31:03 46.41 153.18 2 10 6.5 W 1 1 1 1 0
03/03/08 18:01:41 14.23 56.57 3 10 5.5 B 1 0 1 0 1
03/13/08 13:28:45 -45.49 35.01 3 10 6.0 W 0 1 0 0 0
03/20/08 22:32:58 35.49 81.47 2 10 7.3 S 1 1 1 1 1
03/22/08 21:24:11 52.18 -178.72 1 132 6.2 W 1 1 1 1 1
04/02/08 14:36:11 -4.88 69.24 3 10 5.7 W 1 1 1 1 1
04/16/08 5:54:20 51.88 -179.17 1 13 6.6 W 1 1 1 1 1
04/23/08 18:28:42 22.88 121.62 2 10 6.0 W 1 0 0 1 0
05/02/08 1:33:37 51.86 -177.53 1 14 6.8 S 1 1 1 1 1
05/07/08 16:02:03 36.18 141.55 2 19 6.2 W 1 1 0 1 0
05/12/08 6:28:02 31
05/23/08 19:35:35 7.31
103.32
-34.9
2
2
19
8
8.1 S
6.5 W
1
0
1
1
1
1
1
1
0
0
05/24/08 4:58:19 42.39 -30.52 1 10 5.5 W 0 1 0 1 0
05/25/08 19:18:26 56.09 -153.78 1 22 6.0 W 1 1 1 1 1
05/29/08 15:46:00 64 -21.01 1 9 6.3 W 1 1 1 0 0
06/01/08 1:57:24 20.12 121.35 2 31 6.3 W 1 1 0 1 0
167

06/13/08 23:43:45 39.03 140.88 2 8 7.0 S 1 1 1 1 0
06/22/08 23:56:30 76.7 141.28 1 18 6.1 W 1 0 1 1 1
06/27/08 11:40:14 11.01 91.82 3 17 6.7 S 1 1 1 1 1
07/03/08 6:34:54 10.28 -60.44 1 34 5.8 W 0 1 1 1 0
07/05/08 2:12:04 53.88 152.89 1 633 7.7 W 1 1 1 1 1
07/08/08 7:42:11 27.53 128.33 2 43 6.0 W 1 1 0 0 0
07/13/08 14:58:33 21.01 121.15 2 14 6.3 S 1 1 0 0 0
07/19/08 2:39:29 37.55 142.21 2 22 7.0 W 1 1 1 1 0
07/24/08 1:43:16 50.97 157.58 1 27 6.2 W 1 0 0 1 0
07/27/08 21:15:42 -0.25 -18.29 2 17 5.9 W 0 1 1 1 0
08/05/08 9:49:17 32.76 105.49 2 6 6.0 W 0 1 1 1 0
08/11/08 23:38:38 -1.02 -21.84 2 13 6.0 W 0 0 0 1 0
08/22/08 7:47:40 -17.77 65.39 3 6 6.0 W 1 1 0 0 1
08/25/08 13:21:59 30.9 83.52 2 12 6.7 W 1 0 1 0 0
08/27/08 1:35:32 51.61 104.16 2 16 6.3 W 1 0 1 1 0
08/27/08 6:46:19 -10.75 41.47 3 10 5.9 B 1 1 0 0 0
08/28/08 15:22:23 -0.25 -17.36 2 12 6.3 W 0 1 0 1 0
08/30/08 8:30:53 26.24 101.89 2 11 6.0 W 1 0 1 0 0
09/10/08 11:00:34 26.74 55.83 3 12 6.1 W 1 0 1 0 0
09/10/08 13:08:15 8.09 -38.7 2 9 6.6 W 0 1 1 1 0
09/11/08 0:20:51 41.89 143.75 2 25 7.0 S 1 1 1 1 0
09/26/08 18:46:19 3.07 65.32 3 10 5.7 W 1 1 1 1 1
10/06/08 8:30:46 29.81 90.35 2 12 6.3 W 1 1 1 1 0
10/07/08 10:00:48 79.82 -115.45 1 10 5.7 W 1 1 1 1 1
10/11/08 10:40:14 19.16 -64.83 1 23 6.1 W 0 1 1 1 0
10/28/08 23:09:58 30.64 67.35 2 15 6.6 S 1 0 1 1 0
10/29/08 11:32:43 30.6 67.45 2 14 6.6 S 1 1 1 1 0
11/10/08 1:22:03 37.56 95.83 2 19 6.4 B 1 1 1 1 0
11/17/08 12:55:23 79.65 -116.06 1 7 5.7 W 1 0 1 0 1
11/22/08 18:49:42 -1.23 -13.93 2 10 6.3 W 0 1 0 1 0
11/24/08 9:02:59 54.2 154.32 1 492 7.3 W 1 1 1 1 1
12/07/08 6:23:10 13.35 -44.83 2 10 5.6 W 0 0 1 1 0
12/08/08 1:51:01 13.41 -44.8 2 10 5.6 B 0 1 1 1 0
12/20/08 21:05:16 -31.19 -13.34 2 4 5.8 W 0 1 1 1 0
01/15/09 17:49:39 46.86 155.15 1 36 7.5 S 1 1 1 1 0
03/06/09 10:50:29 80.32 -1.85 1 9 6.6 B 1 0 1 0 1
03/30/09 7:13:07 56.55 -152.74 1 21 6.0 W 1 1 1 0 1
04/05/09 12:56:15 -5.26 68.55 3 10 5.6 W 1 1 0 0 0
Table A.2: Selected events and scattering modes used in 2-D GRT inversion of NL array.
168

A.6 Figures
Figure A.1: Comparison of composite images for SL (a) and NL (b) using different projection line angles from top to bottom as labeled. Colorbars denote d β / β from +/- 5%.
Solid lines highlight Moho of subducted and overriding plates for SL and NL as shown in
Figure 2.3 and 2.5, respectively.
169

Figure A.2: Comparison of individual scattering mode images for SL (a) and NL (b) from top to bottom as labeled. Colorbars denote d α / α and d β / β from +/- 10% and +/- 5%, respectively. Solid lines highlight Moho of subducted and overriding plates for SL and
NL as shown in Figure 2.3 and 2.5, respectively.
170

171

B.1 Introduction
This appendix is comprised of three parts. First, we give a detailed description of the automated multichannel alignment method introduced in Section 3.2.1. Then, we use a synthetic test to benchmark the RF imaging method described in Section 3.2.2. Finally, we perform an error analysis to explore the effects of uncertainty in the discontinuity dip assumed during CCP imaging.
B.2 Automated multichannel alignment method
Here we present an automated multichannel alignment method developed to improve the multichannel preprocessing method described in Section 3.2.1. Multichannel preprocessing takes advantage of waveform coherence across an array of stations to obtain a robust estimate of the incident wavefield. The incident wavefield recorded at each station must be aligned, such that a single reference time describes all stations in the array. Misaligned station can then be defined as any station that is time-shifted relative to the reference time of the array. Alignment errors can severely degrade the phase and the amplitude of the RFs produced via multichannel preprocessing. Therefore, it is critical to identify and remove them prior to imaging. Here, we develop an automated multichannel alignment method designed to remove alignment errors during multichannel preprocessing. It is presented in the following three sections. The first section describes the main steps in the multichannel preprocessing method, defines the different wavefields
172

it uses, and shows how these wavefields are affected by alignment errors. The second section describes our approach for identifying misaligned stations using the deconvolved
P wavefield. The last section provides a detailed description of each step in the automated multichannel alignment method. An example event is used to illustrate the main points of each section. We chose an incident PP phase from an event in the Andean subduction zone. PP phases typically have lower signal-to-noise ratio than P phases, but provide a means of expanding data coverage, both in backazimuth and slowness. Thus, our example demonstrates how the automated multichannel alignment method can provide high quality RFs from phases with low signal-to-noise, thereby improving data coverage.
In this section, we describe each step in the multichannel preprocessing method to define the different wavefields it uses and how these wavefields are influenced by station misalignment. The multichannel preprocessing method is comprised of the following four steps [see Rondenay et al., 2005 for details], each of which defines a wavefield used in our automated multichannel alignment method: (1) transform the recorded wavefield from N-E-Z coordinates to P-SV-SH coordinates and perform an initial alignment, which generates the first estimate of the aligned P wavefield; (2) refine the alignment using multichannel cross-correlation, which produces the second estimate of the aligned P wavefield; (3) estimate the incident wavefield using principal component analysis, which generates the second estimate of the incident wavefield; (4) deconvolve the incident wavefield from the SV and SH components, which produces the second estimate of the
SV and SH receiver functions, respectively. In the following, we define each of these wavefields in greater detail, and use examples of them to illustrate how misaligned stations destabilize RF amplitudes.
Step 1 : Define the first estimate of the aligned P wavefield
In the first step, the raw data is transformed from the N-E-Z coordinate system to the P-SV-SH coordinate system using the free-surface transfer matrix [Kennett, 1991].
173

This produces a P wavefield containing most of the incident wavefield’s energy with minimal leakage onto the SV and SH wavefields. At this stage, we also perform a
“rough” initial alignment of the wavefields by either hand-picking arrival times, as was done here, or using arrival times from a 1-D reference model. Note that all alignment corrections are simultaneously applied to each wavefield component (i.e. P, SV, and SH).
This step produces the first estimate of the aligned P wavefield, P j
A1 , where the subscript j indexes the different stations in the array and the superscript A1 indicates the wavefield has been aligned once. The alignment of the incident wavefield is typically quite poor at this stage (see example of P j
A1 shown in Figure B.1a).
Step 2 : Refine the aligned P wavefield
In the second step, we refine the alignment of the P wavefield by applying multichannel cross-correlation [vanDecar and Crosson, 1990] to a window around the early portion of the P wavefield. The window length is typically ~5 sec, but can be as much as 20 sec for emergent events. This yields a set of optimal delay times (in a leastsquares sense) that are used to shift P j
A1 into better alignment. The time shifts are applied in the frequency domain to produce the second estimate of the aligned P wavefield, P j
A2 ,
P j
A 2
= FFT − 1
(
FFT
( ) 1 exp
(
i
t A 1 j
) )
(B.1) where FFT and FFT -1 denote the fast-Fourier transform and its inverse, respectively; exp is the exponential function; i is the imaginary unit;
Δ t j
P j
A1
A1 is the delay time derived from
where the subscript j indexes each of the different stations in the array; and w is the vector of discrete angular frequencies matching the output of FFT . The multichannel cross-correlation delay times lead to significant overall improvement in the P wavefield alignment, as can be seen in our example (compare Figure B.1a to Figure B.1b), but there are a few stations that appear to be misaligned ( j =14 and 30).
Step 3 : Estimate the incident wavefield
174

In the third step, we produce an estimate the incident wavefield from the first principal component of the aligned P wavefield. This produces an incident wavefield estimate, I j
Ax , of the following form:
I j
Ax
=
j
W
(B.2) where W is a single incident waveform of unit length (i.e. norm equal to one) that
Γ j
is the amplitude estimate of the incident wavefield for the j th station. At this stage, we define the alignment iteration number x such that the superscript Ax refers to the x th estimate of any aligned P wavefield
(i.e. an input of P j
A2 yields and output of I j
A2 ). It is important to note that W defines the reference time of the incident wavefield for all stations in the array (i.e. all RF delay times are measured relative to W for a given event) while the amplitude of the incident wavefield, Γ j
, is station dependent. Γ j
is related to the portion of the aligned P wavefield amplitude at a given station that “matches” W , and is approximately equal to
j
P j
Ax
W
(B.3) where !
denotes the vector dot product (for simplicity, we have omitted the Ax superscript on W
Γ j
). This approximation is generally accurate to within ~2%.
Misaligned stations by definition have poor estimates of the aligned P wavefield that are time-shifted relative to W (i.e. they are out of phase). Thus, according to equation (B.3), misaligned stations will also have a reduced Γ j
. Figure B.2 shows examples of P j
A2 and
I j
A2 . Several individual stations exhibit relatively small G j
values compared to other stations (e.g. j = 14 and 30), which likely result from station misalignment, though it is difficult to observe the actual time shifts relative to other stations in P j
A2 .
Step 4 : Compute receiver functions
In the fourth step, we compute receiver functions by deconvolving the incident wavefield from the aligned SV wavefield. Only the SV receiver functions are discussed
175

here, because the SH RFs lead to the same general conclusions. Deconvolution is done in the frequency domain with regularization (see Rondenay, 2009 and references therein), leading to the following definition of the SV receiver function, SV j
Dx :
SV j
Dx
= FFT − 1 ⎜
⎛
⎜
⎝
I
SV j
Ax
( ω ) I j
Ax
( ω ) I j j
Ax
( ω ) *
Ax
( ω ) *
+ ε j
⎟
⎠
⎞
⎟
(B.4) where SV j
Ax ( w ) and I j
Ax ( w ) are the Fourier transforms (via FFT ) of the x th estimate of the
€
ε j
is damping factor for the j th station; and superscript Dx refers to the x th estimate of any wavefield obtained through deconvolution (SV RFs in this case). ε j
is chosen such that it minimizes unstable “ringing” in each receiver function and is constrained to be less than
5% of the maximum amplitude of I j
Ax ( w )I j
Ax ( w ) * (see Section 2.3.1 for further discussion).
It can be shown by substituting equation (B.2) into (B.4) that the receiver function amplitudes are inversely proportional to the incident wavefield amplitude factor (i.e.
1/ Γ j
). Therefore, the decrease in Γ j
caused by station misalignment will artificially increase (i.e. destabilize) the receiver function amplitudes. This effect is clearly observed in the example receiver functions shown in Figure B.3. The station with the smallest G j value (i.e. j = 14) has a peak RF amplitude, which is nearly an order of magnitude larger than the average peak amplitude of the other stations.
Station misalignment amplifies the receiver functions in a different way than the instability resulting from deconvolution. Misalignments amplify the entire spectrum evenly, with a proportionality of ~1/ Γ j
(Figure B.3a). Conversely, deconvolution-related amplification is the result of dividing by small amplitude values in the spectrum of the incident wavefield (i.e. in the denominator of equation (B.4)). Thus, it is related to details of the incident waveform (i.e. W ) and the presence of noise within SV j
Ax . For this reason, misalignment-related amplification cannot be addressed by varying the damping parameter in equation (B.4) or by implementing a different deconvolution method. Thus, a better approach is needed for identifying misaligned stations so that we can correct them.
176

In this section, we develop an approach for identifying misaligned stations. The aligned
P wavefield contains the imprint of the incident wavefield (i.e. W ) and noise, which make it difficult to identify the subtle time shifts associated with misaligned stations. These issues can be addressed by deconvolving the incident wavefield from the aligned P wavefield, thus producing the deconvolved P wavefield, P j
Dx ,
P j
Dx
= FFT − 1 ⎜
⎛
⎜
⎝
I
P j
Ax ( ω ) I j j
Ax ( ω ) I j
Ax ( ω ) *
Ax ( ω ) *
+ ε j
⎟
⎠
⎞
⎟
(B.5) where P j
Ax ( w ) and I j
Ax ( w ) are the Fourier transforms (via FFT ) of the x th estimate of the
€
ε j
is damping factor for the j th station. Normalizing P j
Ax by the incident wavefield essentially reduces P j
Dx to a single “spike” or, more formally, a band-limited delta function (i.e. compare Figure B.1b with Figure B.4a). The ideal alignment (and shape) of this “spike” is described by the resolution kernel, which is defined as the incident wavefield deconvolved by itself
K j
Dx
= FFT − 1 ⎜
⎛
⎜
⎝
I
I j
Ax ( ω ) I j
Ax ( ω ) I j j
Ax ( ω ) *
Ax ( ω ) *
+ ε j
⎟
⎠
⎞
⎟
(B.6) where the damping factor,
ε j
, is the same as the one used to compute P j
Dx . Time shifts
€
P j
Dx and K j
Dx provide a clear indication of station misalignment.
Figure B.4 shows examples of P j
D2 and K j
D2 to illustrate this point. Several stations exhibit large time shifts (i.e. > 0.5 sec) indicative of alignment errors (e.g. j = 14 and 30).
These stations also have anomalously low Γ j
values (Figure B.2b), as predicted by
177

equation (B.3). Thus, these misaligned stations produce unstable receiver functions, so they must be either corrected or the removed prior to imaging.
Alignment errors are difficult to correct within the confines of the existing multichannel preprocessing steps. They arise primarily from noise within the P window used in the multichannel cross-correlation. Thus they are sensitive to changes in the P window parameters (i.e. location and width) but are only minimally influenced by other parameters. Pearce et al. [2012] (see also Section 2.3.1) used a trial and error approach to reduce alignment errors by repeated steps 2-4 in the multichannel preprocessing with different P windows. They used the deconvolved P wavefield to qualitatively compare the alignment errors generated by each P window. The P window that produced the smallest overall alignment error was selected to generate the final receiver functions for imaging. However, this approach is time consuming, qualitative, and often unable to align many of the stations. Any station that could not be corrected had to be manually removed from the dataset, which is also quite time consuming. To address these issues, we have developed an alternative alignment method, which we describe next.
In this section, an automated multichannel alignment method is developed that minimizes alignment errors below a prescribed threshold for all stations in the array. It has several advantages over the previous approach, namely that it is automated (there are no adjustable parameters) and provides quantitative quality control on the wavefield alignment. The method uses a four step process: (i) estimate alignment time corrections from the deconvolved P wavefield, (ii) use these corrections to improve the estimate of the aligned P wavefield, (iii) update the estimates of the incident and deconvolved P wavefields, and (iv) iterate over steps i-iii until the alignment time corrections fall below a prescribed tolerance. In the following, we describe each step of the automated multichannel alignment method, and demonstrate how it improves the alignment for our example event.
In step (i), we estimate delay time corrections that measure the alignment errors identified in the deconvolved P wavefield. This is done by applying multichannel cross-
178

correlation to the deconvolved P wavefield, P j
D2 (e.g. Figure B.4a), similar to the approach used on P j
A1 in step 2 of the multichannel preprocessing (see Section B1.1).
This yields a second set of optimal delay times derived from the deconvolved P wavefield, Δ t j
D2 , which represent corrections to the first set of delay times derived from the first estimate of the aligned P wavefield,
Δ t j
A1 .
In step (ii), we apply these delay time corrections to obtain a better estimate of the aligned P wavefield. This is done by shifting the first estimate of the aligned P wavefield, P j
A1 , by both Δ t j
A1 and Δ t j
D2 to produce the third estimate of the aligned P wavefield, P j
A3 ,
P j
A 3
= FFT − 1
FFT
( ) exp
(
i
[ t A 1 j
+
t D 2 j
] )
(B.7).
The term in brackets defines the total alignment time correction required to improve the alignment of P j
A1 and is updated with each iteration as described in step (iv).
In step (iii), we use the third estimate of the aligned P wavefield ( P j
A3 ) to produce refined estimates of the incident and deconvolved P wavefields (i.e. I j
A3 and P j
D3 ). P j
A3 is used to update the incident wavefield estimate, I j
A3 , using step 3 of the multichannel preprocessing (see Section B1.1). Then we updated the estimate of the deconvolved P wavefield, P j
D3 , by deconvolving I j
A3 from P j
A3 as described in Section B1.2. This produces a more stable estimate of the incident wavefield (increases G j
values of misaligned stations) and an estimate of the deconvolved P wavefield with smaller alignment errors. However, significant alignment errors may persist at this stage due to
(1) noise in the deconvolved P window, (2) changes in the updated estimate of the incident waveform ( W ), and (3) bias in the least-squares estimates of
Δ t j
D2 from large station misalignments (i.e. outliers) [vanDecar and Crosson, 1990].
In step (iv), we iterate over steps i-iii until the alignment errors for all stations fall below a prescribed tolerance. First, we apply multichannel cross-correlation to the most recent estimate of the deconvolved P wavefield (i.e. P j
D3 ). This generates a third set of time corrections (e.g. Δ t j
D3 ), which are added to the previous two sets of delay times to generate a new total alignment time correction (i.e. the bracketed term in equation (B.7)
179

becomes [ Δ t j
A1 + Δ t j
D2 + Δ t j
D3 ]). These new alignment corrections are used to update the estimates of the aligned P wavefield using step ii, which is then used to update the estimates of the incident and deconvolved P wavefields using step iii. This process is repeated until the delay time corrections for each station (i.e.
Δ t j
Dx ) fall below a prescribed threshold. The threshold is typically chosen to be the data-sampling rate (0.1 sec in our case), which often requires several iterations for an event with low signal-tonoise ratio. However, each iteration only takes a few seconds so the overall computation time is still quite small. Once this stopping criterion is met, the aligned and incident wavefield estimates from the last iteration are used to compute the final receiver functions via step 4 of the multichannel preprocessing.
Finally, we use our example event to demonstrate the effectiveness of the automated multichannel alignment method. Figure B.5 shows the alignment time corrections produced from each iteration. Convergence occurs after three iterations, though the first iteration is responsible for the majority of the total alignment time correction. The alignment errors that are clearly present in the second deconvolved P wavefield (Figure B.4a) have now been eliminated as shown in Figure B.4c. Previously misaligned stations now have incident wavefield amplitudes that are larger and more consistent with the amplitudes observed at other stations (Figure B.2). Removing the alignment errors leads to receiver functions that have nearly uniform peak amplitudes for all stations in the array (Figure B.3). For example, the RF amplitudes for station #14 have fallen by nearly an order of magnitude.
B.3 Synthetic test
A synthetic test is used to benchmark the RF imaging method described in Section 3.2.2.
Synthetic data is generated using the RAYSUM package of Frederiksen and Bostock
[2000], a high-frequency asymptotic method (i.e. assumes ray theory) capable of modeling 3-D dipping discontinuities. A two-step process is used to test the RF imaging method. First, we describe the model used to generate the synthetic RFs, which is designed to capture the main features in the RF images from field data. Second, we use the synthetic RFs to produce CCP images. The imaged discontinuities are compared to
180

the discontinuities in the model to verify the accuracy of the method. In the following, we briefly describe these two steps.
First, we describe the model used to generate synthetic RFs. It is designed to emulate the subducted crust and overriding Moho observed in the RF images (e.g. Figure
3.4). The model geometry is shown in Figure B.6. It is comprised of a 35 km thick top layer representing the crust (a = 6.5 km/s and b = 3.7 km/s), which is underlain by an upper mantle (a = 8.1 km/s and b = 4.7 km/s). A low velocity layer (LVL) representing subducted crust dipping to the east at 17° is embedded within the upper mantle. The
LVL has the same velocities as the top layer and its top is fixed at 60 km depth beneath the first seismic station (X=0 km). The station array is linear, 150 km long, and has a station spacing of 7.5 km. The incident P waveform is comprised of a Gaussian pulse with a width of 0.5 sec. We simulate a total of 24 incident P waves spanning a complete range of backazimuths (0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°) and slownesses
(0.04 s/km, 0.06 s/km, and 0.08 s/km).
Figure B.6 shows the SV and SH images for three cross-sections beneath the array. They are produced by mapping the synthetic RFs to depth using a two-layer model comprised of the crust and mantle described above, but no LVL. Discontinuities are assumed to dip at 17° for all depths. Overall, the SV and SH RF images recover the input velocity structure with great accuracy. For the SV images, a negative discontinuity clearly tracks the top of the LVL in all three-dimensions (i.e. along each cross-section).
This indicates that the delay time and conversion point expressions presented in Section
3.2.2 are equivalent to those of Frederiksen and Bostock [2000], for all P wave backazimuths. The positive discontinuity along the base of the LVL (i.e. slab Moho) has the correct dip but is shifted to slightly greater depth (~2 km) because the background model does not account for waves that propagate within the LVL. The positive discontinuity marking the base of the crust (i.e. overriding Moho) is located at ~34 km depth. This small depth error (< ~1 km) results from the discrepancy between the modeled discontinuity dip (17°) and the actual discontinuity dip (0°) (such errors increase with increasing depth).
The SH images show two interesting features. First, there is no signal on the middle profile (II) and no signal from the overriding Moho. The middle profile lacks
181

signal because it is formed using P waves that sample directly inline with the array.
These waves travel close to the updip (backazimuth = 90°) and downdip (backazimuth =
270°) directions of the dipping discontinuity, and so produce no SH signal. Similarly, the overriding Moho is horizontal and so produces no SH signal. Second, the southern profile (Y=-30 km) exhibits the same LVL polarities as the SV profile (negative top, positive base) but the northern profile (Y=30 km) shows the opposite polarities. These opposing polarities are expected given the P waves used to form these two profiles sample along the strike of the dipping discontinuity, but in opposite directions. However, this result is sign convention dependent. To verify our sign convention is consistent with the field data processing, we first generate all synthetic data from the RAYSUM package in the n-e-z coordinate system. Then we correct it to the n-e-z coordinate system used in the preprocessing described in Section 3.2.1. Finally, the preprocessing is used to transform the data into the SV and SH coordinate system shown in Figure B.6. This procedure guarantees the same SH polarity convention is used for both the field and synthetic data. Thus, the SH polarity in Figure B.6 is indicative of an LVL with isotropic properties, though anisotropy also strongly influences SH polarity [Frederiksen and
Bostock; Park and Levin, 2004].
B.4 Error analysis
In this section, an error analysis is performed to explore the effects of uncertainty in model discontinuity dip, which is defined as the dip assumed for forward calculations of the conversion points versus delay time described in Section 3.2.2. There are two other types of discontinuity dips referred to in this section that relate to the data: 1) the data dip is defined as the dip of the discontinuity that generated the actual RF data (known for synthetic tests but unknown for real data), and 2) the image dip refers to the approximation of the discontinuity dip generated by applying the forward model to data.
In the ideal situation, each of these different discontinuity dips would be equal to each other; however, the data dip is never known in practice and must instead be approximated based on the choice of model dip and the resulting image dip. The error analysis described in this section focuses on two important consequences of uncertainty in model
182

dip: 1) error in the position of the discontinuity (i.e. model vs. data conversion points), and 2) error in the image dip relative to the data dip.
The conversion point error as a function of uncertainty in model dip is found by solving the equation for the delay time assuming two different discontinuity dips as follows:
t
D
= (
S
D
P
D
) z
D (B.8)
t
D
= (
S
M
P
M
) z
M (B.9) where the subscript D denotes variables that are computed with the data discontinuity dip
(i.e. the dip of the discontinuity that generated the data) while the subscript M denote
the model discontinuity depth, z
M
, is equal to the true discontinuity depth, z
D
, plus some model error in discontinuity depth, dz
M
as follows: z
D
= z
M
+ δ z
M (B.10)
Equating (B.8) and (B.9), and then substituting (B.10) yields an expression for the error in discontinuity depth as a function of the effective vertical slownesses and delay time
€
z
M
=
S
D
1
P
D
S
M
1
P
M
t
D
(B.11)
Thus, the model error in conversion depth is a linear function of the PdS-P delay time with a slope that is determined by the difference in the P- and S-wave effective vertical
M
) and y (dy
M
) conversion points may be derived in a similar fashion leading to
€
δ x
M
=
⎛
⎜
⎝ p S x, D
η
S
D
⎞
⎟ z
⎠
D
−
⎛
⎜
⎝ p S x, M
η
S
M
⎞
⎟ z
⎠
M
(B.12),
δ y
M
=
⎛
⎜
⎝ p
S y, D
η
S
D
⎞
⎟ z
⎠
D
−
⎛
⎜
⎝ p
S y, M
η
S
M
⎞
⎟ z
⎠
M
(B.13),
183
€

with x
D
= x
M
+ δ x
M (B.14), y
D
= y
M
+ δ y
M (B.15).
€
€ down-dip conversion point (dx
M
) has the largest error, shifts the model conversion point up-dip, and is relatively insensitive to variations in backazimuth. The error in conversion point depth (dz
M
) shifts the model conversion point to greater depth, with a magnitude that ranges from ~25% of dx
M
for an event traveling up-dip to ~10% of dx
M
for an event traveling down-dip. The error in along-strike conversion point is essentially zero for all event back-azimuths. Thus, a large uncertainty in model dip produces an image discontinuity that is dramatically shifted up-dip but its depth and along-strike positions remain largely unaffected.
Finally, we examine how uncertainty in model discontinuity dip impacts the image discontinuity dip. Let us begin by developing an analytical expression for the image discontinuity dip, q
I
, which is defined as follows: tan( θ
I
) =
Z E 2
M
X E 2
M
− Z
− X
E 1
M
E 1
M (B.16) where the superscripts E
1
and E
2
refer to events with two different down-dip slowness components, p x1
and p x2
, respectively. The expression in (B.16) can be understood as
€ D
, tan( θ
D
) =
Z
D
E 2
X
D
E 2
− Z E 1
D
− X
D
E 1
(B.17) then the conversion points from any event will fall along a line with a slope equal to tan(q
I
) that is identical to the slope of the discontinuity that generated the data, tan(q
D
).
€
184

Thus, we may use (B.16) and (B.17) to derive an analytical expression for the image dip as a function of the data dip and uncertainty in the model dip. To do this, we first express the image dip in terms of the difference in data conversion points and error in conversion points between two events by substituting (B9) and (B13) into (B15), which yields tan(
I
) =
(z E 2
D
( x E 2
D
z E 2
M
x E 2
M
)
(z E 1
D
)
( x E 1
D
z E 1
M
)
x E 1
M
)
(B.18).
To explicitly describe the error in image dip we recast (B.18) in terms of tan(q
D
) and terms involving the data down-dip conversion point, x
D
, and its error dx
M
, which leads to
tan( θ
I
) = tan( θ
D
)( x
D
E 2
( x
D
E 2
− x E 1
D
− x E 1
D
) − ( δ z
) − ( δ x E 2
M
E 2
M
− δ x
− δ
E 1
M
) z E 1
M
)
(B.19).
Then, we simplify (B.19) into the following form
€
I
1
=
1
D
M
(B.20) where tan(dq
M
) is the slope of the error in x and z conversion point between the two events
tan(
M
) =
z E 2
M
x E 2
M
z E 1
M
x E 1
M
(B.21) and
ε
is the difference in x conversion point error relative to the difference in the x conversion points between the two events
=
x E 2
M x
D
E 2
x E 1
M
x
D
E 1
(B.22).
The expression in (B.20) explicitly shows that the image dip converges to the data dip as e approaches zero. Furthermore, e is small under typical receiver function imaging
dip is only a few degrees for any two events. However, the difference between the image dip and data dip obtained for different event pairs may vary significantly in both magnitude and sign, which may significantly degraded image quality. Thus, the expression in (B.20) demonstrates that a reasonable estimate of the data discontinuity dip
185

may be obtained by forming an image assuming an erroneous model discontinuity dip
(i.e. assume a horizontal discontinuity) as the image discontinuity dip will be similar to the data discontinuity dip, and not the model discontinuity dip used to produce the image.
In conclusion, our error analysis demonstrates that errors in model discontinuity dip impact receiver function images in two ways: 1) the discontinuity position has a large error in the down-dip direction, and to a lesser extent its depth, and 2) the dip of the discontinuity in the image is not significantly affected.
186

B.5 Figures
Figure B.1: P wavefield from an example event showing the improvement in alignment from step 2 of the multichannel preprocessing. The example is an incident PP phase from an event in the northern (Peruvian) Andean subduction zone that was recorded by the SL array. Panels a) and b) show the first ( P j
A1 ) and second ( P j
A2 ) estimates of the aligned P wavefield, respectively (see text for definitions). The window used in the multichannel cross-correlation is denoted by the dashed black lines in a). The resulting delay times lead to an improved alignment of the incident wavefield (i.e. (b) is better aligned than
(a)), but it is difficult to gauge the accuracy of the alignment due to the low signal-tonoise ratio. Stations #14 and #30 (black arrows) are examples of misaligned stations that we will follow in subsequent figures. In both panels, the x-axis denotes the station number (i.e. the index j ) and the y-axis is time in seconds (origin is arbitrary). Red and blue colors denote positive and negative P wavefield amplitudes, respectively. The color scale is saturated at the maximum amplitude in each panel.
187

188

Figure B.2: Comparison of the aligned P wavefield (a,c) and incident wavefield (b,d) estimates before (a,b) and after (c,d) application of the automated multichannel alignment method (see event description in Figure B.1). Panels (a) and (b) show the second estimates of the aligned P wavefield, P j
A2 , and incident wavefield, I j
A2 , respectively.
They are produced from steps 2 and 3 of the multichannel preprocessing method (see definitions in Section B1.1). Note that (a) is an expanded version of Figure B.1b. Panels
(c) and (d) show the fifth estimates of the aligned P wavefield, P j wavefield, I j
A5
A5 , and incident
. They are produced using 3 iterations of the automated multichannel alignment method. Note that the incident wavefields shown in (b,d) are each comprised of a single wavelet, W , scaled by an amplitude factor for each station, Γ j
as described in equation B.2. The small Γ j
values in (b) are an indication of misaligned stations (e.g. stations #14 and #30 denoted by black arrows). These misalignments have been corrected in (d), leading to larger and more uniform Γ j
values across the array compared to those in (b).
189

190

Figure B.3: Comparison of SV receiver functions before (a,b) and after (c,d) application of the automated multichannel alignment method (see event description in Figure B.1).
Panels (a) and (b) show the SV RFs from step 4 of the multichannel preprocessing (i.e.
SV j
D2 ) in the frequency domain and time domain, respectively. Panels (c) and (d) show the SV RFs after three iterations of the automated multichannel alignment method (i.e.
SV j
D5 ) in the frequency domain and time domain, respectively (same layout as (a) and
(b)). Panels (b) and (d) are computed by applying FFT -1 to each trace in (a) and (c), respectively (see equation B.4). The automated multichannel alignment method reduces the RF amplitudes by nearly a factor of ~10 for the most severely misaligned stations
(e.g. stations #14 and #30 denoted by black arrows). In all panels, the colorbar denotes the RF amplitude relative to the peak amplitude of the incident wavefield with the colorscale saturated at the maximum amplitude in each panel.
191

Figure B.4: Comparison of the deconvolved P wavefield (a,c) and resolution kernel (b,d) before (a,b) and after (c,d) application of the automated multichannel alignment method
(see event description in Figure B.1). Following the notation described in Section B1, panels (a), (b), (c), and (d) represent P j
D2 , K j
D2 , P j
D5 and K j
D5 , respectively. Ideally, each peak in (a,c) would be centered at the same time as in (b,d). Misaligned stations are identified by deviations from this ideal case (e.g. stations #14 and #30 denoted by black arrows). Multichannel cross-correlation is applied to panels (a) and (c) to produce the correction times shown in Figure B.5 (see the blue and black lines, respectively).
192

Figure B.5: Delay time corrections from our example event as a function of the number of iterations through the automated multichannel alignment method (see Section B1.3).
The y-axis is the delay time correction obtained from multichannel cross correlation applied to the x th estimate of the deconvolved P wavefield, Δ t j
Dx . The number of iterations through the alignment method is equal to x -1 where x is the alignment estimate number (see text for definition). For example, the first iteration produces time corrections indicated by the blue curve, Δ t j
D2 , that are derived from multichannel crosscorrelation applied to the deconvolved P wavefield, P j
D2 . Numerous stations exhibit large alignment errors (e.g. stations #14 and #30 denoted by black arrows). The time corrections indicated by the black curve, Δ t j
D5 all fall below alignment error tolerance of
0.1 sec (dashed black lines), indicating the method has converged after 3 iterations (i.e. three additional time correction terms within the bracketed term in equation B.7, [
Δ t j
A1 +
Δ t j
D2 + Δ t j
D3 + Δ t j
D4 ]).
193

Figure B.6: Test of the RF imaging method using synthetic data from the RAYSUM package of Frederiksen and Bostock [2000] (see text for model description). (a) Profile locations and array geometry with black triangles denoting individual stations. The positive X and Y axes point in the East and North directions, respectively. (b), (c), and
(d) show CCP images for profiles I, II, and III, respectively. The left and right panels present SV and SH images, respectively. The colorbars indicate the RF amplitude as a % of the incident P-wave amplitude. Black lines identify the location of model discontinuities. V
1
indicates regions of the model with P- and S-wave velocities of 6.5 km/s and 3.7 km/s, respectively, while V
2
identifies regions of the model with P- and Swave velocities of 8.1 km/s and 4.7 km/s, respectively.
194

195

Abers, G.A., L.S. MacKenzie, S. Rondenay, Z. Zhang, A.G. Wech, and K.C. Creager
(2009), Imaging the source region of Cascadia tremor and intermediate-depth earthquakes, Geology , 37(12), 1119-1122, DOI: 10.1130/G30143A.1.
Ammon, C. J. (1991). The isolation of receiver effects from teleseismic P waveforms,
Bull. Seism. Soc. Am., 81, 2504-2510.
Armijo, R., B. Meyer, G.C.P. King, A. Rigo and D. Papanastassiou (1996), Quaternary evolution of the Corinth Rift and its implications for the late Cenozoic evolution of the Aegean , Geophys. J. Int.
, 126, 11-53.
Audet, P., M.G. Bostock, N. Christensen, and S. Peacock (2009), Seismic evidence for overpressured subducted oceanic crust and megathrust fault sealing, Nature,
457(7225), 76–78, doi:10.1038/nature07650.
Baker, B., D. Hatzfeld, H. Lyon-Caen, E. Papadimitriou, and A. Rigo (1997), Earthquake mechanisms of the Adriatic Sea and Western Greece: implications for the oceanic subduction-continental collision transition, Geophys. J. Int ., 131(3), 559-594.
Båth, M. and R. Stefánsson (1966), S–P conversions at the base of the crust, Ann. Geofis.
,
19, 119–130.
Becker, T. W., and C. Faccenna (2009), A review of the role of subduction dynamics for regional and global plate motions, in Subduction Zone Geodynamics, edited by S.
E. Lallemand and F. Funiciello, pp. 3 – 34, Springer, Berlin, doi:10.1007/978-3-
540-87974-9.
Bialas, R.W., F. Funiciello, and C. Faccenna, Subduction and exhumation of continental crust: insights from laboratory models, Geophys. J. Int., 184, 43–64, doi:
10.1111/j.1365-246X.2010.04824.x.
Bohnhoff, M., J. Makris, D. Papanikolaou, and G. Stavrakakis (2001), Crustal investigation of the Hellenic subduction zone using wide aperture seismic data,
Tectonophysics, 343, 239–262.
Bostock, M.G. (1998), Mantle stratigraphy and evolution of the Slave province, J
Geophys Res.
, 103(B9), 21183–21200.
196

Bostock, M.G., and S. Rondenay (1999), Migration of scattered teleseismic body waves,
Geophys. J. Int ., 137, 732– 746.
Bostock, M.G., S. Rondenay, and J. Shragge (2001), Multiparameter two-dimensional inversion of scattered teleseismic body waves, 1. Theory for oblique incidence, J.
Geophys. Res., 106(B12), 30771–30782.
Bostock, M. G., R. D. Hyndman, S. Rondenay, and S. M. Peacock (2002), An inverted continental Moho and serpentinization of the forearc mantle, Nature , 417, 536–
538.
Bourova, E., I. Kassaras, H.A. Pedersen, T. Yanovskaya, D. Hatzfeld, and A. Kiratzi
(2005), Constraints on absolute S velocities beneath the Aegean Sea from surface wave analysis, Geophys. J. Int., 160(3), 1006–1019.
Brun, J.P. and C. Faccenna (2008), Exhumation of high-pressure rocks driven by slab rollback, Earth planet. Sci. Lett., 272, 1-7, doi:10.1016/j.epsl.2008.02.038.
Burchfiel, B.C. (1980), Eastern European Alpine system and the Carpathian orocline as an example of collision tectonics, Tectonophysics , 63, 31-61.
Burchfiel et al., in prep.
Burdick, L.J. and C.A. Langston (1977), Modeling crustal structure through the use of converted phases in teleseismic body-wave forms, Bull. Seismol. Soc. Am.
, 67(3),
677–691.
Calcagnile, G., F. D’Ingeo, P. Farrugia, and G.F. Panza (1982), The lithosphere in the central-eastern Mediterranean area, Pure Appl. Geophys., 120(2), 389–406.
Carminati, E., M.J.R. Wortel, W. Spakman, and R. Sabadini (1998), The role of slab detachment processes in the opening of the western–central Mediterranean basins: some geological and geophysical evidence, Earth planet. Sci. Lett., 160, 651–665.
Cassidy, J.F. (1992), Numerical experiments in broadband receiver function analysis,
Bull. Seism. Soc. Am., 82 , 1453–1474.
Cassinis, R., S. Scarascia, and A. Lozej (2003), The deep crustal structure of Italy and surrounding areas from seismic refraction data: a new synthesis, Bollettino di
Geofisica Teorica ed Applicata , 122, 365-375.
Chemenda, A.I., M. Mattauer, and A.N. Bokum (1996) Continental subduction and a mechanism for exhumation of high-pressure metamorphic rocks: new modelling and field data from Oman, Earth planet. Sci. Lett.
, 143, 173–182.
197

Clément, C., A. Hirn, P. Charvis, M. Sachpazi, and F. Marnelis (2000), Seismic structure and the active Hellenic subduction in the Ionian islands, Tectonophysics, 329,
141–156.
Clément, C., M. Sachpazi, P. Charvis, D. Graindorge, M. Laigle, A. Hirn, and G.
Zafiropoulos (2004), Reflection–refraction seismics in the Gulf of Corinth: hints at deep structure and control of the deep marine basin, Tectonophysics, 391, 97–
108.
Crotwell, H. P., T. J. Owens, and J. Ritsema (1999). The TauP Toolkit: Flexible seismic travel-time and ray-path utilities, Seismological Research Letters , 70, 154–160.
Cruciani, C., E. Carminati, and C. Doglioni (2005), Slab dip vs. lithosphere age: no direct function, Earth Planet. Sci. Lett., 238, 298 – 310, doi:10.1016/j.epsl.2005.07.025.
De Franco, R., R. Govers, and M.J.R. Wortel (2008), Dynamics of continental collision: influence of the plate contact, Geophys. J. Int., 174 , 1101–1120, doi:10.1111/j.1365-246X.2008.03857.x. de Voogd, B., C. Truffert, N. Chamot-Rooke, P. Huchon, S. Lallemant, and X. Le Pichon
(1992), Two-ship deep seismic soundings in the basins of the Eastern
Mediterranean Sea (Pasiphae cruise), Geophys. J. Int ., 109(3), 536–552.
Dewey, J.F., and C. Ş engör (1979), Aegean and surrounding regions: Complex multiplate and continuum tectonics in a convergent zone, Geological Society of America
Bulletin , 90(1), 84-92.
Di Luccio, F., and M.E. Pasyanos (2007), Crustal and upper-mantle structure in the
Eastern Mediterranean from the analysis of surface wave dispersion curves,
Geophys. J. Int., 169(3), 1139–1152.
Ellis, S., C. Beaumont, and A. Pfiffner (1999), Geodynamic models of crustal-scale episodic tectonic accretion and underplating in subduction zones, J . Geophys.
Res ., 104(B7), 15169-15190.
Dinter, D.A., and L.H. Royden (1993), Late Cenozoic extension in northeastern Greece:
Strymon Valley detachment system and Rhodope metamorphic complex,
Geology , 21, 45-48.
Dueker, K.G. and A.F. Sheehan (1997) Mantle discontinuity structure from midpoint stacks of converted p to s waves across the Yellowstone hotspot track, J.
Geophys. Res.
, 102, 8313–8327.
Endrun, B., L. Ceranna, T. Meier, M. Bohnhoff, and H.P. Harjes (2005), Modeling the influence of Moho topography on receiver functions: A case study from the central Hellenic subduction zone, Geophys. Res. Lett., 32(12), L12311.
198

Endrun, B., T. Meier, S. Lebedev, M. Bohnhoff, G. Stavrakakis, and H.P. Harjes (2008),
S velocity structure and radial anisotropy in the Aegean region from surface wave dispersion, Geophys. J. Int., 174(2), 593–616.
Faccenna, C. and T.W. Becker (2010), Shaping mobile belts by small-scale convection,
Nature , 465, 602-605, doi:10.1038/nature09064.
Finetti, I. R., and A. Del Ben (2005), Crustal Tectono-Stratigraphic setting of the Adriatic
Sea from new CROP seismic data, in Deep Seismic Exploration of the Central
Mediterranean and Italy, CROP PROJECT , 23, edited by I. R. Finetti, 519–548,
Elsevier.
Finetti, I., D. Papanikolaou, A. Del Ben, and P. Karvelis (1991), Preliminary geotectonic interpretation of the East Mediterranean chain and the Hellenic arc, Bull. Geol.
Soc. Greece, 25(1), 509–526.
Floyd, M. A., H. Billiris, D. Paradissis, G. Veis, A. Avallone, P. Briole, S. McClusky,
J.M. Nocquet, K. Palamartchouk, B. Parsons, and P. C. England (2010), A new velocity field for Greece: Implications for the kinematics and dynamics of the
Aegean, J. Geophys. Res ., 115, B10403, doi:10.1029/2009JB007040
Frederiksen A.W., and M.G. Bostock (2000), Modelling teleseismic waves in dipping anisotropic structures, Geophys. J. Int.
, 141, 401-412.
Fytikas, M., O. Giuliani, F. Innocenti, G. Marinelli, and R. Mazzuoli (1976),
Geochronological data on recent magmatism of the Aegean Sea, Tectonophysics,
31, 29–34.
Fytikas, M., and N.P. Kolios (1979), Preliminary heat flow map of Greece, in Terrestrial
Heat Flow in Europe , edited by V. Cermak, and L. Rybach, Springer-Verlag,
197–205.
Galanopoulos, D., V. Sakkas, D. Kosmatos, and E. Lagios (2005), Tectonophysics , 409,
73–84, doi:10.1016/j.tecto.2005.08.010.
Gallovi č , F., J. Zahradník, D. K ř í ž ová, V. Plicka, E. Sokos, A. Serpetsidaki, and G-A. Tselentis (2009), From Earthquake Centroid to Spatial-Temporal
Rupture Evolution: Mw 6.3 Movri Mountain Earthquake, June 8, 2008, Greece,
Geophysical Research Letters , 36(21), 1-6, doi:10.1029/2009GL040283.
Gautier, P., J.P. Brun, R. Moriceau, D. Sokoutis, J. Martinod, and L. Jolivet (1999),
Timing, kinematics and cause of Aegean extension: a scenario based on a comparison with simple analogue experiments, Tectonophysics , 315,
31-72.
199

Gesret, A., M. Laigle, J. Diaz, M. Sachpazi, and A. Hirn (2010), The oceanic nature of the African slab subducted under Peloponnesus: thin-layer resolution from multiscale analysis of teleseismic P -toS converted waves, Geophys. J. Int .,
183(2), 833-849, doi:10.1111/j.1365-246X.2010.04738.x.
Gesret, A., M. Laigle, J. Diaz, M. Sachpazi, M. Charalampakis, and A. Hirn (2011), Slab tops dips resolved by teleseismic converted waves in the Hellenic subduction zone, Geophysical Research Letters , 38, doi:10.1029/2011GL048996.
Goldsworthy, M., J. Jackson and J. Haines (2002), The continuity of active fault systems in Greece, Geophys. J. Int ., 148(3), 596-618.
Govers, R., and M.J.R. Wortel (2005), Lithosphere tearing at STEP faults: Response to edges of subduction zones, Earth and Planetary Science Letters , 236, 505–523.
Hacker, B.R. (2008), H2O subduction beyond arcs, Geochem. Geophys. Geosyst., 9,
Q03001, doi: 10.1029/2007GC001707.
Hacker, B.R., S.M. Peacock, G.A. Abers, and S. Holloway (2003), Subduction Factory 2.
Intermediate-depth earthquakes in subducting slabs are linked to metamorphic dehydration reactions, Journal of Geophysical Research , 108, doi:10.1029/2001JB001129.
Hacker, B.R., and G.A. Abers (2004), Subduction Factory 3. An Excel worksheet and macro for calculating the densities, seismic wave speeds, and H2O contents of minerals and rocks at pressure and temperature, Geochem. Geophys. Geosyst., 5 ,
Q01005.
Hacker, B.R., P.B. Kelemen, and M.D. Behn (2011), Differentiation of the continental crust by relamination, Earth and Planetary Science Letters , 307, 501–516, doi:10.1016/j.epsl.2011.05.024.
Hager, B.H., (1984), Subducted slabs and the geoid: Constraints on mantle rheology and flow, J. Geophys. Res ., 89, 6003–6015.
Hager, B.H., and R.J. O’Connell (1978), Subduction zone dip angles and flow driven by plate motion, In: M.N. Toksöz (Editor), Numerical Modeling in Geodynamics,
Tectonophysics , 50, 111-133.
Hashida, T., G. Stavrakakis, and K. Shimazaki (1988), Three-dimensional seismic attenuation structure beneath the Aegean region and its tectonic implications,
Tectonophysics , 145, 43-54.
Hirn, A., M. Sachpazi, R. Siliqi, J. McBride, F. Marnelis, L. Cernobori, and the
STREAMERS-PROFILES group (1996), A traverse of the Ionian islands front
200

with coincident normal incidence and wide-angle seismics, Tectonophysics, 264,
35–49.
Hollenstein, C., M.D. Muller, A. Geiger, and H.G. Kahle (2008), Crustal motion and deformation in Greece from a decade of GPS measurements, 1993-2003,
Tectonophysics , 449, 17-40
,
doi:10.1016/j.tecto.2007.12.006.
Hosa, A. (2008), Imaging of the Hellenic subduction zone by seismic tomography, BS thesis , Massachusetts Institute of Technology, Cambridge, MA, USA.
Husson, L. (2006), Dynamic topography above retreating subduction zones, Geology , 34,
741-744, doi:10.1130/G22436.1.
Husson, L., J.P. Brun, P. Yamato and C. Faccenna (2009), Episodic slab rollback fosters
HP-UHP rocks exhumation, Geophys. J. Int.
, 179(3), 1292–1300, doi:
10.1111/j.1365-246X.2009.04372.x.
International Seismological Centre (2001), On-line Bulletin , http://www.isc.ac.uk,
Internatl. Seis. Cent.
, Thatcham, United Kingdom.
Jacobshagen, V., D. Richter, and J. Makris (1978), Alpidic development of the
Peloponnesus, in Alps, Apennines, Hellenides, edited by H. Closs, D. Roeder, and
H. Schmidt, 415–423, E. Schweizerbart’sche Verlagsbuchhandlung, Stuttgart,
Germany.
Jolivet, L., L. Labrousse, P. Agard, O. Lacombe, V. Bailly, E. Lecomte, F. Mouthereau,
C. Mehl (2010), Rifting and shallow-dipping detachments, clues from the Corinth
Rift and the Aegean, Tectonophysics , 483, 287–304.
Kahle, H.G., and S. Mueller (1998), Structure and dynamics of the Eurasian-
African/Arabian plate boundary system: Objectives, tasks and resources of the
WEGENER group, Journal of Geodynamics , 25, 303-325.
Kahle, H.G., M.V. Muller, S. Mueller, and G. Veis (1993), The Kephalonia Transform fault and the rotation of the Apulian platform: evidence from satellite geodesy,
Geophys. Res. Lett., 20(8), 651–654.
Kahle, H.G., M. Cocard, Y. Peter, A. Geiger, R. Reilenger, A. Barka, and G. Veis (2000),
GPS-derived strain rate field within the boundary zones of the Eurasian, African, and Arabian Plates, Journal of Geophysical Research , 105(B10), 23353-23370.
Kahle, H.G., M. Muller, A. Geiger, G. Danuser, S. Mueller, G. Veis, H. Billiris, and D.
Paradissis (1995), The strain field in northwestern Greece and the Ionian Islands: results inferred from GPS measurements, Tectonophysics , 249, 41-52.
201

Karagianni, E.E., C.B. Papazachos, D.G. Panagiotopoulos, P. Suhadolc, A. Vuan, and G.
Panza (2005), Shear velocity structure in the Aegean area obtained by inversion of Rayleigh waves, Geophys. J. Int., 160(1), 127–143.
Karagianni, E.E., and C.B. Papazachos (2007), Shear velocity structure in the Aegean region obtained by joint inversion of Rayleigh and Love waves, Geological
Society, London, Special Publications , 291, 159-181, doi:10.1144/SP291.8.
Karason, H. (2002) Constraints on mantle convection from seismic tomography and flow modeling, PhD thesis , Massachusetts Institute of Technology, Cambridge, MA,
USA.
Kennett, B.L.N. (1991), The removal of free surface interactions from three-component seismograms, Geophys. J. Int.
, 104, 153–163.
Kennett, B.L.N, and E.R. Engdahl (1991), Traveltimes for global earthquake location and phase identification. Geophys. J. Int ., 105(2), 429-465.
King, S.D. (2001), Subduction zones: observations and geodynamic models, Physics of the Earth and Planetary Interiors , 127, 9–24.
Knapmeyer, M., and H. Harjes (2000), Imaging crustal discontinuities and the downgoing slab beneath western Crete, Geophys. J. Int., 143(1), 1–21.
Kokinou, E., A. Vafidis, M. Loucogiannakis, and I. Louis (2003), Deep seismic imaging and velocity estimation in Ionian Sea, J. Balkan Geophys. Soc., 6(2), 100–116.
Kokinou, E., E. Kamberis, A. Vafidis, D. Monopolis, G. Ananiadis, and A. Zelilidis
(2005), Deep seismic reflection data from offshore western Greece: a new crustal model for the Ionian sea, J. Petr. Geol., 28(2), 185–202.
Konstantinou, K.I., and N.S. Melis (2008), High-frequency shear-wave propagation across the Hellenic subduction zone, Bull. Seismol. Soc. Am ., 98(2), 797 – 803, doi:
10.1785/0120060238
Kopf, A., J. Mascle, and D. Klaeschen (2003), The Mediterranean Ridge: a mass balance across the fastest growing accretionary complex on Earth, J. Geophys. Res.,
108(B8), 2372, doi:10.1029/2001JB000473.
Kreemer, C., N. Chamot-Rooke, and X. Le Pichon (2004), Constraints on the evolution and vertical coherency of deformation in the Northern Aegean from a comparison of geodetic, geologic and seismologic data, Earth Planet. Sci. Lett., 225, 329–346.
Lallemand, S., A. Heuret, and D. Boutelier (2005), On the relationships between slab dip, backarc stress, upper plate absolute motion, and crustal nature in subduction zones, Geochem. Geophys. Geosyst., 6 , Q09006, doi:10.1029/2005GC000917.
202

Lallemant, S., C. Truffert, L. Jolivet, P. Henry, N. Chamot-Rooke, and B. de Voogd
(1994), Spatial transition from compression to extension in the Western
Mediterranean Ridge accretionary complex, Tectonophysics , 234, 33-52.
Langston, C.A. (1977), The effect of planar dipping structure on source and receiver responses for the constant ray parameter, Bull. Seismol. Soc. Am., 67 , 1029–1050.
Langston, C.A. (1979), Structure under Mount Rainier, Washington, inferred from teleseismic body waves, J. Geophys. Res ., 84, 4749–4762.
Le Pichon, X., and J. Angelier (1979), The Hellenic arc and trench system: a key to the neotectonic evolution of the eastern Mediterranean area, Tectonophysics , 60, 1-
42.
Li, X., G. Bock, A. Vafidis, R. Kind, H.P. Harjes, W. Hanka, K. Wylegalla, M. van der
Meijde, and X. Yuan (2003), Receiver function study of the Hellenic subduction zone: imaging crustal thickness variations and the oceanic Moho of the descending African lithosphere, Geophys. J. Int., 155(2), 733–748.
Ligdas, C.N., I.G. Main and R.D. Adams (1990), 3-D structure of the lithosphere in the
Aegean region, Geophys. J. Int, 102(1), 219–229.
Lowrie, W. (2007). Fundamental of Geophysics (2 nd ed.). Retrieved from http://elibrary.bsu.az/azad/new/2194.pdf.
MacKenzie, L.M., G.A. Abers, S. Rondenay, and K.M. Fischer (2010), Imaging a steeply dipping subducting slab in southern Central America, Earth Planet. Sci. Lett .,
296, 459–468, doi:10.1016/j.epsl.2010.05.033.
Makris, J. (1978), The crust and upper mantle of the Aegean region from deep seismic soundings, Tectonophysics , 46, 269-284.
Makris, J. (1985), Geophysics and geodynamic implications for the evolution of the
Hellenides, in Geological Evolution of the Mediterranean Basin, edited by D.J.
Stanley and F. Wezel, 231–248, Berlin Springer.
Marone, F., M. van der Meijde, S. van der Lee, and D. Giardini (2003), Joint inversion of local, regional and teleseismic data for crustal thickness in the Eurasia-Africa plate boundary region, Geophys. J. Int., 154(2), 499–514.
Martínez, M.D., J.A. Canas, X. Lana, and J. Badal (2001), Objective regionalization of
Rayleigh wave dispersion data by clustering algorithms: an application to the
Mediterranean basin, Tectonophysics, 330, 245–266.
203

Mascle, J., and E. Chaumillon (1998), An overview of Mediterranean Ridge collisional accretionary complex as deduced from multichannel seismic data, Geo-Marine
Lett., 18(2), 81–89.
McClusky, S., S. Balassanian, A. Barka, C. Demir, S. Ergintav, I. Georgiev, O. Gurkan,
M. Hambourger, K. Hurst, H.G. Kahle, K. Kastens, G. Kekelidge, R. King, V.
Kotzev, O. Lenk, S. Mahmoud, A. Mishin, M. Nadariya, A. Ouzounis, D.
Paradissis, Y. Peter, M. Prilepin, R. Reilinger, I. Sanli, H. Seeger, A. Tealeb, M.
Toksoz, and G. Veis (2000), Global Positioning System constraints on plate kinematics and dynamic in the eastern Mediterranean and Caucasus: Journal of
Geophysical Research , 105 (B3), 5695-5719, doi:10.1016/j.tecto.2007.12.006.
McKenzie, D. (1972), Active tectonics of the Mediterranean region, Geophys. J. R. Astr.
Soc., 30(2), 109–185.
Meier, T., M. Rische, B. Endrun, A. Vafidis, and H.P. Harjes (2004), Seismicity of the
Hellenic subduction zone in the area of western and central Crete observed by temporary local seismic networks, Tectonophysics, 383, 149–169.
Meijer, P.Th., M. J. R. Wortel (1996), Temporal variation in the stress field of the
Aegean region, Geophys. Res. Lett.
, 23(5), 439-442.
Mercier, J.L., D. Sorel, P. Vergely, and K. Simeakis (1989), Extensional tectonic regimes in the Aegean basins during the Cenozoic, Basin Research , 2(1), 49-71.
Meyer, B., R. Armijo, and D. Dimitrov (2002), Active faulting in SW Bulgaria: possible surface rupture of the 1904 Struma earthquakes, Geophys. J. Int.
, 148, 246–255.
Miller, D., M. Oristaglio, and G. Beylkin (1987), A new slant on seismic imaging: migration and integral geometry, Geophysics, 52(7), 943–964.
Molnar, P. (1988), Continental tectonics in the aftermath of plate tectonics, Nature , 335,
131-137.
Monopolis, D., and A. Bruneton (1982), Ionian sea (Western Greece): Its structural outline deduced from drilling and geophysical data, Tectonophysics , 83, 227-242.
Morelli, C., C. Gantar, and M. Pisani (1975), Bathymetry, gravity and magnetism in the
Strait of Sicily and in the Ionian Sea, Bollettino di Geofisica Teorica ed
Applicata, 17, 39-58.
Moretti, I., and L.H. Royden (1988), Deflection, gravity anomalies and tectonics of doubly subducted continental lithosphere: Adriatic and Ionian seas, Tectonics ,
7(4), 875-893.
204

Papanikolaou, D. (1993), Geotectonic evolution of the Aegean, Bull. Soc. Geol. Greece ,
28(1), 33-48.
Papanikolaou, D. (2009), Timing of tectonic emplacement of the ophiolites and terrane paleogeography in the Hellenides, Lithos , 108, 262–280.
Papanikolaou, D., and L. Royden (2007), Disruption of the Hellenic arc: Late Miocene extensional detachment faults and steep Pliocene-Quaternary normal faults—Or what happened at Corinth?, Tectonics, 26, doi:10.1029/2006TC002007.
Papazachos, C., and G. Nolet (1997), P and S deep velocity structure of the Hellenic area obtained by robust nonlinear inversion of travel times, J. Geophys. Res., 102(B4),
8349–8367.
Papazachos, C.B., P.M. Hatzidimitriou, D.G. Panagiotopoulos, and G.N. Tsokas (1995),
Tomography of the crust and upper mantle in southeast Europe, J. Geophys. Res.,
100(B7), 12405-12422.
Papazachos, B.C., V.E. Karakostas, C.B. Papazachos, and E.M. Scordilis (2000), The geometry of the Wadati–Benioff zone and lithospheric kinematics in the Hellenic arc, Tectonophysics, 319(4), 275–300.
Pasyanos, M.E., and W.R. Walter (2002), Crust and upper-mantle structure of North
Africa, Europe and the Middle East from inversion of surface waves, Geophys. J.
Int., 149(2), 463–481.
Pearce, F.D., S. Rondenay, M. Sachpazi, M. Charalampakis, and L.H. Royden (2012),
Seismic investigation of the transition from continental to oceanic subduction along the western Hellenic Subduction Zone. J. Geophys. Res.
, 117(B07306), http://dx.doi.org/10.1029/2011JB009023.
Pe-Piper, G., and D.J.W. Piper (2007), Neogene backarc volcanism of the Aegean: New insights into the relationship between magmatism and tectonics, Geological
Society of America Special Papers, 418, 17–31, doi: 10.1130/2007.2418(02).
Piana Agostinetti, N., I. Bianchi, A. Amato, and C. Chiarabba (2011), Fluid migration in continental subduction: The Northern Apennines case study, Earth planet. Sci.
Lett., 302, 267–278, doi:10.1016/j.epsl.2010.10.039.
Piromallo, C., and A. Morelli (2003), P wave tomography of the mantle under the
Alpine-Mediterranean area, J. Geophys. Res., 108(B2), 2065.
Reilinger, R., S. McClusky, D. Paradissis, S. Ergintav, and P. Vernant (2010), Geodetic constraints on the tectonic evolution of the Aegean region and strain accumulation along the Hellenic subduction zone, Tectonophysics , 488, 22-30, doi:10.1016/j.tecto.2009.05.027.
205

Roberts, S., and J. Jackson (1991), Active normal faulting in central Greece: an overview,
Geological Society, London, Special Publications , 56(1), 125-142.
Rondenay, S. (2009), Upper mantle imaging with array recordings of converted and scattered teleseismic waves, Surv. Geophys ., 30, 377-405, doi:10.1007/s10712-
009-9071-5.
Rondenay, S. M.G. Bostock, T.M. Hearn, D.J. White, and R.M. Ellis (2000),
Lithospheric assembly and modification of the SE Canadian Shield: Abitibi-
Grenville teleseismic experiment, J. Geophys. Res.
, 105(B6), 13,735-13,754
Rondenay, S., M.G. Bostock, and J. Shragge (2001), Multiparameter two-dimensional inversion of scattered teleseismic body waves, 3. Application to the Cascadia
1993 data set, J. Geophys. Res., 106(12), 30795–30807.
Rondenay, S., M.G. Bostock, and K. Fischer (2005), Multichannel inversion of scattered teleseismic body waves: practical considerations and applicability, in Seismic
Earth: Array Analysis of Broadband Seismograms, 157, edited by A. Levander and G. Nolet, AGU Geophysical Monograph Series, 187–204.
Rondenay, S., G.A. Abers, and P. van Keken (2008), Seismic imaging of subduction zone metamorphism, Geology, 36(4), 275–278.
Rondenay, S., L.G.J. Montési, and G.A. Abers (2010), New geophysical insight into the origin of the Denali volcanic gap, Geophys. J. Int ., 182, 613-630.
Rossi, G., G. A. Abers, S. Rondenay, and D. H. Christensen (2006), Unusual mantle
Poisson’s ratio, subduction, and crustal structure in central Alaska, J. Geophys.
Res.
, 111, B09311, doi:10.1029/2005JB003956.
Royden, L.H. (1993a), The tectonic expression of slab pull at continental convergent boundaries, Tectonics , 12, 303-325.
Royden, L.H. (1993b), Evolution of retreating subduction boundaries formed during continental collision, Tectonics , 12, 629-638.
Royden, L.H., and L. Husson (2006), Trench motion, slab geometry and viscous stresses in subduction systems, Geophys. J. Int, 881-905, 167, doi: 10.1111/j.1365-
246X.2006.03079.x.
Royden, L.H., and L. Husson (2009), Subduction with variations in slab buoyancy:
Models and application to the Banda and Apennine systems, in Subduction Zone
Geodynamics, edited by S. E. Lallemand and F. Funiciello, 35 45, Springer,
Berlin, doi:10.1007/978-3-540-87974-9_2.
206

Royden, L.H., and D. Papanikolaou (2011), Slab segmentation and late Cenozoic disruption of the Hellenic arc, Geochem. Geophys. Geosyst.
, 12(3), 1-24, doi:10.1029/2010GC003280.
Rychert, C.A., K.M. Fischer, S. Rondenay (2005), A sharp lithosphere–asthenosphere boundary imaged beneath eastern North America, Nature , 436, 542–545.
Sachpazi, M., A. Galvé, M. Laigle, A. Hirn, E. Sokos, A. Serpetsidaki, J.-M. Marthelot,
J.M. Pi Alperin, B. Zelt, and B. Taylor (2007), Moho topography under central
Greece and its compensation by Pn time-terms for the accurate location of hypocenters: the example of the Gulf of Corinth 1995 Aigion earthquake,
Tectonophysics, 440, 53–65, doi:10.1016/j.tecto.2007.01.009.
Saunders, P., K. Priestley, and T. Taymaz (1998), Variations in the crustal structure beneath western Turkey, Geophys. J. Int., 134(2), 373–389.
Savage, M.K. (1998), Lower crustal anisotropy or dipping boundaries? Effects on receiver functions and a case study in New Zealand, J. Geophys. Res., 103 ,
15,069–15,087.
Schmid, C., S. van der Lee, and D. Giardini (2004), Delay times and shearwave splitting in the Mediterranean region, Geophys. J. Int., 159(1), 275–290.
Schmid, C., S. van der Lee, and D. Giardini (2006), Correlated shear and bulk moduli to
1400 km beneath the Mediterranean region, Phys. Earth planet. Int., 159, 213-
224.
Sengor, A.M.C., N. Gorur, and F. Saroglu (1985), Strike-slip faulting and related basin formation in zones of tectonic escape: Turkey as a case study, in Strike-Slip
Faulting and Basin Formation, edited by K.T. Biddle and N. Christie-Blick, Spec.
Publ. SEPM Soc. Sediment. Geol., 37, 227-264.
Shragge, J., M.G. Bostock, and S. Rondenay (2001), Multiparameter two-dimensional inversion of scattered teleseismic body waves: 2. Numerical examples, J.
Geophys. Res, 106(12), 30783-30793.
Snopek, K., T. Meier, B. Endrun, M. Bohnhoff, and U. Casten (2007), Comparison of gravimetric and seismic constraints on the structure of the Aegean lithosphere in the forearc of the Hellenic subduction zone in the area of Crete, J. Geodyn., 44,
173–185.
Sodoudi, F., R. Kind, D. Hatzfeld, K. Priestley, W. Hanka, K. Wylegalla, G. Stavrakakis,
A. Vafidis, H.P. Harjes, and M. Bohnhoff (2006), Lithospheric structure of the
Aegean obtained from P and S receiver functions, J. Geophys. Res., 111(B12),
B12307, doi:10.1029/2005JB003932.
207

Spakman, W., M.J.R. Wortel, and N.J. Vlaar (1988), The Hellenic subduction zone: a tomographic image and its geodynamic implications, Geophys. Res. Lett., 15(1),
60–63.
Spakman, W., S. van der Lee, and R. van der Hilst (1993), Travel-time tomography of the
European-Mediterranean mantle down to 1400 km, Physics of The Earth and
Planetary Interiors , 79, 3-74.
Suckale, J., S. Rondenay, M. Sachpazi, M. Charalampakis, A. Hosa, and L.H. Royden
(2009), High-resolution seismic imaging of the western Hellenic subduction zone using teleseismic scattered waves, Geophys. J. Int.
, 178(2), 775-791, doi:10.1111/j.1365-246X.2009.04170.x.
Syracuse, E.M., P.E. van Keken, and G.A. Abers (2010), The global range of subduction zone thermal models, Phys. Earth Planet. Inter ., 51(8), 1761–1782, doi:10.1016/j.pepi.2010.02.004.
Taymaz, T., J. Jackson, and D. McKenzie (1991), Active tectonics of the north and central Aegean Sea, Geophys. J. Int.
, 106, 433-490.
Tiberi, C., H. Lyon-Caen, D. Hatzfeld, U. Achauer, E. Karagianni, A. Kiratzi, E. Louvari,
D. Panagiotopoulos, I. Kassaras, G. Kaviris, K. Makropoulos, and P.
Papadimitriou (2000), Crustal and upper mantle structure beneath the Corinth rift
(Greece) from a teleseismic tomography study, J. Geophys. Res., 105(B12),
28159–28171.
Tiberi, C., M. Diament, H. Lyon-Caen, and T. King (2001), Moho topography beneath the Corinth Rift area (Greece) from inversion of gravity data, Geophys. J. Int.,
145(3), 797–808.
Tirel, C., F. Gueydan, C. Tiberi, and J.P. Brun (2004), Aegean crustal thickness inferred from gravity inversion. Geodynamical implications, Earth planet. Sci. Lett., 228,
267–280.
Tirel, C., J.P. Brun, E. Burov, M.J.R. Wortel, and S. Lebedev (2013), A plate tectonics oddity: Caterpillar-walk exhumation of subducted continental, Geology , 41, 555-
558.
Truffert, C., N. Chamot-Rooke, S. Lallemant, B. De Voogd, P. Huchon and X. Le Pichon
(1993), The crust of the Western Mediterranean Ridge from deep seismic data and gravity modeling, Geophys. J. Int.
, 114,360-372.
Tsokas, G.N., and R.O. Hansen (1997), Study of the crustal thickness and the subducting lithosphere in Greece from gravity data, J. Geophys. Res., 102(B9), 20 585–20
597.
208

VanDecar, J. C., and R. S. Crosson (1990), Determination of teleseismic relative phase arrival times using multi-channel cross-correlation and least squares, Bull.
Seismol. Soc. Am ., 80, 150–159. van der Hilst, R.D., S. Widiyantoro, and E.R. Engdahl (1997), Evidence for deep mantle circulation from global tomography, Nature, 386(6625), 578–584. van der Meijde, M., S. van der Lee, and D. Giardini (2003), Crustal structure beneath broad-band seismic stations in the Mediterranean region, Geophys. J. Int., 152(3),
729–739. van Hinsbergen, D.J.J., E. Hafkenscheid, W. Spakman, J.E. Meulenkamp, and M.J.R.
Wortel (2005), Nappe stacking resulting from subduction of oceanic and continental lithosphere below Greece, Geology, 33(4), 325-328, DOI:
10.1130/G20878.1. van Keken, P.E., B.R. Hacker, E.M. Syracuse, and G.A. Abers (2011), Subduction factory: 4. Depth ‐ dependent flux of H2O from subducting slabs worldwide , J.
Geophys. Res ., 116, B01401, doi:10.1029/2010JB007922.
Vassilakis, E., L. Royden, and D. Papanikolaou (2011), Kinematic links between subduction along the Hellenic trench and extension in the Gulf of Corinth,
Greece: A multidisciplinary analysis, Earth Planet. Sci. Lett.
, doi:10.1016/j.epsl.2010.12.054.
Vinnik, L. (1977), Detection of waves converted from P to SV in the mantle, Phys. Earth
Planet Int.
, 15, 39–45.
Wada, I., and K.L. Wang (2009), Common depth of slab ‐ mantle decoupling:
Reconciling diversity and uniformity of subduction zones, Geochem. Geophys.
Geosyst ., 10, Q10009, doi:10.1029/2009GC002570.
Warren, C.J., C. Beaumont, and R.A. Jamieson (2008), Modelling tectonic styles and ultra-high pressure (UHP) rock exhumation in the transition from oceanic subduction to continental collision, Earth Planet. Sci. Lett.
, 267, 129–145, doi:10.1016/j.epsl.2007.11.025.
Waschbusch, P., and C. Beaumont (1996), Effect of slab retreat on crustal deformation in simple regions of plate convergence, J. Geophys. Res ., 101, 28133-28148.
Wortel, M.J.R., and W. Spakman (2000), Subduction and Slab Detachment in the
Mediterranean-Carpathian Region, Science , 290(5498), 1910-1917, doi:10.1126/science.290.5498.1910.
209

Zelt, B.C., B. Taylor, M. Sachpazi, and A. Hirn (2005), Crustal velocity and Moho structure beneath the Gulf of Corinth, Greece, Geophys. J. Int., 162(1), 257–268, doi:10.1111/j.1365-246X.2005.02640.x.
Zhu, L. and H. Kanamori (2000), Moho depth variation in southern California from teleseismic receiver functions, J. Geophys. Res., 105 , 2969–2980.
Zhu, L.P., B.J. Mitchell, N. Akyol, I. Cemen, and K. Kekovali (2006), Crustal thickness variations in the Aegean region and implications for the extension of continental crust, J. Geophys. Res, 111(B1), B0130.
210