Institut de Physique du Globe de Paris · Sorbonne Paris Cité

École doctorale des Sciences de la Terre

en cotutelle avec
Alma Mater Studiorum · Università di Bologna

THÈSE

présentée par
Lucia GUALTIERI
pour l’obtention du titre de
Docteur de l’Institut de Physique du Globe de Paris
et de l’Alma Mater Studiorum - Università di Bologna
Spécialité: Géophysique

Modeling the secondary microseismic noise

generation and propagation

soutenue le vendredi 27 Juin 2014 devant le jury composé de:

Mme. Ana FERREIRA (University College London) . . . . . . . . . . . . . . . . Rapporteur

M. Gilberto SACCOROTTI (INGV, Pisa) . . . . . . . . . . . . . . . . . . . . . . . . . . . Rapporteur
M. Michel CAMPILLO (Université J. Fourier, Grenoble) . . . . . . . . . . . . . . . . Membre
M. Jean-Paul MONTAGNER (IPG Paris) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Membre
M. Marco MUCCIARELLI (Università della Basilicata) . . . . . . . . . . . . . . . . . Membre
M. Daniele SPALLAROSSA (Università di Genova) . . . . . . . . . . . . . . . . . . . . . Membre
Mme. Eléonore STUTZMANN (IPG Paris) . . . . . . . . . . . . . . . . . . .Directrice de thèse
M. Andrea MORELLI (Università di Bologna) . . . . . . . . . . . . . Co-directeur de thèse

Institut de Physique du Globe de Paris - Équipe de Sismologie

1 rue Jussieu, 75238 Paris Cedex 05, France
[email protected]
 “Under the water it rumbled on.”
S. T. Coleridge, The Rime of the Ancient Mariner

3
4
 Abstract
Seismic noise is the continuous oscillation of the Earth recorded worldwide, in
response to the interaction amongst the atmosphere, the ocean and the solid Earth.
The secondary microseismic noise is the strongest signal, generated by the non-linear
interaction among ocean gravity waves, at periods smaller than 12 s.
In this thesis, we deal with the wavefield, the generation and the propagation of the
secondary microseismic noise. Noise sources are schematized as vertical single forces
along the ocean surface, whose amplitude is derived from a realistic ocean wave model.
The amplitude of the three-components of noise spectra is modeled using normal mode
summation. The fundamental mode of Rayleigh waves is the dominant signal. The
discrepancy between real and synthetic spectra on the horizontal components enables
to estimate the amount of Love waves, for which a different source mechanism is
needed. The ocean site effect is computed on Rayleigh waves using normal modes,
in order to study the effect of ocean depth, crust layering and seafloor sediments on
noise amplitude. The ocean site effect on body waves is also computed by defining
the wavefield as the superposition of plane waves. The site effect varies strongly with
period and ocean depth, although in a different way for body waves than for the
different modes of Rayleigh waves, amplifying different source regions at different pe-
riods. Finally, the effect of the bathymetry on the seismic noise wavefield is presented
for varying periods using the spectral-element method. A secondary virtual source is
generated by the bathymetry, which affects the seismic wavefield.

5
 Le bruit sismique est l’oscillation continue enregistrée partout sur la Terre, généré
par l’interaction entre l’atmosphère, l’océan et la terre solide. Le bruit microsismique
secondaire est le signal le plus fort. Il est dû à l’interaction non linéaire entre les
vagues océaniques, à des périodes inférieures à 12 s .
Cette thèse a pour objet l’étude du champ d’onde, la génération et la propagation
du bruit microsismique secondaire. Les sources du bruit sont schématisées par des
forces verticales sur la surface de l’océan, dont l’amplitude est dérivée d’un modèle
de vagues oceanique réaliste. L’amplitude des trois composantes du spectre du bruit
est modélisée par sommation de modes normaux. Le mode fondamental des ondes de
Rayleigh est le signal dominant. L’écart entre les spectres réels et synthétiques des
composantes horizontales permet d’estimer la quantité d’énergie associée aux ondes
de Love, pour lesquelles un mécanisme de source différent est nécessaire. L’effet de site
océanique est calculé pour les ondes de Rayleigh en utilisant les modes normaux, afin
d’étudier l’effet sur l’amplitude du bruit de la profondeur de l’océan, de la croûte et
des sédiments au-dessous du fond océanique. L’effet de site océanique sur les ondes de
volume est également calculé en décrivant le champ d’onde comme la superposition
d’ondes planes. L’effet de site varie fortement avec la période et la profondeur de
l’océan, mais d’une manière différente pour les ondes de volume et les différents
modes d’ondes de Rayleigh, amplifiant les sources dans des régions différentes à des
différentes périodes. Enfin, l’effet de la bathymétrie sur le bruit sismique est présenté
pour différentes périodes en utilisant la méthode des éléments spectraux. Une source
virtuelle secondaire est générée par la bathymétrie et le champ d’onde sismique est
modifié par cette source.

6
 Il rumore sismico è l’oscillazione continua del suolo, ovunque registrata, causata
dall’interazione tra l’atmosfera, l’oceano e la Terra solida. I microsismi oceanici sec-
ondari, dovuti all’interazione non lineare tra onde oceaniche e aventi periodo inferiore
a 12 s, determinano il segnale più forte.
L’obiettivo di questa tesi è lo studio del campo d’onda, della generazione e della
propagazione dei microsismi oceanici secondari. Le sorgenti del rumore sismico sono
rappresentate da forze verticali disposte lungo la superficie dell’oceano, la cui ampiezza
è cacolata a partire da un modello d’oceano realistico. L’ampiezza delle tre compo-
nenti dello spettro del rumore sismico è modellizzata come sovrapposizione di modi
normali. Il modo fondamentale delle onde di Rayleigh risulta essere il segnale domi-
nante. La differenza tra lo spettro reale e quello sintetico delle componenti orizzontali
permette di stimare l’entità dell’energia associata alle onde di Love, la cui generazione
richiede un diverso meccanismo. L’effetto di sito che agisce sulle onde di Rayleigh,
dovuto all’oceano, è stimato usando i modi normali, al fine di studiare l’effetto della
profondità oceanica, della stratificazione crostate e dello strato sedimentario sub-
oceanico sull’ampiezza del rumore sismico. L’effetto di sito dovuto all’oceano che
agisce sulle onde di volume associate al rumore sismico è calcolato definendo il campo
ondulatorio come sovrapposizione di onde piane. L’effetto di sito risulta variare forte-
mente con il periodo della perturbazione e con la profondità dell’oceano, sebbene in
modo diverso per le onde di volume e per le onde di Rayleigh, amplificando regioni
diverse a seconda del periodo. Infine, si analizza l’effetto della batimetria sul rumore
sismico al variare del periodo usando il metodo degli elementi spettrali. Una sor-
gente secondaria virtuale viene prodotta dalla batimetria e il campo d’onda sismico
è modificato dalla presenza di questa sorgente.

7
8
Contents

Introduction 23

1 Secondary microseismic noise 29

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.2 Seismic noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.2.1 Microseism temporal and spatial variability . . . . . . . . . . 31
1.2.2 Microseisms waveform composition . . . . . . . . . . . . . . . 37
1.3 Secondary microseismic noise generation theory . . . . . . . . . . . . 38
1.3.1 A little historical overview . . . . . . . . . . . . . . . . . . . . 38
1.3.2 The cause of seismic noise generation: the second-order pressure
fluctuation at the seafloor . . . . . . . . . . . . . . . . . . . . 39
1.3.3 A statistical description of the origin of the microseisms . . . . 45
1.4 Locating the ocean wave-wave interaction . . . . . . . . . . . . . . . . 48
1.4.1 Ocean wave-wave interaction classes . . . . . . . . . . . . . . . 48
1.4.2 The ocean wave model WAVEWATCH III(R) . . . . . . . . . . 51
1.4.3 Ocean wave height and ocean wave-wave interaction . . . . . . 52
1.5 Seismic noise sources and the ocean site effect . . . . . . . . . . . . . 57
1.5.1 The ocean site effect on Rayleigh waves . . . . . . . . . . . . . 58
1.5.2 The ocean site effect on body waves . . . . . . . . . . . . . . . 61
1.5.3 What about Love waves? . . . . . . . . . . . . . . . . . . . . . 63
1.6 Propagation of seismic noise inside the Earth . . . . . . . . . . . . . . 63

2 Modeling noise Rayleigh waves 65

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.2 Normal modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.2.1 Normal modes and seismic waves . . . . . . . . . . . . . . . . 66
2.2.2 Normal modes and gravity . . . . . . . . . . . . . . . . . . . . 68

9
 2.3 Synthetic seismograms . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.3.1 Generic analytical formulation . . . . . . . . . . . . . . . . . . 69
2.3.2 Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.3 Spherical and canonical coordinates . . . . . . . . . . . . . . . 72
2.4 Synthetic seismogram using a single force . . . . . . . . . . . . . . . . 74
2.4.1 Vertical single force . . . . . . . . . . . . . . . . . . . . . . . . 74
2.5 Earth models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.5.1 Two-layer model . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.5.2 PREM model . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.5.3 Sediments at the ocean seafloor . . . . . . . . . . . . . . . . . 85
2.6 Site effect due to the ocean depth . . . . . . . . . . . . . . . . . . . . 90
2.7 Effect of the ocean depth on Rayleigh wave propagation . . . . . . . . 94
2.8 Normal Mode Summation code . . . . . . . . . . . . . . . . . . . . . 95
2.8.1 Attenuation model . . . . . . . . . . . . . . . . . . . . . . . . 98
2.9 Modeling secondary microseismic noise by normal mode summation . 99
Appendix A: Normal modes associated with Scholte waves . . . . . . . . . 114
Appendix B: Tsunami modes . . . . . . . . . . . . . . . . . . . . . . . . . 115

3 Modeling noise body waves 117

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.2 Transmission and reflection coefficients at a solid-liquid interface . . . 118
3.3 Estimation of the ocean site effect on body waves . . . . . . . . . . . 125
3.3.1 The ocean site effect for varying epicentral distance . . . . . . 128
3.3.2 The ocean site effect for varying frequency and ocean depth . 131
3.4 Comparison between the site-effects on body and Rayleigh waves . . . 134
3.5 Ray-mode duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.6 Modeling the ocean site effect on seismic noise body waves . . . . . . 145

4 The seismic wavefield under the ocean 159

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
4.2 Short description of the spectral-element method . . . . . . . . . . . 160
4.3 Model and source set-up . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.4 The wavefield composition for varying bathymetries . . . . . . . . . . 163
4.4.1 The flat seafloor . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.4.2 A realistic continental slope . . . . . . . . . . . . . . . . . . . 170

10
 4.5 The seismic wavefield for varying source frequencies . . . . . . . . . . 176
4.6 Further analysis and next steps . . . . . . . . . . . . . . . . . . . . . 180

Conclusions and future prospectives 181

Most significant achievements . . . . . . . . . . . . . . . . . . . . . . . . . 181
Ideas for future investigations . . . . . . . . . . . . . . . . . . . . . . . . . 184

Annex I - Finite difference P-wave tomography of the Italian region:

an additional work 187

11
12
List of Figures

1-1 Power spectral density – dB with respect to the acceleration – from

1989 to 2010 for seven stations of the GEOSCOPE global network.
White areas correspond to periods when data were triggered on earth-
quakes or to missing data. The secondary microseismic noise is the
strongest signal, at periods smaller than 10s. After Stutzmann et al.
(2012). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1-2 a) Seismogram recorded at the Geoscope station SSB (Saint Sauveur en
Rue, France) during January 2008. b) Power spectral density measured
in dB with respect to the acceleration. c) Average power spectra over
each day of January 2008 (black lines) and over January 2008 (pink
line). The average power spectra has been computed cutting away the
earthquakes from the seismogram. . . . . . . . . . . . . . . . . . . . 33
1-3 a) Seismogram recorded at the Geoscope station SSB (Saint Sauveur en
Rue, France) during August 2008. b) Power spectral density measured
in dB with respect to the acceleration. c) Average power spectra over
each day of August 2008 (black lines) and over August 2008 (pink
line). The average power spectra has been computed cutting away the
earthquakes from the seismogram. . . . . . . . . . . . . . . . . . . . 34
1-4 Background noise spectra in dB for stations at varying distances from
the coast: a) TAM (Tamanrasset, Algeria); b) SSB (Saint Sauveur en
Rue, France); c) CAN (Canberra, Australia), d) KIP (Kipapa, Hawaii,
USA).The spectra are averaged over 1 day, 2008, January 2nd. Vertical
(blue) and horizontal (red and green) components are shown for each
one of the stations. The dashed lines represent the high and low-noise
models developed by Peterson (1993).
From http://geoscope.ipgp.fr/index.php/en/stations/station-list. . . . 36

13
1-5 Forces acting on a fluid between two fixed points on the horizontal
direction, separated by a wavelength λ. . . . . . . . . . . . . . . . . . 40

1-6 Fluid particle orbits caused by an oscillatory surface wave traveling

to the right for three liquid depths. (a) When the liquid is deep, the
particle orbits are circular and decrease in size with increasing depth.
(b) When the water is shallow, the orbits are thin ellipses that become
thinner with increasing depth. Adapted from Kundu et al. (2012). . . 43

1-7 Schematic illustration of the three classes of ocean wave-wave interac-

tion. The thick arrows indicate the local wind direction and the small
arrows indicate the direction of wave trains. At point A, ocean waves of
the same storm S1 interact with each other (class I). At point B, ocean
waves travelling towards the coast interact with ocean waves reflected
from the coast (class II). At point C, remote ocean waves from the
storm S1 interact with ocean waves generated by the storm S2 (class
III). After Ardhuin et al. (2012). . . . . . . . . . . . . . . . . . . . . . 49

1-8 Wave-wave interaction caused by a moving cyclone. A and B: moving

cyclone locations at two consecutive times. C: location of the ocean
wave-wave interaction when the cyclone is located at B. After Longuet-
Higgins (1952). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

1-9 a) Ocean wave height and b)-d) wave-wave interaction as given by

equation (1.20) on 2008 January 1, 12:00-15:00 UTC. The ocean wave-
wave interaction has been computed by using the numerical ocean wave
model WAVEWATCH III(R) at the following periods: b) T = 3.9 s, c)
T = 5.2 s and d) T = 6.9 s. . . . . . . . . . . . . . . . . . . . . . . . 54

1-10 a) Ocean wave height and b)-d) wave-wave interaction as given by

equation (1.20) on 2008 August 1, 12:00-15:00 UTC. The ocean wave-
wave interaction has been computed by using the numerical ocean wave
model WAVEWATCH III(R) at the following periods: b) T = 3.9 s, c)
T = 5.2 s and d) T = 6.9 s. . . . . . . . . . . . . . . . . . . . . . . . 55

14
1-11 Maps of the ocean wave height (leftmost column) and power spectral
density of the pressure field as given by equation (1.20) at T = 3.9 s
(central column) and T = 5.2 s (rightmost column). The computa-
tion has been performed for three hours (12:00-15:00 UTC) from 2006
September 1st (first row) to 2006 September 5th (last row). The black
line shows the typhoon track. . . . . . . . . . . . . . . . . . . . . . . 56

1-12 The ocean site effect cn as a function of the product between the ocean
depth h and the seismic frequency σ = 2πfs , normalised by the S-wave
velocity in the elastic medium βc . Adapted from Longuet-Higgins (1950). 59

1-13 Maps of the ocean site effect c1 – as it has been tabulated in Longuet-
Higgins (1950), Table 1 – for fixing periods from 3 s to 10 s. . . . . . 60

1-14 World ocean bathymetry. . . . . . . . . . . . . . . . . . . . . . . . . . 61

1-15 Maps of the ocean wave-wave interaction (column on the left) as given
by equation (1.20) and noise sources (column of the right) as given
by equation (1.21) on 2008, January 1st and 2008, August 1st (12:00-
15:00 UTC). The computation has been carried out for three fixed
periods: T = 3.9 s, T = 5.2 s and T = 6.7 s. The maxima of the ocean
wave-wave interaction do not correspond to the maxima of noise sources. 62

2-1 Cartoon showing the surface vibrations associated with the first three
spheroidal modes – 0 S0 , 0 S1 and 0 S2 – (upper row) and toroidal modes
– 0 T1 , 0 T2 and 0 T3 – (lower row) on a sphere. See the text for the details
concerning the corresponding normal modes of the Earth. . . . . . . . 68

2-2 Dispersion diagram showing the spheroidal fundamental mode n = 0

and the first spheroidal overtone of the PREM model in presence of
gravity (red dots) and in absence of gravity (blue dots). The spherical
harmonic degree l is plotted as a function of the eigenfrequency ωk .
The dispersion diagrams have been computed for frequencies a) from
0.0001 Hz to 0.002 Hz (T=500 − 1000 s) b) from 0.1 Hz to 0.2 Hz
(T=5 − 10 s). Only long period normal modes are affected by gravity. 69

15
2-3 Three-component synthetic seismograms computed by normal mode
summation with a vertical point source. The source amplitude is
1 × 1017 N. The epicentral distance is 35◦ . All the three-component
seismograms have been bandpass filtered with corner frequencies of
0.004 Hz and 0.01 Hz. All synthetics here are calculated using the
model PREM (Dziewonski & Anderson, 1981). . . . . . . . . . . . . . 75
2-4 a) Density ρ, b) compressional-wave speed α and c) shear-wave speed β
as a function of depth in the two-layer model. The model is capped by
a 3 km thick uniform oceanic layer. For depths deeper than the core-
mantle boundary (CMB, 2891 km depth), the model consists of the
elastic parameters of the Preliminary Reference Earth Model (PREM).
A zoom in the ocean layer is shown in the upper part of the figures. 77
2-5 Dispersion diagram showing the fundamental mode and the first four
overtones of the two-layer model varying the ocean depth. a) 1 km of
ocean thickness; b) 3 km of ocean thickness; c) 5 km of ocean thickness.
It has been considered only for periods from T = 4 s to T = 10 s. . . 79
2-6 Displacement eigenfunctions U as a function of depth for the funda-
mental mode at a fixed period of 6 s. The computation has been
carried out in the two-layer model in which we vary the ocean depth.
The colour scale is referred to the ocean thickness, which varies from
1 km to 10 km, with discrete steps of 1 km. . . . . . . . . . . . . . . 80
2-7 Group velocity (left column) and phase velocity (right column) as a
function of the period T considering a) 1 km of ocean depth; b) 3 km
of ocean depth; c) 5 km of ocean depth. The Earth model is the two-
layer model. The colour scale is related to the radial order n of the
first five normal modes. . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2-8 a) Density ρ, b) compressional-wave speed α and c) shear-wave speed
β as a function of depth in the PREM model. The model is capped
by a 3 km thick uniform oceanic layer. A zoom of the ocean layer is
shown in the upper part of the figures. . . . . . . . . . . . . . . . . . 82
2-9 Dispersion diagram showing the fundamental mode and the first four
overtones of the PREM model varying the ocean depth. a) 1 km of
ocean thickness; b) 3 km of ocean thickness; c) 5 km of ocean thickness.
It has been considered only for periods from T = 4 s to T = 10 s. . . 84

16
2-10 Displacement eigenfunctions U as a function of depth for the funda-
mental mode (a), the first (b) and the second (c) overtone, at a fixed
period of 6 s (left column) and 10 s (right column). The computation
has been carried out in the PREM model in which we vary the ocean
depth. The colour scale is referred to the ocean thickness, which varies
from 1 km to 10 km, with discrete steps of 1 km. . . . . . . . . . . . 86
2-11 Group velocity (left column) and phase velocity (right column) com-
puted in the PREM model considering a) 1 km of ocean depth; b) 3
km of ocean depth; c) 5 km of ocean depth. The colour scale is related
to the radial order n of the first five normal modes. . . . . . . . . . . 87
2-12 Map of sediment thickness for the world’s oceans compiled by the
National Geophysical Data Center (NGDC), Marine Geology & Geo-
physics Division (http://www.ngdc.noaa.gov/mgg/sedthick/sedthick.html )
(Divins, 2003). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2-13 Displacement eigenfunctions U as a function of depth for the funda-
mental mode of Rayleigh waves. The computation has been carried
out in the PREM model in which has been added a layer of sediments
between the ocean and the crust (red eigenfunctions). The period is
fixed at T = 10 s and the ocean depth at a) 1 km depth and b) 3 km
depth. In grey, it is shown the eigenfunction computed in the PREM
model with the same ocean thickness without the sediment layer. . . 89
2-14 a) Dispersion diagram, b) group velocity and c) phase velocity com-
puted considering the PREM model with the sediment layer (red curves)
and without the sediment layer (grey curves) for the fundamental mode
and the first overtone. . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2-15 Site effect coefficient – as expressed in equation 2.29 – as a function
of the dimensionless quantity ωh/β, where ω is the eigenfrequency, h
is the ocean depth and β is the S-wave velocity in the crust. The
computation of normal modes has been carried out in a) the two-layer
model, b) the PREM model and c) the PREM model in which we have
added a layer of sediments between ocean and crust. The source and
the receiver positions are at the surface of the ocean and at the seafloor,
respectively. Wave periods from 4 s up to 10 s have been considered.
The colour scale is related to the ocean depth. . . . . . . . . . . . . . 92

17
2-16 Site effect on the fundamental mode of Rayleigh waves (equation 2.29)
for periods from 3 s to 10 s. The computation of the eigenfunctions
has been carried out in the PREM model. . . . . . . . . . . . . . . . 93
2-17 Group velocity as a function of the dimensionless quantity ωh/β, where
ω is the eigenfrequency, h is the ocean depth and β is the S-wave
velocity in the crust. The computation has been carried out in a) the
two-layer model (see section 2.5.1), b) the PREM model (see section
2.5.2) and c) the PREM model in which has been added a sediment
layer beneath the seafloor (see section 2.5.3) for periods between T = 4
s and T = 10 s. The colour scale is referred to the ocean thickness. . 96
2-18 a) Phase velocity in the PREM model as a function of the period for
the first six modes for periods shorter than T = 3 s. The horizontal
grey line is referred to the velocity of the compressional wave in the
fluid layer (αw ' 1.5 km/s). b) Radial displacement eigenfunction 0 Ul
as a function of depth for a couple of Stoneley modes, computed at
T = 2 s (blue) and T = 2.5 s (red). A zoom concerning the first 10 km
is also shown for each eigenfunction. . . . . . . . . . . . . . . . . . . 114
2-19 a) Dispersion diagram comparing the tsunami modes and the first ten
spheroidal-mode branches for a two-layer model described in section
2.5.1. The ocean thickness is fixed to 3 km. The tsunami branch ap-
pears to be non-dispersive. b) Tsunami-mode displacement associated
with the radial eigenfunctions U and c) with the eigenfunction V versus
depth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

3-1 Cartoon illustrating the ray paths at the liquid-solid interface. The
down-going incident P-wave (black) is partially reflected as upgoing
P-wave (red), partially transmitted as downgoing P-wave (green) and
S-wave (blue). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3-2 Plane wave reflection and transmission coefficients in terms of potential
at the liquid-solid interface as a function of the incident P-wave take-
off angle θPw ≤ θPS∗w (θPS∗w = 27.95◦ ), as given by the equations (3.9),
(3.10) and (3.11), respectively. The P-wave reflection and transmis-
sion coefficients are shown in red and blue, respectively. The S-wave
transmission coefficient is in green. The vertical black line denotes the
critical P-wave take-off angle in equation (3.14). . . . . . . . . . . . . 123

18
3-3 Plane wave reflection and transmission coefficients in terms of potential
at the liquid-solid interface as a function of the incident P-wave take-
off angle θPw ≥ θPS∗w (θPS∗w = 27.95◦ ), as given by the equations (3.9),
(3.10) and (3.11), respectively. The P-wave reflection and transmis-
sion coefficients are shown in red and blue, respectively. The S-wave
transmission coefficient is in green. . . . . . . . . . . . . . . . . . . . 124

3-4 Cartoon illustrating the liquid-solid interface between a solid half space
and a liquid layer. Seismic P-waves are generated at the source located
at the surface of the liquid layer. Reflection and transmission occur at
the discontinuity. The upper surface of the liquid layer is considered
as free surface, than total reflection occurs. After Gualtieri et al. (2014).126

3-5 Ocean site effect on a) P-waves and b) S-waves as a function of the

epicentral distance (or the take-off angle) and the product between
frequency and ocean depth, normalised by the P-wave velocity in the
ocean layer. The colour scale is referred to the amplitude of these
coefficients as given by equations (3.16) and (3.17). The panel c) dis-
plays in colours the ratio between the CP and CS . Only epicentral
distances larger than 30◦ have been considered to avoid the P- and
S-wave triplication regime. . . . . . . . . . . . . . . . . . . . . . . . . 129

3-6 Ocean site effect on a) PKP-waves and b) SKS-waves as a function of

the epicentral distance (or the take-off angle) and the product between
frequency and ocean depth, normalised by the P-wave velocity in the
ocean layer. The colour scale is referred to the amplitude of these
coefficients as given by equations (3.16) and (3.17). . . . . . . . . . . 130

3-7 P- and S-wave site effect obtained by summing up all the take-off angles
smaller than the P-wave critical one, as given by equations 3.19 and
3.20. The colour scale is referred to the ocean depth. After Gualtieri
et al. (2014). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

3-8 Maps of the site effect on P-waves as given by the equation (3.19) for
periods from 3 to 10 s. . . . . . . . . . . . . . . . . . . . . . . . . . . 132

3-9 Maps of the site effect on S-waves as given by the equation (3.20) for
periods from 3 to 10 s. . . . . . . . . . . . . . . . . . . . . . . . . . . 133

19
3-10 Ocean site effect on Rayleigh waves (left column) and on P-waves (right
column) for periods from T = 3 s to T = 5 s. Both of them have been
normalised to end up with comparable amplitudes. . . . . . . . . . . 135

3-11 Maps of the site effect on P-waves as computed by equation 3.19 and
considering only the contribution due to a) the first peak (f h/αw <
0.5); b) the second peak (0.5 ≤ f h/αw < 1) and c) the third peak
(1.5 ≤ f h/αw < 1) as in Figure 3-7. Each row is referred to a given
period, from 3 s (first row) to 10 s (last row). . . . . . . . . . . . . . 138

3-12 Site effect CS (colour scale) as computed by equation (3.17) as a func-

tion of the take-off angle θPw and the angular frequency ω, considering
an ocean layer of a) 1 km, b) 3 km and c) 5 km depth. The white lines
denote the critical take-off angle of P-waves θPPw∗ = 15.71◦ (equation
3.13) and S-waves θPS∗w = 27.95◦ (equation 3.14). . . . . . . . . . . . . 140

3-13 Site effect CS (colour scale) as computed by equation (3.17) as a func-

tion of the angular order l and the angular frequency ω, considering an
ocean layer of a) 1 km, b) 3 km and c) 5 km depth. The angular order
l has been computed by using equation (3.21). The white lines denote
the critical take-off angle of P-waves θPPw∗ = 15.71◦ (equation 3.13) and
S-waves θPS∗w = 27.95◦ (equation 3.14). The small black boxes display
the dispersion diagrams of the first five normal modes as presented in
Chapter 2, Figure 2-5. . . . . . . . . . . . . . . . . . . . . . . . . . . 141

3-14 Phase velocity as a function f h/αw , where f is the frequency, h is the

ocean depth and αw is the P-wave velocity in the ocean layer. a) Ocean
site effect of P-waves (colour scale); b) Ocean site effect on S-waves
(colour scale); c) normal mode computation, showing the ocean depth
with different colours. The white lines (upper and central figures) and
the black line (bottom figure) at c = 2.8 km/s is referred to the S-wave
velocity in the crust. . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

4-1 Cartoon illustrating the flat seafloor configuration. The seismic source
is located at the surface of the ocean and x = 75 km (red star). The
receivers are located below the liquid-solid interface (blue triangles). 164

20
4-2 Snapshot of the kinetic energy field for the flat-seafloor configuration
(red line) at t = 7 s. The red star at the surface of the ocean and
x = 75 km is the source and the blue triangles under the seafloor are
the seismic stations. The blue letters on the top panel refer to to the
seismic phases illustrated on the left. The cartoon on the left side has
been adapted from Komatitsch et al., 2000. . . . . . . . . . . . . . . . 165

4-3 Snapshots of the kinetic energy field for the flat-seafloor configuration
(red line) at different times: a) t = 10 s, b) t = 20 s, c) t = 30 s and
d) t = 40 s. The red star at the surface of the ocean and x = 75 km
is the source and the blue triangles under the seafloor are the seismic
stations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

4-4 a) Horizontal and b) vertical components of the seismograms recorded

at a line of receivers located 100 m below the seafloor. The inter-station
distance is 1 km. All seismograms have been normalised by a factor of
5 × 10−8 m/s. c) Theoretical arrival-times at the receivers. We recall
that the source is located at the surface of the ocean and x = 75 km. 168

4-5 Cartoon illustrating the realistic ocean-continental slope configuration.

The seismic source is located at the surface of the ocean and x = 75
km (red star). The receivers are located below the liquid-solid interface
(blue triangles). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

4-6 Snapshots of the kinetic energy for the ocean-continental slope con-
figuration (red line) at different times: a) t = 10 s, b) t = 20 s, c)
t = 30 s and d) t = 40 s. The red star at the surface of the ocean and
x = 75 km is the source and the blue triangles under the seafloor are
the seismic stations. A secondary virtual source is generated around
x = 50 km by the continental slope. . . . . . . . . . . . . . . . . . . . 172

4-7 a) Horizontal and b) vertical components of the seismograms recorded

at a line of receivers located 100 m below the seafloor. All seismograms
have been normalised by a factor of 5 × 10−8 m/s. The inter-station
distance is 1 km. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

21
4-8 Horizontal (column on the left) and vertical (column on the right)
components of the seismograms recorded at five receivers located below
the seafloor and at a) x = 5 km, b) x = 30 km, c) x = 45 km, d) x = 60
km and e) x = 95 km. Blue seismograms are related to the flat liquid-
solid interface and the red seismograms to the continental slope. . . 175
4-9 Particle motion diagrams computed from the seismograms recorded
at the station number 80 for varying periods: a) f0 = 0.12 Hz (i.e.
fmax = 0.33 Hz), b) f0 = 0.073 Hz (i.e. fmax = 0.2 Hz), c) f0 = 0.052
Hz (i.e. fmax = 0.14 Hz) and d) f0 = 0.036 Hz (i.e. fmax = 0.1 Hz).
The colour scale shows successive times. The typical particle motion
of Rayleigh waves (elliptical and retrograd) clearly appears. . . . . . . 177
4-10 Vertical component of the seismograms recorded at five receivers lo-
cated below the seafloor. Their horizontal coordinates are: a) x = 5
km, b) x = 30 km, c) x = 45 km, d) x = 60 km and e) x = 95 km.
Blue seismograms are related to the flat liquid-solid interface and the
red seismograms to the continental slope. The cut-off frequency of the
source is f0 = 0.12 Hz (i.e. fmax = 0.33 Hz) for the left column seis-
mograms and f0 = 0.073 Hz (i.e. fmax = 0.2 Hz) for the right column
seismograms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
4-11 Vertical component of the seismograms recorded at five receivers lo-
cated below the seafloor. Their horizontal coordinates are: a) x = 5
km, b) x = 30 km, c) x = 45 km, d) x = 60 km and e) x = 95 km.
Blue seismograms are related to the flat liquid-solid interface and the
red seismograms to the continental slope. The cut-off frequency of the
source is f0 = 0.052 Hz (i.e. fmax = 0.14 Hz) for the left column seis-
mograms and f0 = 0.036 Hz (i.e. fmax = 0.1 Hz) for the right column
seismograms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

22
Introduction

“We define the signal as the desired part of the data and the noise as the unwanted
part. Our definition of signal and noise is subjective in the sense that a given part of
the data is ‘signal’ for those who know how to analyse and interpret the data, but it
is ‘noise’ for those who do not.”

Aki & Richards (2002)

Seismic noise on seismograms has been viewed for many years as a chaotic signal
contaminating records of earthquakes. The less noisy the seismogram was, the better
the record was considered. Then many techniques have been developed in order to
improve the signal-to-noise ratio (e.g. Park et al., 1987), throw away the noisy part
and better isolate the earthquake signals on seismograms.
Seismic noise became a subject of interest once it had been established that it was
not caused by instrumentation resonance (e.g. Bertelli, 1872) and that it was carrier of
important meteorological and oceanic signals. Seismic noise has been largely ignored
for many years because of difficulties on the understanding of its physical origin and
on extracting a deterministic and coherent seismic wavefield. Seismology based on
the seismic noise wavefield is thus a young science and a thriving research field, with
many aspects which are still unknown.
Nowadays, microseisms have been recognised as the continuous incessant quasi-
random oscillation of the Earth caused mostly by ocean wave-wave interaction during
storms over the ocean. Since they link the atmosphere, the ocean and the solid Earth,
they have been used in complementary areas of work, like indicators of climate change
(e.g. Bromirski et al., 1999; Grevemeyer et al., 2000; Grob et al., 2011; Stutzmann
et al., 2009), coastal wave energy variability (e.g. Aster et al., 2010), climate decadal
variability (e.g. Aster et al., 2008) and significant wave height (e.g. Bromirski et al.,
1999; Ardhuin et al., 2012).

23
Introduction

Moreover, it has been discovered that the Green’s function between two receivers
can be approximated by cross-correlating ambient noise that the receivers record
continuously (e.g. Shapiro & Campillo, 2004; Campillo, 2006; Sabra et al., 2005).
This evidence has allowed to investigate the internal structure of our planet, bringing
important advantages with respect to the earthquake-based seismic imaging, like the
independence from sparsely distributed earthquakes, the ability to use short-period
waves and a higher spatial resolution in regions with little or no seismicity (e.g. near-
surface structure: Mordret et al., 2013; crustal structure: Shapiro et al., 2005; mantle
imaging: Nishida et al., 2009; Schimmel et al., 2010; crustal and uppermost mantle
imaging: Tian et al., 2013). Variations of the cross-correlograms over time are used
for monitoring volcanoes (e.g. Snieder, 2004; Brenguier et al., 2008b; Mordret et al.,
2010; Sens Schönfelder & Wegler, 2011; Clarke et al., 2013; Obermann et al., 2013;
Rivet et al., 2014) and seismic faults (e.g. Wegler & Sens-Schönfelder, 2007; Brenguier
et al., 2008a, Froment et al., 2013).
In spite of the large use of seismic noise to infer properties of the Earth, storm
characteristics and climate variability, its generation mechanisms and the causes of
the various physical processes which control microseisms have not been fully explored.
Very little is known about seismic noise sources in the ocean and the efficiency to
transfer energy to the solid Earth.
The aim of this study is then to improve our knowledge of seismic noise sources
and noise waveform composition. In order to do that, we study the main physical
processes which shape the source amplitude and location and we model the seismic
noise amplitude. We deal with the secondary microseisms, which are the strongest
signals in the noise spectra, worldwide continuously recorded by seismic stations at
periods shorter than about 12 s.
The generation of the secondary microseismic noise is related to the storm capa-
bility to give rise to ocean gravity wave-wave interactions at the ocean surface, able
to bring energy to the seafloor. The fundamental theoretical milestones on the under-
standing of these phenomena were achieved in the middle of the 20th century and the
most significant contributions are due to Miche (1944), Longuet-Higgins (1952) and
Hasselmann (1963). Their works allow to identify the sea state necessary conditions
to generate seismic noise and to explain seismic noise as the result of many sources
acting at the same time in different oceanic regions.
Noise body wave sources (e.g. Toksöz & Lacoss, 1968; Haubrich & McCamy,

24
Introduction

1969; Koper & de Foy, 2008; Zhang et al., 2009; Koper et al., 2009; Gerstoft et al.,
2008; Koper et al., 2010; Landès et al., 2010; Obrebski et al., 2013) and surface wave
sources (e.g. Ramirez, 1940a; Ramirez, 1940b; Haubrich & McCamy, 1969; Capon,
1973; Webb & Constable, 1986; Cessaro & Chan, 1989; Cessaro, 1994; Stehly et al.,
2006; Obrebski et al., 2012) have been located by using the frequency-wavenumber
analysis (e.g. beamforming analysis). These techniques are heavily influenced by the
seismic array location and they do not allow one to separate the contribution due to
the sources of seismic noise from propagation effects.
In order to study the location of seismic noise sources, we adopt a very new
approach based on the numerical modeling of the sea state (e.g. Kedar et al., 2008;
Ardhuin et al., 2011), which allows to compute noise sources from the ocean gravity
wave-wave interaction.
The efficiency of noise sources to transfer energy from the ocean surface to the
solid ground depends on many aspects, very little explored so far (e.g. Longuet-
Higgins, 1950; Ardhuin & Herbers, 2013). Firstly, since the noise sources are at the
surface of the ocean, the seismic wavefield below the seafloor is sensitive to the ocean
layer, which acts as waveguide on the compressive or acoustic waves propagating in
it. This effect, that we call ocean site effect, acts on the seismic wavefield generated
below the source, in the solid Earth, and it modulates the shape and position of noise
sources. Moreover, seafloor sediments and the shallowest part of the crust, beneath
the source region, have an important impact on the seismic noise wavefield. In this
work, we demonstrate that, these effects act differently on noise body and Rayleigh
waves.
Modeling the amplitude of the seismic noise has become possible very recently
thanks to numerical ocean models and it allows to investigate the seismic noise wave-
field composition and the aspects which control the propagation of seismic waves
inside the Earth. Since the noise sources are at the surface of the ocean, the ocean
site effect also affects the velocity and attenuation of the noise seismic wavefield below
the seafloor.
This thesis work is structured as follows:

• In Chapter 1, we introduce the concepts of seismic noise and microseisms and

its most significant time-, space- and frequency-dependent features. We review
and explain how the ocean wave-wave interaction is able to generate microseisms
and how this kind of sources can be schematised as a pressure applied at the

25
Introduction

surface of the ocean (Longuet-Higgins, 1950; Hasselmann; 1963). A statistical

description of the origin of microseisms, which is needed in order to use the
numerical ocean wave model for describing the ocean sea state, is then presented.
The last part of this chapter deals with the importance of taking into account
the ocean site effect on the seismic wavefield in the solid Earth, to correctly
locate noise sources.

• Chapter 2 is dedicated to the study of noise Rayleigh waves through normal

modes. Since spheroidal normal modes are equivalent to propagating funda-
mental and higher-order Rayleigh waves, the normal mode approach is a natu-
ral way to study surface waves. Then, we retrieve seismic noise Rayleigh wave
features and we model the amplitude of the secondary microseismic noise by
using normal mode summation. The normal mode summation theory using a
single force is reviewed in order to compute seismic noise synthetic seismograms
and spectra.

The ocean site effect on seismic noise Rayleigh waves is computed in various
Earth models. The effect produced on noise Rayleigh wave sources by ocean
depth, crust layering, seafloor sediments and receiver position are detailed for
varying frequencies. We also investigate the main features of noise Rayleigh
wave propagation, that is group velocity, phase velocity and attenuation.

The most significant findings towards the comprehension of noise Rayleigh

waves, as well as the modeling of the vertical and horizontal components of
the noise spectra, are summarised in Gualtieri et al. (2013), attached at the end
of this chapter.

• Chapter 3 is dedicated to noise body waves. Because we are not interested on

recording seismic noise in the source area, we adopt the far field approximation
and we describe the noise seismic wavefield by using the plane wave decom-
position. The theory for computing the ocean site effect on noise body waves
below the seafloor is detailed both for P and SV waves. This effect, which is the
strongest control on noise sources, is compared to the one which affects Rayleigh
waves. The most amplified Rayleigh and body wave sources are located in dif-
ferent geographical regions. A ray-mode duality, which allows to retrieve the
imprints of Rayleigh waves from the constructive interferences of body waves,
is presented in the last part of this chapter.

26
Introduction

The body wave generation theory, as well as its validation by using beamforming
analysis, is presented in Gualtieri et al. (2014), attached at the end of this
chapter.

• Preliminary results concerning the effect of the bathymetry on the seismic wave-
field are presented in Chapter 4. In order to model the seismic wavefield recorded
below the seafloor, we use the spectral-element method, since it enables to follow
the irregular discontinuities of the Earth, like the seafloor, by adapting the mesh
configuration, and it handles coupled liquid-solid regions. First results about
the seismic wavefield recorded beneath the ocean-crust interface are detailed for
varying periods and seafloor topographies.

• This thesis ends with conclusive remarks and new research lines for future in-
vestigations.

27
Chapter 1

Secondary microseismic noise:

phenomenology and generation
mechanism

1.1 Introduction

In this chapter, we introduce the concepts of seismic noise and microseisms. We

describe both the phenomenology and the generation mechanisms, focusing on periods
from 3 s to 10 s, that is on the so-called secondary microseismic noise.
After a brief introduction about seismic noise and the analysis of the most signifi-
cant features of the noise spectrum, we deal with the spatial and temporal-dependent
characteristics of the secondary microseismic noise, with the aim to show where
seismic noise is generated. Then, we present the composition of the seismic noise
wavefield, by summarising the most significant literature contributions on the under-
standing of the seismic noise waveform (section 1.2). Section 1.3 is dedicated to a
review of the secondary microseisms noise generation theory, where we explain how
seismic noise can be generated in the ocean. A single progressive wave is inefficient
in generating seismic noise because it does not produce any pressure variations at the
ocean seafloor. The interaction of ocean waves is instead the proxy for generating
a non-null pressure variation at the seafloor and then the seismic wavefield in the
solid Earth (section 1.3.2). Because an instantaneous picture of the sea surface does
not indicate in which direction the waves are going, to specify the full sea state, it
is necessary to adopt a statistical description and consider the probability of finding

29
Chapter 1. Secondary microseismic noise

a particular sea state (section 1.3.3). In section 1.4.1, we focus on the classification
of the ocean wave-wave interaction. The ocean wave model we use to compute the
ocean wave-wave interaction and few examples are shown in section 1.4.2 and 1.4.3,
respectively. The last part of this chapter is dedicated to the effect of the ocean depth
under the source region on the seismic noise wavefield, termed the ocean site effect
(section 1.5).

1.2 Seismic noise

The ground is never truly at rest. A random signal, independent from the earth-
quake activity, is recorded everywhere in the world and continuously in time. This
background noise level is mainly due to the atmospheric and ocean activity. Short
period contributions are also due to diurnal heating of the surface and human-induced
vibrations (Ebeling, 2012). Seismic noise intensity varies over time scales as short as
seconds or as long as decades, with seasonal variations up to several orders of mag-
nitude (e.g. Stutzmann et al., 2000; McNamara & Buland, 2004; Schimmel et al.,
2011). The microseismic noise level is temporarily and spatially variable and it is
not uniform at all frequencies. The weakest and strongest ambient noise signals are
known as hum and microseisms, respectively.
For periods smaller then T ' 20 s, the ground acceleration noise spectrum is
typically dominated by two main peaks at about 7 s and 14 s of period, which are
termed the microseismic peaks (e.g. Webb, 1998; Stutzmann et al., 2000; Berger,
2004). The smaller one, which falls in the period band between 10 s and 20 s, is
called primary microseism, while the much larger one, between about 3 s and 10 s,
is called secondary microseism. These maxima in the seismic noise spectra are also
known as the single- and the double-frequency peaks, respectively. The reason of this
terminology – that we will detail in the following – is linked to the frequency of their
sources, that is the gravity ocean waves (e.g. Miche, 1944; Longuet-Higgins, 1950;
Hasselmann, 1963).
At periods from T ' 20 s up to hundreds of seconds, the long period noise or
hum consists of continuously excited free oscillations of the Earth. Their source
mechanisms are still under debate, but they are related to atmospheric perturbations
and ocean activity, involving ocean infragravity waves (e.g. Nawa et al., 1998; Suda
et al., 1998; Tanimoto et al., 1998; Rhie & Romanowicz, 2004; Rhie & Romanowicz,

30
Chapter 1. Secondary microseismic noise

2006; Webb, 2007; Duncan Carr Agnew, 2007; Kurrle & Widmer-Schnidrig, 2008;
Uchiyama & McWilliams, 2008; Webb, 2008; Bromirski & Gerstoft, 2009; Traer et al.,
2012; Nishida, 2013; Traer & Gerstoft, 2014).
In this manuscript, we will only focus on the secondary microseismic noise, con-
sidering periods from 3 s to 10 s. A different generation theory is needed to explain
the other main features of noise spectrum, that is the primary microseisms and the
hum. However, in principle, the theoretical tools we will describe in Chapter 2 and
Chapter 3, may be extended to longer periods.

1.2.1 Microseism temporal and spatial variability

Microseism amplitude varies over time, from days up to decades, and with the
receiver location in relation with the local atmospheric season. We focus on the
secondary microseismis noise, at periods from 3 s up to 10 s.

Microseisms temporal variability

Figure 1-1 displays the noise power spectral density (PSD) for seven stations of
the GEOSCOPE global network from 1989 to 2010 (after Stutzmann et al., 2012).
The PSD is plotted in decibel (dB) with respect to the acceleration. The secondary
microseismic noise is the strongest signal (periods smaller than 10 s) and its seasonal
variability is stable over 20 years.
The northern hemisphere stations SSB (Saint Sauveur en Rue, France) and INU
(Inuyama, Japan) display maxima every December and January. The seismic station
TAM, which is located in the middle of the African continent (Tamanrasset, Algeria),
shows less visible seasonal variations. The station HYB (Hyderabad, India) is also
on the northern hemisphere, but it shows maxima during the southern hemisphere
winter, in July and August, during the monsoon season (Koper & de Foy, 2008).
The station PPT (Pamatai - Papeete - Tahiti island - French Polynesia, France),
which is located in the southern Pacific ocean, shows the strongest signal, constant in
time, without seasonal variations. The station CAN (Canberra, Australia) displays
maxima of noise amplitude each July-August, that is during the local winter. Finally,
the station DRV, located on a small island close to Antarctica (Dumont d’Urville -
Terre Adélie, Antarctica), shows a noise level decreasing during winter, when the sea
ice over several hundred of kilometres around the station prevents the generation of
noise sources near by the station (Stutzmann et al., 2009).

31
Chapter 1. Secondary microseismic noise

Figure 1-1: Power spectral density – dB with respect to the acceleration – from
1989 to 2010 for seven stations of the GEOSCOPE global network. White areas
correspond to periods when data were triggered on earthquakes or to missing data.
The secondary microseismic noise is the strongest signal, at periods smaller than 10s.
After Stutzmann et al. (2012).
32
Chapter 1. Secondary microseismic noise

Figure 1-2: a) Seismogram recorded at the Geoscope station SSB (Saint Sauveur en
Rue, France) during January 2008. b) Power spectral density measured in dB with
respect to the acceleration. c) Average power spectra over each day of January 2008
(black lines) and over January 2008 (pink line). The average power spectra has been
computed cutting away the earthquakes from the seismogram.

33
Chapter 1. Secondary microseismic noise

Figure 1-3: a) Seismogram recorded at the Geoscope station SSB (Saint Sauveur en
Rue, France) during August 2008. b) Power spectral density measured in dB with
respect to the acceleration. c) Average power spectra over each day of August 2008
(black lines) and over August 2008 (pink line). The average power spectra has been
computed cutting away the earthquakes from the seismogram.

34
Chapter 1. Secondary microseismic noise

Figure 1-2 and Figure 1-3 display a) the seismogram, b) the power spectral density
and c) the average power spectra recorded at the Geoscope station SSB (Saint Sauveur
en Rue, France) over one month in 2008. Figure 1-2 is referred to January 2008 and
Figure 1-3 to August 2008. Figure 1-2a and Figure 1-3a display two seismograms (in
digital units) in which seismic noise and earthquakes coexist. The background noise
level is smaller in amplitude with respect to big earthquakes (e.g. January 4th or
August 28th in Figures 1-2a and 1-3a, respectively) and it varies in time. In Figure
1-2b and Figure 1-3b, the power spectral density has been computed day by day as
a function of the period. The earthquakes overcome the noise level especially at long
period and they appear as red straight horizontal lines. Between 10 s and about 20
s, the background seismic noise level – primary microseisms – is weaker by about 20
dB with respect to the secondary microseimic noise. At periods longer then 20 s, the
background noise level – long period noise or hum – decreases by about 60 dB with
respect to the secondary microseisms. Figure 1-2c and Figure 1-3c display the average
spectra as a function of period over January 2008 and over August 2008 (pink lines),
respectively. The average spectra day by day are shown in black. They gives an idea
of the variability of the noise level during a month. These average power spectra have
been computed cutting away all the earthquakes from the seismograms.
Comparing Figure 1-2 and Figure 1-3, the link between the seismic noise level
and the seasonal atmospheric activity is evident. The seismic noise level presents a
smaller amplitude during the local summer (August, Figure 1-3) compared to the
local winter (January, Figure 1-2). The average power spectral density associated
with the secondary microseismic noise decreases by about 10 dB from Figure 1-2c to
Figure 1-3c, that is moving from the local winter to the local summer.

Microseisms spatial variability

The seismic noise level in the period band 3 − 10 s also depends on the seismic
station location. It has been observed that the strongest microseismic signal occurs
at coastal or insular seismic stations, whereas it is weak far away from the coasts (e.g.
Stutzmann et al., 2000).
Figure 1-4 displays the variation of the background noise acceleration in dB as
averaged over one day (January 2nd, 2008) for the vertical (blue) and horizontal (red
and green) components, combining the results for different recording bands (from the
GEOSCOPE network website: http://geoscope.ipgp.fr/index.php/en/stations/station-

35
Chapter 1. Secondary microseismic noise

Figure 1-4: Background noise spectra in dB for stations at varying distances

from the coast: a) TAM (Tamanrasset, Algeria); b) SSB (Saint Sauveur en Rue,
France); c) CAN (Canberra, Australia), d) KIP (Kipapa, Hawaii, USA).The spec-
tra are averaged over 1 day, 2008, January 2nd. Vertical (blue) and horizontal
(red and green) components are shown for each one of the stations. The dashed
lines represent the high and low-noise models developed by Peterson (1993). From
http://geoscope.ipgp.fr/index.php/en/stations/station-list.

list). The average spectra are shown for four GEOSCOPE stations located in different
environments. The four stations are sorted by the distance from the coast, from the
farthest to the nearest station. In particular, a) TAM station is located in the middle
of the african continent, b) SSB station is in France, c) CAN station is at about 160
km from the Australian coasts and d) KIP station is located at one of the Hawaii
islands. The dashed lines represent the high and low-noise models developed by Pe-
terson (1993), that is the maximum and minimal noise level which guarantees the
correct use and functionality of the station. This figure suggests that the island sites
(KIP station, Figure 1-4d) are much noisier than the continental sites located far away
with respect to the coast (TAM station, Figure 1-4a). Stations located at halfway
distances from the coast (SSB and CAN stations, Figures 1-4b and 1-4c, respectively)

36
Chapter 1. Secondary microseismic noise

present an intermediate noise level. These observations suggest that the noise am-
plitude is related to the seasonal atmospheric conditions and the proximity to the
ocean. The strongest noise signal is observed close to the ocean and during the local
winter. As a consequence, the seismic noise sources have to be related to the ocean
activity, as we will see in the following.

1.2.2 Microseisms waveform composition

Although seismic noise appears as a random and chaotic signal, information about
the seismic wave field can be extracted from the seismic noise waveform by analysing
array data in the frequency-wavenumber domain. This technique has permitted to
clarify the complexity of microseism waveforms.
Surface Rayleigh waves have been recognised as the main content of microseisms
(e.g. Haubrich et al., 1963; Haubrich & McCamy; Tanimoto et al., 2006; 1969;
Nishida et al., 2008). Both the fundamental and the overtones of Rayleigh waves
have been found (e.g. Yao et al., 2011), but the amplitude of the fundamental mode
of Rayleigh waves is dominant and it is sufficient to explain the main features of the
noise spectrum amplitude measured on the vertical component (for further details
about noise Rayleigh waves, see Chapter 2 and Gualtieri et al., 2013).
Both data analysis at periods longer than 5 s (e.g. Toksöz & Lacoss, 1968; Lacoss
et al., 1969; Friedrich et al., 1998; Chevrot et al., 2007a; Chevrot et al., 2007b) and
theoretical studies (e.g. Nishida et al., 2008) have also identified the presence of Love
waves in the microseisms waveform. Considering only periods associated with the
secondary microseismis, between 4 and 10 s, Nishida et al. (2008) assert that the
kinetic energy ratio between Love and Rayleigh waves takes values between 0.5 and
0.7, meaning that Love waves have a kinetic energy varying between 50 and 70 per
cent with respect to Rayleigh wave kinetic energy.
Body waves have also been recognised as being part of the seismic noise wavefield.
Body wave signal and their sources have been detected by using the beamforming
analysis (first studies: Lacoss et al., 1969 and Haubrich & McCamy, 1969) and specific
seismic phases have been found. For instance, Zhang et al. (2009), Koper et al. (2009)
detected P-wave sources. Gerstoft et al. (2008), Koper et al. (2010) and Landès et al.
(2010) extracted P, PP and PKP signals. Koper & de Foy (2008) focused instead on
body-wave phases that have interacted with the Earth core, that is PKP and PcP.
For further consideration about noise body waves, see Chapter 3.

37
Chapter 1. Secondary microseismic noise

1.3 Secondary microseismic noise generation the-

ory
In this section, we deal with a brief explanation of the mechanism which allows
to explain the secondary microseismic noise generation.
The energy of the secondary microseisms originates in the atmospheric activity over
the oceans. Storm winds generate ocean gravity waves at the ocean surface, thereby
coupling energy into the water column. This energy is transmitted to the ocean
seafloor and converted into elastic waves propagating at the seafloor and inside the
solid earth.

1.3.1 A little historical overview

First pioneering studies concerning the origin of seismic noise date back to the
end of the XIX century. Understanding of microseisms advanced in parallel with
improvements in seismological instrumentation and they were probably associated
with the observation that a link exists between microseism intensity and regional
meteorological conditions (Bertelli, 1872, 1875, 1878; De Rossi, 1875; Milne, 1883).
Many observations were made during the first half of the XX century (e.g. Wiechert,
1904; Gutenberg, 1936). For a complete review of the early observations on seismic
noise, see Ebeling (2012).
By 1950, seismic noise had been clearly associated with atmospheric disturbances
over the oceans and ocean waves generated by those storms. Microseisms had been
recognized as a continuous global phenomenon of varying amplitudes and with fre-
quencies ranging from about 4 s to 10 s (Ebeling, 2012). The relationship between
dominant microseism frequency and that of ocean gravity waves associated with them
had also shown that the microseism frequency was double than that of the ocean waves
(Deacon, 1947). Microseism intensities were known to vary seasonally and Rayleigh
waves were recognised as the main content (Blaik & Donn, 1953; Gutenberg, 1958;
Leet, 1947; Ramirez, 1940a; Ramirez, 1940b). Some evidence of Love-wave motion
was also found (Darbyshire, 1954; Gutenberg, 1958; Leet, 1947).
But, at that time, the mechanism by which the secondary microseismic noise was
generated remained unexplained since the pressure variations beneath a travelling
ocean wave were known to decrease exponentially with increasing the ocean depth.
Miche (1944) discovered the existence of second order pressure variations associated

38
Chapter 1. Secondary microseismic noise

to a standing ocean wave which are not attenuated with depth. Further contributions
came from Longuet-Higgins & Ursell (1948).
In 1950, Longuet-Higgins extended their results, showing that such pressure vari-
ations are sufficient to produce the observed ground movement if two ocean waves
of the same wavelength, but not necessary of the same amplitude, travel in opposite
directions. He computed the magnitude of the vertical displacement of the ocean
floor associated with the seismic wave field, showing that it was consistent with the
observations.
A different approach was taken by Hasselmann (1963), who expressed the theory of
the origin of microseisms in term of spectral transfer functions and the local energy-
balance equation of the seismic wavefield. This representation allows to adopt a
statistical approach.

1.3.2 The cause of seismic noise generation: the second-order

pressure fluctuation at the seafloor
Longuet-Higgins (1950) showed that the unattenuated second-order pressure term
in a standing wave pattern, firstly found by Miche (1944), was capable of generating
microseisms.
Let us consider a system of horizontal and vertical coordinates x and z, respec-
tively, with z measured downwards from the mean ocean surface. The ocean surface
is at z = 0 and the seafloor is at the depth z = h. The fluid motion is defined by the
velocity of its particles u = {u, w}. The fluid is assumed to be inviscid (i.e. without
viscosity), incompressible (i.e. ∇ · u = 0) and irrotational (i.e. ∇ × u = 0). In this
case, the equation of motion, that is the Navier-Stokes’ equation, can be written as to
the Bernoulli’s equation (see Longuet-Higgins, 1950, p.4; Kundu et al., 2012, p.128):

p − ps ∂Φ 1 2
− gz = − (u + w2 ) + θ(t) (1.1)
ρ ∂t 2

where p denotes the pressure, ps the pressure at the free surface (supposed constant),
ρ the density of the fluid and g the acceleration of gravity. Φ is the potential of the
velocity u, which depends on time t and it is periodic in x. The horizontal and vertical
components of the velocity u = {u, w} are given by u = −∂Φ/∂x and w = −∂Φ/∂z,
respectively. We recognise in 12 (u2 + w2 ) the kinetic energies per unit volume. Finally,
θ(t) is the variation in pressure with time t at a given depth.

39
Chapter 1. Secondary microseismic noise

Figure 1-5: Forces acting on a fluid between two fixed points on the horizontal direc-
tion, separated by a wavelength λ.

The evaluation of the pressure variation θ(t) at the ocean seafloor z = h will give
us the necessary condition which allows the generation of seismic noise:

• if θ(t) is zero or constant in time at z = h, there are no pressure variations and

then the displacement of the solid ground is null;

• if θ(t) varies with time at z = h, the pressure fluctuation in time is associated

with a non-null displacement of the seafloor.

In deep water approximation Φ, u, and w all diminish exponentially with increasing

z (Longuet-Higgins, 1950, p.5). Then, at the seafloor z = h, the Bernoulli’s equation
(1.1) can be simplified as:
p − ps
− gz = θ(t) (1.2)
ρ
To evaluate the pressure variation θ(t) at the ocean seafloor, we consider a portion
of fluid between two fixed points on the horizontal direction x, separated by a wave-
length λ. The fluid motion can be considered periodic, then the results we get for
a wavelength can be easily extended to the entire fluid. The external forces (Figure
1-5) acting on this volume of fluid are (Longuet-Higgins, 1952):

• the downward gravity field, which has a constant amplitude of λ(ρgh);

• the downward atmospheric load, which has a constant amplitude of λps ;

• the pressure fluctuation at the seafloor, equal to λp¯h , which depends on time.

40
Chapter 1. Secondary microseismic noise

Each couple of horizontal forces p1 and p2 acting on a fluid element vanishes each
other. Then, the total force acting on the volume of fluid can be written as:

F = λ(p¯h − ps − ρgh) (1.3)

The expression of F , the vertical component of the external forces acting on the mass
of water in Figure 1-5, is obtained by summing up the equation of motion for each
element of fluid: !
N  N
∂ 2 zi ∂2
X  X
F = mi 2 = 2 mi zi (1.4)
i=1
∂t ∂t i=1

assuming a constant mass mi for each particle of fluid i. The total number of particles
of fluid in the volume is N . The expression inside the brackets on the right side of
equation (1.4) is g −1 times the potential energy of the fluid U = g N
P
i=1 (mi zi ), written
in terms of the mass mi of the fluid particles, that is (Kundu et al., 2012, p. 264,
equation (7.41)):
N Z λ
X 1
(mi zi ) = ρ ξ(x, t)2 dx (1.5)
i=1 0 2

where ξ(x, t) is the vertical displacement of the free surface with respect to the equi-
librium position, that is the ocean surface elevation, as in Figure 1-5.

Then, by comparing equation (1.3) and equation (1.4) and using equation (1.5),
we find out λ
p¯h − ps 1 ∂2
Z
1
− gh = ξ(x, t)2 dx (1.6)
ρ λ ∂t2 0 2
which is the Bernoulli’s equation (1.2) evaluated at the seafloor z = h as average over
a wavelength λ of fluid. Then, the second order pressure variation θ(t), evaluated at
the seafloor, can be written by comparing equation (1.2) and equation (1.6):

λ
1 ∂2
Z
1
θh (t) = ξ(x, t)2 dx (1.7)
λ ∂t2 0 2

The pressure variation at the seafloor θh (t) depends on a second-order function of

the wave amplitude ξ(x, t), which is small with respect to the ocean depth h. This
is the reason why the pressure variation related to the microseisms generation is a
second-order term.

In the following, we are going to evaluate the pressure variation at the seafloor

41
Chapter 1. Secondary microseismic noise

θh (t) by considering three different shape functions for the ocean surface elevation
ξ(x, t): a) a progressive wave, b) the interaction of a progressive and a regressive
wave and c) a standing wave. In such way, we aim to find out the required condition
which allows the generation of the seismic noise wavefield.

a) The progressive gravity wave

Let us consider the case of a single progressive gravity wave. The ocean surface
elevation can be written as:

ξ(x, t) = a cos(kx − ωt) (1.8)

where k = 2π/λ is the wavenumber and ω is the frequency of the ocean wave. The
pressure variation at the ocean seafloor can be evaluated by equation (1.7):

1 ∂2 λ 1 2
Z
θh (t) = a cos2 (kx − ωt) dx
λ ∂t2 0 2
1 ∂2 λ 2
 
= a =0
λ ∂t2 4

Thus, there is no fluctuation in the mean pressure at the seafloor over one wavelength.
As a consequence, the pressure at the seafloor is constant (from equation (1.6)):

p¯h − ps
− gh = 0 → p¯h = ρgh + ps (1.9)
ρ

There is another reason why progressive waves may be expected to be ineffective in

producing microseisms (Longuet-Higgins, 1953): both the first-order pressure fluc-
tuation p and the velocity field u decrease exponentially with depth in deep water,
as it is shown in Figure 1-6a (see Kundu et al., 2012, Chapter 7). The deep water
approximation requires h >> λ, where λ is the wavelength of the ocean waves (tens
of metres for a gravity wave), which is the case almost everywhere in the ocean con-
sidering periods associated with the secondary microseismis. Only in case of shallow
water (Figure 1-6b) a progressive wave does not decay exponentially with depth and
the velocity field at the sea-floor is not null. In such case, the energy transmission to
the seafloor occurs at the same frequency of the ocean waves. This is the generation
mechanism of primary microseisms (Hasselmann, 1963), that we will not address in
this context.

42
Chapter 1. Secondary microseismic noise

Figure 1-6: Fluid particle orbits caused by an oscillatory surface wave traveling to
the right for three liquid depths. (a) When the liquid is deep, the particle orbits are
circular and decrease in size with increasing depth. (b) When the water is shallow,
the orbits are thin ellipses that become thinner with increasing depth. Adapted from
Kundu et al. (2012).

b) The interaction between two progressive gravity waves travelling in

opposite directions

Let us consider now two gravity waves of equal period but unequal amplitude, a1
and a2 , travelling towards opposite directions:

ξ(x, t) = a1 cos(kx − ωt) + a2 cos(kx + ωt) (1.10)

Then, from equation (1.7), we get the second-order pressure variation at the seafloor:

1 ∂2 λ 1
Z
θh (t) = [a1 cos(kx − ωt) + a2 cos(kx + ωt)]2 dx =
λ ∂t2 0 2
1 ∂2 λ 2
 
= 2


a + a2 + 2 a1 a2 cos(2ωt) = (1.11)
λ ∂t2 4 1
= −2 a1 a2 ω 2 cos(2ωt)

which is independent from z and dependent on time t. It fluctuates with an amplitude

proportional to the product between the wave heights a1 and a2 . The frequency of
the pressure variation is twice the frequency of the ocean waves. From equation (1.6),
the average pressure at the seafloor p¯h can be written as:

p¯h − ps
− gh = −2 a1 a2 ω 2 cos(2ωt) (1.12)
ρ

43
Chapter 1. Secondary microseismic noise

This is the type of oscillation required for the generation of the seismic wavefield
at the ocean bottom (Longuet-Higgins, 1950). The effect produced by ocean waves
acting at the surface of the ocean on the fluid particles in depth is the same as would
be produced by an oscillating pressure applied uniformly at the upper surface of the
ocean (Longuet-Higgins, 1952).

c) A standing wave

A particular case occurs when, in addition of having the same wavelength and
travelling towards opposite directions, the two ocean waves have the same amplitude:
a1 = a2 = a/2. In such case, the superposition of the two gravity waves gives rise to
a standing wave:

ξ(x, t) = a/2 cos(kx − ωt) + a/2 cos(kx + ωt) = a cos(kx)cos(ωt) (1.13)

In such case, the deflection of the liquid surface does not travel and the surface simply
oscillates up and down at frequency ω with a spatially-varying amplitude, keeping
the nodal point fixed.
The mean pressure at the ocean bottom, obtained by using equation (1.7), is:

1 ∂2 λ 1
Z
θh (t) = [a cos(kx)cos(ωt)]2 dx =
λ ∂t2 0 2
1 ∂2 λ 1 2
  
= a cos(2ωt) = (1.14)
λ ∂t2 4 2
1
= − a2 ω 2 cos(2ωt)
2

Like in the previous case, for a given wave period, the pressure fluctuation is inde-
pendent of depth. We remind that, so far, the fluid is considered incompressible.
The mean pressure fluctuation p¯h at the seafloor acts sinusoidally with twice the
frequency of the original wave and with the amplitude proportional to the square of
the standing wave amplitude:

p¯h − ps 1
− gh = − a2 ω 2 cos(2ωt) (1.15)
ρ 2

These second-order pressure variations arise only when the crossing ocean waves
(or ocean train waves) of the same wavelength travel in opposite directions. When
opposing wave trains of equal periods are not in direct opposition, the generated

44
Chapter 1. Secondary microseismic noise

pressure field decreases rapidly as the angle between the ocean waves (or ocean train
waves) differs from 180◦ . Is has been shown that, when angles exceed ∼ 10◦ , the
second-order pressure contribution quickly approaches that of the depth-attenuated
first-order term (Figure 1-6) and the seafloor pression variations become negligible
(Vigness et al., 1952).

Incompressible versus compressible ocean

This formulation of the theory, valid for an incompressible ocean, was verified
with laboratory experiments on periodic surface gravity waves by Cooper & Longuet-
Higgins (1951).
We remind that the incompressible hypothesis (∇ · u = 0) is valid as long as the
time t necessary to a disturbance to reach the ocean seafloor and come back to the
ocean surface is small compared to the period T of the wave (i.e the ocean depth h
is small compared to an half of the wavelength λ of the ocean wave). Then, in the
secondary microseismic period band, the incompressible hypothesis is valid as long
as it is true that

2h αw T
if T = 3s : t= << T → h << = 2.25 km
αw 2
(1.16)
2h αw T
if T = 10s : t= << T → h << = 7.5 km
αw 2

where αw = 1.5 km/s is the velocity of the acoustic wave in water. Then, for long
period noise, the incompressible hypothesis is valid almost everywhere, whereas for
short period noise, compressibility has to be taken into account.
The development of a more complete formulation of the generation theory, which
accounts for the ocean compressibility, was developed by Longuet-Higgins (1950).
Owing to the compressibility of the fluid, the displacement at the ocean seafloor due
to the pressure fluctuation at the surface of the ocean becomes ocean depth-dependent
(see section 1.5).

1.3.3 A statistical description of the origin of the microseisms

In general, an instantaneous picture of the sea surface does not indicate in which
direction the ocean waves are going. To specify the full sea state, velocity information
should be given in addition to the height of the sea surface everywhere. Moreover,

45
Chapter 1. Secondary microseismic noise

external conditions, like winds and tides, must be specified. Because it is impossible
to fully specify the sea state in a deterministic way, one normally uses a statistical
description and considers the probability of finding a particular sea state.
Since observations are not made in the storm area itself, a storm can be considered
as a disturbance of finite area, from which energy is propagated outward. As detailed
in the previous section, for ocean waves having equal frequency but opposite direc-
tions, the forcing is equivalent to horizontally uniform pressure oscillations applied
at the surface of the ocean. We point out that the wavelength of an ocean gravity
wave is a few hundred meters, that is small compared to the ocean depth. Then the
ocean wave-wave interaction and the resulting pressure can be considered as to be at
the surface of the ocean. The surface of the ocean can be divided into squares. The
amplitude of the pressure at the ocean surface from any given square is the same as if
the whole force was concentrated at the centre of the square. The displacement gen-
erated by the whole storm will be the summation of all the contributions coming from
the different squares (Miche, 1944; Longuet-Higgins, 1950). A statistical description,
based on the spectral transfer functions and the local energy-balance equation of the
seismic field, enables the determination of the microseismic field at arbitrary positions
within or outside the generating area (Hasselmann, 1963, p. 178).
The local sea-state can be described by using the concept of the directional power
spectrum of the wave-height Fs (f, θ), which takes into account the directionality of
the ocean waves (Ardhuin et al., 2011). The spectral density of the pressure field at
the surface of the ocean written in the frequency-wavenumber domain Fp (k ' 0, fs )
– where fs is the seismic frequency – can be written integrating over all directions
(Hasselmann, 1963, eq. 2.15 p. 191; Komen et al., 1994):
Z π
Fp (k ' 0, fs ) = ρ2w g 2 fs Fs (f, θ)Fs (f, θ + π)dθ (1.17)
0

where ρw is the water density, g is the gravity acceleration, f is the ocean wave
frequency. Fs (f, θ) and Fs (f, θ + π) are the wave-height spectral density of the two
colliding ocean wave trains having the same frequency f and opposite azimuths θ
and (θ + π). These two colliding ocean wave trains generate pressure fluctuations
Fp (k ' 0, fs ) whose wave number k ' 0 is the sum of the wave numbers of the two
opposite ocean waves and whose frequency is fs = 2f . Since the modulus of the two
wavenumbers is nearly the same, k ' 0, that is the pressure spectrum is white and
isotropic with respect to k. Because the integrand has a periodicity of π, it has been

46
Chapter 1. Secondary microseismic noise

used the fact that the integral over [0, 2π] is twice the integral over [0, π] (Ardhuin
et al., 2011).
For practical reasons, the wave-height spectral density Fs (f, θ) can be written as a
product between the frequency spectrum E(f ), which represents the power spectrum
of the ocean surface elevation variance [m2 /Hz], and the directional distribution
M (f, θ), which gives the distribution of the surface elevation variance over all the
directions from 0 to 2π at each frequency. The directional distribution M (f, θ) is
R 2π
dimensionless and it is normalised such that 0 M (f, θ)dθ = 1.
The power spectrum of the surface elevation variance E(f ) can usually be obtained
from the time series recorded by a buoy or satellite altimetry (Ardhuin et al., 2011).
The power spectrum of the elevation variance E(f ) is directly linked to the significant
ocean wave height Hs through the following equation (Ardhuin et al., 2011):
sZ
∞
Hs = 4 E(f )df (1.18)
0

These two quantities, E(f ) and M (f, θ), can be computed by using an ocean wave
model, which then allows to get the power spectral density of the pressure field (see
section 1.4.2).
By defining Fs (f, θ) = E(f )M (f, θ), the spectral density of the pressure field at
the surface of the ocean in equation (1.17) can be written as:
Z π
Fp (k ' 0, fs ) = ρ2w g 2 fs E 2 (f ) M (f, θ)M (f, θ + π)dθ (1.19)
0

where f is the ocean wave frequency and fs = 2f is the seismic wave frequency. The
spectral density FP (k ' 0, fs ) has dimension of [P a2 · m2 · s].
Writing the ocean surface elevation variance E(f ) of each one of the two inter-
acting ocean waves (or ocean train waves) as E1 = a21 /2 and E2 = a22 /2, where a1
and a2 are the amplitude of the ocean waves, one can demonstrate that the second
order pressure variation due to a standing wave at the surface of the ocean, as given
by equation (1.12), can be obtained by integrating Fp (k ' 0, fs ) over frequency and
wavenumber (Ardhuin et al., 2011, their Appendix A).
Predictions of microseismic sources by wave action models have been performed by
e.g. Kedar et al. (2008), Ardhuin et al. (2011), Hillers et al. (2012), Stutzmann et al.
(2012).

47
Chapter 1. Secondary microseismic noise

1.4 Locating the ocean wave-wave interaction

1.4.1 Ocean wave-wave interaction classes

In order to generate gravity wave-wave interaction, the motion of the sea sur-
face should contain at least some wave packets of the same wavelength, travelling in
opposite directions.
As shown in Figure 1-7, the sea state where the opposite wave motion is strong
enough to generate seismic noise can be divided in three classes (Longuet-Higgins,
1950; Longuet-Higgins, 1952; Ardhuin et al., 2011; Ardhuin et al., 2012):

1. Class I: ocean wave-wave interaction within a storm.

The first class occurs when a storm has a wide angular distribution, with ocean
waves coming from many different azimuths. The first class dominates at short
periods due to the wide angular distribution of the short period waves generated
by a constant and steady wind, and it may still be significant at somewhat higher
period. In that case, the interacting waves are within the storm.

2. Class II: ocean wave-wave interaction due to the coastal reflection.

In the second class, ocean waves arrive at the coast, they are reflected and they
meet up with other ocean waves that are travelling towards the coast at the
same time. Then, the interaction area is confined close to the coast.

3. Class III: ocean wave-wave interaction from different storms.

The third class concerns the interaction of ocean waves coming from different
storms, if they co-exists at the same time in different geographical regions.
Ocean waves from a given storm may travel long distances before meeting
up ocean waves generated by another storm. This third class generates the
strongest noise sources anywhere in the ocean basin. Because this class requires
the co-existence of two storms, it occurs quite rarely.

Moving cyclones

Hurricanes, tropical cyclones or typhoons are capable of generating extreme sea

states (Young, 2003). A moving cyclone can be considered a particular case of class
I. In such case, winds on one side of the storm will generate waves travelling in the
direction of motion of the storm (point A in Figure 1-8). When the storm moves

48
Chapter 1. Secondary microseismic noise

Figure 1-7: Schematic illustration of the three classes of ocean wave-wave interaction.
The thick arrows indicate the local wind direction and the small arrows indicate the
direction of wave trains. At point A, ocean waves of the same storm S1 interact with
each other (class I). At point B, ocean waves travelling towards the coast interact
with ocean waves reflected from the coast (class II). At point C, remote ocean waves
from the storm S1 interact with ocean waves generated by the storm S2 (class III).
After Ardhuin et al. (2012).

and turns, winds on the opposite side will generate waves travelling in the opposite
direction (point B in Figure 1-8). If the storm is moving faster than the group
velocity of the ocean waves, there will be a region, located in the tail part of the
cyclone (point C in Figure 1-8), where the two groups of waves will meet. Thus,
the tail of a fast-moving cyclone may expect to be a considerable source of wave-
wave interactions (Longuet-Higgins, 1950; Longuet-Higgins, 1952) . Sources of noise
body waves associated with specific storms have been found by Haubrich & McCamy
(1969), Gerstoft et al. (2006), Koper & de Foy (2008), Zhang et al. (2010a), Zhang
et al. (2010b), Gualtieri et al. (2014).

49
Chapter 1. Secondary microseismic noise

Figure 1-8: Wave-wave interaction caused by a moving cyclone. A and B: mov-

ing cyclone locations at two consecutive times. C: location of the ocean wave-wave
interaction when the cyclone is located at B. After Longuet-Higgins (1952).

Deep water and coastal reflection

Longuet-Higgins (1952) estimates the amplitude of the ground displacement gen-

erated by class I and class II (Longuet-Higgins, 1952, p. 83). He computed the
peak-to-trough amplitude of the displacement due to two ocean waves having slightly
similar periods and ocean depth of h = 3 km in case of class I, h = 0 km in case of
class II. In case of coastal reflection, he assumed the amplitude of the reflected ocean
wave as to 5% smaller than the amplitude of the incident one (amplitude of 0.1 m
for the reflected wave and 2 m for the incident one). The displacement generated by
the class I resulted to be dominant (6.5µm) with respect to the one generated by the
class II (0.3µm).
The coastal reflection coefficient is not well constrained and should be adjusted
as a function of the coast shape (Ardhuin & Roland, 2012). Stutzmann et al. (2012)
provided empirical coastal reflection coefficients associated with seismic noise recorded
at stations in various environments and showed, in agreement with Longuet-Higgins
50 yr earlier, that the strongest noise sources are in deep ocean and that coastal

50
Chapter 1. Secondary microseismic noise

reflection generates numerous smaller sources. The coastal reflection does not give
rise to the largest disturbances in inland stations even though it may be a more
common cause of microseisms near to the coast (Longuet-Higgins, 1952).
Evidence of dominant noise sources due to the class I in deep water have also
been found by Webb & Constable (1986), Cessaro (1994), Stehly et al. (2006), Kedar
et al. (2008), Obrebski et al. (2012) for noise Rayleigh waves and by Gerstoft et al.
(2008), Koper et al. (2009), Koper et al. (2010), Landès et al. (2010) for noise body
waves. Differences in the spatial distribution of Rayleigh and body wave sources were
observed by Obrebski et al. (2013) in the North Atlantic Ocean.
Evidence of strong sources associated with Rayleigh waves due to class II in shallow
water have been documented by Darbyshire (1992), Bromirski et al. (1999), Bromirski
(2001), Bromirski & Duennebier (2002), Bromirski et al. (2005), Essen et al. (2003),
Schulte-Pelkum et al. (2004), Gerstoft & Tanimoto (2007), Tanimoto (2007), Yang &
Ritzwoller (2008).
Evidence of both class I and class II have been found by Haubrich & McCamy
(1969), Friedrich et al. (1998) and Chevrot et al. (2007a).
Haubrich & McCamy (1969) demonstrates that the deep-water generation may
occur only when certain conditions exist, such as when the velocity of the storm
exceeds that of the swell it generates.
Bromirski et al. (2013) show that long period noise (i.e. T > 5 s) can be explained
by deep water source, whereas short period noise (i.e. T < 5 s) are better explained
by coastal sources.
Noise sources of class III have been documented by Obrebski et al. (2012). They
identified such sources located midway between Hawaii and California, and recorded
by stations several thousands of kilometres away.
The relevance of deep water and coastal regions on noise generation is still under
debate.

1.4.2 The ocean wave model WAVEWATCH III(R)

In order to find out the power spectral density of the pressure field generated by
the ocean wave-wave interaction at the ocean surface, the directional power spectrum
of the wave-heigh F (f, θ) = E(f )M (f, θ) in equation (1.19) can be computed by using
the numerical ocean wave model WAVEWATCH III(R) (Tolman, 1991; Tolman, 2009).
This ocean wave model has been developed at the Marine Modeling and Analysis

51
Chapter 1. Secondary microseismic noise

Branch (MMAB) of the Environmental Modeling Center (EMC) of the National Cen-
ters for Environmental Prediction (NCEP). Significant recent improvements concern
the parameterization for wind-wave generation and dissipation in order to reproduce
as well as possible a wide variety of observations (Ardhuin et al., 2010). The model
can be accessed from http://polar.ncep.noaa.gov/waves/wavewatch/distribution.
The model is defined at global scale and it has a resolution of 0.5◦ both in lat-
itude and in longitude. It accounts for 3–hourly wind analysis from the European
Centre for Medium Range Weather Forecasting (ECMWF), daily ECMWF sea ice
concentration analysis and monthly Southern Ocean distribution statistics for small
icebergs (Ardhuin et al., 2011).
At each grid point, the state of the ocean is described as a function of the frequency
f and the directional wave spectrum M (f, θ), with 31 frequencies and 24 azimuths.
The ocean wave frequencies are spaced exponentially between 0.04 Hz (T = 24.4s)
and 0.17 Hz (T = 5.8 s). The seismic noise sources, generated by the superposition of
waves of similar frequencies and opposite directions, are then defined for 16 seismic
frequencies, between 0.08Hz (T = 12.2s) and 0.34 Hz (T = 2.9s).
All the three classes of ocean wave-wave interactions (section 1.4.1) are taken
into account. One key point of this model is that it is the only model to date which
takes into account coastal reflection of ocean waves at the ocean-land boundary (noise
sources of class II, section 1.4.1) and around icebergs (Ardhuin et al., 2010; Ardhuin
et al., 2011). The non-dimensional reflection coefficients, related to the reflected wave
height, can be varied from 0 to 10% (in agreement with Longuet-Higgins (1950) who
used an average value of 5%, as we have seen in the previous section).

1.4.3 Ocean wave height and ocean wave-wave interaction

The spectral density of the pressure field at the ocean surface Fp (k ' 0, fs ),
as given by equation (1.19), can then be evaluated by using the ocean wave model
WAVEWATCH III(R) . The power spectral density of the pressure field, as a func-
tion of the frequency, can be rewritten from the wavenumber to the spatial domain
discretising the surface of the ocean as:

Fp (k ' 0, fs )
|P (f )| = (2π)2 (1.20)
dS

52
Chapter 1. Secondary microseismic noise

where dS = R2 sin(φ)dλdφ is the elementary surface, φ is the co-latitude, λ the

longitude and R is the radius of the Earth. The power spectral density of the pressure
field |P (f )| has dimension of [P a2 · s]. The factor (2π)2 allows to rewrite the spectral
density of the pressure field from the wavenumber domain to the spatial domain
(Gualtieri et al., 2013, p. 2).

Figures 1-9 and 1-10 display the ocean wave height (Figures 1-9a and 1-10a) and
the power spectral density of the pressure field due to the ocean wave-wave interaction
as computed by equation (1.20) for three hours (12:00-15:00 UTC) on 2008 January
1st (Figure 1-9b-d) and 2008 August 1st (Figure 1-10b-d). The power spectral density
has been computed at three (seismic) periods: T = 3.9 s (Figures 1-9b and 1-10b),
T = 5.2 s (Figures 1-9c and 1-10c) and T = 6.9 s (Figures 1-9d and 1-10d).
The location of the highest ocean waves and the maxima of the ocean wave-wave
interactions are related to the local winter (northern hemisphere on January 1st,
Figure 1-9, and southern hemisphere on August 1st, Figure 1-10).
Making a comparison between the ocean wave height (Figures 1-9a and 1-10a) and the
power spectral density (Figures 1-9b-d and 1-10b-d), we observe that their maxima
have a different location. Moreover, the maxima of the ocean wave-wave interaction
vary for varying period.

Figures 1-11 presents the ocean wave height (leftmost column) and the power
spectral density of the pressure field as given by equation (1.20) in case of an extreme
meteorological event: the thypoon Ioke, which occurred at the beginning of September
2006. The power spectral density has been computed at the seismic wave period
T = 3.9 s (central column) and T = 5.2 s (rightmost column). Each row is referred to
three hours (12:00-15:00 UTC) from 2006 September 1st (first row) to 2006 September
5th (last row). The black line shows the typhoon track.

As explained in section 1.4.1, in case of a cyclone moving faster than the group
velocity of the ocean waves, the interaction among ocean waves happens in the tail
region of the cyclone. We observe this effect at both periods of T = 3.9 s and T = 5.2
s in Figure 1-11. Moreover, looking at the same day for both periods (i.e. fixing a
row in Figure 1-11), the maximum power spectral density region at T = 3.9 s (central
column) is different in shape and amplitude with respect to T = 5.2 s of period
(rightmost column). From Figures 1-9, 1-10 and 1-11 we clearly see that the power
spectral density of the pressure field, which is directly linked to the ocean wave-wave
interaction, strongly varies with period.

53
Chapter 1. Secondary microseismic noise

Figure 1-9: a) Ocean wave height and b)-d) wave-wave interaction as given by equa-
tion (1.20) on 2008 January 1, 12:00-15:00 UTC. The ocean wave-wave interaction
has been computed by using the numerical ocean wave model WAVEWATCH III(R)
at the following periods: b) T = 3.9 s, c) T = 5.2 s and d) T = 6.9 s.

54
Chapter 1. Secondary microseismic noise

Figure 1-10: a) Ocean wave height and b)-d) wave-wave interaction as given by
equation (1.20) on 2008 August 1, 12:00-15:00 UTC. The ocean wave-wave interaction
has been computed by using the numerical ocean wave model WAVEWATCH III(R)
at the following periods: b) T = 3.9 s, c) T = 5.2 s and d) T = 6.9 s.

55
Chapter 1. Secondary microseismic noise

Figure 1-11: Maps of the ocean wave height (leftmost column) and power spectral
density of the pressure field as given by equation (1.20) at T = 3.9 s (central column)
and T = 5.2 s (rightmost column). The computation has been performed for three
hours (12:00-15:00 UTC) from 2006 September 1st (first row) to 2006 September 5th
(last row). The black line shows the typhoon track.

56
Chapter 1. Secondary microseismic noise

1.5 Seismic noise sources and the ocean site effect

Seismic noise sources can be considered at the surface of the ocean since the
wavelength of the ocean gravity waves is small with respect to the ocean depths
(hundreds of metres versus kilometres). Because noise sources are located at the
surface which encloses the volume of the Earth, the displacement at a receiver is
simply due to the convolution of the Green function at the free surface and the
source traction. The traction is a pressure perpendicularly oriented with respect to
the surface of the ocean (elastodynamic representation theorem, Aki & Richards,
2002, chapter 2; see also chapter 3, section 3.3 in this manuscript).

The concept of the ocean site effect on seismic waves generated by a noise source
located at the surface of the ocean has been introduced by Longuet-Higgins (1950).
Since the wavelength of the ocean gravity waves is small compared to the ocean depth,
noise sources can be assumed to be at the upper surface of the ocean. The sources
emit P-waves, which are multiply reflected inside the ocean layer and transmitted to
the solid medium as P- and S-waves (see Chapter 3 for further details). Then, the
ocean site effect represents the effect produced by the propagation inside the ocean
on the seismic wavefield generated beneath the seafloor.

In order to compute the noise source amplitude, the wave-wave interaction pres-
sure field has to be modulated by the ocean site effect. Noise sources can be computed
as the product between the power spectral density of the pressure field |P (f )|, as given
by equation (1.20), and the ocean site effect c(fs , h):

S(fs ) = c(fs , h) × |P (fs )| (1.21)

In addition to the variability with the bathymetry, the ocean site effect shows a
frequency-dependent behaviour. It allows to reveal the ocean regions where the
strongest noise generation can potentially occur. The water depth has its strongest
effect on the source regions. As pointed out by Longuet-Higgins (1950, p. 34), “with
an ocean of non-uniform depth, the amplitude [of noise Rayleigh waves] will be af-
fected by the depth of water at all points between the generating area and the observing
station. Since, however, the energy density is greatest near the source of the distur-
bance, the depth of water in the generating area itself may be expected to be of the
most importance.“

57
Chapter 1. Secondary microseismic noise

1.5.1 The ocean site effect on Rayleigh waves

The ocean site effect on the different normal modes – denoted as n – of Rayleigh
waves cn = cn (fs , h) has been computed by Longuet-Higgins (1950). His computation
has been carried out analytically in a simple model of a liquid layer overlying a solid
half space. The elastic properties of the Earth model used by Longuet-Higgins (1950)
are summarised in Table 1.1.

Density P-wave velocity S-wave velocity

Liquid layer ρw = 1.0 g/cm3 αw = 1.4 km/s βw = 0 km/s
3
Solid half space ρc = 2.5 g/cm αc = 4.85 km/s βc = 2.8 km/s

Table 1.1: Density and body wave speeds in the liquid layer and in the solid half
space as used by Longuet-Higgins (1950).

These functions display a resonant behaviour for some combination of frequency

and ocean depth, meaning that, fixing the frequency, there is an ocean depth which
produces the strongest noise signal in the solid crust below the ocean.
The first mode or fundamental mode, modulated by the coefficient c1 , has no nodal
plane in depth between the ocean surface and the seafloor, the second one has one
nodal plane, the third two and so on. When the depth h is small only the fundamental
mode can exist; the others appear successively when the depth increases. Longuet-
Higgins explained the presence of peaks as the effect of resonance in the water layer.
The ocean site effect depends on the ratio σh/βc , where σ = 2πfs is the angular
frequency of the seismic waves, h is the water depth and βc is the S-wave velocity
in the solid halfspace. It accounts for the effect of the ocean depth on the discrete
modes of Rayleigh waves.
Figure 1-12 shows the coefficients c1 , c2 , c3 and c4 as a function of the product
between the ocean depth h and the seismic frequency σ normalised by the S-wave
velocity in the elastic medium βc . The ocean site effect cn related to the overtones
(n ≥ 2) presents two maxima because the group-velocity curves associated with the
overtones have two stationary values (Longuet-Higgins, 1950; Longuet-Higgins, 1952;
Kedar et al., 2008; see also Chapter 2, section 2.6).
Figure 1-13 displays maps of the ocean site effect coefficient c1 – as computed
by Longuet-Higgins (1950), his Table 1, p. 30 – for fixing periods from 3 to 10 s.
Comparing maps in Figure 1-13 with the bathymetry in Figure 1-14, we observe that
the maximum site effect occurs in different regions characterised by different ocean

58
Chapter 1. Secondary microseismic noise

Figure 1-12: The ocean site effect cn as a function of the product between the ocean
depth h and the seismic frequency σ = 2πfs , normalised by the S-wave velocity in
the elastic medium βc . Adapted from Longuet-Higgins (1950).

depths for varying period. The maximum site effect is confined in coastal areas at
short periods, whereas it is spread in wide regions at long periods.
In order to get the location of Rayleigh wave noise sources, we compute the product
between the power spectral density of the pressure field as given by equation (1.20) and
the ocean site effect as computed by Longuet-Higgins, as given by equation (1.21). In
Figure 1-15, on the left column, it has been plotted the wave-wave interaction location
(equation (1.20)) and, on the right column, the noise sources (equation (1.21)). The
computation has been carried out for three fixed seismic periods: T = 3.9 s, T = 5.2
s and T = 6.7 s. The location of noise sources (right column) does not correspond to
the location of the wave-wave interaction maxima. This is due to the ocean site effect
which modulates the power spectral density of the pressure field. For example, on
2008 January 1st, at these frequencies, the north Atlantic Ocean presents quite strong
noise source even if the wave-wave interaction is not so strong. Further considerations
about the ocean site effect on Rayleigh waves in different Earth spherical models,
computed using normal modes, can be found in Chapter 2.

59
Chapter 1. Secondary microseismic noise

Figure 1-13: Maps of the ocean site effect c1 – as it has been tabulated in Longuet-
Higgins (1950), Table 1 – for fixing periods from 3 s to 10 s.

60
Chapter 1. Secondary microseismic noise

Figure 1-14: World ocean bathymetry.

1.5.2 The ocean site effect on body waves

The ocean site effect computed by Longuet-Higgins (1950) only concerns Rayleigh
waves. The effect due to the ocean depth on body waves has not been taken into
account and the Longuet-Higgins’ theory does not predict the observation of body
waves.

Vinnik (1973) proposed a theory for the generation of compressional (P) waves,
but he carried out the computation in a solid half space, without considering the
effect of the water layer. Ardhuin & Herbers (2013) included the water layer and
they computed the site effect on body waves by using the local mode formalism. The
complete computation of the site effect on body waves, carried out by using the plane
wave decomposition can be found in Chapter 3.

The comparison between the ocean site effect on Rayleigh and body waves can be
found in Chapter 3, section 3.4. The ocean site effect on body waves acts differently
with respect to Rayleigh waves, amplifying different depths especially at short period.

61
Chapter 1. Secondary microseismic noise

Figure 1-15: Maps of the ocean wave-wave interaction (column on the left) as given
by equation (1.20) and noise sources (column of the right) as given by equation (1.21)
on 2008, January 1st and 2008, August 1st (12:00-15:00 UTC). The computation has
been carried out for three fixed periods: T = 3.9 s, T = 5.2 s and T = 6.7 s. The
maxima of the ocean wave-wave interaction do not correspond to the maxima of noise
sources.
62
Chapter 1. Secondary microseismic noise

1.5.3 What about Love waves?

Although microseisms are composed predominantly of Rayleigh waves, Love waves
have been observed too. These surface waves arise from constructive interference be-
tween horizontally-polarized S waves, which are not generated by assuming a vertical
force applied at the ocean surface in a 1D model. Then, the model relying on vertical
pressure load cannot be applied straightforwardly to the generation of Love waves
and thus a different mechanism is required. Nowadays, an explanation for Love-wave
generation has not yet been provided, but several mechanisms have been suggested.
A number of studies support Love waves being generated near coastlines (e.g.
Darbyshire, 1954; Haubrich & McCamy, 1969; Lacoss et al., 1969; Rind & Down,
1979; Friedrich et al., 1998; Koper et al., 2009; Koper et al., 2010). These imply that
the generation of Love waves is related to the interaction of ocean gravity waves with
the ocean floor at depths comparable to or less than ocean swell wavelengths.
Love waves may also be generated by the shear component of traction, generated
when the seafloor is not perfectly flat (e.g. Rind & Down, 1979; Nishida et al., 2008;
Fukao et al., 2010; Saito, 2010).
Love waves are not seen at all seismic stations around the world because the veloc-
ity and structural conditions necessary for their propagation do not exist everywhere
(Rind & Down, 1979). Thus, it is difficult to ascertain whether these mechanisms
generate Love waves to a degree consistent with observations.

1.6 Propagation of seismic noise inside the Earth

As a consequence of the elastodynamic representation theorem (Aki & Richards,
2002, chapter 2), the source and the propagation can be decoupled and considered
separately. Like any other elastic wave, microseismic waves dissipate energy both
at tectonic or structural boundaries and along the source-receiver path. Microseisms
traveling as surface waves from the oceanic environment dissipate a significant amount
of energy crossing the continental margin, especially at periods comparable with the
secondary microseisms (McGarr, 1969). This effect is due to lithologic and structural
differences occurring there. However, low attenuation within continental cratons al-
lows the propagation of microseisms to great distances (Given, 1990).

63
Chapter 2

Modeling noise Rayleigh waves

2.1 Introduction
In this chapter, we use normal mode theory to retrieve some features of seismic
noise Rayleigh waves and model the amplitude of the secondary microseismic noise,
whose generation theory was detailed in the previous chapter. The seismic period of
interest goes from 4 s to 10 s.
We first introduce the concept of normal modes of the Earth (section 2.2) and
we present the technique of normal mode summation to compute synthetic seismo-
grams (section 2.3). Then, we detail the theory of normal mode summation focusing
on a specific kind of source: a generically oriented single force (section 2.4) and a
vertical single force (section 2.4.1). A vertical single force allows us to simulate a
noise source, which generates only Rayleigh waves, as we have seen in the previous
chapter. A detailed discussion about the behavior of normal modes associated with
Rayleigh waves in various Earth models is presented in section 2.5. We then compute
the site effect (see Chapter 1 for further explanations) produced on noise Rayleigh
waves in different Earth models (section 2.6) by using normal modes. Doing that, we
demonstrate that the synthetic seismogram computation accounts for the site effect.
The ocean thickness effect on the group velocity and the phase velocity of Rayleigh
waves is also shown.
The normal mode summation code adapted to model the amplitude of noise spec-
tra is described in section 2.8. In the end of this chapter, we present the modeling of
seismic noise amplitude and seasonal variations, as it is in Gualtieri et al. (2013)1 .
1
Gualtieri, L., Stutzmann, E., Capdeville, Y., Ardhuin, F., Schimmel, M., Mangeney,

65
Chapter 2. Modeling noise Rayleigh waves

2.2 Normal modes

2.2.1 Normal modes and seismic waves

Let us consider a spherically symmetric, non-rotating, elastic and isotropic (SNREI)

Earth model. Because of the symmetry of the problem, each property of the Earth,
defined at a given point r = (r, θ, φ), varies only with the radius r. The equation of
motion can be solved writing the displacement as a superposition of normal modes
or eigenfunctions uk , defined at discrete frequencies ωk , called eigenfrequencies:
X
u(r, t) = ak · uk (r) exp(iωk t) (2.1)
k

where ak is the excitation coefficient for the k-th mode related to the source mecha-
nism (Aki & Richards, 2002, pp.342-343; Dahlen & Tromp, 1997, p. 109).
Looking for an oscillatory solution, equation (2.1) has to be transformed from the
time domain to the frequency domain using the Fourier transform (Dahlen & Tromp,
1997, p.109): Z ∞
u(r, ω) = u(r, t) exp(−iωk t)dt (2.2)
−∞

We encapsulate in the index k all quantum numbers (q, n, l, m). The index q is
associated with two families of Earth vibrations, called spheroidal and toroidal modes.
The index n is called radial order and it indicates the number of zero crossings along
the radius of the Earth; l is the angular order and it is related to the number or
nodal motion lines along the surface of the Earth. The quantum numbers n and l
can only take integer values, theoretically up to infinity. The set of (2l + 1) modes
associated with a given value of l form a multiplet, associated with all integer values
of m, from −l to +l. In a SNREI Earth model this multiplet is degenerate – i.e.
the eigenfunction splitting with m does not occur – and the eigenfunctions do not
depend on m (Dahlen & Tromp, 1997, p.272). For this reason, the dependence on m
will be neglected in this manuscript and we encapsulate in k the quantum numbers
k ≡ (q, n, l).
Two eigenfunctions uk and uk0 , associated with distinct eigenfrequencies ω 6= ω 0 ,

A., & Morelli, A., 2013. Modeling secondary microseismic noise by normal mode sum-
mation, Geophysical Journal International, 193(3), 1732-1745, doi:10.1093/gji/ggt090.

66
Chapter 2. Modeling noise Rayleigh waves

are orthogonal in the sense that

Z
ρ(r)u∗k (r) · uk0 (r)dV = 0 if ω 6= ω 0 (2.3)
VE

where ∗ denotes the complex conjugate and VE is the whole volume of the Earth (Aki
& Richards, 2002, p.345; Dahlen & Tromp, 1997, p.279). This is the reason why an
eigenfunction uk is referred to as a “normal mode”.
If uk is an eigenfunction associated with ω and uk0 is associated with the same
eigenfrequency ω = ω 0 , then the normalisation condition can be written as
Z
ρ(r)uk (r)∗ · uk0 (r)dV = 1 if ω = ω0 (2.4)
VE

The normal mode which corresponds to n = 0 is called fundamental mode and it has
no internal nodes within the system, i.e. it has not places where the motion is zero
within the system. Normal modes with n > 0 are called higher modes or overtones.
Each n-order overtone has n internal nodes. Spheroidal modes with l = 0 have all
motion in the radial direction and they are called radial modes.
Normal modes can be classified in two basic types: (1) spheroidal modes, denoted
as n Sl , equivalent to the P , SV and Rayleigh waves, which have a radial motion in
spherical geometry; (2) toroidal modes, denoted as n Tl , equivalente to SH and Love
waves, which have a shear motion parallel to the Earth’s surface. As we will see in
the following sections, these two sets of solutions are independent.
Figure 2-1 summarises the surface motion for the first three spheroidal and toroidal
modes. The first row shows three examples of spheroidal modes. Mode 0 S0 involves
expansion and contraction of a sphere as a whole and it has a period of T0 S0 = 20 min
in the Earth. Mode 0 S1 does not exist for the Earth. It corresponds to a horizontal
shift of a center of gravity, which can occur only in presence of an external force.
Mode 0 S2 is called “football mode” and it has a period of T0 S2 ' 54 min.
The second row is referred to the first three toroidal modes. Mode 0 T1 cannot
exist in the Earth because it violates the conservation of the angular momentum.
It would generate oscillations in the rate of rotation of the whole Earth. Mode
1 T2 corresponds to alternating twisting of the entire upper and lower hemispheres
and it has a period of about T1 T2 ' 13 min. Mode 1 T3 involves the alternating
twisting of polar and equatorial zones of a sphere with a period of T1 T3 ' 12 min
in the Earth. The spheroidal and toroidal multiplets n Sl and n Tl , are equivalent, in

67
Chapter 2. Modeling noise Rayleigh waves

Figure 2-1: Cartoon showing the surface vibrations associated with the first three
spheroidal modes – 0 S0 , 0 S1 and 0 S2 – (upper row) and toroidal modes – 0 T1 ,
0 T2 and 0 T3 – (lower row) on a sphere. See the text for the details concerning the
corresponding normal modes of the Earth.

the limit n << l, to propagating fundamental and higher-mode Rayleigh and Love
waves. A single mode, viewed in isolation, has a standing-wave pattern. Indeed, the
horizontal oscillatory motion at each node is zero, so that a single mode alone can
not propagate horizontally (Lay & Wallace, 1995). The oscillations that move along
the surface of the Earth and that we view as travelling Rayleigh and Love waves
are due to constructive interferences of the coexisting modes. All the internal body-
wave motions can be represented summing up a sufficient number of normal modes,
depending on the epicentral distance and the Earth model.

2.2.2 Normal modes and gravity

Gravity does not influence toroidal modes at all because these modes involve
only shear displacement. Instead, spheroidal modes act along the same direction as
gravity. Then a mutual interaction can occur between the spheroidal modes and
the gravity field. In Figure 2-2, we show the dispersion diagram of the fundamental
and the first higher mode computed considering the PREM model in presence (red)
and in absence (blue) of gravity. The normal mode angular order l is plotted versus

68
Chapter 2. Modeling noise Rayleigh waves

Figure 2-2: Dispersion diagram showing the spheroidal fundamental mode n = 0 and
the first spheroidal overtone of the PREM model in presence of gravity (red dots)
and in absence of gravity (blue dots). The spherical harmonic degree l is plotted as a
function of the eigenfrequency ωk . The dispersion diagrams have been computed for
frequencies a) from 0.0001 Hz to 0.002 Hz (T=500 − 1000 s) b) from 0.1 Hz to 0.2 Hz
(T=5 − 10 s). Only long period normal modes are affected by gravity.

the eigenfrequency ω. Figure 2-2a and Figure 2-2b show two extreme cases in the
frequency domain, very low frequency, ω = (0.1 − 2) × 10−3 Hz, and very high
frequency, ω = 0.1 − 0.2 Hz, respectively. The case in Figure 2-2b corresponds to the
secondary microseismic frequency band.
At very low frequency (Figure 2-2a), the presence of the gravity field (red dots)
produces, for the same frequency, a shift in terms of angular order. Thus, the same
normal modes are excited at different frequencies in presence or in absent of gravity.
Instead, at very high frequency, the influence of gravity becomes negligible. From
Figure 2-2b, we can assert that the secondary microseismic noise is not affected by
gravity.

2.3 Synthetic seismograms

2.3.1 Generic analytical formulation

In this section we deal with the computation of synthetic seismograms s(r, t) by

summing up normal modes. We consider a SNREI – spherically symmetric, non-
rotating, elastic and isotropic – Earth model and, for practical reasons, we employ a
system of spherical-polar coordinates, r = (r, θ, φ), with its origin at the centre of the
Earth r = 0.

69
Chapter 2. Modeling noise Rayleigh waves

Following Gilbert (1970), Gilbert & Dziewonski (1975) and Aki & Richards (2002),
the displacement calculated at a certain time t and at a fixed point having coordinates
r = (r sin θ cos φ, r sin θ sin φ, r cos θ) due to a source f (r) can be written as:

X Z 
s(r, t) = u∗k · f (r) dV uk g(t) (2.5)
k VE

where g(t) is the source time-function. If we consider an instantaneous source at

t = t0 expressed as a Dirac delta-function δ(t − t0 ), g(t) can be written as (Dahlen &
Tromp, 1997, p. 121):
sin(ωk t)
g(t) = H(t − t0 ) (2.6)
ωk
where H(t − t0 ) is the Heaviside step-function.
The integral in equation (2.5) expresses the source contribution in the spatial domain
to the synthetic seismogram. It is called source term and it is valid for any kind of
source mechanism f (r), which occurs instantaneously at t = t0 :
Z
Sk (rs , θs , φs ) = u∗k · f (r) dV (2.7)
VE

The expression of the synthetic seismogram in the time domain can be obtained
by operating on equation (2.5) the projection on the radial, transverse and vertical
receiver components. The “instrumental vector” v is then defined as a unit vector in
the direction of motion applied at the receiver location. Thus, we call receiver term
the internal product:
Rk (rr , θr , φr ) = v · uk (rr , θr , φr ) (2.8)

At this stage, it is possible to write the generic analytical expression of the syn-
thetic seismogram due to a generic source f (r) as:

u(r, t) = v · s(r, t) =
X Z 
∗
= uk · f (r) dV v · uk g(t) =
k VE
X exp(iωk t)
= Rk (rr , θr , φr )Sk (rs , θs , φs ) (2.9)
k
ωk2

in which we recognise the source Sk (rs , θs , φs ) and the receiver Rk (rr , θr , φr ) term (for
further details, see Gualtieri et al., 2013, their Appendix A).

70
Chapter 2. Modeling noise Rayleigh waves

The previous equation highlights that the source and the receiver effects are indepen-
dent and that their contributions can be evaluated separately. The equation (2.9) also
shows that a symmetry existes between the source mechanism f (rs ) and the receiver
vector v(rr ) (Woodhouse & Girnius, 1982).

2.3.2 Propagation

The exponential term in equation (2.9) is related to the propagation of seismic

waves. For a realistic Earth model with anelastic attenuation, the normal mode
eigenfrequency is complex (e.g. Takeuchi & Saito, 1972; Dahlen & Tromp, 1997;
Yang et al., 2010). The real part ωk characterizes the temporal oscillation pattern
and the imaginary part γk the attenuation of the modes:

ωk → ωk + iγk (2.10)

in which:
1
γk = ωk qk−1 (2.11)
2
where qk is the quality factor which corresponds to the k−th mode, called intrinsic
or modal quality factor. The coefficient γk = 21 ωk qk−1 represent a decay rate due to
the attenuation.
The inverse modal quality factor qk−1 is linked to the inverse of the bulk Q−1
K and
the shear Q−1
µ quality factor towards a linear relation:

Q−1 −1 −1
k = fK QK + fµ Qµ (2.12)

where fK and fµ are the compressional and shear energy densities of each modes.
This relation is valid when the modal quality factor Qk is weakly dependent on the
Earth radius, which is the case of the SNREI Earth model (Dahlen & Tromp, 1997,
p. 349).
Introducing in equation (2.9) the complex frequency (2.10), the expression of the
synthetic seismogram for a realistic anaelastic Earth model can be written as:
 h i 
ωk
X exp i ωk + i 2qk t
u(r, t) = Rk (rr , θr , φr )Sk (rs , θs , φs ) h i2 (2.13)
ωk
k ωk + i 2qk

71
Chapter 2. Modeling noise Rayleigh waves

ωk
Since γk = 2qk
<< ωk , for practical reasons, the following equation can be used
(Dahlen & Tromp, 1997, p. 371):
 
ωk
X exp − 2qk
t
u(r, t) = Rk (rr , θr , φr )Sk (rs , θs , φs ) (2.14)
k
ωk2

The effect of the anelasticity on the radial eigenfunctions U , V and W is generally

ignored for practical calculations of synthetic seismograms on a spherical Earth model
and only its effect upon the eigenfrequencies is retained (Dahlen & Tromp, 1997, p.
370).

2.3.3 Spherical and canonical coordinates

At this stage, to find out the oscillatory solution, we need to rewrite the eigenfunc-
tion uk in the frequency domain and spherical coordinates. The spherical symmetry
of the SNREI model allows us to split the contribution due to the vertical and hor-
izontal displacement. The eigenfunction uk can be written in terms of scalar radial
eigenfunctions U (r, ω), V (r, ω) and W (r, ω) (Dahlen & Tromp, 1997):

uk (r, ω) = U k (r, ω)Rk (r, ω) + V k (r, ω)Sk (r, ω) + W k (r, ω)Tk (2.15)

where Rk (r, ω), Sk (r, ω) and Tk (r, ω) of degree 0 ≤ l ≤ +∞ are defined as (Dahlen
& Tromp, 1997, p. 268-269 and Capdeville, 2002);

 Rk (r, ω) = er Yk (Θ, φ)


Sk (r, ω) = ∇1 Yk (Θ, φ) (2.16)

 T (r, ω) = −(e × ∇ )Y (Θ, φ)

k r 1 k

in which Yk (Θ, φ) = Yl (Θ, φ) is the scalar spherical harmonics having angular order l
and eˆr is the unit outwards vector acting perpendicularly to the unit sphere (Dahlen
& Tromp, 1997, Annex B). The surface gradient ∇1 is defined as the gradient on a
unit sphere:
∇1 f (r) = ∂θ f (r)eθ + (sin θ)−1 ∂φ f (r)eφ (2.17)

Combining the equation (2.15) and the (2.16), it is possible to observe that the
spheroidal modes come out for W = 0 and the toroidal modes for U = V = 0. The
two kinds of oscillations are then decoupled and independent.

72
Chapter 2. Modeling noise Rayleigh waves

The linear combination of tensors in equation (2.15) can be simplified adopting

the canonical coordinate system and expanding them in terms of generalized spherical
harmonics (Phinney & Burridge, 1973; Gilbert, 1970). The components of a generic
vector in canonical coordinates can be expressed as a function of the components in
spherical coordinates (e.g. Phinney & Burridge, 1973, p. 452 and their Appendix B):

− √1 (uθ
 u = + iuφ )

 2
u= 0 (2.18)
u = ur

 u+ = √1 (−u + iu )

2 θ φ

and the canonical basis {e+ , e0 , e− } in terms of the spherical basis {er , eθ , eφ }:

√1 (eθ − ieφ )
 e− =

 2
e= e0 = er (2.19)

 e = √1 (−e − ie )

+ 2 θ φ

By using the definition of a vector in canonical coordinates, equation (2.15) becomes:


− l −
 uk = γl Ω0 ( nVl − i nWl )Yl (Θ, φ)


uk (r, ω) = u0k = γl nUl Yl0 (Θ, φ) (2.20)

 u+ = γ Ωl ( V + i W )Y + (Θ, φ)

k l 0 n l n l l

where Yl+ (Θ, φ), Yl0 (Θ, φ) and Yl− (Θ, φ) are the generalised spherical harmonics (see
e.g. Phinney & Burridge,
q 1973, their Appendix A and Dahlen q & Tromp, 1997, their
Annex B) and Ωl0 = l(l+1)
2
. The normalisation constant γl = 2l+1
4π
can be derived
by applying the orthogonality relation of spherical harmonics (Phinney & Burridge,
1973). The orthogonal relation, in terms of the scalar radial eigenfunctions, can be
written as (e.g. Dahlen & Tromp, 1997 and Capdeville, 2002):
Z R Z R
ρ(r) nUl (r) + l(l + 1) nVl (r) r2 dr =
2 2
ρ(r)l(l + 1) nWl 2 (r))r2 dr = 1 (2.21)


0 0

To find out the analytical expression of the synthetic seismogram in frequency domain
and canonical coordinates we need to define the source mechanism.

73
Chapter 2. Modeling noise Rayleigh waves

2.4 Synthetic seismogram using a single force

Let us consider a single point force applied at r0s = (rs , θs , φs ) with components
F = (F+ , F0 , F− ) in canonical coordinates:

f (r) = Fδ(r − r0s ) (2.22)

The source term (2.7) can be easily expressed in canonical coordinates by the internal
product between the complex conjugate of the eigenfunction uk and the point force
F:

−∗ − +∗
Sk (rs , θs , φs ) = F0 · u0∗ +
k (rs , θs , φs ) + F · uk (rs , θs , φs ) + F · uk (rs , θs , φs ) (2.23)

and the receiver term by the internal product between the eigenfunction uk and the
receiver vector v:

Rk (rr , θr , φr ) = v0 · u0 k (rr , θr , φr ) − v− · u+ k (rr , θr , φr ) − v+ · u− k (rr , θr , φr ) (2.24)

where both F and v are in canonical coordinates, as for a generic vector in equation
(2.18). By using equation (2.20), the source and the receiver terms of the synthetic
seismogram in canonical coordinates and considering a single force, are:


 −
 S =
√1 γl Ωl [(FΘ + iFΦ ) V + (−FΦ
2 0 n l + iFΘ ) nWl ]Yk− (Θs , φs )
Sk (rs , θs , φs ) = 0 0 (2.25)
S = γl Fr nUl Yk (Θs , φs )

S + = √12 γl Ωl0 [(−FΘ + iFΦ ) nVl + (FΦ + iFΘ ) nWl ]Yk+ (Θs , φs )





 −
 R =
√1 γl Ωl [(vΘ − ivΦ ) V + (−vΦ
2 0 n l − ivΘ ) nWl ]Yl− (Θr , φr )
Rk (rr , θr , φr ) = 0 0 (2.26)
R = γl vr nUl Yl (Θr , φr )

+
R+ = √12 γl Ωl0 [(−vΘ − ivΦ ) nVl + (vΦ − ivΘ ) nWl ]Yl (Θr , φr )



2.4.1 Vertical single force

In Chapter 1 we have seen that a seismic noise source can be described as a

vertical single force. In this simple case, the unique non-null component of the source
is |F| = F 0 = Fr . Then, the three components of a synthetic seismogram in canonical

74
Chapter 2. Modeling noise Rayleigh waves

Figure 2-3: Three-component synthetic seismograms computed by normal mode sum-

mation with a vertical point source. The source amplitude is 1×1017 N. The epicentral
distance is 35◦ . All the three-component seismograms have been bandpass filtered
with corner frequencies of 0.004 Hz and 0.01 Hz. All synthetics here are calculated
using the model PREM (Dziewonski & Anderson, 1981).

coordinates can be easily written from equations (2.25) and (2.26):


(γ Ωl )2


 u− = l 2 0 Fr nUl (rs )[(vΘ − ivΦ ) nVl (rr )+
 
 ω

− exp − 2qk t
0




 +(−vΦ − ivΘ ) nWl (rr )]Yk (Θs , φs )Yl (Θr , φr ) ωk2
k

 ω
exp − 2qk t
u = γl2 vr Fr nUl (rr ) nUl (rs )Yk0 (Θs , φs )Yl0 (Θr , φr )
0 k (2.27)
 ωk2
(γ Ωl )2

u+ = l 2 0 Fr nUl (rs )[(−vΘ − ivΦ ) nVl (rr )+




  
ω
exp − 2qk t

 +
+(vΦ − ivΘ ) nWl (rr )]Yk0 (Θs , φs )Yl (Θr , φr ) k


ωk2

in which we have also considered the effect due to the attenuation in the propagation
term, as in equation (2.14).
The vertical component of the synthetic seismogram – computed by equation

75
Chapter 2. Modeling noise Rayleigh waves

(2.18) and 2.27 – depends only on the radial eigenfunction U computed at the position
of the source and at the position of the receiver:
 
ωk
exp − 2qk
t
ur = γl2 vr Fr nUl (rr ) nUl (rs )Yl0 (Θs , φs )Yl0 (Θr , φr ) (2.28)
ωk2

A vertical single force only generates P and SV-waves , then only Rayleigh waves. The
transverse component of a synthetic seismogram is then null in such case (Kanamori
& Given, 1982).
In Figure 2-3, we present an example for the radial, transverse and vertical compo-
nent of the synthetic seismograms computed with a vertical single force in the PREM
model (Dziewonski & Anderson, 1981). The chosen source amplitude is |F| = 1×1017
N and the epicentral distance is 35◦ . The radial and vertical components are asso-
ciated with Rayleigh waves, whereas the null transverse component shows that Love
waves are not radiated from this source.

2.5 Earth models

In this section, we investigate some normal mode features varying the Earth’s
structure and considering periods related to the secondary microseismic noise, that is
from 4 s to 10 s. A symmetric, non-rotating, perfectly elastic and isotropic (SNREI)
Earth model is considered, then all the elastic properties only vary as a function of
the radial distance r from the centre of the Earth (Dahlen & Tromp, 1997) and they
do not show lateral variations. Moreover, our Earth models do not account for the
bathymetry.
The computation of normal modes has been carried out using the MINOS code.
This code was developed and improved by F. Gilbert, J. Woodhouse and G. Masters
(e.g. Woodhouse, 1988) and it allows to compute normal modes in a spherical Earth
model. Improvements performed by Yann Capdeville concern the introduction of the
addition theorem for spherical harmonics, which has allowed to speed up significantly
the computation.
Because we are interested in short period normal modes for investigating the sec-
ondary microseismic noise, the eigensolutions are particularly sensitive to the shallow
structure of the Earth. As a consequence, the ocean layer and the upper part of the
crust play a particularly relevant role.

76
Chapter 2. Modeling noise Rayleigh waves

Figure 2-4: a) Density ρ, b) compressional-wave speed α and c) shear-wave speed β

as a function of depth in the two-layer model. The model is capped by a 3 km thick
uniform oceanic layer. For depths deeper than the core-mantle boundary (CMB, 2891
km depth), the model consists of the elastic parameters of the Preliminary Reference
Earth Model (PREM). A zoom in the ocean layer is shown in the upper part of the
figures.

In the following, we study the effect of crust layering, the ocean thickness and the
seafloor sediments on normal modes, considering periods from 4 s to 10 s.

2.5.1 Two-layer model

For investigating the effect of the ocean thickness on normal modes related to the
secondary microseismic noise, we consider a simple Earth model characterized by two
layers above the core-mantle boundary (CMB, 2891 km depth). Starting from the
surface of the Earth, the first layer is fluid and it represents the ocean. In this layer,
the P-wave velocity is set to α = 1.4 km/s and the density to ρ = 1000 kg/m3 . Below
the ocean, we consider a very wide crust from the seafloor down to the CMB, having
the P and S-wave velocity fixed at α = 4.85 km/s and β = 2.8 km/s, respectively.
The density of this layer is ρ = 2500 kg/m3 . Below the CMB, we keep the external
and internal core as in the PREM model (see next section). We remind that these
two very deep layers do not have any influence on the computation of normal modes
at short periods. Because of that, we refer to this model as the two-layer Earth
model. Figure 2-4 summarises the main features of the two-layer model as a function
of depth, considering an ocean thickness of 3 km. Figure 2-4a displays the density ρ,
Figure 2-4b the P-wave velocity vP and Figure 2-4c the S-wave velocity vS . A zoom
in the ocean layer is shown in the upper part of the figures.

77
Chapter 2. Modeling noise Rayleigh waves

The short period normal mode energy and amplitude are partitioned between
motions in the crust and motions in water. The ocean thickness then affects the shape
of the short period eigensolutions. Considering this Earth model, in the following,
we study the effects of the ocean thickness on normal modes moving the ocean-crust
interface and considering an ocean depth from 1 km to 10 km.

In Figure 2-5, we present the dispersion diagrams of the first five spheroidal oscil-
lations in the two-layer model, for varying ocean depths. Only periods from T = 4 s
to T = 10 s have been taken into account. The ocean-crust interface has been placed
at 1 km depth (Figure 2-5a), 3 km depth (Figure 2-5b) and 5 km depth (Figure 2-5c).
In all the three cases, the dispersion branches approach one another increasing the
radial order n, meaning that the higher radial order overtones excite the same fre-
quency for a close angular order l. In case of shallow water, e.g. 1 km depth in Figure
2-5a, all dispersion branches are close together, whereas in case of deep water, e.g. 5
km depth in Figure 2-5c, they appear to be more separated. Moreover, increasing the
ocean depth, the same frequency is excited by modes with increasing angular orders
l, especially for the first three branches (n = 0, 1, 2). In particular, the fundamental
mode associated with the branch n = 0, shows the strongest variability with varying
the ocean depths.

As a consequence of these effects on the dispersion diagrams, the shape and am-
plitude of the displacement eigenfunctions are different for varying water thicknesses.
The amplitude of the radial eigenfunction U of the fundamental mode for a fixed
period of T = 6 s is shown in Figure 2-6. The computation has been carried out in
the two-layer model, in which we have changed the ocean thickness from 1 to 10 km,
with discrete steps of 1 km. The colour scale is referred to the ocean depth. The
eigenfunctions appear to be different in shape and amplitude varying the ocean depth
and the effect due to the ocean on the eigenfunctions is also extended to the crust
below. Each eigenfunction is characterised by a kink, which corresponds, in each one
of the ten models, to the seafloor. The displacement eigenfunctions are non-null both
in the ocean and in the crust, with a percentage of non-null displacement in the solid
part which varies with the ocean depth.

Because part of energy is stored in the ocean and part in the solid crust, for all
periods longer than 4 s, we refer to the corresponding propagating seismic waves
as Rayleigh waves. Stoneley modes, which correspond to seismic waves propagating
along the solid-liquid interface, with energy in water and in the crust which decays

78
Chapter 2. Modeling noise Rayleigh waves

Figure 2-5: Dispersion diagram showing the fundamental mode and the first four
overtones of the two-layer model varying the ocean depth. a) 1 km of ocean thickness;
b) 3 km of ocean thickness; c) 5 km of ocean thickness. It has been considered only
for periods from T = 4 s to T = 10 s.

79
Chapter 2. Modeling noise Rayleigh waves

Figure 2-6: Displacement eigenfunctions U as a function of depth for the fundamental

mode at a fixed period of 6 s. The computation has been carried out in the two-layer
model in which we vary the ocean depth. The colour scale is referred to the ocean
thickness, which varies from 1 km to 10 km, with discrete steps of 1 km.

exponentially from the interface, can be found at shorter period (see Appendix A).
The ocean thickness also affects the propagation of Rayleigh waves. Figure 2-7
presents the group velocity U (left column) and the phase velocity c (right column)
as a function of the period T for the fundamental mode and the first four overtones
of Rayleigh waves. The colour scale is referred to the normal mode radial order n.
Figure 2-7a, Figure 2-7b and Figure 2-7c have been computed in the two-layer
model with an ocean thickness of 1 km, 3 km and 5 km, respectively. Fixing the
ocean depth, a strong variability occurs with respect to the period, especially for the
fundamental mode. The highest radial order overtones are nearly constant varying the
period. Moreover, the group velocity and the phase velocity show a strong variability
with the ocean depth. This variability decreases with increasing the radial order n.
The fundamental mode shows the strongest variation with the ocean thickness. The
overtones denoted by n = 3 and n = 4 are characterised by an almost constant group
velocity and phase velocity, with a few rare exceptions (e.g. group velocity in Figure
2-7b and 2-7c at T = 7 s and T = 4 s, respectively).

80
Chapter 2. Modeling noise Rayleigh waves

Figure 2-7: Group velocity (left column) and phase velocity (right column) as a
function of the period T considering a) 1 km of ocean depth; b) 3 km of ocean depth;
c) 5 km of ocean depth. The Earth model is the two-layer model. The colour scale is
related to the radial order n of the first five normal modes.

81
Chapter 2. Modeling noise Rayleigh waves

Figure 2-8: a) Density ρ, b) compressional-wave speed α and c) shear-wave speed β

as a function of depth in the PREM model. The model is capped by a 3 km thick
uniform oceanic layer. A zoom of the ocean layer is shown in the upper part of the
figures.

2.5.2 PREM model

In this section, we deal with a more realistic 1D Earth model, called Preliminary
Reference Earth Model (PREM) (Dziewonski & Anderson, 1981). With respect to
the previous simple Earth model, the PREM is characterised by a large number of
solid-solid interfaces, including the Mohorovicic discontinuity at a depth of 24.4 km
and upper-mantle discontinuities at depths of 220, 400 and 670 km. The PREM
model has a fixed ocean thickness of 3 km.
This model was designed to fit a variety of different data sets. Data from free
oscillations have been combined with surface and body wave travel time data and
basic astronomical data to profile P, S-velocities, density and attenuation as a function
of depth. In order to simultaneously fit Love and Rayleigh wave observations, the
PREM is transversely anisotropic between 80 and 220 km depth in the upper mantle,
meaning that SH and SV-waves travel with different speeds. But this anisotropic
layer is too deep to yield any effect on the eigensolutions for short periods.
The PREM model is one of the most used reference models and all current Earth
models have values that are reasonably close to the PREM. The largest differences
between the PREM and the other Earth models are in the upper mantle, where, for
example, a discontinuity at 220 km is not found in most models (Shearer, 2009). But
this depth is also too deep to influence our eigensolutions. In Figure 2-8, we present
the main elastic properties of the PREM model as a function of depth. Figure 2-8a

82
Chapter 2. Modeling noise Rayleigh waves

shows the density ρ, Figure 2-8b the compressional P-wave velocity vP and Figure
2-8c the shear S-wave velocity vS .
In Figure 2-9, we present the dispersion diagrams showing the spheroidal oscilla-
tions of the PREM model, for various ocean depths. The colour scale is related to
the radial order of the first five normal modes. Only periods from 4 s to 10 s have
been taken into account. The ocean-crust interface has been placed at 1 km depth
(Figure 2-9a), 3 km depth (Figure 2-9b) and 5 km depth (Figure 2-9c). The 3 km
depth model is the standard PREM model (Dziewonski & Anderson, 1981).
In the following, we aim to present the eigensolutions in the PREM model and to
compare with the previous two-layer model. We consider ocean depths from 1 to 10
km with steps of 1 km. All the other discontinuities (located below the ocean-crust
discontinuity) have been moved for maintaining the same effective thickness of the
layers. Making a comparison with the results obtained with a two-layer model, we
will demonstrate that both the ocean and the first layers of the crust yield an effect
on the eigensolutions.
In the PREM model, as well as in the two-layer model (Figure 2-5), the dispersion
branches approach one another increasing the radial order n, meaning that normal
modes denoted by different radial order n excite the same frequency for a close angular
order l. An exception is observed for deep water (Figure 2-9c) at high frequencies.
The dispersion branches in the PREM model appear to be more separated than in
the two-layer model, especially for the fundamental mode and the first two overtones.
In the shallow water case, we observe the strongest differences between the two models
(Figure 2-5a versus Figure 2-9a). The same frequency and the same radial order n
does not correspond to the same angular order l in the PREM and in the two-layer
model. As in the two-layer model, increasing the ocean depth, the same frequency
is excited by modes with increasing angular orders l, especially for the first three
branches (n = 0, 1, 2). In particular, the fundamental mode associated with the
branch n = 0 shows the strongest variations with varying the ocean depth.
In Figure 2-10, we present the amplitude of the radial eigenfunction U as a function
of depth in the PREM model, for varying ocean depth, the period and the radial order
n. The eigenfunctions of the first three normal modes are presented in Figure 2-10a,
2-10b and 2-10c, respectively. We compute the eigenfunctions for two fixed periods,
T = 6 s (left column) and T = 10 s (right column). The colour scale is referred to
the ocean thickness, which varies from 1 km to 10 km, with discrete steps of 1 km.

83
Chapter 2. Modeling noise Rayleigh waves

Figure 2-9: Dispersion diagram showing the fundamental mode and the first four
overtones of the PREM model varying the ocean depth. a) 1 km of ocean thickness;
b) 3 km of ocean thickness; c) 5 km of ocean thickness. It has been considered only
for periods from T = 4 s to T = 10 s.

84
Chapter 2. Modeling noise Rayleigh waves

Fixing the radial order n (same row), the eigenfunctions have a different amplitude
varying the ocean depth and the period. Fixing the period and the ocean depth (same
column), they also have a different shape varying the radial order n. The amplitude
of the eigenfunctions decreases with increasing the radial order n, especially for long
period (right column). Like in the two-layer model, the eigenfunctions present a kink,
in each one of the ten models, at the seafloor. The kink amplitude decreases with
increasing the overtone number, becoming almost negligible for n = 2 (third row).
The displacement eigenfunctions are non-null both in the ocean and in the crust,
with an ocean depth-dependent amount of non-null displacement also in the solid
part. Because of that, for all periods longer than 4 s, we refer to the corresponding
propagating seismic waves as Rayleigh waves (see Appendix A for Stoneley modes).
As we have already shown for the two-layer model, the ocean thickness also affects
the propagation of Rayleigh waves, i.e. their group and phase velocity. But, doing
the computation in the PREM model, we observe another effect due the the presence
of a solid multi-layer below the ocean.
Figure 2-11 displays the group velocity U (left column) and the phase velocity c
(right column) as a function of the period T for the fundamental mode and the first
four overtones of Rayleigh waves. The colour scale is referred to the radial order n.
The computation has been performed in the PREM model for varying ocean depth.
We present the cases with an ocean thickness of 1 km (Figure 2-11a), 3 km (Figure
2-11b) and 5 km (Figure 2-11c). A strong variability can be observed both varying the
ocean depth and the period. Fixing the ocean depth (same row), a strong variability
occurs with respect to the period. Increasing the ocean depth (same column), both
the group and the phase velocity decrease.
The group and the phase velocity of the overtones in the PREM model present
stronger variations with periods than in the two-layer model (Figure 2-11 versus
Figure 2-7).

2.5.3 Sediments at the ocean seafloor

It is a matter of fact that, below the seafloor, a layer of sediments lies between the
ocean and the crust. Figure 2-12 display the map of the sediment thickness for the
world’s oceans compiled by the National Geophysical Data Center (NGDC), Marine
Geology & Geophysics Division in 2003. Sediment thickness data were compiled from
three principle sources: previous collected data, ocean drilling results and seismic

85
Chapter 2. Modeling noise Rayleigh waves

Figure 2-10: Displacement eigenfunctions U as a function of depth for the fundamental

mode (a), the first (b) and the second (c) overtone, at a fixed period of 6 s (left column)
and 10 s (right column). The computation has been carried out in the PREM model
in which we vary the ocean depth. The colour scale is referred to the ocean thickness,
which varies from 1 km to 10 km, with discrete steps of 1 km.

86
Chapter 2. Modeling noise Rayleigh waves

Figure 2-11: Group velocity (left column) and phase velocity (right column) computed
in the PREM model considering a) 1 km of ocean depth; b) 3 km of ocean depth; c)
5 km of ocean depth. The colour scale is related to the radial order n of the first five
normal modes.

87
Chapter 2. Modeling noise Rayleigh waves

Figure 2-12: Map of sediment thickness for the world’s oceans compiled by the Na-
tional Geophysical Data Center (NGDC), Marine Geology & Geophysics Division
(http://www.ngdc.noaa.gov/mgg/sedthick/sedthick.html ) (Divins, 2003).

reflection profiles. With the exception of the coastal areas, in which the sediment
thickness varies from 1 km up to 20 km (in case of passive margins), the ocean basins
are characterised by a maximum sediment thickness around 500 m.
In this section, we aim to investigate the effect of sediments on the eigensolutions.
We consider the PREM model and we build a sediment layer between the ocean and
the crust. All the solid-solid discontinuities of the PREM model are shifted until
the CMB to keep the same thickness for each layer. Typical elastic properties of
sediments have been chosen for the sediment layer: P-wave velocity vP = 2 km/s,
S-wave velocity vS = 0.6 km/s and density ρ = 1700 kg/m3 .
In Figure 2-13 we present the radial eigenfunction U computed in this model (red
curves). The period is fixed at T = 10 s and the ocean depth at a) 1 km and b) 3
km. In grey, it is shown the eigenfunction computed in the PREM model with the
same ocean thickness and without the sediment layer.
The eigenfunctions computed in the Earth model with the sediment layer show a
second kink a the base of the sediment layer, in addition to the one at the seafloor.
In both cases, the eigenfunctions computed with and without the sediment layer are

88
Chapter 2. Modeling noise Rayleigh waves

Figure 2-13: Displacement eigenfunctions U as a function of depth for the fundamental

mode of Rayleigh waves. The computation has been carried out in the PREM model
in which has been added a layer of sediments between the ocean and the crust (red
eigenfunctions). The period is fixed at T = 10 s and the ocean depth at a) 1 km
depth and b) 3 km depth. In grey, it is shown the eigenfunction computed in the
PREM model with the same ocean thickness without the sediment layer.

nearly the same at the surface of the ocean. A small difference is present at the ocean
surface for shallow water (Figure 2-13a). Immediately below the ocean surface, the
eigenfunctions computed with and without the sediment layer diverge and they show
quite different values at the seafloor. We find the strongest differences at the seafloor
for deep water (Figure 2-13b).The sediment layer also affects the eigenfunction shape
and amplitude in the crust.

Figure 2-14 shows a) the dispersion diagram, b) the group velocity and c) the
phase velocity of the fundamental and the first overtone computed in the PREM
model (grey curves) and in the Earth model in which we have added the sediment
layer (red curves). From Figure 2-14a we observe that, to get the same frequency,
it is necessary to compute more harmonics in the model with sediments than in the
model without sediments, both for the fundamental and the first overtone. The group
velocity (Figure 2-14b) and the phase velocity (Figure 2-14c) are smaller in the model
with sediments, than in the PREM model, with a few rare exceptions.

89
Chapter 2. Modeling noise Rayleigh waves

Figure 2-14: a) Dispersion diagram, b) group velocity and c) phase velocity computed
considering the PREM model with the sediment layer (red curves) and without the
sediment layer (grey curves) for the fundamental mode and the first overtone.

2.6 Site effect due to the ocean depth

In the previous sections, the effect of the ocean thickness on the shape and ampli-
tude of the eigenfunctions has been presented in the secondary microseismic period
band, that is from 4 s to 10 s. In the same section, we have also shown that the
eigensolutions are influenced by the solid structure beneath the ocean and that the
eigenfunction amplitudes and shapes vary with the Earth model. In this section, we
aim to show these effects on the computation of the synthetic seismograms.
If we consider a source as single force, the computation of a synthetic seismogram
depends on the product between the radial eigenfunction U at the position of the
receiver and the radial eigenfunction U at the position of the source, as shown in

90
Chapter 2. Modeling noise Rayleigh waves

section 2.4.1, equation (2.28). Considering the same Earth model for source, receiver
and propagation, the receiver has to be located on the solid ground, that is at the
seafloor. We consider the source at the surface of the ocean, with the purpose of
simulating a noise source, as explained in Chapter 1.
From equation (2.28), we define the quantity:

n Ul (rs ) n Ul (rr )
cn = (2.29)
n ωl

which represents the contributions due to the source and the receiver on the synthetic
seismogram computation. We call cn the site effect on the n−th normal mode of
Rayleigh waves. Because the eigenfunctions U (rs ) and U (rr ) vary with both the ocean
depth and the frequency, the most natural way to represent these double-dependence
is to plot cn as a function of the product between frequency and ocean thickness. We
point out that this is also the natural way to find out the nodes of a standing wave
system.
In Figure 2-15, the coefficient cn has been plotted as a function of the dimensionless
quantity ωh/β, where ω = n ωl is the eigenfrequency, h is the ocean depth and β is
the S-wave velocity in the crust. The computation of normal modes has been carried
out in the three previously described Earth models: a) the two-layer model, b) the
PREM model and c) the PREM model with a layer of sediments between ocean and
crust. The colour scale is related to the ocean depth. The site effect on noise Rayleigh
waves was firstly computed analytically by Longuet-Higgins (Longuet-Higgins, 1950)
in a half space model (see Chapter 1). Because short period eigenfunctions are only
affected by shallow depths, Longuet-Higgins’ half space is equivalent to the two-layer
model.
The fundamental mode of Rayleigh waves – denoted by n = 0 – presents a maxi-
mum around ωh/β ' 0.9 before falling away to zero. This peak is present in all the
three Earth models, with small abscissa variations, especially in the presence of sed-
iments (Figure 2-15c). The peak due to the fundamental mode of Rayleigh waves is
related to water depth up to 4 km, whereas deeper water gives a smaller contribution.
The maximum amplitude of the site effect in the two-layer model (Figure 2-15a) is
comparable with the maximum amplitude in the PREM model (Figure 2-15b). In
the presence of sediments, it increases reaching double values for very shallow water
(around 1 km depth). The overtones display two relative maxima associated with
different ocean depths.

91
Chapter 2. Modeling noise Rayleigh waves

Figure 2-15: Site effect coefficient – as expressed in equation 2.29 – as a function of

the dimensionless quantity ωh/β, where ω is the eigenfrequency, h is the ocean depth
and β is the S-wave velocity in the crust. The computation of normal modes has
been carried out in a) the two-layer model, b) the PREM model and c) the PREM
model in which we have added a layer of sediments between ocean and crust. The
source and the receiver positions are at the surface of the ocean and at the seafloor,
respectively. Wave periods from 4 s up to 10 s have been considered. The colour scale
is related to the ocean depth.
92
Chapter 2. Modeling noise Rayleigh waves

Figure 2-16: Site effect on the fundamental mode of Rayleigh waves (equation 2.29)
for periods from 3 s to 10 s. The computation of the eigenfunctions has been carried
out in the PREM model.

93
Chapter 2. Modeling noise Rayleigh waves

In Figure2-15b, the presence of a multi-layer below the ocean in the PREM model
produces a splitting of the different ocean depth contribution with respect to the
two-layer model in Figure 2-15a. Both the position and the amplitude of the maxima
are comparable, with very few differences, in the two-layer model and in the PREM
model.
In Figure 2-15c, the sediment layer has a significant effect only on the fundamental
mode of Rayleigh waves in shallow water (1 and 2 km). A wider sediment thickness
should be considered for shallow coastal areas (see Figure 2-12). In such case, the
effect due to the sediments should be even bigger (e.g. Webb, 1992).
Fixing the frequency, we can build maps of the site effect. In Figure 2-16, we
present the site effect on the fundamental mode of Rayleigh waves computed in the
PREM model for periods from 3 s to 10 s. The maximum site effect is confined in
coastal areas at short periods (T ≤ 5s), whereas it is spread in wide regions at long
periods (T > 5 s).
Further discussions about the site effect on noise Rayleigh waves, both in the two-
layer model and in the PREM model, have been presented by Gualtieri et al. (2013)
(see section 2.9). In this study, it has also been discussed about the site effect when
the receiver is located on land and at the seafloor (see also Tanimoto, 2013).

2.7 Effect of the ocean depth on Rayleigh wave

propagation
In addition to the effect due to the source and receiver position, the ocean thickness
affects the propagation of Rayleigh waves. In Figure 2-17, the group velocity is shown
versus the dimensionless quantity ωh/β, where ω is the eigenfrequency, h is the ocean
depth and β is the S-wave velocity in the crust.
The computation has been carried out in the two-layer model (Figure 2-17a),
the PREM model (Figure 2-17b) and the PREM model in which has been added a
sediment layer (Figure 2-17c). In each one of the Earth models, the ocean thickness
has been changed from 1 km to 10 km (colour scale). For each one of the three models
and for each one of the ten ocean depths, the computation has been performed for
all periods between 4 s and 10 s.
In the two-layer model, the contributions due to the different ocean depths to the
same normal mode group velocity (referred as n = 0, n = 1 and so on) are aligned

94
Chapter 2. Modeling noise Rayleigh waves

and we distinguish the presence of four separate branches. The group velocity of the
fundamental mode of Rayleigh wave (n = 0) shows a minimum around ωh/β ' 0.9,
the first overtone two minima around ωh/β ' 1 and ωh/β ' 2.8 and so on for the
higher overtones.
Making a comparison between the group velocity in the two-layer model (Figure
2-17a) and in the PREM model (Figure 2-17b), we can evaluate the influence of the
crust on the propagation of Rayleigh waves. In the PREM model, the presence of
multiple solid layers below the ocean adds a minimum to each branch. For example,
the group velocity of the fundamental mode of Rayleigh waves (denoted by n = 0)
presents an additional minimum at ωh/β ' 0.2. The contributions due to the different
ocean thickness appear to be separated. In terms of amplitude, the group velocity
in the PREM model shows some differences with respect to the two-layer model,
especially concerning the higher modes. The amplitude of the group velocity of the
fundamental mode shows very few differences between the two-layer model and the
PREM model.
Adding a sediment layer to the PREM model and varying the ocean depth (Figure
2-17c), multiple effects coexist. They are due to the ocean thickness, the multi-layer
beneath the ocean and the sediment layer. We observe that new minima are added
to the plot and the normal mode branches approach one another. The amplitude of
the group velocity in the presence of the sediment layer, especially for the overtones,
tends to be smaller with respect to the one computed in the PREM model.
Wider sediment layers should be considered, especially for coastal areas (Figure
2-12). In these regions, the layer of sediments beneath the seafloor can be very thick,
varying from 2 km up to 20 km, in passive margin areas. Both the site effect and
the propagation effect on Rayleigh waves can be significantly affected by these thick
layers, producing an amplification which should be taken into account when modeling
seismic noise (e.g. Webb, 1992).

2.8 Normal Mode Summation code

The code which has been used to perform the normal mode summation (NMS)
and compute the synthetic seismograms was originally written by Yann Capdeville.
It allows to use both a moment tensor and a single force. This code, originally
developed for computing synthetic seismograms of earthquakes, has been adapted to

95
Chapter 2. Modeling noise Rayleigh waves

Figure 2-17: Group velocity as a function of the dimensionless quantity ωh/β, where
ω is the eigenfrequency, h is the ocean depth and β is the S-wave velocity in the crust.
The computation has been carried out in a) the two-layer model (see section 2.5.1),
b) the PREM model (see section 2.5.2) and c) the PREM model in which has been
added a sediment layer beneath the seafloor (see section 2.5.3) for periods between
T = 4 s and T = 10 s. The colour scale is referred to the ocean thickness.
96
Chapter 2. Modeling noise Rayleigh waves

noise calculations, as explained in the following.

In this code, when an oceanic Earth model is considered, the receiver position is
set at the seafloor by default. In such way the source term, the receiver term and the
propagation term in equation (2.14) are computed considering the same Earth model,
but in case of a continental station an approximation occurs. We have tested the effect
of this approximation by computing the site effect as defined by the equation (2.29),
in two cases: considering the same oceanic model for the source and the receiver and
considering an oceanic model for the source and a continental one for the receiver.
The difference between these two cases has been shown by Gualtieri et al. (2013),
Figure 4. Our analysis has brought to the conclusion that this difference is small – it
is less than a factor of two – and that it does not strongly affect the computation of
the synthetic seismogram amplitude. Moreover, assuming different Earth models for
the source and the receiver, we might not associate the same angular order l to the
same eigenfrequency ωl , for both the receiver and the source term in equation (2.14).
An approximation, probably even bigger than the receiver at the seafloor, would be
committed in such case.
A future improvement of this aspect may be the computation of seismic noise
synthetic seismograms using local modes and considering different Earth models for
source and receiver.
With the purpose of computing seismic noise spectra in the period band T = 4−10
s, we have added to the normal mode summation code the following aspects:

• a global scale distribution of seismic noise sources at the surface of the ocean,
with a resolution of half a degree in latitude and longitude. The source location
and amplitude have been retrieved from the ocean wave model WAVEWATCH
IIIR (see chapter 1, section 1.4.2);

• a random phase for each noise source;

• a realistic bathymetry for each one of the grid points, reading the normal mode
catalogue computed in the PREM model with a modified bathymetry;

• an attenuation model for short period (see next section);

• the MPI protocol (Message Passing Interface) to speed up the computation and
allow to read different normal modes catalogues.

97
Chapter 2. Modeling noise Rayleigh waves

2.8.1 Attenuation model

All global scale attenuation models are valid for long periods, because they are
derived from long-period Rayleigh and Love wave data. There is no available global
scale attenuation model for short periods.
In our modeling, we consider as reference attenuation model QL6 (Durek & Ek-
ström, 1996), valid for periods 150s ≤ T ≤ 300s, and we empirically adjust the values
of Qµ for the first 100 km to fit the spectra. The calibration is performed for one
station – the Geoscope station SSB, France – and the empirical apparent attenuation
model is then applied to all the other stations (Gualtieri et al., 2013). Our attenua-
tion model values are consistent with those obtained by Stutzmann et al. (2012) and
by Prieto et al. (2011) at the local scale.
In section 2.5, we have seen that the non-null eigenfunction displacement produced
in the solid crust depends on the frequency and the ocean depth and that it is smaller
than 100 km for periods from 4 s to 10 s. A future improvement of this aspect may
be the evaluation of the thickness that has to be changed in the attenuation model to
maintain the same spectra amplitude. An empirical short period attenuation model
could be calibrated in such way.

98
Chapter 2. Modeling noise Rayleigh waves

2.9 Modeling secondary microseismic noise by nor-

mal mode summation

In the following, we attach the paper Gualtieri et al. (2013) in order to show the
main results in terms of noise modeling.

First, this study presents the description of the theory which allows us to compute
noise sources from the ocean wave model. Then, it presents the computation of the
site effect in the two-layer model and in the PREM model. The site effect has been also
evaluated considering a continental model for the receiver. The computation of noise
spectra has been performed both for the vertical and the horizontal components. The
discrepancy between real and modelled spectra in the horizontal components reveals
that our modeling does not account for the entire energy wavefield. We attribute this
missing energy content to Love waves which can not be generated by vertical forces.
Finally, we show that the seasonal variations of noise sources are frequency-dependent
and we display the contribution of the different oceans on noise spectra for a station.

99
Chapter 2. Modeling noise Rayleigh waves

Geophysical Journal International

Geophys. J. Int. (2013) doi: 10.1093/gji/ggt090
Geophysical Journal International Advance Access published April 4, 2013

Modelling secondary microseismic noise by normal mode summation

L. Gualtieri,1,2 E. Stutzmann,1 Y. Capdeville,3 F. Ardhuin,4 M. Schimmel,5 A. Mangeney6

and A. Morelli7
1 Institut de Physique du Globe de Paris (IPGP), UMR 7154 CNRS, F-75005, Paris, France. E-mail: [email protected]
2 Dipartimento di Fisica e Astronomia, Settore di Geofisica, Università di Bologna, Bologna, Italy
3 CNRS, Laboratoire de Planètologie et Gèodynamique de Nantes, Nantes, France
4 Laboratoire d’Océanographie Spatiale, Ifremer, Plouzanè, Brest, France
5 Institute of Earth Sciences Jaume Almera, CSIC, Barcelona, Spain
6 Institut de Physique du Globe de Paris, Sorbonne Paris Cité, Univ Paris Diderot, UMR 7154 CNRS, F-75005, Paris, France
7 Istituto Nazionale di Geofisica e Vulcanologia (INGV), Bologna, Italy

Accepted 2013 March 4. Received 2013 March 1; in original form 2012 December 21

SUMMARY

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

Secondary microseisms recorded by seismic stations are generated in the ocean by the interac-
tion of ocean gravity waves. We present here the theory for modelling secondary microseismic
noise by normal mode summation. We show that the noise sources can be modelled by vertical
forces and how to derive them from a realistic ocean wave model. We then show how to
compute bathymetry excitation effect in a realistic earth model by using normal modes and
a comparison with Longuet–Higgins approach. The strongest excitation areas in the oceans
depends on the bathymetry and period and are different for each seismic mode. Seismic noise
is then modelled by normal mode summation considering varying bathymetry. We derive an
attenuation model that enables to fit well the vertical component spectra whatever the station
location. We show that the fundamental mode of Rayleigh waves is the dominant signal in
seismic noise. There is a discrepancy between real and synthetic spectra on the horizontal
components that enables to estimate the amount of Love waves for which a different source
mechanism is needed. Finally, we investigate noise generated in all the oceans around Africa

GJI Seismology
and show that most of noise recorded in Algeria (TAM station) is generated in the Northern
Atlantic and that there is a seasonal variability of the contribution of each ocean and sea.
Key words: Surface waves and free oscillations; Seismic attenuation; Theoretical seismol-
ogy; Wave propagation.

waves meet each other, at first-order approximation, the resulting

1 I N T RO D U C T I O N
displacement decays exponentially with depth. Longuet-Higgins
Microseisms are the continuous oscillation of the ground recorded (1950) showed that the microseisms are generated by pressure fluc-
everywhere in the world independently from earthquake activity in tuations which can be computed considering the second-order term
the period band between 4 and 20 s (e.g. Webb 1998; Stutzmann of the ocean wave interaction. Secondary microseisms are domi-
et al. 2000; Berger et al. 2004). This seismic background noise nantly Rayleigh waves. Longuet–Higgins computed the excitation
results from the non-linear interaction between the atmosphere, the of Rayleigh waves at the ocean bottom as a function of frequency,
ocean and the solid Earth. The source is the atmospheric wind which bathymetry and S-wave velocity in the crust. Hasselmann (1963)
forces the ocean gravity waves generation (Miche 1944; Longuet- extended the theory to a random ocean wavefield and showed that
Higgins 1950; Hasselmann 1963). Seismic noise spectra display two the matching between ocean waves and seismic waves takes place
main peaks at 14 and 7 s, which denote what are called respectively only when the interaction occurs between two ocean waves with
primary and secondary microseisms. The primary microseismic nearly opposite directions and nearly similar frequencies.
peak is the smaller amplitude hump with periods between 10 and The sea state that generates seismic noise can be classified in
20 s. It is generated near the coast when ocean waves reach shal- three classes (Longuet-Higgins 1950; Ardhuin et al. 2011, 2012).
low water and interact with shallow seafloor (Hasselmann 1963). The first class occurs when a storm has a wide angular distribution,
Both Rayleigh and Love waves are present in significant quantities with ocean waves coming from many different azimuths. The first
(Nishida et al. 2008; Fukao et al. 2010). The secondary microseis- class dominates at frequencies from 0.5 to 2 Hz due to the wide
mic peak is the strongest noise peak with periods between 4 and 10 s. angular distribution of the short waves generated by a constant
It is generated by the interaction of ocean gravity waves having sim- and steady wind, and it may still be significant at somewhat lower
ilar periods and travelling in opposite directions. When two ocean frequencies. In that case, the interacting waves are within the storm.


C 100
2013 The Authors 2013. Published by Oxford University Press on behalf of The Royal Astronomical Society 1
Chapter 2. Modeling noise Rayleigh waves

2 L. Gualtieri et al.

In the second class, ocean waves arrive at the coast, they are reflected pressure at the surface of the ocean can be written (Ardhuin et al.
and they meet up with incident ocean waves. Then, the interaction 2011; Stutzmann et al. 2012),
area is confined close to the coast. The third class concerns the
interaction of ocean waves coming from different storms. Ocean Fp (K  0, f 2 = 2 f ) = ρw2 g 2 f 2 E 2 ( f )
waves from a given storm may travel long distances before meeting  π
ocean waves generated by another storm. This third class generates × M( f, θ )M( f, θ + π )dθ (1)
0
the strongest noise sources and these can be anywhere in the ocean
basin. Obrebski et al. (2012) identified such a source located midway where ρ w is the water density, g is the gravity acceleration, f2 is the
between Hawaii and California, and recorded by stations several pressure field frequency and K is the sum of the wavenumbers of
thousands of kilometres away. the two opposite ocean waves. E( f ) is the surface elevation variance
Secondary microseismic sources have been observed near the of two wave trains and M( f, θ) is the non-dimensional ocean wave
coast (Bromirski & Duennebier 2002; Schulte-Pelkum et al. 2004; energy distribution as a function of ocean wave frequency f and
2
Gerstoft & Tanimoto 2007; Yang & Ritzwoller 2008), in the middle azimuth θ . The unit of the surface elevation variance E( f ) is mH z ,
2
of the ocean (Cessaro 1994; Stelhy et al. 2006; Obrebski et al. whereas the spectral density Fp is in mN2 ·Hz . The integral in eq. (1)
2012) and in both cases (Haubrich & Mc Camy 1969; Friedrich depends only on the azimuthal distribution of the ocean wave energy.
et al. 1998; Chevrot et al. 2007). Kedar et al. (2008) showed the To simplify the notation we call this non-dimensional integral I( f ).
first quantitative modelling of seismic noise using ocean wave model For seismic wavenumbers K = (K x , K y ) of magnitude much
hindcasts. They successfully modelled seismic noise generated in smaller than a typical wavenumber k of the ocean wave, the pres-
deep ocean—North Atlantic ocean—without taking into account sure power spectrum Fp (K, f 2 ) is approximately independent of K.
ocean wave coastal reflections. Ardhuin et al. (2011) introduced Thus, in the limit K  k, the surface pressure field for a frequency
the coastal reflection in the wave model. They showed that seismic interval df2 and over an area dA = R2 sin (Φ  )dλ dφ  —where R is the

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

noise spectra can be modelled with great accuracy and presented radius of the Earth, φ  is the colatitude and λ is the longitude—has
the first global maps of noise sources. Stutzmann et al. (2012) the same power spectrum as the one caused by a single localized
further showed the seasonal variations of noise source location. force with a root mean square value
The coastal reflection coefficient is not well constrained and should 
be adjusted as a function of the coast shape (Ardhuin & Roland Frms ( f 2 , d A, d f 2 ) = 2π Fp (K  0, f 2 )d Ad f 2 , (2)
2012). Stutzmann et al. (2012) provided empirical coastal reflection
as given by Hasselmann (1963, after eq. 1.17). This equality of the
coefficients for stations in various environments and showed, in
two spectra, the one given by eq. (1) and the one associated to a
agreement with Longuet–Higgins 50 yr earlier, that the strongest
point force in the middle of a square having length side L and area
noise sources are in deep ocean and that coastal reflection generates
dA = L2 can be seen by taking a finite pressure distribution over a
numerous smaller sources. The theory was extended to body wave
small square of area 4a2 , that is, p(x, y) = p0 for |x| < a and |y| <
solution by Ardhuin & Herbers (2013).
a and p(x, t) = 0 otherwise. If the small square has length side a <
Previous modelling of seismic noise (Kedar et al. 2008; Webb
L/2, the single-sided power spectrum evaluated over the full square
1992; Ardhuin et al. 2011; Stutzmann et al. 2012) used the Longuet–
is
Higgins excitation coefficients which correspond to a flat two layers
2 2 2
medium at the source. Here, we model seismic noise by normal sin2 (K x a) sin2 (K y a) 4a 2
mode summation in a more realistic spherical earth model and we Fp,a,L (K, f 2 ) = 2 p02 , (3)
Kx a Kya 2π L
show that we can reproduce the main features of noise spectra
by modelling the sources as vertical single forces and taking into with a value 2 p02 [4a 2 /(2π L)]2 for K  0. Thus, the same spectral
account the bathymetry. The source excitation due to the bathymetry density as eq. (1) at K  0 is obtained by setting p0 = 2π L/(4a2 ).
in our realistic model is compared to Longuet–Higgins’ results. Since the force is 4a2 p0 , it is independent of a and we find eq. (2),
Attenuation is not well known in the period band of the secondary which only applies to finite values of K in the limit when a goes
microseisms. We present an apparent attenuation model that enables to zero, which corresponds to a point force. The same is true for
us to compute synthetic noise spectra which fit the real spectra with any spatial distribution, for example, taking a Gaussian pressure
high accuracy in the period band of 4–10 s. We then investigate the distribution instead of a constant.
effect of the fundamental mode and higher modes on the vertical Similarly in the time dimension, the random wave-induced pres-
component of noise spectra. We observe a discrepancy between real sure is equivalent to a temporal variation of the force F that will have
and modelled spectra of the horizontal components which can be the same frequency power spectrum as given by eq. (1). In practice,
explained by the existence of Love waves which cannot be generated our model is formulated in time domain. We obtain a time-series
by vertical forces. Finally, we show that the seasonal variations of of the force that has the required power spectrum by summing dif-
noise sources depend on the period and we show the contribution ferent frequencies for which we specify an amplitude and a phase.
of the different oceans on noise spectra for a station. The phase is drawn randomly between 0 and 2π and the amplitude
is normalized to be
√
2 T H E O RY F( f 2 ) = 2Frms ( f 2 , d A, d f 2 ), (4)
2
so that the variance of the force is indeed Frms ( f 2 , d A, d f 2 ), with
2.1 Deriving force field from the ocean wave model
a power spectrum that will thus equal the wave-induced pressure
Seismic noise is generated by the interaction of ocean waves of sim- power spectrum given by eq. (1).
ilar frequencies and nearly opposite directions. Hasselmann (1963) To compute the force amplitude we need the spectral density
showed how to compute the corresponding pressure field. We use of the equivalent pressure (eq. 1) just below the ocean surface
it to derive the analytical expression of the equivalent vertical force everywhere in the ocean. We use the ocean wave model of Ardhuin
field at the ocean surface. The spectral density of the equivalent et al. (2011). The global ocean wave model has a constant resolution

101
Chapter 2. Modeling noise Rayleigh waves

Modelling seismic noise by normal mode summation 3

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

Figure 1. Spherical grid of noise point sources. All the ocean is discretized with a step of 50 km. This represents a good compromise between solution
convergence and calculation time.

of 0.5◦ both in latitude and in longitude. At each gridpoint, the where r is the radial coordinate, uk (r) is a normalized eigenfunction
ocean state is described by 24 azimuths and 16 frequencies spaced of the earth model and ak (t) is the excitation function of mode k. k
exponentially between 0.04 Hz (T = 24.4 s) and 0.17 Hz (T = encapsulates the notations (q, n, l, m), where q can only take two
5.8 s). One key point of this model is that it is the only model to values, one for spheroidal and the other one for toroidal modes,
date which takes into account coastal reflection of ocean waves. We n is the radial order, l is the angular order and m the azimuthal
compute the spectral density with and without coastal reflections one. Because of the spherical symmetry of the reference model,
for performing a linear combination of the two resulting models eigenfunctions do not depend on the azimuthal order m. We consider
to obtain the equivalent pressure maps corresponding to a given an instantaneous point force f̂(r, t) = f(r)g(t), with
coastal reflection coefficient. We use in our modelling the empirical
g(t) = δ(t) (6)
reflection coefficients determined by Stutzmann et al. (2012) which
are different for different regions. and
For a given area, we convert the equivalent pressure maps into ver-
f(r) = Fδ(r − r0s ), (7)
tical forces located just below the ocean surface. The discretization
in point sources corresponds to dividing a storm area into squares where r0s is the source position. In this case, eq. (5) can be written
and considering a force concentrated in a point at the centre of each as
square. All the oceans are discretized on a grid with a step of 50 km,   t sin ωk t 
as shown in Fig. 1. We choose this grid step as a good compromise s(x, t) = dt u∗k (r ) · f(r) dV, (8)
−∞ ωk VE
between solution accuracy and calculation time. We tested that a k

thinner step does not produce any further constructive interference, where u∗k (r ) is the complex conjugate of uk (r ).
meaning that the solution has already reached the convergence. At The Green’s functions are obtained from the previous equation
each gridpoint, the source corresponds to a vertical force Frms (f2 , setting F = 1. Then, during the normal mode summation we com-
dA, df2 ) with a random phase for each frequency step. These sources pute the spatial convolution with the force f(r). We show the analyt-
are then used for computing synthetic seismograms by normal mode ical expression of the synthetic seismogram computation consider-
summation. ing a single forces in Appendix A. The formulation in the appendix
can be used to compute synthetic seismograms for forces in any
2.2 Normal mode computation direction. Here we only consider vertical forces.
In practice, summation over k cannot be computed numerically
Following Gilbert (1970) and Gilbert & Dziewonski (1975) we cal- without truncations. Synthetic seismograms are computed only up
culate the impulse response of a point source and write the seismic to a given frequency, which enables to define a maximum angular
displacement in a spherically symmetric, non-rotating, elastic and order lmax up to which the sum over l has to be calculated. In that
isotropic (SNREI) earth model as sum of normal modes case, the temporal part of eq. (8) is rewritten for allowing sum

s(x, t) = ak (t)uk (r), (5) truncation as in Capdeville (2005). The total displacement is the
k sum of the synthetic displacement generated by each point source.

102
Chapter 2. Modeling noise Rayleigh waves

4 L. Gualtieri et al.

The temporal part of each source (eq. 6) is a random signal having

variance equal to 1.
The 3-D heterogeneities in the Earth generate focusing and de-
focusing effects of seismic waves. These 3-D effects cannot be ac-
counted for with the normal mode summation method used on this
paper. Although these effects have an important influence on wave-
forms (e.g. Tape et al. 2009, 2010; Fichtner et al. 2012), considering
seismic energy (in dB) with a rough accuracy as it is done here, we
assume these effects to be small and the error made, because of this
approximation, does not affect our conclusions.

3 N O I S E S O U R C E S E X C I TAT I O N
Longuet-Higgins (1950, 1952) first showed that microseismic en-
ergy depends on the ocean wave state and on the bathymetry.
Bathymetry produces an excitation effect which is frequency de-
pendent. The excitation factor has been calculated analytically by
Longuet-Higgins (1950) considering a flat two layers medium as a
function of
x = ωh/β, (9)

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

where ω = 2π f2 is the angular frequency of the ocean pressure field
at the source, h is the ocean depth and β is the S-wave velocity.
This particular combination of frequency and depth in eq. (9) is a
common way for describing a standing wave system and looking for
its nodes. Longuet–Higgins coefficients have been used to analyse
the effect of the bathymetry in all previous studies (e.g. Kedar et al.
2008; Ardhuin et al. 2011; Hillers et al. 2012; Stutzmann et al.
2012). Here we compute the effect of bathymetry using normal
modes and a more realistic spherically symmetric earth model with
varying bathymetry. Comparing Longuet–Higgins’ eq. (183) (p. 28,
Figure 2. Excitation coefficients due to the bathymetry as computed through
Longuet-Higgins 1950) and the synthetic seismogram computation
our normal mode approach as function of ωh/β, where ω is the angular
of Appendix A in case of a vertical force, we obtain the expression frequency, h is the ocean depth and β = 2800 m s−1 is S-wave velocity in
of the excitation coefficient computed by normal modes, the crust. (a) Results obtained for an ocean layer over an half-space, as used
n Ul (rr )n Ul (rs ) in Longuet-Higgins (1950). The vertical point force is located at the top of
cn = , (10) the ocean layer and the water depth is marked by the colour scale. (b) Same
ω
as part (a) but for PREM model. In part (a), we use decreasing thickness for
where ω is the angular frequency of the seismic noise field and n Ul increasing water depth curves to emphasize each ocean depth curve.
is the scalar eigenfunction for a mode with indices (n, l). rr and rs
are related, respectively, to the receiver and source positions. The
division by ω produces the alignment of all ocean depth curves Each curve in Figs 2(a) and (b) is related to a given spheroidal
and enables to plot themas a unique shape. The eigenfunctions are mode. The first coefficient (n = 0) represents the excitation of the
R
normalized considering 0 ρ(r )n Ul 2 (r )r 2 dr = 1, where ρ(r) is the fundamental mode of Rayleigh waves, whereas the others coeffi-
density and R the radius of the Earth. cients (n = 1, n = 2 and so on) are related to the overtones. Our
We compute the excitation coefficients using two different mod- choice of colour scale enables for the first time to understand which
els. The first one is the model used by Longuet-Higgins (1950, p. modes are excited at any ocean depth. The fundamental mode be-
29), hereafter called two layers model, and the second model is comes more and more important for thin ocean layers, whereas for
PREM (Dziewonski & Anderson 1981). We vary the ocean depth deep oceans, the overtones contribute more to the seismic noise
from 1 to 10 km, in discrete steps by kilometre, to simulate the signal.
bathymetry, reproducing intermediate depths by interpolation. We The curves for PREM (Fig. 2b) have a similar shape with respect
compute eigenfunctions from 3 to 12 s and we calculate cn by using to the two layers model ones, with some exceptions. First of all, the
eq. (10) for each depth and frequency. Fig. 2(a) shows the coeffi- different ocean depth curves do not align with each other. This is
cients cn as function of ωh/β for the two layers model and Fig. 2(b) due to the complexity of the stratified PREM model with respect to a
shows the coefficients for PREM. Our normalization is different simple ocean layer above an half-space. The first maximum of each
from that of Longuet–Higgins, but we see on Fig. 2(a) that the shape curve has an amplitude comparable to the two layers case. Moreover,
of the curves is the same. Moreover, the abscissa of the resonant the amplitude of the first five modes are similar to the two layers
peaks is similar between our computation and Longuet–Higgins’ re- case, but this is not the case for the higher overtones. In normal mode
sults, meaning that we reproduce the maximum excitation peaks for language, high-order overtones are related to body waves, meaning
the same combinations of frequency and ocean depth. Differences that their summation corresponds to the body wave packet. The
in shape between Longuet–Higgins and our excitation coefficients sum of all modes, computed for PREM earth model, enables to
are due to the fact that he considered a flat earth model whereas we compute the entire synthetic seismogram, including Rayleigh wave
perform the computation in a spherical earth model. fundamental, higher modes and body waves.

103
Chapter 2. Modeling noise Rayleigh waves

Modelling seismic noise by normal mode summation 5

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

Figure 3. Maps of excitation factor of the fundamental mode n = 0 for a fixed period in the two layers model (first row) and PREM model (second row). The
third row shows the difference between the coefficients calculated in the two layers model and in PREM. The excitation factors for n = 1 are shown in the
fourth row. The two columns correspond to 6 s (left) and 10 s (right) of period.
To determine the regions that excites most the surface waves, Longuet–Higgins coefficients. Similarly to their results, we observe
Fig. 3 shows the excitation coefficients (eq. 10) of the fundamental that, for a period of 6 s, the most excited regions are in the vicinity
mode of Rayleigh waves for the two layers model (top row) and of the ridges or at a few hundred kilometres from the coast. Instead,
the PREM (second row), respectively. These maps are computed for a period of 10 s, the highest excited area covers most of the
for two fixed periods, 6 s (left column) and 10 s (right column) oceans, also further away with respect to the ridges. The differences
using the same colour scale for a given period. These maps can in amplitude between our maps and those by Stutzmann et al. (2012)
be compared to those presented by Stutzmann et al. (2012) using are due to a different normalization.

104
Chapter 2. Modeling noise Rayleigh waves

6 L. Gualtieri et al.

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

Figure 4. (a) Excitation coefficients due to the bathymetry computed in PREM model as a function of ωh/β. ω is the angular frequency, h the ocean depth and
β = 2800 m s−1 is the S-wave velocity in the crust. We compute them by using eq. (10) and considering eigenfunctions of the oceanic model multiplied by
eigenfunctions of the continental model corresponding to the same frequency. In grey, it is shown the case of Fig. 3 with both the eigenfunctions computed in
the oceanic model. (b) Maps of the excitation coefficients at 6 and 10 s of period considering oceanic and continental model for computing the eigenfunctions.

The comparison between the excitation coefficient calculated in The normal mode theory used here assumes a spherically sym-
the two layers model and in PREM (Fig. 3, first two rows) underline metric earth model between source and receiver. When this model
the fact that the excitation depends not only on bathymetry but has a water layer (like PREM model), the station depth is set at the
also on the seismic structure below the seafloor. The maxima are ocean bottom. Therefore, 3-D Earth feature effects on wave prop-
at the same locations for the two models but the amplitudes are in agation, such as ocean–continent transition, can not be accounted
general smaller for PREM. The difference between the excitation for. To estimate this error and check if it does not significantly af-
coefficients calculated in the two layers model and in PREM model fect our conclusions, we compute the excitation coefficients using
(Fig. 3, third row) shows that, for a period of 6 s, the noise sources eq. (10) with eigenfunctions of the oceanic model multiplied by
are more excited in the two layers model with respect to PREM eigenfunctions of the continental model corresponding to the same
everywhere and particularly on both sides of the ridges. For a period frequency.
of 10 s, sources in most of the ocean area are also overexcited in the The excitation coefficients of the Rayleigh waves fundamental
two layers case, except for regions around ridges where sources are mode (yellow curve) are broader and higher (Tanimoto 2012) than
underexcited. in our previous modelling (in grey for comparison). We also ob-
We can also compare the excitation coefficients for the funda- serve that the overtones (green and blue lines) have much smaller
mental mode n = 0 and the most energetic overtone n = 1 (Fig. 3 amplitude.
second and last rows), both of them calculated in PREM. The over- Because we model the amplitude of noise spectra in dB, the
tone n = 1 shows an amplitude of the excitation coefficients smaller difference of the excitation coefficients of the fundamental mode
than the fundamental mode. The maximum amplitudes are of the between the two cases is not so strong and it does not affect our
same order of magnitude for both 6 and 10 s. We also see that final results.
the highest excitation areas are not the same than for the fun- For comparing the spatial sources distribution, Fig. 4(b) shows
damental mode (second row, Fig. 3), especially for the period of two maps of the excitation coefficients at 6 and 10 s of period for
10 s. the fundamental mode computed by using a model without ocean

105
Chapter 2. Modeling noise Rayleigh waves

Modelling seismic noise by normal mode summation 7

for one of the eigenfunctions and with ocean for the other. Making
a comparison with the second line maps in Fig. 3, we observe the
same location of maxima amplitude for 6 s and a different location
for 10 s of period.
However, the use of different models for the source and re-
ceiver eigenfunction—respectively with and without ocean layer—
produces, when we perform normal mode summation, an error
linked with the different discretization of the angular order domain.
This error increases for short period, where the normal modes cat-
alogue becomes dense.

4 SEISMIC NOISE MODELLING

4.1 Group velocity and attenuation

There is no global attenuation model that is accurate in the period
band 3–15 s. Therefore, we started with the QL6 model (Durek
and Ekström 1996), valid for long periods (150 < T < 300 s), and
we modified it in the upper 100 km by decreasing Q. In this way
we find out an apparent attenuation model that enables a good fit
between data amplitude and synthetic spectra. For depths below

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

Figure 6. Attenuation factor and group velocity for the fundamental mode
100 km, Qμ and Qk remain the same as QL6. Decreasing Q, we Rayleigh waves as function of period between 4 and 10 s. The corresponding
maintain the ratio between Qμ and Qk with respect to QL6. Fig. 5 normal mode computation follow Takeuchi and Saito (1972). The different
shows Qμ in QL6 (in black) and in our model (in red) for crustal colours mark the ocean depths between 1 and 10 km.
and mantle depths, shallower than 100 km. We note a difference
of about 90 per cent between these two models. We calibrate our Moskowitz (1963), McCreery et al. (1993) and Webb (1998). We
apparent attenuation model for station SSB and we observe that further show that 0 ql is increasing with increasing water depth for
it is valid also to reproduce data of other stations located in other the fundamental mode (n = 0). The range of 0 ql values presented in
environments. Fig. 6 are consistent with those obtained by Stutzmann et al. (2012)
In Fig. 6, we show group velocity U and attenuation factor 0 ql for and by Prieto et al. (2011) at a local scale.
the fundamental mode of Rayleigh waves calculated with normal In Fig. 6(b) we plot the fundamental mode group velocity as a
modes for periods from 4 to 10 s (Takeuchi & Saito 1972) consider- function of period and water depth for PREM. For a water depth of
ing PREM and varying bathymetry. 0 ql corresponds to the inverse 1 km, group velocity is about 2.5 km s−1 . For a water depth of 2 km,
of the complex part of the eigenfrequencies 0 ωl . We observe in we observe a strong variation of group velocity between 1.1 km s−1
Fig. 6(a) that the attenuation of Rayleigh waves within the Earth for a period of 4 s and 2.4 km s−1 for a period of 10 s. For larger
increases with decreasing period, as already shown by Pierson & water depth, group velocities are all between 1.2 to 1.4 km s−1 in the
entire period range. The varying group velocity as a function of the
bathymetry and period is related to the link between the Rayleigh
wave wavelength and the thickness of the water layer.

4.2 Vertical component of noise spectra

We compute synthetic noise power spectra using normal mode sum-
mation as described in Appendix A. A vertical force Fr is calculated
for each gridpoint on the ocean surface. Synthetic spectra are com-
puted using the ocean wave hindcasts of Ardhuin et al. (2011) and
the coastal reflection coefficients determined by Stutzmann et al.
(2012).
Fig. 7 displays real and synthetic spectra in dB with respect to the
acceleration as function of period for 3 hr of observations at station
SSB (France, Geoscope network). The synthetic spectrum is com-
puted for the sum of the first 100 modes (crosses over a black thin
line) and for each mode separately. The total synthetic spectrum (100
modes) reproduces well the real spectrum. The comparison between
the different modes shows that the fundamental mode is the most
energetic in the entire period band and that the synthetic spectrum
Figure 5. Model of the attenuation parameter Qμ for periods between 4 and amplitude decreases with increasing the overtone order. Moreover,
10 s. In black is shown our starting model QL6 (Durek & Ekström 1996) the difference between them becomes smaller with increasing the
and in red our final apparent attenuation model Qμ that permits best data overtone order: we observe a difference of about 40 dB at 8 s be-
fits using normal mode approach. tween the fundamental mode (yellow line) and the first overtone

106
Chapter 2. Modeling noise Rayleigh waves

8 L. Gualtieri et al.

spectra which is varying with frequency and is about 10 dB at 7 s of

period.
Nishida et al. (2008) showed evidence of Love waves in seis-
mic noise spectra also in the secondary microseismic period band.
Saito (2010) and Fukao et al. (2010) developed a theory of Love-
wave generation based on the fact that a vertical force applied on
a bathymetric shape with a non-null gradient can be separated into
a horizontal and a vertical component. The horizontal component
then excites Love waves and the vertical generates Rayleigh waves.
We make the hypotheses that the discrepancy between real and
synthetic spectra is dominantly due to the presence of Love waves in
recorded signal. Nishida et al. (2008), in their Fig. 2(c), calculated
the kinetic energy ratio between Rayleigh and Love. Considering
only periods between 4 and 10 s, this ratio takes value between 0.5
and 0.7, meaning that Love waves have a kinetic energy varying
between 50 and 70 per cent with respect to Rayleigh wave kinetic
energy. Considering, for example, a period of 7 s, we can write a
relation between the square velocities: v L2 ∼ 0.65 v 2R . Adding to
Figure 7. Vertical component of seismic noise spectra in dB for 3-hr of the energy of Rayleigh waves (black line in Fig. 9) the energy
observations in station SSB (France). We show the comparison between of Love waves calculated from this ratio, we obtain analytically
data (blue dashed line) and synthetic spectra splitting the contribution of
PSE = −134.89 dB for a period of 7 s, which is approximately the

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

fundamental mode and overtones. The amplitude of the synthetic spectra
amplitude of the real spectra in Fig. 9 (blue dashed line). Doing
decreases increasing the overtones number.
this simple analytical computation, we find a proportion of Love
waves energy in agreement with Nishida et al. (2008). Modelling
(green line) and an overlap between the overtones n = 3 (magenta Love waves is beyond the scope of this paper but it can easily be
line) and n = 4 (red line) for the same period. The summation of the done with normal mode theory as shown in Appendix A, consid-
first 100 modes (crosses over a black thin line) is superimposed on ering additional horizontal forces or the components factorization
the spectrum computed with the fundamental mode only, confirm- of vertical forces when they are applied on a bathymetry having a
ing that the fundamental mode of Rayleigh waves is the dominant non-null gradient.
signal in seismic noise recorded by the vertical component. We
also observe that the spectral amplitude computed using only the
4.4 Seasonal variations
first overtone has similar amplitude to the spectrum computed with
overtones from n = 1 to n = 4 (crosses over a thin orange line). Fig. 10 shows background noise seasonal variations for seismic sta-
Fig. 8 shows the power spectral energy in dB with respect to the tion TAM. Schimmel et al. (2011a,b) and Stutzmann et al. (2012)
acceleration as a function of period for the vertical component of analysed noise polarization and concluded that this station records
three Geoscope continental stations, SSB in France (Fig. 8b), TAM noise coming from all surrounding oceans: Atlantic, Indian and
in Algeria (Fig. 8c) and CAN in Australia (Fig. 8d). Seismic spectra Mediterranean Sea. Here, we quantitatively investigate the contri-
are averaged over the year 2010. Data are shown in blue and the bution on noise spectrum amplitude of noise sources located in each
corresponding synthetic spectra in black. Synthetic seismograms ocean. We divide the Atlantic Ocean in two parts, at the Equator,
are calculated between 4 and 10 s. Coastal reflection coefficients to show their influence separately. We calculate average synthetic
are taken from Stutzmann et al. (2012) for each station: 2 per cent spectra for each month of year 2010. In Fig. 10 we show four cases:
for SSB, 2 per cent for TAM and 6 per cent for CAN. The synthetic 2010 January, April, July and December. With blue dashed line we
spectra reproduce quite well both amplitude and shape of the real represent data and with crosses over a thin black line our synthet-
spectra for all stations. We recall that these three spectra, related to ics. We plot on the same figure four other coloured lines, which
stations respectively in Europe, Africa and Oceania, have been cal- represent the contribution of the different portions of the surround-
culated using the same attenuation model, described in the previous ing oceans on spectra amplitude. Coloured lines correspond to the
section and plotted in Fig. 6. In this way, we validate our attenuation coloured oceans in the map.
model for the the secondary microseismic period band. In January (Fig. 10a), the strongest contribution comes from the
Northern Atlantic Ocean (green line) and it is sufficient to repro-
duce the real spectrum. The Southern Atlantic Ocean (red line) and
the Indian Ocean (magenta line) generate noise spectra of similar
4.3 Horizontal components of noise spectra
magnitude within ∼5 dB, the first one being larger above 7 s of
A vertical point force applied on a locally flat seafloor does not period and the second one being larger below 7 s of period. This
generate Love waves. Rayleigh waves instead show energy on the suggests the existence of bigger storms in the Southern Atlantic
vertical and the two horizontal components of the receiver. As we than in the Indian Ocean. But, both spectra are ∼20 dB smaller
consider here only vertical forces, we model only Rayleigh waves than the Northern Atlantic Ocean spectrum. The contribution due
and body waves on the three components. Then, the discrepancy to the Mediterranean Sea (yellow line) is really small for periods
between real and synthetic spectra on the horizontal components can longer than 6 s. This is related to the smaller size of the Mediter-
be used to estimate the amount of Love waves present in the noise. ranean Sea with respect to the oceans which limits the size of the
Fig. 9 shows noise spectra for the East (a) and North (b) components storms and therefore the maximum period of the ocean waves. At
of the Algerian station TAM. Noise spectra are averaged over the very short period—4 and 5 s—the role played by the Mediterranean
year 2010. We observe a discrepancy between the synthetic and real Sea becomes comparable to the contribution of Southern Atlantic

107
Chapter 2. Modeling noise Rayleigh waves

Modelling seismic noise by normal mode summation 9

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

Figure 8. (a) Map with used seismic stations. (b) Vertical component noise spectra in dB for SSB (France). (c) Vertical component noise spectra in dB for TAM
(Algeria). (d) Vertical component noise spectra in dB for CAN (Australia). All seismic spectra are averaged over the year 2010. Data are shown in blue and the
corresponding synthetic spectra in black. The period range of interest is between 4 and 10 s. Synthetic spectra are computed with the IOWAGA ocean wave
model using the same attenuation profiles as function of ocean depth and frequency. Dashed black lines represent respectively the low-noise model (LNM) and
the high-noise model (HNM) spectra (Peterson 1993).

Figure 9. Horizontal components of noise seismic spectra in dB for station TAM. Panel (a) shows the East component and panel (b) the North component. The
difference between theoretical and data spectra is ascribed to the unmodelled Love wave energy inherent to our computation where we use vertical point forces
and a locally flat seafloor. Under this assumption, the estimated Rayleigh wave to Love wave energy ratio is consistent with Nishida et al. (2008). Dashed black
lines represent respectively the low-noise model (LNM) and the high-noise model (HNM) spectra (Peterson 1993).

108
Chapter 2. Modeling noise Rayleigh waves

10 L. Gualtieri et al.

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

Figure 10. Seasonal variations of vertical component of seismic noise spectra for station TAM. Shown are the real (blue dashed line) and synthetic (crosses
over a thin black line) data spectra for (a) January, (b) April, (c) July and (d) December 2010. Coloured lines mark the contribution due to different portions
of the ocean (map above). We observe that, for station TAM, the Northern Atlantic Ocean produces the main contribution in terms of noise source amplitude
almost all the year. Interesting is the increasing noise level in December due to the Mediterranean Sea at short periods. Dashed black lines represent respectively
the low-noise model (LNM) and the high-noise model (HNM) spectra (Peterson 1993).

Ocean and Indian Ocean, whereas at long periods the corresponding During summer (Fig. 10c), we observe a different pattern. North-
spectrum is 40 dB smaller. ern Atlantic Ocean remains the strongest source area, but only for
In April 2010 (Fig. 10b), the spectrum corresponding to the periods below 6 s. At longer period, the energy amount coming from
Northern Atlantic Ocean activity dominates and fits well the ob- the Southern Atlantic Ocean, the Indian Ocean and the North At-
served average spectrum. The spectrum generated by the Mediter- lantic similarly contribute to the total noise spectrum. This is due to
ranean Sea sources has almost the same shape as in 2012 January the fact that during the boreal summer the strongest seismic noise
with a significant contribution only for short periods. The spectrum sources are located on the southern hemisphere (e.g. Stutzmann
generated by sources in the Indian Ocean is ∼10 dB smaller than et al. 2012). Southern hemisphere sources are stronger but they are
the real spectrum, and that generated by South Atlantic sources is further away that the sources in the North Atlantic Ocean and there-
∼15 dB smaller than the real spectrum. fore, they all contribute to the total spectrum. Mediterranean Sea

109
Chapter 2. Modeling noise Rayleigh waves

Modelling seismic noise by normal mode summation 11

We present the first modelling of noise on the horizontal compo-

nents. We consider noise sources as vertical forces and a vertical
point force applied on a locally flat seafloor does not generate Love
waves but they generate Rayleigh and body waves on the vertical and
the two horizontal components. We use the discrepancy between the
real and synthetic spectra of the horizontal components to estimate
the amount of missing Love waves in our synthetic spectra. Our
estimation is in agreement with Nishida et al. (2008). Modelling
Love waves is beyond the scope of this paper but it will be done in
the future with the normal mode theory of Appendix A, consider-
ing additional horizontal forces or the components factorization of
vertical forces related to the topography of the bathymetry.
Figure 11. Vertical force as averaged over 2010 December in the Mediter- We quantify the influence of noise sources in the different oceans
ranean Sea for a fixed period of ∼4.3 s. The red patch near Turkey and Egypt recorded by the station TAM in Algeria. We confirm that this sta-
is related to storm activity and it is the responsible of the increasing noise tion records noise sources from all surrounding oceans and we study
level at short periods due to that area (Fig. 10d). separately the contributions of the different portions of the ocean.
We show that the Northern Atlantic Ocean remains all the time the
contribution is small and the corresponding spectrum is between main noise sources area. Nevertheless, we show that the Mediter-
∼10 dB and ∼20 dB below the real spectrum. ranean Sea also can contribute significantly to the short period noise
Finally in December (Fig. 10d), the Northern Atlantic Ocean in winter.
sources give again a dominant contribution to the spectra as ex- The modelling as presented here can be directly used to investi-

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

pected during the northern hemisphere winter. Indian Ocean and gate in details the Rayleigh wave and body waves generation. The
Southern Atlantic Ocean contributions are again smaller with a formalism presented in the Appendix can be used to investigate the
bigger contribution of Indian Ocean of about 10 dB for most fre- generation of Love waves in relation with the topography of the
quencies. The interesting feature of this month is related to the ocean bottom. Finally, this formalism is well adapted for modelling
Mediterranean Sea. The corresponding spectrum for period shorter the hum considering appropriate sources.
than 8 s becomes 20 dB larger than the other months and larger
than the spectra corresponding to the Southern Atlantic Ocean and
AC K N OW L E D G E M E N T S
Indian Ocean by up to 10 dB. This is due to a strong storm activity
in the Mediterranean sea. Fig. 11 shows, for a period of ∼4 s, the We thank the support from the QUEST Initial Training network
corresponding source area that is located near Turkey and Egypt. founded within the EU Marie Curie Program. M.S. acknowledges
We have plotted here the vertical force derived from the ocean wave support by ILink2010-0112. Most of the figures were realized using
model as average over 2012 December. Generic Mapping Tools (GMT; Wessel & Smith 1998). This is IPGP
contribution number 3386.

5 C O N C LU S I O N S
REFERENCES
In this paper, we present the theory for modelling the secondary Aki, K. & Richards, P.G., 2002. Quantitative seismology, 2nd edn, University
microseisms by normal mode summation. We show that the noise Science Books, Sausalito, California.
sources can be modelled by vertical forces and how to derive them Ardhuin, F., Stutzmann, E., Schimmel, M. & Mangeney, A., 2011.
from a realistic ocean wave model that takes into account coastal Ocean wave sources of seismic noise, J. geophys. Res., 116, C09004,
reflection. We discretize the oceans and we sum up the contribution doi:10.1029/2011JC006952.
of all sources. Bathymetry is an important parameter because it Ardhuin, F., Balanche, A., Stutzmann, E. & Obrebski, M., 2012. From
modulates the excitation of the seismic waves. We show how to seismic noise to ocean wave parameters: general methods and validation,
compute bathymetry excitation effects in a realistic earth model J. geophys. Res., 117, C05002, doi:10.1029/2011JC007449.
Ardhuin, F. & Roland, A., 2012. Coastal wave reflection, directional
using normal modes and we compare our results with Longuet-
spread, and seismoacoustic noise sources, J. geophys. Res., 117, C00J20,
Higgins (1950) computation for a two layers flat model. We show
doi:10.1029/2011JC007832.
that, compared to our more realistic case, the two layers model Ardhuin, F. & Herbers, T.H.C., 2013. Noise generation in the solid Earth,
over predicts the fundamental mode excitation at 6 s of period and oceans and atmosphere, from nonlinear interacting surface gravity waves
that at longer periods, the two layers model either overpredicts or in finite depth, J. Fluid Mech., 716, 316–348.
underpredicts the excitation depending on the area. We also show Berger, J., Davis, P. & Ekstrm, G., 2004. Ambient earth noise: a survey
that strongest excitation areas depend not only on the bathymetry of the global seismographic network, J. geophys. Res., 109, B11307,
and period, but also on the seismic mode. doi:10.1029/2004JB003408.
Seismic noise is then modelled by normal mode summation con- Bromirski, P. & Duennebier, F., 2002. The near-coastal microseism spec-
sidering varying bathymetry. We derive an attenuation model than trum: spatial and temporal wave climate relationships, J. geophys. Res.,
107(B8), ESE 5-1–ESE 5-20.
enables to fit well the vertical component spectra whatever the sta-
Capdeville, Y., 2005. An efficient Born normal mode method to compute
tion location. We select three stations in three different continents
sensitivity kernels and synthetic seismograms in the Earth, Geophys. J.
and we show the good agreement between data and synthetic spec- Int., 163, 639–646.
tra, both in amplitude and in shape, reproducing all the frequency Cessaro, R.K., 1994. Sources of primary and secondary microseisms, Bull.
content of the secondary microseismic noise peak. We show that seism. Soc. Am., 84(1), 142–148.
the fundamental mode of Rayleigh wave is the dominant signal in Chevrot, S., Sylvander, M., Ponsolles, C., Benahmed, S., Lefevre, J.M. &
seismic noise. Paradis, D., 2007. Source locations of secondary microseisms in western

110
Chapter 2. Modeling noise Rayleigh waves

12 L. Gualtieri et al.

Europe: evidence for both coastal and pelagic sources, J. geophys. Res., ocean-generated seismic noise, Geochem. Geophys. Geosyst., 5, Q03004,
112, B11301, doi:10.1029/2007JB005059. doi:10.1029/2003GC000520/.
Durek, J.J. & Ekström, G., 1996. A radial model of anelasticity consistent Schimmel, M., Stutzmann, E., Ardhuin, F. & Gallart, J., 2011a, Earth’s
with long-period surface-wave attenuation, Bull. seism. Soc. Am., 86(1A), ambient microseismic noise, Geochem. Geophys. Geosyst., 12, Q07014,
144–158. doi:10.1029/2011GC003661.
Dziewonski, A. & Anderson, D., 1981. Preliminary reference earth model, Schimmel, M., Stutzmann, E. & Gallart, J., 2011b. Using instantaneous
Phys. Earth planet Inter., 25, 297–356. phase coherence for signal extraction from ambient noise data at a local
Fichtner, A., Trampert, J., Cupillard, P., Saygin, E., Taymaz, T. & Villasenor., to a global scale, Geophys. J. Int., 184, 494–506.
A., 2012. Imaging the North Anatolian Fault Zone with multi-scale full Stehly, L., Campillo, M. & Shapiro, N., 2006. A study of the noise from
waveform inversion, Fall Meeting Abstract 1457037, AGU, San Francisco, its long-range correlation properties, J. geophys. Res., 111, B10306,
CA. doi:10.1029/2005JB004237.
Friedrich, A., Kruger, F. & Klinge, K., 1998. Ocean-generated microseismic Stutzmann, E., Roult, G. & Astiz, L., 2000. Geoscope station noise level,
noise located with the Grafenberg array, J. Seismol., 2, 47–64. Bull. seism. Soc. Am., 90, 690–701.
Fukao, Y., Nishida, K. & Kobayashi, N., 2010. Seafloor topography, ocean Stutzmann, E., Ardhuin, F., Schimmel, M., Mangeney, A. & Patau, G, 2012.
infragravity waves, and background Love and Rayleigh waves, J. geophys. Modeling long-term seismic noise in various environments, Geophys. J.
Res., 115, B04302, doi:10.1029/2009JB006678. Int., 191(2), 707–722.
Gilbert, F., 1970. Excitation of the normal modes of the Earth by earthquake Takeuchi, H. & Saito, M., 1972. Seismic surface waves, Methods Comput.
sources, Geophys. J. R. astr. Soc., 22, 223–226. Phys., 11, 217–295.
Gilbert, F. & Dziewonki, A.M., 1975. An application of normal mode Tanimoto, T., 2012. Seismic noise generation by ocean waves: views from
theory to the retrieval of structural parameters and source mecha- normal-mode theory, Fall Meeting Abstract 1487125, AGU, San Fran-
nisms from seismic spectra, Phil. Trans. R. Soc., 278(A278), 187– cisco, CA.
269. Tape, C., Liu, Q., Maggi, A. & Tromp, J., 2009. Adjoint tomography of the
Gerstoft, P. & Tanimoto, T., 2007. A year of microseisms in southern Cali- southern California crust, Science, 325(5943), 988–992.

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

fornia, Geophys. Res. Lett., 34, L20304, doi:10.1029/2007GL031091. Tape, C., Liu, Q., Maggi, A. & Tromp, J., 2010. Seismic tomography of the
Hasselmann, K., 1963. A statistical analysis of the generation of micro- southern California crust based on spectral-element and adjoint methods,
seisms, Rev. Geophys., 1, 177–209. Geophys. J. Int., 180, 433–462.
Haubrich, R. & McCamy, K., 1969. Microseisms: coastal and pelagic Webb, S.C., 1992. The equilibrium oceanic microseism spectrum, J. acoust.
sources, Rev. Geophys., 7, 539–571. Soc. Am., 92(4), 2141–2158.
Hillers, G., Graham, N., Campillo, M., Kedar, S., Landes, M. & Shapiro, Webb, S., 1998. Broadband seismology and noise under the ocean, Rev.
N., 2012. Global oceanic microsoism sources as seen by seismic arrays Geophys., 36, 105–142.
and predicted by wave action models, Geochem. Geophys. Geosyst., 13, Wessel, P. & Smith, W.H.F., 1998. New improved version of the Generic
Q01021, doi:10.1029/2011GC003875. Mapping Tools released, EOS Trans. Am. geophys. Un., 79(47), 579.
Kedar, S., Longuet-Higgins, M., Graham, F.W.N., Clayton, R. & Jones, C., Woodhouse, J.H. & Girnius, T.P., 1982, Surface waves and free oscillations
2008. The origin of the deep ocean microseisms in the North Atlantic in a regionalized earth model, Geophys. J. E. astr. Soc., 68, 653–673.
Ocean, Proc. R. Soc. Lond. Ser. A., 464(2091), 777–793. Yang, Y. & Ritzwoller, M., 2008. The characteristics of ambient seismic
Longuet-Higgins, M.S., 1950. A theory of the origin of microseisms, Phil. noise as a source for surface wave tomography, Geochem. Geophys.
Trans. Roy. Soc., 243(857), 1–35. Geosyst., 9(2), Q02008, doi:10.1029/2007GC001814.
Longuet-Higgins, M.S., 1952, Can sea waves cause microseisms?, in Pro- Yao, H., Gou(c)dard, P., Collins, J.A., McGuire, J.J. & van der Hilst, R.D.,
ceedings of the Symposium on Microseisms, U.S. Natl. Acad. Sci. Publ., 2011. Structure of young East Pacific Rise lithosphere from ambient noise
Vol. 306, pp. 74–93. correlation analysis of fundamental- and higher-mode Scholte-Rayleigh
McCreery, C., Duennebier, F. & Sutton, G., 1993. Correlation of deep ocean waves, C. R. Geosci., 343, 571–583.
noise (0.4-20Hz) with wind, and the holu spectrum, a world-wide constant,
J. acoust. Soc. Am., 93(5), 2639–2648.
Miche, A., 1944. Mouvements ondulatoires de la mer an profondeur APPENDIX A: NORMAL MODE
croissante ou dècroissante. Première partie. Mouvements ondulatoires S U M M AT I O N T H E O RY U S I N G
pèriodiques et cylindriques en profondeur constante, Annales des Ponts A S I N G L E F O RC E
at Chaussès, 114, 42–78.
Nishida, K., Kawakatsu, H., Fukao, Y. & Obara, K., 2008. Background Love In the following, we derive the elastic displacement field through
and Rayleigh waves simultaneously generated at the Pacific Ocean floors, normal mode summation for a spherical, symmetric, non-rotating,
Geophys. Res. Lett., 35, L16307, doi:10.1029/2008GL034753. perfectly elastic and isotropic (SNREI) earth model. We express
Obrebski, M.J., Ardhuin, F., Stutzmann, E. & Schimmel, M., 2012. How the elastic displacement field by normal modes summation. Be-
moderate sea states can generate loud seismic noise in the deep ocean, cause of the spherical symmetry, the eigenmodes are described by
Geophys. Res. Lett., 39, L11601, doi:10.1029/2012GL051896. only three quantum numbers, n, l and q. n is the radial order, l is
Peterson, J., 1993. Observations and modeling of seismic background noise, the angular order and q represents the type of mode, which can be
USGS Open-File Report 93-322, pp. 95.
spheroidal or toroidal. The eigenfunctions and the associated eigen-
Phinney, R.A. & Burridge, R., 1973. Representation of the elastic-
gravitational excitation of a spherical earth model by generalized spherical
frequencies are indicated as ωk and uk , where k encapsulates all the
harmonics, Geophys. J. R. astr. Soc., 34, 451–487. quantum numbers. Following Gilbert (1970) and Aki & Richard
Pierson, W. & Moskowitz, L., 1963. A proposal spectral for fully developed (2002) the displacement calculated at a certain time t and at a
wind seas based on the similarity theory of s.a. kiaigorodskii, J. geophys. fixed point x = (r sin θ cos φ, r sin θ sin φ, r cos θ) can be expressed
Res., 69, 5181–5190. as,
Prieto, G.A., Denolle, M., Lawrence, J.F. & Beroza, G.C., 2011. On ampli-   
tude information carried by the ambient seismic field, Phys. Earth Int., s(r, t) = u∗k · f(r) dV uk g(t) (A1)
343, 600–614. k VE
Saito, T., 2010. Love-wave excitation due to the interaction between a prop-
agating ocean wave and the sea-bottom topography, Geophys. J. Int., 182, where g(t) is the inverse Laplace–Fourier transform of,

1515–1523. f (s)
Schulte-Pelkum, V., Earle, P. & Vernon, F., 2004. Strong directivity of g(t) = ds (A2)
s 2 − ω2

111
Chapter 2. Modeling noise Rayleigh waves

Modelling seismic noise by normal mode summation 13

which becomes g(t) = sin(ω ωk

k t)
H (t − t0 ) if we consider a point force Comparing equations (A4) and (A5) we are able to define the
in time f (t) = δ(t − t0 ). source and the receiver term in case of a single force,
Moreover, if we consider a point force also in space at source Sk (rs , θs , φs ) = F · u∗k (r0 ) (A6)
coordinates r0s = (rs , θ s , φ s ), then the force can be written as,
f(r) = Fδ(r − r0s ) (A3) Rk (rs , θs , φs ) = v · uk (rr ). (A7)

Knowing the displacement (A1) and the source (A3), a general For the analytical derivation, it is useful to expand a certain lin-
expression of a synthetic seismogram can be written introducing a ear combination of spheroidal components of a tensor in terms of
so-called ‘instrumental vector’ v, which is a unitary displacement generalized spherical harmonics instead to expand the spheroidal
vector in the direction of motion (Woodhouse & Girnius 1982), components of a tensor directly in spherical harmonics (Phinney &
Burridge 1973).
  
sin(ωk t) Then, a general vector can be transformed in canonical coordi-
v · s(rr , t) = u∗k · f(r) dV v · uk H (t) †
k V E
ωk nates as F α = Cαi u i , where α denote the canonical components (+,
0, −) and i the spherical components (r, θ , φ). C† is the hermitian
(A4)
conjugate of the matrix C,
⎛ √ √ ⎞
1/ 2 0 −1/ 2
 exp(iωk t) ⎜ √ √ ⎟
= Rk (rr , θr , φr )Sk (rs , θs , φs )
ωk2
, (A5) C =⎜ ⎝ −i/ 2 0 −i/ 2 ⎠ .
⎟
k
0 1 0
where r0r = (rr , θ r , φ r ) is the receiver position, Rk (rr , θr , φr ) is the
receiver term and Sk (rs , θs , φs ) the source term.

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

The force system, the source and receiver terms can be written in
canonical coordinates as well,
Table A1. Source and receiver coefficients using a point force source.
1
n Ul , n Vl and n Wl are the scalar eigenfunctions for a mode defined by F − = √ (Fθ + i Fφ )
the quantum numbers n and l evaluated at the Earth’s surface for the 2
receiver term and at the source radius for the source term. Fr , Fθ and
Fφ are the components of the force vector and v r , v θ and v φ are the F 0 = Fr (A8)
components of the instrumental vector.
1
N SkN (rs ) RkN (rr ) F + = √ (−Fθ + i Fφ )
2
Spheroidal modes
−1 γl ω0l (F + i F )n Vl γl ω0l (v − iv )n Vl
0 γ l Frn Ul γ l v rn Ul Sk (r0s ) = F + u + + F 0 u 0 + F − u − (A9)
+1 γl ω0l (−F + i F )n Vl γl ω0l (−v − iv )n Vl
Toroidal modes
−1 γl ω0l (−F + i F )n Wl γl ω0l (−v − iv )n Wl Rk (r0r ) = −u + v − + u 0 v 0 − u − v + . (A10)
0 0 0
Following the approach of Phinney & Burridge (1973), that is
+1 γl ω0l (F + i F )n Wl γl ω0l (v − iv )n Wl
using the canonical coordinates, we are now able to rewrite the com-

Figure A1. Eigenfunctions of Rayleigh waves fundamental mode for a period of 6 and 10 s computed by using PREM model in which we set 10 different
ocean depths. The horizontal lines represent the ocean seafloor of each model and their colours are linked with the colour of the respective eigenfunctions. We
observe that the energy is not confined within the ocean layer.

112
Chapter 2. Modeling noise Rayleigh waves

14 L. Gualtieri et al.

ponents of the displacement u introducing the generalized spherical The analytical expression of the vertical component of the synthetic
harmonics YkN (θ, φ), with N = (+, 0, −), seismogram in case of a vertical force can be easily written from
eq. (A5),
u − = γl ω0l (Vk − i Wk )Yk−
v · s = u(r, θ, φ)

u 0 = γl Uk Yk0 = γl2 Fr (rs ) vr Uk (rs ) Uk (rr ) Yk0 ( s,

0
s )Yk ( r, r ) exp(iωk t).

(A12)
u =+
γl ω0l (Vk + i Wk )Yk+ , Due to the vertical point force and the SNREI model only Rayleigh
  waves, P and SV, are excited and recorded at any station.
where ω0l = l(l+1)2
. The normalization coefficient γl = 2l+1 4π
has In Fig. A1 we show the eigenfunctions U of the Rayleigh waves
been found out by applying the orthogonality relation of spherical fundamental mode as a function of depth for two periods, 6 and
harmonics (Phinney & Burridge 1973). 10 s. We compute them by using PREM model in which we change
In Table A1, we summarize the source SkN and the receiver term the thickness of the ocean layer. Horizontal lines represent each
RkN , respectively defined by eqs (A9) and (A10), in canonical coor- ocean seafloor. We use the same colour for the eigenfunction and
dinates for a case of a general point force. the model in which they are computed, identified by the colour
If we consider a purely vertical force, because of the definition of of the ocean seafloor. We observe that the eigenfunctions of the
the force itself (A8) in canonical coordinates, the unique non-null fundamental mode of Rayleigh waves show a cusps at the seafloor.
component of the source term is They are strongly sensitive to the ocean layer (e.g. Yao et al. 2011),
but the amplitude is still significant below the water layer, meaning
n Sl (r s , s, = γl Fr (rs ) Uk (rs ) Yk0 ( s,
0
s) s ). (A11)
that the energy is not confined within the ocean.

Downloaded from http://gji.oxfordjournals.org/ by guest on April 5, 2013

113
Chapter 2. Modeling noise Rayleigh waves

Appendix A: Normal modes associated with Scholte

waves
Scholte waves are non-dispersive elastic waves which propagate along a liquid-solid
interface (Scholte, 1947). These waves are similar to Stoneley waves which propagate
at a solid-solid interface (Stoneley, 1926; Hochstrasser & Stoneley, 1937; Stoneley,
1957). The velocity of Stoneley waves is less than that of the compressional wave in
fluid (αw ' 1.5 km/s) or, in presence of slower sediments below the seafloor, less than
the shear wave velocity in the solid.

Figure 2-18: a) Phase velocity in the PREM model as a function of the period for the
first six modes for periods shorter than T = 3 s. The horizontal grey line is referred to
the velocity of the compressional wave in the fluid layer (αw ' 1.5 km/s). b) Radial
displacement eigenfunction 0 Ul as a function of depth for a couple of Stoneley modes,
computed at T = 2 s (blue) and T = 2.5 s (red). A zoom concerning the first 10 km
is also shown for each eigenfunction.

In the PREM model the phase velocity of Rayleigh waves becomes smaller that
the P-wave velocity αw for periods T . 2.6 s only for the fundamental mode (Figure
2-18a).
Figure 2-18b shows the radial eigenfunctions 0 Ul for a couple of Stoneley modes.
The eigenfunctions are associated with a period of T = 2 s (blue) and T = 2.5 s (red).
A zoom within the first 10 km is also shown for each eigenfunction.
The radial displacement U (Figure 2-18b) exhibits a quasi exponential decrease
from the liquid-solid interface, with a maximum at the boundary. The displacement

114
Chapter 2. Modeling noise Rayleigh waves

becomes more confined to the seafloor with decreasing periods.

Only for periods smaller than T ' 2.6 s Stoneley waves are generated in a simple
model like the PREM. For longer periods, Rayleigh waves propagate, with a non-null
displacement both in the ocean and in the crust (as we have seen in 2.5).

Appendix B: Tsunami modes

The computation of normal modes up to very long periods allows to find out
an ”exotic” kind of modes, called oceanic surface-gravity modes or tsunami modes
(Dahlen & Tromp, 1997, p. 315).
We consider the two-layer model described in section 2.5.1 with a fixed ocean
depth of 3 km and we compute normal modes up to T = 10000 s. In Figure 2-19a,
we present the dispersion diagram of the first ten spheroidal modes. The colour scale
is related to the radial order n of the normal modes.
We immediately observe that, for l = 28 to l = 50, the mode 0 Sl shows a non-
dispersive behaviour, with almost the same eigenfrequency n ωl associated with dif-
ferent angular orders l. Their periods are from T = 9888 s to T = 5526 s and the
associated group velocities are around U = 0.14 km/s= 504 km/h.
The radial eigenfunctions U of the tsunami modes are confined into the homoge-
neous ocean, as illustrated in Figure 2-19b. Their amplitude decreases linearly from
a maximum at the surface to almost zero at the seafloor.
p
Furthermore, the tangential eigenfunctions V – in which the factor l(l + 1) has
been taken into account (see equation 2.21) – are nearly constant within the ocean
and close to zero below the seafloor. The tangential eigenfunctions V at the surface
of the ocean are much bigger – from a factor fifty to a factor one hundred – with
respect to the radial eigenfunctions U.
These modes are the characteristic feature of non-dispersive shallow-water surface-
gravity wave (e.g. Dahlen & Tromp, 1997; Lay & Wallace, 1995).

115
Chapter 2. Modeling noise Rayleigh waves

Figure 2-19: a) Dispersion diagram comparing the tsunami modes and the first ten
spheroidal-mode branches for a two-layer model described in section 2.5.1. The ocean
thickness is fixed to 3 km. The tsunami branch appears to be non-dispersive. b)
Tsunami-mode displacement associated with the radial eigenfunctions U and c) with
the eigenfunction V versus depth.

116
Chapter 3

Modeling the ocean site effect on

noise body waves

3.1 Introduction

In this chapter, we deal with the computation of the ocean site effect on seismic
noise body waves by describing the wavefield as a superposition of plane waves. The
seismic period of interest goes from 3 s to 10 s. This theory is valid for any kind of
source located at the ocean surface.
We firstly detail the computation of the transmission and reflection coefficients at
a solid-liquid interface (section 3.2), which represents the ocean seafloor. Then, we
present the theory for computing the site effect due to the ocean layer upon seismic
body waves generated by a source at the surface of the ocean. We show that the
ocean site effect can be described as the constructive interference of multiply reflected
P waves in the ocean that are then converted to either P or SV waves at the ocean-
crust interface (section 3.3). A detailed discussion about the site effect dependence
on the epicentral distance (section 3.3.1), the frequency and the ocean depth (section
3.3.2) is presented further on. We compare the site effect on Rayleigh waves, as it
has been computed in the previous chapter by using normal modes (section 2.6),
and the site effect on body waves, as it is computed in this chapter by using the
plane wave decomposition, showing that their most amplified sources are located in
different geographical regions (section 3.4). Finally, we show that, by using the ray-
mode duality, we are able to retrieve the imprint of the fundamental mode and the
overtones of Rayleigh waves on the body wave site effect (section 3.5).

117
Chapter 3. Modeling noise body waves

In the end of this chapter, we present a validation of this theory by comparing

the theoretical noise body wave sources with the sources inferred from beamforming
analysis, as it is in Gualtieri et al. (2014)1 .

3.2 Transmission and reflection coefficients at a

solid-liquid interface
In this section, we deal with the computation of the reflection and transmission
coefficient at the liquid-solid interface.
When a wave is incident at the interface between two media, denoted by the subscripts
i = {1, 2}, boundary conditions must be satisfied and the Snell’s law guarantees that
the ray parameter p is constant along the entire ray path:

sin θ1 sin θ2
= =p (3.1)
v1 v2

where vi is the seismic velocity and θi is the take-off angle, with i = {1, 2}. As a
consequence, when a ray traverses materials having different velocities, the take-off
angle of the ray must change in order to maintain the same ray parameter. Knowing
the take-off angles and the velocities, the only remaining parameters that one can
adjust in order to satisfy the boundary conditions are the amplitude of the incident,
the reflected and the transmitted waves at the interface.
Let us consider a 1D model, having a solid half space with a liquid layer on the
top. The liquid layer and the solid half-space schematize the ocean and the crust,
respectively. The main properties of these two media are summarized in Table 3.1.

Density P-wave velocity S-wave velocity

Liquid layer ρw = 1.0 g/cm3 αw = 1.5 km/s βw = 0 km/s
Solid half-space ρc = 2.5 g/cm3 αc = 5.54 km/s βc = 3.2 km/s

Table 3.1: Density and body wave speeds in the liquid layer and in the solid half-space.

We consider a source at the top of the liquid layer to simulate a noise source. A
P-wave is generated at the source and it propagates inside the ocean layer, from the
1
Gualtieri, L., Stutzmann, E., Farra, V., Capdeville, Y., Schimmel, M., Ardhuin, F.,
A., & Morelli, A., 2014. Modeling the ocean site effect on seismic noise body waves,
Geophysical Journal International, 197(2), 1096-1106, doi: 10.1093/gji/ggu042.

118
Chapter 3. Modeling noise body waves

Figure 3-1: Cartoon illustrating the ray paths at the liquid-solid interface. The down-
going incident P-wave (black) is partially reflected as upgoing P-wave (red), partially
transmitted as downgoing P-wave (green) and S-wave (blue).

top to the seafloor. Part of energy is transmitted to the solid half-space as either
downgoing P- and S-waves and part is reflected as upgoing P-waves inside the ocean
layer.

Figure 3-1 illustrates the ray paths at the liquid-solid interface. The take-off
angles are: θPw for the incident and the reflected wave in the liquid layer, θPc for the
transmitted P-wave and θSc for the transmitted S-wave, which are linked by Snell’s
law:
sin θPw sin θPc sin θSc
= = =p (3.2)
αw αc βc
where p is known as the vertical component of the ray parameter. We point out that
the take-off angle of the reflected P-wave is equal to the take-off angle of the incident
wave. We call A0 , A1 , A2 the amplitude of the incident, reflected and transmitted
P-waves, respectively. We call B2 the amplitude of the transmitted S-wave. Because
of the first layer is fluid, no S-wave propagates in there, then B1 = 0.

119
Chapter 3. Modeling noise body waves

Each one of the plane waves can be defined by its potential2 :

• Liquid-layer or ocean :
sin θP cos θP
w w
– Incident downgoing P-wave: φ0 (x, t) = A0 e[iω( αw
x+ αw
z−t)]

sin θP cos θP
w w
– reflected upgoing PV-wave: φ1 (x, t) = A1 e[iω( αw
x− αw
z−t)]

– reflected upgoing S-wave: ψ1 (x, t) = 0

• solid half-space or crust:

sin θP cos θP
c c
– transmitted downgoing P-wave: φ2 (x, t) = A2 e[iω( αc
x+ αc
z−t)]

sin θS cos θS
c c
– transmitted downgoing SV-wave: ψ2 (x, t) = B2 e[iω( βc
x+ βc
z−t)]
,

where ω is the angular frequency, t denotes time, x and z the horizontal and vertical
coordinates, respectively.
The boundary conditions that must be satisfied at the liquid-solid interface are:

1. continuity of the normal displacement at the interface: uzw = uzc ;

2. continuity of the normal traction at the interface: σzzw = σzzc ;

3. vanishing of the tangential traction at the interface: σxz = 0.

The stress tensor components σ = (σzx , σzz ) can be retrieved by using the con-
stitutive law and the strain-displacement relations (e.g. Aki & Richards, 2002) and
they can be written as a function of the displacement components u = (ux , uz ):
 
∂ux ∂uz ∂uz
σzz = λ + + 2µ (3.3)
∂x ∂z ∂z
2
A displacement field can be written using the Helmholtz’s theorem:

u = ∇Φ + ∇ × Ψ

where Φ is the curl-free scalar potential field (∇×Φ = 0) and Ψ is the divergenceness vector potential
field (∇ · Ψ = 0). Physically, Φ does not involve shearing motion and Ψ does not involve volume
changes. They are related to P- and S-waves, respectively. Substituting the Helmholtz’s theorem
into the equation of motion written in terms of the Lamé constants, λ and µ, ρü = (λ + 2µ)∇(∇ ·
u) − µ(∇ × ∇ × u) in which ρ is the density, we obtain two separate equations for Φ and Ψ:
1 1
∇2 Φ − Φ̈ = 0 ∇2 Ψ − Ψ̈ = 0
α2 β2
p p
where α = (λ + 2µ)/ρ and β = µ/ρ are the P- and S-wave velocities, respectively. Then, solving
these equations in terms of potentials, one can determine the displacement field (e.g. Lay & Wallace,
1995; Aki & Richards, 2002). A generic solution of the previous two equation is the harmonic wave
solution.

120
Chapter 3. Modeling noise body waves

 
∂ux ∂uz
σxz = µ + (3.4)
∂z ∂x
where λ and µ are the Lamé constants (see e.g. Lay & Wallace, 1995, p. 51) . The
vertical component of the displacement can be expressed in terms of potential by
using the Helmholtz’s theorem3 :

∂Φ ∂Ψ
uz = − (3.7)
∂z ∂x

Following Geldart & Sheriff (2004) the boundary conditions can be written as:

A1 A2 B2 A0
αw
cos(θPw ) + αc
cos(θPc ) + βc
sin(θSc ) = αw
cos(θPw ) (3.8)
A1 ρw − A2 ρc cos(2θSc ) − B2 ρc sin(2θSc ) = −A0 ρw
A2 βc B2
α2c
sin(2θPc ) − βc
cos(2θSc ) = 0

where A0 , A1 , A2 and B2 are the amplitude of the seismic waves in terms of potential,
θPw , θPc and θSc are the take-off angles as defined in Figure 3-1. The densities and
the seismic wave speeds are defined in Table 3.1.
The ratio between the amplitudes of the reflected wave and the incident wave
is called reflection coefficient R(θPw ) and the ratio between the amplitude of the
transmitted P- and S-waves and the incident wave are called transmission coefficients
TP (θPw ) and TS (θPw ), respectively:

A1 r1 + r2 − r3
R(θPw ) = = , (3.9)
A0 r1 + r2 + r3

A2 2ρw αc cos θPw (1 − 2p2 βc2 )

TP (θPw ) = = , (3.10)
A0 r1 + r2 + r3

3
The components of the gradient of the scalar potential field Φ and the curl of the vector potential
field Ψ are:
 
î ĵ k̂
∂Φ ∂Φ ∂Φ ∂ ∂ ∂  ∂Ψ ∂Ψ
∇Φ = î + ĵ + k̂ ∇ × Ψ = det  ∂x ∂y ∂z
= î − k̂ (3.5)
∂x ∂y ∂z ∂z ∂x
0 −Ψ 0

Then, using the Helmholtz’s theorem, the components of the displacement field are (Geldart &
Sheriff (2004), pp. 28-29):
∂Φ ∂Ψ ∂Φ ∂Ψ
ux = + uz = − (3.6)
∂x ∂z ∂z ∂x

121
Chapter 3. Modeling noise body waves

p
B2 4ρw βc2 p cos θPw 1 − p2 αc2
TS (θPw ) = = . (3.11)
A0 r1 + r2 + r3
where:

r1 = ρc αc (1 − 2p2 βc2 )2 cos θPw (3.12)

p p
r2 = 4βc3 p2 ρc 1 − p2 αc2 1 − p2 βc2 cos θPw
p
r3 = ρw αw 1 − p2 αc2 .

The reflection and transmission coefficients are not frequency-dependent. We note

that in our simple 1D model, the P-to-S transmitted wave at the liquid-solid interface
is a SV-wave.
As a consequence of the Snell’s law (equation 3.2), for an incident P-wave at a
discontinuity to higher velocity from above, there will be a critical take-off angle θP∗ w
at which the take-off angle of the transmitted P-wave becomes θPc = 90◦ :

αw
θPPw∗ = arcsin = 15.71◦ (3.13)
αc

For take-off angles bigger than this, the P-wave in the second medium is evanescent
and it propagates along the interface. The beam width, governed by the horizontal
cos θPc
component of the slowness vector qPc = αc
, becomes zero (Kennett, 2001). Because
the S-wave travels slower with respect to the P-wave in the crust, we expect that the
S-wave critical take-off angle is bigger than the P-wave critical one:

αw
θPS∗w = arcsin = 27.95◦ (3.14)
βc

Figure 3-2 and Figure 3-3 display the plane wave reflection and transmission coef-
ficients in terms of potential at the solid liquid interface as a function of the incident
P-wave take-off angle θPw , as given by equations (3.9), (3.10) and (3.11), respectively.
The P-wave reflection and transmission coefficients are shown in red and blue and
the S-wave transmission coefficient in green. Figure 3-2 is referred to take-off angles
smaller than the S-wave critical one, θPw ≤ θPS∗w and Figure 3-3 to take-off angles
larger than the S-wave critical one, θPw ≥ θPS∗w . The vertical black line in Figure 3-2
corresponds to the critical P-wave take-off angle, that is θPw = θPPw∗ .
When the take-off angle of the incident wave is smaller than the P-wave take-

122
Chapter 3. Modeling noise body waves

Figure 3-2: Plane wave reflection and transmission coefficients in terms of potential at
the liquid-solid interface as a function of the incident P-wave take-off angle θPw ≤ θPS∗w
(θPS∗w = 27.95◦ ), as given by the equations (3.9), (3.10) and (3.11), respectively. The
P-wave reflection and transmission coefficients are shown in red and blue, respectively.
The S-wave transmission coefficient is in green. The vertical black line denotes the
critical P-wave take-off angle in equation (3.14).

123
Chapter 3. Modeling noise body waves

Figure 3-3: Plane wave reflection and transmission coefficients in terms of potential at
the liquid-solid interface as a function of the incident P-wave take-off angle θPw ≥ θPS∗w
(θPS∗w = 27.95◦ ), as given by the equations (3.9), (3.10) and (3.11), respectively. The
P-wave reflection and transmission coefficients are shown in red and blue, respectively.
The S-wave transmission coefficient is in green.

124
Chapter 3. Modeling noise body waves

off critical angle, θPw < θPPw∗ , that is θPw < 15.71◦ , the reflection and transmission
coefficients are real. In turn, when the reflection and transmission coefficients are real,
an incident plane wave is merely scattered in amplitude on reflection and transmission
(Kennett, 2001). Then, P- and S-waves propagate in the solid half space and it does
not occur any kind of interference between them.
When the take-off angle of the incident wave is bigger than the P-wave take-
off critical one, θPw ≥ θPPw∗ , that is θPw ≥ 15.71◦ , the reflection and transmission
coefficients are complex. Once the coefficients become complex, the shape of the
transmitted waves is modified (see, e.g. Hudson, 1962; Kennett, 2001). In this case,
evanescent waves propagate horizontally, along the interface between the liquid layer
and the solid half space. In particular, when θPPw∗ ≤ θPw < θPS∗w , the transmitted P-
wave is evanescent throughout the half-space, but the S-wave has a travelling wave
behaviour. Instead, when θPw ≥ θPS∗w , both P and S-waves are evanescent. In both
cases, when the P-wave is evanescent and the S-wave is propagating and when the P-
and S-waves are both evanescents, interference between P and SV-waves occurs and
Rayleigh wave normal modes are generated (Kennett, 2001, p. 337).
In the following, we will consider only incident take-off angles smaller than the
critical angle θPPw∗ . The emergence of evanescent (or inhomogeneous) waves and the
ray-mode duality will be detailed only in section 3.5.

3.3 Estimation of the ocean site effect on body

waves
Let us consider a pressure source at the surface of the ocean in order to schematize
a noise source. The surface of the ocean is considered as a free surface. The elastody-
namic representation theorem (Aki & Richards, 2002, chapter 2) written in terms of
the Green’s function and satisfying the free surface boundary conditions at the ocean
surface allows us to compute the seismic waves generated by this pressure field. In
our case, the sources are distributed along the ocean surface (e.g. Longuet-Higgins,
1950; Hasselmann, 1963), which corresponds to the external surface S of the physical
domain. There are not body forces within the volume V enclosed by the surface S.
Then, only the surface integral term contributes to the expression of the representa-
tion theorem. From Aki & Richards (2002) (p. 29, equation (2.43)), in this case, the
representation theorem allows to compute the displacement as the convolution of the

125
Chapter 3. Modeling noise body waves

Figure 3-4: Cartoon illustrating the liquid-solid interface between a solid half space
and a liquid layer. Seismic P-waves are generated at the source located at the surface
of the liquid layer. Reflection and transmission occur at the discontinuity. The upper
surface of the liquid layer is considered as free surface, than total reflection occurs.
After Gualtieri et al. (2014).

Green’s function Gfniree (x, t; ξ, τ ) and the traction Ti (u(ξ, τ ), n):

Z ∞ Z Z
u(x, t) = dτ Gfniree (x, t; ξ, τ )Ti (u(ξ, τ ), n)dS (3.15)
−∞ S

where Gfniree (x, t; ξ, τ ) is the Green’s function for a source at (ξ, τ ) and an observation
point at (x, t). Because of the liquid layer, the traction at the surface of the ocean
Ti (u(ξ, τ ), n) acts towards the local normal direction n with respect to the horizontal
surface of the ocean. Then, the traction represents the pressure source at the surface of
the ocean acting along the vertical direction. The Green’s function can be decomposed
as a sum of plane waves. In the far field, this sum can be approximated by using
the stationary phase method. The corresponding expression is called the ray-theory
Green’s function, and it contains only the ray contributions.

In this section, we use the plane-wave decomposition of the Green’s function and
we compute the ocean site effect on one selected plane wave. As for the reflection
and transmission at the liquid-solid interface shown in the previous section, only
plane P-waves are considered in the ocean layer. The previously described liquid-

126
Chapter 3. Modeling noise body waves

solid interface represents the contrast between the ocean and the crust. The up-
going P-wave, generated by the reflection at the liquid-solid interface, is then totally
reflected at the free surface and at the seafloor a new reflection/transmission point
occurs. The second-time reflected P-wave travels upgoing in the liquid layer, it is
totally reflected at the free surface and, traveling again downward, it generates a new
reflection/transmission point at the seafloor. Then, the n-th plane wave is n times
reflected at the ocean bottom and n times totally reflected at the free surface as P-
wave before being transmitted to the beneath solid medium as P and S-waves (Figure
3-4).

Gualtieri et al. (2014) have shown how to compute the P- and S-wave potentials in
the crust just below the ocean bottom by summing up the multiply reflected wave in
the liquid layer (see the paper in the end of this chapter). The potential in the crust
is scaled with respect to the potential in the water layer by a factor which depends
on the geometrical series given by the multiply reflected waves in the liquid layer.

We call this factor ocean site effect on P-waves (Gualtieri et al., 2014):

CP (θPw , h, ω) = TP (θPw ) 1 + (−R(θPw ) eiΦw (h,ω,θPw ) ) + ... + (−R(θPw ) eiΦw (h,ω,θPw ) )n →

 

TP (θPw )
= .
1 + R(θPw ) eiΦw (h,ω,θPw )
(3.16)

and ocean site effect on S-waves:

CS (θPw , h, ω) = TS (θPw ) 1 + (−R(θPw ) eiΦw (h,ω,θPw ) ) + ... + (−R(θPw ) eiΦw (h,ω,θPw ) )n →

 

TS (θPw )
= .
1 + R(θPw )eiΦw (h,ω,θPw )
(3.17)

We recall that R(θPw ), TP (θPw ) and TS (θPw ) are the P-wave reflection coefficient and
the P- and S-wave transmission coefficients, as given by equations (3.9), (3.10) and
(3.11), respectively. They are computed for incident take-off angles smaller than the
P-wave critical one, that is only for θPw ≤ θPPw∗ . Φw (h, ω, θPw ) is the phase shift in the
ocean layer, as given by Gualtieri et al. (2014):

cos θPw
Φw (h, ω, θPw ) = 2ω h (3.18)
αw

127
Chapter 3. Modeling noise body waves

where ω is the angular frequency and h is the liquid layer thickness, that is the ocean
depth. This phase shift introduces in the site effect the dependence on frequency
and ocean depth. These coefficients vary with the incident take-off angle θPw , the
frequency ω and the ocean depth h. More precisely, they vary with the product f h.
We point out that the site effect on body waves computed here, as well as the site
effect on Rayleigh waves computed in the previous chapter (section 2.6), are valid for
a generic kind of source at the ocean surface.

3.3.1 The ocean site effect for varying epicentral distance

The ocean site effect on body waves depends on the take-off angle of the incident
wave, that is on the ray parameter p (from the Snell’s law in equation (3.2)), for
take-off angle smaller than the P-wave critical one (θPw ≤ θPPw∗ ). Given an Earth
model, the ray parameter p is related in turn to the the epicentral distance ∆. Then,
the site effect on P- and S-waves given by equations (3.16) and (3.17) varies with the
epicentral distance as well as with the frequency and the ocean depth.
Direct P-waves and S-waves propagate from the source up to about an epicentral
distance of 95◦ . Then, only for given take-off angles the source generates direct P-
and S-waves. The relation between the epicentral distance and the take-off angle of
the incident P-wave for the direct P- and S-waves, computed in the iasp91 Earth’s
model (Kennett & Engdahl, 1991), can be found in Table 1 of Gualtieri et al. (2014)
(see the paper in the end of this chapter).
Figure 3-5 displays the coefficients CP and CS as given by equations (3.16) and
(3.17) as a function of the epicentral distance and the product between the frequency
f and the ocean depth h, normalised by the P-wave velocity in the ocean layer αw .
The correspondence between the take-off angle and the epicentral distance for P- and
S-waves is shown on the ordinate axes of Figure 3-5a and Figure 3-5b, respectively.
The colour scale in Figure 3-5a and Figure 3-5b is referred to the amplitude of the
site effect CP and CS , respectively. Instead, it is related to the ratio CP /CS in Figure
3-5c. Only epicentral distances larger than 30◦ have been considered to avoid the P-
and S-wave triplication regime, which occurs for epicentral distances smaller than 30◦
(e.g. Shearer, 2009).
The site effect CP is nearly constant for varying epicentral distances (Figure 3-5a),
with some combinations of frequency and ocean depth which are mostly amplified.
Instead, the site effect on S-waves CS (Figure 3-5b) decreases in amplitude with

128
Chapter 3. Modeling noise body waves

Figure 3-5: Ocean site effect on a) P-waves and b) S-waves as a function of the
epicentral distance (or the take-off angle) and the product between frequency and
ocean depth, normalised by the P-wave velocity in the ocean layer. The colour scale
is referred to the amplitude of these coefficients as given by equations (3.16) and
(3.17). The panel c) displays in colours the ratio between the CP and CS . Only
epicentral distances larger than 30◦ have been considered to avoid the P- and S-wave
triplication regime.

129
Chapter 3. Modeling noise body waves

Figure 3-6: Ocean site effect on a) PKP-waves and b) SKS-waves as a function of

the epicentral distance (or the take-off angle) and the product between frequency and
ocean depth, normalised by the P-wave velocity in the ocean layer. The colour scale is
referred to the amplitude of these coefficients as given by equations (3.16) and (3.17).

increasing epicentral distance. The maxima of the S- and P-wave site effect occur at
the same combination of frequency and ocean depth, but the site effect on S-waves
is always weaker then the one on P-waves. At short epicentral distances the two
coefficients are comparable in amplitude, instead their difference increases for large
epicentral distances (Figure 3-5c). This phenomenon, combined with the attenuation
effects, has important consequences on the possibility to retrieve S-wave sources, as
shown by Gualtieri et al. (2014).
Different seismic body waves are generated from different take-off angles (Gualtieri
et al., 2014, Table 1). Then, the same computation shown in Figure 3-5 can be
performed considering other seismic phases. In Figure 3-6, the coefficients CP and

130
Chapter 3. Modeling noise body waves

Figure 3-7: P- and S-wave site effect obtained by summing up all the take-off angles
smaller than the P-wave critical one, as given by equations 3.19 and 3.20. The colour
scale is referred to the ocean depth. After Gualtieri et al. (2014).

CS have been computed for take-off angles which correspond to the emergence of PKP
(Figure 3-6a) and SKS waves (Figure 3-6b). We observe that, as for direct P- and
S-waves in Figure 3-5, some combination of frequency and ocean depth are mostly
amplified. PKP waves show the same amplification for varying epicentral distances,
SKS waves are instead mostly amplified for large epicentral distances. The amplitude
of the ocean site effect on PKP waves is comparable with the one on P-waves (Figure
3-5a). The amplitude of the site effect on SKS waves is instead very small with respect
to the one on S-waves (around a factor of three with respect to Figure 3-5b).

3.3.2 The ocean site effect for varying frequency and ocean
depth
The wavefield generated by a source at the surface of the ocean can be decom-
posed into plane waves with different take-off angles. Then, to evaluate the total
site effect on the transmitted P- and S-wave amplitude, we sum up all the site effect
coefficients due to the different take-off angles. The site effect on the transmitted
P-wave amplitude is:
Z ∗
θP 1/2
w
2
cP (h, ω) = |Cp (θPw , h, ω)| dθPw (3.19)
0

and on the S-wave amplitude is:

131
Chapter 3. Modeling noise body waves

Figure 3-8: Maps of the site effect on P-waves as given by the equation (3.19) for
periods from 3 to 10 s.

132
Chapter 3. Modeling noise body waves

Figure 3-9: Maps of the site effect on S-waves as given by the equation (3.20) for
periods from 3 to 10 s.

133
Chapter 3. Modeling noise body waves

Z ∗
θP 1/2
w
2
cS (h, ω) = |CS (θPw , h, ω)| dθPw (3.20)
0

The fact that the integral runs over take-off angles smaller than θPPw∗ ensure avoiding
evanescent waves.
In Figure 3-7, the coefficients cP and cS have been plotted as a function of the
dimensionless quantity f h/αw , where f is the frequency, h is the ocean depth and αw
is the P-wave velocity in the ocean layer. The colour scale is referred to the ocean
depth.
The site effect on S-waves is always smaller then the one on P-waves, but they
have the same shape, with a few differences for large abscissa (that is for rare ocean
depth and short periods). Only some combination of frequency and ocean depth are
amplified.
Fixing the period, we can build maps of the site effect upon P- and S-waves. We
present the P- and S-wave site effect for periods from 3 s to 10 s in Figure 3-8 and
Figure 3-9, respectively. Fixing the period, the strongest amplification on P- and
S-waves occurs in the same geographical regions. Only the amplitude differs of about
a factor 2.5, as in Figure 3-7. At a given geographical region, a very strong variability
occurs for varying period. Further considerations can be found in Gualtieri et al.
(2014), which is attached at the end of this chapter.

3.4 Comparison between the site-effects on body

and Rayleigh waves
In the previous chapter, the site effect on the fundamental mode of Rayleigh waves
has been computed by using normal modes in different Earth models (see Figure 2-
15). Gualtieri et al. (2013) have shown that the fundamental mode of Rayleigh waves
is able alone to explain the main features of noise spectra, so that we can neglect the
overtones and focus on the fundamental mode.
In this section, we compare the site effect which occurs on the fundamental mode
of Rayleigh waves (Figure 2-15a) and on body waves (Figure 3-7). The site effect
on the fundamental mode of Rayleigh waves presents a main peak related to some
given combination of frequency and ocean depth, before falling away to zero. Instead,
the site effect on body waves shows the presence of several peaks, so that a larger

134
Chapter 3. Modeling noise body waves

Figure 3-10: Ocean site effect on Rayleigh waves (left column) and on P-waves (right
column) for periods from T = 3 s to T = 5 s. Both of them have been normalised to
end up with comparable amplitudes.

135
Chapter 3. Modeling noise body waves

number of combinations of frequency and ocean depth are amplified. The leftmost
peak in Figure 3-7, associated with long period body waves has the same abscissa
and a similar shape to the peak of the fundamental mode of Rayleigh waves in Figure
2-15a. Because of that, for periods longer than about 6 s, the strongest noise sources
of Rayleigh waves (Figure 2-16) and body waves (Figure 3-8) are located at the same
geographical regions. For shorter periods, from T = 3 s to T = 5 s, different locations
occur for the fundamental mode of Rayleigh and body wave sources.

In Figure 3-10, the ocean site effect on Rayleigh waves – computed by using
normal modes – and P-waves – computed by using the plane wave decomposition –
are plotted for periods from T = 3 s to T = 5 s. Both of them have been normalised
to their maxima in order to get comparable amplitudes. Comparing these maps for
the same period, we observe that short period Rayleigh waves are mostly amplified
near to the coast, whereas body waves are mostly related to the deep ocean.

For understanding the reason why the ocean site effect on short period Rayleigh
and on body waves are different, in Figure 3-11 we present maps of the site effect on
P-waves – as given by equation (3.19) – separating the contribution due to the first
three peaks in Figure 3-7. The first column shows maps of the site effect only due to
the leftmost peak in Figure 3-7 with abscissa f h/αw < 0.5, the second column only
due to the second peak with abscissa 0.5 ≤ f h/αw < 1 and the third column only due
to the third peak with abscissa 1 ≤ f h/αw < 1.5. Each row is referred to a period
from T = 3 s (first row) to T = 10 s (last row). At T = 3 s of period (maps on the
first row in Figure 3-11), the main contribution to the body wave site effect comes
from the second and the third peak (i.e. maps on the second and third columns).
For periods from T = 4 s to T = 6 s (maps from the second to the forth row in
Figure 3-11) both the first and the second peaks contribute to the site effect. The
contribution due to the second peak decreases in amplitude for increasing periods.
For periods T > 5 s, its contribution becomes negligible.

The fundamental mode of Rayleigh waves (Figure 2-15a) has only a maximum
around f h/αw = 0.5, with slightly differences in shape with respect to the one of
body waves. Then, the first column in Figure 3-11 displays the geographical regions
where both Rayleigh and body waves are mostly amplified. Instead, the second and
the third column display the geographical regions where body wave noise source are
amplified and Rayleigh waves noise sources are not. Thus, Rayleigh and body wave
noise sources are amplified in a different way only at short period (T ≤ 5 s), whereas

136
Chapter 3. Modeling noise body waves

Figure 3-11: Caption on the following page.

137
Chapter 3. Modeling noise body waves

Figure 3-11: Maps of the site effect on P-waves as computed by equation 3.19 and
considering only the contribution due to a) the first peak (f h/αw < 0.5); b) the
second peak (0.5 ≤ f h/αw < 1) and c) the third peak (1.5 ≤ f h/αw < 1) as in
Figure 3-7. Each row is referred to a given period, from 3 s (first row) to 10 s (last
row).

the strongest amplification occurs at the same geographical regions at long period
(T > 5 s).
The S-wave site effect due to each peak in Figure 3-7 has a similar geographical
distribution since the site effect on S- and P-waves have the same shape as a function
of f h/αw .

3.5 Ray-mode duality

An alternative representation of the complex wavetrains generated by the multiple
reflections in the ocean layer is provided by the superposition of many normal modes.
In section 3.2, we have seen that the reflection and transmission coefficients are real
only for incident take-off angles smaller than the P-wave critical one θPPw∗ . For take-off
angles larger than θPPw∗ , the reflection and transmission coefficients become complex
(Figure 3-2 and Figure 3-3). Evanescent P- and S-waves are then generated and they
propagate along the discontinuity. For a given ray parameter p, P- and SV-waves
interfere with each other and they generate Rayleigh waves.
For a source at the surface of the ocean, plane waves are emitted with all angles

138
Chapter 3. Modeling noise body waves

from θPw = 0◦ up to θPw = 90◦ . These two extreme values are related to the vertical
and the horizontal propagation with respect to the ocean surface, respectively.
In figure 3-12, the site effect on S-waves CS is shown as a function of the take-off
angle θPw and the angular frequency ω, considering an ocean layer of 1 km (Figure
3-12a), 3 km (Figure 3-12b) and 5 km depth (Figure 3-12c). The site effect is shown
as computed in equation (3.17) for all take-off angles from θPw = 0◦ up to θPw = 90◦ .
The white lines indicate the critical take-off angle of P-waves (equation 3.13) and
S-waves (equation 3.14).
If the take-off angle is larger than the S-wave critical one θPS∗w = 27.95◦ , both P-
and S-waves are evanescent. In this case, the amplitude of the site effects (colour
scale) varies with frequency, showing some aligned maxima for varying the take-off
angle. Increasing the ocean depth, that is moving from Figure 3-12a to Figure 3-12c,
these maxima move towards larger take-off angles and new branches appear in case
of deep water.
For take-off angles larger than the critical take-off angle of S-waves θPS∗w , the reflec-
tion coefficient is constant and the P- and S-wave transmission coefficients have the
same shape (Figure 3-3). The P-wave transmission coefficient is always smaller than
the S-wave transmission coefficient because the decay rate with depth is much more
for evanescent P-waves than for evanescent S-waves (Kennett, 1985). The amplitude
of the site effect on P-waves as a function of the take-off angle and frequency is smaller
than the one on S-waves, but it presents the same behaviour.
For take-off angles smaller than the critical take-off angle of S-waves θPS∗w , the
amplitude of the site effect is weaker than for larger take-off angles. Therefore, the
site effect is hidden in the background blue colour.
Body waves are associated with the constructive interaction of high order over-
tones. Gualtieri et al. (2014) have shown how normal modes can be associated with
the generation of the different body wave phases (see Appendix of the paper in the
end of this chapter). Through the wavenumber-ray parameter relationship and the
wavenumber-angular order relationship (their equations (A2) and (A3)), they end up
with the computation of the angular order l as a function of the ray parameter p and
the frequency n ωl : r
1 1
l=− + + n ω l 2 p2 (3.21)
2 4
Gualtieri et al. (2014) have used this equation to identify the normal modes associated
with P-waves at epicentral distances between 30◦ and 90◦ and compute the site effect

139
Chapter 3. Modeling noise body waves

Figure 3-12: Site effect CS (colour scale) as computed by equation (3.17) as a function
of the take-off angle θPw and the angular frequency ω, considering an ocean layer of
a) 1 km, b) 3 km and c) 5 km depth. The white lines denote the critical take-off angle
of P-waves θPPw∗ = 15.71◦ (equation 3.13) and S-waves θPS∗w = 27.95◦ (equation 3.14).

140
Chapter 3. Modeling noise body waves

Figure 3-13: Site effect CS (colour scale) as computed by equation (3.17) as a function
of the angular order l and the angular frequency ω, considering an ocean layer of a)
1 km, b) 3 km and c) 5 km depth. The angular order l has been computed by
using equation (3.21). The white lines denote the critical take-off angle of P-waves
θPPw∗ = 15.71◦ (equation 3.13) and S-waves θPS∗w = 27.95◦ (equation 3.14). The small
black boxes display the dispersion diagrams of the first five normal modes as presented
in Chapter 2, Figure 2-5.
141
Chapter 3. Modeling noise body waves

on body waves by using normal modes (like it has been done in the previous chapter).
Extending the ray parameter range to all possible take-off angles from θPw = 0◦ to
θPw = 90◦ , we include normal modes associated with body waves and Rayleigh waves.
We recall that neither SH waves nor Love waves are generated at our liquid-solid
interface.
Figure 3-13 displays the site effect on S-waves as a function of the angular order
l – as given by the equation (3.21) – and the angular frequency ω, considering an
ocean layer of 1 km (Figure 3-13a), 3 km (Figure 3-13b) and 5 km depth (Figure
3-13c). The white lines denote the critical take-off angle of P-waves (equation 3.13)
and S-waves (equation 3.14). For take-off angles larger than the S-wave critical one
(furthest white line to the right), the fundamental mode and the overtones of Rayleigh
waves clearly appear as strong maxima of CS . We observe that the normal modes
appear to be more separated increasing the ocean depth, that is moving from Figure
3-13a to Figure 3-13c. The same behavior has been previously observed in Chapter
2 concerning Rayleigh waves normal modes (Figure 2-5).
Figure 2-5 – which shows the dispersion diagram for the first five normal modes
computed in the two-layer model varying the ocean depth – has been redrawn in
Figure 3-13 (small black boxes). Making a comparison between them, for each ocean
depth we can identify the furthest branch to the right as to the fundamental mode
of Rayleigh waves. Fixing the ocean depth, we observe that the angular order is
nearly the same fixing the frequency, both for the fundamental mode and the first
overtone. A few small differences are due to the fact that the Earth’s models are
slightly different.
We point out that the amplitude of the site effect associated with the higher
overtone numbers is smaller than the amplitude of the fundamental mode and the
first overtone and they are hidden in the background blue colour.
The phase velocity is defined as c = 1/p = αw / sin(θPw ). In Chapter 2, the phase
velocities associated with the normal modes of Rayleigh waves have been computed
for several Earth’s models (section 2.5). Let us consider the simplest one, the two-
layer model (section 2.5.1) with ρ = 1000 kg/m3 and α = 1.4 km/s in the ocean
layer and ρ = 2500 kg/m3 , α = 4.85 km/s and β = 2.8 km/s in the crust and let us
compute the site effect of body waves CP and CS for take off angles larger than the
critical take-off angle of S-waves in this Earth model.
Figure 3-14 displays the coefficients CP and CS as a function of f h/αw and of the

142
Chapter 3. Modeling noise body waves

phase velocity c. The colour scale is referred to CP (Figure 3-14a) and CS (Figure
3-14b). In Figure 3-14c we plot the phase velocity as a function of f h/αw computed
by using normal modes (as in Chapter 2, section 2.5.1) considering ocean depths from
1 km to 10 km (colour scale). The periods goes from 4 s to 10 s both by using the
plane wave decomposition (i.e. Figure 3-14a and Figure 3-14b) and by using normal
modes (Figure 3-14c). The white line in Figure 3-14a and Figure 3-14b and the black
line in Figure 3-14c, at c = 2.8 km/s, correspond to the critical take-off angle of
S-waves, that is c = 1/p = αw / sin(θPS∗w ) = βc .
In Figure 3-14, we clearly see the first four modes of Rayleigh waves appearing
from the computation of the ocean site effect of P- and S-waves when the phase
velocity is smaller than the S-wave velocity in the crust. The values of the phase
velocity correspond to the ones computed by using normal modes. For larger values
of the phase velocity, only S-waves propagate whereas P-waves are evanescents.
The ocean site effect on Rayleigh waves can be obtained by finding out the poles
or nodes of the ocean site effect CP and CS and by applying the method of residues
(Longuet-Higgins, 1950).

143
Chapter 3. Modeling noise body waves

Figure 3-14: Phase velocity as a function f h/αw , where f is the frequency, h is the
ocean depth and αw is the P-wave velocity in the ocean layer. a) Ocean site effect of
P-waves (colour scale); b) Ocean site effect on S-waves (colour scale); c) normal mode
computation, showing the ocean depth with different colours. The white lines (upper
and central figures) and the black line (bottom figure) at c = 2.8 km/s is referred to
the S-wave velocity in the crust.

144
Chapter 3. Modeling noise body waves

3.6 Modeling the ocean site effect on seismic noise

body waves
In the following, we attach the paper Gualtieri et al. (2014) which presents the
computation and the validation of the site effect on seismic body waves by using the
plane wave superposition of the wavefield. The secondary microseismic period band
and a source at the surface of the ocean have been considered.
After a little review about the main contributions to the scientific literature con-
cerning body wave source detections, this study deals with the modeling of the ocean
site effect on P- and S-waves. It demonstrates that the ocean site effect can be mod-
elled by considering the constructive interferences of multiply reflected P-waves in
the ocean, which are converted into either P and SV-waves at the liquid-solid seafloor
interface.
Then, it presents the quantitative study of the ocean site effect on body waves
considering periods from 3 s to 10 s and ocean depths from 1 to 10 km.
An application concerning the noise body wave sources generated by the typhoon
Ioke (2006, September 4) is shown further on. The theoretical noise body wave sources
have been computed by considering the pressure field due to the interaction of the
ocean gravity waves (see Chapter 1, section 1.5) and by taking into account the ocean
site effect.
For validating the theoretical source location, we analyze the Typhoon Ioke data
recorded by the Southern California Seismic Network, considering the three-component
seismograms. A good match has been observed between the theoretical and observed
noise source detected on the vertical and P-component beamforming power spectra.
On the other hand, the amplitude of the generated SV wave is too small to be detected
on the SV-component beamforming.
Finally, the Appendix of this paper presents a different approach for computing
the body wave site effect. Normal modes related to a specific phase of body waves
can be identified by associating the ray parameter with the horizontal component of
the wavenumber. A comparison between the Rayleigh wave and body wave site effect
is presented in the end of the Appendix.

145
Chapter 3. Modeling noise body waves

146
Chapter 3. Modeling noise body waves

Geophysical Journal International

Geophys. J. Int. (2014) doi: 10.1093/gji/ggu042
Geophysical Journal International Advance Access published March 7, 2014

Modelling the ocean site effect on seismic noise body waves

L. Gualtieri,1,2 E. Stutzmann,1 V. Farra,1 Y. Capdeville,3 M. Schimmel,4 F. Ardhuin5

and A. Morelli6
1 Institut de Physique du Globe de Paris, Sorbonne Paris Cité, UMR 7154 CNRS, Paris, France. E-mail: [email protected]
2 Dipartimento di Fisica e Astronomia, Settore di Geofisica, Università di Bologna, Bologna, Italy
3 Laboratoire de Planétologie et Géodynamique de Nantes, CNRS, Université de Nantes, UMR-6112, France
4 Institute of Earth Sciences Jaume Almera, CSIC, Lluis Sole i Sabaris s/n, E-08028 Barcelona, Spain
5 Ifremer, Laboratoire d’Océanographie Spatiale, F-29280 Plouzané, France
6 Istituto Nazionale di Geofisica e Vulcanologia, Bologna, Italy

Accepted 2014 February 3. Received 2014 January 13; in original form 2013 July 15

SUMMARY

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

Secondary microseismic noise is generated by non-linear interactions between ocean waves
at the ocean surface. We present here the theory for computing the site effect of the ocean
layer upon body waves generated by noise sources distributed along the ocean surface. By
defining the wavefield as the superposition of plane waves, we show that the ocean site effect
can be described as the constructive interference of multiply reflected P waves in the ocean
that are then converted to either P or SV waves at the ocean–crust interface. We observe that
the site effect varies strongly with period and ocean depth, although in a different way for
body waves than for Rayleigh waves. We also show that the ocean site effect is stronger for P
waves than for S waves. We validate our computation by comparing the theoretical noise body
wave sources with the sources inferred from beamforming analysis of the three seismogram
components recorded by the Southern California Seismic Network. We use rotated traces for
the beamforming analysis, and we show that we clearly detect P waves generated by ocean
gravity wave interactions along the track of typhoon Ioke (2006 September). We do not detect
the corresponding SV waves, and we demonstrate that this is because their amplitude is too

GJI Seismology
weak.
Key words: Body waves; Site effects; Theoretical seismology.

many different azimuths. This mechanism dominates at frequen-

1 I N T RO D U C T I O N
cies from 0.5 to 2 Hz, due to the wide angular distribution of the
Microseisms are continuous oscillations of the ground with periods short waves generated by a constant and steady wind, and it can
of between 3 and 20 s that can be detected worldwide (e.g. Gutenberg still be significant at lower frequencies. In this case, the interacting
1936; Webb 1998; Stutzmann et al. 2000; Berger 2004). They can waves are within the storm. For the second class of sea-state con-
be generated by the interactions among the atmosphere, the ocean figuration, ocean gravity waves arrive at the coast, where they are
and the solid Earth. Seismic noise spectra show two main peaks reflected and then meet up with incident ocean gravity waves. The
with periods of about 14 and 7 s, which are known as the primary interaction area is confined close to the coast. The third class of
and secondary microseisms, respectively. sea-state configurations relates to the interactions of ocean gravity
The primary microseismic noise is the smaller amplitude hump waves coming from different storms. Ocean gravity waves from any
that is generated by the interactions between the ocean gravity waves given storm can travel long distances before meeting ocean gravity
at a sloping seafloor, which occurs when the ocean waves reach waves that are generated by another storm. This third class gener-
shallow water. The corresponding seismic waves have the same ates the strongest noise sources and these can occur anywhere in
period as the ocean gravity waves (Hasselmann 1963). the ocean basin. Obrebski et al. (2012) showed an example of this
The secondary microseismic noise is the biggest peak in the noise third class of noise generation and located a source between Hawaii
spectra. Its generation is associated with the interactions between and California that was recorded by stations several thousands of
ocean gravity waves that have similar periods and are travelling kilometres away.
in opposite directions. We can expect three possible sea-state con- Secondary microseisms are mostly dominated by surface waves,
figurations that result in secondary microseismic noise generation and in particular by Rayleigh waves (Nishida et al. 2008).
(Ardhuin et al. 2011). The first class occurs when a storm has a Theoretical studies of surface wave generation were developed by
wide angular distribution, with ocean gravity waves coming from Miche (1944), Longuet-Higgins (1950) and Hasselmann (1963).


C The Authors 2014. Published by Oxford University Press on behalf of The Royal Astronomical Society. 1
147
Chapter 3. Modeling noise body waves

2 L. Gualtieri et al.

Recently, Gualtieri et al. (2013) demonstrated that the fundamental sources that are derived by beamforming analysis, for observed seis-
mode of Rayleigh waves is sufficient to explain the main features of mograms at the southern California seismic array. To identify the
the noise spectrum amplitude measured on the vertical component. detected waves, we consider the three components and rotate them
Over the past decades, many studies have focused on the location to analyse the beamforming along the P- and SV-components. In
of surface wave sources. Rayleigh wave sources have been found in the Appendix, we show that the body wave site effect can also be
shallow water; that is, close to the coast (Bromirski & Duennebier obtained by using normal-mode theory. In the Appendix, we also
2002; Essen et al. 2003; Schulte-Pelkum et al. 2004; Gerstoft & show that the site effect acts differently on body wave and Rayleigh
Tanimoto 2007; Yang & Ritzwoller 2008), in deep water (Cessaro wave sources.
1994; Stehly et al. 2006; Kedar et al. 2008; Obrebski et al. 2012)
and in both cases (Haubrich & McCamy 1969; Friedrich et al. 1998;
Chevrot et al. 2007). Stutzmann et al. (2012) modelled seismic 2 MODELLING OF THE OCEAN SITE
noise surface waves in various environments and showed that the E F F E C T O N B O D Y WAV E S
strongest noise sources are generated in deep water, whereas coastal
Seismic noise sources are due to non-linear interactions between
reflection generates numerous smaller sources that contribute to the
ocean gravity waves and can be represented as a pressure field that
background noise level.
acts on the ocean surface (Hasselmann 1963). To compute the seis-
In this study, we deal with the noise body wave generation mecha-
mic waves generated by this pressure field, it is possible to use the
nisms in the band of the secondary microseismic period. The origin
elastodynamic representation theorem (Aki & Richards 2002, chap-
of noise body waves is still under debate. Sources of body waves
ter 2) written with the Green’s function satisfying the free surface
have been found mostly by beamforming, which enables the deter-
boundary conditions on the ocean surface. In our case, the sources
mination of both the azimuth and the distance between a seismic
are distributed along the ocean surface, so that only the surface
network and a noise source. Probably, the first body wave source

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

integral term contributes to the expression of the representation
detection by beamforming appears to have been reported by Lacoss
theorem.
et al. (1969) and Haubrich & McCamy (1969). Several studies have
The Green’s function can be decomposed as a sum of plane waves.
demonstrated that a significant amount of P-wave microseismic en-
For body waves in the far field, this sum can be approximated by
ergy is generated far from the coast in deep oceans (e.g. Gerstoft
using the stationary phase method; the corresponding expression is
et al. 2008; Koper et al. 2009, 2010; Landès et al. 2010). Sources
called the ray-theory Green’s function, and this contains only the ray
of body waves have also been associated with specific storms (e.g.
contributions. In this section, we use the plane wave decomposition
Schulte-Pelkum et al. 2004; Gerstoft et al. 2006; Koper & de Foy
of the Green’s function, and compute the ocean site effect on one
2008; Zhang et al. 2010a,b).
selected plane wave. As would be expected, only plane P waves are
Specific phases have been detected by beamforming analysis.
considered in the ocean layer.
Zhang et al. (2009) and Koper et al. (2009, 2010) detected P-wave
We consider a 1-D model with an ocean layer and a homogeneous
sources. Gerstoft et al. (2008) extracted P, PP and PKP sources.
isotropic elastic crust below, which are in welded contact on a plane
They compared the noise sources inferred by beamforming with
boundary (Fig. 1). We denote the velocity and density of the water
ocean wave hindcast data, and they showed that these body wave
layer with subscript ‘w’ and the velocity and density of the crust
microseisms are generated close to storms, where the ocean gravity
with subscript ‘c’.
waves are high. Koper & de Foy (2008) focused instead on body
Let us consider a plane P wave that propagates inside the ocean
wave phases that have interacted with the Earth core: PKP and PcP.
layer from the top—which is the location of the microseismic
A comparison between body wave sources in the primary and
source—to the bottom of the ocean. In the ocean layer, upgoing
secondary microseismic frequency band was reported recently by
P waves are then generated by reflection at the seafloor and down-
Landès et al. (2010). They showed that sources of secondary and pri-
going P waves by reflection at the free surface. At the seafloor, P
mary microseismic P waves do not coincide geographically, which
and S waves are transmitted to the medium below. The angles θ Pw ,
indicates different generation mechanisms of these two microseis-
mic peaks.
Hillers et al. (2012) compared body wave source locations
inferred from beamforming and ocean wave model predictions,
but they used the ocean site effect derived by Longuet-Higgins
(1950) for Rayleigh waves. Differences in the spatial distribution of
Rayleigh and body wave sources were observed by Obrebski et al.
(2013) in the North Atlantic Ocean.
We use plane wave decomposition of the wavefield to study
the body waves generated by the interactions between ocean grav-
ity waves. Previous theoretical studies were developed by Vinnik
(1973), who neglected the amplification effect of the water layer and
Ardhuin & Herbers (2013), who included the water layer and used a
local mode formalism. Here, we consider periods from 3 up to 10 s,
and we demonstrate that the ocean site effect upon the wavefield is
the result of constructive interference of multiply reflected P waves
in the ocean, which are converted to either P waves or to SV waves
at the seafloor. To compute the theoretical noise body wave sources, Figure 1. Cartoon illustrating the seismic rays that propagate from the
we consider the pressure field that acts at the ocean surface because source to the receiver. The reflection of P waves in the ocean layer and the
of the gravity wave interaction, and we modulate this through the transmission/conversion of P and S waves in the crust below are taken into
ocean site effect. These theoretical sources are compared with noise account.

148
Chapter 3. Modeling noise body waves

Ocean site effect on noise body waves 3

θ Pc and θ Sc of the rays shown in Fig. 1 can be computed using Snell’s

law:
sin θ Pw sin θ Sc sin θ Pc
= = = p, (1)
αw βc αc
where p is the ray parameter.
Following Aki & Richards (2002), a plane P wave that propagates
in a homogeneous layer (here the water layer) can be defined by its
potential:
φw (x, t) = e[i(k·x−ωt)] = e[iω( px+qw z−t)] , (2)
where k = (k x , k z ) is the wavenumber vector and |k| = |ω/αw |,
where α w is the compressional wave velocity in the ocean layer.
We denote the spatial coordinates with x = (x, z) and time with t.
Frequency and circular frequency are denoted as f and ω = 2π f,
respectively. The imaginary unit is denoted by ‘i’. The horizontal
and vertical slowness vector components are referred to as (p, qw )
and they are related to the propagation direction (defined by the
sin θ cos θ
angle θ Pw ) by the relation ( p, qw ) = ( αwPw , αwPw ).
We consider first the case of the transmitted P waves. We denote Figure 2. Plane wave reflection and transmission coefficients in terms of
with the index n the plane wave that is n times reflected from the the potentials for the liquid–solid interface, with the incident wave being

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

free surface and n times reflected at the ocean bottom as a P wave a P-wave propagating in the liquid. The P-to-P reflection and transmission
before being transmitted from the seafloor in the medium below as a coefficients are red and blue, respectively. The P-to-S transmission coeffi-
cient is green. The take-off angles are smaller than the P-wave critical angle
P wave. We call the reflection coefficient at the ocean bottom R and
θ P∗w  15.71◦ .
the P-wave transmission coefficient at the seafloor TP . To compute
the P-wave potential in the crust just under the ocean bottom, we
can sum up the contributions of all of the P waves that are reflected The P-to-P transmission coefficient TP at the liquid/solid interface
in the water layer before being transmitted in the crust: is

∞ 
∞
 n 2ρw αc cos θ Pw (1 − 2 p2 βc2 )
φc (x, t) = φc[n] = TP (θ Pw ) −R(θ Pw ) eiw eiω( px+qw h−t) TP (θ Pw ) = , (8)
r1 + r2 + r3
n=0 n=0
in which r1 , r2 and r3 are defined by eq. (7).
= C p (θ Pw , h, ω)eiω( px+qw h−t) , (3) The same approach can be adopted for the P-to-S transmitted
where waves by setting in eq. (3) the P-to-S transmission coefficient TS
given by
TP (θ Pw )
C P (θ Pw , h, ω) = . (4) 
1 + R(θ Pw ) eiw (h,ω,θ Pw ) 4ρw βc2 p cos θ Pw 1 − p2 αc2
TS (θ Pw ) = . (9)
The convergence of the series is guaranteed because |R| < 1. The r1 + r2 + r3
minus sign before the coefficient R in the infinite sum is due to the We note that in our simple 1-D model, the P-to-S transmitted waves
reflection coefficient being (−1) on the free surface. We denote the at the ocean bottom will be SV waves.
phase shift due to the propagation within the water layer as w , In our computation, we use ρ w = 1.0 g cm−3 and α w = 1.5 km s−1
which is defined as for water density and P-wave velocity, respectively and
√
cos θ Pw ρ c = 2.5 g cm−3 , α c = 5.54 km s−1 and βc = αc / 3 = 3.2 km s−1
w (h, ω, θ Pw ) = 2ω h = 2ωqw h, (5)
αw for crust density and P- and S-wave velocities, respectively.
where h is the ocean depth (Fig. 1). In Fig. 2, we show the coefficients R, TP and TS , which are
To obtain the reflection and transmission coefficients in terms given by eqs (6), (8) and (9). The results are shown for take-off
of potentials for a solid/liquid interface, we impose that the normal angles smaller than the P-wave critical angle, as denoted by θ P∗w =
displacement and the normal traction are continuous at the boundary arcsin(αw /αc )  15.71◦ . For take-off angles larger than θ P∗w , the
between the ocean and the crust and that the tangential traction reflection and transmission coefficients become complex. As we
vanishes at the same discontinuity (e.g. Geldart & Sheriff 2004). are not interested in evanescent waves, we consider only take-off
The reflection coefficient R of the P wave at the liquid/solid interface angles smaller than θ P∗w . Here, the coefficient TS is always smaller
is than the coefficient TP in the considered take-off angle range.
The coefficients R, TP and TS do not depend on ω for the
r1 + r2 − r3
R(θ Pw ) = , (6) water/rock discontinuity considered here. If a sediment layer is
r1 + r2 + r3 present between the water and the rock, frequency-dependent re-
in which flection and transmission coefficients can be introduced, which are
obtained by following the approach of Červený (1989). This simple
r1 = ρc αc (1 − 2 p2 βc2 )2 cos θ Pw ,
approach is valid if the thickness of the sediment layer is smaller
  than one half of the wavelength of the signal, which is the case
r2 = 4βc3 p2 ρc 1 − p2 αc2 1 − p2 βc2 cos θ Pw ,
under most of the oceans in the period range considered here.
 We obtain the water-layer site effect upon P and S waves by
r3 = ρw αw 1 − p2 αc2 . (7) summing up the site-effect coefficients for all the different take-off

149
Chapter 3. Modeling noise body waves

4 L. Gualtieri et al.

angles. Considering C P (θ Pw , h, ω) given by eq. (4), for P waves we frequency and ocean depth; more precisely, they vary with the prod-
obtain uct fh. From results shown in Fig. 2, we can see that TS is smaller
 ∗ 1/2 than TP in the whole take-off angle range, which means that cS is
θ Pw
c P (h, ω) = |C p (θ Pw , h, ω)|2 dθ Pw , (10) always smaller than cP .
0

and for S waves, we obtain 3 Q UA N T I TAT I V E S T U D Y O F T H E

 ∗ 1/2 O C E A N S I T E E F F E C T U P O N B O DY
θ Pw
c S (h, ω) = |C S (θ Pw , h, ω)|2 dθ Pw , (11) WAV E S
0
In this section, we investigate the ocean site effect upon body waves
where due to the combined effect of bathymetry h and frequency f. We
TS (θ Pw ) consider the secondary microseismic period band; that is, periods
C S (θ Pw , h, ω) = . (12) from 3 to 10 s. We also vary the ocean depth from 1 to 10 km, to
1 + R(θ Pw )eiw (h,ω,θ Pw )
simulate the bathymetry.
The coefficients in eqs (10) and (11) express the site effect due In Fig. 3, we present the P- and S-wave coefficients as a function
to the water layer upon the seismic wavefield generated by the of the product fh/α w , where f is the frequency, h is the ocean depth
noise sources. These coefficients are real numbers and vary with and α w is the P-wave velocity in the ocean.

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

Figure 3. (a) Ocean site effect of P waves for a fixed take-off angle that corresponds to an epicentral distance of 60◦ for the direct P phase. The colour scale
is related to bathymetry, showing that amplification occurs only at certain ocean depths. (b) P-wave site effect for take-off angles from 0◦ to 15◦ (colours).
We observe that considering different ray parameters, we have small differences between the peak abscissas. (c) P- and S-wave site effects considering the
integration over all of the take-off angles (ocean depth in colour). We observe peaks at close abscissas, meaning that they are related to similar combinations
of depth and frequency.

150
Chapter 3. Modeling noise body waves

Ocean site effect on noise body waves 5

Table 1. Take-off angle θ Pw (fourth column) computed for each seismic phase (first column) in a
given range of epicentral distance (second column). In the third column, we show the corresponding
ray parameters. All of these take-off angles are included in our computation of the ocean site-effect
coefficients cP and cS .
Seismic wave Epicentral distance (deg) Ray parameter p (s km−1 ) Take-off angle θ Pw (deg)
P 30◦ –95◦ 0.041–0.080 3.524–6.85
PP 60◦ –180◦ 0.042–0.080 3.59–6.86
PKP 143◦ –175◦ 0.030–0.040 2.54–3.44
PcP 0◦ –70◦ 0.0009–0.038 0.074–3.27

S 30◦ –95◦ 0.080–0.141 6.72–12.22

SS 60◦ –190◦ 0.083–0.141 7.14–12.22
SKS 110◦ –144◦ 0.019–0.037 1.62–3.20
ScS 0◦ –70◦ 0.002–0.071 0.14–6.11

Fig. 3(a) shows the P-wave coefficient |CP | for the take-off angle 1.9 and 5.7 km. The area of strong site effects are different from
θ Pw = 5◦ , which corresponds to the epicentral distance of about 60◦ those observed at period of 4 s, with much stronger amplification in
for the direct P phase. The ocean depth is marked with different northwest Atlantic Ocean close to Canada and in the Pacific Ocean
colours. It can be seen that only some combinations of frequency close to Japan (Fig. 4). Finally, for a period of 6 s, the maximum site
and ocean depth give strong amplification |CP |. Considering that effect corresponds to depths of 2.3 and 6.8 km (Fig. 3c). Bathymetry

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

the average worldwide ocean depth is around 4 km, we are mostly of 6.8 km is rare in the ocean and therefore mostly ridges and coastal
interested in the first two peaks. areas amplify the noise source (Fig. 4). For that period, only the first
In Fig. 3(b), we show the P-wave coefficient |CP | for take-off peak of Fig. 3(c) is important. The existence of two peaks of high
angles between 0◦ and 15◦ (in colours), with 1◦ steps. We observe amplification explains the strong variability of the ocean site effect
that the abscissas of the first two resonant peaks are relatively similar with period.
for the different take-off angles. Prominent differences occur for the
last two peaks, which correspond to relatively unusual ocean depths
(i.e. greater than 6 km in depth). Different seismic phases are related 4 N O I S E B O D Y WAV E S O U R C E S
to different take-off angles. In Table 1, we show the take-off angles
We compute the theoretical noise body wave sources by consid-
θ Pw (fourth column) for different seismic phases (first column) in
ering the pressure field P(f) due to the interaction of the ocean
a given range of epicentral distances (second column). We observe
gravity waves and by taking into account the site effect computed in
that all of the seismic phases are related to take-off angles smaller
Section 2. The modulus of the pressure field is given by
than θ P∗w . A specific phase can therefore be extracted by considering

the appropriate ray parameter range. |P( f )| = A Fp (K  0, f ), (13)
Fig. 3(c) shows the site-effect coefficients cP and cS , given by
eqs (10) and (11), respectively. Ardhuin & Herbers (2013) obtained where Fp (K  0, f ) is the spectral density of the pressure field at the
a similar result by using local modes in a flat medium. Here, by ocean surface due to the ocean wave–wave interaction (Hasselmann
decomposing the wavefield into a sum of plane P waves, we show 1963; Ardhuin et al. 2011), and A is a normalization constant that
that the ocean site effect is the result of the constructive interferences depends on the sampling parameters used in the ocean wave model.
of P-wave multiples reflected in the ocean. The spectral density Fp (K  0, f ) is given by
In the Appendix, we show that the same result can be obtained  π
by using normal modes and selecting them in relation to the seis- Fp (K  0, f ) = ρw2 g 2 f E 2 ( f /2) M( f /2, θ )M( f /2, θ + π )dθ,
0
mic phase. The computation of the site effect for Rayleigh waves
(14)
was reported by Gualtieri et al. (2013, their fig. 2). In Fig. A1, we
compare the site effect for Rayleigh waves (A1b) and body waves where ρ w is the water density, g is the gravity acceleration, f/2
(A1a) by plotting their amplitude as a function of fh/α w . We ob- is the ocean wave frequency, K is the sum of the wave numbers
serve a strong difference in shape between body wave and Rayleigh of the two opposite ocean gravity waves, E(f) is the sea surface
wave site effects, which means that their most amplified sources are elevation variance and M(f, θ ) is the non-dimensional ocean gravity–
potentially located in different geographical regions. wave energy distribution as a function of the ocean gravity–wave
To determine the oceanic regions that produce the strongest site frequency f and the azimuth θ . To compute the spectral density
effect on body waves, we show maps of the site-effect coefficients Fp (K  0, f ), we use the ocean wave model that was developed by
cP and cS , given by eqs (10) and (11), at different periods. In Fig. 4, Ardhuin et al. (2011). This is a global scale model with a constant
we show maps for P waves (left-hand column) and S waves (right- resolution of 0.5◦ in both latitude and longitude. One key point of
hand column) for three fixed periods: 4, 5 and 6 s. We use different this model is that it is the only model to date that takes into account
colour scales for the P and S waves, because the coefficient for P the coastal reflection of ocean gravity waves. In our computation,
waves is larger than that for S waves. We observe strong site-effect we consider 5 per cent coastal reflection.
variability with period. At 4 s, the maximum site effect occurs for We then compute the P- and S-wave noise sources as products
depths of 1.5 and 4.5 km, which correspond to the maxima of the of the modulus of the pressure field |P(f)| (eq. 13) and the mean
first two peaks in Fig. 3(c). Depths around 4.5 km correspond to site-effect coefficients C̄ P = c P /θ P∗w and C̄ S = c S /θ P∗w for the P and
wide areas of the ocean basins (Fig. 4). For a period of 5 s, the S waves, respectively. The unit of noise body wave sources is then
maxima of the first two peaks (Fig. 3c) correspond to depths of Pa · s.

151
Chapter 3. Modeling noise body waves

6 L. Gualtieri et al.

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

Figure 4. Maps of the ocean site-effect coefficients cP (left-hand column) and cS (right-hand column) as a function of the period: 4 s (first row), 5 s (second
row) and 6 s (third row). The same colour scale is used for the same wave type.

In Fig. 5, we present the example of the Typhoon Ioke in the amplification due to the ocean site effect is strong only at the period
western Pacific. Figs 5(a) and (b) show the significant wave height of T = 5 s along the typhoon track.
and the bathymetry (from Amante & Eakins 2009). Figs 5(c) and To validate our modelling, we analyse the Typhoon Ioke data
(d) show the modelled wave interaction for a period of 5 s, aver- recorded by the Southern California Seismic Network (network
aged over 2 hr on the day of 2006 September 4 (Julian day 247) code CI), and we consider the three-component seismograms. We
and the P-wave ocean site-effect coefficient cP as computed by compute the beamforming power spectrum for an angular frequency
eq. (10). Figs 5(e) and (f) show the corresponding P- and S-wave ω as follows:
theoretical sources. For a typhoon, the noise sources are class I;
that is, generated by the interactions of the ocean waves associated 
Ns
B F(ω, s) = | Si (ω)e−iωs·(xi −xc ) |2 , (15)
with the typhoon. Therefore, the strongest wave interactions are in
i=1
the vicinity of the largest significant wave height. The significant
wave height maximum corresponds to the typhoon location, and where Ns is the number of stations, Si (ω) is the seismogram spec-
its successive positions define the typhoon track. The comparison trum that is recorded at station i, s is the slowness vector towards
of Figs 5(a) and (c) shows that the largest wave interactions oc- the source, xi is the position vector of station i and xc is the posi-
cur along the typhoon track and behind the typhoon. The largest tion vector of the network centre. In Fig. 6(a), we show the beam
wave interactions are in the typhoon tail, and the distance between power spectrum computed from the vertical seismograms over the
the locations of the significant wave height maximum and wave same 2-hr time window as in Fig. 5. We observe a maximum for the
interaction maximum varies with the displacement velocity of the slowness modulus of 0.053 s km−1 and the azimuth of 290.8◦ .
typhoon. Zhang et al. (2010b) used a beamforming approach on To determine which wave type corresponds to the detected slow-
the southern California network data to follow this typhoon track ness, we rotate the vertical, north and east components into the so-
over several days. They computed the beam power from the vertical called P, SV and transverse components. The radial (R) and trans-
component seismograms, and they backprojected the beam maxi- verse (T) components are the horizontal components towards the
mum slowness, under the assumption that the corresponding wave source and perpendicular, respectively. For each slowness vector, s,
is a P wave. They showed good agreement between the typhoon we first rotate the north and east components towards the radial and
track and the backprojected source location. transverse components (Fig. 6b). We then rotate the vertical and
In Fig. 5(b), we show that the bathymetry does not change sig- radial components towards the P and SV-components (Fig. 6c). The
nificantly along the typhoon track. Because of that, at a fixed period angle of rotation is the P-wave theoretical angle of incidence i, which
of T = 5 s, the ocean site effect cP does not vary strongly along the is computed for each slowness s = ||s|| using s = sini/α, where α
typhoon track too (Fig. 5d). From Fig. 4, we also observe that the is the P-wave velocity at the receivers. We take α = 5 km s−1 .

152
Chapter 3. Modeling noise body waves

Ocean site effect on noise body waves 7

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

Figure 5. Typhoon Ioke on 2006 September 4, 12:00–14:00. The typhoon track is plotted with the black line. (a) Significant wave height. (b) Bathymetry
(from Amante & Eakins 2009). (c) Modulus of the pressure field spectrum |P(f)| due to the interaction of the ocean gravity waves. (d) P-wave ocean site effect
cP . (e) Theoretical noise sources for the P waves. The white circle shows the location of the P-wave source detected by beamforming analysis. (f) Theoretical
noise sources for the SV waves. Figs (c), (d), (e) and (f) are computed at a period of 5 s.

Incoming P waves are mostly polarized along the P-component, and plitude is 0.9 relative to the vertical component BF maximum. We
incoming SV waves are mostly polarized along the SV-component. find the P-wave source location by backprojecting the slowness and
For each slowness, we compute the beamforming power spectrum azimuth, and we obtain the source coordinates (27.33◦ N, 153.96◦ E).
(BF) of the P, SV and T components using eq. (15). In Figs 6(d) We observe good agreement with the P-wave source derived from
and (e), we show the P-component and SV-component BF. For the wave model (Fig. 5e). On the SV-component BF (Fig. 6e), we
the P-component, we observed a BF maximum at the slowness observe an extremum at the same slowness and azimuth as on the
0.0529 s km−1 and azimuth of 292.2◦ , which are very close to those P and vertical component BFs. Its amplitude is 0.23 relative to the
values obtained from the vertical component BF. The maximum am- vertical component BF maximum, which is much weaker than the

153
Chapter 3. Modeling noise body waves

8 L. Gualtieri et al.

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

Figure 6. (a) Beamforming power spectrum of the vertical component seismograms. The colour scale is normalized to the vertical component maximum. (b),
(c) Cartoon of the coordinates. The Z-axis is vertical and the R-axis is horizontal towards the network-source azimuth. The P and SV-axes are in the vertical
plane defined by Z and R. The P-axis is towards the theoretical direction of the incident P wave, and the SV-axis is perpendicular to P. The P-component (d),
SV-component (e) and T-component (f) beamforming power spectra. The small black circles indicate the beam maximum. The large black circles and circle
arcs correspond to slownesses of 0.04, 0.08 and 0.14 s km−1 . For a distance range of 30◦ to 90◦ , the P-wave slowness is between 0.04 and 0.08 s km−1 , and the
S-wave slowness is between 0.08 and 0.14 s km−1 .

amplitude detected on the P-component BF. We interpret this signal 5 C O N C LU S I O N

as the result of the P-to-SV conversions under the receivers. From
In this study, we used the plane wave superposition of the wavefield
theory, we would also expect to observe an SV wave generated at
to compute the site effect of the ocean layer upon seismic noise
the same source location as the P wave. This would correspond to
body waves. We considered only seismic waves in the secondary
a beam maximum at the slowness of 0.1 s km−1 along the azimuth
microseismic period band. We demonstrate that the ocean site effect
of 292◦ , but it is not observed on the SV-component BF (Fig. 6e).
can be modelled by considering the constructive interferences of
The relative amplitude of the S wave with respect to the P wave is
multiply reflected P waves in the ocean, which are converted into P
given by
and SV waves at the liquid–solid seafloor interface. We show in the
Appendix that the noise body wave site effect can also be retrieved

exp(−π ts∗ /T ) J p C S αc using normal modes.
A S/P = √ , (16) We computed both the P- and S-wave ocean site effects and
exp(−π t p∗ /T ) Js C P βc
we observe that they have the same dependence with respect to
frequency and bathymetry. Fixing the period, we show site-effect
where subscripts S and P are for the S and P waves, respectively, maps where the amplification patterns are relatively similar for both
∗ P and S waves. However, in terms of amplitude, these maps show
T is the
√ period, t characterizes the seismic attenuation along the
path, J is the geometrical spreading, C is the ocean site effect a stronger site effect for the P waves than for the S waves. We
computed with eqs (4) and (12) and α c and β c are the P- and S- also observe important variations with frequency of the site-effect
wave velocities in the crust, respectively. The ratio α c /β c converts maps for the same seismic phase. The Rayleigh wave site effect
the relative ocean site effect CS /CP from potential to displacement. was recently computed by Gualtieri et al. (2013) using normal
The relative geometrical spreading effect is about 1. For the period modes. Comparing Rayleigh wave and body wave site effects (see
of 5 s, the relative ocean site effect is CS /CP = 0.33 and the relative the Appendix), we observe that the local bathymetry produces a
seismic attenuation effect along the path is 0.17; we obtain AS/P  different effect on these.
0.1. We estimate the SV-component BF relative noise level at 0.15. We validate our modelling by computing noise theoretical sources
Therefore, it is impossible to detect the generated SV-wave signal as the product of the pressure field induced by the interaction of
of relative BF amplitude 0.12 = 0.01 because it is too far below ocean gravity waves and the site-effect coefficient. We compare
the noise level. Finally, we checked that we do not observe any SH- our results with the beamforming analysis results for Typhoon Ioke
wave on the transverse component BF at the theoretical slowness of (2006 September) that was recorded by the Southern California
0.1 s km−1 and the azimuth of 292.2◦ (Fig. 6f). Seismic Network. To identify which waves correspond to the BF

154
Chapter 3. Modeling noise body waves

Ocean site effect on noise body waves 9

detected signals, we rotated the traces to compute the P, SV and Friedrich, A., Krüger, F. & Klinge, K., 1998. Ocean-generated microseismic
transverse component BF. We obtain a good match between the noise located with the Gräfenberg array, J. Seismol., 2(1), 47–64.
theoretical and observed noise source detected on the vertical and Geldart, L.P. & Sheriff, R.E., 2004. Problems in Exploration Seismology
P-component beamforming power spectra, which confirms that it and Their Solutions, Society Of Exploration Geophysicists.
Gerstoft, P. & Tanimoto, T., 2007. A year of microseisms in southern Califor-
is a P wave. For the same slowness and azimuth, we also detect
nia, Geophys. Res. Lett., 34(20), L20304, doi:10.1029/2007GL031091.
a signal on the SV-component BF. We interpret this as the result
Gerstoft, P., Fehler, M.C. & Sabra, K.G., 2006. When Katrina hit California,
of the P-to-S conversions in the crust under the seismic array. On Geophys. Res. Lett., 33(17), L17308, doi:10.1029/2006GL027270.
the SV-component BF, we do not detect any signal at the S-wave Gerstoft, P., Shearer, P.M., Harmon, N. & Zhang, J., 2008. Global P, PP,
slowness corresponding to the same source. We demonstrate that and PKP wave microseisms observed from distant storms, Geophys. Res.
the amplitude of the generated SV wave is too small to be detected Lett., 35(23), L23306, doi:10.1029/2008GL036111.
on the SV-component BF. Gualtieri, L., Stutzmann, E., Capdeville, Y., Ardhuin, F., Schimmel, M.,
The computation of the noise body wave site effect can be used to Mangeney, A & Morelli, A., 2013. Modelling secondary microseismic
investigate the body wave source locations in detail, and to compare noise by normal mode summation, Geophys. J. Int., 193(3), 1732–1745.
body wave and Rayleigh wave noise sources. This can also be Gutenberg, B., 1936. On microseisms, Bull. seism. Soc. Am., 26(2), 111–117.
Hasselmann, K., 1963. A statistical analysis of the generation of micro-
used to investigate the energy levels of noise body waves in the
seisms, Rev. Geophys., 1(2), 177–210.
secondary microseismic period band. Moreover, the introduction of
Haubrich, R.A. & McCamy, K., 1969. Microseisms: coastal and pelagic
a sedimentary layer between the ocean and the crust might affect sources, Rev. Geophys., 7(3), 539–571.
the body wave source amplification, especially close to the coast, Hillers, G., Graham, N., Campillo, M., Kedar, S., Landes, M. & Shapiro, N.,
and its study should be the subject of future studies. 2012. Global oceanic microseism sources as seen by seismic arrays and
predicted by wave action models, Geochem. Geophys. Geosyst., 13(1),
doi:10.1029/2011GC003875.

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

AC K N OW L E D G E M E N T S Kedar, S., Longuet-Higgins, M., Webb, F., Graham, N., Clayton, R. & Jones,
C., 2008. The origin of deep ocean microseisms in the North Atlantic
We acknowledge the support of the QUEST Initial Training Network Ocean, Proc. R. Soc. A, 464(2091), 777–793.
funded within the EU Marie Curie Programme. MS acknowledges Koper, K.D. & de Foy, B., 2008. Seasonal anisotropy in short-period seismic
financial support through the projects Rifsis (CGL 2009-09727) and noise recorded in South Asia, Bull. seism. Soc. Am., 98(6), 3033–3045.
Topolberia (CSD2006-00041). Most of the figures were prepared Koper, K.D., de Foy, B. & Benz, H., 2009. Composition and variation of
using Generic Mapping Tools (GMT; Wessel & Smith 1995). We noise recorded at the Yellowknife Seismic Array, 1991–2007, J. geophys.
thank the two anonymous reviewers for their detailed comments Res., 114(B10), B10310, doi:10.1029/2009JB006307.
that helped us to improve our manuscript. This is IPGP contribution Koper, K.D., Seats, K. & Benz, H., 2010. On the composition of Earth’s
number 3498. short-period seismic noise field, Bull. seism. Soc. Am., 100(2), 606–617.
Lacoss, R.T., Kelly, E.J. & Toksöz, M.N., 1969. Estimation of seismic noise
structure using arrays, Geophysics, 34(1), 21–38.
Landès, M., Hubans, F., Shapiro, N.M., Paul, A. & Campillo, M., 2010.
REFERENCES Origin of deep ocean microseisms by using teleseismic body waves,
Aki, K. & Richards, P.G., 2002. Quantitative Seismology, 2nd edn, Univer- J. geophys. Res., 115(B5), B05302, doi:10.1029/2009JB006918.
sity Science Books. Longuet-Higgins, M.S., 1950. A theory of the origin of microseisms, Phil.
Amante, C. & Eakins, B.W., 2009. ETOPO1 1 Arc-Minute Global Relief Trans. R. Soc. Lond., A, 243(857), 1–35.
Model: procedures, data sources and analysis, NOAA Technical Memo- Miche, A., 1944. Mouvements ondulatoires de la mer en profondeur
randum NESDIS NGDC-24, 1–19. croissante ou décroissante. Première partie. Mouvements ondula-
Ardhuin, F. & Herbers, T.H.C., 2013. Noise generation in the solid Earth, toires périodiques et cylindriques en profondeur constante, Ann. Ponts
oceans and atmosphere, from nonlinear interacting surface gravity waves Chaussées, 114, 42–78.
in finite depth, J. Fluid Mech., 716, 316–348. Nishida, K., Kawakatsu, H., Fukao, Y. & Obara, K., 2008. Background Love
Ardhuin, F., Stutzmann, E., Schimmel, M. & Mangeney, A., 2011. Ocean and Rayleigh waves simultaneously generated at the Pacific Ocean floors,
wave sources of seismic noise, J. geophys. Res., 116(C9), C09004, Geophys. Res. Lett., 35(16), L16307, doi:10.1029/2008GL034753.
doi:10.1029/2011JC006952. Obrebski, M., Ardhuin, F., Stutzmann, E. & Schimmel, M., 2013. Detec-
Berger, J., 2004. Ambient Earth noise: a survey of the global tion of microseismic compressional (P) body waves aided by numerical
seismographic network, J. geophys. Res., 109(B11), B11307, modeling of oceanic noise sources, J. geophys. Res., 118(8), 4312–4324.
doi:10.1029/2004JB003408. Obrebski, M.J., Ardhuin, F., Stutzmann, E. & Schimmel, M., 2012. How
Bromirski, P.D. & Duennebier, F.K., 2002. The near-coastal microseism moderate sea states can generate loud seismic noise in the deep ocean,
spectrum: spatial and temporal wave climate relationships, J. geophys. Geophys. Res. Lett., 39(11), L11601, doi:10.1029/2012GL051896.
Res., 107(B8), 2166–2185. Schulte-Pelkum, V., Earle, P.S. & Vernon, F.L., 2004. Strong directivity
Červený, V, 1989. Synthetic body wave seismograms for laterally varying of ocean-generated seismic noise, Geochem. Geophys. Geosyst., 5(3),
media containing thin transition layers, Geophys. J. R. astr. Soc., 99(2), doi:10.1029/2003GC000520.
331–349. Stehly, L., Campillo, M. & Shapiro, N.M., 2006. A study of the seismic noise
Cessaro, R.K., 1994. Sources of primary and secondary microseisms, Bull. from its long-range correlation properties, J. geophys. Res., 111(B10),
seism. Soc. Am., 84(1), 142–148. B10306, doi:10.1029/2005JB004237.
Chevrot, S., Sylvander, M., Benahmed, S., Ponsolles, C., Lefèvre, J.M. & Stutzmann, E., Roult, G. & Astiz, L., 2000. GEOSCOPE station noise levels,
Paradis, D., 2007. Correction to ‘Source locations of secondary micro- Bull. seism. Soc. Am., 90(3), 690–701.
seisms in western Europe: evidence for both coastal and pelagic sources”, Stutzmann, E., Ardhuin, F., Schimmel, M., Mangeney, A. & Patau, G., 2012.
J. geophys. Res., 112(B12), B12398, doi:10.1029/2007JB005481. Modelling long-term seismic noise in various environments, Geophys. J.
Dziewonski, A.M. & Anderson, D.L., 1981. Preliminary reference Earth Int., 191(2), 707–722.
model, Phys. Earth planet. Inter., 25(4), 297–356. Vinnik, L.P., 1973. Sources of microseismic P waves, Pure appl. Geophys.,
Essen, H.-H., Krüger, F., Dahm, T. & Grevemeyer, I., 2003. On the genera- 103(1), 282–289.
tion of secondary microseisms observed in northern and central Europe, Webb, S.C., 1998. Broadband seismology and noise under the ocean, Rev.
J. geophys. Res., 108(B10), doi:10.1029/2002JB002338. Geophys., 36(1), 105–142.

155
Chapter 3. Modeling noise body waves

10 L. Gualtieri et al.

Wessel, P. & Smith, W.H.F., 1995. New version of the Generic Mapping
Tools released, EOS, Trans. Am. geophys. Un., 76(33), 329.
Yang, Y. & Ritzwoller, M.H., 2008. Characteristics of ambient seismic noise
as a source for surface wave tomography, Geochem. Geophys. Geosyst.,
9(2), doi:10.1029/2007GC001814.
Zhang, J., Gerstoft, P. & Shearer, P.M., 2009. High-frequency P-wave seis-
mic noise driven by ocean winds, Geophys. Res. Lett., 36(9), L09302,
doi:10.1029/2009GL037761.
Zhang, J., Gerstoft, P. & Bromirski, P.D., 2010a. Pelagic and coastal sources
of P-wave microseisms: generation under tropical cyclones, Geophys.
Res. Lett., 37(15), L15301, doi:10.1029/2010GL044288.
Zhang, J., Gerstoft, P. & Shearer, P.M., 2010b. Resolving P-wave travel-
time anomalies using seismic array observations of oceanic storms, Earth
planet. Sci. Lett., 292(3–4), 419–427.
Zhao, L. & Dahlen, A., 1995. Asymptotic normal modes of the Earth—II.
Eigenfunctions, Geophys. J. Int., 121, 1–42.
Zhao, L & Dahlen, F.A., 2007. Asymptotic eigenfrequencies of the Earth’s
normal modes, Geophys. J. R. astr. Soc., 115(3), 729–758.

A P P E N D I X : N O I S E B O D Y WAV E S I T E
EFFECT USING NORMAL MODES

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

Figure A1. (a) Comparison between the results obtained using normal
We develop here a method for retrieving the noise body wave site
modes (dots) and the plane wave superposition approach (solid line) for
effect using normal modes, and we compare the result with the site epicentral distances between 30◦ and 90◦ . The amplitude of cn (normal-
effect computed using plane wave superposition. modes approach) is shown on the left and the amplitude of cP (plane wave
Considering noise sources as vertical point forces at the top of superposition approach) on the right. The ocean depth is shown in colour.
the ocean, Gualtieri et al. (2013) provided the expression of the For computing the P-wave site effect using normal modes, it is necessary to
site-effect coefficients using normal-mode theory: integrate all of the contributions shown in this figure (dots), with each one
multiplied by a phase. Small differences are due to the different earth mod-
n Ul (rr )n Ul (rs )
cn = , (A1) els (PREM model with normal modes and an ocean layer over a half-space
n ωl for the plane wave superposition approach). (b) Rayleigh wave site effect
where n ωl is the angular frequency, and n Ul is the radial eigenfunc- computed using normal modes (modified from Gualtieri et al. 2013).
tion for a mode described by quantum numbers (n, l). The receiver
and source positions are denoted by rr and rs , respectively. univocally identify pairs of quantum numbers (n, l) and we obtain
Body waves are generated by the constructive interaction of high- all of the associated normal modes that contribute to generate this
order overtones. To identify the normal modes that provide a con- seismic phase by constructive interference.
tribution to the generation of a specific seismic phase, we use the We perform our computation for the P waves considering the
ray parameter. In a spherically symmetric earth model, the ray pa- PREM model (Dziewonski & Anderson 1981), in which we vary
rameter is defined as p = rsin i(r)/v(r), where v(r) is the seismic the ocean depth from 1 to 10 km, with discrete steps of 1 km, to
velocity field computed at distance r from the centre of the Earth, simulate bathymetry. We set the period range from 4 to 10 s. We
and i(r) is the ray take-off angle. consider here all of the epicentral distances from 30◦ to 90◦ , which
The ray parameter can be associated with the horizontal compo- correspond to the P-wave ray parameters from p = 0.080 s km−1 to
nent of the wavenumber kx , due to the relation: p = 0.042 s km−1 .
R In Fig. A1(a), we use dots to show the contribution to the co-
p= kx (A2)
n ωl efficients cn due to each mode, computed using eq. (A1). Normal
and the wavenumber kx can be defined by using normal-mode modes that contribute to the P-wave generation are selected using
formalism eq. (A5). We plot the coefficient cn as a function of fh/α w , where f
 is the frequency, h is the ocean depth and α w is the P-wave velocity
l(l + 1) in the ocean. The ocean depth is marked with colours. The actual
kx = (A3)
R body wave site effect can be computed by considering the summa-
in which l is the normal-mode angular order and R is the Earth tion of all cn corresponding to the same fh/α w , taking into account
radius. Combining eqs (A2) and (A3), we obtain the phases.
 In Fig. A1(a), we also show for comparison the site-effect coeffi-
l(l + 1)
p= (A4) cient obtained using the plane wave superposition approach (contin-
n ωl uous line). The scale amplitudes on the left refer to the coefficients
from which we can compute the relationship between the angular computed using normal modes, the scale on the right to the coeffi-
order l and the ray parameter p: cients computed using plane wave superposition. We observe that
the same ocean depth involves the same abscissa, with the same
1 1 shape (dots and line colours). Moreover, the coefficients show three
l=− + + n ωl 2 p 2 . (A5)
2 4 peaks in both cases, for the same combination of ocean depth and
For large l, our expression (A5) recovers eq. (5) of Zhao & Dahlen frequency. Differences in amplitude are due to the normal-mode
(1995) (see also Zhao & Dahlen 2007). Fixing the ray parameter, we normalization, whereas differences in abscissa are related to the

156
Chapter 3. Modeling noise body waves

Ocean site effect on noise body waves 11

different chosen earth models (PREM model with normal modes waves, although as a function of fh/α w , as we did for the body
and an ocean layer over a half-space for the plane wave superposi- waves, instead of ωh/β c , like in Gualtieri et al. (2013). We observe
tion approach). that the shape of the Rayleigh wave site effect (Fig. A1b) is very
By using normal modes, the same approach was used by Gualtieri different from the P-wave site effect (Fig. A1a). Then, fixing the
et al. (2013) to compute the ocean site effect of the Rayleigh waves. period and the ocean depth, the site effect acts differently and the
In this case, each Rayleigh wave mode corresponds to a given strongest sources of body and Rayleigh waves are potentially located
radial order n. In Fig. A1(b), we plot the result for the Rayleigh in different geographical regions.

Downloaded from http://gji.oxfordjournals.org/ by guest on March 8, 2014

157
Chapter 4

The seismic wavefield under the

ocean

4.1 Introduction
In this chapter, we deal with preliminary results about the microseismic wavefield
recorded under the seafloor for varying frequencies and seafloor topographies.
Like any other elastic wave, microseismic waves lose energy both at tectonic or
structural boundaries and along the source-receiver path. It has been shown that
microseisms travelling from the oceanic environment lose a significant amount of
energy crossing the continental margin (McGarr, 1969).
In order to model the seismic wavefield for varying bathymetries, we use the
spectral-element method, briefly described in section 4.2. This numerical method
allows to follow the irregular discontinuities of the Earth, such as the ocean-crust
interface, by adapting the mesh configuration and it handles coupled liquid-solid
regions. Thus, it is the perfect candidate for performing wave propagation in presence
of discontinuities, like the seafloor.
In section 4.3, we describe the model and source set-up. In order to identify the
different seismic body waves which compose the seismic wavefield under the ocean, we
perform a computation at high frequency (section 4.4) by considering a flat seafloor
(section 4.4.1) and a realistic continental slope (section 4.4.2). The same computation
has been extended at lower frequencies, with the purpose of studying the amplitude
of noise Rayleigh waves below a flat seafloor and below a continental slope for varying
frequencies.

159
Chapter 4. The seismic wavefield under the ocean

Finally, in section 4.6, we briefly present the future steps that we are going to
explore in order to model the microseismic wavefield for varying seafloor topographies.

4.2 Short description of the spectral-element method

The spectral element method is a high-order numerical method that allows us to
solve the seismic wave equation in a 2D or 3D Earth model. The method enables
adaptation of the mesh to the irregular seafloor topography and the other seismic
discontinuities of the Earth structure and to the variable wavelength inside the Earth.
Moreover, the spectral-element method yields accurate solutions for surface waves
without increasing the number of grid points per wavelength, therefore overcoming
some deficiencies of the finite-difference method (e.g. Fichtner, 2011).
Originally developed for fluid-dynamics (Patera, 1984; Maday & Patera, 1989),
the spectral-element method has been firstly applied in seismology by Seriani et al.
(1995), Faccioli et al. (1997) and Komatitsch (1997). Very accurate 2D and 3D nu-
merical solutions have been presented for example by Komatitsch & Vilotte (1998),
Komatitsch & Tromp (2000), Komatitsch et al. (2000), Komatitsch & Tromp (2002).
Several codes have been developed in order to solve the seismic wave equations at re-
gional scale (e.g. Cupillard et al., 2012) or at global scale (e.g. Komatitsch & Vilotte,
1998; Tromp et al., 2008). Applications and validations of the seismic wave propa-
gation in fluid layers have been discussed by Komatitsch et al. (2000) and Cristini &
Komatitsch (2012).
The spectral-element method (SEM) is based on a high-order polynomial approxi-
mation of the weak formulation of the wave equations, which enables to naturally take
into account both interfaces and free boundary surface conditions. It combines the
accuracy of the pseudo-spectral method with the flexibility of the second-order finite-
element method (e.g. Fichtner, 2011). In this method, the wavefield is represented in
terms of high-degree Lagrange interpolants, and integrals are computed based upon
Gauss-Lobatto-Legendre (GLL) quadrature. The orthogonal basis is represented by
Lagrange polynomials of order N, associated with (N + 1) interpolation nodes. This
combination leads to a perfectly diagonal mass matrix, which in turns leads to a fully
explicit time scheme that leads itself very well to numerical simulations on parallel
computers. The choice of a Lagrangian interpolation associated with the GLL nodes
gives the SEM a very interesting convergence property: an increase of the polynomial

160
Chapter 4. The seismic wavefield under the ocean

order leads to an exponential diminution of the aliasing error. This property, called
spectral precision, gives its name to the method. The SEM is particularly well suited
to handling complex geometries and interface conditions. This formulation enables to
naturally take into account both interface and free boundary surface conditions, al-
lowing a good resolution of evanescent interface and surface waves. As a consequence,
the accurate simulation of surface wave propagation is straightforward without any
additional cost.
The spatial discretisation process implies the decomposition of the spatial domain
into non-overlapping elements. Classical implementations of the SEM are based on
hexahedral elements. The typical element size that is required to generate an accurate
mesh depends on the smallest wavelength of waves travelling in the model and on the
polynomial degree N. This is because each element of the spatial domain contains a
sub grid of (N + 1)2 in 2D, or (N + 1)3 in 3D, GLL discretization points when the
SEM is used with a polynomial order N. An accurate mesh requires about 5 points per
minimum wavelength (Komatitsch & Vilotte, 1998). Then, the size of the elements
dx is constrained by the shortest wavelength λmin propagated in the medium and by
the chosen polynomial order N (e.g. Cupillard et al., 2012):

N N cmin
dx ≤ λmin ≤ (4.1)
5 5 fmax

where cmin is the minimum wave-speed of the model and fmax is the maximum corner
frequency of the source. Because each element contains a subgrid of GLL discretiza-
tion points, very distorted mesh elements can be used and the mesh is said to honour
discontinuities when the boundaries of the elements coincide with the discontinuities.
To ensure the stability of the time-marching, the time step dt of the finite-
difference scheme has to verify the Courant-Friedrichs-Lewy (CFL) condition (Courant
et al. (1928)):  
∆xmin
dt ≤ C (4.2)
cmax
where C is the Courant constant, usually chosen between 0.3 and 0.4, ∆xmin is
the minimum grid spacing (i.e. distance between two GLL nodes) and cmax is the
maximum wave-speed of the model (e.g. Komatitsch et al., 2005 ).
In regional scale applications, absorbing boundaries are introduced throughout
Perfectly Maching Layers (PML, Berenger, 1994) following Komatitsch et al. (2003),
Basu & Chopra (2004) and Festa & Vilotte (2005) in order to avoid parasitic energy

161
Chapter 4. The seismic wavefield under the ocean

reflections from the sides. For accurate reviews on the SEM, see Komatitsch et al.
(2005) and Chaljub et al. (2007).

4.3 Model and source set-up

In this section, we focus on the description of the temporal and spatial configura-
tions that we adopt in order to perform simulations with the SEM. Furthermore, we
describe the Earth model and the source configuration.
We consider a regional Earth model, with horizontal and vertical dimensions of
100 km and 20 km, respectively. We use the 2D SEM code originally written by
G. Festa, E. Delavaud and J.P. Vilotte (called SPEC2D)1 and then modified by Y.
Capdeville (Capdeville, 2002).
The spatial domain is divided in two layers, a liquid shallowest one, which repre-
sents the ocean layer and a solid deepest one, which is referred to as the shallowest
part of the crust. Absorbing boundaries are introduced to avoid parasitic reflections
from the sides. The upper surface of the liquid layer is a free surface where the
pressure is enforced to zero. The elastic properties of these layers are summarised in
Table 4.1.

Density P-wave velocity S-wave velocity

Liquid layer ρw = 1.0 g/cm3 αw = 1.5 km/s βw = 0 km/s
Solid layer ρc = 3.0 g/cm3 αc = 5.8 km/s βc = 3.2 km/s

Table 4.1: Density and body wave speeds in the liquid and in the solid layer.

In such model, the critical take-off angles in the liquid layer are:

αw
θPPw∗ = arcsin = 15◦ (4.3)
αc
αw
θPS∗w = arcsin = 28◦ (4.4)
βc
and the critical take-off angle in the solid layer is:

βc
θPS∗c = arcsin = 33.5◦ (4.5)
αc
1
The original version of the code by G. Festa, E. Delavaud and J.P. Vilotte can be found at:
http://www.spice-rtn.org/library/software/2DSPEC.html

162
Chapter 4. The seismic wavefield under the ocean

We consider a single source acting at the upper surface of the liquid layer with
the purpose of simulating a noise source. In order to evaluate the seismic wavefield
composition in the solid Earth, we set an array of seismic stations 100 m below the
ocean-crust interface. The inter-station distance is 1 km. In the following, we call
each station with the corresponding horizontal coordinate (e.g. the station number
75 is the station located at x = 75 km). The source is a pressure source located at
the surface of the ocean and above the station number 75. The source time-function
g(t) is a Ricker pulse, i.e. the negative normalized second derivative of a Gaussian
function:
2
g(t) = 1 − (2πf0 (t − t0 ))2 e−(2πf0 (t−t0 ))
 
(4.6)

where f0 is the cut-off frequency and t0 is the source time-shift (from the code manual).
With such a function, the maximum corner frequency fmax of the source, necessary
to evaluate the spatial discretisation as given by equation (4.1), can be defined as

fmax ' 2.75f0 (4.7)

The source time-shift t0 has to be adjusted as a function of f0 to get the complete

source time-function.
We consider Lagrange polynomials of degree 8. The highest frequency used in our
simulations is f0 = 1 Hz, then fmax ' 2.75 Hz. Then, in order to satisfy the spatial
and temporal constraints given by equations (4.1) and (4.2), we chose dx = 500 m
and dt = 0.00025 s. The time duration of the records is 150 s.

4.4 The wavefield composition for varying bathyme-

tries

In order to evaluate the seismic wavefield transmitted under the seafloor when a
source acts just below (1 m) the surface of the ocean, in equation (4.6), we fix the
cut-off frequency of the source at f0 = 1 Hz (i.e. fmax = 2.75 Hz) and the source
time-shift at t0 = 3s. This very high frequency enables to evaluate the contribution
due to the different seismic phases separately. As a preliminary study, we consider
two seafloor configurations: 1) a flat seafloor and 2) a slope, which idealises the
ocean-continent boundary.

163
Chapter 4. The seismic wavefield under the ocean

4.4.1 The flat seafloor

A flat liquid-solid interface is considered, with the ocean thickness fixed at 3 km,
as shown in Figure 4-1. The red star at the surface of the ocean and x = 75 km
represents the seismic source and the blue triangles under the seafloor denote the
seismic stations.

Figure 4-1: Cartoon illustrating the flat seafloor configuration. The seismic source
is located at the surface of the ocean and x = 75 km (red star). The receivers are
located below the liquid-solid interface (blue triangles).

Figure 4-2 displays the kinetic energy associated with the wavefield in the 2D
domain at t = 7 s. The blue letters refer to the seismic phases illustrated on the left
side of the figure. The cartoon on the left side has been adapted from Komatitsch
et al. (2000).
Several seismic phases are visible: (a) the direct and (b) the reflected P-wave in
the fluid, (c) the transmitted P-wave, (d) the P-to-S converted wave, (e) the refracted
S-wave in the solid at the critical take-off angle θPS∗c (equation 4.5), (f) the refracted P-
wave in the fluid at the critical take-off angle θPPw∗ (equation 4.3) and (g) the refracted
P-wave in the fluid at the critical take-off angle θPS∗w (equation 4.4). The refracted
waves are also called head-waves.
It is noted that the kinetic energy associated with the direct P-wave (a) propagates
in the ocean along an arc of a circle. Such an arc is confined between the ocean
surface and the seafloor and, as the waves get further from the source, it becomes
more and more indistinguishable from a vertical segment. Furthermore, the kinetic
energy related to the P-to-S wavefield (d) is null for vertical transmissions since the
S-wave transmission coefficient is equal to zero (see Chapter 3, Figure 3-2). At t = 7
s, the downgoing P-wave (b) has been reflected only once at the seafloor. Since the
atmosphere is not taken into account in our model (free surface boundary condition),
a total reflection of P-waves occurs at the ocean surface for the seismic phases denoted

164
Chapter 4. The seismic wavefield under the ocean

Figure 4-2: Snapshot of the kinetic energy field for the flat-seafloor configuration (red
line) at t = 7 s. The red star at the surface of the ocean and x = 75 km is the source
and the blue triangles under the seafloor are the seismic stations. The blue letters on
the top panel refer to to the seismic phases illustrated on the left. The cartoon on
the left side has been adapted from Komatitsch et al., 2000.

as (b), (f) and (g) in Figure 4-2. As time increases, multiply reflected P-waves fill the
ocean.
Figure 4-3 shows four snapshots of the kinetic energy field: a) t = 10 s, b) t = 20
s, c) t = 30 s and d) t = 40 s. Moving from t = 10 s to t = 40 s, i.e. from Figure
4-3a to Figure 4-3d, an increasing number of multiple P-wave reflections occurs in
the water layer.
The direct P-wave, denoted as (a) in Figure 4-2, can be recognised at x ' 65 km
when t = 10 s, at x ' 50 km when t = 20 s, at x ' 35 km when t = 30 s and at
x ' 20 km when t = 40 s. The refracted P-wave in the water layer, denoted as (f)
in Figure 4-2, and its reflection at the free surface are clearly visible, for example,
between x ' 50 km and x ' 60 km when t = 10 s. It has to be noted that the amount
of kinetic energy which passes through the seafloor decreases with increasing time.
It is observed that, in the solid layer, a large energy amount is confined just below
the seafloor at some given points (e.g. x = 60 km when t = 10 s, x = 20 km for
t = 30 s and many others). These finger-shaped patterns represent the regions where
P-waves and S-waves interfere and surface waves are generated by their constructive
interference. The surface waves are linked to very intense head waves in the water
layer, characterise by a large amount of kinetic energy.
Figure 4-4 displays the seismograms of a) the horizontal and b) the vertical com-
ponents of the velocity field recorded at the line of receivers located below the seafloor

165
Chapter 4. The seismic wavefield under the ocean

Figure 4-3: Caption on the following page.

166
Chapter 4. The seismic wavefield under the ocean

Figure 4-3: Snapshots of the kinetic energy field for the flat-seafloor configuration
(red line) at different times: a) t = 10 s, b) t = 20 s, c) t = 30 s and d) t = 40 s.
The red star at the surface of the ocean and x = 75 km is the source and the blue
triangles under the seafloor are the seismic stations.
167
Chapter 4. The seismic wavefield under the ocean

Figure 4-4: a) Horizontal and b) vertical components of the seismograms recorded at

a line of receivers located 100 m below the seafloor. The inter-station distance is 1
km. All seismograms have been normalised by a factor of 5×10−8 m/s. c) Theoretical
arrival-times at the receivers. We recall that the source is located at the surface of
the ocean and x = 75 km.
168
Chapter 4. The seismic wavefield under the ocean

(blue triangles in Figures 4-2 and 4-3). As the receivers are located in the solid layer,
the direct P-wave (a) is not recorded.
In Figure 4-4c, we present the theoretical arrival times of the direct P-waves, the
multiply reflected P-waves in the water layer and the transmitted P- and S-waves,
as recorded at the line of seismic stations located 100 m below the seafloor. In this
case with the term “transmitted” P- and S-waves, we denote the seismic waves which
leave the source with take-off angles θPPw∗ and θPS∗w respectively and travel horizontally
below the seafloor, as it will be detailed in the following. Due to the closeness of the
receivers to the interface, the theoretical arrival times are computed as if the receivers
were located at the interface. The source time-shift t0 = 3s, as used in equation (4.6),
has been taken into account for computing these theoretical arrival-times.
The direct P-wave – green dots in Figure 4-4c – is recorded when the downgoing
P-wave, travelling inside the ocean layer, impacts the seafloor. Because of the ge-
ometry of the numerical experiment, the direct P-wave is only visible on the vertical
component (Figure 4-4b) around the source region (i.e. around the station number
75). For long offsets the direct P-wave can be recorded only on the horizontal compo-
nent of the seismogram because the seismic ray tends to become horizontal for long
distances. However, the amplitude of the direct P-wave becomes so small that it is
no longer visible.
The multiply reflected P-waves in the water layer, denoted as (b) in Figure 4-2,
are only recorded on the vertical component (yellow dots in Figure 4-4c).
In Figure 4-4c, the transmitted P-waves – (c) in Figure 4-2 – are computed as the
ones leaving the source as P-waves with a critical P-wave take-off angle θPPw∗ , reaching
the seafloor and travelling horizontally with the P-wave velocity of the solid layer at
the interface. They are shown as red dots in Figure 4-4c. We compute the transmitted
S-waves – (d) in Figure 4-2 – as the ones leaving the source as P-waves with a critical
S-wave take-off angle θPS∗w , reaching the seafloor and travelling horizontally with the
S-wave velocity of the solid layer at the interface. They are shown as light blue dots
in Figure 4-4c. The transmitted P-waves are visible on the horizontal component,
only for short offsets, and the transmitted S-waves are slightly visible on the vertical
component.
Let us note that, very close to the transmitted S-waves (light blue dots in Figure
4-4c), there is another type of seismic waves travelling slightly slower, at v ' 3 km/s.
They are recorded on both components (Figures 4-4a and 4-4b) and their arrival

169
Chapter 4. The seismic wavefield under the ocean

time corresponds to the localised finger-shaped pattern of kinetic energy just below
the seafloor (e.g. Figure 4-3d, x =' 25 km). Because of all these reasons, such waves
are interpreted as surface waves.

4.4.2 A realistic continental slope

We now perturb the liquid-solid interface in order to reproduce a realistic slope
occurring at the ocean-continent boundary, as shown in Figure 4-5. The seafloor has

Figure 4-5: Cartoon illustrating the realistic ocean-continental slope configuration.

The seismic source is located at the surface of the ocean and x = 75 km (red star).
The receivers are located below the liquid-solid interface (blue triangles).

been perturbed between x = 0 km and x = 50 km, whereas it is still flat between

x = 50 km and x = 100 km. We simulate the beginning of the continental slope
around x = 50 km. The continental shelf – the underwater landmass which extends
from a continent – goes from 0 km to x ' 10 km. The seismic source is located at
the surface of the ocean, at x = 75 km (red star in Figure 4-5), like in the previous
flat-seafloor case. The receivers follow the interface configuration and they are still
located 100 m below the seafloor (blue triangles in Figure 4-5).
Figure 4-6 shows snapshots of the kinetic energy field for the continental-slope
seafloor configuration at: a) t = 10 s, t = 20 s, t = 30 s and t = 40 s. At t =
10 s (Figure 4-6a), the wavefront has not reached the slope yet, then the energy
equipartition is the same as for the flat seafloor (Figure 4-3a). Increasing time, the
presence of the continental slope around x = 50 km generates a secondary virtual
source, whose wavefield below the seafloor meets the one generated by the source.
The two wavefronts are orthogonal each other and they are both clearly visible at
times t ≥ 20 s.
Both the wavefronts are recorded by the seismic stations located below the seafloor.
Figure 4-7 displays a) the horizontal and b) the vertical components of the velocity

170
Chapter 4. The seismic wavefield under the ocean

Figure 4-6: Caption on the following page.

171
Chapter 4. The seismic wavefield under the ocean

Figure 4-6: Snapshots of the kinetic energy for the ocean-continental slope configu-
ration (red line) at different times: a) t = 10 s, b) t = 20 s, c) t = 30 s and d) t = 40
s. The red star at the surface of the ocean and x = 75 km is the source and the blue
triangles under the seafloor are the seismic stations. A secondary virtual source is
generated around x = 50 km by the continental slope.
172
Chapter 4. The seismic wavefield under the ocean

Figure 4-7: a) Horizontal and b) vertical components of the seismograms recorded

at a line of receivers located 100 m below the seafloor. All seismograms have been
normalised by a factor of 5 × 10−8 m/s. The inter-station distance is 1 km.

173
Chapter 4. The seismic wavefield under the ocean

field recorded at the line of receivers located below the seafloor (blue triangles in
Figures 4-2 and 4-3). Comparing the seismograms recorded below the flat seafloor
and below the continental slope – in Figure 4-7 and Figure 4-4, respectively – we
can evaluate how the slope affects the wavefield. In the presence of the continental
slope, we observe that the seismic stations between x = 0 km and x = 50 km get a
strong signal from the secondary wavefront. The wavefront coming from the source is
weaker in amplitude. Instead, the receivers located between x = 50 km and x = 100
km are influenced by the secondary wavefront only late in time and then the wavefront
related to the source is not affected.

In Figure 4-8 we present the horizontal (left column) and vertical (right column)
components of the seismograms recorded below the seafloor at five given stations.
Blue seismograms refer to the flat liquid-solid interface and the red seismograms to
the continental slope. The seismograms recorded when the seafloor is flat contains
only the wavefield due to the source. Then making a comparison with the seismograms
recorded under the continental slope, we are able to determine the amount of seismic
waves emitted by the secondary virtual source.

The amplitude has the same order of magnitude in both cases. For seismic stations
located at x = 5 km and at x = 30 km – Figure 4-8a and Figure 4-8b – the signal
coming from the actual source is much smaller then the one coming from the virtual
source. The wavefront shape is strongly affected by the continental slope.

The secondary source is located between x ' 40 km and x ' 50 km. At x = 45

km – Figure 4-8c – both the wavefronts are mixed up at the same time.

For seismic stations in the region which remains flat also in presence of the con-
tinental slope, i.e. at x = 60 km and x = 95 km – Figure 4-8c and Figure 4-8d,
respectively – the two waveforms are time-shifted on the seismograms. In the pres-
ence of the continental slope, the arrivals due to the actual sources are unaffected
by the virtual source and the first waveform matches quite well with the flat-seafloor
case. The waveform due to the secondary source arrives later in time and both con-
tributions are clearly visible.

We observe that the station number 60 is at an intermediate distance between the

actual and the virtual source. The amplitude of the waveform related to the virtual
source (later in time on the red seismograms in Figure 4-8d) is smaller than the one
related to the actual source.

174
Chapter 4. The seismic wavefield under the ocean

Figure 4-8: Horizontal (column on the left) and vertical (column on the right) com-
ponents of the seismograms recorded at five receivers located below the seafloor and
at a) x = 5 km, b) x = 30 km, c) x = 45 km, d) x = 60 km and e) x = 95 km. Blue
seismograms are related to the flat liquid-solid interface and the red seismograms to
the continental slope.

175
Chapter 4. The seismic wavefield under the ocean

4.5 The seismic wavefield for varying source fre-

quencies

In this section, we compare the seismograms recorded under the flat seafloor and
under the continental slope for varying source frequencies. The configuration source-
receivers is the same than the one described in the previous section: the source is an
explosion at the surface of the ocean and the receivers are located 100 m below the
seafloor, following the interface topography. The source is a Ricker time function as
given by equation (4.6) in which we set t0 = 40 s in order to get the entire source
time-function.
In the previous section, we have shown that, at a very high frequency (f0 = 1
Hz), it is possible to retrieve the different seismic body waves which compose the
wavefield, separately. At this frequency, interferences between P- and S-waves occur
locally, then surface waves are generated only at given points for a given time (see
section 4.4.1). At lower frequencies, we are not able to distinguish these body wave
seismic phases, since they interfere constructively and they generate Rayleigh waves.
Figure 4-9 shows the particle motion diagrams computed from the seismograms
recorded at the station number 80 for varying source frequencies. The cut-off fre-
quencies of the sources are: f0 = 0.12 Hz (Figure 4-9a), f0 = 0.073 Hz (Figure 4-9b),
f0 = 0.052 Hz ( Figure 4-9c) and f0 = 0.036 Hz (Figure 4-9d). The corresponding
maximum corner frequencies, defined by equation (4.7) are: fmax = 0.33 Hz (Figure
4-9a), fmax = 0.2 Hz (Figure 4-9b), fmax = 0.14 Hz (Figure 4-9c) and fmax = 0.1 Hz
(Figure 4-9d). The seafloor is maintained flat everywhere. Different colours indicate
different times. The particle motion at these frequencies clearly indicates an elliptic
and retrograde motion, the typical pattern of Rayleigh waves.
Figure 4-10 displays the seismograms recorded at five stations located below the
seafloor. Their horizontal coordinates are: a) x = 5 km, b) x = 30 km, c) x = 45
km, d) x = 60 km and e) x = 95 km. The cut-off frequency of the source is f0 = 0.12
Hz (i.e. fmax = 0.33 Hz) for the left column seismograms and f0 = 0.073 Hz (i.e.
fmax = 0.2 Hz) for the right column seismograms.
Figure 4-11 is instead related to a cut-off frequency of f0 = 0.052 Hz (i.e. fmax =
0.14 Hz) for the left column seismograms and f0 = 0.036 Hz (i.e. fmax = 0.1 Hz) Hz
for the right column seismograms. For a given period and receiver, we observe that
the amplitude of the seismograms recorded under the flat seafloor (blue seismogram)

176
Chapter 4. The seismic wavefield under the ocean

Figure 4-9: Particle motion diagrams computed from the seismograms recorded at
the station number 80 for varying periods: a) f0 = 0.12 Hz (i.e. fmax = 0.33 Hz), b)
f0 = 0.073 Hz (i.e. fmax = 0.2 Hz), c) f0 = 0.052 Hz (i.e. fmax = 0.14 Hz) and d)
f0 = 0.036 Hz (i.e. fmax = 0.1 Hz). The colour scale shows successive times. The
typical particle motion of Rayleigh waves (elliptical and retrograd) clearly appears.

has the same order of magnitude as the one recorded under the continental slope (red
seismogram). The effect of the secondary virtual source is still visible for frequencies
higher than f0 = 0.073 Hz (i.e. fmax = 0.2 Hz), especially at stations located far away
with respect to the actual source (Figure 4-10d and Figure 4-10e). The continental
slope has instead a negligible effect for decreasing frequencies.

At f0 = 0.052 Hz (i.e. fmax = 0.14 Hz) – Figure 4-11, left column – only seismo-
grams recorded on the slope (red traces), far away the actual source, present a different
waveform (Figure 4-11a and Figure 4-11b, left column). Seismograms recorded below
a deep ocean basin (Figure 4-11d and Figure 4-11e, left column) show a waveform
comparable in amplitude and shape to the flat-seafloor records (blue traces).

At f0 = 0.036 Hz (i.e. fmax = 0.1 Hz) – Figure 4-11, right column – the wavefield
generated under the flat seafloor (blue traces) and the one generated in presence of
the continental slope (red traces) are comparable both in amplitude and shape and
the secondary virtual source does not produce any effect on the seismograms.

177
Chapter 4. The seismic wavefield under the ocean

Figure 4-10: Vertical component of the seismograms recorded at five receivers located
below the seafloor. Their horizontal coordinates are: a) x = 5 km, b) x = 30 km,
c) x = 45 km, d) x = 60 km and e) x = 95 km. Blue seismograms are related to
the flat liquid-solid interface and the red seismograms to the continental slope. The
cut-off frequency of the source is f0 = 0.12 Hz (i.e. fmax = 0.33 Hz) for the left
column seismograms and f0 = 0.073 Hz (i.e. fmax = 0.2 Hz) for the right column
seismograms. 178
Chapter 4. The seismic wavefield under the ocean

Figure 4-11: Vertical component of the seismograms recorded at five receivers located
below the seafloor. Their horizontal coordinates are: a) x = 5 km, b) x = 30 km,
c) x = 45 km, d) x = 60 km and e) x = 95 km. Blue seismograms are related to
the flat liquid-solid interface and the red seismograms to the continental slope. The
cut-off frequency of the source is f0 = 0.052 Hz (i.e. fmax = 0.14 Hz) for the left
column seismograms and f0 = 0.036 Hz (i.e. fmax = 0.1 Hz) for the right column
seismograms.
179
Chapter 4. The seismic wavefield under the ocean

4.6 Further analysis and next steps

The preliminary results presented in this chapter are based on work in progress
and they have the potential for future far-reaching applications.
In order to evaluate the effect of the continental slope on the amplitude of the
microseismic wavefield under the ocean, we plan to compute the amplitude ratio as
a function of frequency between traces recorded in presence and in absence of the
continental slope as well as in deep and shallow water.
Further realistic continental slopes, for varying steepnesses, can be tested as well
as a real bathymetry can be implemented in the SEM. The effect of small and large
scale variations in depth as well as the effect of the ridges and the abyssal hills on the
seismic wavefield can be also studied.
In Chapter 2 and and Chapter 3, we have seen that the ocean depth under the
source affects the seismic noise amplitude. By varying the ocean thickness and main-
taining a flat seafloor this effect may be evaluated on the seismic wavefield recorded
under the ocean.
Moreover, the SEM can be used to evaluate the seismic wavefield generated by a
noise source in shallow water at lower frequency.

180
Conclusions and future
prospectives

“In the observation of these [...] disturbances which are accessible to every one
[...] we have undoubtedly a fruitful source of study.”

Milne (1883)

Most significant achievements

In this thesis, we have modelled and discussed several physical processes which
affect the seismic noise source location and amplitude and the seismic noise wavefield
propagation.
In Chapter 1, we have shown that the noise sources can be modelled by vertical
forces at the surface of the ocean. The amplitude of the power spectral density of
the pressure field can be derived from a realistic numerical ocean wave model, which
simulates the ocean gravity wave-wave interaction.
Chapter 2 presents the theory for modeling the amplitude of the secondary mi-
croseisms by normal mode summation, which has been successfully used by Gualtieri
et al. (2013) for modeling the three components of noise spectra. In this study, select-
ing three stations in three different environments, we have compared the amplitude of
the vertical component of the noise spectra with the one computed by normal mode
summation. In order to do that, we have derived an attenuation model than enables
to fit well the vertical component of noise spectra whatever the station location. A
good fit both in amplitude and shape has been obtained, reproducing all the fre-
quency content of the secondary microseismic noise peak. It has been shown that the
fundamental mode of Rayleigh waves is the dominant signal in seismic noise records
and it is sufficient alone to explain the main features of the noise spectrum.

181
Conclusions and future prospectives

In order to quantify the influence of the noise source located in various oceans
and recorded by a single station, we have performed the modeling of the spectrum
recorded by the Geoscope station TAM by considering the contributions of the dif-
ferent portions of the ocean separately. We confirm that this station records noise
sources from all surrounding oceans. We show that the Northern Atlantic Ocean re-
mains the main noise sources area all the time during a year. Nevertheless, we show
that the Mediterranean Sea also can contribute significantly to the short period noise
in winter.
The first modeling of the secondary microseismic noise amplitude on the horizontal
components and an estimation of the amount of noise Love wave energy is presented.
Because a vertical point force applied on a locally flat seafloor only generates P- and
SV-waves, Love waves are not excited. We use the discrepancy between the real and
synthetic spectrum of the horizontal components to estimate the amount of missing
Love waves in our synthetic spectra, which is in agreement with previous studies (e.g.
Nishida et al., 2008).
Bathymetry is an important parameter because it modulates the shape and the
amplitude of noise sources, as well as the noise wavefield below the seafloor. First,
the ocean site effect on noise Rayleigh waves has been computed by using normal
modes in a simple two-layer model and it has been shown that it varies both with
the ocean depth and the seismic frequency, amplifying sources close to the coast at
short period (T ≤ 5 s) and sources in deep basin regions at long period (T > 5
s). Each overtone is differently amplified with varying ocean depths and frequencies.
Our ocean site effect is in agreement with the analytical computation performed
by Longuet-Higgins (1950). The computation of the ocean site effect has been also
carried out in a realistic Earth model. It has been shown that the two-layer model
overpredicts or underpredicts the fundamental mode ocean site effect depending on
the oceanic regions.
Adding a thin sediment layer between the ocean and the crust, the site effect on
Rayleigh waves shows a stronger amplification especially for shallow ocean depths
with respect to a model without sediments.
The ocean depth, the crust layering and the sediment layer below the seafloor con-
tribute to the amount of recorded noise Rayleigh waves and affect their propagation
velocity in the solid Earth.
The secondary microseismic noise signal from T = 3 s and T = 10 s is dominated

182
Conclusions and future prospectives

by Rayleigh waves. Their eigenfunctions have a non-null energy both in the ocean
layer and in the crust. Normal modes associated with Scholte waves, have been found
only at shorter period (T . 2.6 s).
Chapter 3 has been dedicated to noise body waves. The wavefield below the
ocean has been decomposed in plane waves in order to compute the ocean site effect
on seismic noise body waves (Gualtieri et al., 2014). It has been shown that the
ocean site effect can be modelled by considering the constructive interferences of
multiply reflected P-waves in the ocean, which are converted into P and SV-waves
at the liquid-solid seafloor interface. The ocean site effects on P- and S-waves have
the same dependence with respect to frequency and bathymetry, but, in terms of
amplitude, P-waves are more strongly amplified than S-waves. Fixing the period, we
build maps of the site-effect, which show relatively similar patterns for both P and
S-waves. The site effect on P, S, PKP and SKS waves shows that these waves are
differently amplified with respect to the epicentral distance.
We validate our modeling of noise body wave generation by computing noise
theoretical sources as the product of the pressure field induced by the interaction
of ocean gravity waves and the site-effect on body waves (Gualtieri et al., 2014).
Theoretical P- and SV-wave sources have the same location. These theoretical results
have been compared with the beamforming analysis results for the Typhoon Ioke
(2006 September) that was recorded by the Southern California Seismic Network. The
good match between the theoretical and the observed noise sources on the vertical
component of beamforming analysis allows to say that we correctly detect noise P-
wave sources. We also demonstrate that the amplitude of the generated SV-wave is
too small to be detected.
The comparison between the site effect on Rayleigh and body waves allows to
say that short period (T ≤ 5 s) Rayleigh waves are mostly amplified near to the
coast, whereas short period noise body waves are mostly amplified in the deep ocean
regions. At longer period (T > 5 s), the maxima of the site effect on Rayleigh and
body waves are located at the same oceanic regions. Noise body wave site effect can
also be retrieved by normal modes (Gualtieri et al., 2014, their Appendix A) and
the imprint of Rayleigh wave fundamental and higher order normal modes can be
extracted by considering the constructive interference of P- and SV-waves.
In Chapter 4, the effect of the bathymetry on the seismic wavefield produced by
a single source located at the surface of the ocean has been presented. A secondary

183
Conclusions and future prospectives

virtual source is generated by the seafloor topography at the beginning of a realistic

ocean-continental slope. The amplitude and the shape of the seismograms recorded
below the seafloor are affected by this secondary virtual source, that is by the ocean-
continental slope, only at short period.

Ideas for future investigations

Gualtieri et al. (2013) have shown that the amplitude of the site effect on Rayleigh
waves is slightly different by setting the seismic station at the seafloor and on land.
The fundamental mode of Rayleigh waves as well as the overtones are differently am-
plified (their Figure 4). This little difference does not strongly affect the computation
of the synthetic seismogram amplitude, especially for stations located far away with
respect to the source regions. Our computation of synthetic noise spectra, carried
out by normal mode summation, does not allow to compute synthetic seismogram
considering stations on land, when an oceanic Earth model is adopted. The receiver
position is set at the seafloor by default in order to use the same Earth model for
the source, the receiver and the propagation. Assuming different Earth models for
the source and the receiver, we might not be able to associate the same mode with
the same frequency, for the receiver and the source (see section 2.8). The farther the
station is with respect to the sources, the better this approximation works, because
the propagation effects dominate with respect to the station position effect. In order
to go further, the local mode approach can be adopted for analyzing the gradual
evolution of the surface wavefield from one type of structure to another (e.g. Maupin,
2007). In such way, a continental model can be used for the receiver and an oceanic
one for the source.
Gualtieri et al. (2013) have also presented an empirical attenuation model, which
allows to fit noise data at periods compatible with the secondary microseisms. This
model has been calibrated empirically adjusting the reference attenuation model QL6
(Durek & Ekström, 1996) for lithospheric depths (≤ 100 km depth), but the non-null
eigenfunction displacement involves smaller depth for periods from 4 s to 10 s. A
future improvement may be the evaluation of the thickness that has to be changed in
the attenuation model in order to maintain the same spectra amplitude.
The site effect both on Rayleigh (Chapter 2) and on body waves (Chapter 3)
is strongly affected by the presence of the sediment layer below the seafloor. A

184
Conclusions and future prospectives

preliminary study about Rayleigh wave site effect in presence of seafloor sediments
has been shown in Chapter 2 (section 2.6) for a thin sediment layer. Taking into
account the strong variability of the seafloor thickness, especially for coastal areas
(see Figure 2-12), it could be possible to estimate how they affect the seismic noise
amplitude. Moreover, by introducing frequency dependent reflection and transmission
coefficients in the theory shown in Chapter 3, the effect of the sediment layer on noise
body waves, can be evaluated. These coefficients can be computed by using the
approach developed by Červený (1989).
The ocean site effect on Rayleigh waves has been estimated by using normal
modes (Chapter 2), whereas the ocean site effect on body waves has been found out
by decomposing the wavefield in plane waves (Chapter 3). Gualtieri et al. (2014) in
their Appendix A have shown how to retrieve the ocean site effect on body waves
by using normal modes. In Chapter 3, it has been demonstrated that the imprints
of Rayleigh waves exist on the site effect of body waves computed by plane wave
decomposition. The plane wave decomposition could be used to compute the ocean
site effect on both Rayleigh and body waves in order to compare their amplitude as
well as the location of the strongest amplification regions. The ocean site effect on
Rayleigh waves can be obtained by finding out the poles or nodes of the ocean site
effect on body waves and applying the method of residues (Longuet-Higgins, 1950).
Noise Love wave generation is one of the biggest open problems concerning the
origin of seismic noise. Nowadays, an explanation for Love-wave generation has not
been provided yet. Different mechanisms have been proposed, like the generation in
shallow water, near to the coast, where ocean gravity waves can interact with the
ocean seafloor (Darbyshire, 1954; Haubrich & McCamy, 1969; Lacoss et al., 1969;
Rind & Down, 1979; Friedrich et al., 1998; Koper et al., 2009; Koper et al., 2010) and
the generation due to the bathymetry (Rind & Down, 1979; Nishida et al., 2008). In
order to investigate which mechanism dominates, one can use the 3D spectral-element
method simulating the bathymetry and the ocean-continental slope – as it has been
done in Chapter 4 with the 2D approach – looking at the transverse component of
the seismograms.
Finally, the normal mode summation approach, as presented in Chapter 2, can
be used to model the long period noise, like the primary microseisms and the hum,
considering appropriated sources in the ocean.

185
Annex I
Finite difference P-wave
tomography of the Italian region:
an additional work

The following work on P-wave seismic tomography of the Italian region is not part
of my PhD research project and it is an extension of my Master thesis project. This
study has no link with the subject of this thesis, which is the modeling of microseismic
noise generation and propagation. Nonetheless, since it was finalized during my PhD
time, I decided to include it in this thesis to document this piece of work which led
to a scientific publication in Geochemistry, Geophysics, Geosystems.
Seismic tomography is our most powerful tool to investigate the internal structure
of our planet. Conceptually similar to medical imaging based on ultrasound record-
ings, seismic tomography translates recordings of seismic waves into 3D models of the
subsurface at scales ranging from few metres to thousands of kilometres.
The goal of the following paper is to improve the knowledge of the crustal and
uppermost mantle structure (i.e. the first 50 km) under the Italian region through
a nonlinear iterative travel time tomography. The forward problem has been solved
using an accurate finite-difference travel time calculation scheme and adopting a 3D
reliable initial Earth model, without explicitly imposing any discontinuity. Our result-
ing model consists of an image of P-wave velocity with a very high spatial resolution.
The Moho interface has been retrieved in the final velocity model as the locus of sharp
increase of seismic velocity, the transition from the crust to the mantle.
The main velocity heterogeneities recovered by this work can be related to geo-
logical features on the shallower part of the crust, and to geophysical processes in

187
Annex I - P-wave tomography of the Italian region: an additional work

the deeper one. Our resulting model shows two arcs of relatively low velocity in the
crust running along both the Alps and the Apennines, underlying the collision belts
between plates. Beneath the Western Alps we detect the presence of the Ivrea body,
denoted by a strong high P wave velocity anomaly. We find a complex lithospheric
structure characterised by shallower Moho close by the Tyrrhenian Sea, intermediate
depth along the Adriatic coast, and deepest Moho under the two mountain belts.

188
Annex I - P-wave tomography of the Italian region: an additional work

189
Annex I - P-wave tomography of the Italian region: an additional work

Article
Volume 15, Number 00
00 Month 2014
doi: 10.1002/2013GC004988
ISSN: 1525-2027

Finite-difference P wave travel time seismic

tomography of the crust and uppermost mantle in
the Italian region
L. Gualtieri
Institut de Physique du Globe de Paris, 1 Rue Jussieu, FR-75005 Paris CEDEX 05, France

Dipartimento di Fisica e Astronomia, Settore di Geofisica, Universita di Bologna, Bologna, Italy

([email protected])

P. Serretti and A. Morelli

Istituto Nazionale di Geofisica e Vulcanologia, Bologna, Italy

[1] We present a 3-D P wave velocity model of the crust and shallowest mantle under the Italian region, that
includes a revised Moho depth map, obtained by regional seismic travel time tomography. We invert 191,850
Pn and Pg wave arrival times from 6850 earthquakes that occurred within the region from 1988 to 2007,
recorded by 264 permanent seismic stations. We adopt a high-resolution linear B-spline model representation,
with 0.1 horizontal and 2 km vertical grid spacing, and an accurate finite-difference forward calculation
scheme. Our nonlinear iterative inversion process uses the recent European reference 3-D crustal model
EPcrust as a priori information. Our resulting model shows two arcs of relatively low velocity in the crust
running along both the Alps and the Apennines, underlying the collision belts between plates. Beneath the
Western Alps we detect the presence of the Ivrea body, denoted by a strong high P wave velocity anomaly.
We also map the Moho discontinuity resulting from the inversion, imaged as the relatively sharp transition
between crust and mantle, where P wave velocity steps up to values larger than 8 km/s. This simple condition
yields an image quite in agreement with previous studies that use explicit representations for the discontinuity.
We find a complex lithospheric structure characterized by shallower Moho close by the Tyrrhenian Sea,
intermediate depth along the Adriatic coast, and deepest Moho under the two mountain belts.

Components: 12,888 words, 15 figures, 1 table.

Keywords: seismic tomography; body waves; computational seismology; Moho topography; Italy.
Index Terms: 7270 Tomography: Seismology; 7203 Body waves: Seismology; 7290 Computational seismology: Seis-
mology; 6982 Tomography and imaging: Radio Science; 8180 Tomography: Tectonophysics.
Received 8 August 2013; Revised 15 November 2013; Accepted 17 November 2013; Published 00 Month 2014.

Gualtieri, L., P. Serretti, and A. Morelli (2014), Finite-difference P wave travel time seismic tomography of the crust and
uppermost mantle in the Italian region, Geochem. Geophys. Geosyst., 15, doi:10.1002/2013GC004988.

1. Introduction region. The two main Italian mountain chains—

the Alps and the Apennines, Figure 1—are due to
[2] The Italian peninsula plays a fundamental role complex processes generated by large-scale con-
in the geodynamic evolution of the Mediterranean vergence between the African and Eurasian plates,

© 2013. American Geophysical Union. All Rights Reserved. 1

190
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

presence of a lithospheric slab below 100 km, but

fail to resolve it unambiguously in its shallow
extent. Regional tomographic studies, such as
Mele et al. [1998], Di Stefano et al. [1999], Piro-
mallo and Morelli [2003], and Di Stefano et al.
[2009], have also provided information about the
lithosphere-asthenosphere system, and have shown
the presence of both fast and slow anomalies
beneath the Italian region.
[4] The structure of the crust is not often retrieved
in these studies, and models of its constitution are
frequently used only as a fixed constraint. At this
scale, comprehensive information about thickness,
density, and seismic velocities in the crust is still
rather poor, or missing spatial continuity. Studies
about crustal thickness, or Moho structure, have
Figure 1. Map of the Italian peninsula in the central Medi- been conducted using receiver functions [e.g.,
terranean topographic context. In black we mark the topo- Piana Agostinetti et al., 2002; Mele et al., 2003;
nyms and in red the tectonic plates. Only the region inside the Piana Agostinetti and Amato, 2009; Di Stefano
black rectangle is taken into account during the computation et al., 2011], measuring crustal thickness beneath
and it denotes the studied region. several seismic stations. Piana Agostinetti et al.
[2008] and Bianchi et al. [2010] have found a dip-
ping interface that marks the top of the Apennines
since about 65 My ago [e.g., Malinverno and slab at 40–80 km. Such determinations, however,
Ryan, 1986; Faccenna et al., 2004]. These proc- represent Moho depth at single points, and do not
esses have pushed lithospheric material beneath provide a laterally continuous model unless
the Alps, Apennines, and the Calabrian Arc. Sub- through interpolation across wide geographical
stantial agreement exists over the formation of the gaps [Piana Agostinetti and Amato, 2009].
Alps as a consequence of the closure and ensuing
subduction of the Alpine Tethyan domain [e.g., [5] The aim of this paper is improve the knowl-
Schmid et al., 2005; Handy et al., 2010]. Several edge of the crustal structure of Italy through seis-
competing models have been proposed to explain mic travel time tomography, for which we focus
the formation of the Apennines that have either our attention to the shallowest 50 km. In our study
been related to westward directed subduction since we use travel times of P waves from earthquakes
the Cretaceous, or thought to have developed located inside the model region, retrieved from the
along the retrobelt of the south-westward prolon- EHB seismic catalog [International Seismological
gation of the Alps [Carminati and Doglioni, Centre (ISC), 2009]. As we follow a nonlinear iter-
2012]. Benoit et al. [2011] suggest instead a ative inversion process, with full 3-D finite-differ-
delamination scenario for the northern part of the ence forward travel time calculation, a reliable
Apennines and the presence of a slab foundering initial model is needed. For this reason, we opt to
at its northern terminus. use the recent 3-D European reference model
EPcrust [Molinari and Morelli, 2011] as our a pri-
[3] In recent years, several seismological studies ori. Our model consists of an image of P wave
have revealed with increasing detail the seismic velocity interpolated on a 3-D grid of nodes, where
velocity structure of the mantle inverting both we can identify the surface of the Moho as the
regional and teleseismic data in this crucial part of locus of sharp increase of seismic velocity. In the
the Mediterranean. Among the most striking fea- following, we describe our data and methods,
tures, lithospheric slabs subducting into the upper show results of recovery tests and inversion of real
mantle have been mapped as high vP and vS bodies data, and finally provide a discussion of results.
under the Apennines [e.g., Amato et al., 1993;
Lucente et al., 1999; Cimini and De Gori, 2001;
Piromallo and Morelli, 2003; Montuori et al., 2. Data and Method
2007] and the Alps [e.g., Piromallo and Morelli,
2003; Lippitsch et al., 2003; Kissling et al., 2006; [6] We consider P wave travel times retrieved
Wagner et al., 2012]. These studies reveal the from the EHB Bulletin of the International
2
191
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

Figure 2. (a) Seismic stations. The color scale shows the number of recorded events. (b) Earthquakes epi-
centres. The color scale shows the hypocentral depth. Only the region inside the black rectangle is taken into
account during the computation and it denotes the studied region.

Seismological Centre [ISC, 2009] for the time [7] Figure 3 shows the path distribution as a func-
period between 1988 and 2007. The well-known tion of epicentral distance (top) and path density
EHB Bulletin is a ISC-based earthquake catalog as a function of depth (bottom), as calculated in
reprocessed and updated by Engdahl et al. [1998]. our a priori model, following the technique that
The processing of Engdahl et al. [1998] composed we describe further on. Deeper than 50 km, the ray
of an iterative relocation with dynamic phase iden- coverage becomes insufficient to provide good
tification in a 1-D Earth reference model [ak135, resolution. The picture shows peaks of seismic ray
Kennett et al., 1995]. We select both earthquakes density just below the average depth of the main
and seismographic stations falling inside our study discontinuities of the velocity field. In particular,
area, covering the whole Italian peninsula, stretch- we observe an increase of the average seismic ray
ing 1300 km in latitude and 700 km in longitude, density at the depth of 4 km, where a discontinuity
and extending from the Alps to the Calabrian Arc between sedimentary layer and crystalline upper
and Sicily (Figure 1). We end up with 264 stations crust is generally present, and at the depth of 14–
(see Figure 2a). The Italian region is well covered 16 km, corresponding to upper/lower crust transi-
by seismic stations, although we may note a gap in tion. The main peak at about 36–38 km depth is
the distribution in the Po Plain. We select only where Pn waves concentrate. The Moho region is
earthquakes having a resulting station coverage therefore very well sampled and we may expect to
with a secondary azimuthal gap smaller than 180 . be able to map structure in its vicinity with best
Finally, we reject stations with fewer than five accuracy.
records. The total number of events chosen—and
then used during the inversion step—is 6850, with [8] We calculate first-arrival travel times using a
191,850 total seismic rays, resulting in 28 stations finite-difference method based on numerical solu-
recording the same event on average. According tion on a 3-D grid of the eikonal equation and
with Figure 2a, about 69% of the stations record propagation of the first-arrival wavefront [Vidale,
less than 500 events. We consider the first arrivals, 1988, 1990; Podvin and Lecompte, 1991]. We
corresponding both to Pn and Pg phases. The data specifically use the algorithm proposed by Podvin
set provides good coverage down to about 50 km and Lecompte [1991] that is able to account for
depth. the existence of different wave propagation
3
192
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

depends on the grid step size. Following the

numerical tests shown by Serretti and Morelli
[2011], and favoring precision with respect to
computational cost, we choose a uniform step size
for our grid of 1 km in each direction. The limited
regional range of the model makes the calculation
quite manageable. This computational grid is used
for the finite-difference discretization, needed to
solve the forward problem, and is independent
from the model description—that can be coarser
for economy in the inversion stage—as illustrated
further on. The slowness to be assigned to each
computational element is defined by a linear B-
spline interpolation within the model grid. For
convenience, we rotate the geographical reference
frame to a new system, with a new North aligned
along the main axis of the box. The computational
grid size is thus 1300 km 3 700 km 3 300 km
along rotated latitude, longitude, and depth (see
Figure 2). We take into account Earth curvature by
applying the Earth flattening approximation, and
we use an equidistant azimuthal projection to map
the Cartesian grid to the sphere [Serretti and Mor-
elli, 2011]. Advantages of the use of the numerical
solution of the eikonal equation, over approximate
ray tracing techniques, are particularly evident in
the presence of strong velocity gradients—such as
Figure 3. (a) Ray frequency as a function of epicentral dis-
tance. (b) Corresponding distribution of average seismic ray
in crust and uppermost mantle environments—
density as a function of depth. Density is calculated as an where it handles the nonlinear dependence of the
average of the total ray length within model blocks at a given travel time on the velocity field in full.
depth. Coverage is best down to about 50 km and it has a
peak at 30–40 km depth, corresponding to the depth range [10] The seismic velocity model is represented by
just beneath the Moho discontinuity, due to wide diffusion of a grid of nodes, where actual P wave speed is
Pn phases. specified, and a linear B-spline interpolation is
used to give velocity at any point within each cell.
The model grid step is 0.1 in both latitude and
modes, in its updated version given by Tryggvason longitude, and 2 km in the vertical direction. The
and Bergman [2006] to guarantee reciprocity of model thus consists of 231 3 151 3 301 nodes,
travel time for stations located close to an anomaly for a total of 10,499,181 model parameters. We
region. Upon retrieval of the whole travel time chose this parameterization in order to represent
field, the raypath may be back-tracked following small heterogeneities as well as sharp velocity gra-
the travel time gradient from station to hypocenter dients—such as the transition from crust to mantle.
to find the seismic raypath. By making use of the Note that this parametric continuous representa-
reciprocity property of ray-theoretical travel times, tion of the 3-D velocity field (the tomographic
for each station we calculate the travel time field model) needs to be discretized by the finite-
originating at each station. The raypath, for each difference computational grid, as described above.
station-event pair, is then calculated following the The thin vertical spacing of the model parameter-
gradient of the wavefront. Following the approach ization grid is necessary to work with a structural
detailed by Serretti and Morelli [2011], we com- model without an explicit discontinuity, and con-
pute the gradient to wave fronts by finite differen- sisting of a 3-D continuous wave speed field. We
ces [e.g., Lapidus and Pinder, 1982]. do need a fine vertical spacing both to be able to
[9] This method requires discretization of the locate the Moho with sufficient precision, and to
velocity field in a 3-D Cartesian grid with cubic keep the discontinuity—actually, a strong gradient
cells with constant velocity (or slowness). Obvi- in our model (see section 4.1)—sharp enough.
ously, the precision of travel times and raypaths Using an adequate inversion approach, as shown
4
193
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

before the origin time relocation have a systematic

negative shift meaning that, for this particular
region, as average, seismic rays have a tendency to
travel slower in our a priori model with respect to
the 1-D model used for relocating events in the
EHB Bulletin.
[13] Our aim is then to find a model, m, able to
predict observed travel times, dobs, within uncer-
tainties given by a data covariance matrix, CD,
given that we are able to compute travel times in a
candidate model mn as: dn 5gðmn Þ: Assuming a
Gaussian travel time uncertainty, we can compute
Figure 4. Histograms of travel time residuals at different the solution to our problem through a linearized
stages of the inversion. Blue: initial residuals; red: residuals iterative inversion by which, at each step, we eval-
after the preliminary relocation; black: final residuals after uate the next estimate of the model mn11 by an
model inversion. The mean value drops from 20.812 s to
update to the current model, mn, as:
3.58 3 1023 s after the origin time relocation and it remains
stable in the final model. The RMS reduces from 1.63 s (blue

21 21
histogram) to 1.22 s (black histogram). After the origin time mn11 5mn 2 GTn C21D Gn 1CM
 T 21  (1)
relocation the RMS is 1.34 s (red histogram). Gn CD ðdn 2dobs Þ1C21
M ðmn 2mprior Þ

where mprior is our a priori model, known with

further on, the number of parameters used to repre- its covariance matrix CM ; dobs are the observed
sent the model does not pose critical issues [for a travel times, known with their covariance matrix
review, see e.g., Tarantola, 2005]. CD ; Gn is the matrix of partial derivatives
approximating gðmÞ ’ Gn m near mn [Tarantola,
[11] We use the recent European plate 3-D refer- 2005].
ence crustal model EPcrust [Molinari and Morelli,
2011] as our a priori information. EPcrust integra- [14] We simplify, assuming that:
tes local information from a variety of previous
high-resolution studies and provides seismic 2


C21 2 2 2 T
M 5  =rD I1k =rD L L (2)
speeds (P and S) and density in variable-thickness
sedimentary and crystalline layers, with a sharp where L is the finite-difference expression of the
Moho. We use seismic wave speeds in the under- Laplacian operator, so that parameters  and k, in
lying mantle from sp6 [Morelli and Dziewonski, turn, control vicinity of the model to the a priori
1993]. one, and its smoothness [Serretti and Morelli,
[12] To locate the sources and perform calcula- 2011]. We assume the error on measurements uni-
tions, we keep into account the real elevation of form and uncorrelated. Because of that the covari-
seismic stations and apply corrections for elliptic- ance matrix CD is a diagonal matrix and can be
ity of the Earth. We perform a preliminary reloca- written as CD 5r2D I.
tion of all seismic events in this model. For each
[15] We use the LSQR algorithm [Paige and Saun-
source, we calculate travel times to all seismo-
ders, 1982a, 1982b; Nolet, 1987; Papazachos and
graphic stations using the method illustrated
Nolet, 1997] to solve the linear systems as needed
above, and then correct the origin time to obtain a
by equation (1).
zero average of all the time residuals. This origin
time-relocation step is needed to discard a possible [16] We choose to stop the iterative process after
bias between the global 1-D Earth model in which six iterations, when the data residuals have a varia-
the data set has been located by Engdahl et al. tion by less than 1% with respect to the previous
[1998], and our regional 3-D crustal model. Figure iteration. The ‘‘restraining’’ () and ‘‘smoothing’’
4 shows a frequency histogram of seismic travel (k) parameters of the a priori model covariance
time residuals before (blue curve) and after (red matrix, chosen at each iteration are shown in Table
curve) this origin time relocation. The mean value 1. Notice that, as usually done in iterative optimi-
drops from 20.812 to 3.58 3 1023 s, whereas the zation procedures [e.g., Gill et al., 1982], we pro-
RMS from 1.63 s reduces to 1.34 s. We observe ceed with shorter steps during the first iterations,
that the seismic travel time residuals computed and release this condition only later on in the
5
194
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

Table 1. Restraining and Smoothing Factors Chosen at Each stations as in the real case, and the same inversion
Iteration Based on Trade-Off Curves procedure. Comparison between resulting and
Iteration Number Restraining Factor  Smoothing Factor k input models provides information on the ability
and limitations of the inversion scheme to recon-
1 400 300
2 300 300 struct real structure, and identifies regions with rel-
3 250 200 atively better and worse performance. Such
4 200 200 experiments are most often performed using an
5 150 100
6 100 100 input model with a geometrical checkerboard pat-
tern. We present in Figure 6, the results of such a
test for three representative depths: 14, 18, and 28
process, to avoid instability. Figure 5 shows data km. The input model—Figure 6 (top left)—con-
misfit as a function of model norm variation for sists of an alternating 3-D pattern given by a sine-
each iteration. Data misfit decreases, as model cubed function with period of 1.5 laterally, and
norm gradually increases, until convergence is 10 km in depth, overimposed on our a priori refer-
reached. ence—EPcrust in the crustal region and sp6 in the
[17] The tomographic model data misfit is also mantle below. We add Gaussian noise to the syn-
shown in Figure 4 (black curve) as a frequency thetic travel times, with a standard deviation of 0.8
histogram of seismic travel time residuals. We s to simulate the random data error. The anomalies
observe an RMS decreasing from 1.34 to 1.22 s
with respect to the relocated residuals computed in
the 3-D a priori model (red histogram), whereas
the mean values remain stable. In the a priori
model, the residuals larger than 12 s and smaller
than 22 s are 18% of the total seismic rays num-
ber, whereas after the origin time relocation, they
drop at 8.1%. In the final model, we still find some
residuals larger than 12 s and smaller then 22 s,
but they are only 6.5% of the total seismic rays
number.

3. Synthetic Tests
[18] We perform synthetic tests to estimate the
robustness of our inversion procedure. We attempt
to reconstruct a 3-D structure by inverting a set of
synthetic travel times computed for a known
model, using the same distribution of sources and

Figure 6. Resolution tests performed for synthetic struc-

tures as checkerboard test at three representative depths, 14,
18, and 28 km. The input model pattern is shown on the top
left. We observe that we are able to reconstruct only the conti-
Figure 5. Data misfit as a function of model norm variation nental part of the studied area due to the stations-events distri-
for each iteration (blue numbers). It has been normalized with bution (Figure 2). Only the region inside the black rectangle
respect to the number of seismic rays. Restraining and is taken into account during the computation and it denotes
smoothing factors for each iteration are shown in Table 1. the studied region.

6
195
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

are then plotted as relative velocity variations

from the background model. As expected, in Fig-
ure 6 we observe that only the continental regions
are well reconstructed. Because of the event and
station distribution we miss the oceanic areas. In
continental areas, this specific pattern of velocity
is best reconstructed below the Alpine arc and
North-central Apennines. There appears to be
preferential NE-SO smearing, emphasizing an
alignment of coherent velocity anomalies present
in the input model, oriented diagonally with
respect to the NS-aligned grid.
[19] We also note that the amplitudes of the
anomalies are globally smaller than in the input
model. This fact is commonly observed in these
reconstruction tests, and hints at a general under-
estimation of wave velocity variations with respect
to the background field. Besides, in Figure 6 it is
also clear that positive and negative variations are
not retrieved equally well. This may be ascribed to
wave front healing, making the evolution in space
of the time delay not symmetric with respect to the
sign of the initial anomaly, with the signature of
negative anomalies getting somehow obscured
under some circumstances—even in the regime of
validity of ray theory [Wielandt, 1987; Nolet and
Dahlen, 2000; Hung et al., 2001]. Our finite-
difference forward calculation scheme correctly
models the wave front healing effect—that may
instead be neglected by more approximate ray
tracing methods thus resulting in overly optimistic
outcomes of synthetic tests. Our synthetic data set
includes instead the healing effect in full.
[20] The real significance of this kind of tests has
however been questioned. Lev^eque et al. [1993]
show, for instance, that under some circumstances
small checkerboard patterns may be quite well
Figure 7. Synthetic test at three representative depths, 14,
imaged, even when larger-scale structure is much
28, and 40 km, one for each block of the a priori model. In
more poorly retrieved. Besides, in the nonlinear fact, the synthetic input model has been built considering
travel time inversion, raypaths—bent according to three ranges of depth having three different anomalous blocks,
the wave speed model—assume quite unrealistic from 10 to 18 km, from 20 to 28 km, and from 30 to 48 km.
oscillating shapes [Serretti and Morelli, 2011]. For (left) The synthetic input model and (right) the reconstructed
such reasons we deem better to use a more realistic one. Only the region inside the black rectangle is taken into
model in a reconstruction experiment, with syn- account during the computation and it denotes the studied
thetic structures similar to what we expect to find. region. We use a smaller scale range for plotting the recon-
In this way, we can verify the capacity of the structed model with respect to the synthetic input one.
method to retrieve seismic velocity structures in a
more realistic scenario. Figure 7 shows the input tures present in the peninsula, as described in the
synthetic model on the left. The anomalies are literature. We divided the input synthetic model in
plotted as relative velocity variations with respect three sectors as a function of depth, where we
to the background a priori model, consisting of superimpose different heterogeneous shapes to the
EPcrust in the crustal region and sp6 in the mantle background a priori model. Figure 7 (left) shows
below. We build these shapes taking inspiration the structure at three representative depths, one for
from the main geological and geophysical struc- each of the different sectors of the input model. As
7
196
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

Figure 8. (left) Average correlation coefficient for each iteration. We observe an increase of the correlation
coefficient of about 35%, from 0.42 until 0.65. (right) Correlation coefficient as a function of depth after the
first (blue) and the last (red) iteration. We can observe that, after six iterations, the correlation coefficient
increases particularly in depth, where the ray coverage increases (compare with Figure 3).

before, for this synthetic model we calculate syn- also be affected by this underestimation. As in the
thetic travel times along the paths in the real data previous layer, scattered small-amplitude positive
set. To simulate the random error present in the anomalies are present along the Apennines. These
real data, we added Gaussian noise to the synthetic anomalies are related to the third sector of the
travel times with standard deviation of 0.8 s. As input model, beginning just below this 28 km
we observed for the checkerboard test, the inten- depth. On the other hand, at 40 km depth—repre-
sity of anomalies in the reconstructed model sentative of the depth interval in the input model
(right) is underestimated with respect to the input between 30 and 48 km—both the Alps and the
model (left). This observation will bring some Apennines are quite well retrieved in shape (see
contribution to the discussion on the final tomo- Figure 7), whereas the anomalies are underesti-
graphic model (see next section). Figure 7 (top) mated in intensity.
shows a horizontal cross section at 14 km depth,
[21] To evaluate quantitatively the resemblance
representative of the input model in the range
between output and input models, we compute the
from 10 to 18 km. The synthetic pattern shows the
correlation coefficient between the two fields. It
alternation of positive and negative anomalies,
reaches the value of 65% after six iterations, as
simulating geological structure in the shallow part
shown in Figure 8 (left). Figure 8 also shows the
of the crust. The reconstructed model, to the right,
correlation as a function of depth for the first and
shows good agreement in spatial shape along the
the last iteration. Note that some relatively worse
Apennines. The intensity of anomalies is generally
correlation—between, say, 20 and 30 km—could
lower than in the input model. The reconstructed have been expected as it roughly corresponds to a
model also presents some background noise—with notch in the ray density (Figure 3). The relative
anomalies smaller than 1%—along the Apennines data misfit between travel times calculated in the
and especially along the Alps, in regions where final model and in the synthetic model decreases
the input model has DmP/mP 5 0%. This tendency is by 20% after six iterations, as shown in Figure 9.
due to the fact that, in the input model, few kilo- The misfit reduction is comparable with the data
meters below, the second model sector presents misfit decrease in the tomographic model (Figure
negative anomalies along the Alpine chain. The 5). At each step the relative data misfit decreases
middle row of Figure 7 shows, to the left, the input less and less, reaching convergence after six itera-
model cut at 28 km. In this depth range (from 20 tions. A seventh iteration would produce a
to 28 km) we include a low velocity body to repre- decrease in data misfit by less than 1%.
sent the Alpine roots. The outcome of the recon-
struction test is good below the Northern [22] The correlation between output and input
Apennines and the Alpine Arc in terms of shape of models in Figure 8 is rather high in the depth span
the anomalies, although the amplitude is, again, typical of the crust-to-mantle transition, due to the
underestimated. We may expect that the retrieved good sampling provided by rays in our data set
intensity of seismic velocity heterogeneity will (Figure 3b). We expect to retrieve the velocity
8
197
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

and vertical sections that show the main features

of the reconstructed three-dimensional P wave
velocity field.
[24] Figures 11 and 12 illustrate our resulting
tomographic model (right) compared with the a
priori model (left) at selected depths in terms of
absolute velocity. These depth sections represent
the whole model quite completely, given some
vertical continuity, as shown by the vertical cross
sections described further on. The layered 3-D ref-
erence model EPcrust [Molinari and Morelli,
2011], that we use as our a priori crustal model,
Figure 9. Synthetic relative data misfit, k Ddk2 , as a func- has laterally varying interface depths separating
tion of the norm of the model for the six iterations performed upper from lower crust, and an undulated Moho
(blue numbers). It has been normalized with respect to the discontinuity with the mantle. Seismic velocity
number of seismic rays. Data misfit decreases of about 20%, a assumes rather different values going from upper-
value comparable with data misfit decreasing of the tomo- most crust to mantle, so that to represent subtle
graphic model (Figure 5). The initial data misfit results domi-
nated by the Gaussian error we added to the input model,
having a variance of 0.64 s2. Restraining and smoothing fac-
tors are shown in Table 1 for each iteration.

structure particularly well in this vertical range, so

that, although we do not explicitly honor this tran-
sition as a discontinuity, we may think to search
for a Moho as the locus of points where wave
velocity transitions from crustal to mantle values.
Figure 10 shows the simulated reconstruction of
such a Moho, for the input model (top) and for the
reconstructed model (bottom). Deepening of the
transition, down to 60 km, is well reconstructed,
such as a shallowing of this model Moho—whose
extent appears instead under-estimated by some 5–
7 km. The positive outcome of this further test
induces us to use this criterium to the Moho in the
model resulting from inversion of real data.

4. Results
[23] In this section, we present the results of the
nonlinear tomographic inversion, as a new veloc-
ity model, contributing some interpretation in
terms of the geological structures that we may rec-
ognize in the maps. We also seize this opportunity
to provide a general view of the current state of
knowledge about the structure of the crust and
uppermost mantle in the Italian region. An anima-
tion showing maps of the model, as the depth
sweeps from the surface to the bottom of the
model, is provided as supporting information.1 Figure 10. Simulated reconstruction of crust-mantle transi-
Here we focus our attention on selected horizontal tion in the synthetic test, defined as depth of exceedance of a
threshold wave speed. (This criterium will be used further on
to define a reconstructed Moho for the inversion of real data.)
1
Additional supporting information may be found in the online (top) Interface in the input model. (bottom) Interface deduced
version of this article. from the reconstructed model.

9
198
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

Figure 11. (left) A priori model and (right) tomographic images at 14 and 18 km in terms of absolute P
wave velocity. We represent the main parts of the a priori model—lower crust, upper crust and mantle—by
using three different color scales. Starting from the left, we use colors from pink to violet for the upper crust,
from brown to black for the lower crust and from red to magenta for the mantle. Only the region inside the
black rectangle is taken into account during the computation and it denotes the studied region.

velocity variations we use three different color roughly follow the boundary between upper and
scales for velocity values pertaining to these three lower crustal material. Slower upper crust, to the
domains. We use colors varying from pink to vio- East, is denoted by violet, and it is relative to P
let for the upper crustal velocity range, from velocity values from 5.9 to 6.3 km/s. Following Di
brown to green-black for lower crustal velocity, Stefano et al. [2009], we may note that in the
and from red to blue-magenta for the mantle. upper crust P wave velocity patterns reflect local
geology: highs correspond to limestone or crystal-
[25] The first map in Figure 11 is cut at a constant line basement units, and lows to recent sedimen-
depth of 14 km. At this depth the Apennines tary basins. In our case, this behavior is to some
10
199
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

Figure 12. (left) A priori model and (right) tomographic images at 28 and 40 km in terms of absolute P
wave velocity. We represent the main parts of the a priori model—lower crust and mantle—by using two dif-
ferent color scales. Starting from the left, we use colors from brown to black for the lower crust and from red
to magenta for the mantle. Only the region inside the black rectangle is taken into account during the compu-
tation and it denotes the studied region.

extent inherited from EPcrust [Molinari and Mor- rhenian Sea, in correspondence of the magmatic
elli, 2011]. Lower crustal anomalies—shown with area of the volcanic arc of the Eolian Islands and
a brown-black color scale—are instead relative to the Marsili Seamount—see Figure 1 for toponyms.
values from 6.4 up to 7 km/s in the a priori model Melted mantle is believed to rise in the lower crust
and up to 7.3 km/s in the tomographic model. in the southern Tyrrhenian Sea, above the steeply
Mantle material is present at the center of the Tyr- dipping Ionian, feeding the magmatism of the
11
200
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

volcanic arc of the Marsili Seamount. Its basin is graphic model, the central and eastern part of the
near circular in shape, with a diameter of about Alps are characterized by low velocity roots (in
120 km, in accordance with Marani and Trua dark brown and light violet), in agreement with
[2002]. The presence of mantle is quite limited at previous tomographic studies such as Alessandrini
this depth. Typical mantle velocities are geograph- et al. [1995], Chiarabba and Amato [1996], Wald-
ically quite localized, mostly slow in a relative hauser et al. [2002], and Di Stefano et al. [1999,
sense, with velocities around 8 km/s both in the a 2009], whereas the a priori model shows two
priori and the final model (the structure under the blocks of uniform upper and lower crust velocities.
Tyrrhenian Sea is not well retrieved because of This fact may mean that the Alpine region at 18
scarce ray coverage, as shown in section 3). Fol- km depth is mostly characterized by typical upper
lowing the Apennines Arc, we may observe that in crustal rocks that underthrust the mountain belts
the northern part of the Apennines the border [Di Stefano et al., 2009]. In the Eastern Alps, the
between these two types of crustal material is relationship between the European and Adriatic
moving to the west with respect to the a priori plates is controversial [e.g., Lippitsch et al., 2003;
model. We observe in general that the anomalies Schmid et al., 2005]. Results from migrated
have the tendency to become less smooth, which receiver functions [Kummerow et al., 2004] sug-
is a typical behavior for a tomographic model with gest that the European Moho underthrusts the
respect to a simpler a priori reference. A small Adriatic lithosphere to the south. On the other
slow anomaly appears in the final tomographic hand, Lippitsch et al. [2003] and Schmid et al.
model under the Etna volcano area and it may be [2005] assert that a high-velocity body connected
related to its crustal roots. to the Adriatic lithosphere is subducting northeast
beneath the European plate. At the western part of
[26] The second plot in Figure 11 maps the model the Alpine arc, a lithospheric unit, known as the
at 18 km depth. Compared to the shallower depth, ‘‘Ivrea body’’, has been recognized as consisting
we now observe a considerable decrease of the of peridotitic rocks with high seismic velocity and
upper crustal domain. Along the Apennines we density—typical of lower crust and mantle—at
find quite isolate slow anomalies (dark violet and shallow depth and surrounded by crustal material
dark brown) which follow the mountain belt. On [e.g., Schmid and Kissling, 2000]. In Figure 11 at
the contrary, Di Stefano et al. [2011] have shown 18 km depth, the high velocities (in green-black)
at 22 km depth a quite continuous slow anomaly. below the western part of the Alps are then related
The mantle under the Tyrrhenian is wider (light to this Ivrea body [e.g., Solarino et al., 1997;
green, orange, and red areas) than at 14 km depth. Schmid and Kissling, 2000; Diehl et al., 2009;
Mantle velocities are 8.03 km/s in the a priori Wagner et al., 2012]. A fast anomaly was already
model, but become slower after inversions in the present in the a priori model at this depth, but its
area of Eolian Island and Marsili volcano. Veloc- amplitude, shape, and geographical location have
ity values close to 7.8 km/s or slightly lower— been completely reshaped in the final tomographic
absent in the a priori reference, as quite unrealistic model. In Figure 13, we show a vertical cross sec-
both for crustal and for mantle rocks—may result tion under the western Alps plotting relative veloc-
from the inversion lowering a starting mantle ity changes with respect to the a priori model. The
velocity of, say, 8 km/s. In view of the tendency figure shows then the update improvement to the
to underestimate adjustments to the initial model, reference crustal model due to our inversion. In
quite usual in tomographic inversions and also spite of a varying amplitude—likely reflecting the
suggested by results of the synthetic model recon- seismic ray density—it is clear the presence of a
struction tests shown in section 3, we interpret continuous high-velocity body between about 15
such features as the actual detection of crustal and 40 km depth, with some discontinuous fast
rocks. As our model representation—a continuous anomalies above and below it. This fast seismic
interpolation among nodes of a 3-D mesh, anomaly in the lower crust is inconsistent with
described in section 2—does not impose an common velocity of crustal materials [Christensen
explicit Moho, it may induce some smoothing of and Mooney, 1995], and it is more consistent with
the crust/mantle border as a result of inversion. the typical mantle velocity. The Ivrea body finds
However, a sharp transition from typically crustal an explanation as a piece of rising mantle that
to mantle velocities is generally present. The Etna forms the rigid frontal part of the Adriatic plate
volcano area diverts form the a priori model show- [Schmid and Kissling, 2000; Schmid et al., 2005].
ing a slow velocity patch shifted toward the south The high-velocity anomaly in Figure 13 is in gen-
with respect to 14 km depth. In the final tomo- eral agreement with previously proposed models
12
201
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

rhenum side of the Apennines, marking the roots

of Quaternary volcanoes and the magmatic provin-
ces of Tuscany, Latium, Campania, and the Aeo-
lian Islands. This low-Vp anomaly hence
correlates with Quaternary volcanoes and geother-
mal activity. We may thus interpret these features
as due to thermal anomalies, in agreement with the
observed high heat flow [see Di Stefano et al.,
2009, and references therein). The hypothesized
thermal state reconciles with the ending of mag-
matic activity in this area at about 0.3 Ma [Picci-
nini et al., 2010]. In the Western Alpine region at
this depth, we note that the absolute velocity
moves from a maximum velocity of 6.8 km/s in
the a priori model to a maximum velocity of 7.0
km/s in the final tomographic model. We believe
that this tendency to retrieve high velocities in this
region is associated with the Ivrea-Verbano body.
Wagner et al. [2012] identify the Ivrea-Verbano
body at 28 km depth (their Figure 5) with compa-
rable P wave velocities. The Ivrea-Verbano body,
Figure 13. Relative velocity change, with respect to the a which was already present in the a priori model at
priori model, along a vertical cross section under the Western 28 km as a piece of high crustal velocity, has been
Alps. We find a wide high velocity anomaly positioned completely reshaped after the inversion process by
between 20 and 40 km. We hypothesize that it is related to the the new data set, like at 18 km depth (Figures 11
Ivrea-Verbano body. and 13). In the central part of the Alps, we note
slow anomalies—which are not present in the a
of the Ivrea body, based on deep seismic sounding priori model—mainly associated with the crustal
data [e.g., Solarino et al., 1997; Schmid and Kis- root of the Alps, in substantial accord with Diehl
sling, 2000; Diehl et al., 2009; Wagner et al., et al. [2009].
2012] both in terms of location and vertical exten-
sion of the anomaly. Because of our irregular seis- [28] At 40 km depth (Figure 12) besides clear
mic ray coverage, we are not able to detect the crustal ‘‘roots’’ under the Alps already seen in the
Ivrea body deeper than 40 km and shallower than a priori model, we mostly observe the presence of
15 km, whereas Solarino et al. [1997] claim that mantle rocks. However, a main trait in the final
the body extends from 2 to 62 km depth. tomographic model is given by the slow anomaly
continuously running below the entire Apennines
[27] Figure 12 shows deeper layers, at 28 and 40 crest. This feature is not present in the reference
km. Here the upper crust disappears. Lower crust model, that at this depth consists of laterally uni-
and mantle can easily be distinguished as they are form mantle. We may interpret it as crustal mate-
presented with two different color scales, from rial, as the velocity is too low for mantle rocks.
brown to black and from red to magenta, respec- Such explanation is indeed in agreement with seis-
tively. At 28 km depth, we note quite a sharp edge mic receiver functions [e.g., Piana Agostinetti
running along the Apennines, where both the et al., 2008; Piana Agostinetti and Amato, 2009;
crustal region, to the East, and the mantle, to the Bianchi et al., 2010] and gravity studies [e.g.,
West, are marked by slow anomalies. The continu- Tiberti et al., 2005]. In the following section, we
ous slow anomalies—which were not present in discuss this interpretation, involving deepening of
the a priori model—along all the Tyrrhenian coast the Moho, in more detail. Low seismic velocities
are probably related to strong thermal activities at this depth may also have different origin. Conti-
indicating asthenosphere upwellings [Di Stefano nental crustal material in subduction zones may be
et al., 2009]. This hypothesis is also supported by carried down by the Adriatic lithospheric slab
a concentration of very high attenuation in the [e.g., Di Stefano et al., 1999, and references
Tyrrhenian wedge, above the Adriatic subduction therein]. Mantle seismic anomalies can also be
slab, as explained by Piccinini et al. [2010]. The associated with thermal effects. In this case, the
narrow slow anomalies extend along all the Tyr- deep slow Apenninic anomaly could be associated
13
202
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

with rising hot asthenospheric material in front of from typical crustal to mantle P wave velocity
the Adriatic slab. Several evidences have been appears to be a reliable way to identify the Moho.
found by Mele et al. [1998], Di Stefano et al. We can therefore search the sudden transition
[1999], Piromallo and Morelli [1997], and Piro- from 7.6 to 8.0 km/s in our model, and map its
mallo and Morelli [2003]. In the Alpine region, depth over all the region. Figure 15 shows the
we see the presence of lower crustal material, map of this Moho depth, compared with the a pri-
mostly slower than the a priori model, and then ori model EPcrust [Molinari and Morelli, 2011].
much slower than the Apenninic anomaly. Both The geographical area covered in this paper is
the features found beneath Alps and Apennines marked by a black rectangle. Outside this area,
confirm previous studies about the existence of the Moho depth does not change with respect to
crustal roots nd the presence of a deep Moho in the a priori model. The total range has not
these areas [e.g., Piana Agostinetti and Amato, changed considerably after the inversion, going
2009, for the Apennines Moho; e.g., Stehly et al., from 10 to 50 km in the a priori model, and
2009, for the Alps Moho]. reaching 52 km in the final model, compatibly
with receiver function studies [e.g., Piana Agosti-
netti and Amato, 2009; Di Stefano et al., 2011].
4.1. Moho Depth However, we can point out some significant
[29] Figure 14 shows three vertical, parallel, cross improvement in the final model. The most notice-
sections cut through the a priori model (left) and able variation is under the main reliefs, i.e., Alps
the final tomographic model (right) roughly and Apennines. The deepest Alpine area is wider
orthogonally with respect to the axis of the Apen- in our Moho model than in EPcrust. In particular,
nines. We notice that the crust/mantle discontinu- the deep eastern area is wider both in latitude and
ity in the a priori model is understandably longitude, and is also deeper.
smoothed out in the tomographic result that has no
[31] Indeed, in this area, the purple-colored area is
constraint imposed on the velocity field. However, broader than in the top figure, indicating depths
a sharp transition is still recognizable that can be down to 52 km. The western area, under the
associated to the Moho and that is entirely French-Italian border close to the Ivrea zone,
required by travel time data. This Moho remains shows depth down to about 50 km. These features
shallower at the Tyrrhenian side, about 20 km, are in agreement with other studies concerning
than under the Adriatic, about 40 km, and is deep- this area—e.g., from inversion of seismic ambient
est under the Apennines, where it reaches about 40 noise [Stehly et al., 2009], and from wide-angle
km. Notice a clear indication of increased depth of seismic tomography [Bleibinhaus and Gebrande,
the crust/mantle transition with respect to the ini- 2005]. Making a comparison between Figure 15
tial model. Cross sections AB and CD show indi- and the topography map we can easily observe
cations of deepening of the Adriatic Moho toward that our Moho depth follows the topographic ele-
SW, consistent with models of lithospheric sub- vation in the Alpine region. It is clear that the
duction and receiver function studies [e.g., Piana Moho depth, at a first order, qualitatively accounts
Agostinetti et al., 2008; Piana Agostinetti and for isostatic equilibrium. In the western part of the
Amato, 2009; Bianchi et al., 2010] that also show Alps, we obtain Moho depths shallower by about
a SW-dipping Moho and a flat 20 km thick Tyr- 10 km than some regional studies, such as Wagner
rhenian crust. The deepest Moho along the Apen- et al. [2012]. The Alpine roots show here a quite
nines—46 km—is reached in section EF, in continuous trend. On the contrary, Di Stefano
correspondence to the highest mountains. et al. [2011] present a discontinuity around 46 of
latitude and 8 of longitude, with depths about 10
[30] Moho topography is strongly correlated with km shallower than in our case.
tectonics and geodynamic processes and strongly
influences the measure of most of the geophysical [32] To the South of the Alpine chain, the border
properties of the Earth, like the gravity field and between the Alps roots and the Po Plain region is
the propagation of seismic rays. Hence a detailed clearly marked out. The crustal thickness reaches
Moho map is a key requisite for geodynamic about 30–35 km. This is the signature of the plate
modeling, for understanding the evolution and boundary between the European plate, to the
state of lithosphere, and to correct seismic data to north, and the Adriatic plate. The boundaries of
map the upper mantle structure [Lippitsch et al., these plates correlate with the strike of the Eastern
2003]. A first-order discontinuity is not present in Alps and the Dinarides. The European Moho dips
the tomographic model, but the sharp transition to the south and the Adriatic Moho dips to the
14
203
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

Figure 14. Three parallel vertical cross sections cut along the Northern Apennines. (left) The a priori model
and (right) the final tomographic model. In the central part of all three sections, moving from North to South
(from AB to EF), we find out that the deepest part of the Moho moves toward the East. The Moho depth is
marked by a sharp transition between yellow (lower crust) and blue (mantle).

15
204
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

of 50 km only in three main areas located under

the Northern and Central Apennines. The disconti-
nuity of the Moho depth under the Apennine
Chain has been recognized by previous studies,
e.g., the teleseismic receiver function study by
Piana Agostinetti and Amato [2009]. This feature
suggests a discontinuity of the belt axes in central
southern Italy, between 41 and 42 [Amato et al.,
1993; Piromallo and Morelli, 2003; Di Stefano
et al., 2011]. We can also point out the largest
Moho depth under the Gran Sasso area (about 46
km depth) in Central Italy. This Moho deepening
has also been found by Piana Agostinetti et al.
[2009]. We observe that the sharp boundary steep-
ness of the Adriatic plate approximately decreases
from north to south, as pointed out by Di Stefano
et al. [2009]. The Apennines belt continues the
South in the Calabrian Region, where we do not
observe any significant changes with respect to
EPcrust. The Calabrian Arc is characterised by
Moho depth around 40 km, like in Di Stefano
et al. [2011]. We believe that this lack of variation
of the Moho depth in the Calabrian region may be
mostly due to the poor path coverage of our data
set in this area. We detect instead an increase of
the thickness of the crust under the Etna area,
increasing from about 30 to 35 km. On the north-
east of the Apennines chain, the Ligurian Moho
changes in depth from a about 20 km beneath the
deepest parts of the Ligurian Sea—see Figure 1
for toponyms—to about 30 km beneath the moun-
tain chain. This evidence is in accord with the
result of the regional study developed by Wagner
et al. [2012], despite of their border is shifted
Figure 15. (top) Moho depth in the a priori model. (bottom) toward the north and it is better constrained. All
Moho depth identified as a sudden transition from 7.6 to 8.0 the Tyrrhenian side of the Apennines (i.e., Tus-
km/s in the final model. The studied area in this paper is cany, Latium, and Campania) is characterized by
inside the black rectangle. Outside the Moho depth is the shallow depths, with consistent values between 20
same as in the a priori model. We observe a deeping Moho and 30 km, in accord with results shown by Piana
especially under the central Apennines and the Western Alps. Agostinetti et al. [2009] and assert by Wagner
Only the region inside the black rectangle is taken into
et al. [2012].
account during the computation and it denotes the studied
region. [34] A Moho depth minimum, around 10 km
depth, is imaged in the central Tyrrhenian sea,
where the Seamount Marsili is located. Despite of
north-east. The imprint of the Po plain on the we observe some Vp velocity anomalies in this
Moho topography is partially inherited from area (see Figure 10, 18 km depth), the Moho map
EPcrust [Molinari and Morelli, 2011], but it does not show significative additional anomalies.
appears shallower for a wider area. These Moho depths presented by EPcrust are in
[33] Another important peculiarity of our Moho substantial accord with Di Stefano et al. [2011].
depth is along the Apennines chain, where its
structure significantly differs from the a priori 5. Conclusions
information. In the same way as for the Alpine
belt, here we can observe an arc of deeper Moho [35] In this paper, we present a new model for the
along the Apennines. It reaches depths just short P wave velocity structure of the Earth’s crust and
16
205
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

uppermost mantle beneath Italy, derived by seis- Po Plain area does not permit here to update the
mic travel time tomography with an innovative prior model with lateral variations of seismic wave
implementation consisting of full 3-D finite-differ- velocity.
ence forward calculation and nonlinear iterative
[38] The main velocity heterogeneities recovered
inversion of first-arrival travel times between
by this work can be related to geological evidences
earthquakes and stations located in the study
on the shallower part of the crust, and to geophysi-
region (Figure 2). The finite-difference method
cal processes in the deeper one. Starting from a
used for forward calculations [Podvin and
depth of about 18 km we see a low velocity area—
Lecompte, 1991; Tryggvason and Bergman, 2006] laterally discontinuous in the shallow part, but
is able to tackle strong heterogeneities and discon- much more continuous deeper down—beneath the
tinuities, such as those present in the crust. Alpine and Apenninic belt, likely related to plate
Although computationally more expensive, track- subduction processes [Carminati and Doglioni,
ing the full wave front through the 3-D medium 2012] or delamination [Benoit et al., 2011]. We
avoids inaccuracies and difficulties met by approx- confirm the presence of the Ivrea body, a high-
imate or iterative ray tracing schemes, such as ray velocity body which is not consistent with proper-
bending methods that may not converge to true ties of known crustal rocks, but is rather assimi-
minimum-time paths [Serretti and Morelli, 2011]. lated to mantle material surrounded by crust. We
This is particularly relevant for a strongly hetero- have found the Ivrea body extending from 15 to
geneous setting, such as that encountered in crustal 40 km, in agreement with previous studies [e.g.,
tomography that include modeling of the Moho. Schmid and Kissling, 2000; Schmid et al., 2005;
Coupled to this numerical scheme to solve the for- Nicolich, 2010].
ward problem, we implement a nonlinear iterative
inversion method that uses full 3-D a priori in the [39] We also present a new map of the Moho
form of a reference crustal model, and permits sta- topography for the Italian region, obtained consid-
ble inversion with a detailed model description. ering the velocity transition from 7.6 to 8.0 km/s
retrieved in the final velocity model as the transi-
[36] The resulting model provides a local tion from the crust to the mantle. The map shows a
improvement of the recent 3-D crustal reference shallower Moho on the Tyrrenian side, with about
model EPcrust [Molinari and Morelli, 2011], that 20 km, becoming deeper beneath the Apennines
we use as a priori information in the inversion. chain, to about 40 km. The Adriatic side presents a
Although the model is represented by absolute P Moho depth at about 30 km. The deepening of the
wave velocity on a 3-D grid of nodes, and it does Moho surface under the mountains belt, which
not explicitly include any discontinuity, a reliable corresponds to the low-velocity arc present in the
estimate of the crust-mantle boundary can be tomographic images, is due to the presence of iso-
obtained mapping the strong gradient marking the statically compensated mountains roots. In the
transition from typical crustal velocity values to western part of the Alps we observe deepening of
mantle ones. The resulting map of Moho depth the mountain roots that reach about 50 km in cor-
reveals some significant features, and—as it is not respondence of the Ivrea body. The border
induced by any explicit parameterization—it can between the roots of the Alps and the Po Plain
be considered an unbiased estimate of crustal
region is well marked after the inversion. The
thickness.
Moho depth is about 30–35 km under the Po Plain
[37] The resolution of the model strongly depends and around 50 km in the Alpine region. Along the
on the coverage of seismic rays. Given the limited Apennines, the map shows the presence of a deep
areal extent of the model, and our choice of using Moho arc, which marks the belt trend. It reaches a
only stations and hypocenters located inside the maximum depth of about 50 km only in three areas
study region, first-arrival seismic rays only dip located under the Northern and Central Apennines.
down to about 60 km, but fair coverage is limited This maximum depth is in general comparable
to the first, say, 50 km. The synthetic model recov- with the Alpine Moho, except for the Western
ery test (Figure 7) however shows good capability Alps which record the deepest Moho area in Italy.
of reproducing velocity structures, particularly for These considerations are in agreement with previ-
layers and areas with good ray coverage. Adequate ous studies obtained using receiver functions [e.g.,
presence of seismic stations is essential to guaran- Piana Agostinetti and Amato, 2009; Di Stefano
tee good seismic ray illumination—and hence reli- et al., 2011]. We note that, although regional-
able tomographic images—in the shallower layers distance first-arrival travel times are generally not
of the crust. The scarcity of seismic stations in the deemed the most appropriate data set to map
17
206
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

Moho depth, and in spite of the fact that no discon- Bianchi, I., J. Park, N. Piana Agostinetti, and V. Levin (2010),
tinuity is prebuilt in the model parameterization, Mapping seismic anisotropy using harmonic decomposition
of receiver functions: An application to Northern Apen-
by accurately modeling wave front propagation nines, Italy, J. Geophys. Res., 115, B12317, doi:10.1029/
our method is indeed able to retrieve the crust/ 2009JB007061.
mantle boundary in a geographically extensive Bleibinhaus, F., and H. Gebrande (2005), Crustal structure of
region with good confidence. the Eastern Alps along the TRANSALP profile from wide-
angle seismic tomography, Tectonophysics, 414, 51–69, doi:
[40] The quality of any tomographic model 10.1016/j.tecto.2005.10.028.
depends on an accurate method and on the virtues Carminati, E., and C. Doglioni (2012), Alps vs. Apennines:
of the data set. The performance of our method The paradigm of a tectonically asymmetric Earth, Earth Sci.
has been extensively documented in the paper, and Rev., 112, 67–96, doi:10.1016/j.earscirev.2012.02.004.
Chiarabba, C., and A. Amato (1996), Crustal velocity structure
is also validated by the results, but we have also of the Apennines (Italy) from P-wave travel time tomogra-
commented on some limits posed by the data set. phy, Ann. Geophys., 39(6), 1133–1148, doi:10.4401/
We have considered P wave travel times retrieved ag-4042.
from the high-quality EHB Bulletin of the Interna- Christensen, N. I., and W. D. Mooney (1995), Seismic velocity
tional Seismological Centre [ISC, 2009] for the structure and compositional of the continental crust: A
global view, J. Geophys. Res., 100(B6), 9761–9788, doi:
time period between 1988 and 2007, but some 10.1029/95JB00259.
improvement may in fact be possible. First of all, Cimini, G., and P. De Gori (2001), Nonlinear P-wave tomogra-
the apparent gap in station coverage in Northern phy of subducted lithosphere beneath central-southern
Italy is currently being filled by new seismo- Apennines (Italy), Geophys. Res. Lett., 28(23), 4387–4390,
graphic stations that in the near future may provide doi:10.1029/2001GL013546.
important data. Also, time picks from uniformly Diehl, T., S. Husen, E. Kissling, and N. Deichmann (2009),
High-resolution 3-D P-wave model of the Alpine crust, Geo-
reprocessed seismograms may in fact decrease phys. J. Int., 179(2), 1133–1147, doi:10.1111/j.1365-246X.
the data noise [Di Stefano et al., 2011] and result 2009.04331.x.
in a more accurate inversion. These two improve- Di Stefano, R., C. Chiarabba, F. Lucente, and A. Amato
ments of quantity and quality of the data set may (1999), Crustal and uppermost mantle structure in Italy from
in the future result in an even better tomographic the inversion of P-wave arrival times: Geodynamic implica-
tions, Geophys. J. Int., 139, 483–498, doi:10.1046/j.1365-
model of the crust and uppermost mantle in this 246x.1999.00952.x.
area. Di Stefano, R., E. Kissling, C. Chiarabba, A. Amato, and D.
Giardini (2009), Shallow subduction beneath Italy: Three-
dimensional images of the Adriatic-European-Tyrrhenian
Acknowledgments lithosphere system based on high-quality P wave arrival
times, J. Geophys. Res., 114, B05305, doi:10.1029/2008JB
005641.
[41] We thank the support from the QUEST Initial Training Di Stefano, R., I. Bianchi, M. G. Ciaccio, G. Carrara, and E.
Network funded within the EU Marie Curie Program. All the Kissling (2011), Three-dimensional Moho topography in
figures were realized using Generic Mapping Tools (GMT) Italy: New constraints from receiver functions and con-
[Wessel and Smith, 1998]. We are grateful to Associate Editor trolled source seismology, Geochem. Geophys. Geosyst., 12,
Cin-Ty Lee, Jeffrey Park, and Nicola Piana Agostinetti for Q09006, doi:10.1029/2011GC003649.
insightful comments and suggestions which lead to an Engdahl, E. R., R. Van Der Hilst, and R. Buland (1998),
improvement of the original manuscript. This is the IPGP Global teleseismic earthquake relocation with improved
contribution number 3461. travel times and procedures for depth determination, Bull.
Seimol. Soc. Am., 88(3), 722–743.
Faccenna, C., C. Piromallo, A. Crespo-Blanc, L. Jolivet, and
References F. Rossetti (2004), Lateral slab deformation and the origin of
the western Mediterranean arcs, Tectonics, 23, TC1012, doi:
10.1029/2002TC001488.
Alessandrini, B., L. Beranzoli, and F. M. Mele (1995), 3-D
Gill, P. E., W. Murray, and M. H. Wright, (1982), Practical
crustal P-wave velocity tomography of the Italian region
Optimization, 418 pp., Emerald Group Publ. Ltd, Academic
using local and regional seismicity data, Ann. Geophys.,
Press, London.
38(2), 189–211, doi:10.4401/ag-4119.
Handy, M. R., S. M. Schmid, R. Bousquet, E. Kissling, and D.
Amato, A., B. Alessandrini, G. Cimini, A. Frepoli, and G. Bernoulli (2010), Reconciling plate-tectonic reconstructions
Selvaggi (1993), Active and remnant subducted slabs of Alpine Tethys with the geological-geophysical record of
beneath Italy: Evidence from seismic tomography and seis- spreading and subduction in the Alps, Earth Sci. Rev., 102,
micity, Ann. Geophys., 36(2), 201–214. 121–158, doi:10.1016/j.earscirev.2010.06.002.
Benoit, M. H., M. Torpey, K. Liszewski, V. Levin, and J. Park Hung, S.-H., F. A. Dahlen, and G. Nolet (2001), Wavefront
(2011), P and S wave upper mantle seismic velocity struc- healing: A banana-doughnut perspective, Geophys. J. Int.,
ture beneath the northern Apennines: New evidence for the 146, 289–312.
end of subduction, Geochem. Geophys. Geosyst., 12, International Seismological Centre (2009), EHB Bulletin,
Q06004, doi:10.1029/2010GC003428. Thatcham, U. K. [Available at http://www.isc.ac.uk.].

18
207
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

Kennett, B. L. N., E. R. Engdahl, and R. Buland (1995). Con- Paige, C. C., and M. A. Saunders (1982a), LSQR: An algo-
straints on seismic velocities in the Earth from traveltimes, rithm for sparse linear equations and sparse least squares,
Geophys. J. Int., 122, 108–124. ACM Trans. Math. Software, 8(1), 43–71.
Kissling, E., S. M. Schmid, R. Lippitsch, J. Ansorge, and B. Paige, C. C., and M. A. Saunders (1982b), LSQR: Sparse lin-
Fgenschuh (2006), Lithosphere structure and tectonic evolu- ear equations and least squares problems, ACM Trans. Math.
tion of the Alpine arc: New evidence from high-resolution Software, 8(2), 195–209.
teleseismic tomography, Geol. Soc. London Mem., 32, 129– Papazachos, C. B., and G. Nolet (1997), Non-linear arrival
145, doi:10.1144/GSL.MEM.2006.032.01.08. time tomography, Ann. Geophys., 40(1), 85–97, doi:
Kummerow, J., R. Kinda, O. Oncken, P. Giese, T. Ryberg, K. 10.4401/ag-3937.
Wylegalla, F. Scherbaum, and TRANSALP Working Group Piana Agostinetti, N., and A. Amato (2009), Moho depth and
(2004), A natural and controlled source seismic profile Vp/vS ratio in peninsular Italy from teleseismic receiver func-
through the Eastern Alps: TRANSALP, Earth Planet. Sci. tions, J. Geophys. Res., 114, B06303, doi:10.1029/2008
Lett., 225(12), 115–129. JB005899.
Lapidus, L., and G. F. Pinder (1982), Numerical Solution of Piana Agostinetti, N., F. Lucente, G. Selvaggi, and M. Di
Partial Differential Equations in Science and Engineering, Bona (2002), Crustal structure and Moho geometry beneath
John Wiley, New York. the Northern Apennines (Italy), Geophys. Res. Lett., 29(20),
Lev^eque, J. J., L. Rivera, and G. Wittlinger (1993), On the use 1999, doi:10.1029/2002GL015109.
of the checker-board test to assess the resolution of tomo- Piana Agostinetti, N., V. Levin, and J. Park (2008), Crustal
graphic inversions, Geophys. J. Int., 115, 313–318, doi: structure above a retreating trench: Receiver function study
10.1111/j.1365-246X.1993.tb05605.x. of the northern Apennines orogen, Earth Planet. Sci. Lett.,
Lippitsch, R., E. Kissling, and J. Ansorge (2003), Upper man- 275, 211–220, doi:10.1016/j.epsl.2008.06.02.
tle structure beneath the Alpine orogen from high-resolution Piccinini, D., M. Di Bona, F. P. Lucente, V. Levin, and J. Park
teleseismic tomography, J. Geophys. Res., 108(88), 2376, (2010), Seismic attenuation and mantle wedge temperature
doi:10.1029/2002JB002016. in the northern Apennines subduction zone (Italy) from tele-
Lucente, F., C. Chiarabba, G. Cimini, and D. Giardini (1999), seismic body wave spectra, J. Geophys. Res., 115, B09309,
Tomographic constraints on the geodynamic evolution of doi:10.1029/2009JB007180.
the Italian region, Geophys. Res. Lett., 104(B9), 20,307– Piromallo, C., and A. Morelli (1997), Imaging the Mediterra-
20,327, doi:10.1029/1999JB900147. nean upper mantle by P-wave travel time tomography, Ann.
Malinverno, A., and W. Ryan (1986), Extension in the Tyrrhe- Geophys, 40(4), 963–979, doi:10.4401/ag-3890.
nian Sea and shortening in the Apennines as result of arc Piromallo, C., and A. Morelli (2003), P wave tomography of
migration driven by sinking of the lithosphere, Tectonics, the mantle under the Alpine-Mediterranean area, J. Geophys.
5(2), 227–245, doi:10.1029/TC005i002p00227. Res., 108(B2), 2065, doi:10.1029/2002JB001757.
Marani, M. P., and T. Trua (2002), Thermal constriction and Podvin, P., and I. Lecomte (1991), Finite difference computa-
slab tearing at the origin of a superinflated spreading ridge: tion of traveltimes in very contrasted velocity models: A
Marsili volcano (Tyrrhenian Sea), J. Geophys. Res., massively parallel approach and its associated tools, Geo-
107(B9), 2188, doi:10.1029/2001JB000285. phys. J. Int., 105, 271–284, doi:10.1111/j.1365-246X.1991.
Mele, G., and E. Sandvol (2003), Deep crustal roots beneath tb03461.x.
the northern Apennines inferred from teleseismic receiver Serretti, P., and A. Morelli (2011), Seismic rays and traveltime
functions, Earth Planet. Sci. Lett., 211, 6978, doi:10.1016/ tomography of strongly heterogeneous mantle structure:
S0012-821X(03)00185-7. Application to the Central Mediterranean, Geophys.
Mele, G., A. Rovelli, D. Seber, T. M. Hearn, and M. Barazangi J. Int., 187(3), 1708–1724, doi:10.1111/j.1365-246X.2011.
(1998), Compressional velocity structure and anisotropy in 05242.x.
the uppermost mantle beneath Italy and surrounding regions, Schmid, S., B. F€ ugenschuh, E. Kissling, and R. Schuster
J. Geophys. Res., 103(B6), 12,529–12,543, doi:10.1029/ (2005), Tectonic map and overall architecture of the Alpine
98JB00596. orogen, Eclogae Geol. Helv., 97, 93–117, doi:10.1007/
Molinari, I., and A. Morelli (2011), EPcrust: A reference s00015-004-1113-x.
crustal model for the European Plate, Geophys. J. Int., Schmid, S. M., and E. Kissling (2000), The arc of the western
185(1), 352–364, doi:10.1111/j.1365-246X.2011.04940.x. Alps in the light of geophysical data on deep crustal struc-
Montuori, C., G. Cimini, and P. Favalli (2007), Teleseismic ture, Tectonics, 19(1), 62–85, doi:10.1029/1999TC900057.
tomography of the southern Tyrrhenian subduction zone: Solarino, S., E. Kissling, S. Sellami, G. Smriglio, F.
New results from seaflor and land recordings, J. Geophys. Thouvenot, M. Granet, K. P. Bonjer, and D. Slejko (1997),
Res., 112, B03311, doi:10.1029/2005JB004114. Compilation of a recent seismicity data base of the greater
Morelli, A., and A. Dziewonski (1993), Body wave travel- Alpine region from several seismological networks and pre-
times and a spherically symmetric P- and S-wave velocity liminary 3D tomographic results, Ann. Geophys., 40(1),
model, Geophys. J. Int., 112, 178–194, doi:10.1111/j.1365- 161–174.
246X.1993.tb01448.x. Stehly, L., B. Fry, M. Campillo, N. M. Shapiro, J. Giulbert, L.
Nicolich, R. (2010), Geophysical investigation of the crust of Boschi, and D. Giardini (2009), Tomography of the Alpine
the Upper Adriatic and neighbouring chains, Rend. Fis. Acc. region from observations of seismic ambient noise, Geophys.
Lincei, 21, suppl. 1, 1530, doi:10.1007/s12210-010-0099-8. J. Int., 178(1), 338–350, doi:10.1111/j.1365-246X.2009.
Nolet, G. (1987), Seismic wave propagation and seismic 04132.x.
tomography, in Seismic Tomography, edited by G. Nolet, pp. Tarantola, A. (2005), Inverse Problem Theory and Model
1–23, D. Reidel, Norwell, Mass. Parameter Estimation, Society for Industrial and Applied
Nolet, G., and F. A. Dahlen (2000), Wave front healing and Mathematics, Philadelphia, Penn.
the evolution of seismic delay times, J. Geophys. Res., Tiberti, M., L. Orlando, D. Di Bucci, M. Bernabini, and M.
105(B8), 19,043–19,054. Parotto (2005), Regional gravity anomaly map and crustal

19
208
Annex I - P-wave tomography of the Italian region: an additional work

GUALTIERI ET AL.: P WAVE TOMOGRAPHY OF THE ITALIAN REGION 10.1002/2013GC004988

model of the central southern Apennines (Italy), J. Geodyn., region, Geophys. J. Int., 191(2), 789–802, doi:10.1111/
40, 73–91, doi:10.1016/j.jog.2005.07.014. j.1365-246X.2012.05655.x.
Tryggvason, A., and B. Bergman (2006), A traveltime reci- Waldhauser, F., R. Lippitsch, E. Kissling, and J. Ansorge
procity discrepancy in the Pdvin & Lecomte time3d finite (2002), High resolution teleseismic tomography of upper-
difference algorithm, Geophys. J. Int., 165, 432–435, doi: mantle structure using an a priori three-dimensional crustal
10.1111/j1365-246X.2006.02925.x. model, Geophys. J. Int., 150, 403–414.
Vidale, J. E. (1988), Finite-difference calculation of travel Wessel, P., and W. H. F. Smith (1998), New improved version
times, Bull. Seismol. Soc. Am., 78(6), 2062–2076. of the Generic Mapping Tools released, EOS Trans. AGU,
Vidale, J. E. (1990), Finite-difference calculation of travel 79(47), 579, doi:10.1029/98EO00426.
times in three dimensions, Geophysics, 55(5), 521–526. Wielandt, H. (1987), On the validity of the ray approximation
Wagner, M., E. Kissling, and S. Husen (2012), Combining for interpreting delay times, in Seismic Tomography, edited
controlled-source seismology and local earthquake tomogra- by G. Nolet, pp. 85–98, Reidel Publishing Co., Dordrecht,
phy to derive a 3-D crustal model of the western Alpine Holland.

20
209
Bibliography

Aki, K. & Richards, P. G., 2002. Quantitative Seismology, Second Edition, Sausalito,
California, 2nd edn.

Ardhuin, F. & Herbers, T. H. C., 2013. Noise generation in the solid Earth, oceans
and atmosphere, from nonlinear interacting surface gravity waves in finite depth,
Journal of Fluid Mechanics, 716, 316–348.

Ardhuin, F. & Roland, A., 2012. Coastal wave reflection, directional spread, and seis-
moacoustic noise sources, Journal of Geophysical Research: Oceans (1978–2012),
117(C6).

Ardhuin, F., Rogers, E., Babanin, A. V., Filipot, J.-F., Magne, R., Roland, A., Van
Der Westhuysen, A., Queffeulou, P., Lefevre, J.-M., Aouf, L., & Collard, F., 2010.
Semiempirical Dissipation Source Functions for Ocean Waves. Part I: Definition,
Calibration, and Validation, Journal of physical oceanography, 40(9), 1917–1941.

Ardhuin, F., Stutzmann, E., Schimmel, M., & Mangeney, A., 2011. Ocean wave
sources of seismic noise, Journal of Geophysical Research, 116(C9), C09004.

Ardhuin, F., Balanche, A., Stutzmann, E., & Obrebski, M., 2012. From seismic noise
to ocean wave parameters: General methods and validation, Journal of Geophysical
Research, 117(C5), C05002.

Aster, R. C., McNamara, D. E., & Bromirski, P. D., 2008. Multidecadal Climate-
induced Variability in Microseisms, Seismological Research Letters, 79(2), 194–202.

Aster, R. C., McNamara, D. E., & Bromirski, P. D., 2010. Global trends in extremal
microseism intensity, Geophysical Research Letters, 37(14), L14303.

211
Bibliography

Basu, U. & Chopra, A. K., 2004. Perfectly matched layers for transient elastody-
namics of unbounded domains, International Journal for Numerical Methods in
Engineering, 59(8), 1039–1074.

Berenger, J.-P., 1994. A perfectly matched layer for the absorption of electromagnetic
waves, Journal of computational physics, 114(2), 185–200.

Berger, J., 2004. Ambient Earth noise: A survey of the Global Seismographic Net-
work, Journal of Geophysical Research, 109(B11), B11307.

Bertelli, T., 1872. Osservazioni sui Piccoli Movimenti dei Pendoli in Relazione ad Al-
cuni Fenomeni Meteorologici, Bollettino Meteorologico Dell’Osservatorio Del Col-
legio Romano.

Bertelli, T., 1875. Della Realtà dei Moti Microsismici ed Osservazioni sui Medes-
imi fatte nell’anno 1873-1874 nel Collegio alla Querce Presso Firenze, Atti dell’
Accademia Pontificia de’ nuovi Lincei , pp. 334–375.

Bertelli, T., 1878. Riassunto delle Osservazioni Microsismiche fatte nei Collegio alla
Querce di Firenze, Atti dell’Accademia Pontificia de’ nuovi Lincei , pp. 193–243.

Blaik, M. & Donn, W. L., 1953. Microseism ground motion at palisades and weston,
Tech. rep., DTIC Document.

Brenguier, F., Campillo, M., Hadziioannou, C., Shapiro, N., Nadeau, R. M., & Larose,
E., 2008a. Postseismic relaxation along the san andreas fault at parkfield from
continuous seismological observations, Science, 321(5895), 1478–1481.

Brenguier, F., Shapiro, N. M., Campillo, M., Ferrazzini, V., Duputel, Z., Coutant,
O., & Nercessian, A., 2008b. Towards forecasting volcanic eruptions using seismic
noise, Nature Geoscience, 1(2), 126–130.

Bromirski, P. D., 2001. Vibrations from the “perfect storm”, Geochemistry Geophysics
Geosystems, 2(7).

Bromirski, P. D. & Duennebier, F. K., 2002. The near-coastal microseism spectrum:

Spatial and temporal wave climate relationships, Journal of Geophysical Research,
107(B8), 2166.

212
Bibliography

Bromirski, P. D. & Gerstoft, P., 2009. Dominant source regions of the Earth’s “hum”
are coastal, Geophysical Research Letters, 36(13), L13303.

Bromirski, P. D., Flick, R. E., & Graham, N., 1999. Ocean wave height determined
from inland seismometer data: Implications for investigating wave climate changes
in the NE Pacific, Journal of Geophysical Research: Oceans (1978–2012), 104(C9),
20753–20766.

Bromirski, P. D., Duennebier, F. K., & Stephen, R. A., 2005. Mid-ocean microseisms,
Geochemistry Geophysics Geosystems, 6(4), Q04009.

Bromirski, P. D., Stephen, R. A., & Gerstoft, P., 2013. Are deep-ocean-generated
surface-wave microseisms observed on land?, Journal of Geophysical Research:
Solid Earth, 118(7), 3610–3629.

Campillo, M., 2006. Phase and correlation inrandom’seismic fields and the recon-
struction of the green function, Pure and Applied Geophysics, 163(2-3), 475–502.

Capdeville, Y., 2002. Méthode couplée éléments spectraux – solution modale pour la
propagation d’ondes dans la Terre à l’échelle globale, Ph.D. thesis.

Capon, J., 1973. Analysis of microseismic noise at lasa, norsar and alpa, Geophysical
Journal of the Royal Astronomical Society, 35(1-3), 39–54.

Červený, V., 1989. Synthetic body wave seismograms for laterally varying media
containing thin transition layers, Geophysical Journal of the Royal Astronomical
Society, 99(2), 331–349.

Cessaro, R. K., 1994. Sources of primary and secondary microseisms, Bulletin of the
Seismological Society of America, 84(1), 142–148.

Cessaro, R. K. & Chan, W. W., 1989. Wide-angle triangulation array study of

simultaneous primary microseism sources, Journal of Geophysical Research: Solid
Earth (1978–2012), 94(B11), 15555–15563.

Chaljub, E., Komatitsch, D., Vilotte, J.-P., Capdeville, Y., Valette, B., & Festa,
G., 2007. Spectral-Element Analysis in Seismology, Advances in geophysics, 48,
365–419.

213
Bibliography

Chevrot, S., Sylvander, M., Benahmed, S., Ponsolles, C., Lefèvre, J. M., & Paradis,
D., 2007a. Source locations of secondary microseisms in western Europe: Evidence
for both coastal and pelagic sources, Journal of Geophysical Research, 112(B11),
B11301.

Chevrot, S., Sylvander, M., Benahmed, S., Ponsolles, C., Lefèvre, J. M., & Paradis,
D., 2007b. Correction to “Source locations of secondary microseisms in western
Europe: Evidence for both coastal and pelagic sources”, Journal of Geophysical
Research, 112(B12), B12398.

Clarke, D., Brenguier, F., Froger, J.-L., Shapiro, N., Peltier, A., & Staudacher, T.,
2013. Timing of a large volcanic flank movement at piton de la fournaise vol-
cano using noise-based seismic monitoring and ground deformation measurements,
Geophysical Journal International , 195(2), 1132–1140.

Cooper, R. & Longuet-Higgins, M. S., 1951. An experimental study of the pressure

variations in standing water waves, Proceedings of the Royal Society of London.
Series A. Mathematical and Physical Sciences, 206(1086), 424–435.

Courant, R., Friedrichs, K., & Lewy, H., 1928. Über die partiellen differenzengle-
ichungen der mathematischen physik, Mathematische Annalen, 100(1), 32–74.

Cristini, P. & Komatitsch, D., 2012. Some illustrative examples of the use of a
spectral-element method in ocean acoustics, The Journal of the Acoustical Society
of America, 131(3), EL229.

Cupillard, P., Delavaud, E., Burgos, G., Festa, G., Vilotte, J.-P., Capdeville, Y., &
Montagner, J.-P., 2012. Regsem: a versatile code based on the spectral element
method to compute seismic wave propagation at the regional scale, Geophysical
Journal International , 188(3), 1203–1220.

Dahlen, A. & Tromp, J., 1997. Theoretical Global Seismology, Princton, New Jersey.

Darbyshire, J., 1954. Structure of microseismic waves: Estimation of direction of

approach by comparison of vertical and horizontal components, Proceedings of the
Royal Society of London. Series A. Mathematical and Physical Sciences, 223(1152),
96–111.

214
Bibliography

Darbyshire, J., 1992. Microseisms formed off the coast of Norway, Physics of the earth
and planetary interiors, 73(3), 282–289.

De Rossi, M., 1875. Primi Risultati delle Osservazioni fatte in Roma ed in Rocca di
Papa sulle Oscillazioni Microscopiche dei Pendoli, Atti dell’Accademia Pontificia
de’ nuovi Lincei , pp. 168–204.

Deacon, G., 1947. Relations between sea waves and microseisms, Nature, (4065),
419–421.

Divins, D., 2003. Total Sediment Thickness of the World’s Oceans & Marginal Seas,
NOAA National Geophysical Data Center, Boulder .

Duncan Carr Agnew, J. B., 2007. Vertical Seismic Noise at Very Low Frequencies,
pp. 1–5.

Durek, J. J. & Ekström, G., 1996. A radial model of anelasticity consistent with long-
period surface-wave attenuation, Bulletin of the Seismological Society of America,
86(1A), 144–158.

Dziewonski, A. M. & Anderson, D. L., 1981. Preliminary reference Earth model,

Physics of the earth and planetary interiors, 25(4), 297–356.

Ebeling, C. W., 2012. Inferring Ocean Storm Characteristics from Ambient Seismic
Noise: A Historical Perspective, vol. 53, Elsevier.

Essen, H.-H., Krüger, F., Dahm, T., & Grevemeyer, I., 2003. On the generation of
secondary microseisms observed in northern and central europe, Journal of Geo-
physical Research: Solid Earth (1978–2012), 108(B10).

Faccioli, E., Maggio, F., Paolucci, R., & Quarteroni, A., 1997. 2d and 3d elastic
wave propagation by a pseudo-spectral domain decomposition method, Journal of
seismology, 1(3), 237–251.

Festa, G. & Vilotte, J.-P., 2005. The newmark scheme as velocity–stress time-
staggering: an efficient pml implementation for spectral element simulations of
elastodynamics, Geophysical Journal International , 161(3), 789–812.

Fichtner, A., 2011. Full seismic waveform modelling and inversion, Berlin, Germany.

215
Bibliography

Friedrich, A., Krüger, F., & Klinge, K., 1998. Ocean-generated microseismic noise
located with the gräfenberg array, Journal of Seismology, 2(1), 47–64.

Froment, B., Campillo, M., Chen, J., & Liu, Q., 2013. Deformation at depth asso-
ciated with the 12 may 2008 mw 7.9 wenchuan earthquake from seismic ambient
noise monitoring, Geophysical Research Letters, 40(1), 78–82.

Fukao, Y., Nishida, K., & Kobayashi, N., 2010. Seafloor topography, ocean infra-
gravity waves, and background Love and Rayleigh waves, Journal of Geophysical
Research, 115(B4), B04302.

Geldart, L. P. & Sheriff, R. E., 2004. Problems in Exploration Seismology and Their
Solutions, Society Of Exploration Geophysicists, United States of America.

Gerstoft, P. & Tanimoto, T., 2007. A year of microseisms in southern California,

Geophysical Research Letters, 34(20), L20304.

Gerstoft, P., Fehler, M. C., & Sabra, K. G., 2006. When Katrina hit California,
Geophysical Research Letters, 33(17), L17308.

Gerstoft, P., Shearer, P. M., Harmon, N., & Zhang, J., 2008. Global P, PP, and
PKP wave microseisms observed from distant storms, Geophysical Research Letters,
35(23), L23306.

Gilbert, F., 1970. Excitation of the normal modes of the Earth by earthquake sources,
Geophysical Journal of the Royal Astronomical Society, 22(2), 223–226.

Gilbert, F. & Dziewonski, A. M., 1975. An application of normal mode theory to

the retrieval of structural parameters and source mechanisms from seismic spectra,
Philosophical Transactions for the Royal Society of London. Series A, Mathematical
and Physical Sciences, pp. 187–269.

Given, H. K., 1990. Variations in broadband seismic noise at iris/ida stations in the
ussr with implications for event detection, Bulletin of the Seismological Society of
America, 80(6B), 2072–2088.

Grevemeyer, I., Herber, R., & Essen, H.-H., 2000. Microseismological evidence for a
changing wave climate in the northeast atlantic ocean, Nature, 408(6810), 349–352.

216
Bibliography

Grob, M., Maggi, A., & Stutzmann, E., 2011. Observations of the seasonality of the
antarctic microseismic signal, and its association to sea ice variability, Geophysical
Research Letters, 38(11).

Gualtieri, L., Stutzmann, E., Capdeville, Y., Ardhuin, F., Schimmel, M., Mangeney,
A., & Morelli, A., 2013. Modelling secondary microseismic noise by normal mode
summation, Geophysical Journal International , 193(3), 1732–1745.

Gualtieri, L., Stutzmann, E., Farra, V., Capdeville, Y., Schimmel, M., Ardhuin, F.,
& Morelli, A., 2014. Modelling the ocean site effect on seismic noise body waves,
Geophysical Journal International , 197(2), 1096–1106.

Gutenberg, B., 1936. On microseisms, Bulletin of the Seismological Society of Amer-

ica, 26(2), 111–117.

Gutenberg, B., 1958. Two types of microseisms, Journal of Geophysical Research,

(3), 595–597.

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

of Geophysics, 1(2), 177–210.

Haubrich, R. A. & McCamy, K., 1969. Microseisms: Coastal and pelagic sources,
Reviews of Geophysics, 7(3), 539–571.

Haubrich, R. A., Munk, W. H., & Snodgrass, F. E., 1963. Comparative spectra
of microseisms and swell, Bulletin of the Seismological Society of America, 53(1),
27–37.

Hillers, G., Graham, N., Campillo, M., Kedar, S., Landes, M., & Shapiro, N., 2012.
Global oceanic microseism sources as seen by seismic arrays and predicted by wave
action models, Geochemistry Geophysics Geosystems, 13(1), Q01021.

Hochstrasser, U. & Stoneley, R., 1937. The Transmission of Rayleigh Waves across an
Ocean Floor with Two Surface Layers, Part II: Numerical, Geophysical Supplements
to the Monthly Notices of the Royal Astronomical Society, 4(1), 197–201.

Hudson, J. A., 1962. The total internal reflection of SH waves, Geophysical Journal
International , 6(4), 509–531.

217
Bibliography

Kanamori, H. & Given, J. W., 1982. Analysis of long-period seismic waves excited by
the May 18, 1980, eruption of Mount St. Helens- A terrestrial monopole, Journal
of Geophysical Research, 87, 5422–5432.

Kedar, S., Longuet-Higgins, M., Webb, F., Graham, N., Clayton, R., & Jones, C.,
2008. The origin of deep ocean microseisms in the North Atlantic Ocean, Pro-
ceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences,
464(2091), 777–793.

Kennett, B., 1985. Seismic Wave Propagation in Stratified Media (Cambridge Mono-
graphs on Mechanics) , Cambridge,UK.

Kennett, B., 2001. The seismic wave field , Volume I: Introduction and Theoretical
Development, Cambridge,UK.

Kennett, B. N. & Engdahl, B. N., 1991. Traveltimes for global earthquake location
and phase identification, Geophysical Journal International , 105(2), 429–465.

Komatitsch, D., 1997. Méthodes spectrales et éléments spectraux pour l’équation de

l’élastodynamique 2 D et 3 D en milieu hétérogène, Ph.D. thesis.

Komatitsch, D. & Tromp, J., 2000. The Spectral Element method for three-
dimensional seismic wave propagation, Society of Exploration Geophysicists - Con-
ference paper , pp. 197–192.

Komatitsch, D. & Tromp, J., 2002. Introduction to the spectral element method
for three-dimensional seismic wave propagation, Geophysical Journal of the Royal
Astronomical Society, 139(3), 806–822.

Komatitsch, D. & Vilotte, J.-P., 1998. The spectral element method: An efficient
tool to simulate the seismic response of 2d and 3d geological structures, Bulletin of
the seismological society of America, 88(2), 368–392.

Komatitsch, D., Barnes, C., & Tromp, J., 2000. Wave propagation near a fluid-solid
interface: A spectral-element approach, GEOPHYSICS , 65(2), 623–631.

Komatitsch, D., Tsuboi, S., Ji, C., & Tromp, J., 2003. A 14.6 billion degrees of
freedom, 5 teraflops, 2.5 terabyte earthquake simulation on the earth simulator, in
Proceedings of the 2003 ACM/IEEE conference on Supercomputing, p. 4, ACM.

218
Bibliography

Komatitsch, D., Tsuboi, S., & Tromp, J., 2005. The spectral-element method in
seismology, Geophysical Monograph-AGU , 157, 205.

Komen, G., Cavalieri, L., Donelan, M., Hasselmann, K., Hasselmann, S., & P.A.E.M,
J., 1994. Dynamics and Modelling of Ocean Waves, Cambridge University Press.

Koper, K. D. & de Foy, B., 2008. Seasonal Anisotropy in Short-Period Seismic Noise
Recorded in South Asia, Bulletin of the Seismological Society of America, 98(6),
3033–3045.

Koper, K. D., de Foy, B., & Benz, H., 2009. Composition and variation of noise
recorded at the Yellowknife Seismic Array, 1991–2007, Journal of Geophysical Re-
search, 114(B10), B10310.

Koper, K. D., Seats, K., & Benz, H., 2010. On the Composition of Earth’s Short-
Period Seismic Noise Field, Bulletin of the Seismological Society of America,
100(2), 606–617.

Kundu, P. K., Cohen, I. M., & Dowling, D. R., 2012. Fluid mechanics, Accademic
Press, New York, NY.

Kurrle, D. & Widmer-Schnidrig, R., 2008. The horizontal hum of the Earth: A global
background of spheroidal and toroidal modes, Geophysical Research Letters, 35(6),
L06304.

Lacoss, R., Kelly, E., & Toksöz, M., 1969. Estimation of seismic noise structure using
arrays, Geophysics, 34(1), 21–38.

Landès, M., Hubans, F., Shapiro, N. M., Paul, A., & Campillo, M., 2010. Origin of
deep ocean microseisms by using teleseismic body waves, Journal of Geophysical
Research, 115(B5), B05302.

Lay, T. & Wallace, T., 1995. Modern Global Seismology, San Diego, California.

Leet, L. D., 1947. Microseisms in new england-case history of a storm, Geophysics,

12(4), 639–650.

Longuet-Higgins, M., 1953. On the decrease of velocity with depth in an irrotational

water wave, in Proc. Camb. Phil. Soc, vol. 46, pp. 552–560, Cambridge Univ Press.

219
Bibliography

Longuet-Higgins, M. S., 1950. A theory of the origin of microseisms, Philosophical

Transactions of the Royal Society of London. Series A. Mathematical and Physical
Sciences, pp. 1–35.

Longuet-Higgins, M. S., 1952. Can sea waves cause microseisms?, Symposium on

microseisms held at Arden House, pp. 74–93.

Longuet-Higgins, M. S. & Ursell, F., 1948. Sea waves and microseisms, Nature,
162(4122), 700.

Maday, Y. & Patera, A. T., 1989. Spectral element methods for the incompressible
navier-stokes equations, in IN: State-of-the-art surveys on computational mechanics
(A90-47176 21-64). New York, American Society of Mechanical Engineers, 1989,
p. 71-143. Research supported by DARPA., vol. 1, pp. 71–143.

Maupin, V., 2007. Introduction to mode coupling methods for surface waves, pp.
127–155, Elsevier.

McGarr, A., 1969. Amplitude variations of rayleigh waves-propagation across a conti-

nental margin, Bulletin of the Seismological Society of America, 59(3), 1281–1305.

McNamara, D. E. & Buland, R. P., 2004. Ambient noise levels in the continental
United States, Bulletin of the Seismological Society of America, 94(4), 1517–1527.

Miche, A., 1944. Mouvements ondulatoires de la mer an profondeur croissante ou

décroissante. Premiére partie. Mouvements ondulatoires périodiques et cylindriques
en profondeur constante, Annales des Ponts at Chaussés, 114, 42–78.

Milne, J., 1883. Earth tremors, Transactions of the Seismological Society of Japan,
7(1), 1–15.

Mordret, A., Jolly, A., Duputel, Z., & Fournier, N., 2010. Monitoring of phreatic
eruptions using interferometry on retrieved cross-correlation function from ambient
seismic noise: Results from mt. ruapehu, new zealand, Journal of Volcanology and
Geothermal Research, 191(1), 46–59.

Mordret, A., Landès, M., Shapiro, N., Singh, S., Roux, P., & Barkved, O., 2013. Near-
surface study at the valhall oil field from ambient noise surface wave tomography,
Geophysical Journal International , 193(3), 1627–1643.

220
Bibliography

Nawa, K., Suda, N., Fukao, Y., Sato, T., Aoyama, Y., & Shibuya, K., 1998. Incessant
excitation of the Earth’s free oscillations, Earth Planets Space, 50(1), 3–8.

Nishida, K., 2013. Earth’s background free oscillations, Annual Review of Earth and
Planetary Sciences, 41, 719–740.

Nishida, K., Kawakatsu, H., Fukao, Y., & Obara, K., 2008. Background Love and
Rayleigh waves simultaneously generated at the Pacific Ocean floors, Geophysical
Research Letters, 35(16), L16307.

Nishida, K., Montagner, J. P., & Kawakatsu, H., 2009. Global Surface Wave Tomog-
raphy Using Seismic Hum, Science, 326(5949), 112–112.

Obermann, A., Planès, T., Larose, E., & Campillo, M., 2013. Imaging preeruptive
and coeruptive structural and mechanical changes of a volcano with ambient seismic
noise, Journal of Geophysical Research: Solid Earth, 118(12), 6285–6294.

Obrebski, M., Ardhuin, F., Stutzmann, E., & Schimmel, M., 2013. Detection of
microseismic compressional (p) body waves aided by numerical modeling of oceanic
noise sources, Journal of Geophysical Research: Solid Earth, 118(8), 4312–4324.

Obrebski, M. J., Ardhuin, F., Stutzmann, E., & Schimmel, M., 2012. How moderate
sea states can generate loud seismic noise in the deep ocean, Geophysical Research
Letters, 39(11), L11601.

Park, J., Vernon, F. L., & Lindberg, C. R., 1987. Frequency dependent polarization
analysis of high-frequency seismograms, Journal of Geophysical Research: Solid
Earth (1978–2012), 92(B12), 12664–12674.

Patera, A. T., 1984. A spectral element method for fluid dynamics: laminar flow in
a channel expansion, Journal of computational Physics, 54(3), 468–488.

Peterson, J., 1993. Observations and modeling of seismic background noise, SGS
Open-File Report 93-322 , pp. 1–95.

Phinney, R. A. & Burridge, R., 1973. Representation of the elastic-gravitational ex-

citation of a spherical earth model by generalized spherical harmonics, Geophysical
Journal of the Royal Astronomical Society, 34(4), 451–487.

221
Bibliography

Prieto, G. A., Denolle, M., Lawrence, J. F., & Beroza, G. C., 2011. On amplitude
information carried by the ambient seismic field, Comptes rendus - Geoscience,
343(8-9), 600–614.

Ramirez, J. E., 1940a. An experimental investigation of the nature and origin of

microseisms at St. Louis, Missouri Part one, Bulletin of the Seismological Society
of America, 30(1), 35–84.

Ramirez, J. E., 1940b. An experimental investigation of the nature and origin of

microseisms at St. Louis, Missouri Part two, Bulletin of the Seismological Society
of America, 30(2), 139–178.

Rhie, J. & Romanowicz, B., 2004. Excitation of earth’s continuous free oscillations
by atmosphere–ocean–seafloor coupling, Nature, 431(7008), 552–556.

Rhie, J. & Romanowicz, B., 2006. A study of the relation between ocean storms and
the Earth’s hum, Geochemistry Geophysics Geosystems, 7(10), Q10004.

Rind, D. & Down, W., 1979. Microseisms at palisades: 2. rayleigh wave and love
wave characteristics and the geologic control of propagation, Journal of Geophysical
Research: Solid Earth (1978–2012), 84(B10), 5632–5642.

Rivet, D., Campillo, M., Radiguet, M., Zigone, D., Cruz-Atienza, V., Shapiro, N. M.,
Kostoglodov, V., Cotte, N., Cougoulat, G., Walpersdorf, A., et al., 2014. Seismic
velocity changes, strain rate and non-volcanic tremors during the 2009–2010 slow
slip event in guerrero, mexico, Geophysical Journal International , 196(1), 447–460.

Sabra, K. G., Gerstoft, P., Roux, P., Kuperman, W., & Fehler, M. C., 2005. Extract-
ing time-domain green’s function estimates from ambient seismic noise, Geophysical
Research Letters, 32(3).

Saito, T., 2010. Love-wave excitation due to the interaction between a propagating
ocean wave and the sea-bottom topography, Geophysical Journal International ,
182(3), 1515–1523.

Schimmel, M., Stutzmann, E., & Gallart, J., 2010. Using instantaneous phase co-
herence for signal extraction from ambient noise data at a local to a global scale,
Geophysical Journal International , 184(1), 494–506.

222
Bibliography

Schimmel, M., Stutzmann, E., Ardhuin, F., & Gallart, J., 2011. Polarized Earth’s
ambient microseismic noise, Geochemistry Geophysics Geosystems, 12(7), Q07014.

Scholte, J., 1947. The range of existence of rayleigh and stoneley waves, Geophysical
Journal International , 5(s5), 120–126.

Schulte-Pelkum, V., Earle, P. S., & Vernon, F. L., 2004. Strong directivity of ocean-
generated seismic noise, Geochemistry Geophysics Geosystems, 5(3), Q03004.

Sens Schönfelder, C. & Wegler, U., 2011. Passive image interferometry for monitoring
crustal changes with ambient seismic noise, Comptes rendus - Geoscience, 343(8),
639–651.

Seriani, G., Priolo, E., & Pregarz, A., 1995. Modelling waves in anisotropic media by
a spectral element method, in Proceedings of the third international conference on
mathematical and numerical aspects of wave propagation, pp. 289–298, SIAM.

Shapiro, N. M., Campillo, M., Stehly, L., & Ritzwoller, M. H., 2004. Emergence of
broadband Rayleigh waves from correlations of the ambient seismic noise, Geophys-
ical Research Letters, 307(5715), 1615–1618.

Shapiro, N. M., Campillo, M., Stehly, L., & Ritzwoller, M. H., 2005. High-resolution
surface-wave tomography from ambient seismic noise, Science, 307(5715), 1615–
1618.

Shearer, P. M., 2009. Introduction to seismology, Cambridge University Press.

Snieder, R., 2004. Monitoring change in volcanic interiors using coda wave interfer-
ometry: Application to Arenal Volcano, Costa Rica, Geophysical Research Letters,
31(9), L09608.

Stehly, L., Campillo, M., & Shapiro, N. M., 2006. A study of the seismic noise from
its long-range correlation properties, Journal of Geophysical Research, 111(B10),
B10306.

Stoneley, R., 1926. The Effect of the Ocean on Rayleigh Waves., Geophysical Journal
of the Royal Astronomical Society, 1(s7), 349–356.

Stoneley, R., 1957. The transmission of Rayleigh waves across an ocean floor with two
surface layers Part I: Theoretical, Bulletin of the Seismological Society of America,
47(1), 7–12.

223
Bibliography

Stutzmann, E., Roult, G., & Astiz, L., 2000. GEOSCOPE station noise levels, Bul-
letin of the Seismological Society of America, 90(3), 690–701.

Stutzmann, E., Schimmel, M., Patau, G., & Maggi, A., 2009. Global climate imprint
on seismic noise, Geochemistry Geophysics Geosystems, 10(11), Q11004.

Stutzmann, E., Ardhuin, F., Schimmel, M., Mangeney, A., & Patau, G., 2012. Mod-
elling long-term seismic noise in various environments, Geophysical Journal Inter-
national , 191(2), 707–722.

Suda, Naka, & Fukao, 1998. Earth’s Background Free Oscillation, Science, 279, 1–3.

Takeuchi, H. & Saito, M., 1972. Seismic surface waves, Methods in computational
physics, 11, 217–295.

Tanimoto, T., 2007. Excitation of microseisms, Geophysical Research Letters, 34(5),

L05308.

Tanimoto, T., 2013. Excitation of microseisms: views from the normal-mode ap-
proach, Geophysical Journal of the Royal Astronomical Society, 194(3), 1755–1759.

Tanimoto, T., Um, J., Nishida, K., & Kobayashi, N., 1998. Earth’s continuous
oscillations observed on seismically quiet days, Geophysical research letters, 25(10),
1553–1556.

Tanimoto, T., Ishimaru, S., & Alvizuri, C., 2006. Seasonality in particle motion of
microseisms, Geophysical Journal International , 166(1), 253–266.

Tian, Y., Shen, W., & Ritzwoller, M. H., 2013. Crustal and uppermost mantle shear
velocity structure adjacent to the juan de fuca ridge from ambient seismic noise,
Geochemistry, Geophysics, Geosystems, 14(8), 3221–3233.

Toksöz, M. N. & Lacoss, R. T., 1968. Microseisms: Mode structure and sources,
Science, 159(3817), 872–873.

Tolman, H. L., 1991. A third-generation model for wind waves on slowly varying, un-
steady, and inhomogeneous depths and currents, Journal of Physical Oceanography,
21(6), 782–797.

Tolman, H. L., 2009. User manual and system documentation of WAVEWATCH III
TM version 3.14, Technical note, MMAB Contribution, (276).

224
Bibliography

Traer, J. & Gerstoft, P., 2014. A unified theory of microseisms and hum, Journal of
Geophysical Research: Solid Earth, 119, doi:10.1002/2013JB010504.

Traer, J., Gerstoft, P., Bromirski, P. D., & Shearer, P. M., 2012. Microseisms and
hum from ocean surface gravity waves, Journal of Geophysical Research, 117(B11),
B11307.

Tromp, J., Komattisch, D., & Liu, Q., 2008. Spectral-element and adjoint methods
in seismology, Communications in Computational Physics, 3(1), 1–32.

Uchiyama, Y. & McWilliams, J. C., 2008. Infragravity waves in the deep ocean: Gen-
eration, propagation, and seismic hum excitation, Journal of Geophysical Research,
113(C7), C07029.

Vigness, I., Kammer, E., Dinger, J., & Irving, L., 1952. Sea swell and microseisms,
in Journal of Atmospheric Sciences, vol. 9, pp. 443–444.

Vinnik, L. P., 1973. Sources of microseismicP waves, Pure and Applied Geophysics,
103(1), 282–289.

Webb, S. C., 1992. The equilibrium oceanic microseism spectrum, The Journal of the
Acoustical Society of America, 92, 2141.

Webb, S. C., 1998. Broadband seismology and noise under the ocean, Reviews of
Geophysics, 36(1), 105–142.

Webb, S. C., 2007. The Earth’s ‘hum’ is driven by ocean waves over the continental
shelves, Nature, 445(7129), 754–756.

Webb, S. C., 2008. The Earth’s hum: the excitation of Earth normal modes by ocean
waves, Geophysical Journal International , 174(2), 542–566.

Webb, S. C. & Constable, S. C., 1986. Microseism propagation between two sites on
the deep seafloor, Bulletin of the Seismological Society of America, 76(5), 1433–
1445.

Wegler, U. & Sens-Schönfelder, C., 2007. Fault zone monitoring with passive image
interferometry, Geophysical Journal International , 168(3), 1029–1033.

Wiechert, E., 1904. Verhandlung der zweiten internationalen seismologischen kon-

ferenz, Beitrage zur Geophysik , 2, 41–43.

225
Bibliography

Woodhouse, J. H., 1988. The calculation of eigenfrequencies and eigenfunctions of the

free oscillations of the earth and the sun., In D. J. Doornbos (Ed.), Seismological
algorithms, Academic Press, New York., pp. 321–370.

Woodhouse, J. H. & Girnius, T. P., 1982. Surface waves and free oscillations in a
regionalized Earth model, Geophysical Journal of the Royal Astronomical Society,
68(3), 653–673.

Yang, H.-Y., Zhao, L., & Hung, S.-H., 2010. Synthetic seismograms by normal-
mode summation: a new derivation and numerical examples, Geophysical Journal
International , 183(3), 1613–1632.

Yang, Y. & Ritzwoller, M. H., 2008. Characteristics of ambient seismic noise as a

source for surface wave tomography, Geochemistry Geophysics Geosystems, 9(2).

Yao, H., Gouédard, P., Collins, J. A., McGuire, J. J., & van der Hilst, R. D., 2011.
Structure of young East Pacific Rise lithosphere from ambient noise correlation
analysis of fundamental- and higher-mode Scholte-Rayleigh waves, Comptes rendus
- Geoscience, 343(8-9), 571–583.

Young, I. R., 2003. A review of the sea state generated by hurricanes, Marine struc-
tures, 16(3), 201–218.

Zhang, J., Gerstoft, P., & Shearer, P. M., 2009. High-frequency P-wave seismic noise
driven by ocean winds, Geophysical Research Letters, 36(9), L09302.

Zhang, J., Gerstoft, P., & Bromirski, P. D., 2010a. Pelagic and coastal sources
of P-wave microseisms: Generation under tropical cyclones, Geophysical Research
Letters, 37(15), L15301.

Zhang, J., Gerstoft, P., & Shearer, P. M., 2010b. Resolving P-wave travel-time anoma-
lies using seismic array observations of oceanic storms, Earth and Planetary Science
Letters, 292(3-4), 419–427.

226