Hannemann Thesis

Institut für Erd- und Umweltwissenschaften
Arrayseismologie
Seismological investigation of the oceanic crust and upper

mantle using an ocean bottom station array in the
vicinity of the Gloria fault (eastern mid Atlantic)
Kumulative Dissertation
zur Erlangung des akademischen Grades
“doctorum rerum naturalium”
(Dr.rer.nat.)
in der Wissenschaftsdisziplin “Geophysik”
eingereicht an der
Mathematisch-Naturwissenschaftlichen Fakultät
der Universität Potsdam
von
Katrin Hannemann
Potsdam, den 21.09.2016

2
Contents
1 Zusammenfassung 5
2 Abstract 7
3 Introduction 9
3.1 Challenges at the sea floor . . . . . . . . . . . . . . . . . . . . . 9
3.2 Structure of the oceanic crust and upper mantle . . . . . . . . . 11
3.3 Research questions . . . . . . . . . . . . . . . . . . . . . . . . . 13
4 Measuring clock drift and time offset at OBS 21

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.4.1 Estimation of clock drifts . . . . . . . . . . . . . . . . . . 27
4.4.2 Constant time shifts . . . . . . . . . . . . . . . . . . . . 27
4.4.3 Shifted correlation time window approach . . . . . . . . 28
4.4.4 Processing of data set . . . . . . . . . . . . . . . . . . . 28
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.5.1 Correlation of vertical seismic components . . . . . . . . 29
4.5.2 Shifted correlation time window approach . . . . . . . . 31
4.5.3 Correlation of hydrophone signals . . . . . . . . . . . . . 33
4.6 Conclusion and outlook . . . . . . . . . . . . . . . . . . . . . . . 34
5 Structure of oceanic crust and upper mantle by OBS RF 37

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.S1 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.S2 T components of RF . . . . . . . . . . . . . . . . . . . . . . . . 66
5.S3 Event tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3
Contents
6 Oceanic lithospheric vs from the analysis of Ïp 69

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.1.1 Previous studies of oceanic S wave velocity . . . . . . . . 71
6.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.4 Synthetic tests . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.4.1 Half space S wave velocity . . . . . . . . . . . . . . . . . 77
6.4.2 Depth resolution . . . . . . . . . . . . . . . . . . . . . . 80
6.4.3 Influence of water depth . . . . . . . . . . . . . . . . . . 82
6.4.4 Influence of LVL . . . . . . . . . . . . . . . . . . . . . . 84
6.4.5 Summary of synthetic tests . . . . . . . . . . . . . . . . 86
6.5 Application to real data . . . . . . . . . . . . . . . . . . . . . . 86
6.5.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.5.2 Quantitative modelling approach . . . . . . . . . . . . . 91
6.5.3 Results of quantitative modelling . . . . . . . . . . . . . 93
6.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.6.1 Discussion of quantitative modelling results . . . . . . . 98
6.6.2 Discussion of method . . . . . . . . . . . . . . . . . . . . 101
6.7 Conclusion and Outlook . . . . . . . . . . . . . . . . . . . . . . 102
6.S1 Derivation of equation (3.10) . . . . . . . . . . . . . . . . . . . . 110
6.S2 Additional figures . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7 Discussion 117
8 Conclusion 129
A Data 131
B Orientation of OBS using P phase 145
C Reflection and refraction at the ocean bottom 147

C.1 SH-case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
C.2 P-SV-case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
C.2.1 Apparent angle of incidence (C.48) . . . . . . . . . . . . 170
C.3 SV-P-case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
C.3.1 Apparent angle of incidence (C.82) . . . . . . . . . . . . 192
D Acknowledgement 197
E Eidesstattliche Erklärung 199
4
1. Zusammenfassung
Obwohl der Planet Erde zu mehr als 70% von Ozeanen bedeckt ist, befinden
sich die meisten Seismometer (Messinstrumente zur Aufzeichnung von Bodenbe-
wegungen) an Land. Dies führt dazu, dass trotz eines globalen Netzwerks aus
Seismometern der innere Aufbau der Erde unter den Ozeanen weniger bekannt
ist als unter den Kontinenten. Es gibt spezielle Ozeanbodenstationen (OBS),
die mit einem Seismometer und einem Drucksensor ausgestattet sind und für
den Einsatz am Ozeanboden konzipiert wurden. In dieser Arbeit werden 12
OBSe verwendet, die in der Tiefsee des östlichen Mittelatlantik installiert waren
und sich ca. 100 km nördlich von der eurasisch-afrikanischen Plattengrenze
(Gloriaverwerfung) befunden haben. Für die Bestimmung von Uhrzeitdriften
und konstanten Zeitversätzen habe ich eine Methode entwickelt, die durch die
Anwendung von Kreuzkorrelationen der Bodenunruhe deren Bestimmung allein
anhand der Daten ermöglicht.
Mit den so vorbereiteten Daten können sogenannte Receiver Funktionen
(RF) berechnet werden. Diese dienen dazu die oft sehr kleinen Signale, die
durch eine P Welle an Grenzflächen unterhalb der Station erzeugt werden,
in den Daten zu verstärken und damit analysierbar zu machen. Einen Hin-
weis auf die tatsächlichen seismischen Geschwindigkeiten können die am Seis-
mometer gemessenen Auftauchwinkel der P Welle liefern, welche mit Hilfe
von entsprechenden Relationen in S-Wellengeschwindigkeiten übersetzt werden
können. Ich habe die gültige Relation für den Meeresboden mit Hilfe von
verschiedenen Tiefpassfiltern, sowie einer quantitativen Modellierung der Daten
benutzt, um S-Wellengeschwindigkeitstiefenprofile für jedes OBS zu bestimmen.
Eine Gegenüberstellung der Modelle ergab, dass eine durch Serpentinisierung
verursachte deutliche Geschwindigkeitsabnahme und eine leichte Krustenverdick-
ung in Richtung der Gloriaverwerfung zu beobachten ist. Ich konnte für die
RF zeigen, dass eine Qualitätskontrolle mit verschiedenen Gütekriterien, sowie
eine kohärente Stapelung der Erdbebensignale (Beamforming) die Qualität der
RF verbessern kann. Zusätzlich ist es durch das Beamforming möglich die
Anzahl der zur Analyse bereitstehenden Ereignisse zu erhöhen. Eine gegenüber
von Modellen flachere Lithosphären-Asthenosphärengrenze und eine zusätzlich
beobachtete Grenzfläche in der Asthenosphäre deuten auf das Vorhandensein
von partiellen Schmelzen hin. Zudem kann aus der Beobachtung des Signals
der Asthenosphärenunterkante, auf einen gewissen Anteil an Fluiden im oberen
Erdmantel geschlossen werden. Die Grenzflächen der Mantelübergangszone
werden meist in den erwarteten Tiefen beobachtet, allerdings gibt es für die
untere Grenzfläche bei 660 km für einige Erdbebenrichtungen eine zusätzliches
Signal, dass mit einem Phasenübergang in den Aluminumphasen des Mantels
zusammenhängen kann.
Zusammenfassend bietet diese Arbeit einen Katalog an Methoden zur
Vorbearbeitung von OBS Daten und zur Auswertung von Receiver Funktionen.
5
Chapter 1. Zusammenfassung
6
2. Abstract
In recent years, the investigation of the oceanic crust has been extended
from mainly land based operations to more experiments on the ocean bottom.
The installation of ocean bottom stations (OBS) on the sea floor offers the
opportunity to characterize the oceanic crust and mantle with a higher spatial
resolution compared to global surface wave tomography or underside reflection
analysis. In 2011, twelve broadband OBS were installed in the deep sea (4.5-
5.5 km) of the eastern mid Atlantic approximately 100 km North of the Gloria
fault which coincides with Eurasian-African plate boundary in this region.
During this thesis, I find that it is possible to retrieve lost clock drifts
and estimate static time offsets from OBS data by using ambient noise cross-
correlation. In the course of the analysis of receiver functions (RF) using OBS
data, we find that beamforming is a feasible technique to increase the number
of usable events for the analysis and that the employment of a quality control
facilitates the selection of suiting processing parameters to obtain RF with a
good signal-to-noise ratio (SNR). Furthermore, the analysis of RF revealed
that the oceanic crust has a normal thickness of ≥6.7 km and a slight crustal
thickening towards the Gloria fault is observed from the P wave polarization
analysis. The latter might be related to a known transpression in the study
area. Furthermore, an increase in the uppermost mantle S wave velocities
towards the fault can be identified by the P wave polarization analysis which
is probably related to the increasing amount of serpentinized mantle in the
vicinity of the fault. Moreover, we observe that the lithosphere-asthenosphere
boundary (LAB) is located at shallower depths than would be expected from
models of the oceanic lithosphere. Additionally, we identified a positive RF
phase which we associate to a mid asthenospheric discontinuity (MAD). Both
features indicate a high temperature anomaly and the influence of partial
melt in the upper asthenosphere. The signal of the lower boundary of the
asthenosphere (Lehmann discontinuity) arrives a bit earlier than expected
from global Earth models. This might be related to a wet uppermost mantle.
The RF phases of the mantle discontinuity at 410 km (’410’) and at 660 km
(’660’) at least for the single stations arrives at the expected time. The ’660’
RF phase on the beamformed RFs is slightly delayed compared to the global
Earth model, furthermore, we observe an additional phase for some azimuths.
This might be related to an additional transition in the Aluminium phases
(majorite-perovskite) of the mantle.
In conclusion, this thesis provides methods for the pre-processing and the
analysis of OBS data: (1) estimation of clock drifts and static time offsets, (2)
receiver function analysis using a quality control and beamforming to increase
the signal-to-noise ratio of the RF and (3) analysis of the RF in terms of S
wave velocity-depth models using the P wave polarization.
7
Chapter 2. Abstract
8
3. Introduction
About two third of the Earth are covered by oceans, but most seismological
stations are located on-shore and our knowledge of the oceanic crust and mantle
structures is therefore limited (Webb, 1998). The deployment of a seismological
station at the ocean bottom (ocean bottom station, OBS) has to deal with
high-pressure conditions, usually a lack of permanent power supply, and a lack
of radio, light or GPS signals for communication or data transfer (e.g Suetsugu
and Shiobara, 2014). In the 1980s, short period (1-5 Hz) OBS were developed
which were equipped with pressure cases, stable clocks, and robust and compact
sensors and dataloggers, broad-band OBS were developed based on them in
the 1990s (Suetsugu and Shiobara, 2014). Most of the used OBSs are free-fall
pop-up stations which are installed by releasing them from a ship and a descend
to the sea floor, and which are recovered by an acoustic signal which results in
a detachment of station and anchor, and a rise of the station to the sea surface
(Trnkoczy et al., 2012). In recent years, the number of - especially large scale -
OBS experiments (e.g. RHUM-RUM, Barruol and Sigloch, 2013, Cascadia, Gao
and Schwartz, 2015) increased worldwide. In parallel, the number of studies
applying standard seismological analysis methods to OBS data has also grown
(e.g. Thomas and Laske, 2014; Audet, 2016). This thesis consists of three
publications which deal with the pre-processing of OBS data and the analysis
of receiver functions at the ocean bottom in terms of discontinuity depths and
S wave velocity-depth models.
3.1 Challenges at the sea floor

During the processing of OBS data, several things have to be taken into account:
the position, the orientation and the clock of the station, the influence of the
water column on the recorded data and the noise at the ocean bottom.
The position of an OBS is usually chosen based on bathymetric data which
were acquired before the deployment. These data are used to determine
rather flat areas and to exclude areas below the maximum operation depth of
the instruments (usually ≥6000 m, e.g. Trnkoczy et al., 2012). The positions
themselves are determined by the ship positions during deployment and recovery.
The water depth is estimated from either a local bathymetry or the ship’s echo
sounder (e.g. Trnkoczy et al., 2012). Additionally, the estimated position can be
improved by employing range measurements and perform triangulations similar
to absolute geodetic sea floor measurements (e.g Bürgmann and Chadwell,
2014, red open triangles in Fig. 3.2). This technique is rather time and energy
consuming and therefore often just the ship position during recovery and
deployment are used.
9
Chapter 3. Introduction
In order to analyse earthquake data, we need to know the orientation of the

seismometer components. For three component broad-band stations, a gimbal
mechanism ensures that the vertical component is indeed vertical and is able
to correct for a tilt of .20¶ of the station on the sea floor (e.g Trnkoczy et al.,
2012). During the descend of the OBS to the sea floor, it is likely that the
OBS is rotating around its vertical axis and therefore the orientation of the
horizontal components at the sea floor is unknown. There exist several methods
to estimate the proper orientation along North and East. These methods
employ either earthquake signals like P and Rayleigh waves (e.g Stachnik et al.,
2012, see also appendix 5.S1 in chapter 5) or ambient noise recordings (e.g Zha
et al., 2013). There exist also techniques for the relative orientation between
stations (e.g. Grigoli et al., 2012).
The internal clocks of OBSs can only be synchronized with a GPS signal
during the deployment and the recovery and are therefore required to be stable
i.e. having only a small linear drift (Trnkoczy et al., 2012, e.g). During the
second synchronization with a GPS signal, the clock difference to the GPS time
(skew time) is estimated and used for the time correction during the analysis
of the data by assuming a linear drift (e.g Trnkoczy et al., 2012; Hannemann
et al., 2014). The measured skew times are usually in the order of ≥1-2 s
(e.g. Hannemann et al., 2014). Nevertheless, the measurement might be either
biased by a failed first synchronization or impossible due to a shut-down of the
station (e.g. Hannemann et al., 2014; Stähler et al., 2016).
If seismological analysis techniques are transferred from land stations to
OBSs, the influence of the water column has to be taken into account. This
might be done by either modelling synthetic data for the sea floor case (e.g.
Audet, 2016) or theoretical estimating the effects encountered at the sea floor
(e.g. chapter 6, Hannemann et al., 2016, and appendix C). The obtained
effects have to be taken into account during the analysis of OBS data.
Additionally, there are several sources of noise at the ocean bottom which
have to be taken into account during the processing of OBS data: tilt noise,
compliance noise, microseism noise (Bell et al., 2015) and resonance caused by
sediments with unknown but likely extreme properties (chapter 6 Hannemann
et al., 2016). The tilt noise is created by the movement of the station’s case
by currents at the ocean bottom and is usually stronger on the horizontal
components than on the vertical components of the seismometer (>10 s, e.g.
Crawford and Webb, 2000; Bormann and Wielandt, 2013; Stähler et al., 2016).
The compliance noise is induced by oceanic infragravity waves and its frequency
and amplitude strongly depends on the water depth (e.g Crawford et al., 1998;
Bell et al., 2015). For deep water depths, the frequency of the compliance noise
is lower than for shallow water depths (e.g. ≥100 s for water depths of 4.5-
5.5 km). Microseismic noise is known nearly since the beginning of seismological
instrumental recording and has been identified to be induced by the movement
of water waves (e.g. Longuet-Higgins, 1950; Hasselmann, 1963; Bormann and
Wielandt, 2013). Often, two distinct ocean microseism peaks can be identified
in noise spectra: the primary ocean microseisms at ≥14±2 s and the secondary
ocean microseisms at ≥6±2 s (Bormann and Wielandt, 2013). The primary
ocean microseisms are generated by ocean waves in shallow waters of coastal
areas in which the waves’ energy is transformed into seismic energy by either
vertical pressure variations or interactions with coastlines (e.g. Hasselmann,
1963). The period of the primary ocean microseisms is similar to the period of
10
3.2. Structure of the oceanic crust and upper mantle
the water waves which generate them. For OBS recordings, the primary ocean
microseisms are mainly important for deployments close to coastal regions
as their energy decay rapidly away from their origin or for all deployments
if storm waves interact with remote coastlines (e.g Bormann and Wielandt,
2013). The secondary ocean microseisms are explained by the superposition of
ocean waves with equal period which travel in opposite directions (Longuet-
Higgins, 1950; Hasselmann, 1963). The generated standing gravity waves have
half the period of the ocean waves and induce pressure perturbations which
travel without attenuation to the ocean bottom at which they are converted
to seismic energy (e.g. Bormann and Wielandt, 2013). As for land stations,
the secondary ocean microseisms have larger amplitudes than the primary
ocean microseisms at OBS (e.g. Bormann and Wielandt, 2013). Contrary to
land stations, the spectral peak of the secondary microseisms is wider and
more noise energy is present at frequencies close to ≥1 Hz (e.g. Webb, 1998).
Due to these high noise levels, short period teleseismic body waves are often
hardly detectable unless for larger magnitude earthquakes (mb &7.5, Webb,
1998). Nevertheless, the noise level in the North Atlantic ocean is especially in
calm-weather periods lower than for example in the Pacific ocean (Webb, 1998;
Bormann and Wielandt, 2013). Additionally, long period body and surface
waves are often well recorded by OBS even for teleseismic distances of several
tens of degrees (magnitudes&6, Bormann and Wielandt, 2013). Resonance or
reverberation caused by a sedimentary cover are observed for the OBS data
used in this thesis and we could relate this resonance effect to the sedimentary
cover (chapter 6 Hannemann et al., 2016). It might be caused by trapped wave
energy and is triggered by ambient noise as well as body waves (chapter 6
Hannemann et al., 2016).
3.2 Structure of the oceanic crust and upper

mantle
The velocity-depth model of the oceanic crust and uppermost mantle extracted
from the preliminary reference Earth model (PREM, Dziewonski and Anderson,
1981) has a crustal thickness of 21.4 km. This thickness is large compared to
estimations of seismic refraction experiments (≥7 km, e.g. White, 1984; White
et al., 1992) which have been used to characterize the P wave velocity-depth
model of the oceanic crust and uppermost mantle. In general, the derived
velocity-depth models are divided into four layers. A sedimentary layer with
variable thickness (Layer 1), an upper crustal layer with a thickness of ≥2 km
(Layer 2), a lower crustal layer with a thickness of ≥5 km (Layer 3) and a
mantle layer (Layer 4) below (e.g. White, 1984; White et al., 1992, Fig. 3.1).
Furthermore, Laske et al. (2013) provide a 1-degree global model of Earth’s crust
(CRUST1.0) which is based on the analysis and validation with a surface wave
analysis. In Fig. 3.1, we give a comparison of all mentioned oceanic crust and
uppermost mantle velocity-depth model. In order to provide S wave velocities
for the model of White et al. (1992), we used theÔmudrock-line (Castagna et al.,
1985) for the sedimentary layer, a vp /vs ratio of 3 for the crustal layers and a
vp /vs ratio of 1.8 for the uppermost mantle. The model of White et al. (1992)
and CRUST1.0 are quite similar except for the gradients in the crust and a
slower uppermost mantle velocity in White et al. (1992).
11
0 0
10 10
20 20 PREM
White92
CRUST1.0
0 2 4 6 8 10 0 1 2 3 4 5
Figure 3.1: Velocity-depth models for oceanic crust and uppermost mantle: PREM
(blue, Dziewonski and Anderson, 1981), White92 (red, White et al., 1992, fig. 1)
and CRUST1.0 (yellow, Laske et al., 2013, at 38.5¶ N, -18.5¶ E).
The depths of deeper discontinuities in the upper mantle are known by

receiver function studies (e.g. Shen et al., 1998a; Chevrot et al., 1999; Gilbert
et al., 2001; Suetsugu et al., 2005, 2007; Lawrence and Shearer, 2006; Tauzin
et al., 2008; Suetsugu et al., 2010), precursor studies (e.g. Gossler and Kind, 1996;
Gu et al., 1998; Flanagan and Shearer, 1998; Gu and Dziewonski, 2002; Deuss
et al., 2013; Saki et al., 2015) or global Earth’s models (e.g. Dziewonski and
Anderson, 1981; Kennett et al., 1995) as well as by rheological and petrological
studies of the oceanic lithosphere (e.g Karato and Jung, 1998; Karato, 2012;
Olugboji et al., 2013) and the upper mantle (e.g Karato, 1992; Weidner and
Wang, 1998).
The first major discontinuity in the mantle is the lithosphere-asthenosphere
boundary (LAB) at which a mechanically strong lithosphere above a weak
asthenosphere is observed (e.g. Karato, 2012; Olugboji et al., 2013). The
oceanic LAB is detected in different depths and its cause has been under
debate (e.g Kawakatsu et al., 2009; Fischer et al., 2010; Olugboji et al., 2013).
Comparisons of mineral-physics modelling and seismological signature (Olugboji
et al., 2013) indicate a thermally controlled, age-dependent, diffuse LAB for
young oceanic lithosphere and a sharp, constant (≥70 km) LAB for old oceanic
lithosphere marking an abrupt change in water content from a dry depleted
oceanic lithosphere to a hydrated fertile asthenosphere (e.g. Fischer et al., 2010).
The age, at which the transition from the thermal control to the dominance
of the change in water content happens, depends on the used thermal model
(40-80 Ma, Karato, 2012 or 55-75 Ma, Olugboji et al., 2013).
In the global Earth model PREM (Dziewonski and Anderson, 1981), a
discontinuity at 220 km marks the boundary between an anisotropic crust and
mantle above and an isotropic mantle below. This discontinuity is referred
to as Lehmann discontinuity and was first observed by Lehmann (1961). In
12
3.3. Research questions
continental areas, the discontinuity often coincides with the base of the astheno-
sphere (e.g. Deuss et al., 2013), but it was also observed in oceanic regions (e.g.
Shen et al., 1998a; Deuss and Woodhouse, 2002).
The three global upper mantle discontinuities at ≥410 km depth (’410’),
≥520 km depth (’520’) and ≥660 km depth (’660’) are related to phase tran-
sitions in olivine and Aluminium phases (e.g. garnet) in mantle (e.g Agee,
1998; Helffrich, 2000; Deuss et al., 2013). The part of the upper mantle which
lies between the ’410’ and the ’660’ is often referred to as mantle transition
zone (MTZ). The ’410’ is most likely related to the transition from olivine
(–) to —-spinel (—), the ’520’ to the transition from — to “-spinel (“) and the
’660’ to the transition from either “ to magnesiowustite (mw) and silicate
perovskite (pv), garnet (gt) to ilmenite (il) and at larger depth to pv or in
hotter regions gt to pv (Estabrook and Kind, 1996; Vacher et al., 1998; Agee,
1998; Helffrich, 2000). The phase transitions of the ’410’ and the ’520’ have a
positive Clapeyron slope which means that an elevation of the discontinuity is
expected in cold regions and an depression in warm regions (Helffrich, 2000).
The Clapeyron slopes of the phase transitions of the ’660’ are either negative
(“ to mw and pv) or positive (gt to il to pv or gt to pv) whereby the first three
transitions dominate in cold mantle regions, leading to multiple transitions
(Simmons and Gurrola, 2000; Deuss et al., 2013) and the last is dominant in
extreme hot regions (Weidner and Wang, 1998; Vacher et al., 1998; Hirose, 2002;
Deuss et al., 2006, 2013). All mantle discontinuities have been observed in the
oceans by precursor (e.g Gossler and Kind, 1996; Gu et al., 1998; Flanagan and
Shearer, 1998; Gu and Dziewonski, 2002; Deuss et al., 2006, 2013; Saki et al.,
2015) and RF studies (e.g. Shen et al., 1998a,b; Chevrot et al., 1999; Gilbert
et al., 2001; Suetsugu et al., 2005; Lawrence and Shearer, 2006; Suetsugu et al.,
2007; Tauzin et al., 2008; Silveira et al., 2010; Suetsugu et al., 2010)
3.3 Research questions

This thesis works with a data set collected at 12 OBSs which were located
≥100 km North of the Gloria fault. This fault marks the Eurasian-African
plate boundary in the eastern mid Atlantic (Fig. 3.2). The OBSs were deployed
as part of the DOCTAR (Deep OCean Test ARray) project which consisted of
three array running in parallel on the ocean floor, at an ocean island (Madeira)
and in a continental setting (Portugal). The OBSs were installed to form a mid
aperture array (≥75 km, for further details see appendix A). The main goal of
this thesis is the investigation of the oceanic crust and the upper mantle North
of the Gloria fault. In our investigation, we analyse whether the neighbouring
fault has a visible influence on the crust and upper mantle structure. We focus
on discontinuity depths in the mantle and their implications for temperature
and petrology. Furthermore, we have a look on the sharpness of discontinuities
which gives further evidence for possible temperature gradients.
For this purpose, we use receiver functions (RF), which result from the
deconvolution of teleseismic event recordings (epicentral distances>30¶ ) with
their P wave signal to enhance P-to-S (Ps) conversions from different discon-
tinuities (e.g. Vinnik, 1977; Langston, 1979). The delay time, i.e. the time
difference between P and Ps peak, gives evidence about the depth and the
impedance contrasts of a discontinuity. Furthermore, the RFs can be analysed
13
station positions 30.0 20.0 10.0

deployment OBS array Portugal 40.0
recovery
Azores
range estimate ul t
km Gloria fa
35.0
km Madeira
0 500
38.8
0 12.5 25
D11
125 km
D07 D06
D04 4000
38.4 D12
D10 D01
D03 D02
D08 D05
5000
70 km
D09
38.0
towards Gloria fault

6000
18.8 18.4 18.0
Figure 3.2: Map with OBS positions. The position during deployment is given as
filled triangle, the position during recovery as black open circle and the position
estimated from the range measurements as red open triangle. The colour shading
gives the bathymetry (EMEPC, Task Group for the Extension of the Continental
Shelf). The distance to the Gloria fault (Eurasian-African plate boundary, Bird,
2003) is given by the white line. The location of the OBS array in the eastern mid
Atlantic is shown in the small inset map. Station D05 had two clamped components
and is marked with a filled grey triangle. For station D08, the position estimated
with the range measurements is only based on few measurements towards the North
east of the station (dashed red open triangle).
in terms of S wave velocity depth profiles (e.g. Svenningsen and Jacobsen,

2007). Additionally, we use array methods (e.g. Rost and Thomas, 2002) to
increase the number of events used for the RF analysis. The application of array
methods requires a synchronized network of stations (Rost and Thomas, 2002).
This is often not given for arrays consisting of free-fall pop-up OBSs as used in
this thesis. Furthermore, some skew times estimated during the recovery of our
stations were unusually high or could not be determined due to a shut-down at
one station. In the first publication in this thesis (chapter 4, Hannemann et al.,
2014), we used ambient noise cross-correlation to estimate the clock drifts and
the static time offsets of the OBSs and cross-checked our results with the mea-
sured skew times. The obtained results were used to correct the time at each
station by inserting or deleting individual time samples (e.g. to correct a clock
drift of 1 s a≠1 we have to delete one sample every 3 d 15 h 36 min). With the
synchronized network, we performed a frequency-wavenumber analysis (Rost
and Thomas, 2002) to search for teleseismic event recordings (appendix A).
14
Bibliography
Based on this, we used teleseismic P and Rayleigh wave recordings to estimate

the orientation of the OBSs (appendix 5.S1 in chapter 5).
The calculation of RF at OBSs strongly depends on the signal quality of
the individual teleseismic event recordings (e.g. Audet, 2016). In this thesis,
we used beamforming (Rost and Thomas, 2002) to increase the number and
the quality of the used events and we introduced a quality control employing
different evaluation criteria to choose optimal deconvolution parameters for
both single station recordings and beamformed traces (chapter 5). The final
set of single station RFs and beamformed RF was used to investigate major
discontinuities from the sea-floor down to the mantle transition zone. The
single station RFs were used for the investigation of the crustal thickness and
LAB depths across the study area. FOr this purpose, we also used a common
conversion point stack which results from backprojecting the amplitudes of the
single event recordings along the event’s ray trace (e.g. Knapmeyer-Endrun
et al., 2014) by assuming a global Earth model (PREM, Dziewonski and
Anderson, 1981). The stacked single station RFs and the beamformed RFs
were analysed with a focus on the mantle discontinuities. The analysis of the
different major discontinuities using single station and beamformed RFs is
the content of the second publication in this thesis (chapter 5). In the third
publication in this thesis (chapter 6, Hannemann et al., 2016), we analysed the
RF in terms of S wave velocity-depth models. For this purpose, we obtained
a new relation for the apparent P wave incidence angle (P wave polarization)
which can be linked to the apparent S wave half space velocity and implemented
a progressive low-pass filtering and a quantitative modelling approach to search
for possible S wave velocity-depth models at each OBS.
Bibliography
Agee, C. B. (1998). Phase transformations and seismic structure in the upper
mantle and transition zone. In Hemley, R. J. and Ribbe, P. H., editors,
Ultrahigh-Pressure Mineralogy. Physics and Chemistry of the Earth’s Deep
Interior, volume 37, chapter 5, pages 165–203. Mineralogical Society of
America, Washington, D.C.
Audet, P. (2016). Receiver functions using OBS data: promises and limita-
tions from numerical modelling and examples from the Cascadia Initiative.
Geophysical Journal International, 205(3):1740–1755.
Barruol, G. and Sigloch, K. (2013). Investigating La Réunion hot spot from crust
to core. Eos, Transactions American Geophysical Union, 94(23):205–207.
Bell, S. W., Ruan, Y., and Forsyth, D. W. (2015). Shear Velocity Structure
of Abyssal Plain Sediments in Cascadia. Seismological Research Letters,
86(5):1247–1252.
Bird, P. (2003). An updated digital model of plate boundaries. Geochemistry,

Geophysics, Geosystems, 4(3):1027.
Bormann, P. and Wielandt, E. (2013). Seismic Signals and Noise. In Bormann,

P., editor, New Manual of Seismological Observatory Practice 2, volume 2,
chapter 4, pages 1–64. GeoForschungsZentrum Potsdam GFZ, Potsdam.
15
Bibliography
Bürgmann, R. and Chadwell, D. (2014). Seafloor Geodesy. Annual Review of

Earth and Planetary Sciences, 42(1):509–534.
Castagna, J. P., Batzle, M. L., and Eastwood, R. L. (1985). Relationships

between compressional-wave and shear-wave velocities in clastic silicate rocks.
Geophysics, 50(4):571–581.
Chevrot, S., Vinnik, L., and Montagner, J. P. (1999). Global-scale analysis of the
mantle Pds phases. Journal of Geophysical Research, 104(B9):20203–20219.
Crawford, W. C. and Webb, S. C. (2000). Identifying and Removing Tilt Noise

from Low-Frequency (<0.1 Hz) Seafloor Vertical Seismic Data. Bulletin of
the Seismological Society of America, 90(4):952–963.
Crawford, W. C., Webb, S. C., and Hildebrand, J. A. (1998). Estimating

shear velocities in the oceanic crust from compliance measurements by two-
dimensional finite difference modeling fractured. Journal of Geophysical
Research, 103(B5):9895–9916.
Deuss, A., Andrews, J., and Day, E. (2013). Seismic Observations of Mantle
Discontinuities and Their Mineralogical and Dynamical Interpretation. In
Karato, S.-i., editor, Physics and Chemistry of the Deep Earth, pages 297–323.
John Wiley & Sons, Ltd.
Deuss, A., Redfern, S. A. T., Chambers, K., and Woodhouse, J. H. (2006). The
nature of the 660-kilometer discontinuity in Earth’s mantle from global seismic
observations of PP precursors. Science (New York, N.Y.), 311(5758):198–201.
Deuss, A. and Woodhouse, J. H. (2002). A systematic search for mantle

discontinuities using SS-precursors. Geophysical Research Letters, 29(8):90–
94.
Dziewonski, A. M. and Anderson, D. L. (1981). Preliminary reference Earth

model. Physics of the Earth and Planetary Interiors, 25(4):297–356.
Estabrook, C. H. and Kind, R. (1996). The Nature of the 660-Kilometer Upper-

Mantle Seismic Discontinuity from Precursors to the PP Phase. Science,
274(5290):1179–1182.
Fischer, K. M., Ford, H. A., Abt, D. L., and Rychert, C. A. (2010). The
Lithosphere-Asthenosphere Boundary. Annual Review of Earth and Planetary
Sciences, 38(1):551–575.
Flanagan, M. P. and Shearer, P. M. (1998). Global mapping of topography on

transition zone velocity discontinuities by stacking SS precursors. Journal of
Geophysical Research, 103(B82):2673–2692.
Gao, H. and Schwartz, S. (2015). Preface to the Focus Section on Cascadia

Initiative Preliminary Results. Seismological Research Letters, 86(5):1235–
1237.
Gilbert, H. J., Sheehan, A. F., Wiens, D. A., Dueker, K. G., Dorman, L. M.,
Hildebrand, J., and Webb, S. (2001). Upper mantle discontinuity structure
in the region of the Tonga Subduction Zone. Geophysical Research Letters,
28(9):1855–1858.
16
Bibliography
Gossler, J. and Kind, R. (1996). Seismic evidence for very deep roots of
continents. Earth and Planetary Science Letters, 138(1-4):1–13.
Grigoli, F., Cesca, S., Dahm, T., and Krieger, L. (2012). A complex linear
least-squares method to derive relative and absolute orientations of seismic
sensors. Geophysical Journal International, 188(3):1243–1254.
Gu, Y. J. and Dziewonski, A. M. (2002). Global variability of transition zone

thickness. Journal of Geophysical Research, 107(B7):2135.
Gu, Y. J., Dziewonski, A. M., and Agee, C. B. (1998). Global de-correlation

of transition zone discontinuities. Earth and Planetary Science Letters,
157(1-2):57–67.
Hannemann, K., Krüger, F., and Dahm, T. (2014). Measuring of clock drift
rates and static time offsets of ocean bottom stations by means of ambient
noise. Geophysical Journal International, 196(2):1034–1042.
Hannemann, K., Krüger, F., and Dahm, T. (2016). Oceanic lithospheric S

wave velocities from the analysis of P wave polarization at the ocean floor.
Geophysical Journal International.
Hasselmann, K. (1963). A statistical analysis of the generation of microseisms.

Reviews of Geophysics, 1(2):177.
Helffrich, G. (2000). Topography of the transition zone seismic discontinuities.

Reviews of Geophysics, 38(1):141–158.
Hirose, K. (2002). Phase transitions in pyrolitic mantle around 670-km depth:

Implications for upwelling of plumes from the lower mantle. Journal of
Geophysical Research, 107(B4):2078.
Karato, S.-i. (1992). On The Lehmann Discontinuity. Geophysical Research

Letters, 19(22):2255–2258.
Karato, S.-i. (2012). On the origin of the asthenosphere. Earth and Planetary
Science Letters, 321-322:95–103.
Karato, S.-i. and Jung, H. (1998). Water, partial melting and the origin of the
seismic low velocity and high attenuation zone in the upper mantle. Earth
and Planetary Science Letters, 157(3-4):193–207.
Kawakatsu, H., Kumar, P., Takei, Y., Shinohara, M., Kanazawa, T., Araki,
E., and Suyehiro, K. (2009). Seismic Evidence for Sharp Lithosphere-
Asthenosphere Boundaries of Oceanic Plates. Science, 324(5926):499–502.
Kennett, B. L. N., Engdahl, E. R., and Buland, R. (1995). Constraints

on seismic velocities in the Earth from traveltimes. Geophysical Journal
International, 122(1):108–124.
Knapmeyer-Endrun, B., Kruger, F., Group, T. P. W., Krüger, F., and Group,
P. W. (2014). Moho depth across the Trans-European Suture Zone from
P- and S-receiver functions. Geophysical Journal International, 197(2):1048–
1075.
17
Bibliography
Langston, C. A. C. A. (1979). Structure under Mount Rainier, Washington,

inferred from teleseismic body waves. Journal of Geophysical Research,
84(B9):4749–4762.
Laske, G., Masters, G., Ma, Z., and Pasyanos, M. E. (2013). CRUST1.0 :
An Updated Global Model of Earth ’ s Crust. In Geophys. Res. Abstracts,
volume 15, pages Abstract EGU2013—-2658.
Lawrence, J. F. and Shearer, P. M. (2006). A global study of transition

zone thickness using receiver functions. Journal of Geophysical Research,
111(6):B06307.
Lehmann, I. (1961). S and the Structure of the Upper Mantle. Geophysical

Journal of the Royal Astronomical Society, 4(Supplement 1):124–138.
Longuet-Higgins, M. S. (1950). A Theory of the Origin of Microseisms. Philo-

sophical Transactions of the Royal Society of London. Series A, Mathematical
and Physical Sciences, 243(857):1–35.
Olugboji, T. M., Karato, S., and Park, J. (2013). Structures of the oceanic
lithosphere-asthenosphere boundary: Mineral-physics modeling and seismo-
logical signatures. Geochemistry, Geophysics, Geosystems, 14(4):880–901.
Rost, S. and Thomas, C. (2002). Array seismology: Methods and applications.

Saki, M., Thomas, C., Nippress, S. E. J., and Lessing, S. (2015). Topography
of upper mantle seismic discontinuities beneath the North Atlantic: The
Azores, Canary and Cape Verde plumes. Earth and Planetary Science Letters,
409:193–202.
Shen, Y., Sheehan, A. F., Dueker, K. G., de Groot-Hedlin, C., Gilbert, H.,
de Groot–Hedlin, C., and Gilbert, H. (1998a). Mantle Discontinuity Structure
Beneath the Southern East Pacific Rise from P-to-S Converted Phases.
Science, 280(5367):1232–1235.
Shen, Y., Solomon, S. C., Bjarnason, I. T., and Wolfe, C. J. (1998b). Seismic
evidence for a lower-mantle origin of the Iceland plume. Nature, 395(6697):62–
65.
Silveira, G., Vinnik, L., Stutzmann, E., Farra, V., Kiselev, S., and Morais, I.
(2010). Stratification of the Earth beneath the Azores from P and S receiver
functions. Earth and Planetary Science Letters, 299(1-2):91–103.
Simmons, N. A. and Gurrola, H. (2000). Multiple seismic discontinuities near

the base of the transition zone in the Earth’s mantle. Nature, 405(6786):559–
562.
Stachnik, J. C., Sheehan, A. F., Zietlow, D. W., Yang, Z., Collins, J., and
Ferris, A. (2012). Determination of New Zealand Ocean Bottom Seismometer
Orientation via Rayleigh-Wave Polarization. Seismological Research Letters,
83(4):704–713.
18
Bibliography
Stähler, S. C., Sigloch, K., Hosseini, K., Crawford, W. C., Barruol, G., Schmidt-
Aursch, M. C., Tsekhmistrenko, M., Scholz, J. R., Mazzullo, A., and Deen, M.
(2016). Performance report of the RHUM-RUM ocean bottom seismometer
network around la Réunion, western Indian Ocean. Advances in Geosciences,
41:43–63.
Suetsugu, D., Inoue, T., Obayashi, M., Yamada, A., Shiobara, H., Sugioka,
H., Ito, A., Kanazawa, T., Kawakatsu, H., Shito, A., Fukao, Y., Suetsugu,
D., Bina, C., Inoue, T., Wiens, D., and Jellinek, M. (2010). Depths of the
410-km and 660-km discontinuities in and around the stagnant slab beneath
the Philippine Sea: Is water stored in the stagnant slab? Physics of the
Earth and Planetary Interiors, 183(1-2):270–279.
Suetsugu, D., Shinohara, M., Araki, E., Kanazawa, T., Suyehiro, K., Yamada,
T., Nakahigashi, K., Shiobara, H., Sugioka, H., Kawai, K., and Fukao,
Y. (2005). Mantle discontinuity depths beneath the west Philippine basin
from receiver function analysis of deep-sea borehole and seafloor broadband
waveforms. Bulletin of the Seismological Society of America, 95(5):1947–1956.
Suetsugu, D. and Shiobara, H. (2014). Broadband Ocean-Bottom Seismology.

Annual Review of Earth and Planetary Sciences, 42(1):27–43.
Suetsugu, D., Shiobara, H., Sugioka, H., Fukao, Y., and Kanazawa, T. (2007).
Topography of the mantle discontinuities beneath the South Pacific superswell
as inferred from broadband waveforms on seafloor. Physics of the Earth and
Planetary Interiors, 160(3-5):310–318.
Svenningsen, L. and Jacobsen, B. H. (2007). Absolute S-velocity estimation

from receiver functions. Geophysical Journal International, 170(3):1089–1094.
Tauzin, B., Debayle, E., and Wittlinger, G. (2008). The mantle transition zone
as seen by global Pds phases: No clear evidence for a thin transition zone
beneath hotspots. Journal of Geophysical Research, 113(B8):B08309.
Thomas, C. and Laske, G. (2014). D” observations in the Pacific from PLUME

ocean bottom seismometer recordings. Geophysical Journal International,
200(2):851–862.
Trnkoczy, A., Bormann, P., Hanka, W., Holcomb, L. G., Nigbor, R. L., Shino-
hara, M., Shiobara, H., and Suyehiro, K. (2012). Site Selection, Preparation
and Installation of Seismic Stations. In Bormann, P., editor, New Manual
of Seismological Observatory Practice 2, volume 2, chapter 7, pages 1–139.
GeoForschungsZentrum Potsdam GFZ, Potsdam.
Vacher, P., Mocquet, A., and Sotin, C. (1998). Computation of seismic profiles
from mineral physics: the importance of the non-olivine components for
explaining the 660 km depth discontinuity. Physics of the Earth and Planetary
Interiors, 106(3-4):275–298.
Vinnik, L. P. (1977). Detection of waves converted from P to SV in the mantle.

Physics of the Earth and Planetary Interiors, 15(1):39–45.
Webb, S. C. (1998). Broadband Seismology and Noise under the Ocean. Reviews
of Geophysics, 36(1):105–142.
19
Bibliography
Weidner, D. J. and Wang, Y. (1998). Chemical- and Clapeyron-induced

buoyancy at the 660 km discontinuity. Journal of Geophysical Research,
103(B4):7431.
White, R. S. (1984). Atlantic oceanic crust: seismic structure of a slow-spreading

ridge. Geological Society, London, Special Publications, 13(1):101–111.
White, R. S., McKenzie, D., and O’Nions, R. K. (1992). Oceanic crustal

thickness from seismic measurements and rare earth element inversions.
Journal of Geophysical Research, 97(B13):19683–19715.
Zha, Y., Webb, S. C., and Menke, W. (2013). Determining the orientations
of ocean bottom seismometers using ambient noise correlation. Geophysical
Research Letters, 40(14):3585–3590.
20
4. Measuring of clock drift rates
and static time offsets of ocean
bottom stations by means of
ambient noise
K. Hannemann, F. Krüger, T.Dahm

published in Geophysical Journal International
(doi: 10.1093/gji/ggt434)
Summary
Marine seismology usually relies on temporary deployments of stand alone
seismic ocean bottom stations (OBS), which are initialized and synchronized on
ship before deployment and re-synchronized and stopped on ship after recovery
several months later. In between, the recorder clocks may drift and float at
unknown rates. If the clock drifts are large or not linear and cannot be corrected
for, seismological applications will be limited to methods not requiring precise
common timing. Therefore, for example array seismological methods, which
need very accurate timing between individual stations, would not be applicable
for such deployments.
We use an OBS test-array of 12 stations and 75 km aperture, deployed
for 10 months in the deep sea (4.5 - 5.5 km) of the mid-eastern Atlantic.
The experiment was designed to analyze the potential of broad-band array
seismology at the sea floor. After recovery, we identified some stations which
either show unusual large clock drifts and/ or static time offsets by having a
large difference between the internal clock and the GPS-signal (skew).
We test the approach of ambient noise cross-correlation to synchronize
clocks of a deep water OBS array with km-scale inter-station distances. We
show that small drift rates and static time offsets can be resolved on vertical
components with a standard technique. Larger clock drifts (several seconds
per day) can only be accurately recovered if time windows of one input trace
are shifted according to the expected drift between a station pair before the
cross-correlation. We validate that the drifts extracted from the seismometer
data are linear to first order. The same is valid for most of the hydrophones.
Moreover, we were able to determine the clock drift at a station were no skew
could be measured. Furthermore, we find that instable apparent drift rates at
some hydrophones, which are uncorrelated to the seismometer drift recorded
at the same digitizer, indicate a malfunction of the hydrophone.
21
Chapter 4. Measuring clock drift and time offset at OBS
4.1 Introduction
Within the DOCTAR project (Deep OCean Test ARray), we examine the
potential of broad-band array methods on the ocean floor. The application
of array methods requires a synchronized network (Rost and Thomas, 2002).
On shore, this is usually achieved by continuous clock synchronization with a
GPS signal. In the deep sea, GPS signals cannot be recorded. Therefore, clock
synchronization can only be achieved before deployment and after recovery
of the stations (e.g. Geissler et al., 2010). The measured time difference
between the data logger and GPS after recovery is referred to as skew. Several
studies deal with clock drift measurements and synchronization of seismic
on-shore networks using ambient noise cross-correlation (Stehly et al., 2007;
Sens-Schönfelder, 2008). The advantage of such methods, beside the low cost
aspect, is that they can be applied off-line a posteriori to correct data already
recorded.
Up to our knowledge, there has only been one successful application of
synchronization using ambient noise cross-correlation at an ocean bottom
station (OBS) installation which was at shallow water depth (several tens of
meters) and several meter inter-station distances (Sabra et al., 2005).
In this study, we demonstrate the feasibility of ambient noise clock syn-
chronization for deep water depths and km-scale installations. Such network
geometries require long correlation times (at least 1 d) to achieve good signal-
to-noise-ratios (SNRs) and stable correlation results. We further extend the
standard method to identify static time offsets and large clock drifts which are
present at some stations of our OBS deployment. The method demonstrated
here is useful to estimate the parameters for the correction of time drifts and
offsets, which is a first step to greatly improve the relevance of OBS network
data.
4.2 Data
In 2011, twelve free-fall OBS (DEPAS pool, http://www.awi.de/en/go/depas)
were deployed in the 4.5 km to 5.5 km-deep water of the mid-eastern Atlantic,
north of the Gloria Fault, 800 km off the coast of Portugal. Each station
consists of a broad-band seismometer (Guralp CMG-40T, 60 s-50 Hz) and a
hydrophone (HighTechInc HTI-04-PCA/ULF, 100 s-8 kHz, flat instrument
response down to 5 s, at D08 down to 2 s). The sensors share the same data
logger (Send Geolon MCS, 24 bit, 1 - 1000 Hz, 20 GB; see Dahm et al., 2002,
for detailed instrument description). The stations formed an array with an
aperture of approximately 75 km (see Fig. 4.1).
The stations recorded from 2011 July until 2012 April with a sampling
rate of 100 Hz. During recovery, we identified three stations with problems
concerning the timing. For station D05, no skew measurement was possible.
The recording at this station stopped 1 month before the recovery, possibly
due to a higher energy consumption because of a clamped vertical and one
clamped horizontal component, meaning that these components were at their
maximum values (see Fig. 4.2a, D05 Z). Furthermore, station D08 and D09
show unusually high skews of around 20 min and 10 s (Fig. 4.1 and Table 4.1).
All other stations had ’normal’ skews smaller than 1.5 s (Table 4.1).
22
4.2. Data
Figure 4.1: Configuration of OBS deployment
Table 4.1: Skew measurements tskew after recovery and nominal clock drift d per
year assuming a linear drift.
station tskew [s] d [s a≠1 ]

D01 0.260812 0.319
D02 1.337781 1.638
D03 0.415937 0.510
D04 0.504531 0.615
D05 - -
D06 -0.038125 -0.047
D07 0.377656 0.462
D08 1201.108843 1472.38
D09 -8.36275 -10.318
D10 0.520906 0.640
D11 0.928687 1.138
D12 0.0095 0.012
We checked the relative timing between stations with strong teleseismic

events (Fig. 4.2) by performing a beam for the first arrival (compare captions
in Fig. 4.2 for details and Rost and Thomas, 2002, for method). We choose one
event shortly after the deployment (2011 July 6, Fig. 4.2a) and another slightly
before the recovery (2012, April 11, Fig. 4.2b). Thus, we could verify that D08
shows indeed an extremely high clock drift as indicated by the measured skew
of 20 min. Furthermore, we note that the 10 s skew at D09 seems to be related
to a time offset which is constant over the whole period of the experiment,
since the time offset was already present shortly after the beginning of the
experiment (Fig. 4.2a and time difference shown by red line in Fig. 4.2b). We
checked the log-files of the recorders and find that the first synchronization
(during deployment) was not logged in the file of station D09 as it was in case
of all other stations. This seems to indicate that the first synchronization has
failed at station D09. Besides this, the station worked normally.
23
(a)
(b)
Figure 4.2: Examples for teleseismic events, unfiltered data. 0 s corresponds to origin
time in each case, station D05 has only data for the first event because of the power
loss 1 month before the recovery, red line in (b) shows time of maximum of P-wave for
all stations except for D08 and D09, Mw is the moment magnitude, the distance in
degree and is the azimuth of the event seen from station D01 ( and estimated
from location of USGS (http://earthquake.usgs.gov/earthquakes /eqarchives/epic/,
last accessed 30 October 2013), the slowness values were taken from the AK135 Travel
Time Tables (http://rses.anu.edu.au/seismology/ak135/intro.html, last accessed 30
October 2013): (a) 06.07.2011 19:03:18.26, Mw = 7.6, depth = 17 km, = 159.72¶ ,
= 289.23¶ ; phase PKPdf (PKIKP) with slowness 1.18 s degrees≠1 ; (b) 11.04.2012
08:38:37.36, Mw = 8.6, depth = 22 km, = 105.22¶ , = 74.52¶ ; phase P with
slowness 4.45 s degrees ≠1
4.3 Theory
In this section, we consider some simple examples to make clear which influence
a linear clock drift has on correlation results. These examples deal with
separated, undisturbed signals, but the effects are similar for ambient noise
cross-correlations where several signals from different directions and sources
are superimposed (Stehly et al., 2007; Sens-Schönfelder, 2008).
24
4.3. Theory
The cross-correlation in a time window of length Tw = tE ≠ tB , where tB is

the start and tE is the end of the time window, is defined as follows:
⁄tE
C12 (· ) = f1 (t + · )f2 (t)dt. (4.1)
tB
Herein, f1 is the signal measured at sensor 1 and f2 the signal measured at

sensor 2, · is the lag-time.
We assume that the measured time tÕ at sensor 2 is related to the measured
time t at sensor 1 as follows:
tÕ = t (1 + ”d) + tc , (4.2)
where ”d is the dimensionless relative clock drift and tc is a static time
offset between the sensors.
To illustrate the influence of a clock drift on a correlation, we assume a
homogeneous half-space with velocity c and a delta impulse excited at time t”
and location ˛r” , which is measured by sensor n at location r˛n :
A B
r˛n ≠ r˛”
fn (t) = ” t ≠ t” ≠ (4.3)
c
If we now replace t = t(tÕ ) in equation (4.3) for sensor 2 and insert f2 (tÕ )
and equation (4.3) for sensor 1 in equation (4.1) and assume t = tÕ as we do by
correlating skewed signals, we get a cross-correlation which is non-zero for
A B
r˛2 ≠ r˛1 r˛2 ≠ r˛”
· =≠
Õ
≠ tc ≠ t” + ”d. (4.4)
c c
Therefore, the correlation result is a delta impulse located at the lag-time
which corresponds to the negative travel time between sensor 1 and 2, which is
equivalent to the lag-time of the correlation of unskewed signals, delayed by
the time which corresponds to the influence of the static time offset tc and the
relative clock drift ”d between the sensors on the arrival time of the signal at
sensor 2.
A synthetic example should demonstrate, how a clock drift influences the
correlation results in consecutive time windows and how it can be measured.
We consider a more complex signal at sensors 1 and 2 which consists of several
short wavelets which occur every 20 s starting at 5 s at sensor 1 and at 10 s at
sensor 2 (Fig. 4.3, black solid lines trace 1 and 2). Furthermore, we assume
a skewed signal with a clock drift of 1 s per 100 s at sensor 2 (Fig. 4.3, red
dashed line trace 2). We introduce common, unshifted time windows to cut
out traces before applying the cross-correlation (blue boxes in Fig. 4.3). The
resulting cross-correlation function for each time window are also presented in
Fig. 4.3 (black solid line = unaffected by a clock drift and red dashed line =
influenced by the effect of the clock drift).
We observe that the influence of the clock drift increases for later time
windows. From equation (4.4), we know that the time difference between the
correlation without clock drift and the one with clock drift corresponds to the
influence of the static time offset and the clock drift on the arrival times of
the signal at sensor 2. This effect is linear in ”d and visible in the synthetic
25
Figure 4.3: Synthetic example to visualize the effect of the clock drift, black solid
lines are unskewed data and their correlation results, red dashed line is for skewed
data, blue box shows used time window, green solid line gives correlation result from
shifted time windows (grey shading)
example, too. Therefore, we can estimate the clock drift by comparing the
correlation of the first time window with the correlation results of all consecutive
time windows. On the other hand, this example illustrates that the stack of
correlation traces of consecutive time windows affected by a clock drift would
lead to a correlation trace with a more or less smeared signal depending on the
amount of the clock drift (see stacks of correlation results in Fig. 4.3, compare
black solid (without skew) and red dotted line (with skew)). We would expect
that we will not get any usable correlation signal for large clock drifts (several
seconds per day), because either they lead to time differences which could be
larger than the chosen correlation time window or even if the window was
properly set, the clock drifts would cause a strong smearing.
To overcome this problem, we shift each correlation time window of the
second input trace according to an assumed relative clock drift ”d between the
sensors. The length Tw of the correlation time window will remain unaffected
by the drift correction. The amount of time needed to shift the time window of
the second signal f2 corresponds to the effect of the assumed relative clock drift
”d on the start time tB of the time window which can be calculated analogue
to equation (4.2) assuming tc2 = 0. Therefore,
f2 (t) = f2 (tÕ ≠ tB · ”d) (4.5)
is inserted in equation (4.1).

The gray shading in Fig. 4.3 shows the shifted time windows in case of
the synthetic example. The green solid lines represent the resulting cross-
correlation functions for the shifted time windows. We observe that they are
26
4.4. Method
only influenced by the clock drift within the short time windows and that their
difference in lag-time does not increase for later time windows. By comparing
the stack of the unskewed and the shifted time window correlation results (see
lower panel in Fig. 4.3, green and black solid lines), we find that the signal is
nearly identical except for a small difference in amplitude and lag-time, which
represents the influence of the clock drift in the short time window.
If we deal with a data set where all signals are skewed, we only can
estimate relative clock drifts between the sensor pairs. We need additional
information such as skew measurements or stations with continuous GPS
synchronization to get an absolute timing. Therefore, the stations of a network,
where no additional information concerning the timing is available, can only be
synchronized relatively to one reference station.
4.4 Method
4.4.1 Estimation of clock drifts
In case of OBS data, we usually have no idea how the correlation result without
any clock drift would look like. Therefore, we take the correlation result of
the first time window as our reference trace. The reference trace is correlated
with the correlation results of all time windows of the station pair. The lag
times of the maxima of the resulting traces are determined (hereafter referred
to as tmi ). This gives the influence of the relative clock drift of the station pair
on the correlation result (comparable to the last term in eq. (4.4)). However,
we have to be aware of the influence of the drift on the first correlation time
window. By plotting the resulting clock differences tmi as a function of the time
Ti passed since the start of the first correlation time window, we can verify the
linear drift assumption. In case of a positive result, we are able to calculate the
relative clock drift per second ”d by a linear regression. Moreover, we estimate
an error ‡tm which reflects how well the clock differences tmi over time Ti are
estimated by the trend line resulting from the linear regression.
ı̂
ı 1 ÿN
‡tm = Ù (tmi ≠ (”d · Ti + tdif f0 ))2 (4.6)
N i=0
Herein, tdif f0 is the clock difference at T0 = 0. This parameter is also

estimated by the linear regression and should be close to zero.
If we compare the relative clock difference after 1 yr calculated by the
results of the regression with the expected values estimated from the measured
skews, we can determine all parameters needed for the timing correction of the
data set.
4.4.2 Constant time shifts

Additionally, we know from the beam forming (Fig. 4.2) that station D09
probably has a static time offset tc . To determine its amount, we use the
following relations.
27
tm = Tcorr · ”d , tth = Tcorr · ”dapp (4.7)

tc + ”d · T
with ”dapp = (4.8)
T
tth ≠ tm
∆ tc = ·T (4.9)
Tcorr
Herein, tm is the estimated clock difference from the correlations which

reflects the influence of the actual clock drift ”d over the length Tcorr of the used
time window. The time tth is the expected time difference for the used time
window calculated with the apparent clock drift ”dapp which results from the
measured skews and the time T between synchronization and skew measurement
by neglecting the effect of the static time offset tc .
4.4.3 Shifted correlation time window approach

For station D08, we know from the measured skew (compare Table 4.1) that
the overall clock drift is around 20 min for 10 months of recording, which is
coarsely validated by the beam forming (Fig. 4.2). We use shifted correlation
time windows (see eq. (4.5)) to estimate the clock drift at station D08 for the
vertical seismic component. Hereby, the drift rates of all other stations remain
unchanged and are assumed to be constant.
If the tested skew, which is used to calculate the expected clock difference
by assuming a linear clock drift, is correctly chosen, the lag-time of the result-
ing cross-correlation will only be influenced by the clock drift in the chosen
correlation time window. The method of the shifted correlation time windows
offers the opportunity to systematically test for several clock drifts. This might
be useful in case of no, lost, or erroneous skew measurements. After performing
several tests with different skews and therefore clock drifts, all correlations from
the whole available recording period are stacked for each assumed clock drift
and the maximum amplitudes of the resulting traces are estimated as a quality
measure. The tested value which belongs to the highest maximum amplitude is
assumed to be the most likely (compare stacks of correlation results in synthetic
example Fig. 4.3).
4.4.4 Processing of data set

We decide to cross-correlate the vertical seismic components of our whole data
set day by day. The processing is similar for the unshifted and the shifted time
windows except for some details. The day traces are split into 90 s windows
which overlap by 50 per cent on which we perform the cross-correlation. The
window length was chosen because of the largest interstation distance of 75 km
and an assumed surface wave velocity of 2 km/s. The latter results from the
assumption of a very thin sediment layer which is based on the observation of
the high-frequency content (up 30-40 Hz) of local events and the steep incidence
angles of teleseismic events. In contrast to the unshifted time windows, we have
to check whether the shifted correlation time window for the second input trace
is still within the time range of the chosen trace time window which is in our case
1 d. The resulting correlation traces of the different time windows are stacked.
28
4.5. Discussion
The stacked correlation result with unshifted time windows corresponds to the
correlation result for the whole trace (Bensen et al., 2007), although they are
now probably influenced by the relative clock drift between the stations. We
estimate the correlation traces from -40 s to 40 s lag-time which should be
sufficient for Rayleigh wave propagation between the stations of the array.
Furthermore, we also correlate the hydrophone signals. There, we choose
180 s time windows with 50 per cent overlap and estimate the correlations
from -60 s to 60 s lag-time. The larger time window is necessary because of
the acoustic velocities which are slower than the seismic ones. We perform
the correlations of the hydrophone signals to test whether they show the same
behaviour as the seismic components.
For the pre-processing of the data, we mainly follow the work by Bensen et al.
(2007) and Picozzi et al. (2009). We use the one-bit and spectral normalization
and remove the global and the local offset. Furthermore, we stack the correlation
traces of 20 consecutive days (here referred to as 20 d stack) every 5 d to increase
the SNR of the correlation trace. We apply a low-pass filter with a corner
frequency of 1 Hz before estimating clock differences from the 1 d and the 20 d
stack. For better comparison, we determine the amount of clock difference after
1 year.
4.5 Discussion
4.5.1 Correlation of vertical seismic components
In Fig. 4.4, the correlation result (20 d stack) for the vertical seismic components
of the station pair D01 and D02 is presented as an example. The correlation
signal is asymmetric which is related to a non-isotropic distribution of noise
sources (Stehly et al., 2006). The different amplitudes over time might be
related to changes in the strength and distribution of the noise sources. We
observe a linear shift of the correlation signal with time (Fig. 4.4) which reflects
the influence of a relative clock drift between the stations. For the estimation
Figure 4.4: Amplitude of stacked correlation traces (20 d stack) over lag-time and
starting day of the vertical seismic components of stations D01 and D02, the amplitude
is normalized to the maximum of all correlation traces. The black line is given as a
reference.
29
of the time-shifts, we used a cross-correlation of the first correlation trace and

all consecutive traces without any normalization, any tapering or any offset
removal. The used time windows were 5 s long and had an overlap of 50 per
cent. We also tested longer time windows of 15 s and 40 s which gave similar
results. By plotting the estimated clock differences against the starting day of
the time windows used for the stack (Fig. 4.5), we can validate the linear drift
as expected by eq. (4.4). Moreover, we are able to use a linear regression to
estimate the clock drift per second and calculate from this the clock difference
which would be present after 1 yr of recording.
We observe in Fig. 4.5 that the results from the 20 d stacks (filled circle)
have smaller deviations compared to the trend line than the ones estimated
from the 1 d correlations (open circles). This is related to the better SNR
of the 20 d stacks. Therefore, we will use the results from the 20 d stack for
the following discussion. We made a comparison of the results for the vertical
seismic components for all station pairs with the expected clock differences
(Fig. 4.6). The errors were estimated by using eq. (4.6).
First of all, we find that our estimates are in good agreement with the
expected values for the majority of the stations (Fig. 4.6a, black symbols)
and have small errors. The standard deviation of the difference between the
estimated annual clock differences and the expected ones is ±0.087 s and was
estimated by using the following equation:
ı̂
ı 1 ÿN
‡m = Ù (tmi ≠ tthi )2 (4.10)
N i=0
Herein, tmi is the estimated annual clock difference from the clock drift per
second which was estimated by the correlations of the i-th station pair. The
value tthi is the expected annual clock difference from the skew measurements
of the i-th station pair.
We obtain two groups in Fig. 4.6(b) and (c) which have large differences
between the estimated and the expected annual clock differences. By identifying
the related station pairs, we find that the group with expected values around
Figure 4.5: Comparison of the clock differences estimated from 1 d (open circles)
and 20 d stacks (filled circles) of the vertical seismic components for station pair
D01 and D02. The blue and the green line show the trend lines resulting from linear
regression.
30
4.5. Discussion
1500 s and large errors always includes station pairs involving D08 (Table 4.1
and Fig. 4.6b, green symbols). This result and the corresponding correlations
without any usable signal (not presented here) show that the estimation of
large clock drifts is not possible with ambient noise cross-correlation by using
unshifted correlation time windows.
On the contrary, the estimated annual clock differences for the group with
the expected values around 10 s and small errors (blue symbols in Fig. 4.6c)
indicate a small linear drift for the vertical seismic component. We find that all
station pairs within this group include the station D09 (blue symbols in Fig. 4.6).
Therefore, we confirm our observation from the beam forming (Fig. 4.2) that
this station has a static time offset. Moreover, we can estimate this time
offset from our observation and in addition correct the skew for the timing
correction of the data (Table 4.2). To determine the static time offset, we used
eq. (4.9) for the values of all station pairs involving station D09 and calculated
afterwards the mean of the resulting time offsets. This leads to a static time
offset of -9.213 ± 0.009 s (compare Table 4.2) and a corrected annual drift rate
of 1.049 s a≠1 for station D09.
Additionally, we observe in Fig. 4.6 that the estimated values for station
pairs involving D05 (red symbols) show random deviations from the ’ideal’ line
and large errors. This effect was expected because of the clamped vertical
seismic component as has been mentioned before. Consequently, we are not
able to obtain any information for the timing correction from the correlations
of vertical seismic components and exclude the vertical seismic component of
D05 from further investigations concerning the vertical seismic components.
4.5.2 Shifted correlation time window approach

For the test of the shifted correlation time window approach, we present the
resulting maximum amplitudes of the stacks for the whole recording period
Figure 4.6: Comparison of the estimated annual clock differences of 20 d stacks of

the correlations of the vertical seismic components with expected values from skew
measurements. The error bars give the value of ‡tm as defined in eq. (4.6). Station
pairs involving D05 are plotted red, D08 green and D09 blue, the lines indicate where
estimated and expected values are equal.
31
Table 4.2: Annual clock differences for station D09 as reference station, tm is the
estimated annual clock difference from the 20 d stacks for the vertical seismic
component (the error is estimated by using eq. (4.6)) and tth is the expected annual
clock difference calculated from the measured skews, tc is the constant time shift
calculated according to eq. (4.9)
OBS tm [s] tth [s] tc [s]

D01 0.752 ± 0.032 -10.638 -9.231 ± 0.026
D02 -0.542 ± 0.018 -11.957 -9.252 ± 0.015
D03 0.550 ± 0.024 -10.828 -9.222 ± 0.020
D04 0.483 ± 0.027 -10.934 -9.253 ± 0.022
D06 1.168 ± 0.040 -10.272 -9.271 ± 0.033
D07 0.502 ± 0.039 -10.781 -9.144 ± 0.031
D10 0.216 ± 0.036 -10.959 -9.057 ± 0.029
D11 -0.020 ± 0.028 -11.457 -9.269 ± 0.023
D12 1.041 ± 0.051 -10.330 -9.216 ± 0.042
mean tc [s] -9.213 ± 0.009
measured skew (see Table 4.1) [s] -8.36275
new skew corrected with tc [s] 0.850(25)
new clock drift per year d [s a≠1 ] 1.049
Figure 4.7: Maximum amplitude of the stack of all correlation traces (i.e. for
the whole recording period) for different annual clock drifts tested and stations in
combination with station D08, the distance to station D08 is given in braces. The
tested annual clock drifts are equivalent to the value in the third column of Table 4.1,
the grey line indicates the annual clock drift for D08 as can be calculated by the
measured skew.
over the tested annual clock drifts for station D08 in Fig. 4.7. The highest
value for the maximum amplitude of the stacked correlations is reached for the
annual clock drifts of 1472.24 s a≠1 and 1472.86 s a≠1 . The value estimated by
the measured skew lies at 1472.38 s a≠1 (compare Table 4.1 and grey line in
Fig. 4.7) which is just in-between the two most likely test values. Therefore,
the shifted correlation time window approach and the selected quality criteria
are feasible for the testing of large clock drifts. Furthermore, we observe that
the value of the maximum amplitude is partly related to the distance between
the stations, although it is not direct proportional. Additionally, we took the
32
4.5. Discussion
corrected annual clock drift for station D09 for the calculation of the new
start of the correlation time window on the second input trace which gives the
highest maximum amplitude for an annual clock drift of 1472.24 s a≠1 . This
observation confirms that the correction for the static time offset at station
D09 was successful.
4.5.3 Correlation of hydrophone signals

Hydrophone and seismometer share the same data logger, therefore we expect
to observe a similar drift behaviour and similar drift rates from the correlations
of the hydrophone signals as from the correlations of the vertical seismic
components. Moreover, we hope to extract an information about the drift rate
of station D05 where the vertical seismic component was clamped, but the
hydrophone was working.
Careful data inspection showed that the hydrophones show a malfunction
at two stations (D02 and D12) at the beginning of the recording and that
the hydrophone at D03 shows data errors for most of the recording period.
The stations D02 and D03 show quite small amplitudes and all three stations
have steps in the amplitude which sometimes slowly recover during periods of
malfunction.
If we have a look at the correlation results of the remaining stations, we
find that they show a good agreement in the clock drift behavior between
seismometer and hydrophone. The standard deviation of the difference between
the estimated clock differences and the expected values is ±0.129 s for the
hydrophone signals (calculated with eq. (4.10)). This is slightly larger than
the standard deviation of the differences for results of the vertical seismic
components (±0.087 s).
Furthermore, we are able to give a rough estimate for the annual clock drift
at station D05 of 0.775 ± 0.012 s a≠1 by comparing the estimated and the
expected annual clock differences for all hydrophones with constant drift rates
(Table 4.3).
The partly malfunction of the hydrophones at station D02 and D12 is
reflected by the estimation of instable apparent drift rates for station pairs
involving these stations in the periods of the malfunction (compare Fig. 4.8
first 100 d). Besides this period, the extracted clock differences show a clock
drift which is similar to the clock drift estimated by the correlation of the
vertical seismic data (Fig. 4.8). The nearly total malfunction of the hydrophone
at station D03 leads to a complete vanishing of any signal in the correlation
results after 200 d of recording (mid of 2012 January).
33
Figure 4.8: Comparison of the clock differences estimated from 20 d stacks of the
vertical seismic components (green) and hydrophone signals (black) of station pair
D01 and D02. The arrows indicate the period of malfunction and when no data
errors occured.
Table 4.3: Annual clock differences for station D05 as reference station, tm is the
estimated annual clock difference from the 20 d stacks for the hydrophone signals (the
error is estimated by using eq. (4.6)) and tth is the expected annual clock difference
calculated from the measured skews, tdif f gives the differences between estimated
and expected clock difference.
OBS tm [s] tth [s] tdif f [s]

D01 0.434 ± 0.024 -0.319 0.753 ± 0.024
D04 0.264 ± 0.018 -0.615 0.880 ± 0.018
D06 0.652 ± 0.040 0.047 0.606 ± 0.040
D07 0.277 ± 0.024 -0.462 0.740 ± 0.024
D09 -0.018 ± 0.007 -1.049 1.031 ± 0.007
D10 0.069 ± 0.047 -0.640 0.709 ± 0.047
D11 -0.433 ± 0.047 -1.138 0.705 ± 0.047
mean tdif f [s] 0.775 ± 0.012
4.6 Conclusion and outlook

In summary, we find that the linear clock drift is visible in the correlation results
of the vertical seismic components and that relative annual clock differences
can be estimated except for station pairs involving D05 which had a clamped
vertical seismic component. Furthermore, the comparison with expected annual
clock differences determined by the measured skews and the assumption of
a linear drift shows a good agreement (standard deviation of differences is
±0.087 s). Additionally, we calculate the amount of a static time offset at
station D09. Moreover, we prove that the shifted correlation time window
approach gives usable results for station D08 which had a large clock drift and
can be applied to test for a range of clock drifts to find the most likely. It
should be mentioned that the verification of the clock drift with events might
be helpful, but it is not required. So even in the case of short term deployments
where no or only few large earthquakes occur within the recording period,
34
Bibliography
the application of ambient noise cross-correlation for clock drift estimations is

possible as long as the time period of the recording is sufficient to get a stable
correlation result with a sufficient SNR for several consecutive time windows.
Furthermore, we find that the linear clock drift is also visible for most of
the hydrophones (standard deviation of differences is ±0.129 s). We were able
to estimate an annual drift rate for station D05 which had a clamped vertical
seismic component by using the correlation results of the hydrophone signals.
Furthermore, we find that instable drift rates at some hydrophones indicate
periods of malfunction at these sensors.
Another possibility to estimate drift rates of instruments with clamped ver-
tical seismic components might be the correlation of the horizontal components,
although it is not clear whether and to which degree orientation, which is usu-
ally unknown for free-fall OBS stations, is important for this purpose. Besides,
a test-wise rotation of the horizontal components of one station with known
orientation might offer an additional information about the orientation of the
horizontal components of the other stations. Preliminary tests give promis-
ing results, but need further investigations concerning orientation-dependent
quality measures.
It should be mentioned that the processing of ambient noise cross-correlation
highly depends on the noise field, noise source distribution and strength, the
structure beneath the stations, interstation distances and recording length (for
further details concerning the processing of ambient noise cross-correlation see
Bensen et al., 2007). To determine optimal processing options, we examined
the pseudo-shot gather of the cross-correlation functions of all stations, whether
a propagating wavefront is clearly visible and furthermore, we had a look at the
stability of the cross-correlation results within the consecutive time windows
used for the analysis.
Finally, we conclude that ambient noise cross-correlation is one feasible
method to check the timing and synchronization of deep water, km-scale OBS
arrays. It also can be applied off-line a posteriori to correct data already
recorded. The method works well for vertical seismic components, but needs
further investigations in case of hydrophone signals and horizontal components.
Acknowledgments
The authors want to thank the DEPAS pool for providing the instruments for
the DOCTAR project which is funded by the DFG. The examination of the
data was done using Seismic Handler Stammler (1993). All figures were created
using GMT (Generic Mapping Tools, Wessel and Smith, 1991).
Bibliography
Bensen, G. D., Ritzwoller, M. H., Barmin, M. P., Levshin, A. L., Lin, F.,
Moschetti, M. P., Shapiro, N. M., and Yang, Y. (2007). Processing seismic
ambient noise data to obtain reliable broad-band surface wave dispersion
measurements. Geophysical Journal International, 169(3):1239–1260.
Dahm, T., Thorwart, M., Flueh, E. R., Braun, T., Herber, R., Favali, P.,
Beranzoli, L., D’Anna, G., Frugoni, F., and Smriglio, G. (2002). Ocean bot-
35
Bibliography
tom seismometers deployed in Tyrrhenian Sea. Eos, Transactions American

Geophysical Union, 83(29):309.
Geissler, W. H., Matias, L., Stich, D., Carrilho, F., Jokat, W., Monna, S.,
IbenBrahim, A., Mancilla, F., Gutscher, M.-A., Sallarès, V., and Zitellini, N.
(2010). Focal mechanisms for sub-crustal earthquakes in the Gulf of Cadiz
from a dense OBS deployment. Geophysical Research Letters, 37(18):L18309.
Picozzi, M., Parolai, S., Bindi, D., and Strollo, A. (2009). Characterization of
shallow geology by high-frequency seismic noise tomography. Geophysical
Journal International, 176(1):164–174.

Sabra, K. G., Roux, P., Thode, A. M., D’Spain, G. L., Hodgkiss, W. S.,
and Kuperman, W. A. (2005). Using ocean ambient noise for array self-
localization and self-synchronization. IEEE Journal of Oceanic Engineering,
30(2):338–347.
Sens-Schönfelder, C. (2008). Synchronizing seismic networks with ambient

Stammler, K. (1993). Seismichandler-Programmable multichannel data handler

for interactive and automatic processing of seismological analyses. Computers
and Geosciences, 19(2):135–140.
Stehly, L., Campillo, M., and Shapiro, N. M. (2006). A study of the seismic
noise from its long-range correlation properties. Journal of Geophysical
Research: Solid Earth, 111(10):B10306.
Stehly, L., Campillo, M., and Shapiro, N. M. (2007). Traveltime measurements

from noise correlation: Stability and detection of instrumental time-shifts.
Wessel, P. and Smith, W. H. F. (1991). Free software helps map and display
data. Eos, Transactions American Geophysical Union, 72(41):441.
36
5. Structure of the oceanic crust
and upper mantle north of the
Gloria fault in the eastern mid
Atlantic by receiver function
analysis
K. Hannemann, F. Krüger, T. Dahm, D. Lange

submitted to Journal of Geophysical Research: Solid Earth
Summary
Receiver functions (RF) have been used since several decades to study structures
beneath seismic stations. Although most available stations are deployed on-
shore, the number of ocean bottom station (OBS) experiments increased
worldwide in recent years. Almost all OBS have to deal with higher noise
levels and a limited deployment time (≥1 a), which results in a small number
of usable teleseismic earthquakes. Here, we use OBSs deployed in the deep
ocean (4.5-5.5 km) North of the Gloria fault (eastern mid Atlantic). We exploit
special quality measures for OBS data and beamforming to enhance the quality
of the RFs. Only little Moho depth variations (6-6.5 km) have been observed
across our array. We identify clear signals for the lithosphere asthenosphere
boundary (LAB, ≥40-60 km) and for a discontinuity at ≥90-110 km, which
might be caused by small melt amounts in the upper asthenosphere, probably
also being responsible for elevated LAB depths beneath our array compared to
mineral-physics’ estimates. The lower boundary of the asthenosphere (Lehmann
discontinuity), and mantle discontinuities at ≥410 km and ≥660 km are clearly
identifiable in the stacks of all stations and the beamformed traces. We observe
an additional weak phase eventually being related to the mantle discontinuity at
≥520 km. The phase arrival of the Lehmann discontinuity is slightly advanced
compared to PREM, maybe being related to a wet uppermost mantle, while the
’410’ is found at expected delay times. For some back-azimuths, an additional
phase after the ’660’ indicates the presence of the majorite-perovskite transition.
5.1 Introduction
More than 70% of the Earth are covered by oceans and the majority of the
oceanic crust is not affected by volcanic or tectonic activities. Most of our
37
Chapter 5. Structure of oceanic crust and upper mantle by OBS RF
knowledge of the oceanic mantle is based on global surface wave tomography

(e.g. Romanowicz, 2009) with rather good path coverage in the oceans but
low resolution for sharp discontinuities. Furthermore, studies using land based
stations at teleseismic distances have been conducted to analyze underside
reflections to resolve the oceanic mantle structures (Gossler and Kind, 1996;
Gu et al., 1998; Flanagan and Shearer, 1998; Gu and Dziewonski, 2002; Deuss
et al., 2013; Saki et al., 2015). Both method lack spatial resolution.
The P wave velocity-depth structure of the oceanic crust and uppermost
mantle has been characterized by active geophysical experiment (e.g. White
et al., 1992). In recent years, several passive large scale experiments with ocean
bottom stations (OBS) have been conducted (e.g Friederich and Meier, 2008;
Barruol and Sigloch, 2013; Gao and Schwartz, 2015). Nevertheless, most of the
OBS studies are located at mid oceanic ridges (e.g Shen et al., 1998a; Tilmann
and Dahm, 2008; Jokat et al., 2012; Grevemeyer et al., 2013; Hermann and
Jokat, 2013; Schlindwein et al., 2013, 2015), hot spots (e.g. Suetsugu et al.,
2007; Barruol and Sigloch, 2013; Davy et al., 2014), subduction zones (e.g
Suetsugu et al., 2010; Kopp et al., 2011; Laigle et al., 2013; Ruiz et al., 2013;
Grevemeyer et al., 2015; Janiszewski and Abers, 2015) or the transition from
continent to ocean (e.g. Czuba et al., 2011; Grad et al., 2012; Libak et al.,
2012; Suckro et al., 2012; Monna et al., 2013; Altenbernd et al., 2014; Kalberg
and Gohl, 2014), and are thus not representative for undisturbed oceanic crust
and mantle. Receiver function (RF) analysis provides a strong tool to image
discontinuities in the lithosphere and the upper mantle down to the transition
zone with rather high lateral resolution (e.g Vinnik, 1977; Langston, 1979).
Receiver function (RF) studies of rather undisturbed oceanic lithosphere were
conducted within the Philippine plate and the western Pacific plate using 500 m
deep borehole stations and OBSs (Suetsugu et al., 2005; Kawakatsu et al., 2009;
Olugboji et al., 2016).
This study focus on OBS RFs in the eastern mid Atlantic in the vicinity of
the Eurasian-African plate boundary, in which - up to our knowledge - no OBS
RF study has been done yet. The Mohorovi iÊ discontinuity (Moho) marks the
boundary between oceanic crust and mantle and is expected in depths between
5-8 km (e.g. White et al., 1992), depending on the age of the crust. The RF
phase of the Moho arrives only one or two seconds after the dominant P phase
(e.g Kawakatsu et al., 2009) and therefore requires high frequent data (Audet,
2016) and a good signal-to-noise ratio (SNR) to be detected. Thickened or
thinned oceanic crust may be related to overthrusting, underplating, or basin
formation.
The lithosphere-asthenosphere boundary (LAB) beneath the oceans is
often imaged using surface wave tomography (e.g Romanowicz, 2009) or RFs
employing land stations (e.g. Li et al., 2000; Kumar and Kawakatsu, 2011)
or OBSs (e.g. Kawakatsu et al., 2009; Olugboji et al., 2016). Most of the
discussed models of the LAB (Kawakatsu et al., 2009; Fischer et al., 2010;
Olugboji et al., 2013) include thermal control, changes in rheology, dehydration,
anisotropy or melt fraction. Besides the depth of the LAB, the sharpness of
the discontinuity is of interest. A relatively smooth transition is expected
for a pure thermal control of the LAB (Olugboji et al., 2013) and a sharp
discontinuity for a pure compositional control (e.g. change in water content,
Karato and Jung, 1998). For instance, land based S RF study of oceanic
lithosphere in subduction zones support the model of thermal control, but the
38
5.1. Introduction
observed scatter in their observations indicates additional controlling factors

Kumar and Kawakatsu (2011). Observations of the LAB indicate a diffuse
age-dependent boundary in young oceans and a sharp age-independent LAB at
≥70 km in old oceans (e.g. Fischer et al., 2010; Karato, 2012; Olugboji et al.,
2013). A sub-solidus model which assumes grain boundary sliding (Karato,
2012; Olugboji et al., 2013) indeed predicts a transition from an age-dependent
diffuse LAB roughly following the 1300 K isotherm in young oceans to a sharp
discontinuity at constant depth in old oceans. The age, at which this transition
happens, depends on the thermal model used for the modeling and lies between
40-80 Ma (Karato, 2012) or 55-75 Ma (Olugboji et al., 2013).
The Lehmann discontinuity marks the lower boundary of the asthenosphere
(e.g. Lehmann, 1961; Dziewonski and Anderson, 1981; Deuss et al., 2013).
Only few RF (Shen et al., 1998a) and SS precursor observations (Deuss and
Woodhouse, 2002) exist of the oceanic Lehmann discontinuity. It is located
around 220 km depth. Its cause is still debated (Karato, 1992; Deuss and
Woodhouse, 2004).
The three global mantle discontinuities at approximately 410 km, 520 km
and 660 km depth (referred to as ’410’, ’520’ and ’660’, respectively) are
associated with phase transitions in olivine or the Aluminium phases (e.g.
garnet) of the mantle (e.g Agee, 1998; Helffrich, 2000; Deuss et al., 2013). The
mantle transition zone (MTZ) which is defined by the ’410’ and the ’660’ has
mostly been studied in the ocean by using PP and SS precursors (Gossler and
Kind, 1996; Gu et al., 1998; Flanagan and Shearer, 1998; Gu and Dziewonski,
2002; Deuss et al., 2013; Saki et al., 2015). There are also some global RF
studies of the MTZ (Chevrot et al., 1999; Lawrence and Shearer, 2006; Tauzin
et al., 2008) and local studies focusing on the MTZ using OBS data (Shen et al.,
1998a; Gilbert et al., 2001; Suetsugu et al., 2005, 2007, 2010). Some global
studies of SS precursors suggest a thinner MTZ beneath the oceans than below
the continents (Gossler and Kind, 1996; Gu et al., 1998; Gu and Dziewonski,
2002) whereas Flanagan and Shearer (1998) (SS precursors) and Chevrot et al.
(1999) (RF) could not observe such a correlation. The lack of correlation is
also confirmed by local studies (Shen et al., 1998a,b; Silveira et al., 2010).
Here, we use data from 11 ocean bottom stations (OBS) located in the east-
ern mid Atlantic approximately 100 km North of the Gloria Fault (Figure 5.1)
to investigate the structure of the oceanic crust and upper mantle using array
techniques (e.g. beamforming, Rost and Thomas, 2002) and receiver functions.
Working with data of OBS is usually characterized by a small amount of good
quality events within the short recording period of the employed instruments
(Webb, 1998). One of the main reasons is the often low signal-to-noise ratio
(SNR) at ocean bottom stations (Webb, 1998; Dahm et al., 2006), especially on
the horizontal components. There have been different strategies to increase the
number of usable events: either by reinstalling the OBS at the same site (e.g.
Cascadia Initiative Janiszewski and Abers, 2015) or by using array techniques
(Thomas and Laske, 2015). The latter is known to increase the SNR of an
event by coherent stacking of the observed signals (Rost and Thomas, 2002).
The experiment presented here was designed in such way that the stations
form a mid aperture array with inter-station distances of 10-20 km and an
maximum aperture of 75 km (Fig. 5.1). This design allows us to stack all
stations to enhance phases originating from the deeper parts of the upper
mantle. Additionally, we employ a quality control by using evaluation criteria
39
like relative spike position within the deconvolution time window and SNR of
the RFs to choose optimal deconvolution lengths for each recording.
5.2 Data
In 2011, twelve OBS (Figure 5.1b) were installed in the deep sea (4.5-5.5 km) of
the eastern mid Atlantic. These stations were equipped with three component
broad-band seismometers (Guralp CMG-40T, 60 s - 50 Hz) and hydrophones
(HighTechInc HTI-04-PCA/ULF, 100 s - 8 kHz, flat instrument response
down to 5s, at D08 down to 2 s) and recorded 100 Hz data. To obtain an
accurate clock drift, we used ambient noise cross-correlation and compared it
to the drift calculated from the synchronization with GPS to reveal static time
offsets (Hannemann et al., 2014). Afterwards, we used the pyrocko toolbox
(emolch.github.io/pyrocko) to apply a time correction by inserting and deleting
samples. Station D05 has not been used for the analysis because of two clamped
seismometer components.
18.8 18.4 18.0

N km
38.8 150 km
0 12.5 25
D11
D07 D06
4000
D04
38.4 D10 D01 D12
W E D03 D02
60° D08 D05
120° 70 km
180° D09
38.0
towards Gloria fault 5000
S
40.0 OBS array Portugal
Events used for
Azores
P/ PKP RF (single)
P/ PKP RF (beam) 35.0
km Madeira
0 500 6000
30.0 20.0 10.0
Figure 5.1: (a) Azimuthal equidistant plot of used events for RF analysis given as
filled yellow circles (P and PKP single station RFs) and as open orange circles (P
and PKP beam RFs). The event details are listed in Tab. 5.S3. (b) The top map
shows the array configuration for the OBS. The color scale gives the water depth
from a local bathymetry (EMEPC, Task Group for the Extension of the Continental
Shelf). The white dashed lines give the distance towards the Gloria fault. The black
and white map shows the location of the OBS array within the eastern mid Atlantic
Ocean and the Eurasian-African plate boundary (Gloria Fault, Bird, 2003). The
red triangle marks station D05 which had two clamped seismometer components.
For the used free fall OBS stations, the orientation of the vertical component
is aligned by a gimbaling system (Stähler et al., 2016). Since for OBS data
the orientation of the horizontal components is unknown, we use P-phase
polarization and Rayleigh and Love waves (Thorwart, 2006; Stachnik et al.,
40
5.3. Method
2012; Sumy et al., 2015) to align the horizontal components to North and East
(see appendix A for details).
We used the results of the P phase for the rotation of the horizontal traces,
because we find from a frequency-wavenumber analysis with a moving time
window (Rost and Thomas, 2002) that the estimated back-azimuths of the
P phases are more precise than those of the Rayleigh phases. They show on
average a smaller deviation from the expected back-azimuths (see also Thorwart,
2006).
For the RF calculation, we examine all events which have a body wave
detection in our frequency-wavenumber detector including P between ≥30¶ -90¶
epicentral distance, Pdiff between ≥90¶ -110¶ epicentral distance and PKPdf
between ≥140¶ -160¶ epicentral distance. The events finally used (single: 25,
beams: 37, see Tabs 5.S3 and 5.S4 in the appendix) are chosen based on
evaluation criteria which are described below.
5.3 Method
When the up-going compressional wave (P wave) is incident on an interface
within the Earth, it is partly refracted as a P wave and as a vertical polarized
shear wave (SV wave). The latter are also referred to as P-to-S (Ps) conversion
and are secondary phases which arrive later than the direct P phase (i.e. the
refracted P waves). The amplitudes of the Ps conversions are typically several
ten times smaller than those of the direct P phase, depending on the impedance
contrast at the discontinuity and the ray angle (e.g. Chevrot et al., 1999).
The relative Ps amplitudes can be calculated from the ratio of the refraction
coefficient Ṕ Ṕx and Ṕ Śx for which x indicates the depth of the discontinuity
(see Aki and Richards, 2002, for definition of acute accents). Additionally to
the problem of small relative amplitudes, the identification of Ps phases is
often obscured by ambient noise, multiple reflections at the receiver. Therefore,
specific deconvolution and stacking methods were developed to enhance the
signal-to noise ratio (SNR) of the weak Ps phase. We perform the RF calculation
analogue to Kieling et al. (2011) by rotating the seismograms into the coordinate
system of the ray, LQT (L=longitudinal in ray direction, Q=orthogonal to the
ray in the the vertical plane, T=horizontal transversal). We use the theoretical
incidence angle (epicentral distance < 98¶ Pho and Behe, 1972, else obtained
with horizontal slowness p (IASP91) and crustal velocity vp = 7 km s≠1 ).
The L and Q component are both deconvolved with the L trace using a time
domain Wiener filter (Kind et al., 1995) with damping parameter 0.01. For the
estimation of the Wiener filter, we use the built-in function “spiking” of Seismic
Handler (Stammler, 1993) which performs a least-squares inversion (Berkhout,
1977). This function determines the spike position for which the inversion
is done by estimating the center of the signal tc in the given deconvolution
time window tdec . We have to be aware that the inversion to determine the
Wiener filter works best for minimal phase signals (e.g. Scherbaum, 2007).
The resulting RF shows several spikes representing converted phases and their
multiples from different interfaces/ discontinuities. The spikes of the secondary
phases should be separated from the spike of the direct phase (Vinnik, 1977;
Langston, 1979). The amplitudes and delay times of the spikes of the secondary
phases constrain the impedance contrast and the depth of the interfaces under
41
investigation. The deconvolution method removes source effects from the RFs,
so that RFs from different events can be stacked after the traces have been
stretched to represent time functions on a common ray path. For the distance
moveout correction (Yuan et al., 1997), we use a reference distance of 67¶ and
a global velocity model (oceanic PREM, Dziewonski and Anderson, 1981).
The determination of RFs at OBS might be influenced by water multiples
(e.g. Thorwart and Dahm, 2005) or noise (tilt or water wave compliance, e.g.
Bell et al., 2015) which can be corrected on the vertical component by using
hydrophone data (e.g. Thorwart and Dahm, 2005; Bell et al., 2015) or the
horizontal components (e.g. Bell et al., 2015). We do not observe water multiples
in our teleseismic recordings and therefore we do not apply any correction for
them. Furthermore, the water wave compliance is only present at very low
frequencies (≥100 s) in 4.5-5.5 km water depth (e.g. Crawford et al., 1998;
Bell et al., 2015). We avoid this frequency band by high-pass filtering the RFs.
The tilt noise can not be excluded in our case and might influence the RFs
at periods longer than ≥10 s. Tilting (e.g. movement of the OBS frame by
currents, Webb, 1998; Crawford et al., 1998; Bell et al., 2015) has a higher
influence on the horizontal components than on the verticals and a correction
of the vertical component (e.g. Bell et al., 2015) would probably lead to rather
similar results for the estimated RFs (e.g. Janiszewski and Abers, 2015). We
therefore do not remove the tilt noise from the vertical component.
During processing, we find that the SNR of a RF is mainly determined by
the quality of the earthquake recording and the length of the deconvolution time
window used for the determination of the Wiener filter. To obtain RFs with
sufficient SNR, we follow two approaches: (1) increase the SNR of earthquake
recordings by employing beamforming (Rost and Thomas, 2002) using either the
plain recordings or normalize the recordings to the rms amplitude of the noise
(-200 s to -100 s before P onset) on L or Q to optimize destructive interference
of noise amplitudes and (2) introduce a quality control which employs a set of
evaluation criteria to select a subset of deconvolution lengths for single station
recordings and beams.
To determine whether the chosen time window contains a minimal phase
signal or mainly noise, the first evaluation criterion is the spike position trel
relative to the deconvolution time window tdec in percent:
1 2
100 · tc ≠ tdec
2
trel = . (5.1)
tdec
In case of a mainly minimal phase signal, the spike position is located within
the first half of the deconvolution time window (trel < 0%). On the other hand,
if the time window contains mainly noise, the spike position is in the middle of
the deconvolution time window (trel ¥ 0%).
The success of the deconvolution is estimated by the SNR of the L component
of the RF (SNRL/L ) which is the second evaluation criterion. It is determined
by estimating the ratio of the squared rms amplitudes in the signal time window
(-10 s to 10 s relative to P spike) and noise time window (-55 s to -25 s before
P spike) of the L component.
In order to quantify the success of resolving upper mantle discontinuities,
the third and last evaluation criterion is the ratio of the squared rms amplitudes
of the signal time window on the L component of the RF and the noise time
42
5.3. Method
window on the Q component of the RF (SNRL/Q ). We compare this to the

theoretical ratios of the refraction coefficients ṔṔ ṔŚ410 and ṔṔ ṔŚ660 . We test different
410 660
deconvolution time window lengths tdec starting with 30 s and increasing it
in 5 s steps to a time window length which approximately equals the time
difference between the P onset and the PP phase arrival. The deconvolution
length for each single event recording and beam is chosen by the following four
steps:
1. trel < 0% for minimal phase signals (Figs. 5.2 a and 5.3 a)
2. SNRL/L & 10 for a good deconvolution (Figs. 5.2 b and 5.3 b)
3. SNRL/Q & ṔṔ ṔŚ660 and/ or SNRL/Q & Ṕ Ṕ410

Ṕ Ś410
for resolving upper mantle
660
discontinuities (Figs. 5.2 c and 5.3 c)
4. manual revision of remaining RFs (Figs. 5.2 d and 5.3 d)
The fourth step including the manual revision is required to exclude RFs
which are influenced by high frequency noise or ringing. These disturbed RFs
are often hard to distinguish from undisturbed RFs, even with the simple
evaluation criteria employed here.
For the single stations, we additionally check the stacked spectra of all used
events for each component (Z,N,E) and find that no strong sedimentary rever-
berations with overtones are present in the data. Nevertheless, a probabilistic
power spectral density (PPSD, McNamara, 2004) reveals a resonance like effect
on all three seismometer components of each station (Hannemann et al., 2016).
This effect is also visible in the raw data Figure 5.2 d and can be linked to a
sedimentary cover and is different for each station. The resonance effect mainly
influences the amplitudes of the RFs and can be suppressed by using longer
time windows (Figure 5.2 d). Using synthetic RFs, we find that a sedimentary
cover could lead to a bias in the estimated Moho depth of several kilometers
(± ≥ 3 km, Sheehan et al., 1995). The most prominent feature of a RF
influenced by a sedimentary cover is a time shift in the P spike (Sheehan et al.,
1995), this is hard to evaluate in the unfiltered RFs presented in Figure 5.2 d.
Using the events from Tabs 5.S3 and 5.S4, we find that beamforming
improves - as expected - the SNRL/L and the SNRL/Q and that this effect can
be enhanced by a normalization of the individual traces before stacking (red
and blue lines compared to yellow lines in Figures 5.3 b-c). We therefore just
present the RFs for the beams with pre-normalization to the noise on either L
or Q component.
43
0.5
D03 T 0
0
0.5
D03 Q 0
0.5
D03 L 0
10
0.5
D03 T 0
0.5
D03 Q
20 0
30 60 90 120 D03 L
0.5
0
0.5
60 D03 T 0
0.5
D03 Q 0
40 0.5
D03 L 0
0.5
D03 T 0
20
0.5
D03 Q 0
0 D03 L
0.5
0
30 60 90 120
0.5
D03 T 0
60 D03 Q
0.5
0
0.5
D03 L 0
40
4
D03 T 0 x104
4
4
20 D03 Q 0 x104
4
4
D03 L 0 x104
4
0
30 60 90 120 0 20 40 60
Figure 5.2: Example for evaluation criteria shown against deconvolution time window
length tdec and selection of deconvolution time window length for station D03 and
event # 36. The circles mark the values for the RFs presented in d. The blue circle
gives the finally chosen deconvolution length (95 s). (a) Relative spike position trel
estimated with e. (5.1). (b) The SNR on the L component. (c) The ratio of the
signal time window on L and the noise time window on Q. As guidance the ratios
of the refraction coefficients of the refracted P and SV waves at the ’410’ ( ṔṔ ṔŚ410 ,
410
dotted line) and the ’660’ ( ṔṔ ṔŚ660 ,

dashed line) are given. (d) The unfiltered raw
660
data (in counts) and the RFs (normed to P spike on L) for the different pre-selected
deconvolution lengths (circles in panels a-c).
44
5.3. Method
0.5
QNR T 0
0 0.5
QNR Q 0
0.5
QNR L 0
15
0.5
LNR T 0
0.5
LNR Q
30 0
30 60 90 120 LNR L
0.5
0
0.5
PLN T 0
0.5
200 PLN Q 0
0.5
PLN L 0
20
100 QNR T 0
20
20
QNR Q 0
20
20
0 QNR L 0
30 60 90 120 20
40
LNR T 0
40
40
LNR Q 0
90 40
40
LNR L 0
40
60 4
PLN T 0 x104
4
4
30 PLN Q 0 x104
4
4
PLN L 0 x104
4
0
30 60 90 120 0 20 40 60
Figure 5.3: Same as Figure 5.2 but for traces resulting from beamforming for event
# 36 and different normalization of single station recordings (PLN - no normalization
(yellow), LNR - normalized to rms amplitude of noise on L (red), QNR - normalized
to rms amplitude of noise on Q (blue)). The gray lines give the evaluation criteria for
station D03 (Figures 5.2 a-c). The circles in a-c give the values for a deconvolution
length of 110 s as is used for the RFs in d.
45
5.4 Results
In this study, we mainly interpret the time difference between the converted
(Ps) and the direct (P) phase (hereafter referred to as delay time). We perform
a bootstrap (Efron and Tibshirani, 1986) to estimate uncertainties for the
picked delay times. For this purpose, we randomly choose the receiver functions
(RF) before stacking the distance moveout corrected traces and repeat this
procedure for 300 trials (e.g. Suetsugu et al., 2010).
Before stacking, we employ a distance moveout correction (Yuan et al.,
1997) using a reference distance of 67¶ and a global velocity model (oceanic
PREM, Dziewonski and Anderson, 1981).
37 QNR
37 LNR
25 SUM
11 D12
15 D11
16 D10
14 D09
14 D08
18 D07
17 D06
15 D04
12 D03
17 D02
9 D01
2 0 2 4 6
Figure 5.4: Bandpass filtered P receiver functions (0.5 s to 60 s). The gray traces show
the RFs of single events and the black lines the stacked trace for the corresponding
station, all stations (SUM), beamformed traces normalized to rms amplitude of noise
on L (LNR) and beamformed traces normalized to rms amplitude of noise on Q
(QNR). The numbers in front of the trace names give the number of events which
contributed to the stacked traces. The marker gives the position of the Moho signal
estimated with one standard deviation by picking 300 bootstrapped stacked traces.
In Figure 5.4, we show the bandpass filtered (0.5 s to 60 s) RFs for the
single OBSs (D01-D12), the stack of all stations (SUM) and the stack of the
beamformed traces (LNR and QNR). We observe a positive peak at ≥0 s at
all stations and therefore assume no large biasing influence of a sedimentary
cover (Sheehan et al., 1995). The estimated delay times of the Moho are in
46
5.4. Results
37 QNR
37 LNR
25 SUM
11 D12
15 D11
16 D10
14 D09
14 D08
18 D07
17 D06
15 D04
12 D03
17 D02
9 D01
10 0 10 20
Figure 5.5: Bandpass filtered P receiver functions (4 s to 40 s, black and 6 s

to 40 s, red). Shown are the stacked traces for the single stations, all stations
(SUM), beamformed traces normalized to rms amplitude of noise on L (LNR) and
beamformed traces normalized to rms amplitude of noise on Q (QNR). The numbers
in front of the trace names gives the number of events which contributed to the
stacked trace. The markers indicate the position of the signal corresponding to the
lithosphere-asthenosphere boundary (LAB) and a mid asthenospheric discontinuity
(MAD) estimated with one standard deviation by picking 300 bootstrapped stacked
traces.
the range of 0.41-1.22 s (Figure 5.4 and Tab. 5.1). Using the PREM velocity
model (Dziewonski and Anderson, 1981, P wave velocity vp = 6.8 km s≠1 and
S wave velocity vs = 3.9 km s≠1 ), we convert the delay times at each station
in pseudo depths, i.e. depths estimated with an average velocity model (e.g.
Knapmeyer-Endrun et al., 2014). The resulting pseudo depths lie between 3.5
and 10.7 km (black squares in Figures 5.6 c-d).
In Figure 5.5, we present the RFs filtered in two different period bands
(4-40 s in black and 6-40 s in red) to analyze the stability of the identified
phases. Most of the single station traces and the summed traces as well as
the beamformed traces show a double polarity phase (i.e. a negative phase
followed by a positive phase) with a small peak-to-peak amplitude (14-36% of
direct P phase). We interpret the negative phase as LAB. Except for station
D01 the LAB is at similar positions in both filter bands. The delay times of
47
the LAB (Tab. 5.1) are in the range of 4.48-6.47 s (4-40 s, excluding D01 with
3.65 s) and 4.65-5.86 s (6-40 s). If we use PREM velocities for the lithospheric
mantle (vp = 8.1 km s≠1 and vs = 4.5 km s≠1 ), we get LAB pseudo depths
between 41.6-60.8 km (4-40 s, D01: 29.9 km, black circles in Figures 5.6 e-f)
and 43.7-55.3 km (6-40 s, red circles in Figures 5.6 e-f). The later arriving
small positive phase has similar delay times for both filter bands except for
station D12 at which it is shifted to later times for the filter band 6-40 s
(Tab. 5.1). The estimated delay times are between 9.44-12.22 s (4-40 s, D12:
7.52 s) and 9.75-12.03 s (6-40 s, D12: 8.87 s). If we assume a low velocity
layer (LVL) with vp = 7.9575 km s≠1 and vs = 4.4505 km s≠1 (PREM), the
estimated pseudo thicknesses for this layer are between 39.0-71.6 km (4-40 s,
D12: 23.8 km, pseudo depths: 70.9-115.3 km, black crosses in Figures 5.6 e-f)
and 47.3-53.9 km (6-40 s, D12: 35.4 km, pseudo depths: 83.7-113.5 km, red
crosses in Figures 5.6 e-f). In the following, we refer to this feature as mid
asthenospheric discontinuity (MAD), because its pseudo depths are too shallow
for the Lehmann discontinuity (180-240 km Deuss and Woodhouse, 2004).
Table 5.1: Delay times for different stations, stacked traces and beamformed
traces for Moho, lithosphere-asthenosphere boundary (LAB) and mid asthenospheric
discontinuity (MAD). The times are estimated by picking the corresponding phase
on 300 stacked traces which were formed by bootstrapping the contributing traces.
station tM oho (s) tLAB,4≠40 (s) tLAB,6≠40 (s) tM AD,4≠40 (s) tM AD,6≠40 (s)
D01 1.22±0.06 3.25±0.85 4.71±0.88 10.79±0.60 10.39±0.35
D02 0.62±0.10 4.78±0.10 5.02±0.09 10.28±0.24 10.01±0.26
D03 0.54±0.06 4.53±0.08 4.93±0.35 10.09±0.29 10.41±0.20
D04 0.81±0.10 4.67±0.30 5.72±0.68 10.32±0.59 10.70±0.54
D06 0.77±0.02 4.48±0.34 4.65±0.12 9.44±0.23 9.75±0.66
D07 0.74±0.04 6.47±0.94 5.72±0.32 10.58±1.04 11.39±0.67
D08 0.41±0.04 6.40±0.99 5.86±0.41 11.62±0.59 11.12±0.37
D09 0.72±0.03 4.58±0.25 4.91±0.21 10.47±0.35 10.13±2.13
D10 1.08±0.04 5.84±1.25 5.71±0.37 11.40±0.64 11.63±0.72
D11 0.86±0.06 5.75±0.11 5.38±0.21 12.22±0.40 12.03±0.40
D12 0.64±0.06 5.01±0.23 5.14±0.11 7.52±0.91 8.87±1.15
SUM 0.70±0.04 4.77±0.12 5.01±0.08 10.63±0.32 10.56±0.17
LNR 0.74±0.10 5.24±0.25 5.16±0.13 10.33±0.55 10.79±0.30
QNR 0.67±0.07 5.19±0.20 5.13±0.10 10.37±0.57 10.67±0.32
Additionally, we perform a common conversion point stack (Figure 5.6,

e.g. Knapmeyer-Endrun et al., 2014) in which the delay times are converted
to depths and the single RF is back-projected along the corresponding ray
path using oceanic PREM (Dziewonski and Anderson, 1981). We stack the
amplitudes on a grid with 2 km horizontal and 1 km vertical spacing. This
procedure is done for the direct phase Ps. We generate a N-S and a W-E profile
from the final grid by averaging the amplitudes in 3 km bins along the profile
for grid nodes which are within a distance of ±20 km to the profile.
In Figure 5.6 a (or b), we present 32 cross-profiles of the bathymetry shown
in Figure 5.1 b which are parallel to the N-S profile (or the W-E profile) and
located within a distance of ±20 km to the profile. Below in Figures 5.6 c-d, we
show the resulting N-S and W-E profiles of the CCP stack of 0.5-60 s bandpass
filtered RF data. The most prominent feature in this profiles is the area with
48
5.4. Results
D03 4 4 D01
D03D09 D02
D01 D08 D07
D10
D07 D04
D11
D02 D08 D09 5 5 D12
D10
D12 D11 D06
D06 D04
6 6
38.8 38.6 38.4 38.2 38.0 18.9 18.6 18.3 18.0
38.8 38.6 38.4 38.2 38.0 18.9 18.6 18.3 18.0

0 0
0.06
10 10
0.03
20 20
0.00
30 30
0.03
40 40
0.06
50 50
140 120 100 80 60 0 20 40 60 80
140 120 100 80 60 0 20 40 60 80

0 0
0.02
30 30
0.01
60 60
0.00
90 90
0.01
120 120
0.02
150 150
38.8 38.6 38.4 38.2 38.0 18.9 18.6 18.3 18.0
Figure 5.6: N-S and W-E common conversion point (CCP) stack profiles. The
brighter areas in each profile indicate regions in which the hit-count is below 20% of
the maximum hit-count for the profile. (a) N-S profiles of bathymetry (Figure 5.1 b)
±20 km from CCP profile. (c) CCP N-S profile for band-passed data (0.5 s-60 s).
The black squares indicate the pseudo depths estimated for the Moho at the single
stations. (e) CCP N-S profile for band-passed data (4 s-60 s). The circles indicate
the LAB pseudo depths and the crosses the MAD pseudo depths estimated at the
single stations for 4-40 s (black) and 6-40 s (red) band-passed data. (b), (d) and (f)
same as a, c and e but for W-E profile.
49
positive amplitudes (red colors) between 5 km to 10 km depth. Additionally, we

show the estimated pseudo depths of the Moho for each station as black squares.
As expected, the positions of the squares coincide with the before mentioned
area of positive amplitudes for most cases (Figures 5.6 c-d). Figures 5.6 e-f show
the resulting N-S and W-E profiles of the CCP stack of 4-60 s bandpass filtered
RF data. There we observe two sub-parallel elongated areas for which the upper
one has negative amplitudes (blue colors, 30-60 km) and the lower one has
positive amplitudes (red colors, 80-110 km). Moreover, we show the estimated
pseudo depths for each station for the LAB (black and red circles) and the
MAD (black and red crosses) in Figures 5.6 e-f. As in the case of Figures 5.6 c-d,
the positions of the estimated pseudo depths are in good agreement with the
before mentioned features.
37 QNR
37 LNR
25 SUM
20 30 40 50 60 70 80
360 46/ 143

60/ 154
270 74/ 165
49/ 110
180 18/ 44
90 20/ 55
49/ 143
0
20 30 40 50 60 70 80
Figure 5.7: Bandpass filtered P receiver functions (7 s to 60 s). (a) Shown are
the stacked traces for all stations (SUM, black), beamformed traces normalized to
rms amplitude of noise on L (LNR, red) and beamformed traces normalized to rms
amplitude of noise on Q (QNR, blue). The numbers in front of the trace names gives
the number of events which contributed to the stacked trace. The marker gives the
position of the signal corresponding to the Lehmann discontinuity (220), the ’410’,
the ’660’ and the probable signal of the ’520’ estimated with one standard deviation
by picking 300 bootstrapped stacked traces. (b) Traces, contributing to the stacks in
a, stacked in 90¶ back-azimuth bins every 45¶ . Dashed lines show the estimated mean
delay time of the corresponding phase. The numbers at the end of each azimuthal
stack give the number of single recordings used for the single stations’ stack and the
beamformed traces’ stack.
The amplitudes of the peaks associated with the deeper discontinuities

of the lower part of the upper mantle are several ten times smaller than the
direct P phase (compare ṔṔ ṔŚ410 and ṔṔ ṔŚ660 in Figure 5.2). The usual approach
410 660
to enhance them is stacking. In Figure 5.7 a, we present the bandpass filtered
(7-60 s) stacked RF of the single stations (SUM) and beamformed traces with
different normalizations (LNR and QNR). For the upper mantle, we can clearly
50
5.5. Discussion
identify three positive phases. The first is at ≥22-23 s and is associated with
the Lehmann discontinuity which is usually found at ≥220 km depth. The
second phase is at ≥44-45 s and is linked to the ’410’. The third is at ≥68-69 s
and is associated with the ’660’. Comparing the delay times of these three
phases, we notice a time shift between the single stations’ stack and the stack
of the beamformed traces (≥1 s for Lehmann and ≥1.5 s for ’410’ and ’660’).
We calculate azimuthal stacks with 90¶ bins every 45¶ (Figure 5.7 b). The
mean delay times are given as guide lines and reveal a small deviation of the
signal’s position with back-azimuth. Moreover, the Lehmann discontinuity
(220) disappears for the azimuthal stack at 360¶ (315¶ -45¶ ). Furthermore, the
’660’ seems to be split into two signals for back-azimuths between 180¶ and 315¶ .
There is a fourth weak signal visible at ≥56-57 s which might be related to the
’520’. Transforming delay times of the mantle discontinuities to pseudo depth is
usually strongly influenced by the used velocity model of the uppermost mantle
and crust. It is therefore common practice to employ more robust estimates
like the delay time difference between the ’660’ and the ’410’. This difference
is 24.09±1.26 s for the single station stack (SUM), 23.86±2.95 s for the stack
of the beamformed traces normalized to the noise on the L component (LNR)
and 23.87±1.84 s for the stack of the beamformed traces normalized to the
noise on the Q component (QNR).
Table 5.2: Delay times for stacked traces and beamformed traces for 220 (Lehmann),
’410’, ’520’(?) and ’660’ discontinuities. The times are estimated by picking the
corresponding phase on 300 stacked traces which were formed by bootstrapping the
contributing traces.
station t220 (s) t410 (s) t520(?) (s) t660 (s)

SUM 21.92±0.24 43.71±0.78 56.62±0.68 67.80±0.48
LNR 23.01±0.46 45.46±1.80 56.39±1.62 69.32±1.15
QNR 22.81±0.26 45.02±1.26 57.65±2.22 68.92±0.58
5.5 Discussion
Examining the map showing the piercing points (Figure 5.8), we find that the
RFs mostly sample structures within the Eurasian plate (North of the Gloria
fault) for back-azimuths between ≥200¶ to 80¶ . The azimuthal coverage is
similar for the single stations and the beamformed traces. The gap between
≥80¶ and 200¶ is also visible in the CCP profiles (Figure 5.6 c-f) in which the
Southern and the Eastern part of the profiles have hit-counts which are below
20% of the maximum value which indicates that the results obtained there
have to be handled with more care. Additionally, we present in Figure 5.8 the
azimuthal distribution of the used recordings (inset in Figure 5.8 b). For a
better comparison between single stations and beamformed traces, we give the
fraction of the total number of recordings in percent for 45¶ bins. This azimuthal
distribution shows that the single stations have a slightly higher fraction of
recordings for back-azimuths between 225¶ and 315¶ , whereas the beamformed
traces have a slightly higher fraction of recordings for back-azimuths between
315¶ and 45¶ .
The Moho has an average pseudo depth of ≥6.7 km for the single stations,
as well as 6-6.5 km for the stack of the single stations and the beamformed
51
40 km 140
0 125
120
220 km 410 km
520 km 660 km
36
100
20 16
40 km
0 125 80
60
10%
20%
36
20 16
Figure 5.8: Location of piercing points of upper mantle discontinuities for (a)
single stations and (b) beamformed traces. The azimuthal distribution of the used
recordings (fraction of the total number of recordings in percent for 45¶ bins, black -
single stations, red - LNR, blue - QNR) is given in the inset. The age of the oceanic
lithosphere in million years (Müller et al., 2008) is shown by the color shading. The
piercing points are shown for depth of 220 km (yellow circles, Lehmann discontinuity),
410 km (orange circles), 520 km (black crosses) and 660 km (red circles). The location
of the used OBS are given as black triangles. The position of the Eurasian-African
plate boundary (Gloria Fault, Bird, 2003) is presented as black dashed line.
traces. This crustal thickness corresponds well to the expected values for
oceanic crust (e.g White et al., 1992; Laske et al., 2013). Some single station
estimates are larger than 9 km (D01 and D10) and imply either an apparent
thickening of the crust due to a serpentinization of the uppermost mantle in
the vicinity of the Gloria fault (e.g White et al., 1992) or a sedimentary cover
which might be present at these stations. A change in the thickness of the
sedimentary cover is in good agreement with analog seismic recording of this
area from 1969 (NOAA World Data Service for Geophysics, Marine Seismic
Reflection, Survey ID V2707, ngdc.noaa.gov/mgg/seismicreflection) and the
bathymetry (Figures 5.1 b and 5.6 b) which both indicate an undulation in the
basement topography which leads to the formation of sediment filled troughs
(Hannemann et al., 2016).
The N-S and W-E profiles of the CCP stack of the 0.5-60 s band-passed
data as well as the estimated pseudo depths of the Moho (Figs 5.6 c-d) gives
indications about the depth distribution of the Moho which is rather similar
(6-10 km) for the N-S profile (Figure 5.6 c) and the W-E profile (Figure 5.6 d).
The age differences along both profiles are rather small (5-10 Ma, Figure 5.8,
Müller et al., 2008), therefore the nearly constant Moho depth is reasonable.
In contrast to the observations made by Olugboji et al. (2016), we observe
a double-polarity signal following the Moho signal (LAB and MAD) which
52
5.5. Discussion
indicates in our case the presence of partial melt in the upper part of the
asthenosphere. The already mentioned small amplitudes of the LAB and MAD
signals can be explained by either a rather small amount of melt (few percent)
or a gradient in both boundaries (Olugboji et al., 2013). For ages between
75 Ma and 85 Ma as in this study (Figure 5.8, given by Müller et al., 2008),
the expected lithospheric thickness given by Olugboji et al. (2013) is 70 km
(independent of age and thermal model). The estimated LAB pseudo depths for
the DOCTAR experiment are all shallower than this (≥40-60 km). A possible
explanation could be a thermal anomaly caused by the neighboring fault. A
shallower LAB would indicate an elevated temperature compared to a “normal”
oceanic lithosphere. The broad and small signal which is visible in the longer
periods band (4-40 s and 6-40 s, Figure 5.5), but nearly invisible for the shorter
periods (0.5-60 s, Figure 5.4), is an indication for a gradual velocity change.
This gradual change would imply a more thermal controlled LAB (Karato, 2012;
Olugboji et al., 2013) which would additionally favor the elevated temperature
explanation. The CCP stack (blue colors in Figures 5.6 e-f) shows that the LAB
is found at depths between 40-50 km and does not vary significantly across
the array. As already mentioned, the occurrence of a clear positive signal after
the LAB signal was suggested by (Olugboji et al., 2013) as an evidence for
the presence of partial melt in the asthenosphere. The rather small and broad
peak in the RF is as in the case of the LAB an indication for a more gradual
discontinuity. The positive amplitude (red colors) feature is also visible in the
CCP stack (Figs 5.6 e-f) at depth between 90-110 km.
The delay times of the Lehmann discontinuity (≥22-23 s) are in-line with a
depth of ≥220 km as suggested by PREM (Dziewonski and Anderson, 1981).
According to Karato (1992) and Deuss and Woodhouse (2004), the discontinuity
marks the transition from anisotropic media above to isotropic media below
and most likely coincides with the lower boundary of the asthenosphere. Nev-
ertheless, the estimated delay times are slightly smaller than PREM estimates
(23.49 s). This might be an indication of a wet mantle, a smaller grain size or
a lower stress level (Karato, 1992; Deuss and Woodhouse, 2004). The first two
factors are more likely than the last because below the location of the array
in the vicinity of a major transform fault (Gloria fault) higher stress levels
(Bürgmann and Dresen, 2008) are likely. Moreover, these higher stress levels
would favor smaller grain sizes (Bürgmann and Dresen, 2008).
The expected delay time for the ’410’ assuming PREM velocities is 43.67 s
which agrees quite good with the measured delay time of the single station
stacks and the ≥1.5 s delayed phases on the beamformed traces. The results of
the single stations’ stack indicate a normal depth of the ’410’ which is in good
agreement with observations made by Saki et al. (2015) in their precursor study.
If we calculate the expected delay time for the ’660’ by assuming PREM, we
get 67.98 s. The measured delay times on the single stations’ stack is similar
and the estimates for the stack of the beamformed traces are as in the case
of the ’410’ shifted by ≥1.5 s. The slightly later arrival of the ’660’ at the
beamformed traces compared to PREM and the ’apparent splitting’ of the
’660’ but not of the ’410’ for back-azimuths between 180¶ and 315¶ (Figure 5.7)
might be related to an additional transitions in the Aluminium phase (majorite
garnet to perovskite Simmons and Gurrola, 2000; Deuss et al., 2013). This
hypothesis is in agreement with Saki et al. (2015) who could hardly observe
the ’660’ with their precursors and explained this by the majorite-perovskite
53
transition. Despite this, the delay time difference between the ’410’ and the
’660’ is similar to the theoretical estimate using PREM (24.31 s).
We have examined the corresponding T components of the RFs to explore
the reason for the variation of the delay times of the Lehmann discontinuity,
the ’410’ and the ’660’ (Figure 5.S2 in appendix). There is no clear signal
observed at the determined delay times of these discontinuities which excludes
the possibility the presence of strong or thick anisotropic layers beneath the
array. However, the SNR of the T component RFs is quite high and minor
contributions are not resolvable. We investigate whether the observed time shift
is related to the different processing of the single stations and the beamformed
traces by calculating synthetic RFs using a full wave field reflectivity method
(QSEIS, Wang, 1999) with our station distribution, a global velocity model
(PREM Dziewonski and Anderson, 1981) and a local crustal model (CRUST1.0
Laske et al., 2013). We find no difference in the estimated delay times of the
mantle discontinuities for the beam formed traces’ stack and the stack of the
single stations. The time shift in the delay times between the single stations’
stack and the beamformed traces’ stack (Fig. 5.7) therefore probably indicates
a difference in the velocities or the the thicknesses of the mantle above the
Lehmann discontinuity sampled by the according traces. Moreover, the by 0.5 s
increased delay of the ’410’ and the ’660’ compared to the Lehmann discontinuity
shows that part of the velocity perturbations or thickness variations might
be located below the Lehmann discontinuity but still above the ’410’. The
similar behavior of the ’410’ and the ’660’ is another indication for this latter
hypothesis. Furthermore, the delay times of the MAD RF phase show no
time difference between the single stations’ stack and the beamformed traces’
stack which indicates that the reason for this difference might be located in
the lower asthenosphere. The smaller delay times of the single stations’ stack
compared to the beamformed traces’ stack can indicate either a higher S wave
velocity, a smaller P wave velocity or a smaller thickness of the involved layers.
According to the observations made for the azimuthal distributions (inset in
Figure 5.8 b), the smaller delay times for the single stations’ stacks might
be related to structures West of the array and the larger delay times for the
beam formed traces might be associated with structures North of the array.
Additional information like velocity or thickness estimates from e.g. surface
wave measurements would be required for a further in depth analysis of the
different delay times. This is beyond the scope of this study.
The amplitude of the suspected ’520’ is rather small and therefore it was
hard to estimate its delay time even for the beamformed traces. Flanagan
and Shearer (1998) stated that the ’520’ has a rather small amplitude for SS
precursors and therefore its visibility increases with the number of traces in
the stack. The same is true for RFs (Chevrot et al., 1999). At OBS in general
and in this study, low SNR and few event recordings are common (Webb,
1998), in combination with low amplitudes this results in the ’520’ to be hardly
detectable. The shift between the single stations’ stack and the stack of the
beamformed traces which was observed for the Lehmann discontinuity, the
’410’ and the ’660’ is not clearly observed for the ’520’(?). Furthermore, we
notice that for the azimuthal stacked T components shown in the appendix
(Figure 5.S2) we observe a phase at similar time as as the delay times estimated
for the ’520’(?) on the Q component. This might indicate a dipping of this
interface (Cassidy, 1992) or an out-of-plane arrival. The number of events
54
5.6. Conclusion
and the azimuthal coverage of the data is too limited in our study to further
investigate this arrival.
Overall, we are able to identify the Moho, the LAB, a MAD, the Lehmann
discontinuity, as well as the ’410’ and the ’660’ in our RFs. The beamforming
proved to be an excellent way to increase the SNR of the event recordings and
the number of usable events for the analysis. We could use the single station
recordings to analyze the behavior of the Moho, the LAB and a MAD across
the array. No large changes in the depth of these discontinuities are identified
for the study area. For the mantle discontinuities, we use the stack of all
stations and the beamformed traces to estimate delay times and we compare
them with theoretical delay times calculated with PREM velocities. We find
good agreement for the Lehmann discontinuity, the ’410’ and the ’660’ for the
single stations’ stack and estimated slightly deeper ’660’ for the beamformed
traces. The results for the ’410’ and the ’660’ are in good agreement with Saki
et al. (2015).
5.6 Conclusion
This study shows that we can identify discontinuities in the oceanic crust and
upper mantle down to the MTZ using OBS data. Furthermore, it explores the
advantages of using beamforming to improve the signal quality of the RFs and
a quality control employing evaluation criteria like relative spike position and
SNR to search for the optimal deconvolution length. These techniques prove
to work well in this study.
The first analyzed discontinuity is the Moho for which the pseudo depth
across the array is ≥6.7 km. This nearly constant crustal thickness is observed
from the individual pseudo depths and the CCP profiles for the whole study
area. Nevertheless, the estimated thicknesses are slightly different which might
be influenced by a changing sedimentary cover in agreement with seismic profiles
of this region.
Secondly, the upper boundary of the asthenosphere, as well as a discontinuity
within this LVL are identified as a small peak-to-peak amplitude (14-36% of
direct P phase) double-polarity signal which has rather broad peaks. These
features indicate either the presence of few percents of partial melt in the upper
asthenosphere or gradual interfaces (Olugboji et al., 2013). The estimated
LAB pseudo depths at 44-55 km is elevated compared to the modeled LAB
depth of Olugboji et al. (2013) for normal oceanic crust. We suggest that this
is originating from a thermal anomaly likely related to the neighboring Gloria
fault.
In agreement with continental settings, the lower boundary of the astheno-
sphere is identical to the Lehmann discontinuity which is found at ≥22-23 s in
the RFs. The signal arrives slightly earlier than would be expected by assuming
PREM velocities which might indicate a wet uppermost mantle. Nevertheless,
it is still in agreement with other observations and theoretical modeling of the
Lehmann discontinuity (e.g. Deuss et al., 2013).
We find that the ’410’ is located at normal depth as was also observed by
Saki et al. (2015) using precursors in the same study area. Moreover, the slightly
deeper ’660’ for the beamformed traces and the apparent ’splitting’ of this
discontinuity for back-azimuths between 180¶ and 315¶ matches the hypothesis
55
Bibliography
made by Saki et al. (2015) that an additional transition from majorite garnet
to perovskite might be present in this region. Nevertheless, the MTZ delay
time differences are in-line with the values expected from PREM. Furthermore,
we also observe a weak positive signal in-between the ’410’ and ’660’ which
occurs at delay times which are expected for the ’520’. In accordance with
global observations of this phase (Flanagan and Shearer, 1998), it has a rather
small amplitude and is hard to detect on the rather noisy RFs of the OBSs.
Furthermore, a comparison between Q and T component RFs might indicate a
dipping of the ’520’(?) (Cassidy, 1992).
Despite the ’520’, all other observed signals of the mantle discontinuities
(Lehmann, ’410’, ’660’) show a delay between the signal observed at the
single stations’ stack and the beamformed traces’ stack. This might be an
indication for differently sampled velocity perturbations or thickness variations
in the asthenosphere above the Lehmann discontinuity. Moreover, the ≥0.5 s
smaller delay of the Lehmann discontinuity compared to the ’410’ and ’660’
indicates that part of these changes are most likely located below the Lehmann
discontinuity and above the ’410’.
In conclusion, this study shows that the number of usable events for RF
studies at the ocean bottom can be increased by ≥50% using beamforming
techniques at a mid aperture array. The application of evaluation criteria sup-
ports the selection of optimal deconvolution time window lengths. Nevertheless,
a manual revision of the RFs resulting from the pre-selected deconvolution
lengths is still necessary to exclude RFs from the analysis which are influenced
by high frequency noise. Furthermore, this study proves that the combination
of single OBSs and beamforming techniques gives the opportunity to inves-
tigate structures from the sea-floor down to the MTZ. Moreover, the single
stations offer the possibility to study the lateral variations across the array
for the shallower discontinuities (Moho, LAB, MAD) whereas the combination
of all stations (including beamforming) gives insight into the deeper mantle
discontinuities (Lehmann, ’410’, ’520’(?), ’660’).
Acknowledgments
The authors thank the DEPAS pool for providing the instruments for the
DOCTAR project which was funded by the DFG (KR1935/13, DA 478/21-1)
and by the Leitstelle für Mittelgroße Forschungsschiffe (Poseidon cruises 416
and 431). The first author thanks Brigitte Knapmeyer-Endrun for providing
her scripts for the CCP and answering related questions. The authors thank
EMEPC (Task Group for the Extension of the Continental Shelf) for providing
the bathymetric data and Luis Batista for sharing them. The data processing
was partly done using Seismic Handler (Stammler, 1993). Some figures were
created using GMT (Wessel and Smith, 1991; Wessel et al., 2013).
Bibliography
Agee, C. B. (1998). Phase transformations and seismic structure in the upper
mantle and transition zone. In Hemley, R. J. and Ribbe, P. H., editors,
Ultrahigh-Pressure Mineralogy. Physics and Chemistry of the Earth’s Deep
56
Bibliography
Interior, volume 37, chapter 5, pages 165–203. Mineralogical Society of

America, Washington, D.C.
Aki, K. and Richards, P. G. (2002). Quantitative seismology. University Science

Books, 2nd edition.
Altenbernd, T., Jokat, W., Heyde, I., and Damm, V. (2014). A crustal model
for northern Melville Bay, Baffin Bay. Journal of Geophysical Research B:
Solid Earth, 119(12):8610–8632.
Audet, P. (2016). Receiver functions using OBS data: promises and limita-
tions from numerical modelling and examples from the Cascadia Initiative.
Barruol, G. and Sigloch, K. (2013). Investigating La Réunion hot spot from crust
to core. Eos, Transactions American Geophysical Union, 94(23):205–207.
Bell, S. W., Ruan, Y., and Forsyth, D. W. (2015). Shear Velocity Structure
of Abyssal Plain Sediments in Cascadia. Seismological Research Letters,
86(5):1247–1252.
Berkhout, A. J. (1977). Least-squares inverse filtering and wavelet deconvolution.

Geophysics, 42(7):1369–1383.

Bürgmann, R. and Dresen, G. (2008). Rheology of the Lower Crust and Upper
Mantle: Evidence from Rock Mechanics, Geodesy, and Field Observations.
Cassidy, J. J. F. (1992). Numerical experiments in broadband receiver function

analysis. Bulletin of the Seismological Society of America, 82(3):1453–1474.

Research, 103(B5):9895–9916.
Czuba, W., Grad, M., Mjelde, R., Guterch, A., Libak, A., Krüger, F., Murai,
Y., and Schweitzer, J. (2011). Continent-ocean-transition across a trans-
tensional margin segment: Off Bear Island, Barents Sea. Geophysical Journal
Dahm, T., Tilmann, F., and Morgan, J. P. (2006). Seismic broadband ocean-
bottom data and noise observed with free-fall stations: Experiences from
long-term deployments in the North Atlantic and the Tyrrhenian Sea. Bulletin
of the Seismological Society of America, 96(2):647–664.
Davy, C., Barruol, G., Fontaine, F. R., Sigloch, K., and Stutzmann, E. (2014).
Tracking major storms from microseismic and hydroacoustic observations on
the seafloor. Geophysical Research Letters, 41(24):8825–8831.
57
Bibliography
Deuss, A. and Woodhouse, J. H. (2002). A systematic search for mantle

discontinuities using SS-precursors. Geophysical Research Letters, 29(8):90–
94.
Deuss, A. and Woodhouse, J. H. (2004). The nature of the Lehmann disconti-

nuity from its seismological Clapeyron slopes. Earth and Planetary Science
Letters, 225(3-4):295–304.

Efron, B. and Tibshirani, R. (1986). Bootstrap methods for standard error,

confidence intervals, and other measures of statistical accuracy. Statistical
Science, 1(1):54–75.
Fischer, K. M., Ford, H. A., Abt, D. L., and Rychert, C. A. (2010). The
Lithosphere-Asthenosphere Boundary. Annual Review of Earth and Planetary
Sciences, 38(1):551–575.

Friederich, W. and Meier, T. (2008). Temporary Seismic Broadband Network

Acquired Data on Hellenic Subduction Zone. Eos, Transactions American
Gao, H. and Schwartz, S. (2015). Preface to the Focus Section on Cascadia

Initiative Preliminary Results. Seismological Research Letters, 86(5):1235–
1237.
Gilbert, H. J., Sheehan, A. F., Wiens, D. A., Dueker, K. G., Dorman, L. M.,
Hildebrand, J., and Webb, S. (2001). Upper mantle discontinuity structure
in the region of the Tonga Subduction Zone. Geophysical Research Letters,
28(9):1855–1858.
Gossler, J. and Kind, R. (1996). Seismic evidence for very deep roots of
continents. Earth and Planetary Science Letters, 138(1-4):1–13.
Grad, M., Mjelde, R., Czuba, W., Guterch, A., and the IPY Project Group
(2012). Elastic properties of seafloor sediments from the modelling of ampli-
tudes of multiple water waves recorded on the seafloor off Bear Island, North
Atlantic. Geophysical Prospecting, 60(5):855–869.
Grevemeyer, I., Gràcia, E., Villaseñor, A., Leuchters, W., and Watts, A. B.
(2015). Seismicity and active tectonics in the Alboran Sea, Western Mediter-
ranean: Constraints from an offshore-onshore seismological network and
swath bathymetry data. Journal of Geophysical Research: Solid Earth,
120(12):8348–8365.
58
Bibliography
Grevemeyer, I., Reston, T. J., and Moeller, S. (2013). Microseismicity of

the Mid-Atlantic Ridge at 7°S-8°15’S and at the Logatchev Massif oceanic
core complex at 14°40’N-14°50’N. Geochemistry, Geophysics, Geosystems,
14(9):3532–3554.
Gu, Y. J. and Dziewonski, A. M. (2002). Global variability of transition zone

thickness. Journal of Geophysical Research, 107(B7):2135.
Gu, Y. J., Dziewonski, A. M., and Agee, C. B. (1998). Global de-correlation

of transition zone discontinuities. Earth and Planetary Science Letters,
157(1-2):57–67.
Hannemann, K., Krüger, F., Dahm, T., and Lange, D. (2016). Oceanic
lithospheric S wave velocities from the analysis of P wave polarization at the
ocean floor. Geophysical Journal International, in press.
Helffrich, G. (2000). Topography of the transition zone seismic discontinuities.

Reviews of Geophysics, 38(1):141–158.
Hermann, T. and Jokat, W. (2013). Crustal structures of the boreas basin

and the knipovich ridge,north atlantic. Geophysical Journal International,
193(3):1399–1414.
Janiszewski, H. A. and Abers, G. A. (2015). Imaging the Plate Interface in

the Cascadia Seismogenic Zone: New Constraints from Offshore Receiver
Functions. Seismological Research Letters, 86(5):1261–1269.
Jokat, W., Kollofrath, J., Geissler, W. H., and Jensen, L. (2012). Crustal
thickness and earthquake distribution south of the Logachev Seamount,
Knipovich Ridge. Geophysical Research Letters, 39(8):2–7.
Kalberg, T. and Gohl, K. (2014). The crustal structure and tectonic devel-
opment of the continental margin of the Amundsen sea embayment, West
Antarctica: Implications from geophysical data. Geophysical Journal Inter-
national, 198(1):327–341.

Letters, 19(22):2255–2258.
Karato, S.-i. and Jung, H. (1998). Water, partial melting and the origin of the
seismic low velocity and high attenuation zone in the upper mantle. Earth
59
Bibliography
Kawakatsu, H., Kumar, P., Takei, Y., Shinohara, M., Kanazawa, T., Araki, E.,
and Suyehiro, K. (2009). Seismic evidence for sharp lithosphere-asthenosphere
boundaries of oceanic plates. Science, 324(5926):499–502.
Kieling, K., Rössler, D., and Krüger, F. (2011). Receiver function study in
northern Sumatra. Journal of Seismology, 15(2):235–259.
Kind, R., Kosarev, G. L., and Petersen, N. V. (1995). Receiver functions at
the stations of the German Regional Seismic Network (GRSN). Geophysical
Knapmeyer-Endrun, B., Krüger, F., and Group, P. W. (2014). Moho depth
across the Trans-European Suture Zone from P- and S-receiver functions.
Kopp, H., Weinzierl, W., Becel, A., Charvis, P., Evain, M., Flueh, E. R., Gailler,
A., Galve, A., Hirn, A., Kandilarov, A., Klaeschen, D., Laigle, M., Papenberg,
C., Planert, L., and Roux, E. (2011). Deep structure of the central Lesser
Antilles Island Arc: Relevance for the formation of continental crust. Earth
Kumar, P. and Kawakatsu, H. (2011). Imaging the seismic lithosphere-
asthenosphere boundary of the oceanic plate. Geochemistry, Geophysics,
Geosystems, 12(1):Q01006.
Laigle, M., Hirn, A., Sapin, M., Bécel, A., Charvis, P., Flueh, E., Diaz, J.,
Lebrun, J. F., Gesret, A., Raffaele, R., Galvé, A., Evain, M., Ruiz, M., Kopp,
H., Bayrakci, G., Weinzierl, W., Hello, Y., Lépine, J. C., Viodé, J. P., Sach-
pazi, M., Gallart, J., Kissling, E., and Nicolich, R. (2013). Seismic structure
and activity of the north-central Lesser Antilles subduction zone from an
integrated approach: Similarities with the Tohoku forearc. Tectonophysics,
603:1–20.
Langston, C. A. (1979). Structure under Mount Rainier, Washington, inferred
from teleseismic body waves. Journal of Geophysical Research, 84(B9):4749.
Laske, G., Masters, G., Ma, Z., and Pasyanos, M. E. (2013). CRUST1.0 : An
Updated Global Model of Earth ’ s Crust. In Geophysical Research Abstracts,
volume 15, pages Abstract EGU2013–2658.
Lawrence, J. F. and Shearer, P. M. (2006). A global study of transition
zone thickness using receiver functions. Journal of Geophysical Research,
111(6):B06307.
Li, X., Sobolev, S. V., Kind, R., Yuan, X., and Estabrook, C. (2000). A detailed
receiver function image of the upper mantle discontinuities in the Japan
subduction zone. Earth and Planetary Science Letters, 183(3-4):527–541.
Libak, A., Mjelde, R., Keers, H., Faleide, J. I., and Murai, Y. (2012). An inte-
grated geophysical study of Vestbakken Volcanic Province, western Barents
Sea continental margin, and adjacent oceanic crust. Marine Geophysical
Research, 33(2):185–207.
60
Bibliography
McNamara, D. E. (2004). Ambient Noise Levels in the Continental United

States. Bulletin of the Seismological Society of America, 94(4):1517–1527.
Monna, S., Cimini, G. B., Montuori, C., Matias, L., Geissler, W. H., and Favali,
P. (2013). New insights from seismic tomography on the complex geodynamic
evolution of two adjacent domains: Gulf of Cadiz and Alboran Sea. Journal
of Geophysical Research: Solid Earth, 118(4):1587–1601.
Müller, R. D., Sdrolias, M., Gaina, C., and Roest, W. R. (2008). Age, spreading
rates, and spreading asymmetry of the world’s ocean crust. Geochemistry,
Geophysics, Geosystems, 9(4):Q04006.
Olugboji, T. M., Park, J., ichiro Karato, S., and Shinohara, M. (2016). Nature
of the seismic lithosphere-asthenosphere boundary within normal oceanic
mantle from high-resolution receiver functions. Geochemistry, Geophysics,
Geosystems, 17(4):1265–1282.
Pho, H.-T. and Behe, L. (1972). Extended distances and angles of incidence of
P waves. Bulletin of the Seismological Society of America, 62(4):885–902.
Romanowicz, B. (2009). The Thickness of Tectonic Plates. Science,

324(5926):474–476.

Ruiz, M., Galve, A., Monfret, T., Sapin, M., Charvis, P., Laigle, M., Evain, M.,
Hirn, A., Flueh, E., Gallart, J., Diaz, J., and Lebrun, J. F. (2013). Seismic
activity offshore Martinique and Dominica islands (Central Lesser Antilles
subduction zone) from temporary onshore and offshore seismic networks.
Tectonophysics, 603(April 2007):68–78.
Saki, M., Thomas, C., Nippress, S. E., and Lessing, S. (2015). Topography
409:193–202.
Scherbaum, F. (2007). Of poles and zeros. Fundamentals of Digital Seismology,

volume 15. Springer, Dordrecht, Netherlands, 2nd edition.
Schlindwein, V., Demuth, A., Geissler, W. H., and Jokat, W. (2013). Seismic
gap beneath Logachev Seamount: Indicator for melt focusing at an ultraslow
mid-ocean ridge? Geophysical Research Letters, 40(9):1703–1707.
Schlindwein, V., Demuth, A., Korger, E., L??derach, C., and Schmid, F. (2015).
Seismicity of the Arctic mid-ocean Ridge system. Polar Science, 9(1):146–157.
Sheehan, A. F., Abers, G. a., Jones, C. H., and Lerner-Lam, A. L. (1995).

Crustal thickness variations across the Colorado Rocky Mountains from tele-
seismic receiver functions. Journal of Geophysical Research, 100(B10):20391–
20404.
61
Bibliography
Shen, Y., Sheehan, A. F., Dueker, K. G., de Groot–Hedlin, C., and Gilbert, H.
(1998a). Mantle Discontinuity Structure Beneath the Southern East Pacific
Rise from P-to-S Converted Phases. Science, 280(5367):1232–1235.
Shen, Y., Solomon, S. C., Bjarnason, I. T., and Wolfe, C. J. (1998b). Seismic
evidence for a lower-mantle origin of the Iceland plume. Nature, 395(6697):62–
65.
Silveira, G., Vinnik, L., Stutzmann, E., Farra, V., Kiselev, S., and Morais, I.
(2010). Stratification of the Earth beneath the Azores from P and S receiver
functions. Earth and Planetary Science Letters, 299(1-2):91–103.

562.
83(4):704–713.
41:43–63.

Suckro, S. K., Gohl, K., Funck, T., Heyde, I., Ehrhardt, A., Schreckenberger,
B., Gerlings, J., Damm, V., and Jokat, W. (2012). The crustal structure
of southern Baffin Bay: Implications from a seismic refraction experiment.
Suetsugu, D., Inoue, T., Obayashi, M., Yamada, A., Shiobara, H., Sugioka,
H., Ito, A., Kanazawa, T., Kawakatsu, H., Shito, A., Fukao, Y., Suetsugu,
D., Bina, C., Inoue, T., Wiens, D., and Jellinek, M. (2010). Depths of the
410-km and 660-km discontinuities in and around the stagnant slab beneath
the Philippine Sea: Is water stored in the stagnant slab? Physics of the
Earth and Planetary Interiors, 183(1-2):270–279.
Suetsugu, D., Shinohara, M., Araki, E., Kanazawa, T., Suyehiro, K., Yamada,
T., Nakahigashi, K., Shiobara, H., Sugioka, H., Kawai, K., and Fukao,
Y. (2005). Mantle discontinuity depths beneath the west Philippine basin
from receiver function analysis of deep-sea borehole and seafloor broadband
waveforms. Bulletin of the Seismological Society of America, 95(5):1947–1956.
Suetsugu, D., Shiobara, H., Sugioka, H., Fukao, Y., and Kanazawa, T. (2007).
Topography of the mantle discontinuities beneath the South Pacific superswell
as inferred from broadband waveforms on seafloor. Physics of the Earth and
Planetary Interiors, 160(3-5):310–318.
62
Bibliography
Sumy, D. F., Lodewyk, J. A., Woodward, R. L., and Evers, B. (2015). Ocean-
Bottom Seismograph Performance during the Cascadia Initiative. Seismolog-
ical Research Letters, 86(5):1238–1246.
Tauzin, B., Debayle, E., and Wittlinger, G. (2008). The mantle transition zone
as seen by global Pds phases: No clear evidence for a thin transition zone
beneath hotspots. Journal of Geophysical Research, 113(B8):B08309.
Thomas, C. and Laske, G. (2015). D” observations in the Pacific from PLUME

ocean bottom seismometer recordings. Geophysical Journal International,
200(2):849–860.
Thorwart, M. (2006). Confirming Wavefield methods to analyze passive ocean

bottom seismic data. Application to the Tyrrhenian Sea. PhD thesis, Univer-
sität Hamburg.
Thorwart, M. and Dahm, T. (2005). Wavefield decomposition for passive ocean

bottom seismological data. Geophysical Journal International, 163(2):611–
621.
Tilmann, F. J. and Dahm, T. (2008). Constraints on crustal and mantle

structure of the oceanic plate south of Iceland from ocean bottom recorded
Rayleigh waves. Tectonophysics, 447(1-4):66–79.
Vinnik, L. P. (1977). Detection of waves converted from P to SV in the mantle.

Physics of the Earth and Planetary Interiors, 15(1):39–45.
Wang, R. (1999). A simple orthonormalization method for stable and efficient

computation of Green’s functions. Bulletin of the Seismological Society of
America, 89(3):733–741.
Webb, S. C. (1998). Broadband seismology and noise under the ocean. Reviews
Wessel, P. and Smith, W. H. F. (1991). Free software helps map and display
data. Eos, Transactions American Geophysical Union, 72(41):441.
Wessel, P., Smith, W. H. F., Scharroo, R., Luis, J., and Wobbe, F. (2013).
Generic Mapping Tools: Inproved version released. Eos, Transactions Amer-
ican Geophysical Union, 94(45):409–410.

Yuan, X., Ni, J., Kind, R., Mechie, J., and Sandvol, E. (1997). Lithospheric
and upper mantle structure of southern Tibet from a seismological passive
source experiment. Journal of Geophysical Research, 102(B12):27491–27500.
63
Bibliography
5.S1 Orientation
For the analysis of the P-phase polarization, we measure the amplitudes of
several teleseismic P-phases on all three components for each station. We
estimate the theoretical amplitude distribution of the P-phase on the horizontal
components by using the vertical P-phase polarization and the known back-
azimuth of the earthquake. After a stepwise rotation of the theoretical amplitude
distribution, we calculate the difference (misfit) between the theoretical and
the measured horizontal amplitudes.
We estimate this misfit for several events and calculate the mean and
standard deviation. We use the definitions of mean µ and standard deviation ‡
from directional statistics analogue to Grigoli et al. (2012) for N measurements
of the orientation angle Ïi with weight wi which is chosen based on event
quality.
1 2 Ò
µ = arctan Q
P
· ‡ = 2 · (1 ≠ R) (5.S1)
q qN
with P = N i=1 wi cos Ïi · Q = i=1 wi sin Ïi
1
Ô 2
and R = qN P +Q 2
w
i=1 i
Furthermore, we combine the misfit functions of all events by calculating

the mean of the different misfit functions for each tested angle.
Table 5.S1: Results of orientation of OBS using the P-Phase. We give the results of
the analysis of single events, the combined misfit function and the bootstrap. The
mean and standard deviation are calculated using equation (5.S1).
station # of events single events all events bootstrap

D01 9 144.8¶ ±27.6¶ 144¶ 144.2¶ ± 5.3¶
D02 8 195.5¶ ±17.5¶ 198¶ 198.5¶ ± 3.5¶
D03 13 177.2¶ ±34.2¶ 175¶ 175.5¶ ± 2.3¶
D04 7 94.1¶ ±28.9¶ 98¶ 99.5¶ ± 7.3¶
D06 15 56.1¶ ±49.0¶ 57¶ 57.8¶ ± 5.2¶
D07 14 203.3¶ ±32.8¶ 202¶ 202.1¶ ± 2.8¶
D08 7 239.4¶ ±31.1¶ 228¶ 233.9¶ ±10.6¶
D09 12 142.4¶ ±32.8¶ 139¶ 141.7¶ ± 7.1¶
D10 11 271.1¶ ±51.3¶ 261¶ 262.6¶ ± 5.4¶
D11 11 349.6¶ ±34.2¶ 351¶ 351.4¶ ± 2.9¶
D12 11 289.7¶ ±53.9¶ 300¶ 298.3¶ ± 7.1¶
We also use surface waves to estimate the orientation of the stations (Stach-
nik et al., 2012). The data are filtered with a bandpass between 20 and 60 s and
the horizontal components are rotated using the back-azimuth of the events.
If the horizontal components are properly oriented, the vertical trace will be
identical to the Hilbert transform of the radial trace within the time window
of the Rayleigh phase (Stachnik et al., 2012). We decide to include the Love
phase in our analysis, because its energy should completely vanish from the
radial component if the components have the correct orientation. We use a
normalized zero-lag cross correlation Srz between the Hilbert transform of the
radial trace (R̃) and the vertical trace (Z) (equation (5.S2), Stachnik et al.,
2012; Zha et al., 2013).
64
5.S1. Orientation
fl(R̃,Z )
Srz = fl(Z,Z)
(5.S2)
s
with fl (X, Y ) = tt12 X(t)Y (t)dt
1 2
Herein, fl R̃, Z is the zero-lag cross-correlation between the Hilbert trans-
form of the radial trace and the vertical trace and fl (Z, Z) is the zero-lag
auto-correlation of the vertical trace. Before calculating Srz , we normalize
the traces. The horizontal traces are equally treated to preserve the particle
polarization. Afterwards, we rotate the horizontal traces in one degree steps
and calculate Srz . As for the P-phase, we estimate Srz for several events
and calculated the mean and standard deviation according to equation (5.S1).
Furthermore, we append all event data and processed them together.
orientation
38.8
D11
error
D07
D06
D04
D10 D01 38.4

D12
D03 D02
D08 D05
Pphase
D09 Rayleighphase
38
km
0 10 20 30
18.8 18.4 18
Figure 5.S1: Estimated orientation of OBS by P phase polarization (red arrows and
slices) and Rayleigh wave ellipticity (black arrows and open slices). The arrows give
the estimated directions and the slices the error of the orientation (see also Tabs 5.S1
and 5.S2).
Moreover, we use the bootstrap method and equation (5.S1) to estimate

mean and standard deviation for the combined misfit function for the P-phase
and the correlation coefficient for the Rayleigh phase. The resulting angles are
presented in Tab. 5.S1 for the P-phase and in table 5.S2 for the Rayleigh phase
and in Figure 5.S1.
65
Bibliography
Table 5.S2: Results of orientation of OBS using the Rayleigh phase. We give the
results of the analysis of single events, the correlation coefficient of the concatenated
data and the bootstrap. The mean and standard deviation are calculated using
equation (5.S1).
station # of events single events all events bootstrap

D01 24 140.3¶ ±13.7¶ 142¶ 140.4¶ ±2.2¶
D02 28 193.7¶ ±13.0¶ 195¶ 193.1¶ ±2.3¶
D03 28 164.9¶ ±12.9¶ 166¶ 164.5¶ ±2.2¶
D04 27 99.1¶ ±17.4¶ 98¶ 96.8¶ ±2.2¶
D06 25 48.5¶ ±25.7¶ 49¶ 48.0¶ ±2.5¶
D07 20 197.9¶ ± 8.9¶ 197¶ 196.9¶ ±1.8¶
D08 18 226.2¶ ±11.1¶ 228¶ 225.8¶ ±2.6¶
D09 21 138.6¶ ± 8.0¶ 137¶ 136.6¶ ±2.2¶
D10 11 252.2¶ ±13.9¶ 254¶ 251.2¶ ±3.0¶
D11 18 349.2¶ ±10.5¶ 349¶ 348.5¶ ±2.9¶
D12 21 299.7¶ ±15.5¶ 300¶ 300.2¶ ±3.1¶
5.S2 T components of RF
37 QNR
37 LNR
25 SUM
20 30 40 50 60 70 80
360 46/ 143

60/ 154
270 74/ 165
49/ 110
180 18/ 44
90 20/ 55
49/ 143
0
20 30 40 50 60 70 80
Figure 5.S2: Same as Figure 5.7 but for T components of P receiver functions
(7-60 s). Presented delay times are estimated on Q components (Figure 5.7).
66
5.S3. Event tables
5.S3 Event tables
Table 5.S3: Events used for the P receiver functions, origin time, hypocenter location
and moment magnitude Mw from NEIC catalogue (earthquake.usgs.gov) and is
the distance between earthquake and array location in degree.
# date time lat lon depth Mw

dd.mm.yyyy hh:mm: ss .ss [¶ ] [¶ ] [km] [¶ ]
1 06. 07 .2011 19: 03 :18.26 -29.54 -176.34 17.0 7.6 159.8
2 10. 07 .2011 00: 57 :10.80 38.03 143.26 23.0 7.0 102.1
3 29. 07 .2011 07: 42 :23.40 -23.80 179.75 532.0 6.7 158.8
4 20. 08 .2011 16: 55 :02.81 -18.37 168.14 32.0 7.2 159.3
5 24. 08 .2011 17: 46 :11.65 -7.64 -74.53 147.0 7.0 69.4
6 30. 08 .2011 06: 57 :41.61 -6.36 126.75 469.8 6.9 135.2
7 02. 09 .2011 10: 55 :53.59 52.17 -171.71 32.0 6.9 86.9
8 02. 09 .2011 13: 47 :09.62 -28.40 -63.03 578.9 6.7 78.5
9 03. 09 .2011 22: 55 :40.92 -20.67 169.72 185.1 7.0 161.0
10 05. 09 .2011 09: 52 :01.13 -15.30 -173.62 37.0 6.2 148.3
11 15. 09 .2011 19: 31 :04.08 -21.61 -179.53 644.6 7.3 156.7
12 18. 09 .2011 12: 40 :51.83 27.73 88.16 50.0 6.9 85.0
13 06. 10 .2011 11: 12 :30.07 -24.18 -64.22 15.0 5.8 75.6
14 14. 10 .2011 03: 35 :14.81 -6.57 147.88 37.0 6.5 146.0
15 21. 10 .2011 17: 57 :16.10 -28.99 -176.24 33.0 7.4 159.4
16 23. 10 .2011 10: 41 :23.25 38.72 43.51 18.0 7.1 47.6
17 28. 10 .2011 18: 54 :34.04 -14.44 -75.97 24.0 6.9 75.2
18 22. 11 .2011 18: 48 :16.30 -15.36 -65.09 549.9 6.6 69.1
19 07. 12 .2011 22: 23 :09.73 -27.90 -70.92 20.0 6.1 82.2
20 11. 12 .2011 01: 47 :25.56 17.99 -99.79 59.0 6.5 72.5
21 14. 12 .2011 05: 04 :58.63 -7.55 146.81 135.0 7.1 146.5
22 27. 12 .2011 15: 21 :56.84 51.84 95.91 15.0 6.6 73.5
23 09. 01 .2012 04: 07 :14.67 -10.62 165.16 28.0 6.4 152.2
24 10. 01 .2012 18: 36 :59.08 2.43 93.21 19.0 7.2 105.2
25 30. 01 .2012 05: 11 :00.95 -14.17 -75.64 43.0 6.4 74.8
26 02. 02 .2012 13: 34 :40.65 -17.83 167.13 23.0 7.1 159.0
27 13. 02 .2012 10: 55 :09.44 9.18 -84.12 16.0 5.9 65.4
28 26. 02 .2012 06: 17 :19.76 51.71 95.99 12.0 6.7 73.6
29 14. 03 .2012 09: 08 :35.14 40.89 144.94 12.0 6.9 99.7
30 20. 03 .2012 18: 02 :47.44 16.49 -98.23 20.0 7.4 72.1
31 21. 03 .2012 22: 15 :06.13 -6.24 145.96 118.0 6.6 145.0
32 25. 03 .2012 22: 37 :06.00 -35.20 -72.22 40.7 7.1 88.6
33 11. 04 .2012 08: 38 :36.72 2.33 93.06 20.0 8.6 105.2
34 11. 04 .2012 22: 55 :10.25 18.23 -102.69 20.0 6.5 74.6
35 12. 04 .2012 07: 15 :48.50 28.70 -113.10 13.0 7.0 76.2
36 17. 04 .2012 03: 50 :15.61 -32.63 -71.37 29.0 6.7 86.1
37 17. 04 .2012 07: 13 :49.00 -5.46 147.12 198.0 6.8 144.7
67
Bibliography
Table 5.S4: Events used at single stations and for P beams. Details of events are
listed in Tab. 5.S3.
# D01 D02 D03 D04 D06 D07 D08 D09 D10 D11 D12 P-B
1 X X X X X X X X X X X
2 X
3 X
4 X X
5 X X X X X X X X X X X X
6 X
7 X X X X X X X X
9 X X X X X X
10 X
11 X X X X
12 X X
13 X
14 X
17 X X X X X X X X X X
18 X X X X X X X
19 X
20 X X X X
21 X X X
22 X X
23 X
24 X
25 X X X X X
26 X X X X X
27 X
29 X X X X
31 X X X
32 X X X X X X X X X
33 X X X X X X X X X
34 X X X X X X X
35 X
36 X
total 9 17 11 15 17 18 14 14 16 15 11 37
68
6. Oceanic lithospheric S wave
velocities from the analysis of P
wave polarization at the ocean
floor
K. Hannemann, F. Krüger, T. Dahm, D. Lange

accepted for publication in Geophysical Journal International
(doi:10.1093/gji/ggw342)
Summary
Our knowledge of the absolute S wave velocities of the oceanic lithosphere
is mainly based on global surface wave tomography, local active seismic or
compliance measurements using oceanic infragravity waves. The results of
tomography give a rather smooth picture of the actual S wave velocity structure
and local measurements have limitations regarding the range of elastic parame-
ters or the geometry of the measurement. Here, we use the P wave polarization
(apparent P wave incidence angle) of teleseismic events to investigate the S
wave velocity structure of the oceanic crust and the upper tens of kilometres
of the mantle beneath single stations. In this study, we present an up to our
knowledge new relation of the apparent P wave incidence angle at the ocean
bottom dependent on the half space S wave velocity. We analyse the angle in
different period ranges at ocean bottom stations (OBS) to derive apparent S
wave velocity profiles. These profiles are dependent on the S wave velocity as
well as on the thickness of the layers in the subsurface. Consequently, their
interpretation results in a set of equally valid models.
We analyse the apparent P wave incidence angles of an OBS data set which
was collected in the eastern mid Atlantic. We are able to determine reasonable
S wave velocity-depth models by a three step quantitative modelling after a
manual data quality control, although layer resonance sometimes influences the
estimated apparent S wave velocities. The apparent S wave velocity profiles are
well explained by an oceanic PREM model in which the upper part is replaced
by four layers consisting of a water column, a sediment, a crust and a layer
representing the uppermost mantle. The obtained sediment has a thickness
between 0.3 km and 0.9 km with S wave velocities between 0.7 km s≠1 and
1.4 km s≠1 . The estimated total crustal thickness varies between 4 km and
10 km with S wave velocities between 3.5 km s≠1 and 4.3 km s≠1 . We find a
slight increase of the total crustal thickness from ≥5 km to ≥8 km towards the
South in the direction of a major plate boundary, the Gloria Fault. The observed
69
Chapter 6. Oceanic lithospheric vs from the analysis of Ïp
crustal thickening can be related with the known dominant compression in

the vicinity of the fault. Furthermore, the resulting mantle S wave velocities
decrease from values around 5.5 km s≠1 to 4.5 km s≠1 towards the fault. This
decrease is probably caused by serpentinization and indicates that the oceanic
transform fault affects a broad region in the uppermost mantle.
Conclusively, the presented method is useful for the estimation of the local
S wave velocity structure beneath ocean bottom seismic stations. It is easy
to implement and consists of two main steps: (1) measurement of apparent P
wave incidence angles in different period ranges for real and synthetic data,
and (2) comparison of the determined apparent S wave velocities for real and
synthetic data to estimate S wave velocity-depth models.
6.1 Introduction
The polarization angle of the particle motion (apparent incidence angle Ïp )
of an incoming P (compressional) wave at the free surface or the solid-liquid
interface is the result of a superposition of the displacements of the incident P
wave and the reflected P wave and S (shear) wave. The measured polarization
of the P wave (i.e. apparent P wave incidence angle Ïp ) therefore differs from
the real incidence angle Ïp1 of the incident P wave (Fig. 6.1). Wiechert (1907)
showed that for the case of the free surface, the apparent P wave incidence
angle Ïp is twice the angle of the reflected SV wave (vertically polarized S
wave, Ïs ). The analytical determination of the apparent P wave incidence
angle Ïp is based on the reflection coefficients at the corresponding interface
(i.e. free surface or ocean bottom). Due to the influence of the water column,
the relation of the apparent P wave incidence angle valid for the ocean bottom
has to differ from the free surface relation given by Wiechert (1907).
The apparent incidence angle can be interpreted in different ways. Mea-
surements of the apparent P wave and S wave incidence angles were used by
Nuttli (1961); Nuttli and Whitmore (1962) to determine P wave velocities from
P waves with periods of 3-7 s and S waves with periods in the order of 10 s.
They found P wave velocities over 7 km s≠1 . This result was interpreted by
Phinney (1964) to be an indicator that the polarization is dependent on the
period range used for the analysis and that for shorter periods lower velocities
would be obtained. Krüger (1994) used the P wave polarization to study
the sedimentary structure at the Gräfenberg array in southern Germany by
analysing the steepening of the P wave onset in terms of the ratio between P
wave and S wave velocities. Whereas Svenningsen and Jacobsen (2007) and
Kieling et al. (2011) used a progressive low-pass filtering of receiver functions
(RF) and the relation presented by Wiechert (1907) to perform an inversion for
an S wave velocity-depth model.
Usually, the P wave polarization angle is determined by the measurement
of the particle motion on the vertical (Z) and radial (R) component of a
seismogram (Krüger, 1994). This measurement needs a careful time window
selection and data preparation, because it is influenced by the often complicated
P wave signal. Svenningsen and Jacobsen (2007) proposed to use (Z, R) RF
instead of the raw earthquake signal to avoid this complexity issue and to ease
automatic processing. The earthquake signal is deconvolved either in time
domain (Kind et al., 1995; Kieling et al., 2011) or frequency domain (Ammon,
70
6.1. Introduction
1991). This procedure transforms the P wave signal into a (band limited) spike
like signal on the vertical and radial component of the RF at t = 0. Thus,
the apparent P wave incidence angle can be measured by determining the
amplitudes of the spike on the two components (Svenningsen and Jacobsen,
2007).
Ocean bottom stations (OBS) are sensors constructed for the deployment on
the ocean floor (Webb, 1998; Dahm et al., 2006). Often, OBS recordings have
a poor signal-to-noise ratio (SNR) and suffer from high noise levels especially
on the horizontal components. This results in a small number of usable event
recordings for these sensors which usually operate for one year or less (Webb,
1998). We increase the number of usable recordings by including Pdiff (90-110¶
epicentral distance) and PKP (140-160¶ ) recordings besides P wave recordings
(30-90¶ ) in our analysis. Furthermore, we have to reconsider the apparent P
wave incidence angle relation for the free surface presented by Wiechert (1907)
for the case of the ocean bottom, because the refracted P wave in the water
column has an influence on the reflection coefficients of the ocean bottom. These
coefficients are needed to calculate the displacements within the ocean bottom.
The coefficients for the reflection and refraction at the interface between a solid
and a liquid half space were calculated for specific model parameters by Knott
(1899). Zoeppritz (1919) presented an analytical calculation of the coefficients
which can also be found in some textbooks (Ben-Menahem and Singh, 2012).
These Zoeppritz equations are also used in reflection seismic (e.g Wang, 1999b)
or receiver function studies (Julià, 2007; Kumar et al., 2014; Prakash, 2015) to
analyse impedance contrasts at interfaces.
We use the reflection coefficients to obtain a new relation which enables us
to determine apparent (half space) S wave velocities from P wave polarization
(apparent P wave incidence angle) measurements. We employ this relation
together with a progressive low-pass filtering analogue to Svenningsen and
Jacobsen (2007) and a quantitative modelling to obtain S wave velocity-depth
models.
6.1.1 Previous studies of oceanic S wave velocity

The S wave velocity structure of the oceanic crust and the upper mantle has
mainly been studied by global tomography of surface waves (Romanowicz, 2003;
Laske et al., 2013) using land stations. The results of those global studies are
biased by the poor data coverage in the oceans (Romanowicz, 2003). Moreover,
these studies employ long wavelengths for their investigation and the resolution
of the gained models is therefore rather low (up to several degrees, Romanowicz,
2003; Laske et al., 2013). If phase velocities or group velocities are used to
determine S wave velocity maps, several stations or arrays are needed and the
obtained results reflect more the average velocity between pairs of stations than
single station’s estimates (Weidle and Maupin, 2008; Maupin, 2011; Gao and
Shen, 2015). Another approach to estimate the S wave velocity structure of
the oceanic lithosphere is the sea floor compliance inversion (Yamamoto and
Torii, 1986; Crawford et al., 1998; Webb, 1998). This technique analyses the
ratio of sea floor displacement to pressure loading due to infragravity waves
in a very low frequency band (0.003 to 0.04 Hz, Crawford et al., 1998). The
usage of this technique is limited, because the displacement by ocean surface
71
waves at deep water sites is small and difficult to measure (Crawford et al.,
1998; Webb, 1998).
There have also been attempts to extract information about the shallow S
wave velocity structure from active seismic data (e.g. up to 300 m in Ritzwoller
and Levshin, 2002 and for the upper tens of metres in Nguyen et al., 2009).
The success of these techniques is directly related to the distance of the active
source to the sea floor. The closer the source is located to the sea floor the
more acoustic energy can be converted into S wave energy (Ritzwoller and
Levshin, 2002). The inversion of these active data results in high resolution
S wave velocity models, but it is limited to the upper hundreds of metres
beneath the sea floor. We therefore propose that by using progressive low-pass
filtering (≥0.05 to 2 Hz), the analysis of P wave polarization in terms of S
wave velocities will provide the opportunity to resolve deeper (crustal) S wave
velocity structures than active seismics and will give a better resolution of the
crustal S wave velocity structures than compliance measurements.
First, we use the reflection coefficients provided by Zoeppritz (1919) and
Ben-Menahem and Singh (2012) to find a relation for the apparent P wave
incidence angle at the ocean bottom analogue to the one presented by Wiechert
(1907). Then, we describe the analysis of the P wave polarisation by progressive
low-pass filtering of (Z, R) RF and the estimation of apparent S wave velocities.
We perform several synthetic tests to investigate the resolution of the proposed
method. Finally, we apply the method to real OBS data from the eastern mid
Atlantic Ocean and perform a quantitative modelling to determine sedimentary,
crustal and mantle S wave velocities and the thickness of the sediments and
the oceanic crust.
6.2 Theory
Considering a seismometer which measures the displacement on the sea floor,
we define a local coordinate system with a vertical z-axis pointing upward
and a horizontal r-axis pointing in the horizontal propagation direction of the
wave front (Fig. 6.1, dashed red line). The z-axis is thus parallel to the vertical
(Z) component of the recorded seismogram and the r-axis is parallel to the
radial (R) component. The displacement u is measured along the r-axis and
the displacement w along the z-axis (Fig. 6.1). The tangents of the ratio of
those displacements is used to estimate the P wave polarization Ïp (apparent
incidence angle, Fig. 6.1 and Wiechert 1907).
u
tan Ïp = (6.1)
w
The displacements u and w result from the superposition of the displace-
ments of different elastic waves at the interface between water column and
ocean bottom (z = 0 in Fig. 6.1). The boundary conditions for the displacement
at the interface between a fluid with low viscosity, e.g. water and a solid are
that the displacement normal to the interface (i.e. w) must be continuous,
whereas the tangential components (i.e. u) can be discontinuous (Knott, 1899;
Ben-Menahem and Singh, 2012; Aki and Richards, 2002). Assuming the seis-
mometer of an OBS measures the displacement of the ocean bottom, we have
to consider the amplitudes of the elastic waves in the ocean bottom to obtain
the P wave polarization Ïp at the ocean floor.
72
6.2. Theory
Figure 6.1: Polarities of P waves (red) and SV wave (blue) at the interface between
water column and ocean bottom. The incoming P wave front is represented as
dashed red line. The particle motions of the single wave types are shown as small
black arrows. The normal of the zr-plane n̂ points into the negative transverse
direction. The ocean bottom has the P wave velocity vp1 , the S wave velocity vs and
the density fl1 . The water column has the P wave velocity vp2 and the density fl2 .
The displacements u and w are measured at the sea floor to estimate the P wave
polarization (apparent incidence angle, Ïp ).
In Fig. 6.1, the unit vectors describing the polarization direction of the
incident P wave (k̂p0 ), the reflected P wave (k̂p1 ) and the reflected S wave
(k̂s ◊ n̂) are presented.
Q R Q R Q R
sin Ïp1 sin Ïp1 cos Ïs
c d c d c d
k̂p0 = a 0 b k̂p1 = a 0 b k̂s ◊ n̂ = a 0 b , (6.2)
cos Ïp1 ≠ cos Ïp1 sin Ïs
where Ïp1 and Ïs are the angles of the incident (and reflected) P wave, and the
reflected SV wave, respectively and n̂ denotes the normal of the zr-plane. The
reflection coefficient Ṕ P̀ is defined as the amplitude ratio of the reflected P
wave and the incident P wave and the reflection coefficient Ṕ S̀ is the amplitude
ratio of the reflected SV wave and the incident P wave. Following Aki and
Richards (2002), we use an acute accent (e.g., Ṕ ) to represent an upcoming
wave and a grave accent (e.g.,P̀ ) to denote a down-going wave. Considering
the reflection coefficients and eq. (6.2), eq. (6.1) can be written as:
1 2
1 + Ṕ P̀ sin Ïp1 + Ṕ S̀ cos Ïs
tan Ïp = 1 2 . (6.3)
1 ≠ Ṕ P̀ cos Ïp1 + Ṕ S̀ sin Ïs
The numerator and the denominator of eq. (6.3) are analogue to the displace-
ments in r and z directions provided by Pilant (1979) and Aki and Richards
(2002) for the solid-solid case, and by Ben-Menahem and Singh (2012) for the
solid-liquid case. The signs of cos Ïp1 and cos Ïs are negative in Pilant (1979)
and Ben-Menahem and Singh (2012) in which the z axis is defined downward
instead of upward as in the seismometer based definition used here (Fig. 6.1).
73
We calculated the coefficients Ṕ P̀ and Ṕ S̀ and compared them to the

coefficients published by Zoeppritz (1919):
1 ≠ f (1 ≠ g)
Ṕ P̀ = (6.4)
1 + f (1 + g)
4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs )
Ṕ S̀ = (6.5)
1 + f (1 + g)
vp1 fl1 cos Ïp2 cos2 (2Ïs )
with: f = (6.6)
vp2 fl2 cos Ïp1
A B2
vs sin (2Ïp1 ) tan (2Ïs )
g= . (6.7)
vp1 cos (2Ïs )
They are similar besides that Zoeppritz (1919) provides eq. (6.4) with a
negative sign, because his definition of the polarization direction of the reflected
P wave (k̂p1 ) is opposite to the definition used here (positive in r and negative
in z direction, Fig. 6.1). In eqs (6.4)-(6.7), vp1 , vs and fl1 are the P wave and S
wave velocity as well as the density of the ocean bottom, and vp2 and fl2 are
the P wave velocity and the density of the water column. The coefficients in
eqs (6.4) and (6.5) are equivalent to the coefficients provided by Ben-Menahem
and Singh (2012) for the mantle-core reflection except for the polarity of Ṕ S̀
which can be explained by the before mentioned differing definition of the z
axis.
We insert eqs (6.4)-(6.7) in eq. (6.3) and use Snell’s law
sin Ïp1 sin Ïp2 sin Ïs
= = = p . . . horizontal slowness (6.8)
vp1 vp2 vs
to obtain the relation for the apparent P wave incidence angle at the ocean
bottom (see supplementary material for details of calculation):
fl2 tan Ïp2
tan Ïp = tan (2Ïs ) + . (6.9)
fl1 cos (2Ïs )
The new equation (6.9) has two terms, the first term equals the well known
relation for the free surface of a solid half space (Wiechert, 1907) and the
second term describes the influence of the water column on the apparent P
wave incidence angle.
Using Snell’s law (eq. 6.8), eq. (6.9) is re-written as function of the horizontal
slowness p (see supplementary material for details of calculation):
A Ú B
Ò
1 1
p fl2
vs2
+ 2fl1 vs2
≠ p2 2
vp2
≠ p2
tan Ïp = Ú 1 2 . (6.10)
1 1
fl1 2
vp2
≠ p2 vs2
≠ 2p2
Equation (6.10) shows that the apparent P wave incidence angle Ïp is

independent of the P wave velocity vp1 of the ocean bottom. In Fig. 6.2, we
present a comparison between Ïp for the ocean bottom and the free surface.
It becomes clear that the apparent P wave incidence angles differ especially
for the water/ sediment contrast. Moreover, if apparent S wave velocities are
estimated at the ocean bottom using the free surface relation, the obtained
velocities will be higher than the true values (compare eq. 6.10).
74
6.3. Methodology
90
QSEIS OC
QSEIS OS
OC
FC
OS
60 FS
30
0
0 30 60 90
Figure 6.2: Comparison of eq. (6.9) and Wiechert formula (Ïp = 2Ïs ).The theoretical
apparent P wave incidence angle Ïp on the ocean floor if the P wave incidence angle
Ïp1 is given are shown as solid lines. The Ïp at the free surface if Ïp1 is given are
shown as dashed lines. The values for an oceanic crust (OC: water layer/ crust, FC:
free-surface/ crust, Tab. 6.1) are shown in blue and the values for a sediment (OS:
water layer/ sediment, FS: free-surface/ sediment, Tab. 6.1) in red. The measured
apparent P wave incidence angles from synthetic seismograms (QSEIS) are presented
as circles for the OC model (blue) and the OS model (red).
6.3 Methodology
The estimation of apparent P wave incidence angles (Ïp ) can be done by using
hodographs of the P wave particle motion (Krüger, 1994). The P wave train
can be rather complex because of the influence of the source time function and
the source-to-receiver wave propagation. The analysis of the particle motion
therefore requires a careful data preparation and time window selection. The
processing is eased by employing receiver functions (RF) for which the R
component is deconvolved with the Z component (Svenningsen and Jacobsen,
2007). By this procedure, the P wave signal turns into a zero-phase (band-
limited) spike which arrives at time t = 0 on the vertical (ZRF ) and radial
(RRF ) component. We perform the deconvolution in time domain by using a
Wiener filter (Kind et al., 1995; Kieling et al., 2011). The apparent P wave
incidence angle can be estimated by measuring the amplitudes at t = 0 on ZRF
and RRF (Svenningsen and Jacobsen, 2007). On RRF , additionally a series of P
to S converted signals become visible after the projection of the direct P spike
signal.
The seismic velocities obtained by analysing the P wave polarisation (ap-
parent P wave incidence angle, Ïp ) are dependent on the used period range
(Haskell, 1960; Phinney, 1964). For longer periods (≥5-10 s), the obtained
velocities are typical for the Earth’s mantle (Nuttli, 1961; Nuttli and Whitmore,
1962). If shorter periods are used for the measurement of the P wave polariza-
tion Ïp , the estimated velocities will be similar to crustal velocities (Phinney,
1964; Svenningsen and Jacobsen, 2007). This behaviour can be used to obtain
velocity-depth profiles (Svenningsen and Jacobsen, 2007). In order to analyse
the apparent P wave incidence angles, we apply a set of different low pass filters
to the (Z, R) RF before estimating the angles (Svenningsen and Jacobsen,
2007). We use Butterworth low-pass filters of second order which are applied
75
forwards and backwards in order to get zero phase filters (Scherbaum, 2007).
The corner periods of the filters are chosen to be logarithmically distributed as
suggested by Svenningsen and Jacobsen (2007).
The (Z, R) receiver functions are calculated and filtered with L low pass
filters for N events. After the filtering, the apparent P wave incidence angles
p,n (Tl ) are measured for each corner period Tl at time t = 0 of the filtered
Ïobs
(Z, R) receiver functions. A misfit function m can be formed which compares
the measured apparent P wave incidence angles Ïobs p,n (Tl ) for N different events
with their calculated theoretical equivalent Ïtheo
p using eq. (6.10) for each corner
period Tl :
1 N
ÿ
m(vs , fl1 , Tl ) = N
· (|D (Tl , vs , fl1 , pn )| · wn ) (6.11)
q n=1
wn
n=1
with
D (Tl , vs , fl1 , pn ) = tan Ïobs
p,n (Tl ) ≠ tan Ïp
theo
(vs , fl1 , pn ) .
The weights wn are chosen based on data quality. Standard values (Tab. 6.1)
are used for the P wave velocity vp2 and density fl2 of the water column to
calculate the theoretical angle Ïtheo
p (eq. 6.10). The horizontal slowness pn for
each event n is calculated for global velocity models (AK135, Kennett et al.,
1995). The remaining unknowns in eq. (6.11) are the S wave velocity vs and the
density fl1 of the ocean bottom. In the following section, we perform synthetic
tests to analyse the dependency of the apparent incidence angle on the S wave
velocity vs and the density fl1 , as well as its behaviour in dependence on the
used corner period Tl for multi-layered models.
6.4 Synthetic tests

In this section, we want to investigate typical structures of the ocean bottom
by calculating synthetic data for different oceanic layered velocity models with
a full wave field reflectivity method (QSEIS, Wang, 1999a) using the model
parameters listed in Tab. 6.1. We use a normalised squared half-sinus function
with a length of 0.5 s which has a flat spectra below 2 Hz as source time
function. The reflectivity method is not able to model a liquid layer with an S
wave velocity vs = 0 km s≠1 , instead we use a very soft solid layer for which the
P to S wave velocity ratio is 1000 as suggested by Müller (1985). The sensor
depth is chosen to be 1 m below sea floor (b.s.f.).
In order to save computation time and to obtain high frequency synthetic
data, we first simulate deep regional events (100 km depth) instead of teleseismic
global events. The models for the regional case are listed in Tab. 6.2. In a
first step, we test the accuracy of QSEIS against the theoretical expression
in eq. (6.10) using two half space models. Afterwards, we add one layer to
investigate the depth resolution of the proposed method. In a third step, we
simulate teleseismic global events with a low sampling frequency (8 Hz) to
investigate the influence of water depth and a low velocity layer (Tab. 6.3). For
the synthetic tests in this section, we set all weights (wn in eq. 6.11) to one.
76
6.4. Synthetic tests
Table 6.1: Model parameters for standard values of water column (WC), sediment
(SD), normal oceanic crust (NOC), oceanic crust with 10% reduced velocities and
density (ROC), normal mantle (NM) and mantle with 10% reduced velocities and
density (RM). We give the P wave velocity vp , the S wave velocity vs and the density
fl.
medium vp [km s≠1 ] vs [km s≠1 ] fl [g cm≠3 ]

WC 1.500 0.000 1.000
SD 2.000 0.500 2.000
NOC 6.500 3.750 2.700
ROC 5.850 3.375 2.430
NM 8.120 4.510 3.340
RM 7.308 4.059 3.006
Table 6.2: Model description for synthetic tests (regional case). All models include a
water column (WC) of 5.05 km (layer 1). Layer thickness for sediment (SD), normal
oceanic crust (NOC),and normal mantle (NM) (for model parameters, see Tab. 6.1).
The source depth is given in kilometres below sea floor (b.s.f.).
model layer 2 half source depth

space [km b.s.f.]
OC - NOC 100
OS - SD 100
N 7 km NOC NM 100
S100C 0.1 km SD NOC 100
S200C 0.2 km SD NOC 100
S300C 0.3 km SD NOC 100
S400C 0.4 km SD NOC 100
S600C 0.6 km SD NOC 100
S800C 0.8 km SD NOC 100
S1000C 1 km SD NOC 100
6.4.1 Half space S wave velocity
The first deep regional source models (model OC and OS, Tab. 6.2) consist of
one layer over a half space. Model OC includes a water column (WC, Tab. 6.1)
and a normal oceanic crust (NOC, Tab. 6.1) half space, and model OS a WC
and a sediment (SD, Tab. 6.1) half space. An explosion source is located at
100 km b.s.f. and the receivers are placed in 5 to 100 km epicentral distance
with 5 km inter-station spacing. This setting corresponds to slowness values of
0.9 s/¶ to 12.1 s/¶ for the OC model and 2.8 s/¶ to 39.3 s/¶ for the OS model,
respectively. The sampling rate is 100 Hz.
The apparent P wave incidence angles Ïobs p,n are determined by measuring the
polarisation for the P wave of each synthetic event within a 1 s time window
for unfiltered data on the Z and R components (circles in Fig. 6.2). By directly
comparing, we find a good agreement between measured and theoretical angles
for the OC model (solid blue line and blue circles in Fig. 6.2) and the OS model
(solid red line and red circles in Fig. 6.2). This shows that the apparent P wave
77
Table 6.3: Description of models at the receiver site for synthetic tests (teleseismic
case). All models include a water column (WC) of 5.05 km (layer 1), PREM below
155.05 km, and continental PREM for the source site (supplementary Fig. 6.S1).
Layer thickness for normal oceanic crust (NOC), oceanic crust with 10% reduced
velocities and density (ROC), normal mantle (NM) and mantle with 10% reduced
velocities and density (RM) (for model parameters, see Tab. 6.1).
model layer 2 layer 3 layer 4 layer 5

CM-REF 7 km NOC 143 km NM - -
CI 1 km ROC 6 km NOC 143 km NM -
CII 3 km NOC 1 km ROC 3 km NOC 143 km NM
CIII 6 km NOC 1 km ROC 143 km NM -
MI-50 7 km NOC 50 km RM 93 km NM -
MII-50 7 km NOC 3 km NM 50 km RM 90 km NM
MIII-50 7 km NOC 13 km NM 50 km RM 80 km NM
MIV-50 7 km NOC 23 km NM 50 km RM 70 km NM
MV-50 7 km NOC 43 km NM 50 km RM 50 km NM
MVI-50 7 km NOC 93 km NM 50 km RM -
Table 6.4: Take-off angles and slowness values used for models in Tab. 6.2 (regional
case). The values are given for a half space consisting of either normal oceanic crust
(NOC) or normal mantle (NM).
take-off slowness [s ¶ ]
≠1
event angle NOC NM

1 5 1.49 1.19
2 10 2.97 2.38
3 15 4.43 3.54
4 20 5.85 4.68
5 25 7.23 5.79
6 30 8.55 6.85
7 35 9.81 7.85
8 40 11.00 8.80
9 45 12.10 9.68
incidence angles obtained from synthetic data (QSEIS) are similar to the values
estimated with our theoretical expression in eq. (6.10).
Furthermore, we test the dependency of the misfit m(vs , fl1 ) (eq. 6.11) on
the S wave velocity vs and the density fl1 by using the measured apparent P
wave incidence angles of the OC model.
The misfit is calculated based on a grid search over S wave velocity vs (0.1-
9.0 km s≠1 in 0.1 km s≠1 steps) and density fl1 (1.0-6.0 g cm≠3 in 0.1 g cm≠3
steps). The result shows that the dependency of the misfit function m(vs , fl1 )
and therefore of the apparent P wave incidence angle on the S wave velocity
78
10 10
5
5
5
2
2 0.5
0.5 0.5
0.5 2
3 2
5
5
1 10
10
1 2 3 4 5 6
Figure 6.3: Misfit function m(vs , fl1 ) calculated by eq. (6.11) is shown as blue contours
over S wave velocity vs and density fl1 of the ocean bottom for the OC model with
WC over NOC. The range in the apparent S wave velocity vs,app estimated by the
minimum of the misfit for each tested density is indicated by black dashed lines. The
median of the apparent S wave velocity vs,app for all tested densities is shown in red.
The relation fl1 (vs ) is presented in green and the estimated vs,app for the root search
is depicted as red circle.
vs is much stronger than on the density fl1 (Fig. 6.3). By searching for the
minimum in the misfit for each tested density value fl1 , an S wave velocity
range (black dashed lines in Fig. 6.3) is determined, for which a median (red
line in Fig. 6.3) is estimated. In the presented case, the median of the apparent
S wave velocity vs,app is 3.8 km s≠1 and its range is 3.4 km s≠1 to 3.9 km s≠1 .
If we do not put any constrains on the density for a grid search with a S wave
velocity step size of 0.1 km s≠1 , the obtained median vs,app =3.8 km s≠1 is a
good estimate of the S wave velocity used in the model (vs = 3.75 km s≠1 ).
Instead of a grid search, we can perform a root search to estimate vs,app by
assuming a relation between the density fl1 and the S wave velocity vs . There
are well known empirically derived relations between the density flx and the P
wave velocity vpx (Brocher, 2005, eq. 6.12, for vpx in km s≠1 and flx in g cm≠3 ).
2 3
flx = 1.6612 · vpx ≠ 0.4721 · vpx + 0.0671 · vpx
4 5
≠ 0.0043 · vpx + 0.000106 · vpx (6.12)
To obtain fl1 (vs ), we have to assume vp (vs ). For S wave velocities up to

2.5 km s≠1 , the mud-rock line (vp = 1.16 · vs + 1.36, Castagna et al., 1985)
serves quite well. ForÔlarger S wave velocities, a constant vp /vs ratio could
be assumed (vp /vs = 3 for 2.5 km s≠1 < vs Æ 4.0 km s≠1 and vp /vs = 1.8
for vs > 4.0 km s≠1 ). By using the assumed relation fl1 (vs ) (green line in
Fig. 6.3) in eq. (6.11), the minimum misfit m(vs , fl1 (vs )) for the OC model is
estimated at vs,app = 3.76 km s≠1 (red circle in Fig. 6.3) for an S wave velocity
step size of 0.005 km s≠1 . This velocity is in good agreement with the used
model parameter (vs = 3.75 km s≠1 ) and similar to the median vs,app obtained
by the grid search with a coarser step size.
79
In conclusion, the weak dependency of the misfit function m(vs , fl1 ) (i.e. the
apparent P wave incidence angle) on the density shows that we could hardly
resolve densities with the presented method. We therefore will neglect the weak
influence of the density in the further processing.
In the next section, we analyse the behaviour of vs,app with the period Tl for
a layered model and compare the results obtained by estimating the median
vs,app with the grid search and by determining vs,app using the root search. Both
approaches have proved to give good estimates of the true half space S wave
velocity.
6.4.2 Depth resolution

We use several synthetic models to test the depth resolution of the method.
All models presented in Figs 6.4 and 6.5 include a 5.05 km thick water column
(WC) which is given by CRUST 1.0 for the area of the OBS deployment (Laske
et al., 2013, supplementary, Fig. 6.S1). The angles in Tab. 6.4 give the direction
the ray travels within the half space and are measured towards the normal of
the layer interface (i.e. against vertical). The associated slowness values are
also given in Tab. 6.4. The vertical component of the synthetic data (100 Hz)
is used to create a Wiener filter, choosing a 5 s time window starting at the P
onset. The filter is used to deconvolve the vertical and radial components in
order to estimate (Z, R) RF for the analysis. The RF are filtered with a set of
low pass filters with L corner periods Tl which are logarithmically distributed
(0.5 s to 64 s, 8 filters per octave). The apparent P wave incidence angles
grid search
syn. event 1
syn. event 4
1 syn. event 8
root search
10
3 3 3 4 4 4 5 5 5 6 6
Figure 6.4: Synthetic test for model N (7 km NOC over NM half space below sea
floor, Tabs 6.1 and 6.2). The S wave velocities used in the model are marked by
dashed light grey lines. The median vs,app profiles obtained for synthetic event 1
(dashed yellow, Tab. 6.4), synthetic event 4 (dash-dot red, Tab. 6.4) and synthetic
event 8 (dashed red, Tab. 6.4) are shown. The median vs,app profile obtained from
the total misfit function of all events (grid search, eq. 6.11) is shown in black. The
vs,app profile estimated using the root search is presented in grey.
80
are estimated from the amplitudes on ZRF and RRF at time t = 0 (relative to
deconvolved P spike on ZRF ).
We test model N which consists of 7 km normal oceanic crust (NOC,
Tab. 6.1) over a normal mantle (NM, Tab. 6.1) half space below sea floor (b.s.f.).
We obtain m(vs , fl1 , Tl ) and m(vs , fl1 (vs ), Tl ) for the estimated apparent P wave
incidence angles and different periods Tl . The minima of misfit m(vs , fl1 , Tl ) are
determined for each density fl1 . We obtain the median vs,app for each period Tl
and estimate the roots of m(vs , fl1 (vs ), Tl ) to get the vs,app profiles (black and
grey solid line in Fig. 6.4). To show the variability of the results for slowness
values typical for P and Pdiff (4-9 s/¶ , ≥30-110¶ epicentral distance) and PKPdf
(1-2 s/¶ , ≥140-160¶ epicentral distance), we included the median vs,app profiles
for synthetic event 1 (1.19 s/¶ , dashed yellow line in Fig. 6.4), synthetic event 4
(4.68 s/¶ , dash-dot red line in Fig. 6.4) and synthetic event 8 (8.80 s/¶ , dashed
red line in Fig. 6.4).
The vs,app profiles obtained by the grid search and the root search agree
very well. The only difference is the smoother appearance of the root search
profile due to the smaller step size in vs (0.005 km s≠1 compared to 0.1 km s≠1 ).
Besides this, the overall appearance of both profiles is identical.
The vs,app profiles show the velocity of the upper layer for periods up to
≥2 s. For this period range, all obtained vs,app profiles agree very well. The
kinks of the profiles at which they start to diverge from the S wave velocity
of the upper layer are approximately at 2 · tP s which is twice the delay time
of the Ps conversion for a slowness of 6.36 s/¶ (Fig. 6.4). For longer periods,
the vs,app profiles of model N bump (’overshoot’) before they converge towards
the velocity of the half space (Fig. 6.4). This effect was also described by
Svenningsen and Jacobsen (2007) and was interpreted to be related to the
effect of crustal multiples on the filtered receiver functions for longer periods.
Furthermore, we find that for smaller slowness values (e.g. 1.19 s/¶ , dashed
yellow line in Fig. 6.4) the bump in the vs,app is larger than for larger slowness
values (e.g. 8.80 s/¶ , dashed red line in Fig. 6.4), but the profile with the smaller
slowness value converges faster towards the half space S wave velocity. This
effect can be explained by shorter delay times of crustal multiples for smaller
slowness values.
Due to the agreement of the vs,app profiles obtained by grid search and root
search, we decide to present only the vs,app profiles obtained by the root search
for the following comparison of the different tested models for a better visibility
of the behaviour of the different profiles. In Fig. 6.5, we show the results for a
test of the influence of the upper solid layer thickness on the appearance of the
vs,app profiles. The models named S100C to S1000C consist of a water column
(WC) and a sediment (SD) layer of thickness 100 m to 1000 m over a normal
oceanic crust (NOC) half space (Tabs 6.1 and 6.2). The effect of ’overshooting’,
described for model N in Fig. 6.4, is also visible for the SD-NOC models. The
bump in the vs,app profile is shifted to longer periods for thicker layers, and also
increases in velocity for larger thicknesses (Fig. 6.5). For S100C, the profile
reaches a velocity of 4.13 km s≠1 , whereas for model S1000C, the maximum
velocity lies at 5.365 km s≠1 (Fig. 6.5).
In conclusion, the overall appearance of the vs,app profiles for a model with
a solid layer over half space b.s.f. is determined by the S wave velocity of the
upper solid layer for short periods and the S wave velocity of the half space for
81
0.1 1
1 10
0 2 4 6
S100C
S200C
S300C
S400C
S600C
S800C
S1000C
Figure 6.5: Synthetic tests for two layers (including WC and SD) over NOC half
space. Detailed model description can be found in Tab. 6.2 and used parameters in
Tab. 6.1. Left panel shows used velocity models for depth below sea floor (b.s.f.).
Right panel shows estimated vs,app profiles with corner period T .
longer periods. The thicker the upper solid layer the longer the period and the
larger the maximum vs,app of the ’overshooting’ bump in the vs,app profile get.
6.4.3 Influence of water depth

For the teleseismic tests presented in Figs 6.6 and 6.7, we take advantage of the
ability of QSEIS to use different source and receiver site models to calculate
synthetic body waves for teleseismic distances. We choose continental PREM
(supplementary, Fig. 6.S1, Dziewonski and Anderson, 1981)) on the source site
and utilise different receiver site models for depths above 155.05 km (Tab. 6.3).
We choose a double couple with a dip of 45¶ , a rake of 90¶ and a strike of 0¶ at
15 km depth as source. The source is located at a back azimuth of 69¶ in a
distance of 47.6¶ (p = 7.78 s/¶ ) which corresponds to the values for event #4
of the analysed real data (Tab. 6.5). The synthetic data are sampled with 8 Hz
in order to minimize computation time. For the calculation of (Z, R) RF, we
employ a Wiener filter which is estimated by using the vertical component of
the synthetic seismograms and a 80 s time window starting at the P onset.
We create a reference model (CM-REF, Tab. 6.3) with a 5.05 km thick
water column (WC), a 7 km thick normal oceanic crust (NOC) and a normal
mantle (NM) layer on the receiver site for the comparison of the estimated
vs,app profiles of the models with varied water depth and those including a low
velocity layer (LVL) in either crust or mantle. The reference model is depicted
with a solid blue line in Figs 6.6 a and 6.7.
Figure 6.6 shows the influence of the water depth on the appearance of the
S wave velocity profiles. For this test, we create receiver site models including
a WC with a thickness from 0.55 km to 7.05 km. The models consist of a 7 km
thick NOC and a NM (Tab. 6.1) which extends to a total depth of 155.05 km
82
NOC NM
Water depth R
0.55 km
1 1.05 km
3.05 km
4.05 km
5.05 km Z
6.05 km
7.05 km
Z
10
3.5 4.0 4.5 5.0 20 101010100

100
100 010
010
0101020
1020
10202030
2030
20303040
3040
3040404040
Figure 6.6: (a) Estimated vs,app profiles with corner period T for synthetic tests
for two layers (including WC) over oceanic PREM. Each model contains a 7 km
thick layer of NOC and a NM up to a depth of 155.05 km. The water depth varies
from 0.55 km to 7.05 km. (b) Example RF for water depths 1.05 km, 5.05 km and
7.05 km. The arrival times of the water multiples for each corresponding water depth
are indicated by blue lines.
where PREM takes over. It is visible that the overall appearance of the profiles
is similar (Fig. 6.6 a). All profiles have velocities similar to the NOC for periods
shorter than 2 s. For longer periods, all profiles show a bump in velocity. The
maximum in velocity is similar or larger than the velocity of the NM and
increases for larger water depth. The decrease of vs,app below NM velocities
at longer periods is probably related to the additional (long period) phases
present in the global case (e.g. W phase, Kanamori, 1993) and/ or the possible
incomplete deconvolution of the P wave signal at these periods which depends
on the Wiener filter parameters (supplementary, Fig. 6.S3).
The behaviour of the profiles can be explained by the influence of the water
multiples which is directly related to their travel times. The thinner the water
layer the more water multiples arrive in a shorter time window, e.g. for 0.55 km
the travel time of a water multiple is ≥0.73 s and for 7.05 km ≥9.4 s (vp,W C =
1.5 km s≠1 and p = 7.78 s/¶ , Fig. 6.6 b). The more water multiples arrive in a
shorter time window the more the signal of the direct P wave gets distorted and
this has a direct influence on the Wiener filter estimation for the deconvolution.
This is visible in Fig. 6.6 b in which the RF for water depths of 5.05 km and
7.05 km show a series of regular spaced positive and negative spikes on ZRF
which are expected if the direct P wave is properly deconvolved. For a water
depth of 1.05 km, no such spike series is observed.
Nevertheless, the water depth has only a minor influence on the appearance
of the vs,app profiles at least for the deep ocean, but a removal of the water
multiples (Osen et al., 1999; Thorwart and Dahm, 2005) before deconvolution
might be useful for shallow water depths to prevent influences by wave form
distortions.
83
6.4.4 Influence of LVL

We test the ability of our method to detect a LVL in different depths in either
crust or mantle (Fig. 6.7). All models have again a 5.05 km thick water column
(WC) as is given by CRUST 1.0 for the area of the deployment (supplementary,
Fig. 6.S1, Laske et al., 2013)). The first three models in Fig. 6.7 a (CI, CII,
CIII, Tab. 6.3) consist of a normal oceanic crust (NOC) with a 1 km thick layer
of 10% reduced crustal velocities and density (ROC) in three different depths
ROC NOC NM
0.1
1
1
10 10
100
3.5 4.0 4.5 5.0
CM-REF
CII
CIII
CI
NOC RM NM
0.1
1
1
10 10
100
3.5 4.0 4.5 5.0
CM-REF
MIV-50
MVI-50
MI-50
MII-50
MIII-50
MV-50
NOC RM NM
0.1
1
1
10 10
100
3.5 4.0 4.5 5.0
CM-REF
MIV-1
MIV-5
MIV-10
MIV-20
MIV-50
MIV-100
Figure 6.7: Synthetic test for three to five layers (including WC) over PREM.
Detailed model description can be found in Tab. 6.3 and used parameters in Tab. 6.1.
Left column shows used velocity models with depth below sea floor (b.s.f.). Right
column shows estimated vs,app profiles with corner period T for models in left panel.
S wave velocity of NOC and NM used for forward calculation are given by dashed
lines. The velocity of the LVL introduced in the model is indicated by a dotted line.
(a) Models and vs,app profiles for a 1 km thick LVL at different depths in the crust.
(b) Models and vs,app profiles for a 50 km thick LVL at different depths in the mantle.
(c) Models and vs,app profiles for a LVL with different thickness at a depth of 30 km
b.s.f. in the mantle.
84
below sea floor (0 km, 3 km and 6 km), and a normal mantle (NM) above
PREM. In Fig. 6.7 a, the estimated vs,app profiles of the three models are quite
similar in appearance. For model CI (orange profile in Fig. 6.7 a), we find lower
velocities for the shorter periods (<0.8 s) compared to the reference model
CM-REF (blue profile in Fig. 6.7 a). Model CII (yellow profile in Fig. 6.7 a)
has lower velocities from ≥0.7 s to ≥2 s compared to the reference model.
This appearance might also be explained with a model consisting of two solid
layers over PREM b.s.f. with a lower crustal velocity than the reference model
CM-REF. The last model CIII (purple profile in Fig. 6.7 a) shows nearly the
same appearance as the reference model CM-REF.
The next six receiver site models (MI-50, MII-50, MIII-50, MIV-50, MV-50,
MVI-50, Tab. 6.3 and Fig. 6.7 b) consist of a NOC and a NM with a 50 km
thick layer of 10% reduced mantle velocities and density (RM) in six different
depths below sea floor (7 km, 10 km, 20 km, 30 km, 50 km and 100 km). In
Fig. 6.7 b, all profiles show a velocity of ≥ 3.75 km s≠1 for periods shorter than
≥2 s which corresponds very well to the S wave velocity of the NOC. The vs,app
profiles for model MI-50 and MII-50 (orange and yellow profile) significantly
differ from the reference model (blue profile) for longer periods (> 2 s). Both
profiles show a bump in velocity, which has a maximum velocity similar to
the S wave velocity of the RM. Neither the profile of MI-50 nor MII-50 show
velocities comparable to the NM S wave velocity. The profile of model MII-50
behaves in a similar way like the profile of the model MI-50. This indicates
that the 3 km thick layer of NM in model MII-50 has only small influence on
the appearance of the estimated vs,app profile. The maximum velocity of the
model MII-50 is slightly increased compared to the MI-50 profile. This effect
might also be explained with a model similar to MI-50 but with a faster or
thicker layer than the 50 km RM.
The other models in Fig. 6.7 b (MIII-50, MIV-50, MV-50 and MVI-50) show
a clear bump in their vs,app profiles. Furthermore, their profiles are nearly
identical to the CM-REF profile for periods shorter than ≥5.6 s. The bump in
velocity increases from 4.535 km s≠1 to 4.8 km s≠1 with the thickness of the
upper NM layer (13 km to 93 km). Furthermore, its maximum lies at longer
periods the deeper the location of the LVL. The velocities at periods longer than
16 s increase with the depth of the LVL. Despite a larger maximum velocity,
the MVI-50 profile has a similar appearance as the CM-REF profile. This
indicates a possible trade-off between the depth of the LVL and the uppermost
mantle velocity. It might therefore be explained by a model with a higher
mantle velocity and no LVL if this would be observed for real (noisy) data
(supplementary, Fig. 6.S2).
At last, we tested the influence of a LVL in the mantle at a depth of 30 km
b.s.f. for different layer thickness (1 km, 5 km, 10 km, 20 km, 50 km and
100 km; models MIV-1, MIV-5, MIV-10, MIV-20, MIV-50, MIV-100, Tab. 6.3
and Fig. 6.7 c). The appearance of all tested models in Fig. 6.7 c is similar to
the models MIII-50, MIV-50, MV-50 and MVI-50 discussed before. All profiles
show a similar behaviour to the reference model CM-REF for periods shorter
than ≥5.6 s. The appearance of the bump in velocity differs. For the models
MIV-1 and MIV-5, the profiles are nearly identical to the reference model
CM-REF. The profiles of the other four models (MIV-10, MIV-20, MIV-50 and
MIV-100) mainly differ in the decrease in velocity with increasing thickness of
the LVL for periods longer than ≥ 12 s (Fig. 6.7 c).
85
In conclusion, a thin crustal LVL can be detected in the upper and middle
crust, but not in the lower crust (Fig. 6.7 a). A thin fast velocity layer or LVL
in the uppermost mantle has only minor influence on the vs,app profile. A clear
influence on the vs,app profile is visible for thicker (> 10 km) fast and LVL
above ≥50 km b.s.f..
6.4.5 Summary of synthetic tests

We conclude that the overall appearance of the vs,app profile gives an indication
of the number of layers which should be used to model the profile. The method
is able to resolve either an increase or a decrease in the S wave velocity (e.g.
Figs 6.4 and 6.7) and it is sensitive to the thickness of the layers (e.g. Figs 6.5
and 6.7). Furthermore, we find that the water depth has only a minor influence
on the appearance of the vs,app profile (Fig. 6.6). A 1 km thick LVL with 10%
reduced crustal velocities and densities is detectable in the upper and middle
crust (0-3 km b.s.f., model CI and CII), but it has only a minor influence on
the appearance of the vs,app profile if it is located in the lower crust (7 km
b.s.f., model CIII). A small influence on the vs,app profiles is also observable
for thin (<10 km) fast velocity layers (e.g. MII-50) or LVL (e.g. MIV-1 and
MIV-5) in the uppermost mantle but likely remains undetected if the data are
not compared to those of undisturbed regions. The appearance of the vs,app
profile is clearly influenced by LVL or fast velocity layers which are at least
a few tens of kilometres thick (e.g. MIV-20, MIV-50). We also find that a
50 km thick LVL with 10% reduced mantle velocities and densities has an
influence on the appearance of the vs,app profile if its interface to an upper layer
(either NM or NOC) is located at depths above ≥50 km b.s.f. (e.g. MI-50,
MV-50). On the other hand, it should be noted that the search for a velocity
model which explains a given vs,app profile is non-unique (e.g. similarity of
CIII and CM-REF) due to the trade-off between S wave velocity and layer
thickness. In the following section, we present the vs,app profiles for real OBS
data in the eastern mid Atlantic and describe a quantitative modelling approach
to determine S wave velocity-depth models which is designed based on the
conclusion drawn in this section and the obtained vs,app profiles of the real data.
6.5 Application to real data

6.5.1 Data
Within the DOCTAR project (Deep OCean Test ARray), twelve broadband
ocean bottom stations were deployed in the eastern mid Atlantic (Fig. 6.8)
approximately 60 km to 135 km North of the Gloria Fault which is part of the
Eurasian-African plate boundary (Bird, 2003).
The stations recorded seismometer and hydrophone data from July 2011
until April 2012. The array had an aperture of ≥75 km and was located in
4.5 km to 5.5 km water depth. One of the twelve stations had two clamped
components (filled red triangle in Fig. 6.8, Hannemann et al., 2014), therefore
we do not use the data from this station for our analysis. The data are time
corrected (Hannemann et al., 2014) and the horizontal components are oriented
by using P phases and Rayleigh phases (Stachnik et al., 2012; Sumy et al.,
2015) of know teleseismic events.
86
6.5. Application to real data
30 20 10
40 OBS array Portugal
Azores km
fault
Gloria
35 0 12.5 25
Madeira
38.80
0 500
D11
125 km
D07 D06
4000
D04
38.40
D10 D01 D12
D03 D02
D08 80 km D05
5000
D09
38.00
OBS towards Gloria fault

6000
18.80 18.40 18.00
Figure 6.8: Layout and location of the OBS array. The bathymetry (EMEPC, Task
Group for the Extension of the Continental Shelf) is indicated by the colour. The
OBS positions are marked with triangles. Station D05 had two clamped components
and is not used in the analysis. The distance to the Gloria fault along a N-S profile
is given by the white line. The location of the OBS array and the Eurasian-African
plate boundary (Gloria Fault, Bird, 2003) is shown on the inset map.
For the analysis, we exclude all events for which a strong resonance with
periods between 0.5 s to 4 s (depending on the station, Figs 6.9 a-c for station
D03) is observed and for which this resonance has a clear influence on the
estimated polarization angle (e.g. Fig. 6.9 d). The observed resonance has
a specific period range for each station which can also be identified in the
probabilistic power spectral density (PPSD) of all three components by elevated
amplitudes (Figs 6.9 a-c at ≥3 s for station D03). We think that this resonance is
related to the sedimentary cover in which wave energy is trapped. The resonance
is triggered by ambient noise as well as body waves (compare Figs 6.9 d and e
before and after the P wave arrival). Furthermore, an incoming P wave at
station D03 initially results in a resonance signal on the Z and the R component.
Approximately 9 s later, an increasing resonance is observed on the T component
(Fig. 6.9 e). It is beyond the scope of this study to further describe or analyse
this phenomenon. We only analyse events and period ranges for which the
earthquake signal is visible in the recordings and stronger than the resonance
signal. We also exclude all periods shorter than the corner period of the event
recording from the analysis.
We use one to five events at the different stations and analyse in total 33
events at all stations (Tabs 6.5 and 6.6 and Fig. 6.10). We choose the window
length for the deconvolution for the (Z, R) receiver functions for each event
based on the quality of the recorded signal. For the damping parameter of
the Wiener filter, we use 0.01. Furthermore, we include all apparent P wave
incidence angle measurements in the analysis, for which the SNR on ZRF and
RRF is larger than 4 (signal time window [-10 s,10 s] and noise time window
[-55 s,-25 s] relative to the direct P spike). We select the weight wn in eq. (6.11)
to be the SNR on RRF .
87
30
60 60 60
80 80 80 24
100 100 100
18
120 120 120
[%]
140 140 140 12
160 160 160
6
180 180 180
200 200 200 0

0.01 0.10 1.00 10.00 100.00 0.01 0.10 1.00 10.00 100.00 0.01 0.10 1.00 10.00 100.00
1 1
D03 T D03 T
0 0
1 1
D03 R D03 R
0 0
1 1
D03 Z D03 Z
0 0
1e+04 4e+04
D03 T 0 D03 T 0
1e+04 4e+04
1e+04 4e+04
D03 R 0 D03 R 0
1e+04 4e+04
1e+04 4e+04
D03 Z 0 D03 Z 0
1e+04 4e+04
1e+04 4e+04
D03 T 0 D03 T 0
1e+04 4e+04
1e+04 4e+04
D03 R 0 D03 R 0
1e+04 4e+04
1e+04 4e+04
D03 Z 0 D03 Z 0
1e+04 4e+04
40 20 0 20 40 60 80 100 40 20 0 20 40 60 80 100
Figure 6.9: (a) Probabilistic power spectral density (PPSD, McNamara, 2004) for
vertical component (HHZ) of station D03 for recording period of 10 months. New
High and Low Noise Model (Peterson, 1993) is shown in grey. (b) Same as a, but
for not oriented horizontal component HH3. (c) Same as a, but for not oriented
horizontal component HH4. (d) Data example for event #7 at station D03 which
has not been used in the following analysis. From top to bottom, the three traces for
raw data (in counts), filtered data (in counts, bandpass 2 s to 5 s) and the low pass
filtered (3 s) receiver functions (normalized to P peak on Z) are shown. (e) Same as
d, but for event #8 which has been used in the following analysis.
Table 6.5: Events used for the P wave polarization analysis, origin time, hypocentre
location and moment magnitude Mw from the NEIC catalogue (earthquake.usgs.gov),
is the epicentral distance and p is the horizontal slowness as calculated from the
AK135 travel time tables (rses.anu.edu.au/seismology/ak135).
# origin time lat. lon. depth Mw p

[dd.mm.yyyy hh:mm:ss] [¶ ] [¶ ] [km] [s ¶ ] [¶ ]
≠1
1 24.08.2011 17:46:11 -7.64 -74.53 147.0 7.0 6.14 59.0

2 02.09.2011 10:55:53 52.17 -171.71 32.0 6.9 4.85 86.9
3 02.09.2011 13:47:09 -28.40 -63.03 578.9 6.7 5.32 78.5
4 23.10.2011 10:41:23 38.72 43.51 18.0 7.1 7.78 47.6
5 22.11.2011 18:48:16 -15.36 -65.09 549.9 6.6 6.00 69.1
6 27.12.2011 15:21:56 51.84 95.91 15.0 6.6 5.90 73.5
7 26.02.2012 06:17:19 51.71 95.99 12.0 6.7 5.88 73.6
8 20.03.2012 18:02:47 16.49 -98.23 20.0 7.4 5.98 72.1
9 11.04.2012 08:38:36 2.33 93.06 20.0 8.6 4.44 105.2
10 11.04.2012 22:55:10 18.23 -102.69 20.0 6.5 5.81 74.6
11 17.04.2012 07:13:49 -5.46 147.12 198.0 6.8 1.77 144.7
88
NOC NM NOC NM NOC NM
1 1 1
10 10 10
0.1 22 4 4 6 6 8 8 0.1 22 4 4 6 6 8 8 0.1 22 4 4 6 6 8 8
1 1 1
10 10 10
0.1 22 4 4 6 6 8 8 0.1 22 4 4 6 6 8 8 0.1 22 4 4 6 6 8 8
1 1 1
10 10 10
0.1 22 4 4 6 6 8 8 0.1 22 4 4 6 6 8 8 0.1 22 4 4 6 6 8 8
NOC NM NOC NM
1 1
0 5 10+
10 10
0.1 22 4 4 6 6 8 8 0.1 22 4 4 6 6 8 8
Figure 6.10: Weighted hit-counts of vs,app profiles and median profiles of total misfit
(eq. 6.11) for real data. We used the weights wn which were applied to form the
total misfit to estimate the weighted hit-counts. Station names are given in bold and
number of events in normal font. The specific events are given in Tabs 6.5 and 6.6.
The S wave velocities of normal oceanic crust (NOC) and normal mantle (NM) are
marked with grey dashed lines.
It is likely that serpentinite is present in either the oceanic crust or mantle

close to a major transform fault like the Gloria fault (White et al., 1992).
Therefore, we do not put any constrains on vp /vs ratios by using the grid search
presented in section 6.4 rather than the root search to estimate the median vs,app
profiles shown in Fig. 6.10. We use the period range between 0.5 s to 16.0 s for
the analysis, because of the band limited nature of the earthquake signals and
a known high self-noise level of the used instruments for periods longer than
89
Table 6.6: Events (Tab. 6.5) used at single stations for analysis of P wave polarization
and as presented in Fig. 6.10.
# D01 D02 D03 D04 D06 D07 D08 D09 D10 D11 D12
1 X X X X
2 X
3 X X
4 X X X X X X X
5 X
6 X
7 X X X
8 X X X X X X X X
9 X X X
10 X X
11 X
total 1 3 4 3 5 4 5 1 3 3 1
10 s (Stähler et al., 2016, and Figs 6.9 a-c). In order to show the variability of
the results, we obtain the vs,app profiles for each event (Tabs 6.5 and 6.6) and
each tested density fl1 (1.0-6.0 g cm≠3 in 0.1 g cm≠3 steps). We define a grid
with cells centred at all used corner periods Tl and all possible apparent S wave
velocities vs,app and count the crossings (hits) of all vs,app profiles for each grid
cell (grey-scale plots in Fig. 6.10). This visualisation gives the opportunity to
get an idea about the uncertainties of the obtained result. We observe that the
real data estimates for vs,app often show a multi-modal distribution for each
period Tl which represents the individual events. Therefore, the median profile
of the total misfit and the weighted hit-count of the individual event’s estimates
are, in our opinion, a better representation of the result and its uncertainties
than the mean and its standard deviation.
The S wave velocities of normal oceanic crust (NOC) and normal mantle
(NM) are given by the dashed gray lines in Fig. 6.10. This shows that for
the majority of the OBS (except D07 and D10, Figs 6.10 f and i) the shorter
periods show smaller velocities than NOC. On the other hand, S wave velocities
similar or larger than NM are observed at longer periods (Fig. 6.10). Station
D09 (Fig. 6.10 h) has a data set which is limited to the period range 0.5 s to
8 s. At station D07 (Fig. 6.10 f), the longer periods (>4 s) might be biased by
noise (compare supplementary Fig. 6.S2). The vs,app profiles of single events at
station D11 (Fig. 6.10 j) show S wave velocities close to and larger than mantle
S wave velocities at longer periods (>4 s) which might be an indication for a
different influence of noise on the single events. At stations D06, D08 and D10,
the median vs,app profile is dominated by different events (Fig. 6.10 e, g and i).
This leads to a kink in the vs,app profile at station D08 (≥4 s, Fig. 6.10 g). At
station D06, we observe a jump from vs,app which are larger than mantle S wave
velocities to velocities similar to crustal S wave velocities (≥4 s, Fig. 6.10 e). A
small jump from vs,app which are larger than crustal S wave velocities to crustal
S wave velocities is visible at station D10 (≥0.7 s, Fig. 6.10 i).
Overall, the vs,app profiles can be divided into two groups: those which have
a continuous appearance (D01-D04, D09, D10, D12, Figs 6.10 a-d, h, i, k) and
those which have jumps and kinks (D06, D08 and D11, Figs 6.10 e, g and j) or
are probably influenced by noise at longer periods (D07, Fig. 6.10 f).
90
6.5.2 Quantitative modelling approach

In this section, we develop a quantitative modelling approach based on the
observation made for the vs,app profiles in the previous section (Fig. 6.10) and
the conclusions drawn from the synthetic tests (Figs 6.4-6.7). In order to reduce
the number of forward models needed, we use a unified water depth of 5.05 km
as we saw that in the deep ocean the water depth has only a minor influence on
the appearance of the vs,app profiles (Fig. 6.6 a). For the solid part of the model,
we conclude from the vs,app profiles of most of the OBS which have slower
velocities than normal oceanic crust (NOC) at short periods that at least three
solid layers above PREM b.s.f are needed for the quantitative modelling. These
layers represent sediment, crust and uppermost mantle, and we will search for
the thickness of sediment (ds ) and crust (total crustal thickness d = dc + ds ),
as well as the S wave velocities of all three layers (sediment, vss ; crust, vsc ;
uppermost mantle vsm ). The third solid layer (uppermost mantle) extends to a
depth of 150 km b.s.f at which PREM takes over.
The calculation of the synthetic seismograms is performed by using the
same source site model (PREM without ocean, supplementary, Fig. 6.S1) as
for the synthetic teleseismic tests (Figs 6.6 and 6.7). A double couple with the
same properties as for the synthetic tests in Figs 6.6 and 6.7 is chosen as source,
but this time the source is seen from a back azimuth of 229¶ in an epicentral
distance of 72.1¶ which corresponds to event #8 of the analysed real data
(Tab. 6.5). We model the data by comparing the obtained median vs,app profiles
(ṽs,app
obs
) to the profiles which are determined from the synthetic seismograms
(ṽs,app ). Furthermore, we estimate whether the tested model (ṽs,app
syn syn
) performs
better or worse than a predefined reference model (ṽs,app ) in properly matching
ref
the real data results. For this purpose, we define an objective function R which
is the ratio of the rms value of the weighted difference between ṽs,app
obs
and ṽs,app
syn
0 0 0
5 5 5
1 1 1
10 10 10
10 10 10
0 2 4 6 0 2 4 6 0 2 4 6
0 2 4 6 0 2 4 6 0 2 4 6
Figure 6.11: Illustration of the parameters and weighting used for the three step
quantitative modelling approach. The parameter ranges are shown by red dashed
boxes. The weighting of the different period ranges are given in the inset figure. (a)
First modelling step: Search for the S wave velocity vsc and the thickness ds of the
sediment and the chosen weights wp (Tl ) for this step. (b) Second modelling step:
Search for the S wave velocity of the mantle vsm and the total thickness of the crust
d and the weights wp (Tl ) for this step. (c) Third modelling step: Search for the S
wave velocity of the crust vsc and the equally chosen weights wp (Tl ).
91
(rmsobs≠syn ) and the rms value of the weighted difference between ṽs,app
obs
and
ṽs,app (rmsobs≠ref ).
ref
ı̂ L 1 22
ıq obs (T ) ≠ ṽ syn (T )
ı ṽ
ı l=1 s,app l s,app l · wl
rmsobs≠syn
R= = ı
ıq (6.13)
rmsobs≠ref Ù L 1 obs 22
ṽs,app (Tl ) ≠ ṽs,app (Tl )
ref
· wl
l=1
The objective function R is calculated for L different corner periods Tl . The

weights wl = w̃ · wp (Tl ) are the medians w̃ of the SNRR values (wn in eq. 6.11)
of the contributing measurements at period Tl times an optional, individual
weight for different period ranges (wp (Tl ), Fig. 6.11). The value for R is smaller
than 1 if the median vs,app profile of the current tested model matches ṽs,appobs
better than ṽs,app

ref
. On the other hand, R is larger than 1 if the current tested
model performs worse than the reference model. We therefore search for the
minimum in R to find the best performing model.
In order to limit the number of model parameters, we fix the vp /vs ratio ax
and the density flx which we need for the modelling with QSEIS. The variable
x refers to either sediment (s), crust (c) or mantle (m). The densities flx are
estimated by using the Nafe-Drake curve given by Brocher (2005) (eq. 6.12).
We choose a fine grid to sample our possible model space. We vary the
S wave velocity of the sediment (vss ) between 0.1 and 2 km s≠1 , the S wave
velocity of the crust (vsc ) between 2.5 and 4.5 km s≠1 and the S wave velocity of
the uppermost mantle (vsm ) between 4.0 and 6.0 km s≠1 in steps of 0.1 km s≠1
(Fig. 6.11). The sediment thickness (ds ) is varied between 0.1 and 1 km in steps
of 0.1 km and the total crustal thickness (d) between 4 and 12 km in 0.5 km
steps (Figs 6.11 a-b). This results in nearly 1.5 million models which need to
be modelled and tested to sample the whole 5D model space. To reduce the
computational effort, we therefore use the depth sensitivity of the vs,app profile
for different period ranges by employing a three step quantitative modelling,
although this approach might not be suited to sample the whole 5D model
space.
In the first step, we concentrate on the sedimentary layer and vary its
S wave velocity vss and its thickness ds (0.1 km s≠1 < vss < 2 km s≠1 in
0.1 km s≠1 steps, 0.1 km< ds <1 km in 0.1 km steps, Fig. 6.11 a). The P wave
velocity vps is determined by using an arbitrary vp /vs ratio as which is chosen
in dependence on the P wave velocities of the WC (vpw , Tab. 6.1) and the NOC
(vpc0 , Tab. 6.1):
Y
]4 + n if 4 · vss < vpw with n œ N · n > ≠4
vpw
_
_ vss
as = 4 if 4 · vss > vpw · 4 · vss < vpc0
2
, (6.14)
_
[Ô
_
3 if 4 · vss > vpc0
2
in which the value n is a natural number. The density fls is calculated using

eq. (6.12). The crustal and the mantle layer have the properties of NOC and
NM (Tab. 6.1) and the total crustal thickness d is set to 7 km. As reference
model for estimating ṽs,app
ref
, we choose CM-REF (Tab. 6.3) which has a 5.05 km
thick WC, no sediment layer, a 7 km thick NOC and a NM above PREM.
We search for the minimum in the objective function R(vss , ds ) (eq. 6.13). We
92
learned from the results of the synthetic tests CI and CII (Fig. 6.7) that a LVL
in the upper crust influences mainly the shorter periods of the vs,app profile,
conclusively the weights wp (Tl ) in eq. (6.13) for the individual period ranges
are chosen to be higher for shorter periods than for longer periods (inset in
Fig. 6.11 a). The model resulting from the first step is used as reference model
in the next modelling step.
After determining the sediment properties, a natural procedure would be to
continue with the crustal velocity and thickness. Examining the vs,app profile
of the model CI reveals that the period range in which the crustal properties
influence the appearance of the profile is hard to isolate (Fig. 6.7 a). On the
other hand, the longer periods are clearly influenced by the mantle properties
(Fig. 6.7 b-c). The second modelling step therefore focuses on the uppermost
mantle S wave velocity vsm and the total crustal thickness d (4 km s≠1 < vsm <
6 km s≠1 in 0.5 km s≠1 steps, 4 km< d <12 km in 0.5 km steps, Fig. 6.11 b).
The properties of the sedimentary layer (vss and ds ) are kept constant at the
values estimated in the first modelling step. We select higher weights wp (Tl )
in eq. (6.13) for the longer period range. Moreover, we give higher weights
wp (Tl ) in eq. (6.13) to the shorter periods than the intermediate period range
(inset in Fig. 6.11 b), because the properties of the sedimentary layer have been
determined in the first modelling step. The vp /vs ratio is am = 1.8 which
is typical for oceanic mantle. Equation (6.12) is used to estimate flm . The
velocities and the density of the crust are kept at the values for NOC (Tab. 6.1).
Analogue to the first step, the minimum in the objective function R(vsm , d)
(eq. 6.13) is searched and the corresponding model serves as reference model
for the next step of the modelling.
In the third and last step, we search for the crustal S wave velocity vsc
(2.5 km s≠1 < vsc < 4.5 km s≠1 in 0.1 km s≠1 steps, Fig. 6.11 c). The values for
the S wave velocities of the sediment vss and the mantle vsm , as well as the
thickness of the sediment ds and the total crustal thickness d are kept constant
at the values resulting from the first two modelling steps. The weights wp (Tl )
in eq. (6.13) are equal (inset in Fig. 6.11 c). The P wave
Ô velocity of the crust
is estimated by using a standard vp /vs ratio of ac = 3 and the density flc is
calculated by using eq. (6.12). Similarly to the other steps, the minimum in
R(vsc ) (eq. 6.13) is searched and the resulting model is the best performing
model for this quantitative modelling.
6.5.3 Results of quantitative modelling

The results of the single modelling steps described in section 6.5.2 are shown for
station D03 as an example (Fig. 6.12). In the first step, as described before, we
search for the S wave velocity of the sediment vsc and its thickness ds by looking
for the minimum in the objective function R(vss , ds ) (Fig. 6.12 a). In case of
station D03, this results in a model with vss = 0.7 km s≠1 and ds = 0.6 km (red
cross in Fig. 6.12 a) which serves then as reference model for the next modelling
step.
In the second step, we vary the mantle S wave velocity vsm and the total
crustal thickness d (Fig. 6.12 b). For station D03, the value of R(vsm , d) is
larger than one for all parameter combinations (Fig. 6.12 b) meaning that
the reference model (ṽs,app
ref
) always performs better than each tested model
(ṽs,app ). We therefore keep the model with vss = 0.7 km s≠1 , ds = 0.6 km,
syn
93
0.2
6
0.4
8
0.6
0.8 10
1 12
0.5 1 1.5 2 4 4.5 5 5.5 6
0 0.5 1 1.5 2 0 2 4 6 8 10
NOC NM NOC NM
8
1
6
1
4
10
2
10
100
0
2.5 3 3.5 4 4.5 0.1 2 4 6 0.1 2 4 6
Figure 6.12: Example for modelling steps for station D03: (a) Ratio R (eq. 6.13)
shown in dependence on sediment S wave velocity vss and thickness ds . The reference
model is CM-REF (Tab. 6.3). The model with the smallest value of R (Rmin ) is
marked with a red cross. Additional models tested in step 2 are marked with yellow
circles. The area with R <= Rmin + 0.1 is outlined with black. (b) Ratio R is shown
in dependence on mantle S wave velocity vsm and total crustal thickness d, the best
fitting model of step one is chosen as reference model (red cross in a, vss = 0.7 km s≠1
and ds = 0.6 km). The cross marks Rmin and the area with R <= Rmin + 0.1 has a
black contour. (c) Ratio R is shown in dependence on crustal S wave velocity vsc ,
the best fitting model of step one is chosen as reference model (red cross in a, vss =
0.7 km s≠1 , ds = 0.6 km, vsm = 4.51 km s≠1 and d = 7 km). The red cross marks
Rmin and R = Rmin + 0.1 is indicated by a black line. (d) Weighted hit-count of the
velocity models resulting from all chosen reference models for all three steps (38 in
total, 36 for all reference models with R < Rmin + 0.1). The best performing model
with smallest value R(vsc ) in c is shown as red dash-dot line (vss = 0.7 km s≠1 , ds =
0.6 km, vsm = 4.51 km s≠1 , d = 7 km and vsc = 4.3 km s≠1 ). (e) Weighted hit-count
of corresponding vs,app profiles for all models in d in comparison to real data (orange
dash-dot line). The S wave velocities of normal oceanic crust (NOC) and normal
mantle (NM) are given as grey dashed lines in d and e.
vsm = 4.51 km s≠1 and d = 7 km as reference model for the last step of the
modelling.
In the third step, we look for the crustal S wave velocity vsc by searching the
minimum in R(vsc ) (Fig. 6.12 c). The result, which is found for R(vsc ) slightly
below 1, gives the best performing model for station D03 (vss = 0.7 km s≠1 ,
ds = 0.6 km, vsm = 4.51 km s≠1 , d = 7 km and vsc = 4.3 km s≠1 , Fig. 6.12 c).
94
Instead of concentrating just on the minimum value of the objective function

R (Rmin ), we give an arbitrary range of R which includes all models for which
R <= Rmin + 0.1 (black lines in Figs 6.12 a-c) is valid in order to get an idea of
the stability of the gained results and the sharpness of the obtained minimum
in R. Moreover, we choose additional reference models to repeat step two and
three of the modelling. These models represent alternative allowed parameter
combinations in R(vss , ds ) (e.g. yellow circles in Fig. 6.12 a) or R(vsm , d) (e.g.
black cross in Fig. 6.12 b). In the latter case, they are sometimes chosen outside
the Rmin + 0.1 range if R(vsm , d) shows several areas with small values close
to Rmin + 0.1. Furthermore, if the models with small R(vsm , d) values have
high mantle S wave velocities (>5.5 km s≠1 ) or a large total crustal thickness
(>10 km), we perform an additional test in the third modelling step with a
model which has NM properties and a total crustal thickness of 7 km and the
sedimentary properties resulting from step one.
In order to get an idea about the variability of the obtained models and to
present all models at once, we compare all vs,app syn
profiles of all models which
met the criteria R <= Rmin + 0.1 for each step of the quantitative modelling
to the real data vs,app
obs
profile with equal weights in all period ranges. For
this comparison, the objective function R (eq. 6.13) is calculated by using
the model CM-REF (5.05 km WC, 7 km NOC and NM above PREM) as
reference model. The resulting value of R is used to calculate a weighting
factor (1 ≠ R) for a weighted hit-count of the used velocity models. For the
visualisation (Fig. 6.12 d), we additionally weight the different layers according
to the different period ranges (insets in Fig. 6.11), e.g. for a model, which
results from the first step of the quantitative modelling, the sediment layer
is weighted with 20 as for the period range from 0.5 s to 2 s, the crust with
10 as for the period range from 2 s to 4 s and the mantle with 1 as for the
period range from 4 s to 16 s. The models resulting from the other steps are
treated in a similar way. This procedure is applied to reduce the biasing effect
of models with NM and NOC properties in the first two steps on the hit count.
The red dash-dot line in Fig. 6.12 d shows the model with vss = 0.7 km s≠1 ,
ds = 0.6 km, vsm = 4.51 km s≠1 , d = 7 km and vsc = 4.3 km s≠1 (Tab. 6.7)
which has the smallest value of R(vsc ) in Fig. 6.12 c. Furthermore, we utilize
the weighting factor (1 ≠ R) to estimate the weighted hit-count of the vs,app
profiles (Fig. 6.12 e).
It is clear by comparing Figs 6.12 a-b that if more models are lying within
the R < Rmin + 0.1 area, the minimum in the ratio R is less sharp. If a smaller
number of models fulfils this requirement, the resulting velocity models can
be regarded as better constrained by the available data. The total number
of models estimated for station D03 is 38. This number includes 36 models
which result from choosing reference models in step one and two within the
R < Rmin + 0.1 range and 2 models which result from an additional reference
model chosen in step two from a separate area in R with small R values close
to the R < Rmin + 0.1 range.
We repeat all three steps of the modelling for each station (Fig. 6.10) and
present the obtained results (velocity models and their corresponding vs,app
profiles, as well as number of models) in Fig. 6.13 and the parameters of the
best performing models in Tab. 6.7.
95
NOC NM NOC NM NOC NM NOC NM

0.1 0.1
1 1
1 1
10 10
10 10
100 100
0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8

0.1 0.1
1 1
1 1
10 10
10 10
100 100
0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8

0.1 0.1
1 1
1 1
10 10
10 10
100 100
0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8

0.1 0.1
1 1
1 1
10 10
10 10
100 100
0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8

0.1 0.1
1 1
1 1
10 10
10 10
100 100
0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8 0.1 2 4 6 8
0 2 4 6 8 10+
Figure 6.13: Weighted hit-count for velocity models and their corresponding vs,app
profiles for all stations besides D03. Same as Fig. 6.12 d and e. Station names are
given in bold. The number of models for which R < Rmin + 0.1 is valid are given for
each station: those with reference models which fulfil R < Rmin + 0.1 in the previous
modelling step and in brackets those for all used reference models.
96
Table 6.7: Model parameters for best performing models for each station as presented
in Figs 6.12 and 6.13: S wave velocity of sediment vss , mantle vsm and crust vsc ,
and sediment thickness ds and total crustal thickness d. Poorly resolved model
parameters are indicated by an asterisk.
station vss ds vsm d vsc

D01 1.2 0.8 5.3 4.5 3.5
D02 1.0 0.7 5.2 7.0 3.75
D03 0.7 0.6 4.51 7.0 4.3
D04 1.2 0.8 4.8 5.0 4.1
D06 0.9 0.4 5.3 10.0* 3.8
D07 1.4 0.3 5.9* 5.5 4.1
D08 0.8 0.3 4.7 9.0* 3.6
D09 1.0 0.7 4.0* 7.0 3.9
D10 1.4* 0.3* 5.0 6.5 3.75
D11 1.2 0.9 5.5* 4.0 3.7
D12 0.9 0.8 4.6 5.5 3.75
97
6.6 Discussion
6.6.1 Discussion of quantitative modelling results
The best performing models at most stations show three distinct solid layers
(Figs 6.12 d and 6.13). The stations D03 and D09 are exceptions to this overall
trend. At these stations the crustal and the mantle S wave velocities are
quite similar (Figs 6.12 d and 6.13 g). For station D03, we have a crustal S
wave velocity of 4.3 km s≠1 which is closer to typical mantle S wave velocities.
The weighted hit-count plot for all models within the R < Rmin + 0.1 range
(Fig. 6.12 d) shows that the crustal S wave velocity varies between 3.5 km s≠1
and 4.3 km s≠1 and the mantle S wave velocity lies in the range of 4.5 km s≠1
to 4.9 km s≠1 . The velocities of the best performing velocity model therefore
represent the fastest crustal and the slowest mantle S wave velocities at station
D03. On the other hand, we get a mantle S wave velocity of 4.0 km s≠1 for
station D09 which is closer to a typical oceanic crustal velocity (Fig. 6.13 g).
This can be explained by the fact that for station D09, our result only relies on
a vs,app profile obtained from a single event in the period range between 0.5 s
and 8 s. It therefore is possible that the data set at station D09 is insufficient
to constrain the uppermost mantle S wave velocity at this station and just the
sedimentary and crustal S wave velocity are properly estimated.
D11 D07 D06 D04 D01 D03 D02 D08 D09
10
15
38.6 38.5 38.4 38.3 38.2 38.1 38
D10 D08 D07 D03 D04 D01 D02 D06 D12

water
5 sediment
crust
10
15
uppermost mantle
-18.8 -18.7 -18.6 -18.5 -18.4 -18.3 -18.2 -18.1 -18
0 1 2 3 4 5 6
Figure 6.14: S wave velocity depth profiles in (a) N-S and (b) W-E direction for
best performing models as presented in Figs 6.12 e and 6.13. Models are labelled
with station names and water depth are taken from the ship echo-sounder. Poorly
resolved model parameters mainly influenced by noise are superimposed with grey
overlays.
98
6.6. Discussion
At the stations D07 and D11, we probably observe the influence of noise on
the longer periods (>4 s) of the median vs,app profiles which leads to either higher
vs,app estimates (D07, Fig. 6.10 f) or different velocity estimates for individual
events (D11, Fig. 6.10 j). In both cases, the influence of noise on the longer
periods (>4 s) probably causes that the uppermost mantle S wave velocities
estimated by the quantitative modelling are rather high (Figs 6.13 e and i).
The median vs,app profiles at stations D06 and D08 show kinks and jumps
at ≥4 s. The discontinuity in the profiles probably leads to biased estimates
of the total crustal thickness. This becomes evident considering that at both
station, models with large crustal thicknesses (9-10 km, Figs 6.13 d and f) are
favoured by the quantitative modelling.
Station D10 also shows a jump in its median vs,app profile at 0.7 s (Fig. 6.10 i)
which is rather small compared to station D06. An influence on the estimated
sedimentary model parameters might be possible. The overall appearance of
the short periods (<2 s) of the vs,app profiles at station D10 and D07 are similar
and also the obtained sedimentary model parameters are in good agreement
(Figs 6.13 e and h), which indicates only a minor influence of the discontinuous
vs,app profile at station D10 on the obtained sedimentary models.
At station D04, the comparison of the median vs,app profile and the modelled
profiles shows slightly higher velocities in the period range between 2 s to 4 s for
the real data (Fig. 6.13 c). These elevated velocities are probably related to the
effect of the before mentioned resonance. Nevertheless, the good resemblance
of the asymptotic behaviour of the median vs,app profile at the short and long
periods by the modelled profiles (Fig. 6.13 c) assures us that the obtained models
are reliable.
In order to consider all discussed effects at the single stations, poorly
resolved model parameters are indicated by either superimposed grey shadings
in Fig. 6.14 or lighter colours in Fig. 6.15.
In Fig. 6.14, we present a N-S and a W-E profile in which the best performing
models are shown as columns. The sediment layers can be divided into two
groups. The first group has a thickness between 0.3 km and 0.4 km (D06,
D07, D08, D10) and the second between 0.6 km and 0.9 km (Fig. 6.14 and
Tab. 6.7). The S wave velocities in the sediments vary between 0.7 km s≠1
and 1.4 km s≠1 (second layer in Fig. 6.14). We observe no clear correlation
between water depth or location and sediment thickness or S wave velocity. The
different sediment thicknesses and varying S wave velocities might be related
to a rough basement topography with sediment filled depressions as can be
seen on analog seismic recordings of the area under investigation which were
recorded in 1969 (NOAA World Data Service for Geophysics, Marine Seismic
Reflection, Survey ID V2707, ngdc.noaa.gov). Moreover, we can relate the
resonance observed at the OBS to the properties of the sedimentary layer. If
we take the sediment thickness ds = 0.6 km and the sediment S wave velocity
vss = 0.7 km s≠1 at station D03 and a relation for the resonance frequency of
sediments fr = vss /(4 · ds ) (Parolai et al., 2002), we get a resonance period of
3.4 s which is similar to the observed period of the resonance at station D03
(≥3 s, Fig. 6.9). The periods obtained for the other stations also agree quite
well with the observed periods of the resonance at each station.
The uppermost mantle S wave velocities vsm clearly decrease towards the
South (lower layer in Fig. 6.14, Fig. 6.15 a). In Fig. 6.15 a, we present all
estimated vsm which fulfil the criteria R < Rmin + 0.1 in the second step of the
99
120120 100100 80 80 60 60
6.0
5.5
5.0
4.5
4.0
38.638.6 38.438.4 38.238.2 38.038.0
38.638.6 38.438.4 38.238.2 38.038.0
12
10
120120 100100 80 80 60 60
Figure 6.15: Estimated model parameters of second modelling step along N-S profile
(Figs 6.11 b and 6.14 a) and against distance towards Gloria fault. All models fulfil
R < Rmin in the second quantitative modelling step. Analysing the parameters by
assuming a linear trend gives the grey dashed lines. Poorly resolved model parameters
are shown in lighter colours. (a) Uppermost mantle S wave velocity vsm . (b) Total
crustal thickness d.
quantitative modelling. Assuming a linear trend gives a decrease of ≥23 m s≠1

per km along the N-S profile. We interpret this decrease in uppermost mantle
S wave velocity as the influence of the Gloria Fault which lies South of the
OBS array (Fig. 6.8). This decrease might be related to either biasing effects
of anisotropy or out of plane arrivals, or changes in rheology, hydration or
melt fraction. The azimuthal coverage of our data is too limited to allow
any conclusions whether the observed decrease in S wave velocity towards the
fault is caused by anisotropy which might be present close to a major plate
boundary like the Gloria fault (e.g. San Andreas fault, Ozacar and Zandt,
2009). Furthermore, we exclude any influence by out of plane arrivals due
to the small deviations (<3¶ ) in back azimuth of the analysed events from
theoretical values identified using a frequency-wave number analysis (Rost and
Thomas, 2002) on the vertical components of the OBS, and the similar results
of the orientations using P wave polarization and Rayleigh wave characteristics
(Stachnik et al., 2012). On the other hand, a reduction in the S velocity can
also be caused by partial melt, different water contents, serpentinization or a
change in grain sizes (Horen et al., 1996; Jung, 2001; Faul and Jackson, 2005;
Stixrude and Lithgow-Bertelloni, 2005; Karato, 2012). Karato (2012) suggests
100
6.6. Discussion
that the amount of partial melt generated away from mid-ocean ridges is small
and that the seismic velocity reduction due to small amounts of melt (<0.1%) is
small if grain boundaries are not completely wetted by the melt. In the vicinity
of the Gloria Fault, fractures can be generated which can serve as pathways
for the seawater to penetrate the oceanic lithosphere. This can change the
fabric of the mantle minerals (Jung, 2001) or leads to serpentinization (Fryer,
2002). Both processes result in lower seismic velocities (Horen et al., 1996;
Jung, 2001). Fryer (2002) (and references therein) also gives an example in
which the degree of serpentinization increases towards a shear zone, in-line with
our observation of decreasing velocities towards the plate boundary (Horen
et al., 1996). Another common feature of shear zones is the reduction of grain
size in high shear strain areas by the formation of mylonites (Bürgmann and
Dresen, 2008). The grain size reduction causes a decrease in the shear modulus
which is directly related to a decrease in seismic velocities (Faul and Jackson,
2005). This effect may be a direct indication of an increase in shear strain in the
direction of the plate boundary (Bürgmann and Dresen, 2008). In conclusion,
the observed velocity decrease towards the Gloria Fault probably results from
the combined effects of serpentinization, an increase in water content and grain
size reduction towards the fault. The serpentinization is probably the strongest
of all effects discussed here, because of the abundance of serpentinite close to
fractures in the oceanic crust and mantle (White et al., 1992).
The crustal S wave velocities vsc (third layer in Fig. 6.14) are quite similar
across the array (3.5 km s≠1 to 4.3 km s≠1 ). The total crustal thickness d
varies for most stations between 4 km and 7 km and shows larger values at
station D08 (9 km) and D06 (10 km). The thickness d shows no clear trend
from the western to the eastern end of the array (Fig. 6.14). This behaviour is
reasonable given that the difference in the crustal ages between the western
station D10 (≥76.4 Ma, Müller et al., 2008) and the eastern station D12
(≥82.9 Ma, Müller et al., 2008) are quite small. In contrast, we observe a
possible slight increase in the total crustal thickness towards the South, i.e.
towards the Gloria Fault (grey dashed line in Fig. 6.15 b and Fig. 6.8) although
the differences in crustal age are similar to the W-E direction. In Fig. 6.15 b,
all total crustal thicknesses d, which fulfil the criteria R < Rmin + 0.1 in step
two of the quantitative modelling, are shown as a function of distance along
the N-S profile. The weak trend is still visible, although station D06 shows
larger thicknesses than would be expected by the weak trend and station D09
has a large scatter in the estimated thickness. In the bathymetry (Fig. 6.8),
nearly NE-SW striking bathymetric heights are visible which are an evidence
for the known transpression in this region (Zitellini et al., 2009, fig. 6). This
mechanism leads to a shortening in the crust which is larger in the vicinity of
the fault. We interpret the gradual thickening of the estimated crustal thickness
from N to S with the gradual shortening of the crust towards the Gloria fault.
6.6.2 Discussion of method

The proposed method relies on the proper estimation of the apparent P wave
incidence angle using RF which implies that data quality is essential especially
in case of OBS at which only few event recordings are available (Webb, 1998).
In order to exclude recordings or period ranges which are highly influenced
by noise or resonance effects (e.g. Fig. 6.9), a careful data review is necessary.
101
Some vs,app profiles might be still influenced by noise especially at long periods
(e.g. D07, >4 s, Fig. 6.10 f and supplementary Fig. 6.S2) or by resonance effects
at intermediate periods (D04, Fig. 6.10 d). This needs to be considered during
the interpretation of modelling results. Moreover, we notice that the behaviour
of the vs,app at longer periods might be influenced by long period global phases
(e.g. W phase, Kanamori, 1993) or whether the deconvolution filter is correctly
determined at these periods which is influenced by the Wiener filter parameters
(supplementary, Fig. 6.S3). To ensure that apparent P wave incidence angle
measurement with a sufficient data quality are considered, we employ the SNR
of the RF as criteria to select the data (SNR>4 at ZRF and RRF ) and as
weighting in eqs (6.11) and (6.13).
We also have to consider the corner period of the event’s source spectrum
and exclude all periods shorter than the event’s corner period from the analysis.
Additionally, the effect of the employed filter on the P spike might have an
influence on the appearance of the vs,app profile. Both, the P spike of a RF and
the impulse response of a forward and backward applied low pass filter can
be regarded as having a similar bell shape like a Gaussian function. Keeping
this in mind, the convolution of a P spike with a corner period Tcp and an
impulse response of a zero phase low pass filter with a corner period Tcf would
result in a signal with an actual corner period Tca 2
= Tcp
2
+ Tcf
2
. This effect is
important for actual corner periods Tca smaller than 2-3 times Tcp . We tested
the correction of the corner periods for our first quantitative modelling step
and found similar results as presented in section 6.5. We conclude from this
that the corner periods of the real and the synthetic data are similar enough
that the effect of the filtering can be considered to be the same and therefore
has not to be corrected.
In our case, all used ZRF have an approximate bell shape. If an observed
ZRF would oscillate and deviate from a simple bell shape, the quantitative
modelling presented here can still be applied by convolving the modelled RF
with the observed ZRF (Schiffer et al., 2015) before applying the low pass filters
and comparing the vs,app profiles.
We use an arbitrary vp /vs ratio for all three steps in the quantitative
Ô
modelling of the OBS data (eq. 6.14 for the sediments, vp /vs = 3 for the
crust and vp /vs = 1.8 for the mantle) to estimate the P wave velocity vp and
the density fl. If we replace our vp /vs ratio for the sediments (eq. 6.14) with
the mud-rock line (vp = 1.16 · vs + 1.36, Castagna et al., 1985), the obtained
results for the first step of the quantitative modelling are similar to the results
presented here. This underlines that the apparent P wave incidence angle Ïp is
independent of the P wave velocity vp of the ocean bottom which was already
indicated by our relation in eq. 6.10, and the weak dependence of Ïp on the
density fl of the ocean bottom as already shown in section 6.4.
6.7 Conclusion and Outlook

We derive the relations for the apparent P wave incidence angle at the ocean
bottom in dependence on the elastic parameters, the different reflection and
refraction angles (eq. 6.9) and the slowness (eq. 6.10), respectively. The second
relation reveals that the apparent P wave incidence angle is independent of the
P wave velocity of the ocean bottom and theoretically depends on the apparent
102
6.7. Conclusion and Outlook
half space S wave velocity and the density of the ocean bottom. We show by
employing synthetic half space velocity models that the dependence on the
S wave velocity is much stronger than on the density (Fig. 6.3). Moreover,
we observe a clear dependence of the obtained apparent S wave velocities on
the corner periods of the used low pass filters (Fig. 6.4). This confirms the
hypothesis of Phinney (1964) that obtained seismic velocities from the analysis
of the P wave polarisation depend on the used period range. Furthermore, it is
in good agreement with the observations made for land stations by Svenningsen
and Jacobsen (2007) and Kieling et al. (2011). The investigation of the influence
of different parameters of a layered subsurface model on the appearance of the
vs,app profiles leads to the conclusion that a velocity increase and decrease can
be identified and that the observed ’overshoot’ in the vs,app profiles is influenced
by the layer thickness. Moreover, the vs,app profiles show the tendency that the
shorter periods converge towards the velocity of the uppermost layers and the
longer periods toward the velocity of deeper layers. Besides a small influence
of the water depth on the appearance of the vs,app profiles, we find that the
removal of water multiples might be useful for water depth shallower than
≥1 km (Fig. 6.6, Osen et al., 1999; Thorwart and Dahm, 2005) in order to
prevent a distortion of the obtained results. We find from our synthetic test
that the uppermost mantle shear wave velocity can be identified. However, low
velocity layers deeper than ≥50 km are hardly resolvable (Fig 6.7).
We analyse OBS data which were recorded in the eastern mid Atlantic
≥60 to 135 km North of the Gloria fault. We use a grid search for S wave
velocities in a layered Earth model for the quantitative modelling of the real
ocean bottom data. Additionally, we identify a resonance signal with dominant
periods between 0.5 s and 4 s as a main reason for amplitude distortions and
therefore erroneous apparent P wave incidence angles. Events showing a clear
influence of this resonance phenomenon on the obtained angles are removed
from the analysis.
Overall, we find that models consisting of three solid layers (sediment, crust
and uppermost mantle) over PREM b.s.f. are well suited to model the real
data. The sediments can be grouped in two thickness ranges (0.3-0.4 km and
0.6-0.9 km), there is no water depth dependency which is in good agreement to
known seismic profiles of the area. Furthermore, we find that the uppermost
mantle shear wave velocity decreases towards the Gloria Fault. Additionally,
we observe a crustal thickening towards the fault which agrees well with the
known transpressive character of the plate boundary (Zitellini et al., 2009). We
suggest that the decrease in uppermost mantle shear velocity and the crustal
thickening towards the fault is deformation related.
The presented method of S wave velocity estimation using P wave polariza-
tion can be combined with surface wave methods and/ or receiver functions in a
joint inversion (e.g. Du and Foulger, 1999; Julià et al., 2000; Schiffer et al., 2015)
in order to better constrain the estimated model parameters. The obtained S
wave velocities can also be useful for further analysis of the experimental area.
Receiver functions can be used to estimate Moho depth and the vp/vs ratio
of the crust by employing an amplitude stack of the direct crustal phase and
its multiples (Zhu and Kanamori, 2000). Usually, the needed delay times of
the direct phase and the multiples are calculated by giving a P wave velocity
for the crust. An alternative approach might be that the obtained S wave
103
Bibliography
velocities of the apparent P wave incidence angle analysis are used instead of
an arbitrary P wave velocity.
In conclusion, the proposed method is usable for single station estimates
of the local S wave velocity structure beneath the ocean bottom. The im-
plementation of the method is easy and the processing is performed in two
steps:
(1). measurement of apparent P wave incidence angles in different period

ranges for real and synthetic data
(2). comparison of determined apparent S wave velocity profiles for real and
synthetic data.
The second step leads to the estimation of S wave velocity-depth models.
Acknowledgments
The first authors thanks the Niedersächsische Staats- und Universitätsbib-
liothek in Göttingen for providing the digital copy of Wiechert (1907) and
Zoeppritz (1919), as well as the Thüringer Universitäts- und Landesbiblitothek
for providing the digital copy of Knott (1899). The first author thanks Uwe
Altenberger for discussing probable reasons for the change in the uppermost
mantle velocities and Dietrich Lange for fruitful discussions during the review of
the manuscript. The authors thank EMEPC (Task Group for the Extension of
the Continental Shelf) for providing the bathymetric data and Luis Batista for
sharing them. The authors are grateful to Bo Holm Jacobsen and an anonymous
reviewer for their comments which helped to improve the manuscript. The
data processing was partly done using Seismic Handler (Stammler, 1993). The
power spectral densities were obtained with obsPy (Beyreuther et al., 2010).
Some figures were created using GMT (Wessel et al., 2013). The authors thank
the DEPAS pool for providing the instruments for the DOCTAR project which
was funded by the DFG (KR1935/13, DA 478/21-1) and by the Leitstelle für
Mittelgroße Forschungsschiffe (Poseidon cruises 416 and 431).
Bibliography
Books, 2nd edition.
Ammon, C. J. (1991). the Isolation of Receiver Effects From Teleseismic P

Waveforms. Bulletin of the Seismological Society of America, 81(6):2504–
2510.
Ben-Menahem, A. and Singh, S. J. (2012). Seismic Waves and Sources. Springer

Verlag New York Heidelberg Berlin.
Beyreuther, M., Barsch, R., Krischer, L., Megies, T., Behr, Y., and Wassermann,
J. (2010). ObsPy: A Python Toolbox for Seismology. Seismological Research
Letters, 81(3):530–533.
104
Bibliography

Brocher, T. M. (2005). Empirical relations between elastic wavespeeds and

density in the Earth’s crust. Bulletin of the Seismological Society of America,
95(6):2081–2092.
Castagna, J. P., Batzle, M. L., and Eastwood, R. L. (1985). Relationships

between compressional-wave and shear-wave velocities in clastic silicate rocks.
Geophysics, 50(4):571–581.

Research, 103(B5):9895–9916.
Dahm, T., Tilmann, F., and Morgan, J. P. (2006). Seismic broadband ocean-
bottom data and noise observed with free-fall stations: Experiences from
long-term deployments in the North Atlantic and the Tyrrhenian Sea. Bulletin
of the Seismological Society of America, 96(2):647–664.
Du, Z. J. and Foulger, G. R. (1999). The crustal structure beneath the northwest
fjords, Iceland, from receiver functions and surface waves. Geophysical Journal

Faul, U. H. and Jackson, I. (2005). The seismological signature of temperature

and grain size variations in the upper mantle. Earth and Planetary Science
Letters, 234(1-2):119–134.
Fryer, P. B. (2002). Recent Studies of Serpentinite Occurrences in the Oceans:

Mantle-Ocean Interactions in the Plate Tectonic Cycle. Chemie der Erde -
Geochemistry, 62(4):257–302.
Gao, H. and Shen, Y. (2015). A Preliminary Full-Wave Ambient-Noise Tomogra-

phy Model Spanning from the Juan de Fuca and Gorda Spreading Centers to
the Cascadia Volcanic Arc. Seismological Research Letters, 86(5):1253–1260.
Haskell, N. A. (1960). Crustal reflection of plane SH waves. Journal of

Geophysical Research, 65(12):4147–4150.
Horen, H., Zamora, M., and Dubuisson, G. (1996). Seismic waves velocities and
anisotropy in serpentinized peridotites from xigaze ophiolite: Abundance of
serpentine in slow spreading ridge. Geophysical Research Letters, 23(1):9.
105
Bibliography
Julià, J. (2007). Constraining velocity and density contrasts across the crust-
mantle boundary with receiver function amplitudes. Geophysical Journal
Julià, J., Ammon, C. J., Herrmann, R. B., and Correig, A. M. (2000). Joint
inversion of receiver function and surface wave dispersion observations. Geo-
physical Journal International, 143(1):99–112.
Jung, H. (2001). Water-Induced Fabric Transitions in Olivine. Science,
293(5534):1460–1463.
Kanamori, H. (1993). W Phase. Geophysical Research Letters, 20(16):1691–
1694.
Kennett, B. L. N., Engdahl, E. R., and Buland, R. (1995). Constraints
on seismic velocities in the Earth from traveltimes. Geophysical Journal
Kieling, K., Rössler, D., and Krüger, F. (2011). Receiver function study in
northern Sumatra. Journal of Seismology, 15(2):235–259.
Kind, R., Kosarev, G. L., and Petersen, N. V. (1995). Receiver functions at
the stations of the German Regional Seismic Network (GRSN). Geophysical
Knott, C. G. (1899). Reflection and refraction of elastic waves with seismological
application. Philosophical Magazine Series 5, 48(290):64–97.
Krüger, F. (1994). Sediment structure at GRF from polarization analysis of P
waves of nuclear explosions. Bulletin of the Seismological Society of America,
84(1):149–170.
Kumar, P., Sen, M. K., and Haldar, C. (2014). Estimation of shear veloc-
ity contrast from transmitted Ps amplitude variation with ray-parameter.
Maupin, V. (2011). Upper-mantle structure in southern Norway from beam-
forming of Rayleigh wave data presenting multipathing. Geophysical Journal
McNamara, D. E. (2004). Ambient Noise Levels in the Continental United
States. Bulletin of the Seismological Society of America, 94(4):1517–1527.
Müller, G. (1985). The reflectivity method: a tutorial. Journal of Geophysics,
58(1-3):153–174.
106
Bibliography
Nguyen, X. N., Dahm, T., and Grevemeyer, I. (2009). Inversion of Scholte wave
dispersion and waveform modeling for shallow structure of the Ninetyeast
Ridge. Journal of Seismology, 13(4):543–559.
Nuttli, O. (1961). The effect of the earth’s surface on the S wave particle
motion. Bulletin of the Seismological Society of America, 51(2):237–246.
Nuttli, O. and Whitmore, J. D. (1962). On the determination of the polarization

angle of the S wave. Bulletin of the Seismological Society of America, 52(1):95–
107.
Osen, A., Amundsen, L., and Reitan, A. (1999). Removal of water-layer

multiples from multicomponent sea-bottom data. Geophysics, 64(3):838.
Ozacar, A. A. and Zandt, G. (2009). Crustal structure and seismic anisotropy

near the San Andreas Fault at Parkfield, California. Geophysical Journal
Parolai, S., Bormann, P., and Milkereit, C. (2002). New relationships between
Vs, thickness of sediments, and resonance frequency calculated by the H/V
ratio of seismic noise for the cologne area (Germany). Bulletin of the
Seismological Society of America, 92(6):2521–2527.
Peterson, J. (1993). Observations and modeling of seismic background noise.

Technical report, U.S. Department of Interior Geological Survey, Albuquerque,
New Mexico.
Phinney, R. A. (1964). Structure of the Earth’s crust from spectral behavior of

long-period body waves. Journal of Geophysical Research, 69(14):2997–3017.
Pilant, W. L. (1979). Elastic waves in the earth, Volume 11. Elsevier Scientific
Publishing Company.
Prakash, K. (2015). Estimation of shear velocity contrast for dipping or

anisotropic medium from transmitted Ps amplitude variation with ray-
parameter. Geophysical Journal International, 203(3):2248–2260.
Ritzwoller, M. H. and Levshin, A. L. (2002). Estimating shallow shear velocities

with marine multicomponent seismic data. Geophysics, 67(6):1991–2004.
Romanowicz, B. (2003). Global mantle tomography: progress status in the past

10 years. Annual Review of Earth and Planetary Sciences, 31(1):303–328.

Scherbaum, F. (2007). Of poles and zeros. Fundamentals of Digital Seismology,

volume 15. Springer, Dordrecht, Netherlands, 2nd edition.
Schiffer, C., Jacobsen, B. H., Balling, N., Ebbing, J., and Nielsen, S. B. (2015).
The East Greenland Caledonides—teleseismic signature, gravity and isostasy.
107
Bibliography
83(4):704–713.
41:43–63.
Stixrude, L. and Lithgow-Bertelloni, C. (2005). Thermodynamics of man-
tle minerals - I. Physical properties. Geophysical Journal International,
162(2):610–632.
Sumy, D. F., Lodewyk, J. A., Woodward, R. L., and Evers, B. (2015). Ocean-
Bottom Seismograph Performance during the Cascadia Initiative. Seismolog-
ical Research Letters, 86(5):1238–1246.
Thorwart, M. and Dahm, T. (2005). Wavefield decomposition for passive ocean
bottom seismological data. Geophysical Journal International, 163(2):611–
621.
Wang, R. (1999a). A simple orthonormalization method for stable and efficient
computation of Green’s functions. Bulletin of the Seismological Society of
America, 89(3):733–741.
Wang, Y. (1999b). Approximations to the Zoeppritz equations and their use in
AVO analysis. Geophysics, 64(6):1920.
Webb, S. C. (1998). Broadband seismology and noise under the ocean. Reviews
Weidle, C. and Maupin, V. (2008). An upper-mantle S-wave velocity model
for Northern Europe from Love and Rayleigh group velocities. Geophysical
Wessel, P., Smith, W. H. F., Scharroo, R., Luis, J., and Wobbe, F. (2013).
Generic Mapping Tools: Inproved version released. Eos, Transactions Amer-
ican Geophysical Union, 94(45):409–410.
Wiechert, E. (1907). Über Erdbebenwellen. Part I: Theoretisches über die
Ausbreitung der Erdbebenwellen. Nachrichten von der Gesellschaft der
Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse, (1907):415–
529.
108
Bibliography
Yamamoto, T. and Torii, T. (1986). Seabed shear modulus profile inversion

using surface gravity (water) wave-induced bottom motion. Geophysical
Journal - Royal Astronomical Society, 85(2):413–431.
Zhu, L. and Kanamori, H. (2000). Moho depth variation in southern Cali-

fornia from teleseismic receiver functions. Journal of Geophysical Research,
105(B2):2969–2980.
Zitellini, N., Gràcia, E., Matias, L., Terrinha, P., Abreu, M. A., De Alteriis,
G., Henriet, J. P., Dañobeitia, J. J., Masson, D. G., Mulder, T., Ramella,
R., Somoza, L., and Diez, S. (2009). The quest for the Africa-Eurasia plate
boundary west of the Strait of Gibraltar. Earth and Planetary Science Letters,
280(1-4):13–50.
Zoeppritz, K. (1919). Über Erdbebenwellen Part VII b . Über Reflexion

und Durchgang seismischer Wellen durch Unstetigkeitsflächen. Nachrichten
von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-
Physikalische Klasse, (1919):66–84.
109
Bibliography
6.S1 Derivation of equation (3.10)

Taking equation (6.3) from the paper:
1 2
1 + Ṕ P̀ sin Ïp1 + Ṕ S̀ cos Ïs
tan Ïp = 1 2 . (6.3)
1 ≠ Ṕ P̀ cos Ïp1 + Ṕ S̀ sin Ïs
and inserting equations (6.4) and (6.5) from the paper:
1 ≠ f (1 ≠ g)
Ṕ P̀ = (6.4)
1 + f (1 + g)
4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs )
Ṕ S̀ = (6.5)
1 + f (1 + g)
vp1 fl1 cos Ïp2 cos2 (2Ïs )
with: f = (6.6)
vp2 fl2 cos Ïp1
A B2
vs sin (2Ïp1 ) tan (2Ïs )
g= . (6.7)
vp1 cos (2Ïs )
gives:
1 2 4v s
v fl1
sin Ïp1 cos Ïp2 cos(2Ïs )
1≠f (1≠g)
1+ 1+f (1+g)
sin Ïp1 + p2 fl2
1+f (1+g)
cos Ïs
tan Ïp = 1 2 4 v s fl1
v fl
sin Ïp1 cos Ïp2 cos(2Ïs )
1≠f (1≠g)
1≠ 1+f (1+g)
cos Ïp1 + p2 2
1+f (1+g)
sin Ïs
We notice that multiplying numerator and denominator with 1 + f (1 + g)

results in:
(1 + f (1 + g) + (1 ≠ f (1 ≠ g))) sin Ïp1 + 4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs ) cos Ïs
tan Ïp =
(1 + f (1 + g) ≠ (1 ≠ f (1 ≠ g))) cos Ïp1 + 4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs ) sin Ïs
2 (1 + f g) sin Ïp1 + 4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs ) cos Ïs
tan Ïp = (6.S1)
2f cos Ïp1 + 4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs ) sin Ïs
Now inserting equations (6.6) and (6.7) from the paper, in order to keep
the equations less busy, we first have a look on 2 (1 + f g) sin Ïp1 :
Q A B2 R
vp1 fl1 cos Ïp2 cos2 (2Ïs ) vs sin (2Ïp1 ) tan (2Ïs ) b
2 (1 + f g) sin Ïp1 = 2 a1 + sin Ïp1
vp2 fl2 cos Ïp1 vp1 cos (2Ïs )
A B
vs fl1 vs cos Ïp2 cos (2Ïs ) sin (2Ïp1 ) tan (2Ïs )
=2 1+ sin Ïp1
vp2 fl2 vp1 cos Ïp1
sin(2Ïs )
Using tan (2Ïs ) = cos(2Ïs )
and double-angle formulae sin (2Ïp1 ) = 2 sin Ïp1 cos Ïp1 :
Q sin(2Ïs )
R
vs fl1 vs cos Ïp2 cos (2Ïs ) 2 sin Ïp1 cos Ïp1 cos(2Ïs ) b
2 (1 + f g) sin Ïp1 = 2 a1 + sin Ïp1
vp2 fl2 vp1 cos Ïp1
A B
vs fl1 vs
2 (1 + f g) sin Ïp1 =2 1+2 cos Ïp2 sin Ïp1 sin (2Ïs ) sin Ïp1 (6.S2)
vp2 fl2 vp1
110
6.S1. Derivation of equation (3.10)
Now inserting equations (6.S2) and (6.6) in equation (6.S1):
1 2
2 1 + 2 vvp2s flfl12 vvp1s cos Ïp2 sin Ïp1 sin (2Ïs ) sin Ïp1
tan Ïp = 2
fl1 cos Ïp2 cos (2Ïs )
2 vvp1
p2 fl2 cos Ïp1
cos Ïp1 + 4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs ) sin Ïs
4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs ) cos Ïs
+ 2
fl1 cos Ïp2 cos (2Ïs )
2 vvp1
p2 fl2 cos Ïp1
cos Ïp1 + 4 vvp2s flfl12 sin Ïp1 cos Ïp2 cos (2Ïs ) sin Ïs
1
multiply numerator and denominator with
2 sin Ïp1 cos Ïp2
1
cos Ïp2
+ 2 vp2 fl2 vp1 sin Ïp1 sin (2Ïs ) + 2 vp2 fl2 cos (2Ïs ) cos Ïs
v s fl1 v s v s fl1
tan Ïp = vp1 fl1 cos2 (2Ïs )
vp2 fl2 sin Ïp1
+ 2 vvp2s flfl12 cos (2Ïs ) sin Ïs
sin Ïp1 sin Ïs

Using Snell’s law vp1
= vs
:
1
cos Ïp2
+ 2 vvp2s flfl12 sin Ïs sin (2Ïs ) + 2 vvp2s flfl12 cos (2Ïs ) cos Ïs
tan Ïp = vs fl1 cos2 (2Ïs )
+ 2 vvp2s flfl12 cos (2Ïs ) sin Ïs
vp2 fl2 sin Ïs
vp2 fl2 sin Ïs
multiply numerator and denominator with
vs fl1 cos (2Ïs )
sin Ïs sin(2Ïs )
vp2 fl2
vs fl1 cos Ïp2 cos(2Ïs )
+ 2 sin2 Ïs cos(2Ï s)
+ 2 cos Ïs sin Ïs
tan Ïp =
cos (2Ïs ) + 2 sin2 Ïs
Using Snell’s law sinvsÏs = sinvp2

Ïp2
and double angle formulas
sin (2 Ïs ) = 2 sin Ïs cos Ïs and cos (2Ïs ) = 1 ≠ 2 sin2 Ïs
sin Ïp2 sin(2Ïs )

fl2
fl1 cos Ïp2 cos(2Ïs )
+ 2 sin2 Ïs cos(2Ï s)
+ sin (2 Ïs )
tan Ïp =
1 ≠ 2 sin2 Ïs + 2 sin2 Ïs
sin Ïp2
with tan Ïp2 =
cos Ïp2
1 2
fl2 tan Ïp2 2 sin2 Ïs
fl1 cos(2Ïs )
+ sin (2Ïs ) cos(2Ïs )
+1
=
1 A B
fl2 tan Ïp2 2 sin2 Ïs + cos (2Ïs )
= + sin (2Ïs )
fl1 cos (2Ïs ) cos (2Ïs )
A B
fl2 tan Ïp2 2 sin Ïs + 1 ≠ 2 sin2 Ïs
2
= + sin (2Ïs )
fl2 tan Ïp2 sin (2Ïs )
= +
fl2 tan Ïp2
tan Ïp = + tan (2Ïs ) (6.9)
fl1 cos (2Ïs )
To get equation (6.10) in the paper, we need Snell’s law sin Ïx = vx · p:
111
Bibliography
-
- 1
sin Ïp2 sin Ïp2 vp2 · p - vp2
tan Ïp2 = =Ò =Ò -·
- 1
cos Ïp2 1 ≠ sin2 Ïp2 1 ≠ vp2
2
· p2 - vp2
p
tan Ïp2 =Ú (6.S3)
1
v2
≠ p2
p2
cos (2Ïs ) = 1 ≠ 2 sin2 Ïs = 1 ≠ 2p2 vs2

A B
1
cos (2Ïs ) = vs2 ≠ 2p2 (6.S4)
vs2
Ò
sin (2Ïs ) 2 sin Ïs cos Ïs 2 sin Ïs 1 ≠ sin2 Ïs
tan (2Ïs ) = = =
cos (2Ïs ) cos (2Ïs ) cos (2Ïs )
Ò Ò
1
2pvs 1 ≠ p2 vs2 vs2 2p vs2
≠ p2
= 1
1
2 = 1
1
2
vs2 vs2
≠ 2p2 vs2 vs2
≠ 2p2
Ò
1
2p vs2
≠ p2
tan (2Ïs ) = 1 (6.S5)
vs2
≠ 2p2
Using equations (6.S3), (6.S4) and (6.S5) in equation (6.9):
Ò
1
fl2
p
vs2
2p vs2
≠ p2
tan Ïp = Ú 1 2 + 1
fl1 1
2 ≠ p2 1
≠ 2p2 vs2
≠ 2p2
vp2 vs2
A Ú B
Ò
1 1
p fl2
vs2
+ 2fl1 vs2
≠ p2 2
vp2
≠ p2
tan Ïp = Ú 1 2 (6.10)
1 1
fl1 2
vp2
≠ p2 vs2
≠ 2p2
6.S2 Additional figures
112
6.S2. Additional figures
0
5 10
1
source model:
receiver model:
10
100
1000
0
5 10
Figure 6.S1: Global velocity model used for synthetic seismograms (teleseismic
case). The model is split into a source and a receiver model. The source model is
continental PREM (dashed lines, Dziewonski and Anderson, 1981). Here, we present
a possible receiver model above 155.05 km depth with the CRUST 1.0 model for the
location 38.5¶ N and -18.5¶ E (solid lines, Laske et al., 2013).
113
Bibliography
10
0 2000 4000 6000 8000
Figure 6.S2: Test for influence of 5% noise on the appearance of vs,app profiles for
the synthetic model CM-REF. The spectral appearance of the noise was fitted to the
actual noise spectrum using the sum of a low pass filtered (0.02 Hz, order 1) and
band pass filtered (0.2-0.5 Hz, order 2) random noise. The grey dots show the results
of 300 trails. The blue curve shows the vs,app profile for the undisturbed CM-REF
data. The yellow curve shows the mean and the orange the median of all estimates.
114
6.S2. Additional figures
NOC NM NOC NM
1 1
10 10
3.5 4.0 4.5 5.0 3.5 4.0 4.5 5.0
NOC NM NOC NM
1 1
10 10
3.5 4.0 4.5 5.0 3.5 4.0 4.5 5.0
NOC NM
1
0.0001
0.001
10 0.01
0.1
1.
3.5 4.0 4.5 5.0
Figure 6.S3: Test for influence of Wiener filter parameter (deconvolution time
window length and damping parameter) on the appearance of vs,app profiles. The
colours indicate the damping parameter. (a)-(e) Grid search results for deconvolution
time window length between 30 s and 230 s.
115
Bibliography
Bibliography
116
7. Discussion
The main goal of this thesis is the investigation of the oceanic crust and upper
mantle structures North of the Gloria fault which marks the Eurasian-African
plate boundary in the eastern mid Atlantic. For this purpose, OBS data is used.
Before the further analysis could be conducted, some pre-processing steps are
essential: we need (1) to synchronize the data of the network, (2) to examine
the influence of the water column on the analysis of P wave polarization and
(3) to estimate the orientation of the horizontal components.
For the ’synchronization’ of OBS data, a linear clock drift is usually assumed.
The unusual high clock differences (Tab. 4.1) estimated at two of our twelve
stations lead to the conclusion that we have to test the hypothesis of a linear
clock drift before we could correct the timing. We do this by using ambient
noise cross-correlation (chapter 4, Hannemann et al., 2014) and could confirm
a small linear clock drift (1-2 s a≠1 ) for most of the stations. Nevertheless, we
detect a static time offset of ≥9 s at one station and confirm a large clock drift
of 20 min after 10 months at another station with our method. Furthermore,
we could prove that it is possible to retrieve the clock drift for stations which
had no chance for a second synchronization with a GPS signal. This newly
introduced method for estimating the relative clock drift at each station by
ambient noise cross-correlation has been successfully used to synchronize the
SWIR sub-array of the RHUM-RUM network (Stähler et al., 2016). Moreover,
the necessity for such a method to synchronize OBS networks was confirmed by
the study of Gouedard et al. (2014) which was published half a year after the
study presented in chapter 4 (Hannemann et al., 2014). This study also dealt
with the estimation of clock errors of OBS using ambient noise cross-correlation
not only estimating the linear drift, but additionally investigating a smaller
non-linear drift (≥10% of linear drift, Gouedard et al., 2014).
In order to examine the influence of the water column on the P wave
polarization, we use the so-called Zoeppritz equations (Zoeppritz, 1919; Aki
and Richards, 2002) to estimate the relationship between the measured P wave
polarization (apparent P wave incidence angle Ïp ) at the ocean bottom and
the elastic properties (eq. 6.10, section 6.S1 and appendix C). The obtained
relationship shows a dependence on the S wave velocity and the density of
the ocean bottom and an independence on the P wave velocity of the ocean
bottom of Ïp given that the parameters of the water column are assumed
as constant (in deep water ≥4.5-5.5 km sound speed varies about ≥1%, e.g.
Tolstoy, 1989; chapter 6,Hannemann et al., 2016). Furthermore, we perform
different synthetic tests to investigate the dependence of the P wave polarization
on the S wave velocity and the density and to examine the behaviour of the
P wave polarization at the ocean bottom for a progressive low-pass filtering
as was done for land-based stations by Svenningsen and Jacobsen (2007). We
find that the dependence of the P wave polarization on the S wave velocity is
117
Chapter 7. Discussion
much stronger than on the density (Fig. 6.3) and that the estimated S wave
velocities are a good approximation of the half space S wave velocity for the
simple case of an ocean-half space model. The progressive low-pass filtering of
the (Z, R) RF and the following estimation of the P wave polarization results in
the determination of apparent S wave velocity profiles which show the change
in the estimated S wave velocities with the used corner period of the filter
(Svenningsen and Jacobsen, 2007, chapter 6, Hannemann et al., 2016). The
examination of synthetic models with multiple layers reveals that the estimated
apparent S wave velocity profiles converge towards the S wave velocity of the
upper layer for the shorter periods and towards the S wave velocity of the
deeper layers for longer periods (chapter 6, Hannemann et al., 2016). A
similar behaviour of apparent S wave velocities has already been reported by
Svenningsen and Jacobsen (2007) for the analysis of land stations. Furthermore,
we use a stepwise quantitative modelling to estimate S wave velocity-depth
models for the crust and uppermost mantle beneath each OBS. As we obtain
differing models for stations being ≥10-20 km apart, we can assume a lateral
resolution capability in the same order. A detailed examination of the piercing
point distribution for the Ps phase and its multiples might enable us to further
investigate the sensitivity of the method and could probably also be used to
analyse the effect of layer dipping or undulation on the estimated P wave
polarization in different period ranges. Furthermore, the relation for the SV
wave polarization to the elastic parameters (eq. C.94) might be used for a
similar polarization analysis like the P wave polarization analysis presented
here. For this purpose, it would be necessary to perform similar synthetic tests
as for the P wave polarization and it might also be evaluated whether a joint
inversion of both polarizations is possible or necessary.
For the later on calculation of receiver functions (RFs), we need the correct
orientation of the three component seismogram. As already mentioned in the
introduction and the appendix 5.S1 in chapter 5, there exist several methods for
the estimation of the orientation (e.g Grigoli et al., 2012; Stachnik et al., 2012;
Zha et al., 2013). We use earthquake recordings to perform our orientation.
The synchronized OBS network is used to perform a frequency-wave number
analysis with a moving time window (Rost and Thomas, 2002) for band-pass
filtered data (7-25 s). By introducing a threshold for the extracted semblance
(Neidell and Taner, 1971), we identify 148 earthquakes recorded at the OBS
array for the recording period of 10 months (Tab. A.1). More than half of
these events are just detected by their surface waves which confirms Bormann
and Wielandt (2013) which reported that long period surface waves are easier
to detect at OBSs. The detected event recordings are manually revised to
search for suitable recordings for the orientation at the single stations. The
orientation using P waves and Rayleigh waves gives similar results (Fig. 5.S1
in appendix 5.S1, chapter 5). We finally decide to use the estimates of the
analysis of the P wave polarization, because we observe a smaller deviation in
the estimated back-azimuth for the P wave recordings compared to the Rayleigh
wave recordings for the frequency-wave number analysis (see appendix A) which
might be related to crustal and/ or uppermost mantle anisotropy. In future, it
might be possible to combine relative orientation (e.g. Grigoli et al., 2012) of
seismometer components with absolute orientation (e.g. Stachnik et al., 2012,
or appendix B) to obtain a homogenously oriented data set. Further possible
work-flows for the estimation of sensor orientation might be the application
118
of techniques like a principal component analysis (e.g. Wold et al., 1987) or a
polarization analysis of three component single stations or arrays (e.g. Jurkevics,
1988) on e.g. P wave data.
A further examination of the detected event recordings regarding their
usability for a RF analysis shows that we could use 25 events at the single
stations and that we are able to increase the number of events to 37 by
using beamforming (see Tabs. 5.S3 and 5.S4 in appendix 5.S3, chapter 5).
Furthermore, employing a quality control procedure by using different evaluation
criteria helps to pre-select possible deconvolution time window lengths for the
RF analysis (chapter 5). The first criterium employed is the relative spike
position within the deconvolution time window, the second the SNR of the
L component and the third the ratio of the rms amplitude of the signal time
window on the L component and the noise time window on the Q component.
Additionally, it could be proven that for one and the same event the SNR of
the beamformed trace is indeed higher than for the single station. The SNR
can be even improved by normalizing the single traces to the noise on either L
or Q component before stacking (Fig. 3 in chapter 5). From this observation,
we conclude that array techniques can indeed help to increase the number of
events and the signal quality of the single events for the analysis of OBS data.
The RF resulting from the pre-selected deconvolution lengths are manually
revised to choose the final processing parameter. This last step ensures that no
high frequent noise or resonance effect influences the RFs.
After all these steps, which can be considered as pre-processing, we could
calculate (L, Q) and (Z, R) RF for the investigation of the oceanic crust and
upper mantle structures. For the RF analysis presented in chapter 5, we use
the (L, Q) RF for which the secondary P-to-S converted (Ps) phases - especially
for the Moho - are better separated from the direct P phase which facilitates
their identification. As we are analysing RF stacks the single (L, Q) RFs have
been moveout correct (Yuan et al., 1997) for a reference distance of 67¶ and a
global velocity model (oceanic PREM, Dziewonski and Anderson, 1981) before
stacking. On the other hand, we use the (Z, R) RF for the analysis of the P
wave polarization presented in chapter 6 (Hannemann et al., 2016), because
the SNR of the direct P phase is higher for the R component than for the Q
component and therefore the estimation of the P wave polarization (apparent
P wave incidence angle) for single event recordings is less influenced by noise
for the R component.
In the following, I will discuss the results for the RF analysis which are
presented in chapter 5 and the analysis of the P wave polarization which are
presented in chapter 6 (Hannemann et al., 2016) regarding both the structure
of oceanic crust and upper mantle. The number of events used differs between
the RF analysis and the analysis of the P wave polarization, for which the 25
events used for the RF analysis had to be further reduced to 11 events (Tabs. 6.5
and 6.6) due to the influence of a resonance effect (Fig. 6.9). This effect is directly
related to the sedimentary cover at each station (see chapter 6, Hannemann
et al., 2016, for further details) and mainly influences the amplitudes of the
RF. We observe that the resonance signal can be suppressed by using larger
deconvolution lengths (chapter 5). The resulting RF can be used for the
investigation of the delay times, i.e. the time differences between the direct
P phase and the secondary Ps phases. Nevertheless, the influence on the
amplitudes obscures the estimated P wave polarization. We therefore exclude
119
those events from our analysis of the P wave polarization which show a strong
influence of the resonance (chapter 6, Hannemann et al., 2016). An approach
to deal with recordings influenced by resonance effects might be to convolve
synthetic RF with the Z (or L) component of the real data RF in a specific time
window (e.g. Schiffer et al., 2015) to ’model’ the effects encountered through the
resonance. Nevertheless, the optimal processing parameters like time window
length of Z (or L) component of the RF or eventually taper used before the
convolution have to be investigated. The removal of the resonance signal might
also be possible by using e.g. the Kalman filter (Kalman, 1960).
In Fig. 7.1, the results of the analysis of the P wave polarization (Ïp , grey
lines, chapter 6, Hannemann et al., 2016) and the Moho pseudo depths
estimated during the RF analysis (black circles with error bars, chapter 5) are
presented in comparison to the velocity-depth models extracted from PREM
(blue, Dziewonski and Anderson, 1981), White et al. (1992) (White92, red)
and CRUST1.0 (yellow, Laske et al., 2013). The overall appearance of the
velocity-depth models from the P wave polarization analysis, CRUST1.0 and
White92 is similar (Fig. 7.1).
The sedimentary velocities estimated by the P wave polarization analysis
are similar or higher than the sediment velocities in the models presented by
White et al. (1992) and Laske et al. (2013) (Fig. 7.1). The sediment thicknesses
resulting from the quantitative modelling of the P wave polarization data are
between 0.3 km and 0.9 km and similar to the sediment thickness presented by
White et al. (1992) (≥0.5 km) and the combined thickness of the sediment layer
0 0
10 10
20 20
PREM
results White92
pseudo depth CRUST1.0
0 2 4 6 8 10 12 0 1 2 3 4 5 6
Figure 7.1: Comparison of the velocity-depth models obtained by the analysis of

P wave polarization (Ïp , grey lines, Figs 6.12 d and 6.13), the Moho pseudo depths
estimated during the RF analysis (black circles with error bars Tab. 5.1, chapter 5)
and the velocity-depth models presented in Fig. 3.1: PREM (blue, Dziewonski
and Anderson, 1981), White92 (red, White et al., 1992, fig. 1) and CRUST1.0
(yellow, Laske et al., 2013, at 38.5¶ N, -18.5¶ E). The pseudo depths were estimated
with vp =6.8 km s≠1 and vs =3.9 km s≠1 (PREM) and for better visibility they are
presented arbitrarily distributed around these velocities.
120
and the upper crustal layer in CRUST1.0 (≥0.8 km, Laske et al., 2013). The
latter together with the sometimes higher estimates of the sediment velocities
can be an indication that the modelled ’sediment’ layer below the OBSs might
represent a transition from soft water saturated sediments to either compacted
probable lithified sediments or an upper crustal layer. The observed variation in
sediment thickness is in good agreement with an analog seismic recording from
1969 (NOAA World Data Service for Geophysics, Marine Seismic Reflection,
Survey ID V2707, ngdc.noaa.gov) in which an undulating basement topography
with sediment filled troughs can be identified. For the band-pass filtered (L, Q)
RF data (0.5-60 s), a peak at ≥0 s is observed (Fig. 5.4), we therefore assume
no biasing effect of the sediments on the estimates for the Moho depth of the
RF analysis (Sheehan et al., 1995).
The estimated Moho depths (below sea floor, Fig. 7.1) for all stations show
a good agreement between the results of the P wave polarization (4-11.5 km)
and the pseudo depths of the RF analysis (3.5-10.7 km). The average depth of
both estimates is ≥6.7 km and matches the total crustal thickness of ≥7 km
of the models presented by White et al. (1992) and Laske et al. (2013). For
the crustal thickness estimated by the analysis of the P wave polarization,
we observe a weak increase towards the Gloria fault in the South along a
N-S striking profile (Fig. 6.15 b). This weak trend in the crustal thickness is
hard to detect in the Moho pseudo depths estimated by the RF analysis, but
might be better appreciated in the N-S common conversion point (CCP) profile
(Fig. 5.6), although the southern part of the profile is less well constrained due
to the azimuthal gap of the RF data between ≥80¶ and 200¶ . We suggested
in chapter 6 (Hannemann et al., 2016) that this weak trend might be caused
by a crustal shortening due to a known transpression in the vicinity of the
Gloria fault (Zitellini et al., 2009, fig. 6). The crustal S wave velocity obtained
by the quantitative modelling from the P wave polarization varies between
3.1-4.5 km s≠1 with an average velocity of ≥3.8 km s≠1 which is in the same
range as the lower crust S wave velocity given by PREM (3.9 km s≠1 Dziewonski
and Anderson, 1981) and the crustal Ô velocities given by White et al. (1992)
(2.71-4.04 km s≠1 assuming vp /vs = 3), as well as the S wave velocities of the
middle and lower crust given by CRUST1.0 (3.7 km s≠1 and 4.05 km s≠1 Laske
et al., 2013).
For the uppermost mantle S wave velocities, we obtain a decrease from
velocities around 5.5 km s≠1 in the north of the OBS array to 4.5 km s≠1 in the
south of the OBS array by the quantitative modelling of the P wave polarization
data (Fig. 6.15 a). Furthermore, we conclude that the observed velocity decrease
towards the Gloria fault in the South is in-line with the combined effects of
serpentinization, an increase in water content and grain size reduction due to
the deformation caused by the neighbouring fault (chapter 6, Hannemann
et al., 2016). Additionally, we suggest that the serpentinization might be the
strongest of the given effects (chapter 6, Hannemann et al., 2016) which is in
good agreement with the occurrence of serpentinite close to fractures in oceanic
crust and mantle (White et al., 1992).
For all deeper structures, the analysis of the P wave polarization could not
be used, because the employed filter periods for the analysis are shorter than
16 s (chapter 6, Hannemann et al., 2016) due to the band limited nature of the
earthquake recordings and known high self-noise levels of the seismometers used
(Stähler et al., 2016). Otherwise, it might have been possible to model the low
121
velocity layer (LVL) which was detected to start at depths between 40-60 km
(LAB, Fig. 5.6). This is not possible for the presented data set, because the
examination of synthetic data (Fig. 6.7) shows that for periods shorter than
16 s the presence of a LVL 50 km below sea floor is hard to distinguish from a
model without a LVL by analysing P wave polarization data. An extension of
the period range of the analysis especially to longer periods might be possible
either by using instruments with a wider pass-band (e.g. Trillium Compact 120
sec, NAMMU, Schwenk, 2016) or by a combined analysis of ’short’ period P
wave polarization and ’long’ period Rayleigh wave ellipticty (e.g. Boore and
Toksöz, 1969; Tanimoto and Rivera, 2008; Yano et al., 2009).
For the LAB, we obtain two depth estimates: a pseudo depth of 40-60 km
based on assuming an average velocity model for the study area and an estimate
of 40-50 km from the CCP profiles (Fig. 5.6). Both depth estimates are much
shallower than 70 km which would be expected by Olugboji et al. (2013) for
oceanic lithosphere with an age between 75 Ma and 85 Ma (Fig. 5.8 Müller
et al., 2008). The elevated LAB depth might indicate a high temperature
anomaly which is probably related to the neighbouring fault. The rather broad
small amplitude RF phase associated with the LAB is another indication for
this hypothesis. Furthermore, we observe a positive broad small amplitude RF
phase following rather close in time after the LAB RF phase (Fig. 5.5). The
estimated pseudo depths (70-115 km) and the depth estimated from the CCP
70
60
50
SUM
40
LNR
QNR
30
20
10
0
0 100 200 300 400 500 600 700
Figure 7.2: Estimated delay times of RF phases for single stations stack (black,
SUM) and beamformed traces (red, LNR - normalized to noise on L component
and blue QNR - normalized to noise on Q component). The values for Moho, LAB
(6-40 s) and MAD (6-40 s) are presented at the estimated pseudo depths and the
values for the Lehmann discontinuity, the ’410’, ’520’(?) and ’660’ are shown at
the average global depths of these discontinuities. The estimated theoretical delay
times for PREM (Dziewonski and Anderson, 1981) and a reference distance of 67¶
(horizontal slowness p=6.4 s ¶ ) are presented as a grey line.
≠1
122
stack (90-110 km, Fig. 5.6) are too shallow to be associated with the Lehmann
discontinuity (180-240, Deuss and Woodhouse, 2004) and we therefore refer to
this feature as mid asthenospheric discontinuity (MAD). The small amplitude
double-polarity signal formed by the negative RF phase of the LAB and the
positive RF phase of the MAD can be caused by a small amount of melt in
the upper asthenosphere (Olugboji et al., 2013). The LAB and the MAD RF
phases could still be identified on the single station RFs, but no clear depth
variation could be identified for both discontinuities (Fig. 5.6). Our observations
of a small double-polarity signal are in contrast to the observations made by
Olugboji et al. (2016) at most of their normal-ocean stations. Nevertheless, the
possible presence of partial melt in the upper asthenosphere is in-line with the
estimated shallower LAB depths which indicate a high temperature anomaly.
We observe a positive RF phase in the stacked single stations’ RF and the
stacked beamformed traces’ RFs at ≥22-23 s (Fig. 7 in chapter 5) which can
be associated with the Lehmann discontinuity. Following the interpretation of
Karato (1992) and Deuss and Woodhouse (2004), the Lehmann discontinuity
marks a change from anisotropic material above to isotropic material below
in PREM (Dziewonski and Anderson, 1981) at ≥220 km. Moreover, it is
associated with the lower boundary of the asthenosphere (Lehmann, 1961;
Deuss et al., 2013). Our observed RF phase arrives a bit earlier than expected
by PREM (23.5 s, Fig. 7.2) which can indicate a wet mantle or a smaller grain
size (Karato, 1992; Deuss and Woodhouse, 2004). The smaller grain sizes can
be caused by higher stress levels (Bürgmann and Dresen, 2008) which might
be present in the vicinity of a major transform fault like the Gloria fault.
The estimated delay times for the ’410’ (≥43.7-45.5 s) are similar to the
expected values assuming PREM (43.7 s, Dziewonski and Anderson, 1981) by
considering the delay time errors (≥ ±1.0 s) which is also visible in Fig. 7.2.
This result is in good agreement with the obtained depths by (Saki et al., 2015)
who detected the ’410’ at normal depths in areas away from hot spot locations
in the eastern mid Atlantic. For the ’660’, the obtained delay time for the single
station stack (67.8 s) is in good agreement with the PREM estimates (68.0 s).
Even if the errors are considered (≥ ±0.5 s, Fig. 7.2), the ’660’ RF phase on
the beamformed traces (≥69.0 s) arrives slightly later than expected by PREM.
Furthermore, we observe an additional phase which arrives delayed after the
’660’ RF phase for back-azimuths between 180¶ and 315¶ . Both features, the
slightly delayed phase and the additional phase indicate that an additional
transition in the Aluminium phase from majorite garnet to perovskite is present
at the base of the transition zone (Simmons and Gurrola, 2000) in the study
area. This hypothesis is supported by the study of Saki et al. (2015) in the
eastern mid Atlantic in which the majorite-perovskite transition was used as
explanation for the hardly observable ’660’. Pseudo depths estimated for the
’410’ and the ’660’ are highly influenced by the crust and uppermost mantle
models under consideration, therefore the time difference between the ’660’
and the ’410’ is used as a more robust measure for the characterization of the
mantle transition zone (MTZ). The estimated time difference for the single
stations’ stack as well as the beamformed traces are similar (≥24.0 s) and show
a good correspondence to the expected values using PREM (24.3 s). This
indicates a normal thickness of the MTZ.
The delay times of the beamformed traces’ stacks (Figs 5.7 a and 7.2)
show a small time difference to the times estimated for the single stations’
123
stack for the Lehmann discontinuity (≥1 s), the ’410’ (≥1.5 s) and the ’660’
(≥1.5 s). The examination of azimuthal stacks for the Q (Fig. 5.7 b) and the T
component (Fig. 5.S2 b) does not give evidence for anisotropy. Furthermore, a
comparison of the synthetic RF calculated with the same station distribution
as in the experiment revealed no time difference between single stations’ stack
RF and beamformed traces’ RF. We therefore suggest in chapter 5 that the
delay time difference might be related to velocity perturbations or thickness
variations above the Lehmann discontinuity and the ’410’. The changes in
either velocity or thickness between the Lehmann discontinuity and the ’410’
are indicated by the larger time differences between the beamformed traces’
stack and the single stations’ stack for the ’410’ compared to the Lehmann
discontinuity. Furthermore, comparing the fraction of event recordings with
back-azimuth for the single stations’ stack and the beamformed traces’ stacks
(Fig. 5.8) shows that the single stations’ stack has a higher fraction of recordings
which samples structures West of the OBS array and the beamformed traces’
stack has a higher fraction of recordings which samples structures North of the
OBS array. This indicates that the structures in the mantle above the Lehmann
discontinuity West of the OBS array show either higher S wave velocities, slower
P wave velocities or smaller thicknesses than the structures North of the OBS
array. Nevertheless, there is no obvious reason for this in age (Fig. 5.8) or
bathymetry (Fig. 5.1). The delay times of the Moho, the LAB and the MAD
agree quite well between the single stations’ stack and the beamformed traces
stack, therefore the decrease in the uppermost S wave velocity and the increase
in the crustal thickness towards the Gloria fault South of the array which was
detected by the analysis of the P wave polarization (chapter 5, Hannemann
et al., 2016) is probably not responsible for the delay time difference of the
mantle discontinuities. Furthermore, this might indicate that one reason for
the delay time differences of the Lehmann discontinuity, the ’410’ and the ’660’
is located in the lower asthenosphere (Fig. 7.2). Additional detailed information
concerning the velocity-depth structure (e.g. from the analysis of surface waves)
would be needed to further investigate the cause of the delay time differences.
Besides the Lehmann discontinuity, the ’410’ and the ’660’, an additional
weak RF phase at ≥56-57 s is observed in the single stations’ stack and the
beam formed traces’ stack (Fig. 5.7). This RF phase might be related to the
’520’ which is a global discontinuity with small amplitude signals (Flanagan
and Shearer, 1998; Chevrot et al., 1999). For the ’520’(?), we could not
clearly observe the shift in the delay times between single stations’ stack and
beamformed traces’ stacks (Fig. 5.7). On the other hand, we identify a RF
phase at similar delay times on the T components (Fig. 5.S2). This might
indicate a dipping of the ’520’(?) (Cassidy, 1992), but the RF data and its
azimuthal coverage is too limited to investigate this any further.
For the RF analysis, it might be an additional option to apply the technique
introduced by (Tao et al., 2014) to remove sediment reverberations to obtain
so-called subsurface receiver functions which might facilitate the analysis of the
Moho and the lithosphere asthenosphere boundary (LAB) and this might also
help to better resolve the mantle structures. Concerning the signal quality at
OBS arrays, we observe from the employed evaluation criteria (Fig. 5.3) that
it is possible to increase the SNR on the L component of the RF by a factor
≥5 (at 110 s, QNR/ D03) and the ratio of the rms amplitude in the signal
time window and the one in the noise time window by a factor ≥3.5 (at 110 s,
124
Bibliography
QNR/ D03) and the number of usable events for the RF analysis by ≥50%. It
would be interesting for future OBS experiments to analyse in detail how many
stations are needed to ensure a specific improvement factor between a single
station recording and the beam. This might be accomplished by randomly
choosing a specific number of stations, beamforming of the resulting set of
station recordings and systematically comparing the obtained results.
In summary, this thesis contains methods for the pre-processing of OBS data
and the investigation of the oceanic crust and mantle using earthquake data.
Here, I present a work-flow which was successfully used to a posteriori estimate
clock drifts at OBSs. The estimation of these drifts in case of erroneous or
lost timing is essential for the further analysis of OBS data especially if several
stations should be combined. This combination is achieved by employing
beamforming which is a known approach to enhance the SNR of transient
signals like earthquakes. I show that it is a feasible method to increase the
number of usable events for the analysis of RF at OBSs especially to investigate
structures in Earth’s upper mantle. Furthermore, I modify the analysis of
P wave polarisation in terms of S wave velocities in a way that it can be
employed on the sea floor. For this purpose, I estimate the valid theoretical
relation and use a stepwise quantitative modelling to search for reasonable
S wave velocity-depth models of the oceanic crust and uppermost mantle.
In conclusion, the investigation of the crustal and upper mantle structures
North of the Gloria fault reveals the presence of a rather normal oceanic crust.
Nevertheless, the influence of this major plate boundary can be noticed by a
probable serpentinization of the uppermost mantle towards the fault and that
the upper asthenosphere might contain partial melts. Major discontinuities like
the Lehmann discontinuity, the ’410’ and the ’660’ could be identified in the
processed OBS data and the analysis also confirms the presence of an additional
transition in the Aluminium phase (majorite to perovskite) at the bottom of
the MTZ in the eastern mid Atlantic.
Bibliography
Aki, K. and Richards, P. G. (2002). Quantitative Seismology. University Science
Books, 2nd edition.
Boore, D. M. D. and Toksöz, M. N. (1969). Rayleigh wave particle motion
and crustal structure. Bulletin of the Seismological Society of America,
59(1):331–346.
Bormann, P. and Wielandt, E. (2013). Seismic Signals and Noise. In Bormann,
P., editor, New Manual of Seismological Observatory Practice 2, volume 2,
chapter 4, pages 1–64. GeoForschungsZentrum Potsdam GFZ, Potsdam.
Cassidy, J. J. F. (1992). Numerical experiments in broadband receiver function
analysis. Bulletin of the Seismological Society of America, 82(3):1453–1474.
125
Bibliography

Letters, 225(3-4):295–304.


Gouedard, P., Seher, T., McGuire, J. J., Collins, J. A., van der Hilst, R. D.,
Gouédard, P., Seher, T., McGuire, J. J., Collins, J. A., and van der Hilst,
R. D. (2014). Correction of ocean-bottom seismometer instrumental clock
errors using ambient seismic noise. Bulletin of the Seismological Society of
America, 104(3):1276–1288.
Hannemann, K., Krüger, F., and Dahm, T. (2016). Oceanic lithospheric S

wave velocities from the analysis of P wave polarization at the ocean floor.
Geophysical Journal International.
Jurkevics, A. (1988). Polarization analysis of three-component array data.

Bulletin of the Seismological Society of America, 78(5):1725–1743.
Kalman, R. E. (1960). A New Approach to Linear Filtering and Prediction

Problems. Journal of Basic Engineering, 82(1):35.

Letters, 19(22):2255–2258.
Laske, G., Masters, G., Ma, Z., and Pasyanos, M. E. (2013). CRUST1.0 :
An Updated Global Model of Earth ’ s Crust. In Geophys. Res. Abstracts,
volume 15, pages Abstract EGU2013—-2658.

126
Bibliography
Neidell, N. S. and Taner, M. T. (1971). Semblance and other coherency measures

for multichannel data. GEOPHYSICS, 36(3):482–497.
Olugboji, T. M., Park, J., ichiro Karato, S., Shinohara, M., ichiro Karato, S.,
and Shinohara, M. (2016). Nature of the seismic lithosphere-asthenosphere
boundary within normal oceanic mantle from high-resolution receiver func-
tions. Geochemistry, Geophysics, Geosystems, 17(4):1265–1282.

Saki, M., Thomas, C., Nippress, S. E. J., and Lessing, S. (2015). Topography
409:193–202.
Schiffer, C., Jacobsen, B. H., Balling, N., Ebbing, J., and Nielsen, S. B. (2015).
The East Greenland Caledonides—teleseismic signature, gravity and isostasy.
Schwenk, A. (2016). LOBSTER - The Next Generation. In Geophys. Res.

Abstracts, pages Abstract EGU2016–11915.
Sheehan, A. F., Abers, G. a., Jones, C. H., and Lerner-Lam, A. L. (1995).

Crustal thickness variations across the Colorado Rocky Mountains from tele-
seismic receiver functions. Journal of Geophysical Research, 100(B10):20391–
20404.

562.
83(4):704–713.
41:43–63.

Tanimoto, T. and Rivera, L. (2008). The ZH ratio method for long-period

seismic data: Sensitivity kernels and observational techniques. Geophysical
127
Bibliography
Tao, K., Liu, T., Ning, J., and Niu, F. (2014). Estimating sedimentary
and crustal structure using wavefield continuation: theory, techniques and
applications. Geophysical Journal International, 197(1):443–457.
Tolstoy, A. (1989). Sensitivity of matched field processing to sound-speed

profile mismatch for vertical arrays in a deep water Pacific environment. The
Journal of the Acoustical Society of America, 85(6):2394.

Wold, S., Esbensen, K., and Geladi, P. (1987). Principal component analysis.
Chemometrics and Intelligent Laboratory Systems, 2(1-3):37–52.
Yano, T., Tanimoto, T., and Rivera, L. (2009). The ZH ratio method for
long-period seismic data: Inversion for S-wave velocity structure. Geophysical
Yuan, X., Ni, J., Kind, R., Mechie, J., and Sandvol, E. (1997). Lithospheric
and upper mantle structure of southern Tibet from a seismological passive
source experiment. Journal of Geophysical Research, 102(B12):27491.
Zitellini, N., Gràcia, E., Matias, L., Terrinha, P., Abreu, M., DeAlteriis, G.,
Henriet, J. P., Dañobeitia, J. J., Masson, D. G., and Mulder, T. (2009). The
quest for the Africa–Eurasia plate boundary west of the Strait of Gibraltar.
Earth and Planetary Science Letters, 280(1-4):13–50.

128
8. Conclusion
In this thesis, the analysis of OBS data is used to investigate the nature of the
oceanic crust and upper mantle in the eastern mid Atlantic, North of the Gloria
fault. In order to later on synchronize all stations, we estimated clock drifts and
static time offsets at OBSs using ambient noise cross-correlation. This method
is feasible for cross-checking the estimates gained through the synchronization
of the internal clocks with a GPS signal to reveal erroneous estimates and also
to a posteriori retrieve lost timing due to an too early shut-down. We have
sucessfully transfered the analysis of P wave polarization in terms of S wave
velocities (Svenningsen and Jacobsen, 2007) to the ocean bottom by estimating
its theoretical relation at the ocean bottom and by using a stepwise quantitative
modelling to search for possible S wave velocity-depth models.
The synchronized OBS network was sucessfully used to detect teleseismic
earthquakes by performing a frequency-wave number analysis with a moving
time window. The detected events were examined to select usable recordings at
single stations to calculate receiver functions (RF). Furthermore, the application
of beamforming proved to be a feasible method to increase the number of usable
earthquake data for the RF analysis and to increase the signal-to-noise ratio
(SNR) of the event recordings. The SNR can be even more improved by pre-
normalizing the single station recordings to the noise on either L or Q component
before stacking. Additionally, we used a quality control with different evaluation
criteria and a manual revision step to choose the time window length for the
deconvolution of the earthquake data to obtain RF.
The analysis of the RF revealed a rather normal oceanic crustal structure,
but an elevated LAB depth and the detection of a small amplitude double
polarity (i.e. a negative LAB RF phase followed by a positive RF phase) which
might indicate the presence of partial melt in the upper asthenosphere. This
might be caused due to the influence of the neighbouring fault. The probable
influence of the fault is also visible in the uppermost mantle velocities which
were obtained by the analysis of the P wave polarization data and show a
decrease towards the Gloria fault in the South. This might be related to the
growing influence of serpentinization towards the fault.
We could identify major discontinuities like the Lehmann discontinuity,
the ’410’ and the ’660’ in the RFs. The Lehmann discontinuity arrived at
slightly earlier times than would be expected from PREM. This might be
related to a wet mantle above the Lehmann discontinuity (Karato, 1992; Deuss
and Woodhouse, 2004). The ’410’ is detected at the expected depth and for
the ’660’ an additional RF phase at slightly later times is observed for some
back-azimuths which might indicate the presence of an additional transition in
the Aluminium phase of the mantle (majorite garnet to perovskite, Weidner
and Wang, 1998; Vacher et al., 1998; Hirose, 2002; Deuss et al., 2006, 2013).
129
Bibliography
Furthermore, we observe a weak signal between the ’410’ and the ’660’ which
might be associated to the ’520.
This thesis investigates overall methods for the pre-processing of OBS data
and the analysis of the oceanic crust and mantle using earthquake data. It shows
the sucessful application of ambient noise cross-correlation for the estimation
of clock drifts and static time offsets at OBSs. Furthermore, we present the
usage of RF for the analysis of P wave polarization at the ocean bottom and
the identification of major discontinuities in oceanic crust and mantle. We also
presented that the number of events and their SNR can be increased by using
beamforming before the calculation of RF.
Bibliography
Deuss, A., Redfern, S. A. T., Chambers, K., and Woodhouse, J. H. (2006). The
Nature of the 660-Kilometer Discontinuity in Earth’s Mantle from Global
Seismic Observations of PP Precursors. Science, 311(5758):198–201.

Letters, 225(3-4):295–304.
Hirose, K. (2002). Phase transitions in pyrolitic mantle around 670-km depth:

Implications for upwelling of plumes from the lower mantle. Journal of
Geophysical Research, 107(B4):2078.

Letters, 19(22):2255–2258.

Vacher, P., Mocquet, A., and Sotin, C. (1998). Computation of seismic profiles
from mineral physics: the importance of the non-olivine components for
explaining the 660 km depth discontinuity. Physics of the Earth and Planetary
Interiors, 106(3-4):275–298.
Weidner, D. J. and Wang, Y. (1998). Chemical- and Clapeyron-induced

buoyancy at the 660 km discontinuity. Journal of Geophysical Research,
103(B4):7431.
130
A. Data
The data used for this thesis were recorded with broad-band OBS from the
German instrument pool for amphibian seismology (“Deutscher Geräte-Pool
für amphibische Seismologie”, DEPAS, Stähler et al., 2016). Twelve OBS were
installed in the deep sea (4.5-5.5 km) of the eastern mid Atlantic ocean, North
of the European African plate boundary (Gloria fault, Bird, 2003, Fig. 3.2),
≥800 km off-shore Portugal. Each station is equipped with a three component
broad-band seismometer (Guralp CMG-40T, 0.167 Hz-50 Hz) and a hydrophone
(HighTechInc HTI-04-PCA/ULF, 0.01 Hz-8 kHz, flat instrument response down
to 0.2 Hz, at D08 down to 0.5 Hz). The hydrophone and the seismometer
share the same datalogger (Send Geolon MCS, 24 bit, 1 - 1000 Hz, 20 GB; see
Dahm et al., 2002; Stähler et al., 2016, for detailed instrument description).
The seismometer, the datalogger and the batteries are stored in pressure tubes
which together with the hydrophone are attached to a floating unit. The
station is connected via an acoustic releaser to an anchor which remains on
the sea floor after the recovery (Stähler et al., 2016). All four channels (3
seismometer components and hydrophone) were recorded with 100 Hz. The
stations were installed during the RV Poseidon cruise POS 416 in 2011 July
and recovered during the RV Poseidon cruise POS 431 in 2012 April. Most
stations worked fine for the whole recording period. Station D05 had a clamped
vertical and one clamped horizontal component, i.e. the recorded amplitudes
are at their maximum amplitude. The station stopped recording one month
before recovery due to the high energy consumption because of the clamped
components. At station D03, the hydrophone had a severe malfunction after
200 days of recording (mid of 2012 January).
The positions of the stations were determined by the ship’s position during
deployment and recovery (filled triangles and open circles in Fig. 3.2). Addi-
tionally, range measurements to the acoustic releaser were performed during
cruising around the deployment position after deployment and during the
ascending of the station. These range measurements give the distance between
ship and station by assuming a constant sound speed velocity (1.5 km s≠1 ). By
using gradient descent (Dahm, 2012) and the deployment position as starting
point, the stations’ positions at the sea floor can be re-estimated (red open
triangles in Fig. 3.2).
We performed a frequency-wavenumber analysis with a moving time window
(Rost and Thomas, 2002) with 40 s time windows and 10 s steps on band-pass
filtered data (7-25 s). For each time window, the maximum semblance and
the correponding back azimuth and slowness were stored. In order to detect
teleseismic events, we introduce a threshold (semblance&0.2).
131
Table A.1: Events detected with frequency-wavenumber analysis with a moving time window (NEIC catalog, earthquake.usgs.gov). Magnitudes are given
as moment magnitude (mw?) or short period body wave magnitude (mb).
# date time latitude longitude depth mag magtype

1 2011-07-06 19:03:18.26 -29.54 -176.34 17.00 7.6 mww
2 2011-07-07 09:29:59.56 -29.15 -176.94 30.80 5.8 mwc
3 2011-07-08 05:53:03.81 0.96 -26.42 10.00 5.6 mwc
4 2011-07-09 07:08:18.82 -21.16 -174.64 10.00 5.6 mwc
5 2011-07-09 13:54:23.57 -29.39 -177.12 19.00 5.9 mww
6 2011-07-09 15:02:27.24 -29.34 -177.05 14.00 6.0 mww
7 2011-07-09 19:35:18.75 -29.44 -177.01 15.20 6.0 mww
8 2011-07-10 00:57:10.80 38.03 143.26 23.00 7.0 mww
9 2011-07-11 07:43:28.89 -29.28 -176.84 10.00 5.1 mwc
10 2011-07-11 20:47:04.30 9.51 122.18 19.00 6.4 mww
11 2011-07-12 20:11:01.26 10.65 -85.20 10.00 5.5 mwc
132
12 2011-07-16 00:26:12.64 -33.82 -71.83 20.00 6.0 mww
13 2011-07-16 19:59:12.89 54.79 -161.29 36.00 6.1 mww
14 2011-07-18 22:48:25.50 51.27 179.05 50.80 5.4 mwc
15 2011-07-19 19:35:43.48 40.08 71.41 20.00 6.1 mww
16 2011-07-23 04:34:24.18 38.90 141.82 41.00 6.3 mww
17 2011-07-24 18:51:25.07 37.73 141.39 40.00 6.3 mww
18 2011-07-24 19:05:29.78 45.74 90.20 10.00 4.9 mb
19 2011-07-25 00:50:47.59 -3.18 150.61 10.00 6.3 mww
20 2011-07-25 17:15:40.81 14.95 120.04 35.00 5.9 mww
21 2011-07-26 17:44:20.38 25.10 -109.53 12.00 6.0 mww
22 2011-07-27 23:00:29.67 10.80 -43.39 6.00 5.9 mww
23 2011-07-28 19:50:20.06 -35.77 -73.12 35.00 5.7 mww
Appendix A. Data
24 2011-07-29 07:42:23.40 -23.80 179.75 532.00 6.7 mww


25 2011-07-30 18:53:50.72 36.94 140.96 38.00 6.3 mww
26 2011-07-31 14:34:47.32 -17.02 171.58 10.00 6.1 mww
27 2011-07-31 23:56:36.43 0.03 99.22 79.40 5.1 mb
28 2011-08-01 13:44:47.30 39.84 142.08 40.00 5.7 mww
29 2011-08-01 18:20:05.03 51.79 -171.27 41.20 5.6 mwc
30 2011-08-04 13:51:34.56 48.83 154.77 36.00 6.1 mww
31 2011-08-05 16:08:45.58 -29.99 -176.73 10.00 5.7 mwc
32 2011-08-07 15:57:29.07 -55.87 -27.05 44.60 4.5 mb
33 2011-08-10 23:45:43.04 -7.04 -12.62 10.00 6.0 mww
34 2011-08-17 11:44:08.37 36.77 143.77 9.00 6.1 mww
133
35 2011-08-19 05:36:33.04 37.67 141.65 47.00 6.2 mww
36 2011-08-20 16:55:02.81 -18.37 168.14 32.00 7.2 mww
37 2011-08-20 18:19:23.55 -18.31 168.22 28.00 7.1 mww
38 2011-08-22 09:38:37.31 -29.03 -176.68 10.00 5.7 mwc
39 2011-08-22 11:23:35.25 36.08 141.69 12.00 5.9 mww
40 2011-08-22 20:12:20.95 -6.28 104.05 29.00 6.1 mww
41 2011-08-23 04:56:52.87 12.01 44.09 10.00 4.9 mb
42 2011-08-23 17:51:04.25 37.91 -77.94 0.02 5.8 mw
43 2011-08-24 17:46:11.65 -7.64 -74.53 147.00 7.0 mww
44 2011-08-24 23:06:17.09 -18.16 167.73 13.00 6.2 mww
45 2011-08-30 06:57:41.61 -6.36 126.75 469.80 6.9 mww
46 2011-08-31 12:17:27.01 43.59 -28.90 10.00 5.5 mww
47 2011-09-02 10:55:53.59 52.17 -171.71 32.00 6.9 mww
48 2011-09-02 13:47:09.62 -28.40 -63.03 578.90 6.7 mww

49 2011-09-03 04:48:57.31 -56.45 -26.85 84.00 6.4 mww
50 2011-09-03 16:20:41.00 -38.44 -74.91 12.00 5.8 mww
51 2011-09-03 22:55:40.92 -20.67 169.72 185.10 7.0 mww
52 2011-09-05 09:52:01.13 -15.30 -173.62 37.00 6.2 mww
53 2011-09-09 19:41:34.15 49.54 -126.89 22.00 6.4 mww
54 2011-09-14 07:03:51.00 -32.70 -71.80 37.00 5.8 mww
55 2011-09-14 18:10:09.00 53.12 172.98 15.00 6.0 mww
56 2011-09-15 08:00:09.64 36.26 141.34 28.00 6.1 mww
57 2011-09-15 08:43:07.25 19.57 -77.94 10.00 5.2 mwr
58 2011-09-15 19:31:04.08 -21.61 -179.53 644.60 7.3 mww
59 2011-09-16 19:26:40.26 40.27 142.78 30.00 6.7 mwc
134
60 2011-09-16 20:11:16.09 40.24 143.24 35.00 5.7 mwc
61 2011-09-18 12:40:51.83 27.73 88.16 50.00 6.9 mww
62 2011-09-19 08:14:14.76 52.04 -171.98 31.00 5.6 mww
63 2011-09-19 18:33:55.87 14.19 -90.24 9.00 5.6 mwr
64 2011-09-22 03:22:36.07 39.79 38.84 5.00 5.5 mwb
65 2011-09-22 23:07:03.57 -15.44 -175.31 10.00 6.4 mww
66 2011-09-26 01:02:56.25 63.43 -126.28 1.00 5.3 mwr
67 2011-09-28 22:40:12.86 -37.95 -73.85 10.00 5.6 mwb
68 2011-10-05 23:52:20.01 57.87 -32.54 10.00 5.4 mwc
69 2011-10-06 07:37:01.42 9.70 138.25 20.00 5.8 mww
70 2011-10-06 11:12:30.07 -24.18 -64.22 15.00 5.9 mww
71 2011-10-08 08:53:11.96 -20.60 -173.22 6.00 5.9 mww
Appendix A. Data
72 2011-10-14 03:35:14.81 -6.57 147.88 37.00 6.5 mww


73 2011-10-14 06:10:14.60 54.08 123.72 12.00 6.0 mww
74 2011-10-19 06:23:42.97 38.00 -31.37 10.00 5.0 mwc
75 2011-10-21 17:57:16.10 -28.99 -176.24 33.00 7.4 mww
76 2011-10-23 10:41:23.25 38.72 43.51 18.00 7.1 mww
77 2011-10-28 18:54:34.04 -14.44 -75.97 24.00 6.9 mww
78 2011-10-29 18:45:48.89 38.62 43.18 5.00 4.7 mb
79 2011-10-29 21:01:52.11 -47.50 100.12 10.00 5.3 mwc
80 2011-11-01 12:32:00.43 19.83 -109.21 10.00 6.3 mww
81 2011-11-06 03:53:10.00 35.53 -96.77 5.20 5.7 mww
82 2011-11-08 02:59:08.51 27.32 125.62 224.90 6.9 mww
135
83 2011-11-09 19:23:33.24 38.43 43.23 5.00 5.6 mww
84 2011-11-14 04:05:11.39 -0.95 126.91 17.00 6.3 mww
85 2011-11-14 11:34:57.42 39.83 -29.58 10.00 5.1 mwc
86 2011-11-18 04:34:03.02 -37.42 179.99 33.00 5.6 mww
87 2011-11-22 18:48:16.30 -15.36 -65.09 549.90 6.6 mww
88 2011-11-22 19:10:50.69 -2.29 99.51 35.00 5.2 mb
89 2011-11-24 10:25:34.03 41.90 142.64 38.00 6.2 mww
90 2011-11-27 04:21:53.00 7.45 -36.65 10.00 4.4 mb
91 2011-11-28 12:26:45.45 -5.48 153.73 25.00 6.1 mww
92 2011-11-29 00:30:29.12 -1.60 -15.45 10.00 5.9 mww
93 2011-11-30 00:27:06.99 15.46 119.00 9.00 6.0 mww
94 2011-12-07 22:23:09.73 -27.90 -70.92 20.00 6.1 mww
95 2011-12-08 08:54:46.32 38.02 -30.85 10.00 5.1 mwc
96 2011-12-11 01:47:25.56 17.99 -99.79 59.00 6.5 mww

97 2011-12-14 05:04:58.63 -7.55 146.81 135.00 7.1 mww
98 2011-12-23 00:58:38.31 -43.49 172.80 9.70 5.8 mww
99 2011-12-23 02:18:03.73 -43.53 172.74 6.90 5.9 mww
100 2011-12-23 19:12:34.01 -52.12 27.96 10.00 5.9 mww
101 2011-12-27 15:21:56.84 51.84 95.91 15.00 6.6 mww
102 2012-01-01 05:27:55.98 31.46 138.07 365.30 6.8 mww
103 2012-01-09 04:07:14.67 -10.62 165.16 28.00 6.4 mww
104 2012-01-10 18:36:59.08 2.43 93.21 19.00 7.2 mww
105 2012-01-14 16:36:21.04 19.20 121.16 17.00 5.8 mww
106 2012-01-15 13:40:19.54 -60.95 -56.11 8.00 6.6 mww
107 2012-01-17 23:21:35.04 -31.66 -71.50 32.90 5.6 mww
136
108 2012-01-19 06:48:48.75 -46.69 165.78 20.00 5.9 mww
109 2012-01-21 18:47:11.56 14.87 -93.01 45.00 6.2 mww
110 2012-01-22 06:00:05.41 -56.66 -24.90 10.00 5.5 mwc
111 2012-01-23 16:04:52.98 -36.41 -73.03 20.00 6.1 mww
112 2012-01-30 05:11:00.95 -14.17 -75.64 43.00 6.4 mww
113 2012-02-02 13:34:40.65 -17.83 167.13 23.00 7.1 mww
114 2012-02-03 03:46:21.15 -17.38 167.28 8.00 6.1 mww
115 2012-02-04 07:40:12.80 -20.54 -174.04 8.00 5.8 mwb
116 2012-02-04 20:05:31.71 48.89 -127.91 14.60 5.6 mww
117 2012-02-06 03:49:12.52 10.00 123.21 11.00 6.7 mww
118 2012-02-06 10:10:19.85 9.89 123.10 9.00 6.0 mww
119 2012-02-13 10:55:09.44 9.18 -84.12 16.00 5.9 mww
Appendix A. Data
120 2012-02-13 20:03:15.07 -33.74 -178.47 43.00 4.4 mb


121 2012-02-14 06:22:01.17 36.21 141.39 28.00 5.8 mww
122 2012-02-14 08:19:55.47 -10.39 161.10 51.00 6.4 mww
123 2012-02-15 03:31:20.61 43.63 -127.52 11.00 5.8 mww
124 2012-02-26 06:17:19.76 51.71 95.99 12.00 6.7 mww
125 2012-03-03 12:19:55.09 -22.14 170.34 14.00 6.6 mww
126 2012-03-07 21:12:41.29 -22.11 170.12 10.00 4.1 mb
127 2012-03-08 03:51:35.04 -17.90 167.15 28.90 5.2 mb
128 2012-03-08 10:57:43.22 61.01 -150.92 9.60 4.1 mwr
129 2012-03-09 07:09:50.95 -19.13 169.61 16.00 6.7 mww
130 2012-03-12 06:06:40.64 36.74 73.15 11.00 5.6 mww
137
131 2012-03-14 09:08:35.14 40.89 144.94 12.00 6.9 mww
132 2012-03-20 18:02:47.44 16.49 -98.23 20.00 7.4 mww
133 2012-03-21 22:15:06.13 -6.24 145.96 118.00 6.6 mww
134 2012-03-25 22:37:06.00 -35.20 -72.22 40.70 7.1 mww
135 2012-03-26 16:58:10.63 -30.04 60.65 10.00 5.6 mwb
136 2012-03-27 11:00:44.50 39.86 142.02 15.00 6.1 mwc
137 2012-04-02 17:36:42.06 16.40 -98.32 9.00 6.0 mww
138 2012-04-08 21:43:30.90 24.00 122.35 10.00 5.3 mwb
139 2012-04-10 05:09:08.41 -1.26 -13.97 10.00 5.8 mww
140 2012-04-11 08:38:36.72 2.33 93.06 20.00 8.6 mww
141 2012-04-11 10:43:10.85 0.80 92.46 25.10 8.2 mwc
142 2012-04-11 22:55:10.25 18.23 -102.69 20.00 6.5 mwb
143 2012-04-12 07:15:48.50 28.70 -113.10 13.00 7.0 mww
144 2012-04-14 10:56:19.38 -57.68 -65.31 15.00 6.2 mww

145 2012-04-14 22:05:26.43 -18.97 168.74 11.00 6.2 mww
146 2012-04-15 05:57:40.06 2.58 90.27 25.00 6.2 mww
147 2012-04-17 03:50:15.61 -32.63 -71.37 29.00 6.7 mww
148 2012-04-17 07:13:49.00 -5.46 147.12 198.00 6.8 mww
Table A.2: Slowness p and back-azimuth estimates of events detected with frequency-wavenumber analysis with a moving time window. SW=surface
wave.
# theo. Pp P Sp S SW p SW PP p PP other phases

1 289.2 1.38 283.4 - - 20.33 296.7 5.38 285.6
138
2 290.9 - - - - 28.31 308.9 - -
3 193.0 - - - - 28.19 190.9 - -
4 303.2 - - - - 26.27 296.6 - -
5 290.6 - - - - 28.33 309.3 - -
6 290.7 - - - - 27.73 305.3 - -
7 290.3 - - - - 29.21 308.3 - -
8 14.6 - - 8.17 19.8 25.78 13.0 7.04 14.5
9 291.4 - - - - 29.16 314.6 - -
10 46.2 - - - - 24.87 25.3 7.30 46.2
11 264.0 - - - - 26.92 262.1 - -
12 222.0 - - - - 26.78 216.3 - -
13 339.4 - - - - 25.32 338.1 - -
14 349.2 - - - - 25.32 347.6 - -
Appendix A. Data
15 56.7 - - - - 27.46 27.9 - -

wave.

16 15.6 - - - - 25.03 10.4 - -
17 16.3 - - - - 31.05 3.5 - -
18 43.5 - - - - 28.61 34.0 - -
19 18.9 - - - - 29.01 11.0 - -
20 44.6 - - - - 29.44 21.5 - -
21 290.7 - - - - 25.73 290.7 - -
22 225.9 - - - - 27.13 228.7 - -
23 223.6 - - - - 28.18 221.2 - -
24 308.1 1.40 280.2 - - - - - -
25 16.6 - - - - 21.21 26.4 - -
139
26 334.9 - - - - 26.25 326.5 - -
27 72.1 - - - - 28.53 29.8 5.03 360.0
28 15.1 - - - - 36.27 346.6 - -
29 343.8 - - - - 29.77 340.8 - -
30 4.5 - - - - 24.08 2.7 - -
31 288.6 - - - - 26.94 300.7 - -
32 184.9 - - - - 25.07 40.6 - -
33 172.0 - - - - 27.44 159.2 - -
34 14.7 - - - - 27.25 355.3 7.97 10.9
35 16.1 - - - - 25.60 3.0 - -
36 342.3 1.85 332.4 2.00 332.6 20.41 328.6 - -
37 342.1 - - - - 20.03 345.4 - -
38 290.9 - - - - 20.09 298.6 - -
39 16.5 - - - - 26.87 3.2 - -
wave.

40 73.8 - - - - 23.59 28.4 - -
41 97.7 - - - - 28.49 285.2 - -
42 288.9 - - - - 27.73 286.9 - -
43 241.6 6.86 240.4 11.55 239.0 22.30 207.1 10.73 222.8
44 343.5 - - - - 27.96 305.4 - -
45 53.7 3.32 47.0 2.75 59.9 28.84 27.1 5.66 64.9
46 306.6 13.60 301.2 27.98 306.2 - - - -
47 343.9 4.98 345.9 8.98 339.9 21.71 339.7 - -
48 219.3 6.05 222.4 - - 26.04 214.3 - -
49 184.7 - - - - 27.28 185.8 - -
50 221.0 - - - - 28.05 201.1 - -
140
51 336.1 1.48 340.9 - - 19.12 344.2 5.32 342.3
52 309.7 2.79 294.7 - - 26.73 309.7 - -
53 319.5 - - 10.45 315.3 27.76 314.2 - -
54 222.6 - - - - 26.31 229.4 - -
55 353.2 - - - - 29.11 338.9 - -
56 16.7 - - - - 26.96 21.3 - -
57 267.6 - - - - 25.09 291.0 - -
58 310.5 1.70 313.4 - - 27.61 212.5 5.54 291.2
59 14.4 - - - - 27.64 359.1 7.96 31.5
60 14.1 - - - - 28.93 205.5 - -
61 58.5 5.61 58.8 9.56 58.1 27.44 50.0 - -
62 344.0 - - - - 26.12 342.9 - -
Appendix A. Data
63 270.3 - - - - 28.16 269.9 - -

wave.

64 69.6 - - - - 27.98 75.6 - -
65 311.7 3.68 305.4 - - 25.38 290.5 - -
66 331.4 - - - - 28.92 329.3 - -
67 220.7 - - - - 28.64 213.2 - -
68 339.2 - - - - 28.43 338.7 - -
69 29.5 - - - - 28.54 31.4 - -
70 222.6 6.43 224.9 - - 26.29 220.1 - -
71 302.0 - - - - 27.85 299.5 - -
72 24.9 2.46 38.6 - - 27.98 35.4 - -
73 21.4 - - - - 27.87 12.7 - -
141
74 271.9 - - 28.93 278.8 - - - -
75 290.4 1.17 289.8 - - 26.07 264.6 5.84 296.8
76 69.2 7.90 66.4 14.26 70.4 27.66 65.8 - -
77 237.8 6.20 240.3 13.49 238.9 26.59 238.6 - -
78 69.5 - - - - 32.93 229.5 - -
79 122.4 - - - - 47.54 132.5 - -
80 286.2 - - - - 29.60 298.8 - -
81 293.9 - - - - 28.41 292.1 - -
82 33.0 - - - - 26.66 28.0 6.87 31.7
83 69.7 - - - - 28.09 70.4 - -
84 280.9 - - - - - - - - SKP: p=3.54 s/¶ , =53.3¶
85 283.0 - - 28.31 281.3 - - - -
86 268.0 - - - - 26.46 254.6 - -
87 228.8 6.49 234.1 - - - - 8.06 217.3
wave.

88 73.7 2.37 20.3 - - - - - -
89 14.1 - - - - 25.08 12.2 - -
90 212.9 - - - - 22.20 237.1 - -
91 14.2 3.46 14.1 - - 27.29 322.9 - -
92 175.5 - - - - 26.26 183.3 - -
93 45.3 - - - - 25.21 34.6 - -
94 225.2 5.11 229.6 - - 25.32 220.4 - -
95 271.9 - - 26.45 267.8 - - - -
96 279.2 6.45 279.5 12.74 272.1 27.64 281.5 - -
97 27.4 3.78 23.5 - - - - - -
98 235.3 - - - - 27.87 202.6 - -
142
99 235.0 - - - - 25.57 205.3 - -
100 153.2 - - - - 23.84 167.7 - -
101 36.1 - - - - 24.81 36.2 - - SS: p=14.93 s/¶ , =41.1¶
102 20.5 - - - - 25.52 34.0 7.58 15.0
103 352.5 3.53 358.1 - - 29.32 2.4 - -
104 74.3 - - 8.04 78.4 21.01 82.7 8.32 71.1 SS: p=16.44 s/¶ , =76.8¶
105 41.2 - - - - 27.15 24.6 - -
106 197.9 - - - - 27.16 190.9 - -
107 223.2 - - - - 27.03 216.9 - -
108 198.9 - - - - 26.87 174.1 - -
109 272.6 - - 10.38 263.2 26.43 266.0 - -
110 183.6 - - - - 28.33 187.2 - -
Appendix A. Data
111 221.1 - - - - 27.05 217.4 - -

wave.

112 237.7 6.48 245.5 10.35 243.5 26.62 236.1 - -
113 345.2 1.53 339.4 - - 24.68 355.7 - -
114 345.1 3.75 337.6 - - 32.76 161.5 - -
115 303.3 - - - - 27.97 306.8 - -
116 319.4 - - - - 28.39 315.7 - -
117 45.0 - - 7.67 16.7 26.09 23.5 6.77 43.1
118 45.1 - - - - 25.47 24.3 - -
119 262.0 7.08 251.2 - - 28.25 261.0 - -
120 279.9 - - - - 28.54 311.8 - -
121 16.7 - - - - 25.89 13.2 - -
143
122 1.1 - - - - 24.07 353.3 5.72 13.5
123 315.1 - - - - 27.06 314.7 - -
124 36.1 5.88 33.6 - - 28.44 35.0 - - PcP: p=4.51 s/¶ , =38.7¶
SS: p=14.55 s/¶ , =43.9¶
125 332.7 3.26 324.8 - - 23.73 344.9 - -
126 333.3 - - - - 28.55 9.3 - -
127 345.1 - - - - 26.42 353.5 - -
128 338.0 - - - - 28.25 356.8 - -
129 337.9 - - - - 27.27 346.4 - -
130 59.0 - - - - 24.45 28.3 - -
131 12.8 4.32 17.8 8.97 39.2 28.03 12.4 7.64 12.7
132 277.1 5.45 279.6 11.54 280.8 23.22 265.3 7.48 276.0
133 27.9 3.21 29.4 - - - - - -
134 221.4 5.36 221.0 - - 20.93 217.9 - -
wave.

135 120.1 - - - - 28.05 124.9 - -
136 15.1 - - - - 27.79 19.6 - -
137 277.3 6.86 274.6 - - 25.49 277.7 - -
138 37.5 - - - - 24.12 37.3 - -
139 173.1 - - - - 27.02 166.7 - - SS: p=22.59 s/¶ , =175.8¶
140 74.5 4.37 65.4 8.11 44.2 24.03 80.6 7.19 73.0 SS: p=12.93 s/¶ , =66.8¶
141 76.0 - - 8.52 33.7 24.79 75.7 7.50 72.3 SS: p=9.30 s/¶ , =81.8¶
142 281.3 6.17 280.2 11.02 269.1 26.77 271.8 - -
143 295.6 - - 11.70 298.5 27.05 297.9 7.30 302.9 PcP: p=4.84 s/¶ , =283.7¶
SKS/ScS?: p=6.31 s/¶ , =270.8¶
SS: p=16.16 s/¶ , =285.6¶
144
144 203.8 - - - - 29.16 199.0 - -
145 340.2 1.38 317.7 - - 25.16 334.2 5.02 339.7
146 76.2 - - - - 25.95 74.2 - -
147 222.5 4.97 227.4 8.93 220.9 26.90 224.2 - -
148 25.5 3.36 21.0 - - 27.27 27.2 6.73 18.3
Appendix A. Data
B. Orientation of OBS using P
phase
The following chapter contains the detailled explanation how the orientation
using P phases was done.
I assume that I know how the amplitudes of a P-phase of a specific event are
distributed on the horizontal components given that I know the backazimuth
of the event and the polarisation of the P-phase on the vertical component.
Let AZ , AN and AE be the measured amplitudes of the different components
and is the theoretical backazimuth of the event. The theoretical distribution
for the amplitudes of the North AtN and East AtE component are given by the
following equation.
I
AtE = sin
If AZ <= 0 (B.1)
AtN = cos
I
AtE = ≠ sin
If AZ > 0 (B.2)
AtN = ≠ cos
These theoretical values are now rotated around the origin in one degree steps.
ArtE (Ï) = cos Ï · AtE ≠ sin Ï · AtN (B.3)

ArtN (Ï) = sin Ï · AtE + cos Ï · AtN (B.4)
Herein, Ï is the rotation angle. By scaling the theoretical amplitudes with

the measured horizontal amplitude Ah , I am able to compare the measured
with the theoretical values to find which rotation angle I have to apply for the
right orientation of the stations’ components. The calculated deviation d (Ï)
between theoretical and measured values for each rotation angle Ï is normed
with the horizontal amplitude Ah .
Ò
Ah = A2E + A2N (B.5)
Ú
1 1 22 1 22
d (Ï) = · ArtE (Ï) · Ah ≠ AE + ArtN (Ï) · Ah ≠ AN (B.6)
Ah
The angle of rotation Ï which leads to the smallest deviation between theoretical
and measured values is the most likely one. If I do this procedure for several
time points of the P-phase and also for several events, I can calculate averaged
deviations d¯(Ï) for each angle of rotation Ï. This averaging can be improved
by weighting (with weight w) the different events according to there quality.
Therein, events which have few noise and clear wave forms which are easy to
identify get a one and events with very noisy waveforms but where I still can
identify a signal on all three components get a four. Events which are so noisy
145
Appendix B. Orientation of OBS using P phase
that I can hardly identify any signal and if yes then only on some but not all
components get a five and if there is nothing at all visible they get a six.
I use the following equation to calculate the weighted average of the devia-
tions d¯(Ï) for each rotation angle Ï for N picked events.
N
ÿ
di (Ï) · (7 ≠ wi )
d¯(Ï) = i=0
N
(B.7)
ÿ
(7 ≠ wi )
i=0
To determine the most likely angle for the orientation of the station, I have
to look for the smallest averaged deviation between theoretical and measured
values. I estimate the value of the averaged deviations between theoretical and
measured values where the value is equal to the minimum plus its standard
deviation to determine how “sharp” our estimated minimum is.
146
C. Reflection and refraction at
the ocean bottom
The following chapter contains the detailled theoretical calculations concerning

the apparent incidence angles of the SH wave, the P wave, and the SV wave in
an isotropic medium.
Usually, we are dealing with the free surface and assume that there are
no waves in the upper half space which is know as the atmosphere. If we are
working with ocean bottom data, we should take into account that water is a
more viscous fluid than air and therefore the behaviour of the wave field changes.
In this context, the question arises how this different behaviour influences the
polarisation angle (also referred to as apparent angle of incidence). First of
all, some basic principles which have already been shown by Müller (2007)
(see chapter of Reflection and refraction of plane waves at plane interfaces for
comparison).
We can express plane waves with arbitrary directions of propagation by the
following equations:
Compression:
S Q RT
˛x · ˛k bV
= A · exp UiÊ at ≠ (C.1a)
–
1 ˆ2
which satisfies — = (C.1b)
–2 ˆt2
Shear:
S Q RT
˛
˛ = B · exp UiÊ at ≠ ˛x · k bV ˛n (C.1c)
—
1 ˆ2 j
which satisfies — = (C.1d)
— 2 ˆt2
j
In this case, wavefronts are surfaces of constant phase Ê(t ≠ ˛xc·k ), herein c
˛
is 1the velocity
2 according to the wave type. So these surfaces are defined by
d
dt
t ≠ c = 0. These surfaces are perpendicular to ˛k which describes the
x·˛k
˛
direction of propagation and they move parallel to themeselves with phase

velocity c.
147
Appendix C. Reflection and refraction at the ocean bottom
The polarisation direction is defined as follows:

S Q RT
iÊ ˛x · ˛k bV ˛
Ò = ≠ A exp UiÊ at ≠ k (C.2a)
– –
S Q RT
iÊ ˛x · ˛k bV ˛
Ò ◊ ˛ = ≠ B exp UiÊ at ≠ k ◊ ˛n (C.2b)
— —
with rot (f · ˛n) = f · Ò ◊ ˛n ≠˛n ◊ Òf The additional condition of orthogonality

¸ ˚˙ ˝
here =0
of ˛k and ˛n can be introduced (seperation of ˛n in components parallel and
perpendicular to ˛k) without loss of generality.
We assume two halfspaces and that the components are independent of

y (plane problem, on all planes parallel to x-z plane the same conditions are
valid). The displacement ˛u = (u, v, w) can be written as follows:
˛u = Ò + Ò ◊ ˛ (C.3a)
ˆ ˆ 2 ˆ 1 ˆ 3 ˆ ˆ 2
u= ≠ v= ≠ w= + (C.3b)
ˆx ˆz ˆz ˆx ˆz ˆx
Note that 1 & 3 do not occur in u and w.
For v, we write the equation of motion without body forces and get the wave
equation.
1 ˆ2v
—v = 2 2 (C.4)
— ˆt
Just to remember: the stress and strain are defined as follows:
pij = ⁄◊”ij + 2µ‘ij (C.5a)

û ˆv ˆw û ˆw
◊ = ‘11 + ‘22 + ‘33 = + + = + = Ò˛u (C.5b)
ˆx ˆy ˆz ˆx ˆz
A B
1 ûi ûj
‘ij = + (C.5c)
2 ˆxj ˆxi
At the boundary z=0, the continuity of the stress components is given:

A B
ˆw ˆw û ˆv
pzz = ⁄Ò · ˛u + 2µ pzx =µ + pzy = µ (C.6)
ˆz ˆx ˆz ˆz
Some basic equations for reformulating eq.(C.6):
1 ˆ2 1 ˆ2 1 ˆ2v
— = — = —v = (C.7a)
–2 ˆt2 — 2 ˆt2 — 2 ˆt2
ˆ2 ˆ2 ˆ ˆ ˆ ˆ
with — = + u= ≠ w= + (C.7b)
ˆx2 ˆz 2 ˆx ˆz ˆz ˆx
û ˆw ˆ2 ˆ2 ˆ2 ˆ2
Ò · ˛u = + = ≠ + 2 + (C.7c)
ˆx ˆz ˆx2 ˆxˆz ˆz ˆzˆx
2 2
ˆ ˆ
= 2
+ 2 (C.7d)
ˆx ˆz
148
C.1. SH-case
The resulting stress components are:

A B
ˆw ⁄ ˆ2 ˆ2 ˆ2
pzz = ⁄— + 2µ = 2 2 + 2µ + (C.8a)
ˆz – ˆt ˆz 2 ˆzˆx
A B
2 2 2
ˆ ˆ ˆ ˆv
pzx =µ 2 + 2
≠ 2
pzy = µ (C.8b)
ˆxˆz ˆx ˆz ˆz
As we can see the equations can be splitted in two cases, one for v (SH-
waves) and one for u and w (P-/SV-waves).
The angle of incidence Ï is defined as the angle between the ray and the
normal of the interface.
˛u = (sin Ï, 0, cos Ï) (C.9)
C.1 SH-case
We assumes two halfspaces:
1. ocean floor (crust) with µ1 , —1 , fl1
2. ocean with µ2 , —2 , fl2
3. interface between crust and ocean is at
z=0
Layer 2 corresponds to the ocean therefore:
µ2 = 0 =∆ —2 = 0
For simplicity, we write:
Figure C.1: Incident and reflected
µ1 = µ · —1 = —
SH-waves
In Figure C.1, Ï1 is the reflection an-
gle.
incident wave:
C A BD
sin Ï cos Ï
v0 = C0 exp iÊ t ≠ x≠ z (C.10a)
— —
reflected wave:
C A BD
sin Ï1 cos Ï1
v1 = C1 exp iÊ t ≠ x+ z (C.10b)
— —
The refracted wave does not exist because of µ2 = 0.
The boundary conditions for z=0 is that we have a vanishing tangential

stress (pzy = 0):
ˆ
µ (v0 + v1 ) = 0 (C.11)
ˆz
This leads to the following for z=0:
A C A BD
cos Ï sin Ï
µ ≠ C0 iÊ exp iÊ t ≠ x
— —
C A BDB
cos Ï1 sin Ï1
+ C1 iÊ exp iÊ t ≠ x =0 (C.12)
— —
149
C0 and C1 should be independent of location:

sin Ï sin Ï1
= Snell’s Law =∆ Ï = Ï1 (C.13)
— —
cos Ï cos Ï1
=∆ ≠ C0 + C1 = 0
— —
with (C.13): Ï = Ï1 =∆ C0 = C1
C1 0
=∆ rss = = 1 · bss = =0 (C.14)
C0 C0
So incident and reflected wave have the same amplitude and polarisation.
We have total reflection.
150
C.2. P-SV-case
C.2 P-SV-case
We have again two halfspaces. If we take into
account what has been said in subsection C.1,
we get the following.
1. ocean floor (crust) with –1 , —, fl1 , µ, ⁄1
2. ocean with –2 , fl2 , ⁄2

z=0
In Figure C.2, Ï1 is the reflection angle Figure C.2: Incident, reflected and
of the P-wave (red), ÏÕ1 the reflection angle refracted waves
of the SV-wave (blue) and Ï2 is the refrac-
tion angle of the P-wave in the water col-
umn.
incident P-wave:
5 3 46
sin Ï cos Ï
0 = A0 exp iÊ t ≠ x≠ z (C.15a)
–1 –1
reflected P-wave:
5 3 46
sin Ï1 cos Ï1
1 = A1 exp iÊ t ≠ x+ z (C.15b)
–1 –1
refracted P-wave:
5 3 46
sin Ï2 cos Ï2
2 = A2 exp iÊ t ≠ x≠ z (C.15c)
–2 –2
reflected SV-wave:
C A BD
sin ÏÕ1 cos ÏÕ1
1 = B1 exp iÊ t ≠ x+ z (C.15d)
— —
0 + 1= f P-Potential of ocean floor

2 = w P-Potential of water column
1 = f S-Potential of ocean floor (y-component of ˛ )
The boundary conditions for z=0 are as follows:

vertical stress and displacement have to be continous, the continuity of the
tangential displacement can be neglected because of the low viscosity of water,
the tangential stress vanishs (Aki and Richards, 2002)
ˆ f ˆ f ˆ w
wf = ww =∆ + = (C.16)
ˆz ˆx A ˆz B
⁄1 ˆ 2 f ˆ2 f ˆ2 f ⁄2 ˆ 2 w
pzz,f = pzz,w =∆ 2 + 2µ + = (C.17)
–1 ˆt2 ˆz 2 ˆzˆx –22 ˆt2
A B
ˆ2 f ˆ2 f ˆ2 f
pzx,f = 0 =∆ µ 2 + ≠ =0 (C.18)
ˆxˆz ˆx2 ˆz 2
151
Derivatives
Ï Ï cos Ï1 sin Ï cos Ï

= + A0 exp iÊ t≠ sin –
x≠ cos
–
z +iÊ –
A1 exp iÊ t≠ – 1 x+ – 1 z
1 1 1 1 1 1
=iÊ ≠ –
ˆz ˆz ˆz
ˆ f ˆ 0 ˆ 1 1 cos Ï 2 Ë 1 2È 1 2 Ë 1 2È
ˆ f cos Ï sin Ï cos Ï1 sin Ï1

x + iÊ x (C.19a)
ˆz
(z = 0) = ≠ iÊ A0 exp iÊ t ≠ A1 exp iÊ t ≠
–1 –1 –1 –1
5 3 46 5 3 46
sin ÏÕ cos ÏÕ
= =iÊ ≠ — B1 exp iÊ t≠ — 1 x+ — 1 z
ˆx ˆx
ˆ f ˆ 1 1 sin ÏÕ1 2 Ë 1 2È
ˆ f sin ÏÕ1 sin ÏÕ1

x (C.19b)
ˆx — —
(z = 0) = ≠iÊ B1 exp iÊ t ≠
C A BD
sin Ï cos Ï
= =iÊ ≠ –
2
A2 exp iÊ t≠ – 2 x≠ – 2 z
2 2
ˆz ˆz
ˆ w ˆ 2 1 cos Ï2 2 Ë 1 2È
ˆ w cos Ï2 sin Ï2
x (C.19c)
152
ˆz
(z = 0) = ≠ iÊ A2 exp iÊ t ≠
–2 –2
5 3 46
ˆ2 f ˆ2 0 ˆ2 1 Ï cos Ï
= + =≠Ê 2 A0 exp iÊ t≠ sin –1
x≠ –1
z ≠Ê 2 A exp iÊ t≠ sin Ï1 x+ cos Ï1 z
1 –1 –1
ˆt2 ˆt2 ˆt2
Ë 1 2È Ë 1 2È
2
ˆ 2 sin Ï sin Ï1
2
x (C.19d)
ˆt
f (z = 0) = ≠ Ê A0 exp iÊ t ≠ x ≠ Ê 2 A1 exp iÊ t ≠
–1 –1
5 3 46 5 3 46
ˆ2 f ˆ2 0 ˆ2 1 Ï sin Ï cos Ï 2 cos Ï1 sin Ï1 cos Ï1

= + =≠Ê 2 cos–1
A0 exp iÊ t≠ –1
x≠ –1
z ≠Ê –1
A 1 exp iÊ t≠ –1
x+ –1
z
ˆz 2 ˆz 2 ˆz 2
1 22 Ë 1 2È 1 22 Ë 1 2È
2
ˆ f 2 cos Ï sin Ï 2 cos Ï1 sin Ï1
x (C.19e)
ˆz 2
(z = 0) = ≠ Ê A0 exp iÊ t ≠ x ≠Ê A1 exp iÊ t ≠
–1 –1 –1 –1
3 42 5 3 46 3 42 5 3 46
ˆ 2 f ˆ 2 1 2 sin ÏÕ1 cos ÏÕ1 sin ÏÕ cos ÏÕ

= =Ê B1 exp iÊ t≠ — 1 x+ — 1 z
—2
ˆzˆx ˆzˆx
Ë 1 2È
Õ Õ
ˆ2 f 2 sin Ï1 cos Ï1 sin ÏÕ1
(z = 0) =Ê x (C.19f)
ˆzˆx —2 —
B1 exp iÊ t ≠
C A BD
ˆ2 w ˆ2 2
2
= 2 =≠Ê2 A2 exp iÊ t≠ sin–2Ï2 x≠ cos–2Ï2 z
ˆt ˆt
Ë 1 2È
2
ˆ w 2 sin Ï2
x (C.19g)
ˆt2
(z = 0) = ≠ Ê A2 exp iÊ t ≠
–2
5 3 46
ˆ2 f ˆ2 0 ˆ2 1 Ï Ï cos Ï1 sin Ï1 sin Ï cos Ï

x≠ cos z +Ê 2
C.2. P-SV-case
= + =≠Ê 2 cos Ï 2sin Ï A0 exp iÊ t≠ sin

– –1 –1 – 2 A1 exp iÊ t≠ – 1 x+ – 1 z
1 1
ˆxˆz ˆxˆz ˆxˆz 1 1
1 2 Ë 1 2È 1 2 Ë 1 2È
2
ˆ f cos Ï sin Ï sin Ï cos Ï1 sin Ï1 sin Ï1
2
x + Ê2 2
x (C.19h)
ˆxˆz
(z = 0) = ≠ Ê 2 A0 exp iÊ t ≠ A1 exp iÊ t ≠
–1 –1 –1 –1
5 3 46 5 3 46
ˆ2 f ˆ2 1 sin ÏÕ1 sin ÏÕ1 cos ÏÕ1

= =≠Ê 2 —
B 1 exp iÊ t≠ —
x+ —
z
ˆx2 ˆx2
1 22 Ë 1 2È
Õ
ˆ2 f 2 sin Ï1 sin ÏÕ1
x (C.19i)
ˆx2 — —
(z = 0) = ≠ Ê B1 exp iÊ t ≠
A B2 C A BD
ˆ2 f ˆ2 1
153
cos ÏÕ1 sin ÏÕ cos ÏÕ
= =≠Ê 2 —
B1 exp iÊ t≠ — 1 x+ — 1 z
ˆz 2 ˆz 2
1 22 Ë 1 2È
2 Õ 2
ˆ f 2 cos Ï1 sin ÏÕ1
x (C.19j)
ˆz 2 — —
(z = 0) = ≠ Ê B1 exp iÊ t ≠
A B C A BD
Now inserting the derivatives (C.19a)-(C.19j) into (C.16)-(C.18).

vertical displacement:
ˆ f ˆ f ˆ w
(C.16): + =
ˆz ˆx ˆz
cos Ï sin Ï cos Ï1 sin Ï1
x + iÊ x
– –
=∆ ≠ iÊ A0 exp iÊ t ≠ A1 exp iÊ t ≠
–1 –1
5 3 46 5 3 46
sin ÏÕ1 sin ÏÕ1 cos Ï2 sin Ï2

x (C.20)
— —
≠ iÊ B1 exp iÊ t ≠ x = ≠iÊ A2 exp iÊ t ≠
C A 1 BD 1
–2 –2
5 3 46
As in the SH-case the Amplitudes A0 , A1 , A2 and B1 have to be location

independent.
sin Ï sin Ï1 sin ÏÕ1 sin Ï2

=∆ = = = Snell’s Law (C.21)
–1 –1 — –2
Furthermore this implies:
Ï1 = Ï (C.22a)
A B
—
ÏÕ1 = arcsin sin Ï (C.22b)
–1
3 4
–2
Ï2 = arcsin sin Ï (C.22c)
–1
with (C.21) equation (C.20) becomes:
cos Ï cos Ï1 sin ÏÕ1 cos Ï2

≠A0 +A1 ≠ B1 = ≠A2
–1 –1 — –2
cos Ï sin Ï1
Õ
cos Ï2
with (C.22a) (A1 ≠ A0 ) ≠ B1 = ≠A2 (C.23)
–1 — –2
vertical stress:
A B
⁄1 ˆ 2 f ˆ2 f ˆ2 f ⁄2 ˆ 2 w
(C.17): 2 + 2µ + = 2
with (C.21) · (C.22a)
A3 42 B
cos Ï sin ÏÕ1 cos ÏÕ1 ⁄1 ⁄2
=∆ 2µ (A0 + A1 ) ≠ 2
B1 + 2 (A0 + A1 ) = 2 A2
–1 — –1 –2
(C.24)
tangential stress:
A B
ˆ2 f ˆ2 f ˆ2 f
(C.18): µ 2 + ≠ =0
ˆxˆz ˆx2 ˆz 2
with (C.21) · (C.22a)
A B
2 cos Ï sin Ï cos2 ÏÕ1 ≠ sin2 ÏÕ1
µ (A 1 ≠ A0 ) + B1 = 0
–12 —2
with cos2 Ï ≠ sin2 Ï = cos (2Ï)
A B
2 cos Ï sin Ï cos (2ÏÕ1 )
µ (A 1 ≠ A0 ) + B1 = 0 (C.25)
–12 —2
154
C.2. P-SV-case
We now have three equations and additional conditions which need to be

fulfilled:
cos Ï sin ÏÕ1 cos Ï2
(C.23): (A1 ≠ A0 ) ≠ B1 = ≠A2
–1 — –2
A3 42 B
cos Ï sin Ï1 cos ÏÕ1
Õ
(C.24): 2µ (A0 + A1 ) ≠ B1
–1 —2
⁄1 ⁄2
+ 2 (A0 + A1 ) = 2 A2
–1 –2
A B
2 cos Ï sin Ï cos (2ÏÕ1 )
(C.25): µ (A1 ≠ A0 ) + B1 = 0
–12 —2
A B 3 4
— –2
(C.22b) · (C.22c): Ï1 = arcsin
Õ
sin Ï · Ï2 = arcsin sin Ï
–1 –1
We are interested in the amplitude ratios A 1

, B1 and A
A0 A0
2
A0
to determine the
reflection and refraction coefficients.
cos Ï sin ÏÕ1 cos Ï2 cos Ï
(C.23) ∆ A1 ≠ B1 + A2 = A0
–1 — –2 –1
A1 –1 B1 – 1 A 2
cos Ï ≠ sin ÏÕ1 + cos Ï2 = cos Ï (C.26)
A0 — A0 –2 A0
2µ cos2 Ï + ⁄1 2µ sin ÏÕ1 cos ÏÕ1 ⁄2
(C.24) ∆ 2
A 1 ≠ 2
B1 ≠ 2 A2
–1 — –2
2µ cos Ï + ⁄1
2
=≠ A0
–12
A B2 3 42
–1 B1 1 2A
1 –1 A2
(2µ sin ÏÕ1 cos ÏÕ1 ) ≠ 2µ cos2 Ï + ⁄1 + ⁄2
— A0 A0 –2 A0
2
= 2µ cos Ï + ⁄1
with sin (2Ï) = 2 sin Ï cos Ï · cos2 Ï = 1 ≠ sin2 Ï
A B2 3 42
–1 B1 1 1 2 2A
1 –1 A2
µ sin (2ÏÕ1 ) ≠ 2µ 1 ≠ sin2 Ï + ⁄1 + ⁄2
— A0 A0 –2 A0
1 2
= 2µ 1 ≠ sin2 Ï + ⁄1
with ⁄1 + 2µ = –12 fl1 · µ = — 2 fl1 · ⁄2 = –22 fl2
B1 1 2 2A
1 A2
–12 fl1 sin (2ÏÕ1 ) + 2— fl1 sin2 Ï ≠ –12 fl1 + –12 fl2
A0 A0 A0
2 2 2
= –1 fl1 ≠ 2— fl1 sin Ï (C.27)
2µ cos Ï sin Ï µ cos (2ÏÕ1 ) 2µ cos Ï sin Ï
(C.25) ∆ A1 + B1 = A0
–12 —2 –12
A B2
A1 –1 B1
2µ cos Ï sin Ï + µ cos (2ÏÕ1 ) = 2µ cos Ï sin Ï
A0 — A0
with sin (2Ï) = 2 sin Ï cos Ï
A B2
A1 –1 B1
µ sin (2Ï) + µ cos (2ÏÕ1 ) = µ sin (2Ï)
A0 — A0
A B2 3 4
B1 — sin (2Ï) A1
∆ = 1≠ (C.28)
A0 –1 cos (2Ï1 )
Õ
A0
155
We now have the following equations which need to be fulfilled:

A1 –1 B1 –1 A2
(C.26): cos Ï ≠ sin ÏÕ1 + cos Ï2 = cos Ï
A0 — A0 –2 A0
B1 1 2 2A
1 A2
(C.27): –12 fl1 sin (2ÏÕ1 ) + 2— fl1 sin2 Ï ≠ –12 fl1 + –12 fl2 = –12 fl1 ≠ 2— 2 fl1 sin2 Ï
A0 A0 A0
A B2 3 4
B1 — sin (2Ï) A1
(C.28): = 1≠
A0 –1 cos (2ÏÕ1 ) A0
Auxilliary Calculation
A B 3 4
— –2
with (C.22b) · (C.22c): = arcsin sin Ï · Ï2 = arcsin
ÏÕ1 sin Ï
–1 –1
A A BB
— —
sin Ï1 = sin arcsin
Õ
sin Ï = sin Ï (C.29a)
–1 –1
Û 3 3 44 Û 3 42
Ò
–2 –2
cos Ï2 = 1 ≠ sin Ï2 = 2
1 ≠ sin 2
arcsin sin Ï = 1≠ sin2 Ï
–1 –1
Û3 4
–2 –1 2
cos Ï2 = ≠ sin2 Ï (C.29b)
–1 –2
with (C.29a) · (C.29b) · (C.22b):
ı̂A B2
— — ı –1
sin (2ÏÕ1 ) = 2 sin ÏÕ1 cos ÏÕ1 = 2 sin Ï Ù ≠ sin2 Ï
–1 –1 —
A B2 ı̂A B2
— ı –1
sin (2ÏÕ1 ) =2 sin Ï Ù
≠ sin2 Ï (C.29c)
–1 —
with (C.29a) · (C.22b):

C A A BBD2 A B2
2 — —
cos (2ÏÕ1 ) = 1 ≠ 2 sin ÏÕ1 = 1 ≠ 2 sin arcsin sin Ï =1≠2 sin2 Ï
–1 –1
A B2 QA B2 R
— –1
cos (2ÏÕ1 ) = a ≠ 2 sin2 Ïb (C.29d)
–1 —
Inserting (C.29a)-(C.29d) in (C.26)-(C.28):
A1 –1 B1 –1 A2
(C.26): cos Ï ≠ sin ÏÕ1 + cos Ï2 = cos Ï
A0 — A0 –2 A0
Û3 42
A1 –1 — B1 –1 –2 –1 A2
∆ cos Ï ≠ sin Ï + ≠ sin2 Ï = cos Ï
A0 — –1 A0 –2 –1 –2 A0
Û3 42
A1 B1 –1 A2
cos Ï ≠ sin Ï + ≠ sin2 Ï = cos Ï
A0 A0 –2 A0
A B2 3 4
–1 –1 2
with “ = ·÷ =
— –2
A1 B1 Ò A2
cos Ï ≠ sin Ï + ÷ ≠ sin2 Ï = cos Ï (C.30)
A0 A0 A0
156
1 A2
C.2. P-SV-case
(C.27): –12 fl1 sin (2ÏÕ1 ) + 2— fl1 sin2 Ï ≠ –12 fl1 + –12 fl2 = –12 fl1 ≠ 2— 2 fl1 sin2 Ï

A0 A0 A0
B1 1 2 2A
–1 1 A2
sin Ï + –12 fl2
— —
∆–12 fl1 2 ≠ sin2 Ï + 2— fl1 sin2 Ï ≠ –12 fl1 = –12 fl1 ≠ 2— 2 fl1 sin2 Ï
A B2
A0 A0 A0
ı̂A B2
ı –1
Ù B1 1 2 2A
2
–1 — –1 A1 –1 A2
2fl1 sin Ï + 2 + fl2
R
— — — — —
≠ sin2 Ï fl1 sin2 Ï ≠
A B2 A B2 Q A B A B2 A B2
–1 A0 A0 A0
ı̂A B2
ı –1
Ù B1 a —
fl1 b
–1 —
= fl1 sin2 Ï
157
— —
fl1 ≠ 2
A B2 A B2
–1 –1 2
with “ =
—
·÷ =
A B2
–2
3 4
1 A2
2fl1 sin Ï “ ≠ sin2 Ï + 2fl1 sin2 Ï ≠ “fl1 + “fl2 = “fl1 ≠ 2fl1 sin2 Ï (C.31)
A0 A0 A0
Ò
B1 1 2A
B1 — sin (2Ï) A1
(C.28): = 1≠
A B2
A0 –1 cos (2ÏÕ1 ) A0
3 4
B1 — sin (2Ï) A1
=
—
∆ 1≠
–1
A B2
A0 –1 A0
3 4
–1 —
≠ 2 sin2 Ï
1 2 2 31 22 4
B1 sin (2Ï) A1
–1
1≠
A0 A0
3 4
—
≠ 2 sin2 Ï
= 1 22
–1 –1 2
with “ =
—
·÷ =
A B2
–2
3 4
B1 sin (2Ï) A1
= 1≠ (C.32)
A0 A0
3 4
“ ≠ 2 sin2 Ï
Inserting (C.28) in (C.26) and (C.27)
A1 –1 — sin (2Ï) A1 –1 A2
(C.28) in (C.26): cos Ï sin ÏÕ1 + cos Ï2 = cos Ï
—
≠ Õ
1≠
A B2
A0 –1 cos (2Ï1 ) A0 –2 A0
3 4
Õ — sin (2Ï) A1 –1 A2 Õ — sin (2Ï)

cos Ï + sin Ï1 + cos Ï 2 = cos Ï + sin Ï 1 (C.33)
C D
–1 cos (2ÏÕ1 ) A0 –2 A0 –1 cos (2ÏÕ1 )

—
sin (2Ï) A1 1 A2
(C.28) in (C.27): –12 fl1 sin (2ÏÕ1 ) + –12 fl2
158
Õ
1≠ + 2— 2 fl1 sin2 Ï ≠ –12 fl1 = –12 fl1 ≠ 2— 2 fl1 sin2 Ï
A B2
–1
cos (2Ï1 ) A0 A0 A0
3 1 4 2A
1
–12 fl2 + — fl1 2 sin2 Ï ≠ tan (2ÏÕ1 ) sin (2Ï) ≠ –12 fl1 = –12 fl1 ≠ — 2 fl1 tan (2ÏÕ1 ) sin (2Ï) + 2 sin2 Ï (C.34)
A0 A0
A2 Ë 2 Ë È ÈA Ë È
A1
Determination of A0
cos Ï2
(C.33) · –1 fl2 ≠ (C.34) ·
–2
Õ — sin (2Ï) 2 Õ 2 1
–1 fl2 cos Ï + sin Ï1 Õ
≠ — fl1 2 sin Ï ≠ tan (2Ï1 ) sin (2Ï) ≠ –1 fl1
I C D J
–1 cos (2Ï1 ) –2 A0
cos Ï2 Ë 2 Ë È È A
— sin (2Ï) cos Ï2 2 2 Õ 2

= –1 fl2 cos Ï + sin ÏÕ1 ≠ –1 fl 1 ≠ — fl 1 tan (2Ï 1 ) sin (2Ï) + 2 sin Ï
C D
–1 cos (2ÏÕ1 ) –2
Ë Ë ÈÈ
sin(2Ï) cos Ï2
A1 –1 fl2 cos Ï + sin ÏÕ1 –—1 cos –2
(2ÏÕ1 )
≠ [–12 fl1 ≠ — 2 fl1 [tan (2ÏÕ1 ) sin (2Ï) + 2 sin2 Ï]]
5 6
= (C.35)
A0 sin(2Ï) cos Ï2
–2
–1 fl2 cos Ï + sin ÏÕ1 –—1 cos
(2ÏÕ1 )
≠ [— 2 fl1 [2 sin2 Ï ≠ tan (2ÏÕ1 ) sin (2Ï)] ≠ –12 fl1 ]
5 6
C.2. P-SV-case
with (C.29c) · (C.29d)

— 2 –1 2
sin (2ÏÕ1 ) 2 sin Ï –1
≠ sin Ï —
2 sin Ï –—1 ≠ sin2 Ï
Õ
tan (2Ï1 ) = (C.36a)
1 2 Ú1 22 Ú1 22
cos (2ÏÕ1 ) — –1 2 –1
–1 — —
≠ 2 sin Ï ≠ 2 sin2 Ï
= 1 2 2 31 2 2 4 = 1 22
159
A1
Determination of A0
sin(2Ï) cos Ï2
A1 –1 fl2 cos Ï + sin ÏÕ1 –—1 cos –2
(2ÏÕ1 )
≠ [–12 fl1 ≠ — 2 fl1 [tan (2ÏÕ1 ) sin (2Ï) + 2 sin2 Ï]]
5 6
(C.35): =
A0 sin(2Ï) cos Ï2
–2
–1 fl2 cos Ï + sin ÏÕ1 –—1 cos
(2ÏÕ1 )
≠ [— 2 fl1 [2 sin2 Ï ≠ tan (2ÏÕ1 ) sin (2Ï)] ≠ –12 fl1 ]
5 6
with (C.29a) - C.36a):

–2 –1 2
–1 –2
≠sin2 Ï 2 sin Ï
— ( –—1 ) ≠sin2 Ï
1 2
S T Ú1 22
–1 — –2
S S Ò TT
–1 2 2 ( –—1 ) ≠2 sin2 Ï
A1 –1 ( — ) ≠2 sin Ï
–1 fl2 W X
=
Ucos Ï + sin Ï –—1 1 22 1 sin(2Ï) 2V ≠ U–2 fl1 ≠ — 2 fl1 U sin (2Ï) + 2 sin2 ÏVV
A0 –2 –1 2
–1 –2
≠sin2 Ï 2 sin Ï
— ( –—1 ) ≠sin2 Ï
2
S T Ú1 22
–1 — –2
S S Ò T T
–1 2 2 ( –—1 ) ≠2 sin2 Ï
–1 ( — ) ≠2 sin Ï
W X U— 2 fl1 U2 sin2 Ï ≠
–1 fl2 Ucos Ï + sin Ï –—1 1 22 1 sin(2Ï) 2V ≠ sin (2Ï)V ≠ –12 fl1 V
–1 2
–2
≠sin2 Ï 2 sin Ï
sin Ï sin(2Ï) ( –—1 ) ≠sin2 Ï
–1 fl2 cos Ï + 2 1 2
Ú1 22
–1
S S Ò TT
≠
( –—1 )
C D
≠2 sin2 Ï ( –—1 ) ≠2 sin2 Ï

A1
=
U–2 fl1 ≠ — 2 fl1 U sin (2Ï) + 2 sin2 ÏVV
A0 –1 2
–2
≠sin2 Ï 2 sin Ï
sin Ï sin(2Ï) ( –—1 ) ≠sin2 Ï
–1 fl2 cos Ï + 2 2
Ú1 22
–1
S S Ò T T
≠
( –—1 )
C D
≠2 sin2 Ï ( –—1 ) ≠2 sin2 Ï

U— 2 fl1 U2 sin2 Ï ≠ sin (2Ï)V ≠ –12 fl1 V
2
2 sin Ï
sin Ï sin(2Ï) –1 — ( –—1 ) ≠sin2 Ï
fl2 cos Ï + 2 –2 –1 2
S S Ò TT
≠ fl1
( –—1 )
C D
≠2 sin2 Ï ( –—1 ) ≠2 sin2 Ï

A1
Ú1 22 1 22
= 2
≠ sin2 Ï U1 ≠ U sin (2Ï) + 2 sin2 ÏVV
A0 –
sin Ï sin(2Ï)
2 sin Ï ( —1 ) ≠sin2 Ï
–1
fl2 cos Ï + 2 –2 –1 2
S S Ò T T
≠ fl1
( –—1 ) ( –—1 )
C D
≠2 sin2 Ï ≠2 sin2 Ï
160
Ú1 22 1 22
≠ sin2 Ï U — U2 sin2 Ï ≠ sin (2Ï)V ≠ 1V
–1 –1
with “ =
—
·÷ =
A B2
–2
3 42
2 sin Ï sin(2Ï) “≠sin2 Ï

Ô
sin Ï sin(2Ï) 2 1
fl2 cos Ï + “≠2 sin2 Ï
≠ fl1 ÷ ≠ sin Ï 1 ≠ “ “≠2 sin2 Ï
+ 2 sin2 Ï
5 5 66
=
Ë È Ò
2 sin Ï sin(2Ï) “≠sin2 Ï

Ô
sin Ï sin(2Ï) 2 1 2
fl2 cos Ï + “≠2 sin2 Ï
≠ fl1 ÷ ≠ sin Ï “
2 sin Ï ≠ “≠2 sin2 Ï
≠1
Ë È Ò 5 5 6 6
2 2 2 2 2 2 2 2
fl2 [(“ ≠ 2“ sin Ï) cos Ï + “ sin Ï sin (2Ï)] ≠ fl1 ÷ ≠ sin Ï “ ≠ 2“ sin Ï ≠ 2 sin Ï sin (2Ï) “ ≠ sin Ï ≠ 2 (“ ≠ 2 sin Ï) sin Ï
5 6
=
Ò Ò
2 2 2 2 2 2
fl2 [(“ 2 “2 + 2“ sin Ï
≠ 2“ sin Ï) cos Ï + “ sin Ï sin (2Ï)] ≠ fl1 ÷ ≠ sin Ï 2 (“ ≠ 2 sin Ï) sin Ï ≠ 2 sin Ï sin (2Ï) “ ≠ sin Ï ≠
Ò 5 Ò 6
with sin Ï sin (2Ï) = sin Ï2 sin Ï cos Ï = 2 sin2 Ï cos Ï
2 2 2 2 2 2 2 2 2 2
A1 fl2 [(“ ≠ 2“ sin Ï) cos Ï + 2“ sin Ï cos Ï] ≠ fl1 ÷ ≠ sin Ï “ ≠ 2“ sin Ï ≠ 4 sin Ï cos Ï “ ≠ sin Ï ≠ 2 (“ ≠ 2 sin Ï) sin Ï
5 6
=
Ò Ò
A0 fl2 [(“ 2 ≠ 2“ sin2 Ï) cos Ï + 2“ sin2 Ï cos Ï] ≠ fl1 ÷ ≠ sin2 Ï 2 (“ ≠ 2 sin2 Ï) sin2 Ï ≠ 4 sin2 Ï cos Ï “ ≠ sin2 Ï ≠ “ 2 + 2“ sin2 Ï
Ò 5 Ò 6
C.2. P-SV-case
2 2 2 2 2 2
A1 fl2 “ cos Ï ≠ fl1 ÷ ≠ sin Ï (“ ≠ 2 sin Ï) ≠ 4 sin Ï cos Ï “ ≠ sin Ï
5 6
= (C.37)
Ò Ò
A0 2 2 2 2 2 2
fl2 “ cos Ï + fl1 ÷ ≠ sin Ï 4 sin Ï cos Ï “ ≠ sin Ï + (“ ≠ 2 sin Ï)
Ò 5 Ò 6
A2
Determination of A0
161
— sin (2Ï) A1 –1 A2 — sin (2Ï)
(C.33): cos Ï + sin ÏÕ1 Õ
+ cos Ï2 = cos Ï + sin ÏÕ1
C D
–1 cos (2Ï1 ) A0 –2 A0 –1 cos (2ÏÕ1 )

with (C.29a)- (C.29d):
— sin (2Ï) A1 –1 A2
2 — sin (2Ï)
S T
+ = cos Ï +
— 2 — 2
≠ sin Ï
–1 –1
A B2 A B2
–1 A0 –2 A0 –1
Û3 42
–1 — –1 —
W X
≠ 2 sin Ï ≠ 2 sin Ï
Wcos Ï + sin Ï 1 2 2 31 22 4X sin Ï 1 2 2 31 22 4
U V
A1 –1 A2 sin Ï sin (2Ï)

+
S T
–1 –1
≠ sin2 Ï
–2 A0
Û3 42
— —
W sin Ï sin (2Ï) X
≠ 2 sin2 Ï A0 ≠ 2 sin2 Ï
Ucos Ï + 1 22 V = cos Ï + 1 22
A2 1
SQ R T
2 –1 2
∆
A0 –1 2 A0
3 4
—
2 sin Ï
–2
Wc sin Ï sin (2Ï) d A1 X
≠ sin Ï ≠
= Ú1 2 Uacos Ï + 1 22 b 1≠ V
–1 –1
with “ =
—
·÷ = · sin Ï sin (2Ï) = 2 sin2 Ï cos Ï
A B2
–2
3 42
A2 cos Ï“ A1
∆ 1≠ (C.38)
A0 A0
3 4
÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
=Ò
1
(C.34): –12 fl2 + — fl1 2 sin2 Ï ≠ tan (2ÏÕ1 ) sin (2Ï) ≠ –12 fl1 = –12 fl1 ≠ — 2 fl1 tan (2ÏÕ1 ) sin (2Ï) + 2 sin2 Ï
A0 A0
A2 Ë 2 Ë È ÈA Ë È
with (C.36a):
–1 2 –1
2 sin Ï sin (2Ï) — A1 —
2
≠ sin2 Ï
S S T T S T
–12 fl2
Ú1 22 Ú1 22
A0 –1 A0 –1
— —
A2 W W ≠ sin Ï X X W 2 sin Ï sin (2Ï) X
≠ 2 sin2 Ï ≠ 2 sin2 Ï
+W
U— 2 fl1 W
U2 sin Ï ≠ 1 22 X ≠ – 2 fl1 X
V 1 V = –12 fl1 ≠ — 2 fl1 W
U 1 22 + 2 sin2 ÏX
V
–1 2 –1 2
A2 2 sin Ï sin (2Ï) — A1 — —
≠ sin Ï
S S T T S S TT
fl2
Ú1 22 Ú1 22
162
A – –1 – 1
1≠
A B2 A B2
0 1 A0 –1
— —
W — W ≠ sin Ï X X W W 2 sin Ï sin (2Ï) XX
≠ 2 sin2 Ï ≠ 2 sin2 Ï
+ fl1 W
U
W2 sin2 Ï ≠
U 1 22 X ≠ 1X
V V = fl1 W
U
W
U 1 22 + 2 sin2 ÏXX
VV
–1 –1
with “ =
—
·÷ = · sin Ï sin (2Ï) = sin Ï2 sin Ï cos Ï = 2 sin2 Ï cos Ï
A B2
–2
3 42
A2 2
4 sin2 Ï cos Ï “ ≠ sin2 Ï A1
fl2 + fl1 2 2
S S Ò T T S S Ò TT
“ “
2 sin Ï ≠ ≠1 = fl1 1 ≠
A0 “ ≠ 2 sin Ï A0 “ ≠ 2 sin Ï
U
1U V V U
1 U 4 sin2 Ï cos Ï “ ≠ sin2 Ï
+ 2 sin2 ÏVV
A2 A1
“ 2 ≠ 2“ sin2 Ï fl2 + fl1 2 “ ≠ 2 sin2 Ï sin2 Ï ≠ 4 sin2 Ï cos Ï “ ≠ sin2 Ï ≠ “ 2 + 2“ sin2 Ï
A0 A0
1 2 1 5 2 Ò 6
2 2 2 2 2 2
= fl1 “ ≠ 2“ sin Ï ≠ 4 sin Ï cos Ï “ ≠ sin Ï ≠ 2 “ ≠ 2 sin Ï sin Ï
5 Ò 1 2 6
2 A2 2 A1
“ ≠ 2“ sin Ï fl2 + fl1 4 sin2 Ï “ ≠ sin2 Ï ≠ cos Ï “ ≠ sin2 Ï ≠ “ 2 = fl1 “ 2 ≠ 4 sin2 Ï “ + cos Ï “ ≠ sin2 Ï ≠ sin2 Ï
A0 A0
5 3 4 6 5 3 46
1 2 Ò Ò
2
fl1 “ 2 ≠ 4 sin2 Ï “ + cos Ï “ ≠ sin2 Ï ≠ sin2 Ï ≠ fl1 4 sin2 Ï “ ≠ sin2 Ï ≠ cos Ï “ ≠ sin2 Ï ≠ “ 2 = “ ≠ 2“ sin2 Ï fl2
A0 A0
5 3 Ò 46 5 3 Ò 4 6
A1 1 2 2 A
A2 fl1 2 A1 2 2 A1
∆ = 2 “ ≠ 2 sin Ï 1+ ≠ 4 sin Ï cos Ï “ ≠ sin Ï 1 ≠ (C.39)
A0 A0 A0
51 22 3 4 Ò 3 46
(“ ≠ 2“ sin2 Ï) fl2
2 2 2 2 2 2
C.2. P-SV-case
A1 fl2 “ cos Ï ≠ fl1 ÷ ≠ sin Ï (“ ≠ 2 sin ) ≠ 4 sin Ï cos Ï “ ≠ sin Ï

Ò 5 Ò 6
2 2 2 2 2
with (C.37): 1 ≠ =1≠
A0
3 4
fl2 “2 cos Ï + fl1 ÷ ≠ sin Ï 4 sin Ï cos Ï “ ≠ sin Ï + (“ ≠ 2 sin Ï)

Ò 5 Ò 6
2 2 2 2 2 2 2 2 2
fl1 ÷ ≠ sin Ï (“ ≠ 2 sin Ï) ≠ 4 sin Ï cos Ï “ ≠ sin Ï + 4 sin Ï cos Ï “ ≠ sin Ï + (“ ≠ 2 sin Ï)
5 6
=
Ò Ò Ò
2
fl2 “ 2 cos Ï + fl1 ÷ ≠ sin2 Ï 4 sin2 Ï cos Ï “ ≠ sin2 Ï + (“ ≠ 2 sin2 Ï)
Ò 5 Ò 6
2
A1 2fl1 ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
= (C.40)
163
Ò
2 2 2 2 2
1≠
A0
3 4
fl2 “2 cos Ï + fl1 ÷ ≠ sin Ï 4 sin Ï cos Ï “ ≠ sin Ï ≠ (“ ≠ 2 sin Ï)

Ò 5 Ò 6
2
A1 fl2 “ 2 cos Ï ≠ fl1 ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï) ≠ 4 sin2 Ï cos Ï “ ≠ sin2 Ï
5 6
1+ =1+
Ò Ò
A0 2 2 2 2 2
3 4

Ò 5 Ò 6
2 2 2 2 2 2 2 2 2 2
2fl2 “ cos Ï + fl1 ÷ ≠ sin Ï 4 sin Ï cos Ï “ ≠ sin Ï + (“ ≠ 2 sin Ï) ≠ (“ ≠ 2 sin Ï) + 4 sin Ï cos Ï “ ≠ sin Ï
5 6
=
Ò Ò Ò
2 2 2 2 2
Ò 5 Ò 6
2fl2 “ 2 cos Ï + 8fl1 ÷ ≠ sin2 Ï sin2 Ï cos Ï “ ≠ sin2 Ï

=
Ò Ò
2 2 2 2 2
Ò 5 Ò 6
A1 2 cos Ï fl2 “ 2 + 4fl1 sin2 Ï (÷ ≠ sin2 Ï) (“ ≠ sin2 Ï)
5 6
1+ = (C.41)
Ò
A0 2
3 4

Ò 5 Ò 6
A2
Determination of A0
(C.40) in (C.38):
2
A2 cos Ï“ 2fl1 ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
Ò
A0 ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï) fl2 “ 2 cos Ï + fl1 ÷ ≠ sin2 Ï 4 sin2 Ï cos Ï “ ≠ sin2 Ï + (“ ≠ sin2 Ï)2

=Ò Ò 5 Ò 6
164
= (C.42)
2fl1 cos Ï (“ 2 ≠ 2“ sin2 Ï)
2
Ò 5 Ò 6
(C.40) · (C.41) in (C.39):
2
A2 fl1 (“ ≠ 2 sin2 Ï) 2 cos Ï fl2 “ 2 + 4fl1 sin2 Ï (÷ ≠ sin2 Ï) (“ ≠ sin2 Ï)
Y 5 5 Ò 66
= 2
2 2 2 2 2
_
A0 2
_
]
Ò 5 Ò 6
(“ ≠ 2“ sin2 Ï) fl2 _
2
_
2 2 2 2
[ fl2 “ cos Ï + fl1 ÷ ≠ sin Ï 4 sin Ï cos Ï “ ≠ sin Ï + (“ ≠ 2 sin Ï)
4 sin Ï cos Ï “ ≠ sin Ï 2fl1 ÷ ≠ sin Ï (“ ≠ 2 sin Ï)

Ò 5 Ò 6 Z
2 2 2 2 2
≠
_
_
fl2 “2
^
Ò 5 Ò 6
_
_
cos Ï + fl1 ÷ ≠ sin Ï 4 sin Ï cos Ï “ ≠ sin Ï + (“ ≠ 2 sin Ï) \
A2 fl1 2“ 2 fl2 cos Ï (“ ≠ 2 sin2 Ï) + 8fl1 sin2 Ï cos Ï (“ ≠ sin2 Ï) (÷ ≠ sin2 Ï) (“ ≠ 2 sin2 Ï)
Y
=
Ò
2 2 2 2 2
_
A0
_
fl2 “2
]
cos Ï + fl1 ÷ ≠ sin Ï 4 sin Ï cos Ï “ ≠ sin Ï + (“ ≠ 2 sin Ï)

Ò 5 Ò 6
“fl2 _
_
[
2 2 2 2
C.2. P-SV-case
8fl1 sin Ï cos Ï (“ ≠ sin Ï) (÷ ≠ sin Ï) (“ ≠ 2 sin Ï)

Ò Z
≠
_
_
^
Ò 5 Ò 6
2 _
fl2 “ 2 cos Ï + fl1 ÷ ≠ sin2 Ï 4 sin2 Ï cos Ï “ ≠ sin2 Ï + (“ ≠ 2 sin2 Ï) _\
2 2
A2 fl1
Y Z
=
2“ fl2 cos Ï (“ ≠ 2 sin Ï)
_ _
A0
_ _
2 1
] ^
Ò 5 Ò 6
“fl2 _
_ fl “ 2 cos Ï + fl ÷ ≠ sin2 Ï 4 sin2 Ï cos Ï “ ≠ sin2 Ï + (“ ≠ 2 sin2 Ï)2 _
[ _
\
leads to similiar result as (C.42)
165
A2
(C.42): =
2fl1 cos Ï (“ 2 ≠ 2“ sin2 Ï)
A0 A2 2
fl2 “ 2 cos Ï + fl1 ÷ ≠ sin2 Ï 4 sin2 Ï cos Ï “ ≠ sin2 A0
Ï + (“ ≠ 2 sin2 Ï)
Ò Ë Ò È
B1
Determination of A0
B1 — sin (2Ï) A1 B1 sin (2Ï) A1 –1 B1 sin (2Ï) A1

(C.28): = with “ = = 2
–1 —
Õ
1≠ with (C.29d): ∆ 1≠ ∆ 1≠
A B2 A B2
A0 –1 cos (2Ï1 ) A0 A0 A0 A0 A0
3 4 3 4 3 4
—
≠ 2 sin2 Ï “ ≠ 2 sin Ï
= 1 22
with (C.40)
Ò
2
B1 sin (2Ï) 2fl1 ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
= Ò 5 Ò 6
A0 “ ≠ 2 sin2 Ï fl “ 2 cos Ï + fl ÷ ≠ sin2 Ï 4 sin2 Ï cos Ï “ ≠ sin2 Ï + (“ ≠ 2 sin2 Ï)2
2 1
Ò
2fl1 sin (2Ï) ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
= Ò 5 Ò 6
2
2 2 2 2
(C.43)
Determination of Coefficients To determine the coefficients from the

ratios of the amplitudes we need the displacement amplitudes of the incident,
reflected and refracted waves (compare (C.2a) and (C.2b)). For the incident
P-Wave it would be ≠ –iÊ1 A0 . The displacement amplitudes of the other waves
will be formed respectively. The reflection and refraction coefficients result
from the ratio of the displacement amplitude of the incident wave and the
reflected or refracted wave.
≠ –iÊ1 A1 A1
Rpp = = (C.44a)
≠ –iÊ1 A0 A0
≠ iÊ B1 –1 B1 Ô B1
Rps = —
iÊ = = “ (C.44b)
≠ –1 A0 — A0 A0
≠ –iÊ2 A2 –1 A2 Ô A2
Bpp = = = ÷ (C.44c)
≠ –iÊ1 A0 –2 A0 A0
So the reflection and refraction coefficients are real and frequency independent
for all angles of incidence Ï. Furthermore, our coefficients are identical to those
determined by Zoeppritz (1919).
Rps and Bpp are always positive. We have a look at the values of the coefficients
for Ï = 0 and Ï = fi2 .
Ï=0
Ô Ô
A1 “ 2 fl2 ≠ ÷fl1 “ 2 fl2 ≠ ÷fl1
Rpp : = 2 Ô = Ô
A0 “ fl2 + ÷fl1 “ 2 fl2 + ÷fl1
Ô Ô
A2 Ô 2“ 2 fl1 ÷ 2fl1 ÷
Bpp : · ÷= 2 Ô = Ô
A0 “ fl2 + ÷fl1 “ 2 fl2 + ÷fl1
B1 Ô
Rps : · “=0
A0
fi
Ï=
2
Ô
A1 ≠fl1 ÷ ≠ 1 [“ 2 ≠ 4 (“ ≠ 1)] (“ ≠ 2)2
Rpp : = Ô =≠ = ≠1
A0 ≠fl1 ÷ ≠ 1 (4 [“ ≠ 1) ≠ “ 2 ] (“ ≠ 2)2
A2 Ô
Bpp : · ÷=0
A0
B1 Ô
Rps : “=0
A0
166
C.2. P-SV-case
From this, we find that for an angle of incidence Ï = 0 and Ï = fi2 there exist
no reflected SV-wave. In the case of Ï = fi2 only a P-wave is reflected.
The displacement of the different wave types at the interface between crust
and ocean can be described as follows.
5 3 46
iÊ sin Ï cos Ï
u˛0 = Ò 0 =≠ A0 exp iÊ t ≠ x≠ z k˛0 (C.45a)
–1 –1 –1
5 3 46
iÊ sin Ï1 cos Ï1
u˛1 = Ò 1 = ≠ A0 Rpp exp iÊ t ≠ x+ z k˛1 (C.45b)
–1 –1 –1
5 3 46
iÊ sin Ï2 cos Ï2
u˛2 = Ò 2 = ≠ A0 Bpp exp iÊ t ≠ x≠ z k˛2 (C.45c)
–1 –2 –2
Q R
0 C A BD
c d iÊ sin ÏÕ1 cos ÏÕ1
u˛Õ1 = Ò ◊ a b = ≠ A0 Rps exp iÊ t ≠ x+ z k˛1Õ ◊ ˛n
–1 — —
0
(C.45d)
Herein, the vectors are defined as follows (compare also Fig. C.3):
Q R Q R
sin Ï sin Ï1
c d c d
k0 = a 0 b k˛1
˛ =a 0 b
cos Ï ≠ cos Ï1
Q R Q R
sin Ï2 cos ÏÕ1
k2 = a 0 d
˛ c
b k˛1Õ ◊ ˛
n =a 0 d
c
b
cos Ï2 sin ÏÕ1
Figure C.3: Polarity of P- and SV-waves
167
Due to the fact that the reflection and refraction coefficients are frequency
independent the reflected and refracted waves have always the same form as
the incident wave.
3 4
sin Ï cos Ï
u˛0 =F t≠ x≠ z k˛0 (C.46a)
–1 –1
3 4
sin Ï cos Ï
with (C.22a): u˛1 = Rpp F t ≠ x+ z k˛1 (C.46b)
–1 –1
3 4
sin Ï2 cos Ï2
u˛2 = Bpp F t ≠ x≠ z k˛2 (C.46c)
–2 –2
A B
sin ÏÕ1 cos ÏÕ1
u˛Õ1 = Rps F t ≠ x+ z k˛1Õ ◊ ˛n (C.46d)
— —
A B
—
(C.22b): ÏÕ1 = arcsin sin Ï
–1
3 4
–2
(C.22c): Ï2 = arcsin sin Ï
–1
To get the apparent angle of incidence at the interface between ocean and crust,
we need the resulting displacement of the different waves superimposing at
the interface (for z=0). We choose as boundary conditions that the vertical
displacement is continous while crossing the interface and that the continuity
of the tangential displacement can be neglected. Therefore, we just have to
consider the displacements related to the waves within the ocean bottom.
Horizontal displacement (positive in x-direction):

3 4
sin Ï
u = [(1 + Rpp ) sin Ï + Rps cos ÏÕ1 ] F t ≠ x (C.47a)
–1
Vertical displacement (positive in z-direction):
3 4
sin Ï
w = [(1 ≠ Rpp ) cos Ï + Rps sin ÏÕ1 ] F t ≠ x 00 (C.47b)
–1
The polarisation angle or apparent angle of incidence is given by:
3 4 A B
u (1 + Rpp ) sin Ï + Rps cos ÏÕ1
‘ = arctan = arctan (C.48)
w (1 ≠ Rpp ) cos Ï + Rps sin ÏÕ1
1 2
For the calculation, we can use that (1 + Rpp ) = 1 + A1
A0
and (1 ≠ Rpp ) =
1 2
1≠ A1
A0
which we already calculated (see (C.41) and (C.40)).
A B
—
with (C.22b) · (C.22c): ÏÕ1 = arcsin sin Ï
–1
—
sin ÏÕ1 = sin Ï see (C.29a)
–1
Ò ı̂ A A BB
ı —
cos ÏÕ1 = 1 ≠ sin 2
ÏÕ1 = Ù1 ≠ sin2 arcsin sin Ï
–1
ı̂ A B2 ı̂A B2
ı — — ı –1
= Ù
1≠ sin Ï = Ù2
≠ sin2 Ï (C.49a)
–1 –1 —
168
C.2. P-SV-case
Horizontal displacement (C.47a)
3 4
sin Ï
u = fu (Ï) F t ≠ x ∆ fu = (1 + Rpp ) sin Ï + Rps cos ÏÕ1
–1
ı̂A B2
— ı –1
with (C.49a) ∆ fu = (1 + Rpp ) sin Ï + Rps Ù ≠ sin2 Ï
–1 —
A B2
–1 1 Ò
with “ = ∆ fu = (1 + Rpp ) sin Ï + Rps Ô “ ≠ sin2 Ï
— “
with (C.44a) · (C.44b)
3 4
A1 B1 Ô 1 Ò
fu = 1 + sin Ï + “ Ô “ ≠ sin2 Ï
A0 A0 “
3 4 Ò
A1 B1
= 1+ sin Ï + “ ≠ sin2 Ï
A0 A0
with (C.41) · (C.43)
3 Ò 4
2 cos Ï sin Ï fl2 “ 2 + 4fl1 sin2 Ï (“ ≠ sin2 Ï) (÷ ≠ sin2 Ï)
fu = Ò 5 Ò 6
2
2 2 2 2
Ò Ò
2fl1 sin (2Ï) ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï) “ ≠ sin2 Ï
+ Ò 5 Ò 6
2
with sin (2Ï) = 2 sin Ï cos Ï
5 Ò 6
“ sin (2Ï) fl2 “ + 2fl1 (“ ≠ sin Ï) (÷ ≠ sin Ï)
2 2
fu = Ò 5 Ò 6
2
2 2 2 2
(C.50)
Vertical displacement (C.47b)
3 4
sin Ï
w = fw F t ≠ x ∆ fw = (1 ≠ Rpp ) cos Ï + Rps sin ÏÕ1
–1
—
with (C.29a) ∆ fw = (1 ≠ Rpp ) cos Ï + +Rps sin Ï
–1
A B2
–1 1
with “ = ∆ fw = (1 ≠ Rpp ) cos Ï + Rps Ô sin Ï
— “
3 4
A1 B1 Ô 1
with (C.44a) · (C.44c) ∆ fw = 1 ≠ cos Ï + “ Ô sin Ï
A0 A0 “
3 4
A1 B1
= 1≠ cos Ï + sin Ï
A0 A0
169
with (C.40) · (C.43)

Ò
2
2fl1 cos Ï ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
Ò
+ 2fl1 sin Ï sin (2Ï) ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
fw = Ò 5 Ò 6
2
2 2 2 2
with sin Ï sin (2Ï) = 2 sin2 Ï cos Ï

Ò
2fl1 cos Ï ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï) [“ ≠ 2 sin2 Ï + 2 sin2 Ï]
= Ò 5 Ò 6
2
2 2 2 2
Ò
2fl1 “ cos Ï ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
= Ò 5 Ò 6
2
2 2 2 2
(C.51)
C.2.1 Apparent angle of incidence (C.48)
3 4 A B
u (1 + Rpp ) sin Ï + Rps cos ÏÕ1
‘ = arctan = arctan
w (1 ≠ Rpp ) cos Ï + Rps sin ÏÕ1
with (C.50) · (C.51) · sin (2Ï) = 2 sin Ï cos Ï
Q 5 Ò 6R
c sin Ï fl2 “ + 2fl1 (÷ ≠ sin Ï) (“ ≠ sin Ï) d
2 2
‘ = arctan c
a Ò d
b (C.52)
fl1 ÷ ≠ sin Ï (“ ≠ 2 sin Ï)
2 2
Q Ò R
2 sin Ï “ ≠ sin2 Ï fl2 “ sin Ï
= arctan a + Ò b (C.53)
“ ≠ 2 sin Ï
2
fl1 ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)
with (C.36a), (C.29d), (C.29b) · (C.22c)
A B
fl2 tan Ï2
= arctan tan (2ÏÕ1 ) + (C.54)
fl1 cos (2ÏÕ1 )
We see from equation (C.53) that we can split the ratio of the horizontal and
the vertical displacement in a term which is identical to the expression for the
free surface (compare Müller (2007)) and a term which describes the influence
of the water column on the apparent angle of incidence. Furthermore, we also
find that in case of the free surface the apparent angle of incidence is equal to
twice the angle of the reflected SV-wave (C.54). This has already been shown
by Wiechert (1907).
testing values From a simple awk script:

Parameters:
Ë È Ë È Ë È 1 22 1 22 Ë È Ë È
–1 ms –2 ms — ms “= –1
—
÷= –1
–2
kg
fl1 m3
kg
fl2 m3
6000 1500 6000 · Ô13 3 16 2700 1000
This test shows that we will get larger dipping apparent angles of inci-
dence at the interface between ocean and crust for incidence angles smaller
170
C.2. P-SV-case
Ï ‘ Ï ‘ Ï ‘ Ï ‘
0 0 25 30.1034 50 54.7736 75 69.5485
5 6.20651 30 35.6274 55 58.6508 80 70.9763
10 12.3657 35 40.8919 60 62.0977 85 71.8409
15 18.431 40 45.8608 65 65.0828 90 72.1302
20 24.3578 45 50.4992 70 67.5758
¥65¶ and for grazing incidence smaller apparent angles of incidence (Fig. C.4a).
How the results will change if we take a sediment layer instead of a normal
crust?
Parameters:
Ë È Ë È Ë È 1 22 1 22 Ë È Ë È
–1 ms –2 ms — ms “= –1
—
÷= –1
–2
kg
fl1 m3
kg
fl2 m3
2000 1500 2000 · 14 16 16

9
2000 1000
Ï ‘ Ï ‘ Ï ‘ Ï ‘
0 0 25 21.3538 50 38.5132 75 48.8234
5 4.42456 30 25.2246 55 41.1961 80 49.8438
10 8.80928 35 28.8958 60 43.5889 85 50.4657
15 13.1157 40 32.3463 65 45.6725 90 50.6747
20 17.308 45 35.5578 70 47.4251
This test shows that we get steeper dipping apparent angles for a sedimentary
cover (Fig. C.4b).
(a) Oceanic Crust (b) Sediments
Figure C.4: Behavior of apparent angle of incidence in dependency on the angle of

incidence Ï for ocean bottom and free surface
Furthermore, the comparison of the apparent angles for the ocean bottom
and the free surface show that the apparent angles in case of an oceanic crust
are quite similiar for both types of interfaces, but in case of the sedimentary
cover we expect differences up to 20¶ .
171
Behavior of coefficients The following figures should demonstrate the

behavior of the reflection and refraction coefficients in dependency of the angle
of incidence.
(a) Rpp (b) Rps
(c) Bpp
Figure C.5: Behavior of reflection and transmission coefficients in dependency on

the angle of incidence Ï for ocean bottom
Additionally,
1 22
we have a look at the behavior of the coefficients in dependency
1 22
on “ = –—1 and ‰ = –—2 (see Fig. C.6).
Reflection and Refraction coefficients in dependency on the ray pa-

rameter p: The ray parameter p is defined by Aki and Richards (2002)
as:
sin Ï
=p (C.55)
–1
172
C.2. P-SV-case
(a) Rpp for different ‰ (b) Rpp for different “
(c) Rps for different ‰ (d) Rps for different “
(e) Bpp for different ‰ (f) Bpp for different “
1 22 1 2
— 2
Figure C.6: Behavior for different values for “ = –1
— and ‰ = –2
173
We use this to reformulated the equations (C.37), (C.42), (C.43) and (C.52) as functions of the ray parameter p.
2
A1 fl2 “ 2 cos Ï ≠ fl1 ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï) ≠ 4 sin2 Ï cos Ï “ ≠ sin2 Ï
5 6
(C.37): =
Ò Ò
A0 2 2 2 2 2 2
Ò 5 Ò 6
1
with (C.55): Ï = arcsin (p–1 ) ∆ cos Ï = –1 ≠ p2
Û
–12
fl2 1 1 1 2 2 1 1
A1 —4 –21
≠ p2 ≠ fl1 –22
≠ p2 —2
≠ 2p ≠ 4p –21
≠ p2 —2
≠ p2
Rpp = = (C.56)
Ò Ò 51 22 Ú1 21 26
A0 fl2 1 1 1 1 1
174
—4 –21
≠ p 2 + fl1 –22
≠ p2 —2
≠ 2p2 + 4p2 –21
≠ p2 —2
≠ p2
Ò Ò 51 22 Ú1 21 26
A2
(C.42): =
2fl1 cos Ï (“ 2 ≠ 2“ sin2 Ï)
A0 2
Ò 5 Ò 6
1
Û
–12
1 1
—2
2 ––12 fl1 –21
≠ p2 ≠ 2p2
÷= (C.57)
A2 Ô
Bpp =
Ò 1 2
A0 fl2 1 1 1 1 1
—4 –21
≠ p2 + fl1 –22
≠ p2 —2
≠ 2p2 + 4p2 –21
≠ p2 —2
≠ p2
Ò Ò 51 22 Ú1 21 26
B1 2fl1 sin (2Ï) ÷ ≠ sin2 Ï (“ ≠ 2 sin2 Ï)

(C.43): =
Ò
A0 2 2 2 2 2 2

Ò 5 Ò 6
1
Û
–12
1 1 1
4 –—1 fl1 p –21
≠ p2 –22
≠ p2 —2
≠ 2p2
= “= (C.58)
B1 Ô
Rps
Ò Ò 1 2
A0 fl2 1 1 1 1 1
—4 –21
p2 + fl1 –22
p2 —2
2p2 + 4p2 –21
p2 —2
p2
C.2. P-SV-case
≠ ≠ ≠ ≠ ≠
Ò Ò 51 22 Ú1 21 26
2 2
Q 5 Ò 6R
2 2
c sin Ï fl2 “ + 2fl1 (÷ ≠ sin Ï) (“ ≠ sin Ï) d
fl1 ÷ ≠ sin Ï (“ ≠ 2 sin Ï)
(C.52): ‘ = arctan c
a Ò d
b
with (C.55): Ï = arcsin (p–1 )

fl2 1 1
—2
+ 2fl1 –22
≠ p2 —2
≠ p2
Q 5
(C.59)
Ú1 21 26 R
1 1
fl1
175
—2
cp d
–22
≠ p2 ≠ 2p2
‘ = arctan c
a Ú1 21 2
d
b
C.3 SV-P-case
We have again two halfspaces. If we take into
account what has been said in subsection C.1,
we get the following.
1. ocean floor (crust) with –1 , —, fl1 , µ, ⁄1
2. ocean with –2 , fl2 , ⁄2

z=0
In Figure C.7, Ï1S is the reflection an- Figure C.7: Incident, reflected and
gle of the SV-wave (blue), ÏÕ1S the reflection refracted waves
angle of the P-wave (blue) and Ï2S is the
refraction angle of the P-wave in the water
column.
incident SV-wave:
C A BD
sin ÏS cos ÏS
0 = B0 exp iÊ t ≠ x≠ z (C.60a)
— —
reflected SV-wave:
C A BD
sin Ï1S cos Ï1S
1 = B1 exp iÊ t ≠ x+ z (C.60b)
— —
reflected P-wave:
C A BD
sin ÏÕ1S cos ÏÕ1S
1 = A1 exp iÊ t ≠ x+ z (C.60c)
–1 –1
refracted P-wave:
5 3 46
sin Ï2S cos Ï2S
2 = A2 exp iÊ t ≠ x≠ z (C.60d)
–2 –2
1 =P-Potential of ocean floor

f
2 = w P-Potential of water column
0+ 1 = f S-Potential of ocean floor (y-component of
˛)
The boundary conditions for z=0 are same as for the incident P-wave:
vertical stress and displacement have to be continous, the continuity of the
tangential displacement can be neglected because of the low viscosity of water,
the tangential stress vanishs (Aki and Richards, 2002)
ˆ f ˆ f ˆ w
(C.16):wf = ww =∆ + =
ˆz ˆx ˆzA B
2
⁄1 ˆ f ˆ2 f ˆ2 f ⁄2 ˆ 2 w
(C.17):pzz,f = pzz,w =∆ 2 + 2µ + = 2
A B
ˆ2 f ˆ2 f ˆ2 f
(C.18):pzx,f = 0 =∆ µ 2 + ≠ =0
ˆxˆz ˆx2 ˆz 2
176
Derivatives
cos ÏÕ
= =iÊ –1
1S x+
–1
1S z
ˆz ˆz
ˆ f ˆ 1 1 cos ÏÕ1S 2 Ë 1 sin ÏÕ 2È
Õ Õ
ˆ f cos Ï1S sin Ï1S
(z = 0) =iÊ x (C.61a)
ˆz
C.3. SV-P-case
A1 exp iÊ t ≠
C A BD
–1 –1
ˆ f ˆ 0 ˆ 1 sin Ï sin Ï cos Ï sin Ï sin Ï cos Ï
= + =iÊ (≠ — S )B0 exp[iÊ (t≠ — S x≠ — S z )]+iÊ (≠ — 1S )B1 exp[iÊ (t≠ — 1S x+ — 1S z )]
ˆx ˆx ˆx
ˆ f sin ÏS sin ÏS sin Ï1S sin Ï1S
x (C.61b)
ˆx — — — —
(z = 0) = ≠iÊ B0 exp iÊ t ≠ x ≠ iÊ B1 exp iÊ t ≠
C A BD C A BD
sin Ï cos Ï
= =iÊ ≠ –
2
A2 exp iÊ t≠ – 2S x≠ – 2S z
2 2
ˆz ˆz
ˆ w ˆ 2 1 cos Ï2S 2 Ë 1 2È
ˆ w cos Ï2S sin Ï2S

x (C.61c)
177
ˆz
(z = 0) = ≠ iÊ A2 exp iÊ t ≠
–2 –2
5 3 46
2 2
ˆ f ˆ 1 Õ Õ
2
= 2 =≠Ê2 A1 exp iÊ t≠ sin–Ï11S x+ cos–Ï11S z
ˆt ˆt
Ë 1 2È
2
ˆ f sin Ï Õ
1S
x (C.61d)
ˆt2
(z = 0) = ≠ Ê 2 A1 exp iÊ t ≠
C A BD
–1
ˆ2 f ˆ2 1 1S
cos ÏÕ
= =≠Ê 2 –1
1S x+
–1
1S z
ˆz 2 ˆz 2
1 cos ÏÕ 22 Ë 1 sin ÏÕ 2È
Õ
ˆ2 f 2 cos Ï1S sin ÏÕ1S
x (C.61e)
ˆz 2
(z = 0) = ≠ Ê A1 exp iÊ t ≠
A B2 C A BD
–1 –1
ˆ2 f ˆ2 0 ˆ2 1 sin ÏS cos ÏS sin Ï cos Ï sin Ï1S cos Ï1S sin Ï cos Ï
= + =≠Ê 2 B0 exp[iÊ (t≠ — S x≠ — S z )]+Ê 2 B1 exp[iÊ (t≠ — 1S x+ — 1S z )]
—2 —2
ˆzˆx ˆzˆx ˆzˆx
ˆ2 f 2 sin ÏS cos ÏS sin ÏS 2 sin Ï1S cos Ï1S sin Ï1S
x +Ê x (C.61f)
ˆzˆx —2 — —2 —
(z = 0) = ≠ Ê B0 exp iÊ t ≠ B1 exp iÊ t ≠
C A BD C A BD
ˆ2 w ˆ2 2
2
= 2 =≠Ê2 A2 exp iÊ t≠ sin–Ï22S x≠ cos–Ï22S z
ˆt ˆt
Ë 1 2È
2
ˆ w 2 sin Ï2S
x (C.61g)
ˆt2
(z = 0) = ≠ Ê A2 exp iÊ t ≠
–2
5 3 46
cos ÏÕ
= =Ê 2 A1 exp iÊ t≠ – 1S x+ – 1S z
– 1 1
ˆxˆz ˆxˆz 1
ˆ 2 f ˆ 2 1 2 1 cos ÏÕ1S sin ÏÕ1S 2 Ë 1 sin ÏÕ 2È
2 Õ Õ Õ
ˆ f cos Ï1S sin Ï1S sin Ï1S
(z = 0) =Ê 2 2
x (C.61h)
ˆxˆz
A1 exp iÊ t ≠
C A BD
–1 –1
ˆ2 f ˆ2 0 ˆ2 1 sin Ï 2 sin Ï cos Ï sin Ï 2 sin Ï cos Ï
= + =≠Ê 2 ( — S ) B0 exp[iÊ (t≠ — S x≠ — S z )]≠Ê 2 ( — 1S ) B1 exp[iÊ (t≠ — 1S x+ — 1S z )]
ˆx2 ˆx2 ˆx2
ˆ2 f 2 sin ÏS sin ÏS 2 sin Ï1S sin Ï1S
x (C.61i)
ˆx2 — — — —
(z = 0) = ≠ Ê B0 exp iÊ t ≠ x ≠Ê B1 exp iÊ t ≠
A B2 C A BD A B2 C A BD
ˆ2 f ˆ2 0 ˆ2 1 cos Ï 2 sin Ï cos Ï cos Ï 2 sin Ï cos Ï
178
2
= 2 + 2
=≠Ê 2 ( — S ) B0 exp[iÊ (t≠ — S x≠ — S z )]≠Ê 2 ( — 1S ) B1 exp[iÊ (t≠ — 1S x+ — 1S z )]
ˆz ˆz ˆz
ˆ2 f cos Ï S sin Ï S cos Ï 1S sin Ï 1S
x (C.61j)
ˆz 2 — — — —
(z = 0) = ≠ Ê 2 B0 exp iÊ t ≠ x ≠ Ê2 B1 exp iÊ t ≠
A B2 C A BD A B2 C A BD
C.3. SV-P-case
Now inserting the derivatives (C.61a)-(C.61j) into (C.16)-(C.18).

vertical displacement:
ˆ f ˆ f ˆ w
(C.16): + =
ˆz ˆx Cˆz A BD C A BD
cos Ï1S
Õ
sin ÏÕ1S sin ÏS sin ÏS
=∆ iÊ A1 exp iÊ t ≠ x ≠ iÊ B0 exp iÊ t ≠ x
–1 –1 — —
C A BD
sin Ï1S sin Ï1S
≠ iÊ B1 exp iÊ t ≠ x
— —
5 3 46
cos Ï2S sin Ï2S
= ≠iÊ A2 exp iÊ t ≠ x (C.62)
–2 –2
As in the SH-case and the P-SV-case the amplitudes B0 , B1 , A1 and A2
have to be location independent.
sin ÏS sin Ï1S sin ÏÕ1S sin Ï2S
=∆ = = = Snell’s Law (C.63)
— — –1 –2
Furthermore this implies:
Ï1S = ÏS (C.64a)
A B
–1
ÏÕ1S = arcsin sin ÏS (C.64b)
—
A B
–2
Ï2S = arcsin sin ÏS (C.64c)
—
with (C.63) equation (C.62) becomes:

cos ÏÕ1S sin ÏS sin Ï1S cos Ï2S
A1 ≠ B0 ≠ B1 = ≠A2
–1 — — –2
cos Ï1S
Õ
sin ÏS cos Ï2S
with (C.64a) A1 ≠ (B0 + B1 ) = ≠A2 (C.65)
–1 — –2
vertical stress:
A B
⁄1 ˆ 2 f ˆ2 f ˆ2 f ⁄2 ˆ 2 w
(C.17): 2 + 2µ + =
with (C.63) · (C.64a)
QA B2 R
cos ÏÕ1S sin ÏS cos ÏS ⁄1 ⁄2
=∆ 2µ a A1 + 2
(B0 ≠ B1 )b + 2 A1 = 2 A2
–1 — –1 –2
(C.66)
tangential stress:
A B
ˆ2 f ˆ2 f ˆ2 f
(C.18): µ 2 + ≠ =0
ˆxˆz ˆx2 ˆz 2
with (C.63) · (C.64a)
A B
2 cos ÏÕ1S sin ÏÕ1S cos2 ÏS ≠ sin2 ÏS
µ A 1 + (B0 + B1 ) = 0
–12 —2
with cos2 ÏS ≠ sin2 ÏS = cos (2ÏS )
A B
2 cos ÏÕ1S sin ÏÕ1S cos (2ÏS )
µ 2
A1 + (B0 + B1 ) = 0 (C.67)
–1 —2
179
We now have three equations and additional conditions which need to be

fulfilled:
cos ÏÕ1S sin ÏS cos Ï2S
(C.65): A1 ≠ (B0 + B1 ) = ≠A2
–1 — –2
QA B2 R
cos ÏÕ1S sin ÏS cos ÏS ⁄1 ⁄2
(C.66): 2µ a A1 + 2
(B0 ≠ B1 )b + 2 A1 = 2 A2
–1 — –1 –2
A B
2 cos ÏÕ1S sin ÏÕ1S cos (2ÏS )
(C.67): µ 2
A1 + (B0 + B1 ) = 0
–1 —2
A B A B
–1 –2
(C.64b) · (C.64c): Ï1S = arcsin
Õ
sin ÏS · Ï2S = arcsin sin ÏS
— —
We are interested in the amplitude ratios B 1
, A1 and B
B0 B0
A2
0
to determine the
reflection and refraction coefficients.
cos ÏÕ1S sin ÏS cos Ï2S sin ÏS
(C.65) ∆ A1 ≠ B1 + A2 = B0
–1 — –2 —
— A1 B1 — A2
cos ÏÕ1S ≠ sin ÏS + cos Ï2S = sin ÏS (C.68)
–1 B0 B0 –2 B0
2µ cos2 ÏÕ1S + ⁄1 2µ sin ÏS cos ÏS ⁄2 2µ sin ÏS cos ÏS
(C.66) ∆ 2
A1 ≠ 2
B1 ≠ 2 A2 = ≠ B0
–1 — –2 —2
A B2 A B2
B1 — 1 2A — A2
2 Õ 1
2µ sin ÏS cos ÏS ≠ 2µ cos Ï1S + ⁄1 + ⁄2
B0 –1 B0 –2 B0
= 2µ sin ÏS cos ÏS
with sin (2ÏS ) = 2 sin ÏS cos ÏS · cos2 ÏS = 1 ≠ sin2 ÏS
A B2 A B2
B1 — 1 1 2 2A — A2
1
µ sin (2ÏS ) ≠ 2µ 1 ≠ sin2 ÏÕ1S + ⁄1 + ⁄2
B0 –1 B0 –2 B0
= µ sin (2ÏS )
with ⁄1 + 2µ = –12 fl1 · µ = — 2 fl1 · ⁄2 = –22 fl2
A B2
B1 — 1 2 A1 A2
2
— fl1 sin (2ÏS ) + 2— 2 fl1 sin2 ÏÕ1S ≠ –12 fl1 + — 2 fl2
B0 –1 B0 B0
= — 2 fl1 sin (2ÏS )
Q A B2 R
B1 a — A1
fl1 sin (2ÏS ) + 2 fl1 sin2 ÏÕ1S ≠ fl1 b
B0 –1 B0
A2
+ fl2 = fl1 sin (2ÏS ) (C.69)
B0
2µ cos ÏÕ1S sin ÏÕ1S µ cos (2ÏS ) µ cos (2ÏS )
(C.67) ∆ 2
A1 + 2
B1 = ≠ B0
–1 — —2
A B2
— A1 B1
2 µ cos ÏÕ1S sin ÏÕ1S + µ cos (2ÏS ) = ≠µ cos (2ÏS )
–1 B0 B0
with sin (2ÏS ) = 2 sin ÏS cos ÏS
A B2
— A1 B1
µ sin (2ÏÕ1S ) + µ cos (2ÏS ) = ≠µ cos (2ÏS )
–1 B0 B0
A B2 3 4
A1 –1 cos (2ÏS ) B1
∆ =≠ 1+ (C.70)
B0 — sin (2Ï1S )
Õ
B0
180
C.3. SV-P-case
We now have the following equations which need to be fulfilled:

— A1 B1 — A2
(C.68): cos ÏÕ1S ≠ sin ÏS + cos Ï2S = sin ÏS
–1 B0 B0 –2 B0
Q A B2 R
B1 a — 2 A1 A2
(C.69): fl1 sin (2ÏS ) + 2 fl1 sin ÏÕ1S ≠ fl1 b + fl2 = fl1 sin (2ÏS )
B0 –1 B0 B0
A B2 3 4
A1 –1 cos (2ÏS ) B1
(C.70): =≠ 1+
B0 — sin (2Ï1S )
Õ
B0
A B A B
–1 –2
with (C.64b) · (C.64c): ÏÕ1S = arcsin sin ÏS · Ï2S = arcsin sin ÏS
— —
Ò ı̂ A A BB
ı
cos ÏÕ1S = 1 ≠ sin 2
ÏÕ1S = Ù1 ≠ sin2 arcsin –1 sin ÏS
—
ı̂ A B2
ı –1
= Ù
1≠ sin2 ÏS
—
ı̂A B2
–1 ı —
cos ÏÕ1S = Ù ≠ sin2 ÏS (C.71a)
— –1
with (C.71a) · (C.64c):
ı̂A B2
–2 ı —
cos Ï2S = Ù ≠ sin2 ÏS (C.71b)
— –2
A B2
–1
sin2 ÏÕ1S = sin2 ÏS (C.71c)
—
with (C.71a):
ı̂A B2
–1 –1 ı —
sin (2ÏÕ1S ) = 2 sin ÏÕ1S cos ÏÕ1S = 2 sin ÏS Ù ≠ sin2 ÏS
— — –1
A B2 ı̂A B2
–1 ı —
sin (2ÏÕ1S ) =2 sin ÏS Ù
≠ sin2 ÏS (C.71d)
— –1
Inserting (C.71a)-(C.71d) in (C.68)-(C.70):

— A1 B1 — A2
(C.68): cos ÏÕ1S ≠ sin ÏS + cos Ï2S = sin ÏS
–1 B0 B0 –2 B0
ı̂A B2 ı̂A B2
— –1 ı
Ù — A1 B1 — –2 ı
Ù — A2
∆ ≠ sin ÏS ≠ sin ÏS
2
+ ≠ sin2 ÏS = sin ÏS
–1 — –1 B0 B0 –2 — –2 B0
ı̂A B2 ı̂A B2
ı — A1 B1 ı — A2
Ù
≠ sin2 ÏS ≠ sin ÏS +Ù ≠ sin2 ÏS = sin ÏS
–1 B0 B0 –2 B0
A B2 A B A B2
1
— —
with ’ = = ·‰=
–1
“ –2
Ò
A1 B1 Ò A2
’ ≠ sin2 ÏS ≠ sin ÏS + ‰ ≠ sin2 ÏS = sin ÏS (C.72)
B0 B0 B0
181
— A1 A2
(C.69): fl1 sin (2ÏS ) + 2 + fl2 = fl1 sin (2ÏS )
Q A B2 R
B0 –1 B0 B0
B1 a
fl1 sin2 ÏÕ1S ≠ fl1 b
— –1 A1 A2
+ 2 fl1 + fl2 = fl1 sin (2ÏS )
R
—
∆fl1 sin (2ÏS )
Q A B2 A B2
B0 –1 B0 B0
B1 a
sin2 ÏS ≠ fl1 b
1 A2
fl1 sin (2ÏS ) + 2fl1 sin2 ÏS ≠ fl1 + fl2 = fl1 sin (2ÏS )
B0 B0 B0
B1 1 2A
B1 1 A2
fl1 sin (2ÏS ) ≠ fl1 1 ≠ 2 sin2 ÏS + fl2 = fl1 sin (2ÏS )
B0 B0 B0
1 2A
with cos (2ÏS ) = 1 ≠ 2 sin2 ÏS
182
B1 A1 A2
fl1 sin (2ÏS ) ≠ fl1 cos (2ÏS ) + fl2 = fl1 sin (2ÏS ) (C.73)
B0 B0 B0
A1 –1 cos (2ÏS ) B1
(C.70): 1 +
—
=≠
A B2
B0 sin (2ÏÕ1S ) B0
3 4
A1 –1 cos (2ÏS ) B1
1+
—
∆ =≠
A B2
B0 –1 — 2 B0
3 4
2 —
sin ÏS –1
≠ sin2 ÏS
1 22 Ú1 2
A1 cos (2ÏS ) B1
=≠ 1+
B0 — 2 B0
3 4
2 sin ÏS –1
≠ sin2 ÏS
Ú1 2
— 1 A1 cos (2ÏS ) B1
with ’ = = 1+ (C.74)
“
∆ =≠
A B2 A B
–1 B0 B0
3 4
2 sin ÏS ’ ≠ sin2 ÏS
Ò
Results obtained with Matlab function solve:
B1 fl1 cos2 (2ÏS ) ‰ ≠ sin2 ÏS + fl2 cos (2ÏS ) ’ ≠ sin2 ÏS + 2fl2 sin2 ÏS ’ ≠ sin2 ÏS ≠ 2fl1 sin (2ÏS ) sin ÏS ‰ ≠ sin2 ÏS ’ ≠ sin2 ÏS
Ò Ò Ò Ò Ò
=≠
B0 fl1 cos2 (2ÏS ) ‰ ≠ sin2 ÏS + fl2 cos (2ÏS ) ’ ≠ sin2 ÏS + 2fl2 sin2 ÏS ’ ≠ sin2 ÏS + 2fl1 sin (2ÏS ) sin ÏS ‰ ≠ sin2 ÏS ’ ≠ sin2 ÏS
Ò Ò Ò Ò Ò
2 2 2 2
C.3. SV-P-case
B1 fl1 ‰ ≠ sin ÏS 4 cos ÏS sin ÏS ’ ≠ sin ÏS ≠ cos (2ÏS ) ≠ fl2 ’ ≠ sin2 ÏS

5 6
(C.75)
Ò Ò Ò
B0 fl1 ‰ ≠ sin2 ÏS 4 cos ÏS sin2 ÏS ’ ≠ sin2 ÏS + cos2 (2ÏS ) + fl2 ’ ≠ sin2 ÏS

= Ò 5 Ò 6 Ò
A1 2fl1 cos (2ÏS ) sin (2ÏS ) ‰ ≠ sin2 ÏS

Ò
=≠
Ò Ò Ò Ò Ò
A1 fl1 sin (4ÏS ) ‰ ≠ sin2 ÏS

(C.76)
Ò
B0
183
fl1 ‰ ≠ sin2 ÏS 4 cos ÏS sin2 ÏS ’ ≠ sin2 ÏS + cos2 (2ÏS ) + fl2 ’ ≠ sin2 ÏS
=≠ Ò 5 Ò 6 Ò
A2 2fl1 (2 sin (2ÏS ) sin2 ÏS + cos (2ÏS ) sin (2ÏS )) ’ ≠ sin2 ÏS

=
Ò
Ò Ò Ò Ò Ò
A2 2fl1 sin (2ÏS ) ’ ≠ sin2 ÏS

(C.77)
Ò
B0 2 2 2 2
fl1 ‰ ≠ sin ÏS 4 cos ÏS sin ÏS ’ ≠ sin ÏS + cos (2ÏS ) + fl2 ’ ≠ sin2 ÏS
= Ò 5 Ò 6 Ò
Critical reflected and refracted waves Usually — is smaller than 1 2–1 (’ <
1), therefore we know that there exists an angle ÏS úrefl = arcsin –—1 which
is called critical angle, and that for all angles larger ÏS úrefl there exists no
reflected P-wave. Furthermore,1 if 2— is smaller than –2 (‰ < 1), we get a second
critical angle ÏS úrefr = arcsin –—2 . For every angle larger ÏS úrefr there exists
no refracted P-wave. Due to the fact that usually –1 Ø –2 it follows that ‰ Ø ’
and therefore that ÏS úrefr Ø ÏS úrefl if ÏS úrefr exists.
Determination of Coefficients To determine the coefficients from the

ratios of the amplitudes we need the displacement amplitudes of the incident,
reflected and refracted waves (compare (C.2a) and (C.2b)). For the incident
SV-Wave it would be ≠ iÊ —
B0 . The displacement amplitudes of the other waves
will be formed respectively. The reflection and refraction coefficients result
from the ratio of the displacement amplitude of the incident wave and the
reflected or refracted wave.
≠ iÊ B1 B1
Rss = —
= (C.78a)
≠ iÊ
—
B0 B0
iÊ
≠ –1 A1 — A1 Ò A1
Rsp = = = ’ (C.78b)
≠ iÊ
—
B0 –1 B0 B0
≠ –iÊ2 A2 — A2 Ô A2
Bsp = = = ‰ (C.78c)
≠ iÊ
—
B0 –2 B0 B0
So the reflection and refraction coefficients are frequency independent for all
angles of incidence ÏS . Furthermore, our coefficients are identical to those
determined by Zoeppritz (1919). The coefficients Rss , Rsp and Bpp get complex
for ÏS > ÏS úrefl (or ÏS > ÏS úrefr ). We have a look at the values of the
coefficients for ÏS = 0 and ÏS = fi2 .
ÏS = 0
Ô Ô
B1 fl1 ‰ + fl2 ’
Rss : =≠ Ô Ô = ≠1
B0 fl1 ‰ + fl2 ’
A2 Ô
Bsp : · ‰=0
B0
A1 Ò
Rsp : · ’=0
B0
fi
ÏS =
2
Ô Ô
B1 fl1 ‰ ≠ 1 + fl2 ’ ≠ 1
Rss : =≠ Ô Ô = ≠1
B0 fl1 ‰ ≠ 1 + fl2 ’ ≠ 1
A2 Ô
Bsp : · ‰=0
B0
A1 Ò
Rps : ’=0
B0
From this, we find that for an angle of incidence ÏS = 0 and ÏS = fi

2
there exist
no reflected or refracted P-wave only a SV-wave is reflected.
184
C.3. SV-P-case
The displacement of the different wave types at the interface between crust
and ocean can be described as follows.
Q R
0 C A BD
c d iÊ sin ÏS cos ÏS
u˛0 = Ò ◊ a 0b = ≠ B0 exp iÊ t ≠ x≠ z k˛0 ◊ ˛n (C.79a)
— — —
0
Q R
0 C A BD
c d iÊ sin Ï1S cos Ï1S
u˛1 = Ò ◊ a 1b = ≠ B0 Rss exp iÊ t ≠ x+ z k˛1 ◊ ˛n
— — —
0
(C.79b)
5 3 46
iÊ sin Ï2S cos Ï2S
u˛2 = Ò 2 =≠ B0 Bsp exp iÊ t ≠ x≠ z k˛2 (C.79c)
— –2 –2
C A BD
iÊ sin ÏÕ1S cos ÏÕ1S
u˛Õ1 = Ò 1 = ≠ B0 Rsp exp iÊ t ≠ x+ z k˛1Õ (C.79d)
— –1 –1
Herein, the vectors are defined as follows (compare also Fig. C.8):
Q R Q R
≠ cos ÏS cos Ï1S
c d ˛ c d
k˛0 ◊ ˛n = a 0 b k1 ◊ ˛
n =a 0 b
sin ÏS sin Ï1S
Q R Q R
sin Ï2S sin ÏÕ1S
c d c d
k2 = a 0 b k˛1Õ
˛ =a 0 b
cos Ï2S ≠ cos Ï1S
Õ
Figure C.8: Polarity of P- and SV-waves
185
Due to the fact that the reflection and refraction coefficients are frequency
independent the reflected and refracted waves have always the same form as
the incident wave.
A B
sin ÏS cos ÏS
u˛0 =G t≠ x≠ z k˛0 ◊ ˛n (C.80a)
— —
A B
sin ÏS cos ÏS
with (C.64a): u˛1 = Rss F t ≠ x+ z k˛1 ◊ ˛n (C.80b)
— —
3 4
sin Ï2S cos Ï2S
u˛2 = Bsp F t ≠ x≠ z k˛2 (C.80c)
–2 –2
A B
sin ÏÕ1S cos ÏÕ1S
u˛Õ1 = Rsp F t ≠ x+ z k˛1Õ (C.80d)
–1 –1
A B
–1
(C.64b): ÏÕ1S = arcsin sin ÏS
—
A B
–2
(C.64c): Ï2S = arcsin sin ÏS
—
To get the apparent angle of incidence at the interface between ocean and crust,
we need the resulting displacement of the different waves superimposing at
the interface (for z=0). We choose as boundary conditions that the vertical
displacement is continous while crossing the interface and that the continuity
of the tangential displacement can be neglected. Therefore, we just have to
consider the displacements related to the waves within the ocean bottom.
Horizontal displacement (positive in x-direction):

A B
sin ÏS
u = [(Rss ≠ 1) cos ÏS + Rsp sin ÏÕ1S ] F t ≠ x (C.81a)
—
Vertical displacement (positive in z-direction):
A B
sin ÏS
w = [(Rss + 1) sin ÏS + ≠Rsp cos ÏÕ1S ] F t ≠ x (C.81b)
—
The polarisation angle or apparent angle of incidence is given by:
3 4 A B
u (Rss ≠ 1) cos ÏS + Rsp sin ÏÕ1S
› = arctan = arctan (C.82)
w (Rss + 1) sin ÏS ≠ Rsp cos ÏÕ1S
A B
–1
with (C.64b) · (C.64c): ÏÕ1S = arcsin sin ÏS
—
–1
sin ÏÕ1S = sin ÏS (C.83a)
—
Ò ı̂A B2
–1 ı —
cos ÏÕ1S = 1 ≠ sin ÏÕ1S
2
= Ù ≠ sin2 ÏS see (C.71a)
— –1
186
C.3. SV-P-case
Horizontal displacement (C.81a)

A B
sin ÏS
u = fu (ÏS ) F t ≠ x ∆ fu = (Rss ≠ 1) cos ÏS + Rsp sin ÏÕ1S
—
–1
with (C.83a) ∆ fu = (Rss ≠ 1) cos ÏS + Rsp sin ÏS
—
A B2 A B
— 1 1
with ’ = = ∆ fu = (Rss ≠ 1) cos ÏS + Rsp Ô sin ÏS
–1 “ ’
with (C.78a) · (C.78b)
3 4
B1 A1 Ò 1
fu = ≠ 1 cos ÏS + ’ Ô sin ÏS
B0 B0 ’
3 4
B1 A1
fu = ≠ 1 cos ÏS + sin ÏS (C.84)
B0 B0
Vertical displacement (C.81b)

A B
sin ÏS
w = fw F t ≠ x ∆ fw = (Rss + 1) sin ÏS ≠ Rsp cos ÏÕ1S
—
with (C.49a)
ı̂A B2
–1 ı —
∆fw = (Rss + 1) sin ÏS ≠ Rsp Ù ≠ sin2 ÏS
— –1
A B2 A B
— 1
with ’ = =
–1 “
1 Ò
∆fw = (Rss + 1) sin ÏS ≠ Rsp Ô ’ ≠ sin2 ÏS
’
with (C.78a) · (C.78b)
3 4
B1 A1 Ò 1 Ò
fw = + 1 sin ÏS ≠ ’ Ô ’ ≠ sin2 ÏS
B0 B0 ’
3 4
B1 A1 Ò
fw = + 1 sin ÏS ≠ ’ ≠ sin2 ÏS (C.85)
B0 B0
187
Auxilliary Calculations:
with (C.75)
2 2 2 2
Ò 5 Ò 6 Ò
2 2 2 2
≠1
B0 2
3 4
fl1 ‰ ≠ sin ÏS 4 cos ÏS sin ÏS ’ ≠ sin ÏS + cos (2ÏS ) + fl2 ’ ≠ sin ÏS

≠1 = Ò 5 Ò 6 Ò
2 2 2 2
fl1 ‰ ≠ sin ÏS 4 cos ÏS sin ÏS ’ ≠ sin ÏS ≠ cos (2ÏS ) ≠ fl2 ’ ≠ sin2 ÏS
5 6
=
Ò Ò Ò
188
Ò 5 Ò 6 Ò

Ò 5 Ò 6 Ò
2 2 2
≠
fl1 ‰ ≠ sin ÏS 4 cos ÏS sin ÏS ’ ≠ sin ÏS + cos2 (2ÏS ) + fl2 ’ ≠ sin2 ÏS
Ò 5 Ò 6 Ò
2 2 2
B1 2 fl1 ‰ ≠ sin ÏS cos (2ÏS ) + fl2 ’ ≠ sin ÏS
3 4
(C.86)
Ò Ò
B0
3 4

≠1 =≠ Ò 5 Ò 6 Ò
with (C.75)
2 2 2 2
5 6
+1
Ò Ò Ò
B0 2 2 2 2 2
3 4
fl1 ‰ ≠ sin ÏS 4 cos ÏS sin ÏS ’ ≠ sin ÏS + cos (2ÏS ) + fl2 ’ ≠ sin ÏS

+1 = Ò 5 Ò 6 Ò
2 2 2 2
Ò 5 Ò 6 Ò
B0
3 4

+1 = Ò 5 Ò 6 Ò

C.3. SV-P-case
5 6
+
Ò Ò Ò
2 2 2
Ò 5 Ò 6 Ò
8fl1 ‰ ≠ sin2 ÏS cos ÏS sin2 ÏS ’ ≠ sin2 ÏS

= (C.87)
Ò Ò

Ò 5 Ò 6 Ò
189
Horizontal displacement (C.84)
B1 A1
fu = ≠ 1 cos ÏS + sin ÏS
B0 B0
3 4
with (C.76) · (C.86)

2 2 2
2 cos ÏS fl1 ‰ ≠ sin ÏS cos (2ÏS ) + fl2 ’ ≠ sin ÏS
3 Ò Ò 4
2 2 2
=≠
Ò 5 Ò 6 Ò
fl1 sin ÏS sin (4ÏS ) ‰ ≠ sin2 ÏS

Ò
2 2 2
≠
Ò 5 Ò 6 Ò
with sin (4ÏS ) = 2 sin (2ÏS ) cos (2ÏS )
2 2
2 fl1 cos (2ÏS ) ‰ ≠ sin ÏS [sin ÏS sin (2ÏS ) + cos ÏS cos (2ÏS )] + cos ÏS fl2 ’ ≠ sin ÏS
3 Ò Ò 4
=≠
Ò 5 Ò 6 Ò
with sin (2ÏS ) = 2 sin ÏS cos ÏS · cos (2ÏS ) = 1 ≠ 2 sin2 ÏS

2 2
2 cos ÏS fl1 cos (2ÏS ) ‰ ≠ sin ÏS + fl2 ’ ≠ sin ÏS
3 4
(C.88)
Ò Ò
fu = ≠
Ò 5 Ò 6 Ò
190
Vertical displacement (C.85)
B1
fw = + 1 sin ÏS ≠ ’ ≠ sin2 ÏS
B0 B0
3 4
A1 Ò
with (C.76) · (C.87)

8fl1 sin ÏS ‰ ≠ sin2 ÏS cos ÏS sin2 ÏS ’ ≠ sin2 ÏS
=
Ò Ò

Ò 5 Ò 6 Ò
fl1 ’ ≠ sin2 ÏS sin (4ÏS ) ‰ ≠ sin2 ÏS

+
Ò Ò
2 2 2
cos2
fl1 ‰ ≠ sin ÏS 4 cos ÏS sin ÏS ’ ≠ sin ÏS + (2ÏS ) + fl2 ’ ≠ sin2 ÏS

Ò 5 Ò 6 Ò
with sin (4ÏS ) = 2 sin (2ÏS ) cos (2ÏS ) · sin (2ÏS ) = 2 sin ÏS cos ÏS
2fl1 sin (2ÏS ) ‰ ≠ sin2 ÏS ’ ≠ sin2 ÏS [2 sin2 ÏS + cos (2ÏS )]
fw =
Ò Ò
2 2 2
Ò 5 Ò 6 Ò
C.3. SV-P-case
with cos (2ÏS ) = 1 ≠ 2 sin2 ÏS

2fl1 sin (2ÏS ) ‰ ≠ sin2 ÏS ’ ≠ sin2 ÏS
fw = (C.89)
Ò Ò
2 2 2 2
fl1 ‰ ≠ sin ÏS 4 cos ÏS sin ÏS ’ ≠ sin ÏS + cos2 (2ÏS ) + fl2 ’ ≠ sin ÏS
Ò 5 Ò 6 Ò
191
C.3.1 Apparent angle of incidence (C.82)
with (C.88), (C.89) · sin (2ÏS ) = 2 sin ÏS cos ÏS

Q Ò Ò R
fl1 cos (2ÏS ) ‰ ≠ sin ÏS + fl2 ’ ≠ sin2 ÏS
2
› = arctan a≠ Ò Ò b (C.90)
2fl1 sin ÏS ‰ ≠ sin ÏS ’ ≠ sin ÏS
2 2
Q R
cos (2ÏS )
fl2 1
= arctan a≠ Ò ≠ Ò b (C.91)
2 sin ÏS ’ ≠ sin2 ÏS fl1 2 sin ÏS ‰ ≠ sin2 ÏS
with (C.71d) · (C.64a)
A B
cos (2ÏS ) fl2 1
= arctan ≠ ≠ (C.92)
’ sin (2Ï1S ) fl1 ‰ sin (2Ï2S )
Õ
As for the P-SV-case, we can split the ratio of the horizontal and the vertical
displacement (see equation (C.92) in a term which is identical to the expression
for the free surface (see Wiechert (1907) for comparison) and a second term
which describes the influence of the water column on the apparent incidence
angle. A similar expression has already been given by Zoeppritz (1919).
Wiechert (1907) showed that the the apparent incidence angle › as calulated
by equation (C.82) is negative, because the transversal nature of the wave is
ignored and the apparent incidence angle is estimated as if it is a longnitudal
wave. We have to consider the apparent incidence angle › Õ = 90¶ + › to account
for the transversal character of the wave.
with › Õ = 90¶ + ›
1
tan › Õ = tan (90¶ + ›) with tan (90¶ + ›) = ≠ cot › = ≠ (C.93)
tan ›
w
with (C.82) ∆ tan › Õ = ≠ with (C.88), (C.89)
Q Òu Ò R
2fl1 sin ÏS ‰ ≠ sin2 ÏS ’ ≠ sin2 ÏS
∆ › = arctan a
Õ
Ò Ò b (C.94)
fl1 cos (2ÏS ) ‰ ≠ sin ÏS + fl2 ’ ≠ sin ÏS
2 2
testing values From a simple awk script:

Parameters:
Ë È Ë È Ë È 1 2 1 2 Ë È Ë È
— 2 — 2
–1 ms –2 ms — ms ’= –1
‰= –2
kg
fl1 m3
kg
fl2 m3
6000 1500 6000 · Ô13 1

3
16
3
2700 1000
Ï ›Õ Ï ›Õ
0 0 20 20.7936
5 5.30185 25 25.283
10 10.5775 30 27.9021
15 15.7797 35 12.1076
The critical angle lies at ÏS úrefl = arcsin ’ = 35.3¶ , therefore we only give the
apparent angle of incidence up to 35¶ . Afterwards, all coefficients get complex
192
C.3. SV-P-case
and we will get interface waves which travel along the ocean bottom. This
makes the theoretical estimation of the apparent incidence angle complicated.
How the results will change if we take a sediment layer instead of a normal
crust?
Parameters:
Ë È Ë È Ë È 1 2 1 2 Ë È Ë È
— 2 — 2
–1 ms –2 ms — ms ’= –1
‰= –2
kg
fl1 m3
kg
fl2 m3
2000 1500 2000 · 14 1

16
1
9
2000 1000
Ï ›Õ
0 0
5 1.72037
10 2.82628
14 1.70523
In this case, the critical angle is even smaller and lies at ÏS úrefl = arcsin ’ =
14.5¶ , therefore we only give the apparent angle of incidence up to 14¶ . After-
wards, all coefficients get complex (see Figs. C.10 and C.11 where kinks and
dashed lines mark the critical angles) and we will get interface waves which
travel along the ocean bottom. This makes the theoretical estimation of the
apparent incidence angle more complicated.
(a) Oceanic Crust (b) Sediments
Figure C.9: Behavior of apparent angle of incidence in dependency on the angle of

incidence Ï for ocean bottom and free surface in case of an incident SV wave
193
Figure C.10: Behavior of absolute value of reflection and refraction coefficients at

the ocean bottom in case of an incident SV-wave (normal crust)
Figure C.11: Behavior of absolute value of reflection and refraction coefficients at

the ocean bottom in case of an incident SV-wave (sedimentary cover)
194
Bibliography
Bibliography
Books, 2nd edition.

Dahm, T. (2012). Geophysikalische Inversionsprobleme.
Dahm, T., Thorwart, M., Flueh, E. R., Braun, T., Herber, R., Favali, P.,
Beranzoli, L., D’Anna, G., Frugoni, F., and Smriglio, G. (2002). Ocean bot-
tom seismometers deployed in Tyrrhenian Sea. Eos, Transactions American
Müller, G. (2007). Theory of elastic waves. GeoForschungsZentrum Potsdam,

Potsdam, Frankfurt, Hamburg.

41:43–63.
Wiechert, E. (1907). Über Erdbebenwellen. Part I: Theoretisches über die

Ausbreitung der Erdbebenwellen. Nachrichten von der Gesellschaft der
Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse, (1907):415–
529.

195
Bibliography
196
D. Acknowledgement
I thank Frank Krüger and Torsten Dahm for having the idea to initialize the
DOCTAR project and for their willingness to give me the oppertunity to be
their PhD student. During the course of this thesis, they teached me a lot about
seismology and how to work scientificly on my own. We had many fruitful
discussion during the last four and a half years and I enjoyed that they always
shared their knowledge and experiences with me. They always supported me in
achieving the next goal and always took the time to answer questions in great
detail. I especially thank Frank Krüger for supporting my interest in teaching.
I thank Heidrun Kopp and Dietrich Lange from the GEOMAR Helmholtz
Centre for Ocean Research Kiel for giving me the opportunity to work with
them in Kiel, but also giving me time to finish this thesis.
I thank the Potsdam Graduate School for the financial support for two
month with a stipendia and for offering several helpful workshops.
I also thank my family for their support through all the years and their
believe in me. I am really lucky to be a part of such a great family. I always
know that no matter what happens I will always have a place which I can call
home and at which people wait for me and will be there for me.
I thank Niklas Allroggen whom I met at the beginning of my PhD thesis
and with whom I shared a lot of happy and sad moments since then. He always
is there to lend a ear, to give advices or simply an encouragement if I need it.
I also thank my colleagues and former colleagues at the University Potsdam
and especially the general geophysics working group. I thank Matthias Ohrn-
berger for introducing me to this great working group, for being my mentor
during the Junior Teaching Program and for always having time for answering
questions. I thank Brigitte Knapmeyer-Endrun for taking the time to answer
questions concering receiver functions and for the time in which we shared an
office. I thank Frank Scherbaum during whoms lecture in the third semester I
realized that I want to become a geophysicst and who put the trust in me to
give the short course on Digital Seismology which was one of the toughest but
also one of the best experiences I made during my PhD studies. I thank all
those wonderful persons I had the oppertunity to share an office with (Nico-
las Kühn, Sara Mata Sosa, Lilian Blaser, Stefanie Donner, Galina Kulikova,
Annabel Händel (and Alena Händel), Marius Kriegerowski, Conny Hammer,
Antonia Runge, Martin Zeckra). I also thank Daniel Vollmer for being such an
optimistic person, a great technican and a great support in the field. I thank
Helmut Staedtke for reworking the QSEISR and QSEISS code to be run on
several cores. I also thank the remaining members and former members of
the general geophysics working group (Agostiny Lontsi, Christian Molkenthin,
Sanjay S. Bora, Kristin Vogel, Carsten Riggelsen, Nikolaos Gianniotis and
everybody I might have missed). I thank Uwe Altenberger for having time to
197
Appendix D. Acknowledgement
discuss mineralogical questions and for encouraging me to further practice Qi

Gong. I also thank all the Bachelor, Master and Diploma students I have met
during my time at the University Potsdam, even while teaching I learned a lot
about geophysics, communication or how to teach.
I thank my friends for being there for me and for sharing the joys and
sorrows which nearly everybody faces during a PhD study.
198
E. Eidesstattliche Erklärung
Hiermit versichere ich, dass ich die vorliegende Doktorarbeit selbstständig

verfasst und keine anderen als die angegebenen Quellen und Hilfsmittel be-
nutzt habe, alle Ausführungen, die anderen Schriften wörtlich oder sinngemäß
entnommen wurden, kenntlich gemacht sind und die Arbeit in gleicher oder
ähnlicher Fassung noch nicht Bestandteil einer Studien- oder Prüfungsleistung
war.
——————————————–
Bestätigung des Doktoranden
199

Hannemann Thesis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hannemann Thesis

Uploaded by

Copyright:

Available Formats

Institut für Erd- und Umweltwissenschaften

Seismological investigation of the oceanic crust and upper

Potsdam, den 21.09.2016

4 Measuring clock drift and time offset at OBS 21

5 Structure of oceanic crust and upper mantle by OBS RF 37

6 Oceanic lithospheric vs from the analysis of Ïp 69

B Orientation of OBS using P phase 145

C Reflection and refraction at the ocean bottom 147

E Eidesstattliche Erklärung 199

3.1 Challenges at the sea floor

In order to analyse earthquake data, we need to know the orientation of the

3.2 Structure of the oceanic crust and upper

The depths of deeper discontinuities in the upper mantle are known by

3.3 Research questions

station positions 30.0 20.0 10.0

towards Gloria fault

in terms of S wave velocity depth profiles (e.g. Svenningsen and Jacobsen,

Based on this, we used teleseismic P and Rayleigh wave recordings to estimate

Bird, P. (2003). An updated digital model of plate boundaries. Geochemistry,

Bormann, P. and Wielandt, E. (2013). Seismic Signals and Noise. In Bormann,

Bürgmann, R. and Chadwell, D. (2014). Seafloor Geodesy. Annual Review of

Castagna, J. P., Batzle, M. L., and Eastwood, R. L. (1985). Relationships

Crawford, W. C. and Webb, S. C. (2000). Identifying and Removing Tilt Noise

Crawford, W. C., Webb, S. C., and Hildebrand, J. A. (1998). Estimating

Deuss, A. and Woodhouse, J. H. (2002). A systematic search for mantle

Dziewonski, A. M. and Anderson, D. L. (1981). Preliminary reference Earth

Estabrook, C. H. and Kind, R. (1996). The Nature of the 660-Kilometer Upper-

Flanagan, M. P. and Shearer, P. M. (1998). Global mapping of topography on

Gao, H. and Schwartz, S. (2015). Preface to the Focus Section on Cascadia

Gu, Y. J. and Dziewonski, A. M. (2002). Global variability of transition zone

Gu, Y. J., Dziewonski, A. M., and Agee, C. B. (1998). Global de-correlation

Hannemann, K., Krüger, F., and Dahm, T. (2016). Oceanic lithospheric S

Hasselmann, K. (1963). A statistical analysis of the generation of microseisms.

Helffrich, G. (2000). Topography of the transition zone seismic discontinuities.

Hirose, K. (2002). Phase transitions in pyrolitic mantle around 670-km depth:

Karato, S.-i. (1992). On The Lehmann Discontinuity. Geophysical Research

Kennett, B. L. N., Engdahl, E. R., and Buland, R. (1995). Constraints

Langston, C. A. C. A. (1979). Structure under Mount Rainier, Washington,

Lawrence, J. F. and Shearer, P. M. (2006). A global study of transition

Lehmann, I. (1961). S and the Structure of the Upper Mantle. Geophysical

Longuet-Higgins, M. S. (1950). A Theory of the Origin of Microseisms. Philo-

Rost, S. and Thomas, C. (2002). Array seismology: Methods and applications.

Simmons, N. A. and Gurrola, H. (2000). Multiple seismic discontinuities near

Suetsugu, D. and Shiobara, H. (2014). Broadband Ocean-Bottom Seismology.

Svenningsen, L. and Jacobsen, B. H. (2007). Absolute S-velocity estimation

Thomas, C. and Laske, G. (2014). D” observations in the Pacific from PLUME

Vinnik, L. P. (1977). Detection of waves converted from P to SV in the mantle.

Weidner, D. J. and Wang, Y. (1998). Chemical- and Clapeyron-induced

White, R. S. (1984). Atlantic oceanic crust: seismic structure of a slow-spreading

White, R. S., McKenzie, D., and O’Nions, R. K. (1992). Oceanic crustal

K. Hannemann, F. Krüger, T.Dahm

Figure 4.1: Configuration of OBS deployment

station tskew [s] d [s a≠1 ]

We checked the relative timing between stations with strong teleseismic

The cross-correlation in a time window of length Tw = tE ≠ tB , where tB is

Herein, f1 is the signal measured at sensor 1 and f2 the signal measured at

f2 (t) = f2 (tÕ ≠ tB · ”d) (4.5)

is inserted in equation (4.1).

Herein, tdif f0 is the clock difference at T0 = 0. This parameter is also

4.4.2 Constant time shifts

tm = Tcorr · ”d , tth = Tcorr · ”dapp (4.7)

Herein, tm is the estimated clock difference from the correlations which

4.4.3 Shifted correlation time window approach

4.4.4 Processing of data set