Ambient Noise Multimode Surface Wave Tomography: Review Open Access

Nishida et al.
Progress in Earth and

Progress in Earth and Planetary Science (2024) 11:4
https://doi.org/10.1186/s40645-023-00605-8 Planetary Science
REVIEW Open Access
Ambient noise multimode surface wave

tomography
Kiwamu Nishida1* , Ryota Takagi2 and Akiko Takeo1
Abstract
Seismic techniques using earthquakes are powerful tools for exploring the Earth’s internal structure. However,
the earthquake distribution limits the spatial resolution. In recent years, ambient noise surface wave tomography
using ambient seismic wave field has resolved these limitations. A typical ambient seismic wave field is microseisms
excited by ocean swell activities. Ambient noise surface wave tomography is a technique in seismic interferometry
that extracts seismic wave propagation between pairs of stations by cross-correlating the seismic records. The cross-
correlation function can be interpreted as an impulsive response at a station with a virtual source at the other station.
This technique became standard with the development of modern dense seismic networks. This paper reviews a the-
ory of seismic interferometry for ambient noise surface wave tomography and procedures for practical data process-
ing to calculate cross-correlation functions. The tomographic method typically consists of four steps: (1) the construc-
tion of reference 1-D models, (2) phase velocity measurements for each path, (3) 2-D phase velocity inversions, and (4)
the construction of a 3-D S-wave tomographic model obtained from series of local 1-D inversions at all the grids. This
paper presents the feasibility of multimode surface wave dispersion measurements for improving depth resolution.
Keywords Ambient seismic noise, Multimode surface wave, Seismic interferometry, Surface wave tomography
1 Introduction frequently. The source and station distributions limit the

Seismic techniques are powerful tools for exploring the spatial resolution of the tomographic images.
Earth’s internal structure from local to global scales. Even on seismically quiet days, the Earth persistently
Earthquakes are a primary illumination source of the oscillates because ocean swells excite seismic waves. They
Earth’s interior. Seismic tomography using earthquakes are known as microseisms (e.g., Nishida 2017), which is a
revealed the lateral heterogeneities of the Earth on vari- kind of ambient seismic wave field (also known as ambi-
ous scales (e.g., Romanowicz 2003; Thurber and Ritsema ent seismic noise). Seismic interferometry (SI) provided
2015). Although the method has produced many signifi- a clue to overcome the limitations due to the earthquake
cant results, it has inherent limitations attributed to the distribution because SI produces a virtual earthquake
earthquake distribution. Earthquakes occur only in tec- record from the ambient seismic wave field. An applica-
tonically active areas, and large earthquakes do not occur tion of SI for imaging Earth’s internal structure is known
as ambient noise surface wave tomography (ANT). This
section briefly introduces SI, ambient seismic wave field,
*Correspondence: and ANT followed by the scope of this paper.
Kiwamu Nishida
knishida@eri.u-tokyo.ac.jp
1
Earthquake Research Institute, The University of Tokyo, 1‑1‑1 Yayoi, 1.1 Brief history of seismic interferometry
Bunkyo‑ku, Tokyo 113‑0032, Japan
2
In seismology, an origin of SI is Aki’s Ph.D. thesis (Aki
Research Center for Prediction of Earthquakes and Volcanic Eruptions,
Graduate School of Science, Tohoku University, 6‑6 Aza‑Aoba, Aramaki, 1957), inspired by the seminal book Cybernetics (Wie-
Aoba‑ku, Sendai 980‑8578, Japan ner 1947). He proposed the SPatial AutoCorrelation
(SPAC) method; later studies show that the method is
© The Author(s) 2024. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://creativecommons.org/licenses/by/4.0/.
Nishida et al. Progress in Earth and Planetary Science (2024) 11:4 Page 2 of 42
mathematically equivalent to SI (e.g., Chávez-Garía and seismic hum, through the topographic coupling on the
Luzón 2005; Harmon et al. 2010; Tsai and Moschetti seafloor (Nishida et al. 2008b; Fukao et al. 2010; Nishida
2010) under certain conditions. Aki’s idea had not been 2013). Ocean swell activities excite microseisms from
paid attention until the 1980s. After 26 years, a group at 0.05 to 0.5 Hz (Nishida 2017). Primary microseisms are
Hokkaido University (Okada and Sakajiri 1983) devel- excited by shoaling waves through the topographic cou-
oped a survey method for shallow structure using micro- pling on the sea floor (e.g., Hasselmann 1963; Ardhuin
tremor, now known as the microtremor survey. After this et al. 2015). Secondary microseisms are excited by ocean
work, this method became popular for surveying sub- swell activities in both the pelagic and coastal regions
surface structures in seismic engineering (e.g., Cho et al. through the nonlinear interaction (Longuet-Higgins
2006), which was also extended to multimode inversion 1950; Hasselmann 1963). In the history of earthquake
(Tokimatsu et al. 1992). seismology, secondary microseisms have been a promi-
A key for SI is a cross-correlation function (CCF) nent noise source for earthquake signals. Historical
between a pair of stations. The CCF exhibits an impul- seismological studies took different approaches depend-
sive response at a station by a virtual source at the other ing on two frequency bands below and above the typical
station. The idea of SI dates back to the 1950s in differ- frequency of secondary microseisms. This frequency is
ent research fields: acoustic (Eckart 1953), ocean acous- also important in characterizing the physical nature of
tic (Cox 1973), seismic exploration (Claerbout 1968), seismic waves (Aki 2003). Above the frequency, strong
and seismology (Aki 1957). Although the ideas were pro- lateral heterogeneities in the crust and sediment make
posed independently, they are mathematically identical the seismic wave field complex on a regional or global
under certain conditions (e.g., Chávez-Garía and Luzón scale (see Sect. 6.2 for details), stimulating the develop-
2005; Harmon et al. 2010; Tsai and Moschetti 2010). SI ment of a stochastic approach. Below the frequency, the
is applied not only to the Earth but also to experimen- waveform can be reproduced in a deterministic manner.
tal studies, such as ultrasonic waves (Lobkis and Weaver SI turns stochastic wave fields into a deterministic signal
2001), building responses (Snieder and Wapenaar 2010; as a virtual seismic record, which can be applied to seis-
Nakata et al. 2013), helioseismology (Gizon and Birch mic tomography. Thus, the operation of cross-correlat-
2002; Duvall et al. 1993; Hanasoge et al. 2016) and ocean ing seismic wave fields, or SI, bridges the stochastic and
acoustic (Cox 1973). Later studies (e.g., Godin 2007; deterministic approaches in seismology, which had been
Snieder and Larose 2013) show that the principle of SI incompatible, as noted by Aki (2003).
is akin to the fluctuation–dissipation theorem in phys-
ics (e.g., Callen and Welton 1951), which was generalized 1.3 ANT
as linear response theory by Kubo (1957). The theorem Campillo and Paul (2003) demonstrated that SI can
establishes a connection between the response to an extract virtual seismic records from the ambient seismic
external force and the CCF of specific fluctuating proper- wave field. This paper describes the analysis of the coda
ties when the system reaches thermal equilibrium. wave, a diffusive wave field, in Mexico. Cross-correlating
the seismic records of coda waves extracted clear Love
1.2 Ambient seismic wave field and Rayleigh wave propagations between every pair of
Ocean wave activities excite ambient seismic wave fields, stations. Although this paper analyzed earthquake coda,
even on seismically quiet days. Based on the types of the method can be applied to ambient seismic wave
ocean surface gravity waves and excitation mechanisms, fields. Shapiro and Campillo (2004) subsequently showed
we classified ambient seismic wave fields into (1) seismic that the CCFs of the microseisms also exhibited clear
hum from 1 × 10−3 to 0.02 Hz, (2) primary microseisms surface propagations between every pair of stations. This
from 0.02 to 0.1 Hz, and (3) secondary microseisms result suggested the possibility of seismic imaging with-
between 0.1 and 0.5 Hz (Nishida 2017). out earthquakes, now known as ANT.
The ocean surface gravity waves can be classified as Shapiro et al. (2005) achieved the milestone of the
ocean infragravity (IG) waves below 0.02 Hz and ocean ANT studies. They demonstrated the feasibility of ANT
swell above the frequency. IG wave is a shallow-water using a modern dense seismic network. After this study,
wave, whereas the ocean swell is physically a deep-water ANT became a standard technique for dense broad-
wave. The pressure fluctuations of the IG waves reach band observations because ANT does not require wait-
the seafloor in the pelagic and coastal regions, whereas ing for earthquakes, i.e., a long observation period. After
those of the ocean swells cannot reach the seafloor in the the paper, ANT was applied to many regions: the USA
pelagic regions. (e.g., Bensen et al. 2007; Moschetti et al. 2007; Liang and
IG waves excite background Love waves and Ray- Langston 2008; Lin et al. 2008; Yang et al. 2008), Australia
leigh waves predominantly from 1 to 20 mHz, known as (e.g., Saygin and Kennett 2010), Europe (e.g., Yang et al.
2007), China (e.g., Zheng et al. 2008), Japan (e.g., Nimiya noise cross-correlations (e.g., Lin et al. 2012b; Zhou et al.
et al. 2020; Nishida et al. 2009), and the global scale 2020; Liu et al. 2021) have made it reliable. Because this
(Nishida et al. 2009; Haned et al. 2016). With the increase topic is beyond the scope of this paper, we refer only to
in seismic stations with continuous observations at many the aforementioned papers.
seismic stations in the 2000s, ANT has become a stand- Section 2 summarizes a theory of SI for ANT. Sec-
ard analysis method. Today, ANT is first applied after tion 3 describes the procedures for practical data pro-
deploying a dense seismic array. cessing to calculate CCFs. Ambient noise multimode
surface wave tomography (multimode ANT) consists of
1.4 What is covered/not covered in this review four steps (Fig. 1). The first step (Sect. 4) is the measure-
There are already many review papers on SI and ment of multimode surface wave dispersion for a local
ANT (e.g., Snieder and Larose 2013; Wapenaar et al. seismic array. The dispersion curves are also crucial for
2010a, 2010b) and textbooks (e.g., Schuster 2009, Ritz- constructing local 1-D structures, which can be an initial
woller and Feng 2019, Nakata et al. 2019, and Chapter 10 model of ANT. The second step (Sect. 5) is the dispersion
in Sato et al. 2012). This review focuses on a consistent measurements for each path. The third and fourth steps
theoretical treatment and systematic comparison among (Sects. 6 and 7) are how to infer 3-D seismic velocity
different phase velocity measurement methods. models from the dispersion measurements. The proce-
This review also focuses on multimode measure- dures consist of two steps: (1) the 2-D inversion of phase/
ments. Recent developments in dense arrays enable us group velocities in Sect. 6 and (2) a local 1-D inversion
to extract multimode dispersion (e.g., Spica et al. 2018; at each grid in Sect. 7. Figure 1 shows such procedures
Chmiel et al. 2019; Savage et al. 2013; Jiang and Denolle based on our previous studies (Nishida et al. 2008a;
2022; Socco et al. 2010). Multimode inversions have sig- Nagaoka et al. 2012; Takagi and Nishida 2022; Takeo
nificantly improved depth resolution, although the dis- et al. 2022; Yamaya et al. 2021). This review will focus on
persion of fundamental mode branches alone has poor and compare our previous studies with other studies.
vertical resolution. As multimode dispersion measure-
ment has inherent difficulties, this review aims to provide
clues for practical applications. 2 A brief review of SI
Because SI does not require earthquakes, it acceler- Theories of SI originate from various backgrounds and
ates the progress in monitoring the temporal change in have developed independently. Although they require
seismic velocities. Monitoring the temporal change with different assumptions in different settings (e.g., an open
events requires repeating earthquakes or repeating active or closed system), they are closely related. Theoretically,
sources. In most cases, they are not realistic. In the last it is natural to consider a closed system for the global
ten years, SI has become a standard technique for moni- scale, whereas it is natural to consider an open system for
toring the temporal change in seismic velocities asso- the regional or local scale. This section provides an over-
ciated with environmental origins (e.g., precipitation, view of the theories of SI in a closed system and an open
groundwater, and temperature), volcanic eruptions, and system to understand the physical pictures. To model
earthquakes (e.g., Sens-Schönfelder and Wegler 2006a; CCFs for ANT, we formulate synthetic CCFs for the
Wegler et al. 2007; Brenguier et al. 2008a, b, Wang et al. homogeneous source distribution in a 2-D homogeneous
2017). Because this review does not cover such topics, see medium. However, the heterogeneous source distribu-
review articles (e.g., Sens-Schönfelder and Wegler 2011; tion realistically causes an apparent travel-time anomaly
Obermann and Hillers 2019) for further information. from the synthetic CCF for the homogeneous source dis-
This paper does not cover conventional surface wave tribution. We evaluate such biases based on the analytic
tomography using earthquake data. Because many excel- formulation.
lent reviews are already available (e.g., Romanowicz
2003, 2020, 2021; Laske and Widmer-Schnidrig 2015;
Levshin et al. 2018; Barmin et al. 2001), please refer to 2.1 SI in a closed system
those references. Here, we consider SI in a closed system. In the case of a
This paper also does not cover attenuation tomogra- finite body, we evaluate CCFs based on a normal mode
phy using CCFs of ambient seismic noise. Some studies approach (Lobkis and Weaver 2001). Because the Earth
inferred the attenuation structure from the amplitude is a finite-size sphere, this approach is also feasible for
information of CCFs (Prieto et al. 2009, 2011; Lin et al. multi-orbit propagations on a global scale (Nishida et al.
2011). Although the effects of source heterogeneities 2002, 2009). For simplicity, this subsection describes a
(e.g., Tsai 2011) can bias the estimation, recent develop- scalar 1-D case, but it can be easily extended to elastic
ments to extract amplitude information from ambient 2-D and 3-D cases.
(a) 1st step: secion 4 (b) 2nd step: section 5

Construction of a reference Phase velosity anomaly
dispersion curves of an array for station pairs
ZZ, ZR, ZZ, ZR,
Distance
Distance
RR, TT RR, TT
Lagtime Lagtime
Reference dispersion curves Dispersion curves for each path

Rayleigh wave Love wave Rayleigh wave Love wave
Phase vel.
Phase vel.
Phase vel.
Phase vel.
Frequency Frequency Frequency Frequency
(c) 3nd step: section 6 (d) 4th step: section 7

Phase velosity tomography 1D inversion at all the grid points
0th Rayleigh 1st Rayleigh
3D Vs structure
1D inversion
Fig. 1 a Construction of reference 1-D models. b Phase/group velocity measurements for each path. c 2-D phase/group velocity inversions
for multimodes. d A 3-D S-wave tomographic model is obtained from a collection of local 1-D inversions at all the grids. Modified from Takagi
and Nishida (2022)
2.1.1 Repeating seismic experiments where κ is the elastic constant, ρ is the density and uk is
Virtually, we consider repeating seismic experiments in the displacement at the kth experiment. After the force
a closed system of a perfect elastic body. Before the initial is applied, the displacement is measured at the stations.
time t = 0, the body did not deform, and a random force These experiments are repeated K times. Note that we
f k (x) was applied to the body at t = 0 in the kth experi- cannot consider a persistent force in this system because
ment, where x is the spatial location. We consider a finite no attenuation leads to an infinite increase in the ampli-
body from x = 0 to x = L and impose rigid or free bound- tude over time.
ary conditions at both ends. The equation of motions is We evaluated the displacement uk (x, t) by a convolu-
given by tion between the force f k (x) and the Green’s function.
The Green’s function can be written in terms of normal
∂ 2 uk (x, t) ∂ 2 uk (x, t)
ρ(x) = κ(x) + f k (x)δ(t), (1) mode theory (Dahlen and Tromp 1998) as
∂t 2 ∂x2
Un (x)Un (x′ ) where I is the number of the force. We also assume that
G(x, x′ ; t) = sin(ωn t), t ≥ 0, (2) Fik (x) is white noise as
n
ωn
where ωn is nth eigenfrequency and Un is nth eigenfunc- �Fik Fik′ �k = F̄ 2 δii′ , (10)
tion, which satisfies orthonormality: where k is ensemble average with respect to k. To sim-
L plify the problem, we consider the constant density ρ0.
ρ(x)Un (x)Un′ (x)dx = δnn′ . (3) The expected value of the cross-correlation of An can be
0
evaluated as follows.
Then, the displacement can be represented by the
I F̄ 2 L I F̄ 2

convolution k k
An An′ ∼ Un (x)Un′ (x)dx = δnn′ ≡ E δnn′ ,
k Lρ0 0 Lρ0
sin(ωn t) (11)
uk (x, t) = Akn Un (x) , (4)
n
ωn where E is the modal energy. The amplitudes of the dif-
L ferent modes An do not correlate with each other, and
where Akn = 0 Un (x)f k (x)dx . We consider the energy the total energy of each mode is distributed equally. The
partition of the modes for a random external force in the expected value of the modal energy is constant for each
next section. mode when the external force f is white noise. When a
system meets this condition, we call the state the equi-
2.1.2 Energy partition of modal energy partition of energy. We note that the fluctuation–dis-
This subsection discusses the energy balance of each sipation theorem in physics (Callen and Welton 1951)
mode: how the work done by external forces is distrib- requires a thermal equilibrium that satisfies the equi-
uted to the kinematic and elastic energy. First, we evalu- partition of energy. Equation 10 can be derived from the
ate the work done by the external force can be given by principle of equal a priori probabilities in the case.
∞ L
vnk (x, t)f k (x)δ(t)dxdt = (Ank )2 , (5) 2.1.3 CCFs under the equipartition of energy
0 0 Here we define a CCF φ k (x1 , x2 ; τ ) between uk (x1 ) and
where the particle velocity vnk of nth mode in the kth uk (x2 ) under the equipartition of energy as,
experiment is written by vnk = Akn Un (x) cos(ωn t). 1 T k

k
The kinetic energy of nth mode Tn can be evaluated by φ (x1 , x2 ; τ ) ≡ lim u (x1 , t)uk (x2 , t + τ )dt,
T →∞ T 0
integrating the kinetic energy density in space as
(12)
where x1 shows the location of station 1, and x2 shows
L
ρ k 2 1
Tn = (vn ) dx = (Ank )2 cos2 (ωn t). (6) that of station 2. We note that there are two types of CCF
0 2 2
definitions. The sign of the second term of the other type
The elastic energy Vn, on the other hand, can be evalu- is flipped.
ated by integrating the strain energy density in space. The The ensemble average of φ(x1 , x2 ; t) over K time experi-
partial integral with the boundary condition leads to ments is defined by the ensemble average φ k (x1 , x2 ; τ ) k .

L 2 Then, the CCF can be represented by

κ k ∂ Un sin(ωn t) 1
Vn = An dx = (Ank )2 sin2 (ωn t). d E
0 2 ∂x ωn 2 φ(x1 , x2 ; τ ) = − (G(x1 , x2 ; τ ) − G(x2 , x1 ; −τ )),
(7) dτ 2
(13)
The total energy Tn + Vn is
which relates the CCF to Green’s function (Snieder
Tn + Vn = (Akn )2 , (8) 2004). The frequency dependence of this equation differs
slightly from Aki’s formulation (e.g., Haney et al. 2012).
which balances the work done by external forces. Although the formulation showed that the Hilbert trans-
Here, we consider an excitation by random force Fik at form of the CCF can be related to Green’s function, the
location xi as difference can be attributed to spectral normalization.
I−1 Compared to CCFs for real data, the biggest problem
is the assumption of the equipartition of energy. Because

f k (x) = Fik δ(xi ), (9)
i=0 the excitation sources are distributed near the surface,
fundamental modes dominate the observed wave field.
Energy is not partitioned equally among different mode open system. When external force acts on the body at a
branches. Indeed, the observed dominance of fundamen- location r and time t = 0, the 2-D wave equation is given
tal modes (e.g., Nishida 2017) shows that the energy is by
not equally distributed in the radial direction. When con- ∞
sidering a single-mode branch, the energy is distributed ∂ 2 ψ(r, t)
2
= C(r)∇ 2 ψ(r, t) + δ(ri − r)δ(t)fi
equally in the horizontal direction if the external force is ∂t
i=0
white noise distributed on the entire surface. (14)
Although the theory assumed K experiments, practi- where C(r) is phase velocity, and fi is ith external force
cally, only one observation is possible. Therefore, the at the location ri (i = 0, · · · , ∞). The potential ψ(x, t) is
ensemble averages need to be replaced by time averages given by
to calculate the CCFs of observed data for persistent
∞
external forces (e.g., ocean waves). However, no attenu-
ation in a closed system causes the problem of diverging ψ(r, t) = g 2D (r − ri ; t)fi , (15)
i=0
amplitudes without attenuation; that is, the seismic wave
field in a closed system never meets the equilibrium for where g 2D is the Green’s function in time domain.
persistent sources. Physically, it is natural to consider the The Fourier component of the potential ψ(r, t) is writ-
equilibrium between the energy dissipation due to atten- ten by
uation and the work performed by external forces (Kob-
∞
ayashi and Nishida 1998; Fukao et al. 2002).
�(r, ω) = G 2D (r − ri ; ω)fi , (16)
We note that the term Green’s function is often used
i=0
in a less mathematically rigorous sense in studies on SI.
Mathematically, a CCF converges to the Green’s function where G 2D is the Green’s function in frequency domain.
only in limited cases. A similar situation occurs in quan- In this paper, we define the Fourier transform as follows
tum field theory. The Green’s function refers to CCFs, (Dahlen and Tromp 1998),
although it does not satisfy the mathematical definition ∞
(Zagoskin 2014). X(ω) = x(t)e−iωt dt. (17)
∞
2.2 SI in an open system We note that the Fourier convention depends on the lit-
Next, we consider SI in an open system. The formulations erature. For example, Aki and Richards ’s definition has
in an open system depend on the source-receiver con- the opposite sign in the exponential term.
figuration: (1) random point sources distributed over the Here, we consider a potential ψ(ro , t0 ) in a simplified
whole space, (2) random sources distributed on a closed case with constant phase velocity as C(r) = C0, where ro
curve, and (3) uncorrelated plane wave incidents from is the location of the origin and time denotes an arbitrary
various directions. This subsection describes the rela- positive time. The ψ(ro , t0 ) is represented by the sum of
tionship between the different configurations of an open the arrivals excited by sources along the concentric cir-
system. Mathematically, the theory of an open system cle with radius r = C0 t0. The typical separation of the
differs from that of a closed system. We cannot use nor- sources is assumed to be x . Within the circle with the
mal mode theory for an open system because the system band x , about 2π r/�x sources are distributed (Fig. 2
loses energy from the radiation boundary. left). Because the amplitude decay is proportional to
For ANT, CCFs are usually formulated in an open sys- r 1/2, the mean square amplitude ψ(ro , t0 ) is estimated
tem, which better approximates the source-receiver con- to be about 2π r/(�x(r 1/2 )2 ) = 2π/�x , which does not
figuration. Because the surface wave can be formulated depend on the distance r. Therefore, after t > 0, the fluc-
as a 2-D problem with a membrane approximation (e.g., tuations of ψ last for a semi-infinite time with the same
Tanimoto 1990; Tromp and Dahlen 1992b, 1992a, 1993), mean squared amplitudes (Fig. 2 right).
we can consider them as a 2-D potential problem in an
open system. For ANT, we explicitly express the mixed- 2.2.2 Random sources distributed on a closed curve
component CCFs of surface waves in a simplified case at The second source configuration is random sources on an
the end of the subsection. arbitrary curve enclosing stations. Now, we observe the
potential ψ at ro within the circle with radius r (Fig. 2).
Based on the representation theorem, the wave excited by
2.2.1 Seismic excitation by an infinite number of sources distributed sources outside the circle can be completely
We consider one realization of the background seismic reproduced from the stresses and displacements on the
wave field excited by an infinite number of sources in an circle (e.g., Aki and Richards 1980). The contribution of
(a) (b)
ro t
∆x
t0
Fig. 2 Schematic figure for random sources in 2-dimension. a Source distribution in space. The red triangle shows the observed station. b Potential
ψ(ro , t) observed at the origin r at time t. The wave packet at time t0 is traveled from the source within the concentric circle with radius r shown
in (a)
the sources within the circle for the CCF can be neglected configuration to represent CCFs based on plane wave
because the contribution from the sources outside the incidents in the following subsections.
circle becomes infinitely large when considering infinitely
long times. These features mean that it is equivalent to 2.2.4 CCFs for homogeneous source distribution in a 2‑D
persistent sources distributed only on the circle. Suppose homogeneous medium
that uncorrelated random sources with white spectra on For ANT, we explicitly express the CCFs between all
an arbitrary surface enclose stations in a heterogene- pairs of three-component seismometers in an open sys-
ous medium. In this case, the time derivative of the cor- tem when multimode Love and Rayleigh waves dominate
responding CCF represents the exact Green’s function the ambient seismic wave field. We evaluated the CCFs
between a pair of stations (Wapenaar et al. 2010a). for the surface waves in a 2-D problem with a membrane
approximation (e.g., Tanimoto 1990; Tromp and Dahlen
1992a, b, 1993). We show an expression of the mixed-
2.2.3 Uncorrelated plane wave incidents component CCFs of surface waves.
The third source configuration is uncorrelated plane wave An arbitrary seismic wave field u(r, θ ; ω) in 2-D can be
incidents from various directions. Assuming that the represented by a superposition of multimode Love and
source-station distances are sufficiently longer than the Rayleigh waves as
aperture of the seismic array, the assumption leads to a ∞
∞
plane-wave approximation (Nakahara 2006; Haney et al. Ray

u(r, θ ; ω) = fn,m (ω)[Un (ω)Pm (r, θ ; ω)
2012). The above discussions for a homogeneous source n=0 m=−∞
distribution lead to the identical formulation of CCFs +Vn (ω)Bm (r, θ ; ω)]
regardless of the source configurations, as shown in the Love
following subsection. When we consider a heterogeneous + fn,m (ω)Wn (ω)Cm (r, θ ; ω),
source distribution, the first source configuration is the (18)
Ray
most flexible to represent the source, and the second and where fn,m (ω) is forcing for Rayleigh waves, and fn,m
Love
third are gradually less flexible. The flexibility of the first is forcing of Love waves. Un and Vn are eigenfunctions
two source configurations causes a complex dependence of the nth overtone of the Rayleigh wave, and Wn is the
of a CCF on the locations of the two stations. In contrast, eigenfunction of the nth overtone of the Love wave,
the inflexibility of the third source configuration causes a which has a real value. The basis functions Pm, Bm and
simple dependence of the CCF only on the relative loca- Wm are given by
tion between the stations (only the distance and the azi-
muth). In many cases, the ocean swell activities are far ωr
Pm (r, θ ; ω) = ẑJm Ray eimθ , (19)
enough away from the stations to approximate the phe- cn (ω)
nomenon well. For simplicity, we will use the third source
Ray Transverse

cn (ω) ∂ θ̂ ∂ ωr
Bm (r, θ ; ω) = r̂ + Jm Ray eimθ , Radial
ωr ∂r r ∂θ cn (ω)
r
(20) 2nd station
cLove (ω) r̂ ∂

∂ ωr
Cm (r, θ; ω) = n − θ̂ Jm Love eimθ , 1st station θ
ωr r ∂θ ∂r cn (ω)
(21) Radial
Ray
where cn (ω) is the phase velocity of nth overtone of Transverse
Rayleigh wave, and cnLove (ω) is nth overtone of Love wave. Fig. 3 Schematic of the geometry of a pair of stations
An assumption of equipartition of energy leads to with separation distance r and the coordinate. The figure shows
the radial and transverse directions for the station pair. To simplify
Ray∗ Ray Ray the equation, the signs of the directions are different from those
�fn,m fn′ ,m′ �Un (ω)Un′ (ω) ≡ Pn (ω)δnn′ δmm′ , (22) of other studies. The difference changes the signs of CCF
in the components of RR, TT, RZ, and ZR
Love∗ Love
�fn,m fn′ ,m′ �Wn′ (ω)Wn (ω) ≡ PnLove (ω)δnn′ δmm′ , (23)
Ray
Ray∗ Ray
difference of sign RZ originates from the definition
�fn,m fnLove Love∗
′ ,m′ � = �fn,m fn′ ,m′ � = 0, (24) of the radial direction, which differs from this study.
Using this formulation, cross-spectra for Rayleigh
Rn (ω) ≡ Vn (ω)/Un (ω), (25) waves lead to a symmetric relation between the RZ and
Ray Ray
Ray
ZR components as RZ = ZR , whereas cross-spectra
where Pn (ω) is
the power spectrum of nth overtone of for body waves lead to an antisymmetric relation as
Rayleigh wave, and PnLove is the power spectrum of nth body body
RZ = −ZR (Takagi et al. 2014).
overtone of Love wave. Attenuation becomes significant when consider-
The mixed-component cross-spectra for multimode ing a seismic wave field in sediment above 0.1 Hz
Rayleigh waves are written as, (Nishida et al. 2008a; Prieto et al. 2009, 2011). However,
  � � � � 
ωr |ω|r
 J0 cnRay −Rn J1 Ray 0

 

 
Ray Ray Ray
 
� cn �
 
� �ZR �ZT ∞ 
 
 ZZ
 � � 
2
 
Ray |ω|r R
�
Ray Ray Ray  ωr

n
 �RZ �RR �RT  =  −Rn J1 cnRay
Pn (ω) 2 J0−2 cRay 0  ,
 (26)
Ray Ray Ray  n � 
�TZ �TR �TT n=0 
 
2
� 

  Rn ωr

0 0 J
 
2 0+2

 Ray


cn
where a cross-spectrum represents the CCF in frequency physically plausible attenuation measurements are
domain. The mixed-component cross-spectra for multi- practically difficult (e.g., Liu and Ben-Zion 2013; Tsai
mode Love waves are written as,
  
0 0 0
�Love �Love �Love
 
∞ 
 � � 

ZZ ZR ZT  1 ωr 
0 2 J0+2 0 (27)
�
Love

�Love �Love �Love  = Pn (ω)  cnLove  ,
RZ RR RT � �
�Love �Love Love

�TT 1 ωr

n=0  0 0 2 J0−2
 
TZ TR Love
cn

where J0−2 (z) ≡ J0 (z) − J2 (z) and J0+2 (z) ≡ J0 (z) + J2 (z). 2011) because there is ambiguity between the source
Z represents the vertical component, and R and T repre- heterogeneities and the attenuation. Although new
sent the horizontal components according to the polari- techniques have been developed to overcome this prob-
zation direction (Fig. 3). lem (e.g., Liu et al. 2021; Magrini and Boschi 2021;
These equations are identical to the result of Haney Bowden et al. 2015), they are beyond the scope of
et al. (2012). A similar formulation for DAS observa- this study. Even if the physically plausible attenuation
tion is given by Nakahara et al. (2021). We note that the estimation is difficult, apparent Q measurements are
feasible for better fitting a synthetic cross-spectrum to

mπ 4m2 − 1

2 mπ
an observation (e.g., Nishida et al. 2008a). Jm (kr) ≈ cos χ − − sin χ − ,
πkr 2 8kr 2
(31)
2.3 Travel‑time anomalies due to the source
mπ 4m2 − 1

2 mπ
heterogeneities

Nm (kr) ≈ sin χ − + cos χ − ,
In this subsection, we evaluate travel-time anomalies due π kr 2 8kr 2
to source heterogeneities. Although the mode decompo- (32)
sition in the previous subsection can be applied to more where Nm is the Bessel function of the second kind (also
general source distributions, the resultant CCF for het- known as the Neumann function) and χ ≡ kr − π/4 . The
erogeneous source distribution becomes complex. With real part of the cross-spectrum can be evaluated by only
a stronger assumption of the 2-D ambient seismic wave even orders as,
field described by a superposition of uncorrelated plane ∞
waves from all azimuths (the third source-receiver con-

Re[�(r, θ ; ω)] = (−1)m J2m (kr)β2m (θ). (33)
figuration), the resultant CCF depends only on the rela- m=0
tive location between the station pair. The condition that
the excitation sources exist at infinite distances simpli- Insertion of the asymptotic Bessel function into the
fies the problem (see Sect. 2.2.3 for details). Reducing above equation leads to the following equations. The real
variables simplifies the effects on the source heterogenei- part of the cross-spectrum is given by
ties. This assumption is sufficiently realistic, because we ∞

consider the on-land observation of the ambient seismic Re[�(r, θ ; ω)] = (−1)m J2m (kr)β2m (θ)
wave field excited by distant ocean swells. This estimate is m=0
consistent with Weaver et al. (2009), but with a different

2 sin χ ′′
derivation. ≈ J0 (kr)Bev (θ) + B ,
πkr 2kr ev
We assumed that the sources were located along a (34)
closed curve surrounding the stations. If the source is and the imaginary part of the cross-spectrum is given by
distant enough compared with the wavelength, the cross-
spectrum between a pair of stations is given by ∞

Im[�(r, θ ; ω)] = i (−1)m iJ2m+1 (kr)β2m+1 (θ)
∞
m=1
�(r, θ ; ω) = im Jm (kr)βm (θ), (28)
m=0
2 cos(χ) ′′
≈ N0 (kr)iBod (θ) − i B ,
πkr 2kr od
where wavenumber k(ω) is ω/c(ω), m is the azimuthal (35)
order, θ is azimuth (Fig. 4), and the Fourier coefficients of where Bod and Bev are defined by
the source intensities βm (θ) are real value as
∞

a
0
m = 0, Bod ≡ β2m+1 (θ), (36)
am cos(mθ) + bm sin(mθ) m �= 0, (29)
βm (θ) = 2
m=0
where am and bm are Fourier coefficients of the source ∞

distribution (Harmon et al. 2010; Cox 1973). The spatially Bev ≡ β2m (θ). (37)
symmetric part of the cross-spectrum with even m is a m=0
real function, whereas the antisymmetric part with odd
The real and imaginary parts lead to the following cross-
m is an imaginary function. The symmetric feature origi-
spectrum as,
nated from the plane wave decomposition by the Bessel
function as (1)
�(r, θ ; ω) ≈ H0 (kr)B(θ)
∞ (2)
+ H0 (kr)B(θ + π )
eikr cos θ = im Jm (kr)e−imθ . (30)
i
m=−∞ − eiχ B′′ (θ) − e−iχ B′′ (θ + π ) ,
2π(kr)3
Here, we evaluate a travel-time anomaly due to source
(38)
heterogeneities of CCFs by the asymptotic expansion
of the Bessel function for a large argument given by where B ≡ Bev (θ) + Bod (θ). B(θ) represents the intensity
Eqs. 9.2.5 and 9.2.6 of Abramowitz et al. (1988), of incident plane waves as a function of azimuth θ . Here,
we use the following relations.
y the higher frequency, the travel-time anomaly decreases

for a longer distance and a higher frequency. The equa-
tion does not include the first derivative of the intensity B
B(θ) because the symmetry cancels the contribution (Weaver
et al. 2009). Equation 42 gives an error proportional to
δθ the second derivative of the intensity B to the azimuth.
θ This equation can estimate the phase velocity bias due
r x
to source heterogeneities and correction (Weaver et al.
2009; Froment et al. 2010).
Finally, the analytic representation of a cross-spectrum
for heterogeneous sources is given by
π
Fig. 4 Schematic figure of the geometry of stations �(r, θ ; ω) ≈ 4 e− 2 i G ∗ (r; ω)e−iωδt(r,θ ) B(θ)
and the stationary phase regions shown in blue. The sources are π
(43)
located sufficiently distant from the stations, and the distribution +e 2 i G(r; ω)eiωδt(r,θ +π ) B(θ + π ) ,
is given by B(θ ) as a function of the incident azimuth θ
where Green’s function in 2-D is given by
(2)
B(θ) + B(θ + π ) = 2Bev (θ), H0 (kr)
G(r; ω) = −i . (44)
B(θ) − B(θ + π ) = 2Bod (θ). 4
To evaluate the travel-time anomalies due to source This subsection shows how to estimate travel-time
heterogeneity, we consider the far-field approximation anomalies due to source heterogeneities. We note that
(kr ≫ 1) of the Hankel function of the first kind: the CCFs still satisfy the original wave equation even if
the source distribution is heterogeneous. This property
ensures that phase velocity measurements can be pos-

2 i χ − 8kr
1
(1)
H0 (kr) ≈ e . (39)
π kr sible even for a heterogeneous source distribution if the
station density is sufficiently high and the excitation
We also assume that the source heterogeneity is weak as sources are located outside the station array. Follow-
1 d2B ing Lin et al. (2013), we briefly show the relation in the
≪ 1. (40) following.
krB dθ 2
Here we consider potential ψ, which satisfies a wave
The causal part of the cross-spectra (Eq. 38) can be writ- equation L as
ten by 2
∂ 2
ieiχ B′′ (θ )
L[ψ(r, t)] ≡ − C(r)∇ ψ(r, t) = 0. (45)
(1)
H0 (kr)B(θ ) − ∂t 2
2π(kr)3
We also assume that no sources are distributed within
B′′ (θ )

2 1 the station array. The CCF φ is defined by
≈ B(θ ) exp i χ − − .
π kr 8kr 2krB(θ )
(41) 1 T

φ(r1 , r2 , t) = lim ψ(r1 , τ )ψ(r2 , t + τ )dτ
Therefore, the travel-time anomaly δt of the causal part T →∞ T 0
can be estimated by (46)
The CCF satisfies the wave equation as
B′′ (θ)
δt(r, θ) ≈ , (42) T
2ωkrB(θ) 1

L1 [φ(r1 , r2 , t)] = lim L1 [ψ(r1 , τ )]ψ(r2 , t + τ )dt = 0,
T →∞ T 0
where t is the travel time and ω is the nominal angular (47)
frequency. Figure 4 shows the stationary phase regions,
which dominate the contribution of the CCF (Snieder where L1 is the wave equation concerning r1. The CCF
2006). The aperture δθ of the stationary phase regions is also satisfies the wave equation for r2. Because the CCF
proportional to the square root of the ratio between the satisfies the wave equation only without spectral whiten-
wavelength and the station separation distance r (Fro- ing and one-bit normalization, these procedures often
ment et al. 2010). Because the narrower the aperture at create unphysical phases (Nakata et al. 2013).
3 Data processing 3.1.2 Instrumental corrections

If the seismic wave field is stochastic stationary, data In ambient noise cross-correlation analysis, small-ampli-
processing for the CCF calculation is straightforward. tude seismic wave signals are used compared to regular
In a realistic situation, the spectral structure of seismic earthquake signals. Moreover, the signals may fall within
records changes with time. Moreover, seismometers the lower frequency band than the natural frequency of
record transient phenomena such as earthquakes, which the seismometers, potentially residing below the instru-
are noise for the CCF calculations. Instrumental noise is ment noise level. In such cases, instrument noise can sig-
also problematic. To improve the quality of CCF, we need nificantly impact the quality of CCFs. Instrument noise
to select and regularize seismic records (Bensen et al. can be categorized into incoherent and coherent noises
2007; Ritzwoller and Feng 2019). This section explains across observation stations. Incoherent noise between
the practical data processing. stations decreases with the square root of the number of
stacks of CCFs. However, coherent noise does not dimin-
3.1 Instrument ish and remains in the CCFs even with many stacks.
Before processing data, we must consider the influence Takagi et al. (2015) conducted an ambient noise cross-
of instrumental characteristics. This review considers correlation analysis using Hi-net short-period seismom-
the geophysical observable of ground motions and pres- eters in Japan and found periodic pulses at 60-second
sure in the ocean. The seismometer records the ground and 1-second intervals within the CCFs. The stacking
motions, and they can be categorized into strong motion of raw records revealed that these periodic pulses were
sensors, short-period sensors, and broad-band sensors not a result of data analysis but were inherent in the raw
according to the frequency and amplitude ranges. records shared across all stations. Although the ampli-
tudes were extremely small, less than 1 bit, they appeared
3.1.1 Sensors in the CCFs at frequencies lower than the natural fre-
Today, on a global scale, broad-band seismometers are quency band because of their coherence across stations.
commonly used. Because instrumental responses depend Similar instrument noise and its impact on CCFs have
on sensor types, the responses should be removed prior also been reported in other observation systems (e.g.,
to data analysis. Typically, broad-band sensors have a flat Wang et al. 2017). This coherent periodic instrument
response in particle velocity above the corner frequency noise is attributed to load fluctuations in data loggers,
of about 10 to 100 s, and the response decreases with such as time calibration and data recording. There is also
decreasing frequency below the corner frequency. temporally random but coherent instrument noise across
Short-period sensors (typically velocity seismom- stations. The coherent random instrument noise appears
eters with a corner frequency of 1 Hz) are also com- as pulses near the zero lag in the CCFs. Such coherent
monly used on a local scale. In the recent development random noise was found in DAS observations (e.g., Trib-
of dense observation with more than 1000 stations (also aldos and Ajo-Franklin 2021) and cabled ocean-bottom
known as a large-N array, e.g., Lin 2013), geophones seafloor seismometers (Takagi et al. 2021).
with a natural frequency of 10 Hz are also often used. It is crucial to remove coherent instrument noise to
Although only nominal responses are available in many utilize CCFs to image subsurface structure. Several meth-
cases, the responses are slightly different from one to ods have been proposed to deal with periodic coherent
another (Ueno et al. 2015; Takeo et al. 2022). When we instrument noise. One method involves creating wave-
use the data with a frequency close to the natural fre- forms of periodic coherent instrument noise by stacking
quency, the difference might cause an apparent phase many raw records and subtracting them from the raw
shift, which could be problematic for a short separation observation records (Takagi et al. 2015). Alternatively,
distance. instead of computing instrument noise waveforms, one
When a sensor is deployed on the seafloor, pressure can calculate the difference of raw observation records
gauges are also commonly used (Webb 1998). Cross-cor- (e.g., the difference between two adjacent days) and then
relation analysis of ocean bottom pressure gauges exhibits compute the CCFs of these differential waveforms (Tak-
Rayleigh wave propagations by cross-correlation analysis agi et al. 2021). To reduce randomly coherent instrument
(e.g., Takeo et al. 2014). When a sensor is deployed on the noise, practical methods such as modeling and estimat-
seafloor or bottom of a borehole, the orientation of the ing the spectral shape of instrument noise (Takagi et al.
sensor is unknown (e.g., Takagi et al. 2019). In such cases, 2021) or subtracting the median of CCFs from each trace
the orientations should be determined by known events, (Tribaldos and Ajo-Franklin 2021) have been proposed.
such as an active shot or teleseismic events. By taking advantage of dense DAS observations, the F-K
filtering of fast-propagating phases before computing the normalizations among components cause problems in
CCFs is effective (Fukushima et al. 2022). In summary, the cross-correlation analysis of multicomponent data.
while coherent periodic or coherent random instrument
noise across observation stations can contaminate CCFs, 3.2.2 RMS‑based data selection
understanding the characteristics of the noise and apply- If the signal obeys a stationary process, we can reject out-
ing appropriate removal methods enable us to expand the liers using an RMS threshold. In a strict sense, secondary
analyzable frequency range and ultimately improve the microseisms do not follow a stationary process but are in
resolution of structural imaging. a local stationary state: RMS changes significantly by sev-
The time reliability is crucial for the cross-correlation eral orders of magnitude on a timescale of several days
analysis between stations. For onshore observations, (Fig. 5). We must define the local background levels for
satellite systems, such as the Global Navigation Satellite the rejection, which change slowly over several days.
System (GNSS), usually give a precise time stamp. When Here, we introduce an example for estimating the local
GNSS reception fails, the temporal change in the CCFs background level by Nishida and Takagi (2022). To define
has been used to estimate the error in the time stamps the typical background level of the whole network at the
(Sens-Schönfelder 2008; Hirose and Ueda 2023). For off- ith time step, we calculate the median of MS amplitudes
shore observations, precise time stamps are mostly given for all stations Pi . Here, we consider a situation: ith time
at deployment and recovery, and the drift of the internal step is the latest accepted time step, and we reject n suc-
clock can be linearly corrected. When the time stamp cessive time steps. If Pi+n changes suddenly, we reject the
cannot be obtained either at the beginning or at the end (i + n + 1)th time step with the threshold ǫ:
of the observation, the temporal change in the CCFs can
be used again to estimate the drift (Hannemann et al. | ln(Pi+1 ) − ln(Pi )| > nǫ, (48)
2014; Gouédard et al. 2014; Takeo et al. 2014). The time we reject all data at time step i + n + 1 with the criterion
asymmetric shape of the CCFs can also be used to esti- proportional to the rejection duration (orange points in
mate unknown instrumental responses such as the con- Fig. 5).
stant time shift of the logger or the frequency-dependent
phase response of the differential pressure gauges (Takeo 3.2.3 Data rejection associated with large earthquake using
et al. 2014). a catalog
Seismometers record global propagations of many seis-
3.2 Data selection mic phases excited by large earthquakes, typically with
We must choose seismic data that satisfy a local station- moment magnitudes greater than 6 (Ekström et al. 2012).
ary state to apply the SI. Outliers such as earthquake In most frequency ranges, the RMS criterion can reject
records and instrumental glitches decrease the SNR. the corresponding data. However, the secondary micro-
This subsection introduces (1) one-bit normalization, (2) seisms are still large enough to hide some earthquakes.
RMS-based data selection, and (3) polarization-based Even smaller amplitudes can bias the CCFs because the
data selection. earthquake signals are coherent among the stations.
Careful rejections of large earthquakes improve the qual-
3.2.1 One‑bit normalization ity. Using a global earthquake catalog (e.g., the global
To suppress the effects of transient phenomena such Centroid Moment Tensor (CMT) catalog (Ekström et al.
as earthquakes, one-bit normalization (e.g., Aki 1955; 2012)), data rejection based on the magnitude in the cat-
Bensen et al. 2007; Cupillard et al. 2011) keeps only the alog is feasible to exclude hidden seismic phases.
sign of the original information, changing all positive Approximately the amplitude of earthquake data
values to 1 and all negative values to -1. Because this decays exponentially with time. The typical duration De
method is simple, it has been widely used. When tran- is:
sient phenomena (e.g., many aftershocks and packet loss
during data logging) occur frequently, their contributions ln(JM/1023 )
De = 2Q , (49)
decrease the quality of the CCFs. In such a case, care- (2π f )
ful reduction of transient signals allows us to improve
where f is the dominant frequency, Q is the typical qual-
the CCF’s quality. A disadvantage of this method is the
ity factor, M is the moment, and J is a typical geometrical
loss of amplitude information. Different amplitude
Fig. 5 Example of the data selection process. The orange dots show data excluded using the global CMT catalog (Ekström et al. 2012), and the blue
dots show data transients when the amplitude changes suddenly. The black dots represent selected segments. The vertical axis shows the relative
power normalized by the Peterson NLNM (Peterson 1993)), which represents the lowest ground noise on Earth. Taken from Nishida and Takagi
(2022)
spreading (Nishida and Kobayashi 1999; Tanimoto and nor signal behaves well in a realistic situation. Regulari-
Um 1999). Data were rejected for De seconds after the zation of data is required as part of the preprocessing
arrival. The orange dots in Fig. 5 show a typical example (Bensen et al. 2007).
of rejected data (Nishida and Takagi 2022). Data can be
rejected until the earthquake signal is much smaller than
the background noise levels. 3.3.1 Weighting of cross‑spectra
The weighting on the data is important when we
calculate a cross-spectrum between a station pair
3.2.4 Polarization‑based data selection (Nishida 2014). The amplitudes of microseisms at
If the excitation processes are stochastic stationary in frequencies around 0.1 Hz change with time, which
time and space, the energy partition among the mode reaches more than one order of magnitude on a time-
branches should be constant. The energy partition scale of one day. Spectral whitening efficiently reduces
changes over time because they are not stationary in the non-stationarity (e.g., Bensen et al. 2007). The
a realistic situation. For example, Takagi et al. (2018) amplitudes of the seismic hum in the mHz band are
pointed out that the energy partition of P-wave micro- stationary, although the local noise level is higher than
seisms at periods of 4–8 s becomes more significant on the signal levels. In this case, the weighting of the data
seismically quiet days based on the polarization analy- depending on the local noise level can be effective
sis of Hi-net data in Japan. When we want to empha- (Nishida 2014; Takeo et al. 2013). In the mHz band,
size overtones of seismic surface waves, data selection the noise levels of the horizontal components are
based on polarization information can potentially orders of magnitude higher than those of the vertical
improve the detection of overtone branches. Pedersen components. For the calculation of the cross-spec-
et al. (2023) proposed the data selection based on the trum, we suppressed noisy Fourier components using
H/V spectral ratio to extract teleseismic body wave the data weighting as follows.
microseisms, and this strategy may be applicable for We calculated a weighted cross-spectrum �ij (f )
detecting overtone branches. between the ith and jth stations at a frequency f as
1
3.3 Regularization: weight and normalization of CCFs

�ij (f ) = wijk (f )Uik∗ (f )Ujk (f ),
If both the signal and the noise are subjected to the wijk (f ) k
(50)
Gaussian distribution, calculating the CCFs from k
selected data is straightforward. However, neither noise
where Uik (f ) is a Fourier spectrum of ground acceleration where Nij (f ) is number of stacked traces (Takeo et al.
of the kth segment at the ith station. The weighting fac- 2013). The corresponding CCF φij (t) is calculated by the
tor wijk (f ) depends on the situation. If the noise is much inverse Fourier transform of �ij (f ) (e.g., Fig. 6).
larger than the signal, the weight can be estimated by
1 1 3.3.2 Spectral whitening
wijk (f ) = . (51) If the signal is larger than the noise, the assumption of
|Uik (f )| |Ujk (f )| the previous Sect. 3.3.1 is broken down. Although sim-
ple stacking works for signals with Gaussian distribu-
We can evaluate the uncertainty σij (f ) of the resultant
tion, the signal level changes significantly with time. In
�ij (f ) as
particular, the amplitudes of secondary microseisms

Nij
−1 can change by several orders of magnitude over a time-
1 1 � scale of a few days. Without regularizing the amplitude,
σij (f ) = �  wijk (f ) , (52)
Nij (f ) Nij (f ) k=1
the resultant CCFs emphasize days of high-ocean-swell
activities too much. For the amplitude regularization, the
Fig.
6 Cross-correlation functions as a function of interstation distances filtered at 20–30 s, 10–20 s, and 5–10 s. The amplitudes are multiplied
by wijk (f ) in the frequency domain to produce similar noise levels for different pairs of stations. The dashed pink line shows the group velocity
k
of each specified mode at a typical period for the 1-D isotropic model. The modes include the fundamental mode and first-higher-mode Rayleigh
waves, 0 S, and 1 S. The TT components show wave packets of Love waves xT, which could not be identified as a single mode due to possible
interference between the fundamental mode and the higher mode with very similar group and phase velocities. Taken from Takeo et al. (2016)
cross-spectrum ij is normalized by the amplitude (e.g.,

1
φij+ (t) = 2 φij (t) + φij (−t) t ≥ 0 , (54)
Aki 1957; Nishida et al. 2008a; Prieto et al. 2009) as, 0 t<0
Uik∗ (f )Ujk (f ) and the corresponding Fourier component is defined
�ij (f ) = wijk (f )Uik∗ (f )Ujk (f ) = .
|Uik (f )||Ujk (f )| by �+ij (f ). The folding procedure distorts the waveform
k k
(53) when the separation distance is shorter than about one
wavelength. The tail of the acausal wave packet contami-
The ij with spectral whitening is also known as cross-
nates the causal peak because the CCF does not satisfy
coherency.The equation is similar to the weighting, but
causality, as in the case of Green’s function.
the factor k wijk is not required.
A group velocity-period diagram can be calculated for
Although direct waves are accurately reconstructed each CCF. The general outline of the procedures is as fol-
with spectral whitening, they cause many pseudo-arriv- lows, although there are some other implementations.
als. Therefore, if we ignore the pseudo-arrivals, all nor- Figure 7 shows an example of a synthetic test.
malizations work practically (Nakata 2020). To mitigate At a given center frequency f, a narrow-band filter with
the pseudo-arrivals, smoothing of wijk in the frequency the center frequency f0 is applied to the CCF. For exam-
domain was proposed (Tauzin et al. 2019). ple FTAN (Levshin et al. 2018; Ritzwoller and Feng 2019)
Another implementation is performed by spectral use a Gaussian filter
whitening of the individual seismic trace before calcu- 2
lating CCFs (Bensen et al. 2007). Specifically, the inverse

f −f0
Fb (f , f0 ) = e
−α f0
, (55)
Fourier transform of Uik (f )/|Uik∗ (f )| corresponds to the
whitened seismic trace. where α is the coefficient, determining the bandwidth
(Fig. 7a, b). Because a smaller α broadens the bandwidth,
3.3.3 Temporal flattening a smaller α provides a stable estimate for noisy data. At
Different amplitude normalizations at different stations the same time, smaller α can bias the estimation in cases
become problematic if we utilize amplitude information of strong dispersion. α often changes with the separation
(e.g., attenuation). Although this topic is beyond the scope. distance (Ritzwoller et al. 2011) as
Temporal flattening (Zhou et al. 2020; Weaver 2011) was
proposed to preserve the amplitude information. α = α0 r/r0 , (56)
where α0 is the reference value at the separation distance

4 Dispersion measurement of surface waves r0. In a synthetic test shown in Fig. 7, α0 is 40, and r0 is
for constructing local 1‑D structures 200 km.
In the first step of ANT (Fig. 1a), a reference dispersion The envelope functions of the filtered CCFs are plot-
curve of phase or group velocities is measured using the ted against the periods and corresponding group veloci-
CCFs of an array. This section briefly summarizes the ties calculated from the separation distance and the lag
measurement methods: (1) frequency–time analysis, (2) time (Fig. 7c). Correction for f0 using the instantaneous
slant stack technique, (3) SPAC method, (4) FJ method, frequency based on the analytical signal is feasible for a
and (5) waveform fitting. The dispersion curves are also better estimate (Shapiro and Singh 1999; Levshin et al.
important for the construction of local 1-D structures, 2018). In this case, a synthetic CCF is given by
which can be an initial model of ANT (see Sect. 6.3.1).
�syn (ω) = J0 (k(ω)r) + n(ω), (57)
4.1 Frequency–time analysis (FTAN) where n is Gaussian noise with a standard deviation of
Frequency–time analysis is a common method for meas- 0.03. We can measure group velocities by tracing the loci
uring the dispersion curves of earthquake data (Levshin of a local maximum as a function of the period. Because
et al. 1992; Cotte and Laske 2002; Romanowicz 2020). a strong frequency dependence on the spectral content
This method is also feasible for ambient noise CCFs can bias the group velocity measurements, the correc-
(Bensen et al. 2007; Harmon et al. 2007; Yao et al. 2011; tion is necessary for accurate measurements. AFTAN
Spica et al. 2018). For the measurements, we usually use (Automatic Frequency–Time Analysis) package (Barmine
the time-symmetric part of CCF to reduce the effects 2018) is available for this type of measurement. We show
of source heterogeneities (see equations 34 and 35 in an example of the estimation errors at 0.1 Hz as a func-
Sect. 2.3). In this subsection, we use the folded CCF to tion of the separation distance (Fig. 7d).
lag time 0 φij+ as,
(a) Gaussian filter (b) Gaussian filter

1.0 1.0
10km
20km
50km
100km
200km
0.5 500km 0.5
0.0 0.0
0.00 0.05 0.10 0.15 0.20 0.0 0.1 0.2 0.3 0.4 0.5
Frequency [Hz] Frequency [Hz]
(c) FTAN diagram (d) Anomaly of phase velocity and group velocity
3
3.4 ∆= 3 λ Group velocity
Phase velocity
Group velocity [km/s]
3.2 Anomaly [%] 2
3.0
1
2.8
2.6 0
2.4
−1
2.2
2.0 −2
2 4 6 8 10 12 14 16 18 20 0 200 400 600 800 1000
Period [s] Distance [km]
∆= 3 λ (e) Phase (f) Errors of Group/phase velocity
180
5 Phase velocity
Phase [degree]
Number density
90 4
3
0
2
Group velocity
− 90
1
− 180 0
0 200 400 600 800 1000 − 1.0 − 0.5 0.0 0.5 1.0 1.5 2.0
Distance [km] Anomaly [%]

Fig. 7 a Gaussian filters for various separation distances at 0.1 Hz. b Gaussian filter for r = 200 km. c FTAN diagram for a synthetic CCF
with the observation error is 1%. The separation distance is r = 200 km. The black line shows the theoretical group velocity and the red dots show
measured values using synthetic data. d Group and phase velocity anomalies for the synthetic CCFs with 1% measurement errors as a function
of separation distance at 0.1 Hz. e Measured phases (blue dots) and the theoretical values (dashed lines) at 0.1 Hz. f Histogram of group and phase
velocities estimation errors at 0.1 Hz
We can also measure the phase velocity from the we introduce two typical filters: (1) time variable filter
phase of �+ ij (f ) (e.g., Levshin 2018; Ritzwoller and Feng (Landisman et al. 1969) and (2) floating filter (Levshin
2019) if a single mode is dominant. Multimode inter- et al. 1992).
ference is problematic for such measurements. To iso- The time variable filter (Harmon et al. 2007; Landis-
late a single mode, we must apply a proper filter. Here, man et al. 1969) isolates a single-mode wave packet by
windowing the CCF around the arrival time predicted

by the group velocity measurements. Window length is
set according to the dispersion because the stronger dis-
persion makes the duration of the wave packet longer. A
typical window length (Harmon et al. 2007) Tw is given by
dU
Tw = 160 × , (58)
dτ
where U is group velocity, and τ is period.
Floating filtering is also a common procedure for
extracting a single mode (Levshin et al. 1992). Based on
the group velocity measurement, phase correction ψ(ω)
is defined by
ω1
ψ(ω) = r U −1 dω + t1 ω, (59) Fig. 8 Group and phase velocities and the average spectrogram
ω0 of raw data. Gray-scale filled contours represent the averaged
spectrograms of all the CCF that met the SNR requirements,
where r is the separation distance, and t1 is a constant. where the maximum value for each period in the individual
�+ij (ω) is multiplied by e to compress the signal around
iψ
spectrograms was normalized to one. The fundamental and second
the average group time. t1 is chosen to focus on the aver- mode Rayleigh wave is visible in these data as the broad regions.
Filled black circles with error bars indicate the average group
age group arrival. After cleaning up the signal, φ + is velocities for the fundamental mode, while open circles with error
recovered by the inverse Fourier transform with e−iψ. bars indicate the average group velocities for the higher-mode
The phase velocity C can be measured from the phase Rayleigh wave. Black triangles with error bars indicate fundamental
of �+ (ω) (Lin et al. 2008; Harmon et al. 2007) as, mode uncorrected phase velocities; gray triangles with error bars
indicate −π/4 corrected phase velocities for higher and fundamental
ωr mode phase velocities, and open triangles with error bars indicate
C(ω) = π , (60)
− arg(�+ (ω)) + 2π N + higher mode phase velocities. Taken from Harmon et al. (2007)
4
where N is an integer number. The π/4 originates from

the Bessel function, and the 2π ambiguity can be solved if pointed out, the FTAN method also has the disadvantage
a good starting model is available. Note that �+ (ω) is out that the dispersion measurements are difficult for sepa-
of phase by π/2 with the corresponding empirical Green’s ration distances shorter than about three wavelengths.
function (see Sect. 2 of Lin et al. (2008)). FTAN method is still feasible for measuring multimode
Figure 7e shows the phase of the synthetic test and the dispersion if a wave packet of each mode branch is iso-
theoretical values. Although they agree, the finite band- lated in the time domain. If multiple phases overlap in
width of the Gaussian filter biases the measured phases. time domain, an array-based method (slant stack or
At higher frequencies, estimating the phase ambiguity N SPAC) is appropriate.
is difficult due to the ambiguity of an initial model. If a Figure 8 shows an example of a multimode Rayleigh
dense array is available, the array methods, as discussed wave dispersion diagram measured using ocean-bottom
in the following subsections, are feasible for measur- seismometers. The figure shows two clear branches. The
ing the phase velocity at higher frequencies. The meas- fundamental mode with a period 8 s has energy in the
urement using FTAN has a limitation for the separation ocean, while it has energy in the crust and the upper-
distance of CCFs. The group velocity measurements are most mantle with the longer period. The overtone also
accurate when the separation distance exceeds three has energy in the crust and the uppermost mantle. This
wavelengths. For this reason, (Bensen et al. 2007) recom- behavior of a multimode Rayleigh wave in an oceanic
mended that the longest period should be r/12. region is common, as shown in Fig. 6. A seismic array
Histograms in Fig. 7f show the estimation errors of on a soft sediment layer records multiple mode branches
the group and phase velocities. Although the measured above about 0.2 Hz (e.g., Spica et al. 2018).
phase velocities were relatively accurate, the dispersion
caused bias. When applying this method to strongly dis- 4.2 Slant stack technique
persive wave fields, we must carefully consider bias, par- Suppose that a dense array with station spacing compa-
ticularly for low-SNR data. rable to the wavelength is available. In this case, the slant
An advantage of the FTAN method is the ability to stack technique, also known as the frequency–wavenum-
measure the dispersion of a single station pair. As already ber (F-K) method, is a common array processing (Rost
and Thomas 2002; Gouédard et al. 2008). This method re-sample the CCFs on regular grids if all the station sepa-
also assumes that the superposition of plane waves can ration distances are shorter than half the wavelength. CCFs
represent the wave field. The beam power Bp is defined can be mapped from the spatial domain to the F-K domain
by a sum of time-shifted waveforms �+ (ω) with expected using the 2-D fast Fourier transform (FFT) after a correc-
travel-time anomaly ωp · xi tion r 1/2 for geometrical spreading (Gabriels et al. 1987).
2 This method is applied to joint inversion of the first over-
√
+

iωp·(xk −xl ) tone and fundamental mode for deep imaging in the Val-
Bp (p, ω) = rkl �kl (ω)e , (61)

hall oil field using ambient seismic noise (Tomar et al. 2017
kl
2018).
where k and l represent the station number, rkl is the
separation distance between kth station and lth station, 4.3 SPAC method
and xk is the location vector of kth station represented by This subsection explains the implementation of the SPAC
a Cartesian coordinate with the origin at the array cen- method. In the case of homogeneous source distribution
√ syn
in a stratified medium, the synthetic cross-spectrum ρZZ
troid. The term rkl corrects the amplitude decay due to
geometrical spreading for a surface wave. between vertical components can be represented by a Bes-
The disadvantage of this method is a potential bias due sel function (see Sect. 2.2.4) as
to the plane wave assumption. Equation 31 shows that syn
ωr
the J0 (vertical component) the J0−2 /2 (radial and trans- ρZZ (r, a, C; ω) = a(ω)J0 , (64)
C
verse component) can be approximated as
where r is the separation distance of the station pair.
√ π 1 π
rJ0 (kr) ≈ cos(kr − ) + sin(kr − ), (62) Assuming an arbitrary wavenumber k(ω) at a given fre-
4 8kr 4 quency, we determine the optimum amplitude a, by mini-
√ mizing the squared difference S between the observed
r π 7 π cross-spectrum i of ith pair and ρ syn (ri , a, C; ω) (e.g.,
J0−2 (kr) ≈ cos(kr − ) − sin(kr − ). (63)
2 4 8kr 4 Nishida et al. 2008b) with weight wi . S is given by
The contribution of the terms (8kr)−1 and 7(8kr)−1 is sig- N
syn 2
nificant for a shorter station pair. These equations sug- S(a, C; ω) = wi ρZZ (ri , a, C; ω) − �i (ω) , (65)
gest that the bias of the horizontal components could i=1
be larger. We tested the bias of the vertical component where N is number of the pairs. By minimizing S at a
using a synthetic CCF at 0.15 Hz given by Eq. 57 and the given frequency ω , we infer the phase velocity C(ω) and
slowness of 0.3 s/km. Figure 9 shows the station distribu- the amplitude a(ω). Regarding a, S can be minimized
tion of the array (Hi-net stations with a radius of about analytically as
100 km) and the result. The beam peaks at 0.3 s/km, con-
sistent with the theoretical value. To verify the bias, we ∂S
= 0. (66)
measured the peak slowness for 10,000 experiments. Fig- ∂a
ure 10 shows the histogram. Although the measurements
We obtain the optimum a for given C and ω as
fluctuate, the central value is 0.05% lower than the theo-
retical value. The bias of about 0.05% is consistent with N ωr
i

−(8k 2 r 2 )−1 estimated from equation 62 with a typical wi �i (ω)J0
C
separation distance of 60 km. aopt (C, ω) = i=1
. (67)
This method can reveal multimode dispersion on a N
ωr 2
i
scale of 100 m to 100 km: DAS observation on a 100 m w i J0
C
scale (Dou et al. 2017), a 10 km scale ocean bottom DAS i=1
(Viens et al. 2022; Williams et al. 2021), and a 100 km To minimize S with respect to C in ω , we maximize the
scale basin structure (e.g., Boué et al. 2016; Jiang and variance reduction VR given by
Denolle 2022). Li et al. (2020) demonstrated that super-
vised machine learning methods are feasible for separat- N
ωr
i
2
ing multimode information. wi aopt (ω)J0 − �i (ω)
C
There is another implementation to calculate the F-K VR (C, ω) = 1 − i=1
.
N
spectrum. When calculating the slant stack, the separa-
2
tion distances between stations are not regular. We can wi �i (ω)
i=1
(68)
(a) Station distribution (b) Histogram of separation distance

38°N 500
400
37°N
300
Counts
36°N
200
35°N
100
34°N 0
135°E 136°E 137°E 138°E 139°E 140°E 0 50 100 150 200
Distance [km]
(c) Plot in slowness domain
1.0
SPAC
Slant stack
FJ
0.5
Amplitude
0.0
0.1 0.2 0.3 0.4 0.5

Slowness [s/km]
Fig. 9 Synthetic test of slant stack, SPAC, and FJ methods. We assumed an ambient seismic wave field with a slowness of 0.3 s/km. a Station
distribution. b Histogram of separation distance between station pairs. c Plot in slowness domain: slant stack, variance reduction of the SPAC
method, and the FJ spectrum as a function of slowness. The noise is assumed to be 3% of the signal
SPAC: VR The maximum is inferred using the grid search method.

The estimate can be refined using a generalized least-
squares method (Menke and Jin 2015).
20 Figure 9 shows a test of the SPAC method using a
synthetic cross-spectrum at 0.15 Hz given by Eq. 57
Number density
and the slowness of 0.3 s/km. We used an array with a

Beamforming
FJ: Hu et al. (2023)
radius of about 100 km shown in Fig. 9a. Most of the
10 separation distance ranges from 40 to 130 km (Fig. 9b).
FJ: Nimiya et al. (2023)
Figure 9c shows VR as a function of slowness (1/C).
Because the amplitude of the surface wave decreases
√
with 1/ r , the appropriate modeling by SPAC sharpens
0
the peak compared to the beam power.
-0.1 0 0.1 0.2 0.4 We measured the maximum of VR for 10,000 experi-
Slowness anomaly [%] ments to evaluate the estimation errors. Figure 10
Fig. 10 Bootstrap errors of the estimated slowness by SPAC, shows the histogram. The measurements fluctuate
slant stacking, and FJ method. The noise is 3% of the signal,
and the sampling number for the bootstrap method is 10,000
within 0.05%, and the median value is around the theo-
retical value.
Modeling RR and TT in horizontal components is On a global scale, cross-spectra can be modeled using
more complicated because both Love and Rayleigh waves the Legendre function instead of the Bessel function
contribute to these components, as shown by Eqs. 26 and (Nishida et al. 2002; Nishida 2014).
syn
27. The corresponding synthetic cross-spectra ρRR and
syn
ρTT can be represented by
 � ωr � � � 4.4 FJ method
ωr
J J The frequency–Bessel (FJ) method (Wang et al. 2019a;
� syn � � R �
ρRR 0−2 R� 0+2 cL � a (ω)
= c . (69)
syn ωr aL (ω) Hu et al. 2020; Nimiya et al. 2023) is closely related to
 � �
ρTT J0+2 R J0−2 ωr
c c L
Aki’s SPAC method. Here, we reinterpret the FJ method
with the SPAC method. For simplicity, we consider the
When we analyze the station with a separation distance
ZZ component in this subsection. Refer to Hu et al.
shorter than the wavelength, we must consider the Love/
(2020) for other components.
Rayleigh wave for the RR/TT component, respectively.
If we observed φ(r, ω) as a function of the separation
Although the Love and Rayleigh waves are coupled in
distance r, the squared difference with weight r can be
this system, we can solve this equation directly. The
evaluated by the following integration,
maximum search becomes complex because the variance
reduction at a given frequency is a function of C R and C L.
rN ωr 2
Here, we consider a simplified problem with a far- S= a(ω)J0 − �(r, ω) rdr. (70)
0 C
field approximation. Because J0+2 is proportional to
r −3/2, the term becomes negligible for r longer than the By minimizing S for given C and ω , we obtain
wavelength (e.g., Takeo et al. 2013). In this case, we can rN
ωr
simply apply the SPAC method to these components J0 �(r, ω)rdr
separately by replacing J0 of the vertical component with C
aopt (C, ω) = 0 rN 2 (71)
ωr
J0−2. Figure 11 shows VR using the SPAC method in Japan J0 rdr
(Nishida et al. 2008a). The diagram shows clear Ray- 0 C
leigh and Love branches. Although this method assumed If rN is much longer than the wavelength, the denomina-
a single-mode branch, the figure exhibits multimode tor can be approximated by
branches, which enable us to increase depth resolution
(e.g., Ikeda et al. 2012). The SPAC method is also applied
rN
ωr 2 rN C
to seismic data from the ocean floor (Takeo et al. 2013,
J0 rdr ≈ . (72)
0 C πω
2014; Lin et al. 2016; Takeo et al. 2016, 2018; Kawano
et al. 2023). To separate multimode Rayleigh waves, aopt (C, ω) can be approximated by the integral (the
polarization information could also be useful (Nayak and numerator) evaluated at discrete points (Hu et al. 2020)
Thurber 2020). as
Fig. 11 Frequency–phase velocity spectra using all pairs of stations. a A spectrum of radial components shows a clear Rayleigh wave branch,
and b that of transverse components, which shows a fundamental Love wave branch, first overtone branch, and second overtone branch. Taken
from Nishida et al. (2008a)
πω than the theoretical value. The bias is caused by the fall-

ωr
i
aopt (C, ω) ≈ �i (ω)J0 r̄i �ri
rN C
i
C off of the spectra at low phase velocity of C −2 (Shap-
πω ωr r 2 + 2ri (ri+1 − ri−1 ) − r 2
i i+1 i−1
iro and Singh 1999). The drop in the C −2 term (Nimiya
= �(ri )J0 , et al. 2023) of equation 78 significantly reduces bias, as
rN C C 8
i
(73) shown in Fig. 10. The greater variation in the estimates
is due to the greater weight of the smaller amplitudes at
where
longer separation distances (Eq. 77). When the observa-
1 rj−1 + rj

rj + rj+1
tion error is sufficiently small, the high resolution of the
r̄i ≡ + , (74) FJ method in the slowness domain will be beneficial for
2 2 2
separating multimode surface waves. When the observa-
tion error is significant, the SPAC method can consider
1
�r ≡ (rj+1 − rj−1 ), (75) the noise using appropriate weights.
2 Figure 12 shows an example of field data (Hu et al.
and i is the cross-spectrum of the ith station pair. FJ 2020). Although this subsection mentioned only the ZZ
spectrum I(C, ω) defined by Nimiya et al. (2023) can be components, it can be applied to other components.
approximated by Because the RR and TT components record both Love
and Rayleigh waves, as shown by Eq. 69, their coupling
ω2 rN rN ω opt complicates the estimation of the FJ spectra in the RR
ωr
I(ω, C) ≡ �(r, ω)J0 rdr ≈ a (C, ω).
C 0 C π and TT components. Hu et al. (2020) proposed pro-
(76) cedures to separate the Love and Rayleigh wave using
Although Hu et al. (2020)’s definition of the coefficient orthogonal relations of the Bessel functions (Fig. 12).
ω2 /C 3 instead of ω2 /C has a physical meaning, the
emphasis on small phase velocity C may bias the estima- 4.5 Multimode dispersion measurements by waveform
tion, as shown below. We can interpret the FJ spectrum fitting in a model space
as a variant of SPAC method with weight wi defined by The frequency–phase velocity diagram does not show
2 + 2r (r 2
clear mode branches in regions with strong lateral het-
π ω ri+1 i i+1 − ri−1 ) − ri−1 erogeneities, such as volcanoes and ocean sediments
wi = . (77)
rN C 8 (Takeo et al. 2022; Takagi and Nishida 2022). For example,
Fig. 13 shows an example of SPAC diagrams in an oceanic
We again tested the FJ method using a synthetic cross-
area offshore the central part of northern Japan near the
spectrum given by Eq. 57 and the slowness of 0.3 s/km.
Japan Trench. They show fundamental Rayleigh and Love
Figure 9c shows the FJ spectrum. The peak width is nar-
waves below 0.08 Hz, but other mode branches are inter-
rower, but the side lobes are more significant than the
mittent and ambiguous. Lateral variations of the shallow
others. Let us evaluate the side lobes using Eq. 71. If rN is
soft sediment distort the dispersion curves of multiple
sufficiently longer than the wavelength, the FJ spectrum
modes. Moreover, multimode interference also complexes
ω for synthetic data J (ωri /C0 ) can be approximated (see
the diagrams. In this case, the physically plausible phase
Eq. 31) by
velocity measurements are difficult. A physically plausible
constraint on the dispersion is crucial for better measure-

2ω ωr
i π ωri π
I(C, ω) ≈ cos − cos − �ri . ments. One strategy is to skip phase velocity measure-
π C 4 C0 4
i
ments using a SPAC method. For a given S-wave velocity
(78)
structure, the corresponding phase velocity of jth over-
This integral can be interpreted as the Fourier series. tone C j (β; ω) can be evaluated, where β is a vector with
Fourier series without a taper function cause significant components of S-wave velocity at each depth. The vari-
spectral leakage (Oppenheim and Schafer 2014). If rN is ance reduction given by Eq. 68 is summed over in a fre-
sufficiently large, I(C, ω) has value only at C = C0 like quency range from ω0 to ω1 and the N mode branches as
the delta function, due to the orthogonality of the cosine
function (more precisely, the orthogonality of the Bessel N
−1 ω1
function). This means that the FJ method has, in princi- VRall (β) = VR (C j (β; ω); ω)dω, (79)
ω0
ple, the highest resolution in phase velocity. j=0
To evaluate the accuracy, we measured the peak slow- where VRall is the summed variance reduction. By maxi-
ness for 10,000 experiments using the FJ method. Fig-
mizing VRall for β, we can infer the local 1-D struc-
ure 10 shows the histogram. The estimated value of
ture. This method can be applied to the RR and TT
the FJ spectrum of Hu et al. (2020) is about 0.2% larger
Fig. 12 Multicomponent frequency–Bessel spectrograms of field data in North America. a–d are the FJ spectrograms IR0, IR1, IR2, and IL0, respectively.
Here, IR0 is FJ spectrum of Rayleigh wave using ZZ components, IR1 is FJ spectrum of Rayleigh wave using RZ and ZR components, IR2 is FJ spectrum
of Rayleigh wave using RR and TT components, and IL0 is FJ spectrum of Love wave using RR and TT components. The black arrow in (a) indicates
the location where the fundamental mode Rayleigh wave is bifurcated into two branches. The ith overtone is labeled next to each dispersion curve,
and 0 indicates the fundamental mode. Taken from Hu et al. (2020)
Fig. 13 Array-based dispersion measurement of the real data. a 1-D shear velocity structure modeled with six layers of variable thickness
(parameterization A). The thick and thin black lines represent the mean and standard deviation of the estimated models from the 100
bootstrap samples. b Rayleigh wave dispersion curves estimated with model A. The thick and thin white curves represent the mean and three
times the standard deviation of the 100 bootstrap results. The background image is the average of the variance reductions of the conventional
cross-spectral fitting for the ZZ, ZR, and RR components. The frequency range without the white shadow is used in the dispersion measurement. c
Love wave dispersion curves estimated with model A. Taken from Takagi and Nishida (2022)
components even if Love and Rayleigh waves appear in ωn r

Cm (ωn ) = , (80)
both the RR and TT components with shorter separa- zn+2m
tion distances. Figure 13 shows an example. Figure 13a
shows the S-wave velocity models, and Fig. 13b, c shows where zn of is the nth zero of the Bessel function J0,
the corresponding mode branches. Along the calcu- m = 0, ±1, ±2, which represents the ambiguity of n.
lated mode branches, VR is summed over the frequency This formulation allows missed or extra zero crossings
and all wave types. By maximizing VRall , we can infer the because observed noise makes measurement of zn diffi-
local 1-D model. This method searches for the global cult. The messed or extra zero crossings 2m arise from
maximum of VRall in the model space. Physically implau- the distinction between positive-to-negative zero and
sible models can be rejected automatically. The plot of VR negative-to-positive zero. The advantages of this method
against frequencies and phase velocities, as shown in the over the FTAN method are (i) applicability even for a
figures, is feasible to check the validity of the estimated shorter separation distance (typically shorter than three
model. Liu et al. (2023) proposed an inversion method, wavelengths) and (ii) smaller bias for shorter distances
which directly compared multimode dispersion diagrams owing to no far-field approximation. Although zero-
with the synthetic kernel of the Green’s function, which crossing measurements are practically robust (Ekström
shares a similar idea as the method here. et al. 2009), the low-SNR CCF behaviors become unsta-
ble. In particular, because strong lateral heterogeneities
5 Dispersion measurements for each path distort the phase information, appropriate filtering prior
In the second step of ANT (Fig. 1b), phase or group to application is crucial for stable measurements.
velocity is measured for each path. This section intro-
duces four methods to measure dispersion for each CCF:
(1) FTAN, (2) the zero-crossing method, (3) the SPAC- 5.3 SPAC‑based waveform fitting in phase velocity domain
based waveform fitting in the phase velocity domain, and This method measures the phase velocity anomaly
(4) the waveform fitting in a model space. FTAN and the in the ith frequency band from ωi to ωi+1 under the
zero-crossing method (Ekström et al. 2009) are feasible to assumption of single-mode dominance. If we already
measure the phase velocity of a single mode. The latter inferred local 1-D models based on the SPAC method,
two methods measure phase velocities by SPAC-based the synthetic cross-spectrum can be represented by
waveform fitting. The first measures the phase velocity
ωr

directory under an assumption of single-mode domi- �synth (ǫ; ω) = a(ω)J0 , (81)
C(1 + ǫ)
nance, whereas the second measures by fitting synthetic
CCF to the observed one with the help of reference CCF where ǫ is phase velocity anomaly. We infer δa and ǫ by
constructed by the local 1-D structural inversion. minimizing the squared difference defined by
ωi+1 2
5.1 FTAN S(αi , ǫi ) ≡ (1 + αi )�synth (ǫi ; ω) − �obs (ω) dω,
FTAN explained in Sect. 4.1 can also be used for phase ωi
and group velocity measurements of a single trace (e.g., (82)
Lin et al. 2008). We can measure the group velocities of where ǫi is the phase velocity anomaly at ωi , and αi is an
multimode surface waves by picking up local maxima amplitude correction term at ωi . αi is estimated first by
from a stacked FTAN diagram over an area (Spica et al. the least-squares method for ǫi . With the estimated αi , ǫi
2018). However, multimode interference makes phase is measured using the grid search method (Nagaoka et al.
velocity measurements using FTAN difficult. Even in the 2012; Yamaya et al. 2021).
case of multimodes, if we can isolate a single mode by a
time variable filter (Landisman et al. 1969) or (2) a float- 5.4 Multimode dispersion measurements by waveform
ing filter (Levshin et al. 1992), we can measure the phase fitting in a model space
velocities. In the case of the dense spacing of the mode branches
in the phase velocity domain, there is a method for esti-
5.2 Zero‑crossing method mating the multimode dispersion curve by waveform
The zero-crossing method (Ekström et al. 2009) is briefly fitting using a 1-D velocity structure as model param-
summarized here. We will consider the ZZ component eters. Yoshizawa and Kennett (2002) developed the
for simplicity, but it can be easily extended to other com- original idea of measuring the phase velocity disper-
ponents. For the frequency ωn of the nth zero-crossing sion curve for each path of teleseismic surface waves.
observed, the phase velocity can be evaluated by We applied this method to the cross-spectra of ambient
noise continuously (Takagi and Nishida 2022; Takeo et al.

2022). The advantages of this method are that it provides
physically achievable dispersion curves and can simul-
taneously measure dispersion curves for multimode
surface waves. Given the 1-D velocity structure, it is pos-
sible to calculate dispersion curves for all correspond-
ing higher-order modes, explaining observed waveforms
through the superposition of multimode surface waves.
For the superposition of multiple modes, it is necessary
to know the excitation amplitudes for each mode. In the
case of teleseismic surface waves, the excitation ampli-
tudes for each surface wave mode can be calculated from
the seismic source mechanism. We can use the ampli-
tudes aj calculated from the array-based SPAC method
for the cross-spectrum of ambient noise (Eq. 67). Given a
reference 1-D structure of the first step (Sect. 4), the syn-
thetic cross-spectrum can be represented by
N −1
synth

j ωr
� (δβ; ω) = a (ω)J0 , (83)
C j (β + δβ; ω)
j=0
Fig. 14 Dispersion measurement for a single path by waveform
fitting in a model space. a Modeled pair-specific 1-D shear
where δβ is a perturbation of 1-D structure, N is the num- velocity structure for the S-net station pair N.S2N08-N.S3N20.
ber of the mode branches, and j represents the order of The thick and thin black lines represent the mean and standard
the mode branch. We infer δβ by minimizing the residual deviation of the 1-D structures estimated by the bootstrap
sum of squares given by method. The gray line results from the reference model obtained
by the array-based SPAC method in Fig. 13a. b Phase velocity
dispersion curves of the fundamental-mode Rayleigh wave (0R),
2
ω1 w(ω)2 �synth (δβ; ω) − �obs (ω)
the first-overtone Rayleigh wave (1R), and the fundamental-mode
S(δβ) = dω. Love wave (0 L). The thin blue and red lines are the means and three
ω0 w(ω)2
times standard deviations of the estimated pair-specific dispersion
(84) curves. The thick blue and red curves represent the frequency
It is worth noting that the 1-D velocity structure, which ranges that meet the quality control criteria based on variance
serves as a model parameter, does not necessarily rep- reductions within narrow frequency bands. The gray ones are
the results of the array-based measurement in Fig. 13b, c. c Fitting
resent the true Earth structure (Yoshizawa and Ken- of the cross-spectra for this pair. The black and red are the data
nett 2002). The 1-D structure model is not unique with and the modeled cross-spectra, respectively. d Same as in (c),
respect to band-limited dispersion curves; different 1-D but both cross-spectra are normalized by the standard deviation
models can reproduce similar dispersion curves. The of the cross-spectral data estimated from the array-based SPAC
main products here are the multimode dispersion curves method. Taken from Takagi and Nishida (2022)
calculated by the 1-D structure. Thus, the search range of
the model space should be wide enough to reproduce a
dispersion curve that nicely fits the waveform. by utilizing the entire waveform of multicomponent
These equations correspond to the vertical compo- CCFs and applying physical constraints based on 1-D
nent but can be readily extended to other components. structure.
In waveform fitting, a relatively wide frequency range is
often handled, and multicomponent data can be simul- 6 2‑D multimode ANT
taneously processed. Therefore, it is important to con- In the third step of ANT (Fig. 1c), we conduct a 2-D
sider the weighting of the data in practical applications. inversion of the phase/group velocity of multimode sur-
The error estimate of the SPAC method based on the face waves from the phase/group velocity measurements
array can be utilized as the weights (Takagi and Nishida for all pairs of stations. The method has been developed
2022). to explore the mantle structure using earthquake data
Figure 14 exemplifies its application to S-net data (e.g., Woodhouse and Dziewonski 1984; Nakanishi and
(Takagi and Nishida 2022). Even in this case, where Anderson 1982; Tanimoto and Anderson 1985; Ekström
the signal-to-noise ratio of the CCFs is low, it is pos- et al. 1997; Nolet 2008). Because review papers covering
sible to robustly estimate multimode dispersion curves ANT are available (e.g., Barmin et al. 2001; Montagner,
2015 Rawlinson et al. 2010; Thurber and Ritsema 2015; ), 2005) enables us to separate the logarithmic amplitude
see them for details. term and the phase term.
For inversion, we should consider the sensitivity ker- Here, we consider the Born sensitivity kernel K, which
nels of the phase/group velocity. Although in an ideal- relates the phase velocity perturbation and the cross-
ized situation, the kernel is identical to a finite-frequency spectrum perturbation δ� as
kernel for an earthquake, a heterogeneous distribution of
δC
the noise source distorts the kernels (e.g., Nishida 2011; δ�(r1 , r2 ; ω) = K (r1 , r2 , r3 ; ω)d�3 , (85)
Tromp et al. 2010; Hanasoge 2013; Fichtner 2014; 2015; C
Fichtner et al. 2016; de Vos et al. 2013). where C is phase velocity and δC is the perturbation. The
The first subsection describes how source heteroge- locations r1, r2 , and r3 are shown in Fig. 15a. The kernel K
neities affect the kernel using an analytical method for a is given by
simplified case. The next section explains the conditions
for the application of surface wave tomography. The last K (r1 , r2 , r3 ; ω) = − 2k 2 �(r2 , r3 ; ω)G ∗ (r1 , r3 ; ω)

subsection briefly summarizes the 2-D inversion meth- +�∗ (r1 , r3 ; ω)G(r2 , r3 ; ω) ,

ods: local 1-D inversion, ray-theoretical inversion, and
(86)
finite-frequency inversion.
where �(r1 , r2 ; ω) is the cross-spectrum between r1 and
r2 at frequency ω (e.g., Nishida 2011), and G is the Green’s
6.1 2‑D phase velocity sensitivity kernel for ANT
function in frequency domain. Although the kernel was
The sensitivity kernel for 2-D phase velocity is evaluated generally calculated numerically, it can be estimated by
based on the Born and Rytov approximations (Born et al. observed data in principle if dense data were available
1999; Ishimaru 1997; Nolet 2008). The Born approxima- (Chmiel et al. 2018).
tion relates the scattered wave to the phase velocity per- With an assumption of the 2-D ambient seismic wave
turbations with the first-order approximation, but it does field described by a superposition of uncorrelated plane
not separate the effects on the amplitude and the phase. waves as in Sect. 2.3, Eq. 43 gives the analytic representa-
On the other hand, the Rytov approximation by taking tion of . Assuming that ωδt and kr are small enough, the
the logarithm of the wave (e.g., Yoshizawa and Kennett kernel can be rewritten by
−ik +i(χ23 +χ13 )

K (r1 , r2 , r3 ; ω) ≈ √ e B(θ2 ) − e−i(χ13 +χ23 ) B(θ1 )
π r13 r23 (87)

+ei(χ13 −χ23 ) (B(θ2 + π ) − B(θ1 + π )) ,
(a) (b)
r 13 < r 23 r 13 > r 23
Station 1 Station 2
θ1 θ2
r 12
r1 r2 r1 r2
r 13
r 12
r3
Fig. 15 a Schematic map of the geometry of stations. The star symbols show the station locations at r1 and r2. The open circle shows the location
of a phase velocity anomaly at r3. b Elliptic curves with constant |r13 − r23 |, and hyperbolic curves with constant |r13 − r23 |
where B(θ) represents the intensity of incident plane Most studies on ANT used the time-symmetric part of
waves as a function of azimuth θ , and rij is the distance the cross-correlation function to measure phase velocity
between ri and rj , χij ≡ krij − π/4 (see Fig. 15a). The anomalies. The time-symmetric part of cross-spectrum ¯
form of a phase sensitivity kernel for a homogeneous is defined by
source distribution is the same as that of the earthquake
�c + �ac∗
data (Nishida 2011). ¯
�(ω) ≡ , (93)
The Rytov approximation is feasible to evaluate the 2
phase sensitivity kernel (e.g., Yoshizawa and Ken- We consider the relation between perturbation of
¯ and
nett 2005; Nishida 2011). Although the approximation the corresponding Born kernel K̄ as
requires a single-wave packet, the cross-spectrum
given by Eq. 43 (θ = 0, and r = r12) includes both the δc
¯ =
δ� K̄ (r1 , r2 , r3 ; ω)d�3 , (94)
causal and acausal parts. We divide the cross-spectrum c
and the kernel K into causal and acausal parts accord-
where
ing to the sign of the exponent, approximately. The causal
part can be represented by e−kr+ωt , while the acausal part δ�c + δ�ac∗
can be represented by ekr+ωt , where the wavenumber k
¯
δ �(ω) ≡ (95)
2
and the angular frequency ω are positive. The cross-spec-
trum is divided into the causal part c and the acausal K a (r1 , r2 , r3 ; ω) + K ac∗ (r1 , r2 , r3 ; ω)
part ac as, K̄ (r1 , r2 , r3 ; ω) ≡ .
2
(2) (96)
�c (ω) = H0 (kr12 )eiωδt(r12 ,π ) B(π )
(1) (88) The Rytov approximation is employed to obtain a
�ac (ω) = H0 (kr12 )e−iωδt(r12 ,π ) B(0). phase sensitivity kernel for phase velocity perturba-
tions (Nishida 2011). In the Rytov method, the loga-
The Born sensitivity kernel K is also divided into the
rithm of the cross-spectrum ¯ is considered instead of
causal part K c and the acausal part K ac according to the
sign of the exponent as,
√−ik −ik(χ13 +χ23 ) B(θ ) + eik(χ13 −χ23 ) (B(θ

c π r13 r23 −e 1 2 + π ) − B(θ1 + π )) r13 < r23
(89)
K ≡ −k
−ik(χ +χ ) ,
√
π r13 r23 −e 13 23 B(θ )
1 r13 ≥ r23
and
√−ik ik(χ13 +χ23 ) B(θ )

ac π r13 r23 e 2 r13 < r23
(90)
K ≡ −k +χ ,
ik(χ ) B(θ2 ) + eik(χ13 −χ23 ) (B(θ2

π r13 r23 e
√ 13 23 + π ) − B(θ1 + π )) r13 ≥ r23
where the region r13 > r23 corresponds to the first and
the cross-spectrum itself. By taking the logarithm,
¯ can
fourth quadrants, whereas the region r13 < r23 corre-
be divided into real (amplitude) and imaginary (phase)
sponds to the second and third quadrants (Fig. 15b).
parts, as follows:
The Born kernels also satisfy K = K a + K ac from the
definition. ¯ = ln(Ae−iψ ) = ln A − iψ,
ln � (97)
The causal perturbation δ�a and acausal perturbation
δ�ac are related to phase velocity anomalies as where A is the amplitude of the causal part, and ψ is its
phase. The phase perturbation δψ can be written by
c δc c
δ� = K (r1 , r2 , r3 ; ω)d�3 , (91)
δc
c
δψ = Kp (r1 , r2 , r3 ; ω)d�3 . (98)
c

δc ac
δ� ac
= K (r1 , r2 , r3 ; ω)d�3 . (92) where the phase sensitivity kernel Kp is the imaginary
c part of −K̄ /�
¯.
(a) The first term (b) The second term (c) The whole kernel
x3
100
50
Distance [km]
r1 r2
0
− 50
− 100
− 100 − 50 0 50 100 − 100 − 50 0 50 100 − 100 − 50 0 50 100

Distance [km] Distance [km] Distance [km]
−1 0 1
Amplitude
Fig. 16 a The first term of the phase kernel Kp for B = cos θ + 1. The first term has an elliptical shape. The yellow triangles show the station location.
b The second term of the phase kernel Kp. The second term has a hyperbolic shape. c The whole kernel of the phase kernel Kp. The background
phase velocity is 3 km/s. The frequency range is from 0.1 to 0.2 Hz. We also apply a cosine taper for averaging in the frequency range
Then, the phase kernel Kp can be written by
k 3/2

r1 π B(θ2 ) + B(θ1 )
Kp = √ − cos k(r13 + r23 − r12 ) −
2π r13 r23 4 B(0) + B(π )
(99)
π B(θ2 + π ) − B(θ1 + π )
+ sin k(|r13 − r23 | + r12 ) + ,
4 B(0) + B(π )
where we neglect the first-order term. kernel of an earthquake (e.g., Spetzler et al. 2002; Yoshi-
Figure 16 shows an example of sensitivity kernels. The zawa and Kennett 2005). Because an approximation of
separation distance of the station pair is 150 km, and the the finite-frequency kernel of an earthquake works for
phase velocity is 3 km/s. The source B(θ) is assumed to most cases practically, most studies did not consider the
be cos θ + 1. source heterogeneity effect of the kernel. To check the
The first term has an elliptical shape (Fig. 15b), as effects, travel-time anomalies due to the source hetero-
shown in Fig. 16a, corresponding to the finite-frequency geneities given by Eq. 42 can be a common criterion. In
kernel of an earthquake. The kernel along the line particular, because source heterogeneities cause apparent
between the station pair (θ1 = 0 and θ2 = π ) is identi- azimuthal anisotropy, the travel-time anomalies given by
cal to the finite-frequency kernel of an earthquake. The Eq. 42 are useful to evaluate accuracy. Although attenu-
width of the kernel at the midpoint is proportional to
√ ation measurements are beyond the scope, the corre-
r12 , where is the wavelength 2π/k . As a location sponding amplitude kernel is more sensitive to the source
moves away from the line between, the source heteroge- distribution, so that the source heterogeneities should be
neity increases the sensitivity on the r2 side. considered.
The second term with the dependency of |r13 − r23 |
exhibits a hyperbolic shape (Fig. 15b) caused by source
6.2 Conditions for application of surface wave
heterogeneity (Fig. 16b), and the term oscillates rapidly
tomography
outside the stationary zones behind it. The effect tends to
This subsection briefly summarizes the conditions for
be smoothed out when the typical spatial scale of the lat-
applying the 2-D inversion method, known as phase/
eral heterogeneity is larger than the wavelength.
group velocity tomography. Surface wave tomogra-
When the spatial scale of the source distribution is
phy requires weak scattering of the surface waves dur-
large (i.e., the azimuthal dependence of B(θ) is smooth),
ing propagation because we measure phase or group
the shape of the kernel is similar to the finite-frequency
Fig. 17 Log–log plot of reported isotropic scattering coefficients against frequency. The left vertical axis shows the isotropic scattering coefficient
giso, and the right vertical axis shows the corresponding mean free path ( g−1
iso ). Grey lines are for the lithosphere. Taken from Fig. 1 of Sato (2019)
velocity anomalies of the direct waves. Because the prop- free path. Otherwise, strong scattering makes phase or
agation distance ranges from hundreds to several thou- group velocity measurements difficult. Figure 17 shows
sand km, the distance should be shorter than the mean the mean free path as a function of the frequency com-
piled by Sato (2019). When we explore the mantle struc-
ture using earthquake data, we use surface waves below
0.05 Hz. The mean free path longer than 1000 km at the
frequency validates the application of the method of sur-
face wave tomography.
To understand applicability, Fig. 18 shows a classifica-
tion of wave propagation and the applicable method in
ka − kL diagram (Aki and Richards 1980), where k is the
wavenumber, a is the scale of heterogeneities, and L is
the propagation distance. Here, we assume that the mean
free path is comparable to a, although this estimate is
very crude. The mean free path of northern Japan from
0.5–1Hz is estimated to be about 30 km (Hirose et al.
2020), and that of Germany is estimated at 0.2 Hz to be
about 500 km (Sens-Schönfelder and Wegler 2006b). The
red lines show propagation distances from 10 to 150 km,
and the blue broken lines show those from 150 to
2000 km. At 0.03 Hz, the surface wave propagation at tel-
eseismic distance beyond 2000 km is within the regime of
Fig. 18 ka − kL diagram (Aki and Richards 1980) for a classification ray theory. At 0.2 Hz, the strong scattering makes meas-
of wave propagation, where k is the wavenumber, a is the scale
of heterogeneities, and L is the propagation distance. Red lines show
urements difficult for propagation distances longer than
propagation distances from 10 to 150 km, and blue lines show those 200 km. This is the reason why earthquake surface wave
from 150 to 2000 km. The mean free path at 0.03 Hz is 5000 km, tomography is practically difficult above 0.1 Hz. Before
that at 0.1 is 500 km, and that at 0.2 Hz is 30 km
(a) Earthquake surface wave tomography (b) ANT
Fig. 19 a A typical example of earthquake-receiver geometry for earthquakes. The gray lines show ray paths between the earthquakes
and the stations. Red stars show the earthquakes, and black dots show stations. b A typical example of station geometry for ANT. The lines show ray
paths between the station pairs
applying ANT, setting up an appropriate problem based the phase differences between seismograms for two sta-
on the observed situation is important. tions on a common great-circle path. The drawbacks
Let us consider the difference in ray geometry between of the two-station method are (1) that it requires both
ANT and earthquake surface wave tomography. In the stations to be located close to the common great-circle
case of regional-scale earthquake surface wave tomog- path, which reduces the available ray paths, and (2) that
raphy with source-receiver paths, the target area is sur- the longer earthquake-receiver distance tends to cause
rounded by earthquakes, which are limited along active the wave propagation to enter the regime of multiple
tectonic areas (Fig. 19a). Because most ray paths travel scattering at a higher frequency (Fig. 18).
across the area, the travel-time anomalies reflect the The shape of the sensitivity kernel is elliptical along
seismic structure along lines over the scale. Because the the ray path,
√ and the width at the midpoint is propor-
perturbations of the inferred phase/group velocity map tional to r12 , where is the wavelength. Because the
strongly depend on the damping of the tomographic lengths of ray paths of earthquake data become longer
inversion, the absolute velocities tend to be ambigu- in general, the widths become wider than those of
ous. Although the damping problem is common in ray- ANT. The wider width tends to average the anomalies
theoretical inversion, including ANT, an initial local 1-D within the fat ray, and the averaging over a long dis-
model for ANT constructed from multimode dispersion tance makes restoring the small-scale image difficult.
measurements using subarrays (see Sect. 6.3.1) can miti- The advantages of ANT are the homogeneous path dis-
gate the ambiguity. tribution and the shorter ray paths. These features enable
In contrast, in the case of ANT with a dense array, us to estimate phase/group velocity maps above 0.1 Hz.
the distribution of the ray path is uniform (Fig. 19b).
Above 0.05 Hz, CCFs with longer separation distances 6.3 Implementations of the 2‑D phase/group velocity
(typically longer than 1000 km) become complex. Ray inversions
paths with separation distances shorter than 1000 km The subsection briefly summarizes implementations of
enable us to infer tomographic maps even in regions 2-D phase/group velocity inversions using phase/group
with strong lateral heterogeneities above 0.1 Hz. This velocity anomalies measured from ambient noise CCFs:
situation is similar to earthquake tomography using (1) regionalization using subarrays, (2) the ray-theoreti-
the two-station method (e.g., Dziewonski and Hales cal inversion of ANT, and (3) finite-frequency inversion
1972; Hamada and Yoshizawa 2015), which measures of ANT.
6.3.1 Regionalization using subarrays

If dense stations are distributed over a large area, the sta-
tions can be divided into local subarrays. Phase/group
velocity maps can be inferred from phase/group veloc-
ity measurements with the subarrays, such as F-K analy-
sis (Jiang and Denolle 2022), FJ method (Fu et al. 2022;
Nimiya et al. 2023), and slant stack (Qin et al. 2022).
These methods can easily incorporate information on
multimodes with smaller amplitudes because array analy-
sis enhances the SN ratios.
The inferred map is also helpful as an initial model
for further inversions. Most ANT studies employ lin-
earized inversion to obtain 2-D phase/group veloc-
ity maps (e.g., Rawlinson and Sambridge 2004) because
the computational cost of fully nonlinear inversion (e.g.,
Markov Chain Monte Carlo (MCMC); Bodin and Sam-
bridge 2009; Bodin et al. 2012; Galetti et al. 2016) is still
high. A good initial model is crucial for better modeling
for linearized inversion, because an inappropriate initial
model may lead to a solution with a local minimum. An
inappropriate initial model sometimes requires strong
damping to stabilize the inversion. The appropriate initial
model could mitigate the degradation of a tomographic
image.
6.3.2 Ray‑theoretical inversion of ANT

The most common method of tomographic inversion of
Fig. 20 a Evolution of the wavefront in the case of strongly
phase/group velocity anomalies is ray-theoretical inver- heterogeneous media. The station with a green star acts
sion with the assumption of a high-frequency limit. For as the source, and the red triangles show the stations
example, tomo_sp_cu_s is a program (Barmin 2018) concerning the advancing wavefronts. Note the triplication effects
designed for estimating 2-D models from regional- or around the low-velocity zones. b Histograms of passage time
residuals for measurements at 10 s for Rayleigh waves, initially
global-scale surface wave phase/group velocity measure-
compared to a uniform reference model with a velocity of 2.8 km/s
ments (Barmin et al. 2001; Ritzwoller 2002). The program with great-circle paths, and after the inversion with 6 iterations
also assumed a straight ray for a weakly heterogeneous of the subspace method with deviated paths. Taken from Fig. 4
medium. Another implementation to solve the eikonal of Saygin and Kennett (2012)
equation is the fast marching method (Sethian 1996;
Rawlinson and Sambridge 2004; White et al. 2020). SUR-
FTOMO (Rawlinson 2008) is an iterative nonlinear travel- eikonal tomographic method tracks the evolution of the
time tomography code in 2-D spherical shell coordinates. wavefront using only local information with dense array
Figure 20 shows an example of wavefront tracking using data. The phase velocity can be estimated by the in situ
the final tomographic results. Although the figure shows spatial gradient of the phase travel time without an
the concentric wavefront within the arc distance of 10◦ , it inversion. The method can be applied to a more general
shows kinks in Eastern Australia, which can be explained condition of the source distribution than equal energy
by triplication due to the low-velocity zones. This figure distribution because the method does not require CCFs
also suggests that CCFs with a separation distance longer to be related to the corresponding Green’s function. This
than 10◦ are difficult to model above 0.1 Hz without the method works as long as the phase travel time satisfies
help of a good initial model. An advantage of ANT is that the eikonal equation (see the end of Sect. 2.3; Snieder
we can choose ray paths that meet the conditions accord- et al. 2006; Lin et al. 2013; Lehujeur and Chevrot 2020).
ing to the problem setting owing to the homogeneous If a dense array satisfies the station spacing shorter than
distribution of the ray paths. the wavelength, this method is feasible to infer the phase
The eikonal tomographic method (for example, Lin velocity map (e.g., de Ridder and Dellinger 2011; Lin et al.
et al. 2009) also assumes a high-frequency limit with a 2013; Ritzwoller and Feng 2019; Chen et al. 2022).
single mode and solves the eikonal equation directly. The
6.3.3 Finite‑frequency inversion of ANT 7 3‑D seismic velocity inversion using

Although the patterns of the ray-theoretical inversions the multimode dispersion
are robust in different studies, the retrieved strength of In the fourth step of ANT (Fig. 1d), a 3-D seismic veloc-
phase/group velocity perturbations differs (e.g., Yoshi- ity structure is constructed from the inferred phase/
zawa and Kennett 2004). Finite-frequency inversion is group velocity maps in the previous step, which exhibit
crucial to improve the amplitude recovery of velocity per- the dispersion curves at each grid point. The local 1-D
turbations. As shown in Sect. 6.1, the sensitivity kernel structure at each grid point can be inverted from the
has an elliptical√ shape, and the width at the midpoint is dispersion curves. 3-D seismic velocity structures can
proportional to r12 . For a heterogeneous data set with then be constructed by collecting local 1-D structures.
multiple scales, the difference in the width of the sensitiv- In most cases, only the dispersion of fundamental modes
ity kernels becomes problematic, because it becomes dif- (Love and Rayleigh waves) is utilized for the inversion
ficult to recover the amplitude of velocity perturbations because the excitation sources are distributed on the sur-
by a ray-theoretical inversion. This subsection summa- face. Recent studies reveal that a dense array enables us
rizes recent developments on finite-frequency inversions. to utilize multimode dispersion. Joint inversion with the
As explained in Sect. 6.1, a finite-frequency sensitivity H/V or receiver function is feasible to improve the depth
kernel of a CCF also depends on the source distribution. resolution.
Consequently, the phase/group velocity structure has a
trade-off with the source distribution (Fichtner 2015). To 7.1 1‑D seismic velocity inversion at each grid
improve the structural image, we must infer the source The measured dispersion curves are inverted to obtain
distribution simultaneously (Fichtner et al. 2016; Sager local 1-D structures. Inversion methods can be catego-
et al. 2020, 2017). To mitigate the effects of source het- rized into a Monte Carlo method (Shapiro and Ritzwoller
erogeneities, measurements of differential travel times 2002; Maraschini and Foti 2010) and a linearized method
between two pairs of three stations were proposed (Liu (Tarantola and Valette 1982; Tarantola 2005; Montag-
2020). The sensitivity kernels of the three-station method ner 2015). Because the local 1-D inversion is identical to
significantly reduce the effect of the heterogeneous the earthquake data, this section briefly summarizes the
source distribution. Because CCF amplitudes are more related studies mentioned above.
sensitive to source heterogeneities (Tsai 2011), the con- All 1-D inversions require a code to calculate the sur-
siderations of the source information will become more face wave dispersion (e.g., DISPER80, Saito 1988; sen-
crucial in inferring the attenuation structure. skernel-1.0, Levshin 2018; Computer Programs
Because the computational cost of the sensitivity ker- in Seismology, Herrmann 2013). These codes also
nels for both the structure and the source distribution calculate the depth sensitivity kernel, which is also use-
is still high, the earthquake sensitivity kernel for phase/ ful for a Monte Carlo inversion. Figure 21 shows a typical
group velocities is instead used practically (Gao and example of depth sensitivity kernels for a model based on
Shen 2015a; Yang and Gao 2018; Covellone et al. 2015; a reference Earth model PREM (Dziewonski and Ander-
Gao and Shen 2015b; Emry et al. 2019; Yang et al. 2022; son 1981), where the thickness of the ocean is changed
Gao and Shen 2014; Wang et al. 2019b 2020; Lu et al. from 3.0 to 4.6 km, and the thickness of the crust is
2020; Crosbie et al. 2019; Janiszewski et al. 2019; Russell reduced from 22 to 6 km (Takeo et al. 2013). Fundamen-
and Gaherty 2021; Zhao et al. 2020). The approximation tal Love and Rayleigh waves have sensitivity at shallower
becomes exact under the condition of equipartition of depths. The higher modes are not sensitive to the S-wave
energy (Nishida 2011). Although the kernel does not con- structure at shallower depths, and they are more sensitive
sider the source heterogeneities, it improves the image to the S-wave velocity structure at greater depths than
from ray-theoretical inversion. the corresponding fundamental models. With increas-
Another implementation of the finite-frequency effect ing periods, the kernels have higher sensitivity at greater
of ANT is Helmholtz tomography (Lin and Ritzwoller depths. A multimode inversion can improve depth reso-
2011b; Lin et al. 2012b; Mordret et al. 2013). This method lution because overtones have different sensitivities from
is a natural extension of eikonal tomography to consider fundamental modes.
finite-frequency effects by introducing the amplitude Figure 22 shows an example of the 1-D inversion
correction term. Although this method requires dense (Yamaya et al. 2021) using multimode dispersion under
data, the phase velocity can be estimated without the the ocean. The study used a computer program package
sensitivity kernel and the inversion, as in the case of eiko- SEIS_FILO (Akuhara 2022) for the inversion, which
nal tomography. This method can be applied to a more implements a transdimensional Bayesian interface and
general condition of the source distribution (see the end reversible-jump MCMC (rj-MCMC) (Green 1995). They
of Sect. 2.3). assumed a uniform distribution for the prior probability
(a) (b) 0th Rayleigh 1st Rayleigh

6 0 0
10s
5 1st Love 25s 5s 10s
Depth (km)
0th Love 5 5
c (km/s)
4 h
leig 5s
Ray 0th Rayleigh
3 1st
10 10
2
1 15 15
5 10 20 50 100 0.0 0.5 1.0 0.0 0.5 1.0
Period (s) Kα (km−1) Kα (km−1)
(c) 0th Rayleigh 1st Rayleigh 0th Love 1st Love

0 0 0 0
5s 5s 10s 5s
50
10s
Depth (km)
10s
100
10s
25s 25s
150
5s
25s
200
50 50 50
0.00 0.05 0.10 0.00 0.05 0.10 0.00 0.05 0.10 0.000 0.005
Kβ (km−1) Kβ (km−1) Kβ (km−1) Kβ (km−1)
Fig. 21 a Phase velocities of the fundamental (0th) and first-higher modes of surface waves for the modified PREM (Dziewonski and Anderson
1981). The sudden phase velocity decrease marked by the ellipse corresponds to the transition of modal energy from solid to ocean. b, c
Sensitivities of phase velocities to P-wave and S-wave velocities (Kα and Kβ , respectively). The two lines in each panel show the depths of the seafloor
and Moho. Taken from Takeo et al. (2013)
distribution (1–10 for the number of layers; 2.3–10 km this paper focuses on multimode ANT. Since there is no
for depths of the discontinuities; and 2.3 km for seafloor space to discuss individual models, we introduce Data
depth and 0.1–5 km/s for S-wave velocities). P-wave Services Products: EMC, A repository of
velocities and densities were scaled by the empirical Earth models (Trabant et al. 2012; IRIS DMC 2011),
relation (Brocher 2005). The multimode dispersion con- which is a resource that offers access to different Earth
straints enable us to infer both sediment thickness and models, along with visualization tools to preview mod-
S-wave velocity. To demonstrate the feasibility of mul- els, facilities to extract data and metadata from the mod-
timode measurements to improve depth resolution, let els, and access to software and scripts that others in the
us compare the inversions of Fig. 22 with and without research community have contributed.
higher-mode surface wave dispersion. Figure 23d shows Seismic anisotropy gives us a clue to understanding the
the inversion result using multimode dispersion, whereas rheology and deformation of the Earth. In some cases,
Fig. 23e shows the result using only fundamental modes. an isotropic S-wave velocity model cannot explain the
These figures show that the inversion using multimode observed discrepancy between the Love wave dispersion
dispersion is feasible to constrain the shallow sediment and the Rayleigh wave dispersion, which can be explained
structure, improving the resolution of the crustal struc- by radial anisotropy (e.g., Aki and Kaminuma 1963;
ture. Although Rayleigh waves also have sensitivity to Anderson 1962). Radial anisotropy in the upper mantle
the P-wave velocity, the limited sensitivity only at shal- can be explained by lattice-preferred orientation in the
low depths (Fig. 21b) makes it difficult to constrain the low-velocity zone or partial melt under shear stresses
P-wave velocity structure practically. With an empirical (e.g., Montagner 2015). Radial anisotropy also exists in
relation (e.g., Brocher 2005), the P-wave velocity and the the lower crust, which can be explained by the lattice-
density are scaled to the S-wave velocity to reduce the preferred orientation (LPO) of anisotropic crustal min-
number of parameters. erals under extension (Moschetti et al. 2010a, b; Huang
Many groups have already published crust and upper- et al. 2010). ANT also revealed radial anisotropy in vol-
most mantle models using fundamental modes, although cano areas (Jaxybulatov et al. 2014; Harmon and Rychert
Fig. 22 a Stations of the ocean bottom seismometers array in the region off Ibaraki. The station intervals are about 6 km. The yellow triangles show
the stations deployed on October 17, 2010, and the red triangles show those deployed on February 15, 2011. b, c S-wave velocities along the A-A’
and B-B’ lines. The red, orange, and green regions reflect layer 1, layer 2, and layer 3, respectively. We interpret the red and orange regions to be
sedimentary layers, the green region to be Cretaceous sediment, and the blue region to be the upper crust, respectively. The black lines show
the depth of the acoustic basement (between the orange and green regions) and the top of the upper crust (between the green and blue regions).
Modified from Figs. 1 and 8 of Yamaya et al. (2021)
2015; Spica et al. 2017; Lynner et al. 2018; Jiang and 7.2 Joint inversion with other observations
Denolle 2022; Jiang et al. 2023, 2018; Miller et al. 2020). Although multimode ANT improves depth resolution,
The layering of a partial melt layer in the magma reser- surface wave tomography has an inherent low depth res-
voir can explain them. Thus, radial anisotropy is feasible olution. Joint inversions with other geophysical observa-
for discussing the stress state and the texture. Azimuthal bles are feasible for a better depth resolution at the 4th
anisotropy can also provide information on rheology and step.
deformation (Lin et al. 2009; Yao et al. 2010; Ritzwoller A thin low-velocity layer near the surface is problem-
and Feng 2019; Russell and Gaherty 2021): for example, atic. Rayleigh waves are essentially less sensitive to the
the flow in the uppermost mantle related to plate motion S-wave velocity structure near the surface, as shown in
(Takeo et al. 2016). We note that apparent anisotropy Fig. 21. The poor sensitivity of Rayleigh waves at shal-
due to the lateral heterogeneities of the isotropic S-wave low depths makes resolving the shallow structure dif-
velocity structure is problematic (Lin and Ritzwoller ficult, although the layer can still change the dispersion
2011a; Fichtner et al. 2013). When estimating anisotropy curves of surface waves. The polarization of the Rayleigh
by the tomographic method, the anisotropy has trade- wave provides information on the shallow S-wave veloc-
offs with the lateral heterogeneities. Moreover, ANT has ity structure beneath a seismic station. If the amplitude
another difficulty that source heterogeneity can yield up ratios between the horizontal and vertical components
to about 1% apparent azimuthal anisotropy (Harmon (ellipticity) for individual mode branches are available,
et al. 2010; Takeo et al. 2014, 2016), which can be esti- the joint inversion can constrain shallower depths (Lin
mated by Eq. 42 (equivalent to Weaver et al. (2009)). et al. 2012a). Even if the ratios of individual modes are
Phase velocity (km/s)
(a) Fundamental Rayleigh wave (d) Multi-mode (e) Fundamental mode
Posterior probability
4 0
0
3 0.1
2
0.1
1
0 layer 1
0.1 0.2
Frequency layer 2 0.05
4 (b) First overtone 0.1
Depth (km)
Depth (km)
3
5 layer 3 5
2
1
0
0.2 0.3
Frequency
(c)Fundamental Rayleigh wave

3 0.1
2 10 0 10 0
0 1 2 3 4 5 0 1 2 3 4 5
1 S wave velocity (km/s) S wave velocity (km/s)
0
0.1 0.2
Frequency
Fig. 23 Comparison of 1-D S-wave velocity structure inversions using both the fundamental and the first-higher modes of Rayleigh wave (a,
b, d) with that using only the fundamental modes (c, e). a Phase velocity of the fundamental mode of the Rayleigh wave. The blue points show
the average phase velocities. The Markov chain Monte Carlo method calculated the posterior probabilities for 1-D average S-wave velocity
structures. b Phase velocity of the first-higher mode of the Rayleigh wave. Blue points show the 1-D average phase velocities. c The phase velocity
of the fundamental mode of Rayleigh waves with posterior probabilities using only fundamental modes. d S-wave velocity structure inferred
by the MCMC method using both the fundamental and the first-higher modes of Rayleigh wave. The blue line shows the median velocity at each
0.1 km depth grid point. Layers 1 and 2 are well constrained despite the given loosely bounded uniform priors. e S-wave velocity structure inferred
by the MCMC method using only the fundamental modes. Modified from Figs. 5 and 10 of Yamaya et al. (2021)
difficult to measure, the spectral ratio of the horizontal 8 Conclusions

to vertical (H/V) components at a station (Nakamura The first section introduces theories of SI for ANT.
1989; Sánchez-Sesma et al. 2011; Bahavar et al. 2020) Although they are based on different assumptions, the
can be robustly measurable. The H/V ratio allows us to resultant CCFs can be related to the Green’s function
constrain the shallow structure by joint inversion with an under the equipartition of energy. In a realistic situa-
assumption of equipartition of modes (Spica et al. 2018). tion, although the modal energy is not equally distrib-
We note that the ellipticity also has information on the uted among different mode branches, they work for the
azimuthal anisotropy (Lin and Schmandt 2014) because individual mode branches. Heterogeneities of the noise
the ellipticity also depends on the azimuth from one sta- sources break the equipartition of energy even for a sin-
tion to the other station. gle-mode branch. We evaluated the apparent anomaly
In particular, single-mode surface wave inversion is dif- caused by the source heterogeneities. The important
ficult to distinguish between the smooth Vs model and point is that the CCF still satisfies the wave equation even
the layered model because the dispersion curves of sur- for a heterogeneous source distribution, leading to the
face waves are insensitive to velocity jumps. A joint inver- robustness of the ANT results.
sion with the receiver function allows us to improve the We summarize practical data processing on the calcula-
resolution of the seismic discontinuities (e.g., Bodin et al. tion of CCFs. For calculating a good quality CCF (satisfy-
2012; Gao and Shen 2015b; Zhou et al. 2020; Akuhara ing the theoretical assumptions), it is important to choose
et al. 2023; Ai et al. 2023). A joint inversion of multimode appropriate data processing depending on the behavior of
dispersion with the receiver function for earthquake the data and the noise. Because recent phase velocity meas-
data (e.g., Taira and Yoshizawa 2020) could be feasible urement methods used waveform information for precise
to improve depth resolution, although not yet applied to measurements, amplitude information should be consid-
ANT. ered by careful data selection and normalization of CCFs.
Since most conventional ANT studies utilized only the MCMC Markov Chain Monte Carlo
MS Mean Squared amplitude
fundamental modes, it has inherited a great uncertainty multimode ANT Ambient noise multimode surface wave tomography
regarding the depth structure. Recent developments in NLNM New Low Noise Model (Peterson 1993)
measurement techniques and dense observations ena- PREM Preliminary Reference Earth Model (Dziewonski and
Anderson 1981)
ble us to utilize information on multimode dispersion. rj-MCMC Reversible jump Markov Chain Monte Carlo
We show a typical inversion of multimode ANT, which RMS Root Mean Squared amplitude
consists of the following four steps. The first step is the SI Seismic Interferometry
SPAC SPatial AutoCorrelation method
multimode dispersion measurement of surface waves USArray A 15-year program to place a dense network of perma-
to construct local 1-D structures. The second step is to nent and portable seismographs across the continental
measure the phase/group velocity for each path. Mul- USA. Seismographs record local, regional, and distant (tel-
eseismic) earthquakes. https://doi.org/10.7914/SN/TA
timode interference was problematic in measuring
multimode dispersion because most studies implicitly Acknowledgements
assumed the dominance of the fundamental mode. The The authors thank many people for maintaining the networks and making the
data readily available. The authors thank N. Nakata, K. Yoshizawa, and T. Tani-
dispersion measurements by waveform fitting in a model moto for constructive comments that significantly improved the quality of the
space are feasible for the multimode case. The third step manuscript. The preprocessing and postprocessing of the data analysis were
is the 2-D phase/group velocity inversions. Ray-theo- carried out using ObsPy (Krischer et al. 2015), NumPy (van der Walt et al. 2011),
pyGMT (Tian et al. 2023, Wessel et al. 2019), and SciPy (Virtanen et al. 2020).
retical inversion was commonly used in most cases, but
the finite-frequency effect has been considered recently. Author Contributions
To evaluate the effects of source heterogeneities on the KN contributed to conceptualization; RT and AT contributed to the writing of
the original draft preparation. All authors read, commented on, and approved
sensitivity kernel, we show an analytic kernel with source the final manuscript.
heterogeneities in a simplified case. Helmholtz tomog-
raphy is also feasible for considering finite-frequency Funding
K. Nishida was supported by JSPS KAKENHI (Grand No. 21K03715), R. Takagi
effects and the source heterogeneities for sufficiently was supported by JSPS KAKENHI (Grand No. 23K03522), and A. Takeo was
dense stations compared to the wavelength because supported by JSPS KAKENHI (Grand No. 21K14009).
CCFs for a heterogeneous source distribution still satisfy
Availability of data and materials
the wave equation. The fourth step is a local 1-D inver- The Python code of the synthetic tests (Figs. 7, 9, 10 and 16) will be available
sion on each grid to construct a 3-D S-wave tomographic on the Zenodo page https://doi.org/10.5281/zenodo.10313732.
model. The multimode inversion improves the depth res-
olution of the S-wave velocity structure. Joint inversions Declarations
with other geophysical observations, such as the H/V
Conflict of interests
ratio and receiver functions, could be feasible to improve The authors declare that they have no conflict of interest.
depth resolution further.
Received: 29 September 2023 Accepted: 23 December 2023

Abbreviations
1-D, 2-D, 3-D One-dimensional, two-dimensional, and therein-dimensional
ANT Ambient noise surface wave tomography
CCF Cross-Correlation Function, which is the time-domain
representation of a cross-spectrum References
CMT Centroid Moment Tensor Abramowitz M, Stegun IA, Romer RH (1988) Handbook of mathematical
FFT Fast Fourier transform functions with formulas, graphs, and mathematical tables. Am J Phys
FJ method The frequency–Bessel method 56(10):958. https://doi.org/10.1119/1.15378
FTAN Frequency–time analysis Ai S, Akuhara T, Morishige M, Yoshizawa K, Shinohara M, Nakahigashi K
Hi-net A high-sensitivity seismograph network of nearly 800 (2023) Layered evolution of the oceanic lithosphere beneath the
stations with an average spacing of 20 km. Seismom- Japan basin, the sea of Japan. J Geophys Res. https://doi.org/10.
eters are installed at the bottom of boreholes at a depth 1029/2022jb025581
of 100–3500 m to reduce noise caused by winds, ocean Aki K (1955) Relay computer for correlogram analysis. Zisin 8(1):56–58
waves, and human activity. Observed data is digitized Aki K (1957) Space and time spectra of stationary stochastic waves, with spe-
at each station and continuously transmitted to NIED. cial reference to microseisms. Bull Earthq Res Inst 35:415–457
https://doi.org/10.17598/NIED.0003 Aki K (2003) Earthquake prediction by monitoring the loading process of
H/V spectral ratio The ratio between the Fourier amplitude spectra of the plate-driving forces at the Brittle-Ductile transition zone. In: Seventh
horizontal and vertical components of microtremors. workshop on non-linear dynamics and earthquake prediction. Trieste,
IRIS IRIS (Incorporated Research Institutions for Seismology[1]) Italy, pp 1–166
was a university research consortium dedicated to explor- Aki K, Kaminuma K (1963) Phase velocity of low waves in Japan (part 1):
ing the interior of the Earth by collecting and distributing Love waves from the Aleutian shock of march 9, 1957. PhD thesis, The
seismographic data. On January 1, 2023, IRIS merged with University of Tokyo
UNAVCO to form the EarthScope Consortium. Aki K, Richards PG (1980) Quantitative seismology, vol 2. San Francisco, W. H.
LPO Lattice-preferred orientation Freeman, p 557
mHz Milli-Hertz (10−3 Hz)
Akuhara T (2022) SEISmological transdimensional inversion tools for Flat and processes through active seismic interferometry. Geophys J Int
Isotropic Layered structures in the Ocean (SEIS\_FILO). https://doi.org/ 214(2):1468–1480. https://doi.org/10.1093/gji/ggy211
10.5281/zenodo.6330840 Chmiel M, Mordret A, Boué P, Brenguier F, Lecocq T, Courbis R, Hollis D, Camp-
Akuhara T, Yamashita Y, Ohyanagi S, Sawaki Y, Yamada T, Shinohara M (2023) man X, Romijn R, Van der Veen W (2019) Ambient noise multimode
Shallow low-velocity layer in the Hyuga-nada accretionary prism and Rayleigh and Love wave tomography to determine the shear velocity
its hydrological implications: insights from a passive seismic array. J structure above the Groningen gas field. Geophys J Int 218(3):1781–
Geophys Res. https://doi.org/10.1029/2022jb026298 1795. https://doi.org/10.1093/gji/ggz237
Anderson DL (1962) Love wave dispersion in heterogeneous anisotropic Cho I, Tada T, Shinozaki Y (2006) Centerless circular array method: inferring
media. Geophysics 27(4):445–454. https://doi.org/10.1190/1.1439042 phase velocities of Rayleigh waves in broad wavelength ranges using
Ardhuin F, Gualtieri L, Stutzmann E (2015) How ocean waves rock the earth: microtremor records. J Geophys Res. https://doi.org/10.1029/2005j
Two mechanisms explain microseisms with periods 3 to 300 s. Geophys b004235
Res Lett 42(3):765–772. https://doi.org/10.1002/2014GL062782 Claerbout JF (1968) Synthesis of a layered medium from its acoustic transmis-
Bahavar M, Spica ZJ, Sánchez-Sesma FJ, Trabant C, Zandieh A, Toro G (2020) sion response. Geophysics 33(2):264–269. https://doi.org/10.1190/1.
Horizontal-to-vertical spectral ratio (HVSR) IRIS station toolbox. Seismol 1439927
Res Lett 91(6):3539–3549. https://doi.org/10.1785/0220200047 Cotte N, Laske G (2002) Testing group velocity maps for Eurasia. Geophys J Int
Barmin MP (2023) RayTomo: seismic surface wave isotropic/azimuthally aniso- 150(3):639–650. https://doi.org/10.1046/j.1365-246X.2002.01719.x
tropic group/phase speed 2-D tomography. https://github.com/Noise Covellone BM, Savage B, Shen Y (2015) Seismic wave speed structure of the
CIEI/RayTomo. Accessed 9 Aug 2023 Ontong Java Plateau. Earth Planet Sci Lett 420:140–150. https://doi.org/
Barmin MP, Ritzwoller MH, Levshin AL (2001) A fast and reliable method for 10.1016/j.epsl.2015.03.033
surface wave tomography. Pure Appl Geophys 158:1351–1375. https:// Cox H (1973) Spatial correlation in arbitrary noise fields with application to
doi.org/10.1007/PL00001225 ambient sea noise. J Acoust Soc Am 54(5):1289–1301. https://doi.org/
Barmine M (2023) AFTAN: seismic surface wave dispersion measurements 10.1121/1.1914426
with automatic frequency time analysis. https://github.com/NoiseCIEI/ Crosbie KJ, Abers GA, Mann ME, Janiszewski HA, Creager KC, Ulberg CW, Moran
AFTAN. Accessed 12 July 2023 SC (2019) Shear velocity structure from ambient noise and teleseismic
Bensen GD, Ritzwoller MH, Barmin MP, Levshin AL, Lin F, Moschetti MP, Shapiro surface wave tomography in the cascades around mount St Helens. J
NM, Yang Y (2007) Processing seismic ambient noise data to obtain reli- Geophys Res 124(8):8358–8375. https://doi.org/10.1029/2019jb017836
able broad-band surface wave dispersion measurements. Geophys J Int Cupillard P, Stehly L, Romanowicz B (2011) The one-bit noise correlation:
169(3):1239–1260. https://doi.org/10.1111/j.1365-246X.2007.03374.x a theory based on the concepts of coherent and incoherent noise.
Bodin T, Sambridge M (2009) Seismic tomography with the reversible jump Geophys J Int 184(3):1397–1414. https://doi.org/10.1111/j.1365-246X.
algorithm. Geophys J Int 178(3):1411–1436. https://doi.org/10.1111/j. 2010.04923.x
1365-246X.2009.04226.x Dahlen FA, Tromp J (1998) Theoretical global seismology. Princeton University
Bodin T, Sambridge M, Tkalćić H, Arroucau P, Gallagher K, Rawlinson N (2012) Press, Princeton, p 1025
Transdimensional inversion of receiver functions and surface wave de Ridder S, Dellinger J (2011) Ambient seismic noise Eikonal tomography for
dispersion. J Geophys Res 117(2):1–24. https://doi.org/10.1029/2011J near-surface imaging at Valhall. Lead Edge 30(5):506–512. https://doi.
B008560 org/10.1190/1.3589108
Born M, Wolf E, Bhatia AB, Clemmow PC, Gabor D, Stokes AR, Taylor AM, de Vos D, Paulssen H, Fichtner A (2013) Finite-frequency sensitivity kernels for
Wayman PA, Wilcock WL, (1999) Principles of optics: electromagnetic two-station surface wave measurements. Geophys J Int 194(2):1042–
theory of propagation, interference and diffraction of light, 7th edn. 1049. https://doi.org/10.1093/gji/ggt144
Cambridge University Press, Cambridge, p 985. https://doi.org/10.1017/ Dou S, Lindsey N, Wagner AM, Daley TM, Freifeld B, Robertson M, Peterson
CBO9781139644181 J, Ulrich C, Martin ER, Ajo-Franklin JB (2017) Distributed acoustic
Boué P, Denolle M, Hirata N, Nakagawa S, Beroza GC (2016) Beyond basin sensing for seismic monitoring of the near surface: a traffic-noise
resonance: characterizing wave propagation using a dense array and interferometry case study. Sci Rep 7(1):11620. https://doi.org/10.1038/
the ambient seismic field. Geophys J Int 206(2):1261–1272. https://doi. s41598-017-11986-4
org/10.1093/gji/ggw205 Duvall TL, Jeffferies SM, Harvey JW, Pomerantz MA (1993) Time-distance
Bowden DC, Tsai VC, Lin FC (2015) Site amplification, attenuation, and scat- helioseismology. Nature 362(6419):430–432. https://doi.org/10.1038/
tering from noise correlation amplitudes across a dense array in long 362430a0
beach. CA Geophys Res Lett 42(5):1360–1367. https://doi.org/10.1002/ Dziewonski A, Anderson D (1981) Preliminary reference earth model. Phys
2014gl062662 Earth Planet Inter 25:297–356
Brenguier F, Campillo M, Hadziioannou C, Shapiro NM, Nadeau RM, Larose E Dziewonski AM, Hales AL (1972) Numerical analysis of dispersed seismic
(2008) Postseismic relaxation along the San Andreas fault at Parkfield waves. In: Bolt BA (ed) Methods in computational physics: advances
from continuous seismological observations. Science 321(5895):1478– in research and applications, vol 11. Elsevier, Amsterdam, pp 39–85.
1481. https://doi.org/10.1126/science.1160943 https://doi.org/10.1016/B978-0-12-460811-5.50007-6
Brenguier F, Shapiro NM, Campillo M, Ferrazzini V, Duputel Z, Coutant O, Ner- Eckart C (1953) The theory of noise in continuous media. J Acoust Soc Am
cessian A (2008) Towards forecasting volcanic eruptions using seismic 25(2):195–199. https://doi.org/10.1121/1.1907018
noise. Nat Geosci 1(2):126–130. https://doi.org/10.1038/ngeo104 Ekström G, Tromp J, Larson EWF (1997) Measurements and global models of
Brocher TM (2005) Empirical relations between elastic wavespeeds and den- surface wave propagation. J Geophys Res 19(17):15705–15710
sity in the earth’s crust. Bull Seismol Soc Am 95(6):2081–2092. https:// Ekström G, Ga Abers, Webb SC (2009) Determination of surface-wave phase
doi.org/10.1785/0120050077 velocities across USArray from noise and Aki’s spectral formulation.
Callen HB, Welton TA (1951) Irreversibility and generalized noise. Phys Rev Geophys Res Lett 36(18):5–9. https://doi.org/10.1029/2009GL039131
83(1):34–40. https://doi.org/10.1103/PhysRev.83.34 Ekström G, Nettles M, Dziewoński AM (2012) The global CMT project 2004–
Campillo M, Paul A (2003) Long-range correlations in the diffuse seismic coda. 2010: centroid-moment tensors for 13,017 earthquakes. Phys Earth
Science 299(5606):547–549. https://doi.org/10.1126/science.1078551 Planet Inter 200–201:1–9. https://doi.org/10.1016/j.pepi.2012.04.002
Chávez-García FJ, Luzón F (2005) On the correlation of seismic microtremors. J Emry EL, Shen Y, Nyblade AA, Flinders A, Bao X (2019) Upper mantle earth
Geophys Res 110(B11):1–12. https://doi.org/10.1029/2005JB003671 structure in Africa from full-wave ambient noise tomography. Geochem
Chen Y, de Ridder SAL, Rost S, Guo Z, Wu X, Chen Y (2022) Eikonal tomography Geophys Geosyst 20(1):120–147. https://doi.org/10.1029/2018gc007804
with physics-informed neural networks: Rayleigh wave phase velocity Fichtner A (2014) Source and processing effects on noise correlations. Geo-
in the northeastern margin of the Tibetan plateau. Geophys Res Lett. phys J Int 197(3):1527–1531. https://doi.org/10.1093/gji/ggu093
https://doi.org/10.1029/2022gl099053 Fichtner A (2015) Source-structure trade-offs in ambient noise correlations.
Chmiel M, Roux P, Herrmann P, Rondeleux B, Wathelet M (2018) Data-based Geophys J Int 202(1):678–694. https://doi.org/10.1093/gji/ggv182
diffraction kernels for surface waves from convolution and correlation
Fichtner A, Kennett BLN, Trampert J (2013) Separating intrinsic and apparent Haned A, Stutzmann E, Schimmel M, Kiselev S, Davaille A, Yelles-Chaouche
anisotropy. Phys Earth Planet Inter 219:11–20. https://doi.org/10.1016/j. A (2016) Global tomography using seismic hum. Geophys J Int
pepi.2013.03.006 204(2):1222–1236. https://doi.org/10.1093/gji/ggv516
Fichtner A, Stehly L, Ermert L, Boehm C (2016) Generalised interferometry—I. Haney MM, Mikesell TD, van Wijk K, Nakahara H (2012) Extension of the spatial
Theory for inter-station correlations. Geophys J Int. https://doi.org/10. autocorrelation (SPAC) method to mixed-component correlations of
1093/gji/ggw420 surface waves. Geophys J Int 191(1):189–206. https://doi.org/10.1111/j.
Froment B, Campillo M, Roux P, Gouédard P, Verdel A, Weaver RL (2010) 1365-246X.2012.05597.x
Estimation of the effect of nonisotropically distributed energy on the Hannemann K, Krüger F, Dahm T (2014) Measuring of clock drift rates and
apparent arrival time in correlations. Geophysics 75(5):85–93. https:// static time offsets of ocean bottom stations by means of ambient noise.
doi.org/10.1190/1.3483102 Geophys J Int 196(2):1034–1042
Fu L, Pan L, Li Z, Dong S, Ma Q, Chen X (2022) Improved high-resolution 3D vs Harmon N, Rychert CA (2015) Seismic imaging of deep crustal melt sills
model of long beach, CA: inversion of multimodal dispersion curves beneath Costa Rica suggests a method for the formation of the
from ambient noise of a dense array. Geophys Res Lett. https://doi.org/ Archean continental crust. Earth Planet Sci Lett 430:140–148. https://
10.1029/2021gl097619 doi.org/10.1016/j.epsl.2015.07.062
Fukao Y, Nishida K, Suda N, Nawa K, Kobayashi N (2002) A theory of the earth’s Harmon N, Forsyth D, Webb S (2007) Using ambient seismic noise to
background free oscillations. J Geophys Res 107(B9):2206 determine short-period phase velocities and shallow shear velocities
Fukao Y, Nishida K, Kobayashi N (2010) Seafloor topography, ocean infragrav- in young oceanic lithosphere. Bull Seismol Soc Am 97(6):2009–2023.
ity waves, and background Love and Rayleigh waves. J Geophys Res https://doi.org/10.1785/0120070050
115(B4):04302. https://doi.org/10.1029/2009JB006678 Harmon N, Rychert C, Gerstoft P (2010) Distribution of noise sources for seis-
Fukushima S, Shinohara M, Nishida K, Takeo A, Yamada T, Yomogida K (2022) mic interferometry. Geophys J Int 183(3):1470–1484. https://doi.org/10.
Detailed s-wave velocity structure of sediment and crust off Sanriku, 1111/j.1365-246X.2010.04802.x
Japan by a new analysis method for distributed acoustic sensing data Hasselmann K (1963) A statistical analysis of the generation of microseisms.
using a seafloor cable and seismic interferometry. Earth Planets Space Rev Geophys 1(2):177. https://doi.org/10.1029/RG001i002p00177
74(1):1–11. https://doi.org/10.1186/s40623-022-01652-z Herrmann RB (2013) Computer programs in seismology: an evolving tool for
Gabriels P, Snieder R, Nolet G (1987) In situ measurements of shear-wave instruction and research. Seismol Res Lett 84(6):1081–1088. https://doi.
velocity in sediments with higher-mode Rayleigh waves. Geophys Pros- org/10.1785/0220110096
pect 35(2):187–196. https://doi.org/10.1111/j.1365-2478.1987.tb00812.x Hirose T, Ueda H (2023) Continuous monitoring of instrumental clock errors at
Galetti E, Curtis A, Baptie B, Jenkins D, Nicolson H (2016) Transdimensional 50 volcanoes in Japan based on seismic interferometry. Earth Planets
Love-wave tomography of the British isles and shear-velocity structure Space 75(1):1–12. https://doi.org/10.1186/s40623-023-01798-4
of the east Irish sea basin from ambient-noise interferometry. Geophys Hirose T, Nakahara H, Nishimura T, Campillo M (2020) Locating spatial changes
J Int. https://doi.org/10.1093/gji/ggw286 of seismic scattering property by sparse modeling of seismic ambient
Gao H, Shen Y (2014) Upper mantle structure of the cascades from full-wave noise cross-correlation functions: application to the 2008 Iwate-Miyagi
ambient noise tomography: evidence for 3D mantle upwelling in the Nairiku (M w 6.9), Japan, earthquake. J Geophys Res. https://doi.org/10.
back-arc. Earth Planet Sci Lett 390:222–233. https://doi.org/10.1016/j. 1029/2019jb019307
epsl.2014.01.012 Hu S, Luo S, Yao H (2020) The frequency-Bessel spectrograms of multicompo-
Gao H, Shen Y (2015) A preliminary Full-Wave Ambient-Noise tomography nent cross-correlation functions from seismic ambient noise. J Geophys
model spanning from the Juan de Fuca and Gorda spreading centers to Res. https://doi.org/10.1029/2020jb019630
the Cascadia volcanic arc. Seismol Res Lett 86(5):1253–1260. https://doi. Huang H, Yao H, van der Hilst RD (2010) Radial anisotropy in the crust of SE
org/10.1785/0220150103 Tibet and SW China from ambient noise interferometry. Geophys Res
Gao H, Shen Y (2015) Validation of recent shear wave velocity models in the Lett 37(21):1–5. https://doi.org/10.1029/2010GL044981
United States with full-wave simulation. J Geophys Res 120(1):344–358. Ikeda T, Matsuoka T, Tsuji T, Hayashi K (2012) Multimode inversion with ampli-
https://doi.org/10.1002/2014jb011369 tude response of surface waves in the spatial autocorrelation method.
Gizon L, Birch AC (2002) Time–distance helioseismology: the forward problem Geophys J Int 190(1):541–552. https://doi.org/10.1111/j.1365-246X.
for random distributed sources. Astrophys J 571(2):966–986. https://doi. 2012.05496.x
org/10.1086/340015 IRIS DMC Data Services Products: EMC, A repository of Earth model. https://
Godin OA (2007) Emergence of the acoustic Green’s function from thermal doi.org/10.17611/DP/EMC.1
noise. J Acoust Soc Am 121(2):96–102. https://doi.org/10.1121/1.24307 Ishimaru A (1997) Wave propagation and scattering in random media. IEEE
64 Press, New York
Gouédard P, Stehly L, Brenguier F, Campillo M, Colin de Verdière Y, Larose E, Janiszewski HA, Gaherty JB, Abers GA, Gao H, Eilon ZC (2019) Amphibious
Margerin L, Roux P, Sánchez-Sesma FJ, Shapiro NM, Weaver RL (2008) surface-wave phase-velocity measurements of the Cascadia subduc-
Cross-correlation of random fields: mathematical approach and appli- tion zone. Geophys J Int 217(3):1929–1948. https://doi.org/10.1093/gji/
cations. Geophys Prospect 56(3):375–393. https://doi.org/10.1111/j. ggz051
1365-2478.2007.00684.x Jaxybulatov K, Shapiro NM, Koulakov I, Mordret A, Landes M, Sens-Schonfelder
Gouédard P, Seher T, McGuire JJ, Collins JA, van der Hilst RD (2014) Correction C (2014) A large magmatic sill complex beneath the Toba caldera. Sci-
of ocean-bottom seismometer instrumental clock errors using ambient ence 346(6209):617–619. https://doi.org/10.1126/science.1258582
seismic noise. Bull Seism Soc Am 104(3):1276–1288 Jiang C, Denolle MA (2022) Pronounced seismic anisotropy in Kanto sedimen-
Green PJ (1995) Reversible jump Markov chain Monte Carlo computation and tary basin: a case study of using dense arrays, ambient noise seismol-
Bayesian model determination. Biometrika 82(4):711–732. https://doi. ogy, and multi-modal surface-wave imaging. J Geophys Res. https://
org/10.1093/biomet/82.4.711 doi.org/10.1029/2022jb024613
Hamada K, Yoshizawa K (2015) Interstation phase speed and amplitude meas- Jiang C, Schmandt B, Farrell J, Lin F-C, Ward KM (2018) Seismically anisotropic
urements of surface waves with nonlinear waveform fitting: application magma reservoirs underlying silicic calderas. Geology 46(8):727–730.
to USArray. Geophys J Int 202(3):1463–1482. https://doi.org/10.1093/ https://doi.org/10.1130/G45104.1
gji/ggv213 Jiang C, Schmandt B, Abers GA, Kiser E, Miller MS (2023) Segmentation and
Hanasoge SM (2013) Measurements and kernels for source-structure inver- radial anisotropy of the deep crustal magmatic system beneath the
sions in noise tomography. Geophys J Int 196(2):971–985. https://doi. cascades arc. Geochem Geophys Geosyst. https://doi.org/10.1029/
org/10.1093/gji/ggt411 2022gc010738
Hanasoge S, Gizon L, Sreenivasan KR (2016) Seismic sounding of convection in Kawano Y, Isse T, Takeo A, Kawakatsu H, Morishige M, Shiobara H, Takeuchi
the sun. Annu Rev Fluid Mech 48(1):191–217. https://doi.org/10.1146/ N, Sugioka H, Kim Y, Utada H, Lee S-M (2023) Seismic structure of
annurev-fluid-122414-034534 the lithosphere-asthenosphere system beneath the oldest seafloor
revealed by Rayleigh-wave dispersion analysis. J Geophys Res. https:// Lin P-YP, Gaherty JB, Jin G, Collins JA, Lizarralde D, Evans RL, Hirth G (2016)
doi.org/10.1029/2023jb026529 High-resolution seismic constraints on flow dynamics in the oceanic
Kobayashi N, Nishida K (1998) Continuous excitation of planetary free oscilla- asthenosphere. Nature 535(7613):538–541. https://doi.org/10.1038/
tions by atmospheric disturbances. Nature 395(September):357–360. nature18012
https://doi.org/10.1038/26427 Liu X (2020) Finite-frequency sensitivity kernels for seismic noise interferom-
Krischer L, Megies T, Barsch R, Beyreuther M, Lecocq T, Caudron C, Wassermann etry based on differential time measurements. J Geophys Res. https://
J (2015) ObsPy: a bridge for seismology into the scientific python doi.org/10.1029/2019jb018932
ecosystem. Comput Sci Discov 8(1):014003. https://doi.org/10.1088/ Liu X, Ben-Zion Y (2013) Theoretical and numerical results on effects of attenu-
1749-4699/8/1/014003 ation on correlation functions of ambient seismic noise. Geophys J Int
Kubo R (1957) Statistical-Mechanical theory of irreversible processes. I. general 194(3):1966–1983. https://doi.org/10.1093/gji/ggt215
theory and simple applications to magnetic and conduction problems. Liu X, Beroza GC, Yang L, Ellsworth WL (2021) Ambient noise Love wave
J Phys Soc Jpn 12(6):570–586. https://doi.org/10.1143/JPSJ.12.570 attenuation tomography for the LASSIE array across the Los Angeles
Landisman M, Dziewonski A, Satô Y (1969) Recent improvements in the analy- basin. Sci Adv. https://doi.org/10.1126/sciadv.abe1030
sis of surface wave observations. Geophys J Int 17(4):369–403. https:// Liu Q, Chen X, Gao L, Yu Z, Chen J (2023) Direct image dissimilarity inversion of
doi.org/10.1111/j.1365-246X.1969.tb00246.x ambient noise multimodal dispersion spectrograms. Bull Seismol Soc
Laske G, Widmer-Schnidrig R (2015) 1.04—theory and observations: normal Am. https://doi.org/10.1785/0120230012
mode and surface wave observations. In: Schubert G (ed) Treatise on Lobkis OI, Weaver RL (2001) On the emergence of the green’s function in the
geophysics, 2nd edn. Elsevier, Oxford, pp 117–167. https://doi.org/10. correlations of a diffuse field. J Acoust Soc Am 110(6):3011. https://doi.
1016/B978-0-444-53802-4.00003-8 org/10.1121/1.1417528
Lehujeur M, Chevrot S (2020) Eikonal tomography using coherent surface Longuet-Higgins MS (1950) A theory of the origin of microseisms. Philos Trans
waves extracted from ambient noise by iterative matched filtering— R Soc Lond 243(857):1–35. https://doi.org/10.1098/rsta.1950.0012
application to the large-N Maupasacq array. J Geophys Res. https://doi. Lu, Y, Stehly, L, Brossier, R, Paul, A, AlpArray Working Group (2020) Imaging
org/10.1029/2020jb019363 alpine crust using ambient noise wave-equation tomography. Geophys
Levshin AL, Barmin MP, Ritzwoller MH (2018) Tutorial review of seismic surface J Int 222(1):69–85. https://doi.org/10.1093/gji/ggaa145
waves’ phenomenology. J Seismol 22(2):519–537. https://doi.org/10. Lynner C, Beck SL, Zandt G, Porritt RW, Lin FC, Eilon ZC (2018) Midcrustal
1007/s10950-017-9716-7 deformation in the central Andes constrained by radial anisotropy. J
Levshin, A SensKernel: 1-D seismic surface wave isotropicgroup/phase speed Geophys Res 123(6):4798–4813. https://doi.org/10.1029/2017JB014936
dispersion/eigenfunction/sensitivity kernel. https://github.com/Noise Magrini F, Boschi L (2021) Surface-wave attenuation from seismic ambient
CIEI/SensKernel noise: numerical validation and application. J Geophys Res. https://doi.
Levshin A, Ratnikova L, Berger J (1992) Peculiarities of surface-wave propaga- org/10.1029/2020jb019865
tion across central Eurasia. Bull Seismol Soc Am 82(6):2464–2493. Maraschini M, Foti S (2010) A Monte Carlo multimodal inversion of surface
https://doi.org/10.1785/BSSA0820062464 waves. Geophys J Int 182(3):1557–1566. https://doi.org/10.1111/j.1365-
Li J, Chen Y, Schuster GT (2020) Separation of multi-mode surface waves by 246X.2010.04703.x
supervised machine learning methods. Geophys Prospect 68(4):1270– Menke W, Jin G (2015) Waveform fitting of cross spectra to determine phase
1280. https://doi.org/10.1111/1365-2478.12927 velocity using Aki’s formula. Bull Seismol Soc Am 105(3):1619–1627.
Liang C, Langston C (2008) Ambient seismic noise tomography and structure https://doi.org/10.1785/0120140245
of eastern North America. J Geophys Res 113(B3):1–18. https://doi.org/ Miller D, Bennington N, Haney M, Bedrosian P, Key K, Thurber C, Hart L, Ohlen-
10.1029/2007JB005350 dorf S (2020) Linking magma storage and ascent to eruption volume
Lin F-C, Ritzwoller MH (2011) Apparent anisotropy in inhomogeneous and composition at an arc caldera. Geophys Res Lett. https://doi.org/10.
isotropic media. Geophys J Int 186(3):1205–1219. https://doi.org/10. 1029/2020gl088122
1111/j.1365-246X.2011.05100.x Montagner J-P (2015) Deep earth structure—upper mantle structure: global
Lin FC, Ritzwoller MH (2011) Helmholtz surface wave tomography for isotropic isotropic and anisotropic elastic tomography. In: Schubert G (ed) Trea-
and azimuthally anisotropic structure. Geophys J Int 186(3):1104–1120. tise on geophysics, vol 1, 2nd edn. Elsevier, Amsterdam, pp 613–639.
https://doi.org/10.1111/j.1365-246X.2011.05070.x https://doi.org/10.1016/B978-0-444-53802-4.00016-6
Lin F-C, Schmandt B (2014) Upper crustal azimuthal anisotropy across the Mordret A, Shapiro NM, Singh SS, Roux P, Barkved OI (2013) Helmholtz
contiguous U.S. determined by Rayleigh wave ellipticity. Geophys Res tomography of ambient noise surface wave data to estimate Scholte
Lett 41(23):8301–8307. https://doi.org/10.1002/2014gl062362 wave phase velocity at Valhall life of the field. Geophysics 78(2):99–109.
Lin F-C, Moschetti MP, Ritzwoller MH (2008) Surface wave tomography of the https://doi.org/10.1190/geo2012-0303.1
western United States from ambient seismic noise: Rayleigh and Love Moschetti MP, Ritzwoller MH, Shapiro NM (2007) Surface wave tomography of
wave phase velocity maps. Geophys J Int 173(1):281–298. https://doi. the western United States from ambient seismic noise: Rayleigh wave
org/10.1111/j.1365-246X.2008.03720.x group velocity maps. Geochem Geophys Geosyst. https://doi.org/10.
Lin F-C, Ritzwoller MH, Snieder R (2009) Eikonal tomography: surface wave 1029/2007gc001655
tomography by phase front tracking across a regional broad-band Moschetti MP, Ritzwoller MH, Lin F-C, Yang Y (2010) Crustal shear wave velocity
seismic array. Geophys J Int 177(3):1091–1110. https://doi.org/10.1111/j. structure of the western United States inferred from ambient seismic
1365-246X.2009.04105.x noise and earthquake data. J Geophys Res. https://doi.org/10.1029/
Lin F-C, Ritzwoller MH, Shen W (2011) On the reliability of attenuation meas- 2010jb007448
urements from ambient noise cross-correlations. Geophys Res Lett. Moschetti MP, Ritzwoller MH, Lin F, Yang Y (2010) Seismic evidence for wide-
https://doi.org/10.1029/2011gl047366 spread western-US deep-crustal deformation caused by extension.
Lin F-C, Schmandt B, Tsai VC (2012) Joint inversion of Rayleigh wave phase Nature 464(7290):885–889. https://doi.org/10.1038/nature08951
velocity and ellipticity using USArray: constraining velocity and density Nagaoka Y, Nishida K, Aoki Y, Takeo M, Ohminato T (2012) Seismic imaging
structure in the upper crust. Geophys Res Lett 39:1–7. https://doi.org/ of magma chamber beneath an active volcano. Earth Planet Sci Lett
10.1029/2012GL052196 333–334:1–8. https://doi.org/10.1016/j.epsl.2012.03.034
Lin F-C, Tsai VC, Ritzwoller MH (2012) The local amplification of surface waves: Nakahara H (2006) A systematic study of theoretical relations between spatial
a new observable to constrain elastic velocities, density, and anelastic correlation and Green’s function in one-, two- and three-dimensional
attenuation. J Geophys Res 117:B06302. https://doi.org/10.1029/2012J random scalar wavefields. Geophys J Int 165(3):1097–1105. https://doi.
B009208 org/10.1111/j.1365-246X.2006.03170.x
Lin F-C, Li D, Clayton RW, Hollis D (2013) High-resolution 3D shallow crustal Nakahara H, Emoto K, Nishimura T (2021) Extending the formulation of the
structure in long beach, California: application of ambient noise tomog- spatial autocorrelation (SPAC) method to strain, rotation and tilt. Geo-
raphy on a dense seismic array. Geophysics 78(4):45–56. https://doi.org/ phys J Int 227(1):287–302. https://doi.org/10.1093/gji/ggab217
10.1190/geo2012-0453.1
Nakamura Y (1989) A method for dynamic characteristics estimation of sub- Prieto GA, Denolle M, Lawrence JF, Beroza GC (2011) On amplitude informa-
surface using microtremor on the ground surface. Q Rep RTRI 30(1) tion carried by the ambient seismic field. C R Geosci 343(8–9):600–614.
Nakanishi I, Anderson DL (1982) Worldwide distribution of group velocity of https://doi.org/10.1016/j.crte.2011.03.006
mantle Rayleigh waves as determined by spherical harmonic inversion. Qin T, Lu L, Ding Z, Feng X, Zhang Y (2022) High-resolution 3D shallow S wave
Bull Seismol Soc Am 72(4):1185–1194. https://doi.org/10.1785/BSSA0 velocity structure of Tongzhou, subcenter of Beijing, inferred from mul-
720041185 timode Rayleigh waves by beamforming seismic noise at a dense array.
Nakata N (2020) Pseudo arrivals generated by frequency normalization for J Geophys Res. https://doi.org/10.1029/2021jb023689
seismic interferometry with scattered waves—stationary-phase analy- Rawlinson N (2023) Surface wave tomography code. http://rses.anu.edu.au/
sis. SEG Tech Program Expanded Abstr 2020:2085–2089. https://doi.org/ ~nick/surftomo.html. Accessed 9 Aug 2023
10.1190/segam2020-3427903.1 Rawlinson N, Sambridge M (2004) Wave front evolution in strongly heteroge-
Nakata N, Snieder R, Kuroda S, Ito S, Aizawa T, Kunimi T (2013) Monitoring neous layered media using the fast marching method. Geophys J Int
a building using deconvolution interferometry. I: earthquake-data 156(3):631–647. https://doi.org/10.1111/j.1365-246X.2004.02153.x
analysis. Bull Seismol Soc Am 103(3):1662–1678. https://doi.org/10. Rawlinson N, Pozgay S, Fishwick S (2010) Seismic tomography: a window into
1785/0120120291 deep earth. Phys Earth Planet Inter 178(3–4):101–135. https://doi.org/
Nakata N, Gualtieri L, Fichtner A (2019) Seismic ambient noise. Cambridge 10.1016/j.pepi.2009.10.002
University Press, Cambridge Ritzwoller MH (2002) Global surface wave diffraction tomography. J Geophys
Nayak A, Thurber CH (2020) Using multicomponent ambient seismic noise Res 107(B12):1–13. https://doi.org/10.1029/2002JB001777
cross-correlations to identify higher mode Rayleigh waves and improve Ritzwoller M, Feng L (2019) Overview of pre- and post-processing of
dispersion measurements. Geophys J Int 222(3):1590–1605. https://doi. ambient-noise correlations. In: Nakata N, Gualtieri L, Fichtner A (eds)
org/10.1093/gji/ggaa270 Seismic ambient noise. Cambridge University Press, Cambridge, pp
Nimiya H, Ikeda T, Tsuji T (2020) Three-dimensional S wave velocity structure of 144–187. https://doi.org/10.1017/9781108264808.007
central Japan estimated by surface-wave tomography using ambient Ritzwoller MH, Lin F-CC, Shen W (2011) Ambient noise tomography with a
noise. J Geophys Res 125(4):2019–019043. https://doi.org/10.1029/ large seismic array. C R Geosci 343(8–9):558–570. https://doi.org/10.
2019jb019043 1016/j.crte.2011.03.007
Nimiya H, Ikeda T, Tsuji T (2023) Multimodal Rayleigh and Love wave joint Romanowicz B (2003) Global mantle tomography: progress status in the
inversion for s-wave velocity structures in Kanto basin, Japan. J Geophys past 10 years. Annu Rev Earth Planet Sci 31(1):303–328. https://doi.
Res. https://doi.org/10.1029/2022jb025017 org/10.1146/annurev.earth.31.091602.113555
Nishida K (2011) Two-dimensional sensitivity kernels for cross-correlation Romanowicz B (2020) Surface waves. In: Gupta HK (ed) Encyclopedia of
functions of background surface waves. C R Geosci 343(8–9):584–590. solid earth geophysics. Springer, Cham, pp 1–13. https://doi.org/10.
https://doi.org/10.1016/j.crte.2011.02.004 1007/978-3-030-10475-7_143-1
Nishida K (2013) Earth’s background free oscillations. Annu Rev Earth Planet Sci Romanowicz B (2021) Seismic tomography of the earth’s mantle. In: Alder-
41(1):719–740. https://doi.org/10.1146/annurev-earth-050212-124020 ton D, Elias SA (eds) Encyclopedia of geology, 2nd edn. Academic
Nishida K (2014) Source spectra of seismic hum. Geophys J Int 199(1):416–429. Press, Oxford, pp 587–609. https://doi.org/10.1016/B978-0-08-
https://doi.org/10.1093/gji/ggu272 102908-4.00169-7
Nishida K (2017) Ambient seismic wave field. Proc Jpn Acad Ser B Phys Biol Sci Rost S, Thomas C (2002) Array seismology: methods and applications. Rev
93(7):423–448. https://doi.org/10.2183/pjab.93.026 Geophys. https://doi.org/10.1029/2000RG000100
Nishida K, Kobayashi N (1999) Statistical features of earth’s continuous free Russell JB, Gaherty JB (2021) Lithosphere structure and seismic anisotropy
oscillations. J Geophys Res 104(B12):28741–28 offshore eastern North America: implications for continental breakup
Nishida K, Takagi R (2022) A global centroid single force catalog of p-wave and ultra-slow spreading dynamics. J Geophys Res. https://doi.org/
microseisms. J Geophys Res. https://doi.org/10.1029/2021jb023484 10.1029/2021jb022955
Nishida K, Kobayashi N, Fukao Y (2002) Origin of earth’s ground noise from 2 to Sager K, Ermert L, Boehm C, Fichtner A (2017) Towards full waveform ambi-
20 mHz. Geophys Res Lett 29(10):1413 ent noise inversion. Geophys J Int 212(1):566–590. https://doi.org/10.
Nishida K, Kawakatsu H, Obara K (2008) Three-dimensional crustal S wave 1093/gji/ggx429
velocity structure in japan using microseismic data recorded by hi-net Sager K, Boehm C, Ermert L, Krischer L, Fichtner A (2020) Global-scale full-
tiltmeters. J Geophys Res 113(B10):10302. https://doi.org/10.1029/2007J waveform ambient noise inversion. J Geophys Res. https://doi.org/10.
B005395 1029/2019jb018644
Nishida K, Kawakatsu H, Fukao Y, Obara K (2008) Background Love and Saito M (1988) DISPER80; a subroutine package for calculation of seismic
Rayleigh waves simultaneously generated at the pacific ocean floors. normal-mode solutions. Seismological algorithms; computational
Geophys Res Lett 35(16):16307. https://doi.org/10.1029/2008gl034753 methods and computer programs. Academic Press, San Diego, pp
Nishida K, Montagner J-P, Kawakatsu H (2009) Global surface wave tomogra- 293–319
phy using seismic hum. Science 326(5949):112. https://doi.org/10.1126/ Sánchez-Sesma FJ, Rodríguez M, Iturrarán-Viveros U, Luzón F, Campillo M,
science.1176389 Margerin L, García-Jerez A, Suarez M, Santoyo MA, Rodríguez-Castel-
Nolet G (2008) A breviary of seismic tomography: imaging the interior of the lanos A (2011) A theory for microtremor H/V spectral ratio: applica-
earth and sun. Cambridge University Press, Cambridge. https://doi.org/ tion for a layered medium. Geophys J Int 186(1):221–225. https://doi.
10.1017/CBO9780511984709 org/10.1111/j.1365-246X.2011.05064.x
Obermann A, Hillers G (2019) Chapter two—seismic time-lapse interferometry Sato H, Fehler MC, Maeda T (2012) Seismic wave propagation and scattering
across scales. In: Schmelzbach C (ed) Advances in geophysics, vol 60. Else- in the heterogeneous earth. Springer, Heidelberg, p 496
vier, Amsterdam, pp 65–143. https://doi.org/10.1016/bs.agph.2019.06.001 Sato H (2019) Isotropic scattering coefficient of the solid earth. Geophys J
Okada H, Sakajiri N (1983) Estimates of an s-wave velocity distribution using Int 218(3):2079–2088. https://doi.org/10.1093/gji/ggz266
long-period microtremors. Geophys Bull Hokkaido Univ 42:119–143 Savage MK, Lin F-C, Townend J (2013) Ambient noise cross-correlation
Oppenheim AV, Schafer RW (2014) Discrete-time signal processing. Pearson, Essex observations of fundamental and higher-mode Rayleigh wave
Pedersen HA, Mattern F, Poli P, Stehly L (2023) Imaging with seismic noise: propagation governed by basement resonance. Geophys Res Lett
improving extraction of body wave phases from the deep earth 40(14):3556–3561. https://doi.org/10.1002/grl.50678
through selective stacking based on H/V ratios. Geophys J Int. https:// Saygin E, Kennett BLN (2012) Crustal structure of Australia from ambient
doi.org/10.1093/gji/ggac388 seismic noise tomography. J Geophys Res 117(1):1–15. https://doi.
Peterson JR (1993) Observations and modeling of seismic background noise. org/10.1029/2011JB008403
U.S. Geol Surv Open File Rep 93–322:1–94. https://doi.org/10.3133/ Saygin E, Kennett BLN (2010) Ambient seismic noise tomography of Austral-
ofr93322 ian continent. Tectonophysics 481(1):116–125. https://doi.org/10.
Prieto GA, Lawrence JF, Beroza GC (2009) Anelastic earth structure from the 1016/j.tecto.2008.11.013
coherency of the ambient seismic field. J Geophys Res 114(B7):1–15. Schuster GT (2009) Seismic interferometry. Cambridge University Press,
https://doi.org/10.1029/2008JB006067 Cambridge. https://doi.org/10.1017/cbo9780511581557
Sens-Schönfelder C (2008) Synchronizing seismic networks with ambient Takagi R, Nishida K, Maeda T, Obara K (2018) Ambient seismic noise wavefield
noise. Geophys J Int 174(3):966–970. https://doi.org/10.1111/j.1365- in japan characterized by polarization analysis of hi-net records. Geo-
246X.2008.03842.x phys J Int 215(3):1682–1699. https://doi.org/10.1093/gji/ggy334
Sens-Schönfelder C, Wegler U (2006) Passive image interferometry and Takagi R, Uchida N, Nakayama T, Azuma R, Ishigami A, Okada T, Nakamura
seasonal variations of seismic velocities at Merapi volcano, Indonesia. T, Shiomi K (2019) Estimation of the orientations of the s-net cabled
Geophys Res Lett 33(21):21302. https://doi.org/10.1029/2006GL0277 Ocean-Bottom sensors. Seismol Res Lett 90(6):2175–2187. https://doi.
97 org/10.1785/0220190093
Sens-Schönfelder C, Wegler U (2006) Radiative transfer theory for estimation Takagi R, Toyokuni G, Chikasada N (2021) Ambient noise correlation analysis of
of the seismic moment. Geophys J Int 167(3):1363–1372. https://doi. s-net records: extracting surface wave signals below instrument noise
org/10.1111/j.1365-246X.2006.03139.x levels. Geophys J Int 224(3):1640–1657. https://doi.org/10.1093/gji/
Sens-Schönfelder C, Wegler U (2011) Passive image interferometry for ggaa548
monitoring crustal changes with ambient seismic noise. C R Geosci Takeo A, Nishida K, Isse T, Kawakatsu H, Shiobara H, Sugioka H, Kanazawa T
343(8–9):639–651. https://doi.org/10.1016/j.crte.2011.02.005 (2013) Radially anisotropic structure beneath the Shikoku basin from
Sethian JA (1996) A fast marching level set method for monotonically broadband surface wave analysis of ocean bottom seismometer
advancing fronts. Proc Natl Acad Sci U S A 93(4):1591–1595. https:// records. J Geophys Res 118(6):2878–2892. https://doi.org/10.1002/jgrb.
doi.org/10.1073/pnas.93.4.1591 50219
Shapiro NM, Campillo M (2004) Emergence of broadband Rayleigh waves from Takeo A, Kawakatsu H, Isse T, Nishida K, Sugioka H, Ito A, Shiobara H,
correlations of the ambient seismic noise. Geophys Res Lett 31(7):1615– Suetsugu D (2016) Seismic azimuthal anisotropy in the oceanic litho-
1619. https://doi.org/10.1029/2004GL019491 sphere and asthenosphere from broadband surface wave analysis of
Shapiro NM, Singh SK (1999) A systematic error in estimating surface-wave OBS array records at 60 ma seafloor. J Geophys Res 121(3):1927–1947.
group-velocity dispersion curves and a procedure for its correction. https://doi.org/10.1002/2015JB012429
Bull Seismol Soc Am 89(4):1138–1142. https://doi.org/10.1785/BSSA0 Takeo A, Forsyth DW, Weeraratne DS, Nishida K (2014) Estimation of azi-
890041138 muthal anisotropy in the NW pacific from seismic ambient noise in
Shapiro NMNM, Ritzwoller MHH (2002) Monte-Carlo inversion for a global seafloor records. Geophys J Int 199(1):11–22. https://doi.org/10.1093/
shear-velocity model of the crust and upper mantle. Geophys J Int gji/ggu240
151(1):88–105. https://doi.org/10.1046/j.1365-246X.2002.01742.x Takeo A, Kawakatsu H, Isse T, Nishida K, Shiobara H, Sugioka H, Ito A, Utada
Shapiro NM, Campillo M, Stehly L, Ritzwoller MH (2005) High-resolution H (2018) In situ characterization of the Lithosphere–Asthenosphere
surface-wave tomography from ambient seismic noise. Science system beneath NW pacific ocean via broadband dispersion survey
307(5715):1615–1618. https://doi.org/10.1126/science.1108339 with two OBS arrays. Geochem Geophys Geosyst. https://doi.org/10.
Snieder R (2004) Extracting the Green’s function from the correlation of coda 1029/2018GC007588
waves: a derivation based on stationary phase. Phys Rev E 69(4):046610. Takeo A, Nishida K, Aoyama H, Ishise M, Kai T, Kurihara R, Maeda T, Mizu-
https://doi.org/10.1103/PhysRevE.69.046610 tani Y, Nakashima Y, Nagahara S, Wang X, Ye L, Akuhara T, Aoki Y
Snieder R (2006) Retrieving the Green’s function of the diffusion equation from (2022) S-wave modelling of the Showa-Shinzan lava dome in Usu
the response to a random forcing. Phys Rev E 74(4):046620. https://doi. Volcano, northern Japan, from seismic observations. Geophys J Int
org/10.1103/PhysRevE.74.046620 230(3):1662–1678. https://doi.org/10.1093/gji/ggac111
Snieder R, Larose E (2013) Extracting earth’s elastic wave response from noise Tanimoto T (1990) Modelling curved surface wave paths: membrane surface
measurements. Annu Rev Earth Planet Sci 41(1):183–206. https://doi. wave synthetics. Geophys J Int 102:89–100. https://doi.org/10.1111/j.
org/10.1146/annurev-earth-050212-123936 1365-246X.1990.tb00532.x
Snieder R, Wapenaar K (2010) Imaging with ambient noise. Phys Today Tanimoto T, Anderson DL (1985) Lateral heterogeneity and azimuthal ani-
63(9):44–49. https://doi.org/10.1063/1.3490500 sotropy of the upper mantle: Love and Rayleigh waves 100–250 s. J
Snieder R, Sheiman J, Calvert R (2006) Equivalence of the virtual-source Geophys Res 90(B2):1842. https://doi.org/10.1029/jb090ib02p01842
method and wave-field deconvolution in seismic interferometry. Phys Tanimoto T, Um J (1999) Cause of continuous oscillations of the earth. J
Rev E 73(6 Pt 2):066620. https://doi.org/10.1103/PhysRevE.73.066620 Geophys Res 104(B12):28723–28739. https://doi.org/10.1029/1999J
Socco LV, Foti S, Boiero D (2010) Surface-wave analysis for building near-sur- B900252
face velocity models—established approaches and new perspectives. Tarantola A (2005) Inverse problem theory and methods for model parameter
Geophysics 75(5):75–8375102. https://doi.org/10.1190/1.3479491 estimation. Other titles in applied mathematics. Society for Industrial
Spetzler J, Trampert J, Snieder R (2002) The effect of scattering in surface wave and Applied Mathematics, Philadelphia. https://doi.org/10.1137/1.
tomography. Geophys J Int 149(3):755–767 9780898717921
Spica Z, Perton M, Legrand D (2017) Anatomy of the Colima volcano mag- Tarantola A, Valette B (1982) Generalized nonlinear inverse problems solved
matic system, Mexico. Earth Planet Sci Lett 459:1–13. https://doi.org/10. using the least squares criterion. Rev Geophys 20(2):219. https://doi.
1016/j.epsl.2016.11.010 org/10.1029/RG020i002p00219
Spica Z, Perton M, Nakata N, Liu X, Beroza GC (2018) Shallow VS imaging of Tauzin B, Pham T-S, Tkalčić H (2019) Receiver functions from seismic interfer-
the Groningen area from joint inversion of multimode surface waves ometry: a practical guide. Geophys J Int 217(1):1–24. https://doi.org/10.
and H/V spectral ratios. Seismol Res Lett. https://doi.org/10.1785/02201 1093/gji/ggz002
80060 Thurber C, Ritsema J (2015) Theory and observations-seismic tomography and
Taira T, Yoshizawa K (2020) Upper-mantle discontinuities beneath Australia inverse methods. Elsevier, Amsterdam, pp 307–337. https://doi.org/10.
from transdimensional Bayesian inversions using multimode surface 1016/B978-0-444-53802-4.00009-9
waves and receiver functions. Geophys J Int 223(3):2085–2100. https:// Tian D, Uieda L, Leong WJ, Schlitzer W, FroÅNhlich Y, Grund M, Jones M, Toney
doi.org/10.1093/gji/ggaa442 L, Yao J, Magen Y, Tong 1924 J-H, Materna K, Belem A, Newton T, Anant
Takagi R, Nishida K (2022) Multi-mode dispersion measurement of surface A, Ziebarth M, Quinn J, Wessel P (2023) PyGMT: A Python 1925 Interface
waves extracted by multi-component ambient noise cross-correlation for the Generic Mapping Tools. https://doi.org/10.5281/zenodo.83031
functions. Geophys J Int. https://doi.org/10.1093/gji/ggac225 86
Takagi R, Nishida K, Aoki Y, Maeda T, Masuda K, Takeo M, Obara K, Shiomi K, Tokimatsu K, Tamura S, Kojima H (1992) Effects of multiple modes on Rayleigh
Sato M, Saito K (2015) A single bit matters: coherent noise of seismic wave dispersion characteristics. J Geotech Eng 118(10):1529–1543.
data loggers. Seismol Res Lett 86(3):901–907. https://doi.org/10.1785/ https://doi.org/10.1061/(asce)0733-9410(1992)118:10(1529)
0220150030 Tomar G, Shapiro NM, Mordret A, Singh SC, Montagner J-P (2017) Radial
Takagi R, Nakahara H, Kono T, Okada T (2014) Separating body and Rayleigh anisotropy in Valhall: ambient noise-based studies of Scholte and Love
waves with cross terms of the cross-correlation tensor of ambient waves. Geophys J Int 208(3):1524–1539. https://doi.org/10.1093/gji/
noise. J Geophys Res 119(3):2005–2018. https://doi.org/10.1002/2013J ggw480
B010824
Tomar G, Stutzmann E, Mordret A, Montagner J-P, Singh SC, Shapiro NM (2018) Wapenaar K, Slob E, Snieder R, Curtis A (2010) Tutorial on seismic interfer-
Joint inversion of the first overtone and fundamental mode for deep ometry: part 2-underlying theory and new advances. Geophysics
imaging at the Valhall oil field using ambient noise. Geophys J Int 75(5):75–211
214(1):122–132. https://doi.org/10.1093/gji/ggy122 Weaver RL (2011) On the amplitudes of correlations and the inference of
Trabant C, Hutko AR, Bahavar M, Karstens R, Ahern T, Aster R (2012) Data attenuations, specific intensities and site factors from ambient noise. C
products at the IRIS DMC: stepping stones for research and other R Geosci 343(8):615–622. https://doi.org/10.1016/j.crte.2011.07.001
applications. Seismol Res Lett 83(5):846–854. https://doi.org/10.1785/ Weaver R, Froment B, Campillo M (2009) On the correlation of non-isotropically
0220120032 distributed ballistic scalar diffuse waves. J Acoust Soc Am 126(4):1817.
Tribaldos RV, Ajo-Franklin JB (2021) Aquifer monitoring using ambient seismic https://doi.org/10.1121/1.3203359
noise recorded with distributed acoustic sensing (DAS) deployed on Webb SC (1998) Broadband seismology and noise under the ocean. Rev
dark fiber. J Geophys Res. https://doi.org/10.1029/2020jb021004 Geophys 36(1):105–142
Tromp J, Dahlen FA (1992) Variational principles for surface wave propagation Wegler U, Sens-Schönfelder C, Sens-Schönfelder C (2007) Fault zone monitor-
on a laterally heterogeneous Earth—II. Frequency-domain JWKB theory. ing with passive image interferometry. Geophys J Int 168(3):1029–1033.
Geophys J Int 109(3):599–619. https://doi.org/10.1111/j.1365-246X. https://doi.org/10.1111/j.1365-246X.2006.03284.x
1992.tb00120.x Wessel P, Luis JF, Uieda L, Scharroo R, Wobbe F, Smith WHF, Tian D (2019)
Tromp J, Fa Dahlen (1992) Variational principles for surface wave propaga- The Generic Mapping Tools Version 6. Geochem Geophys Geosyst
tion on a laterally heterogeneous Earth-I. Time-domain JWKB theory. 20(11):5556–64. https://doi.org/10.1029/2019gc008515
Geophys J Int 109(3):581–598. https://doi.org/10.1111/j.1365-246X. White MCA, Fang H, Nakata N, Ben-Zion Y (2020) PyKonal: a python package
1992.tb00119.x for solving the Eikonal equation in spherical and cartesian coordinates
Tromp J, Fa Dahlen (1993) Variational principles for surface wave propaga- using the fast marching method. Seismol Res Lett. https://doi.org/10.
tion on a laterally heterogeneous Earth-III. Potential representation. 1785/0220190318
Geophys J Int 112(2):195–209. https://doi.org/10.1111/j.1365-246X. Wiener N (1947) Cybernetics: or control and communication in the animal and
1993.tb01449.x the machine. MIT Press, Cambridge
Tromp J, Luo Y, Hanasoge S, Peter D (2010) Noise cross-correlation sensitivity Williams EF, Fernández-Ruiz MR, Magalhaes R, Vanthillo R, Zhan Z, González-
kernels. Geophys J Int 183(2):791–819. https://doi.org/10.1111/j.1365- Herráez M, Martins HF (2021) Scholte wave inversion and passive
246X.2010.04721.x source imaging with ocean-bottom DAS. Lead Edge 40(8):576–583.
Tsai VC (2011) Understanding the amplitudes of noise correlation measure- https://doi.org/10.1190/tle40080576.1
ments. J Geophys Res 116(B9):09311. https://doi.org/10.1029/2011J Woodhouse JH, Dziewonski AM (1984) Mapping the upper mantle: Three-
B008483 dimensional modeling of earth structure by inversion of seismic
Tsai VC, Moschetti MP (2010) An explicit relationship between time-domain waveforms. J Geophys Res 89(B7):5953–5986. https://doi.org/10.1029/
noise correlation and spatial autocorrelation (SPAC) results. Geophys J jb089ib07p05953
Int. https://doi.org/10.1111/j.1365-246X.2010.04633.x Yamaya L, Mochizuki K, Akuhara T, Nishida K (2021) Sedimentary structure
Ueno T, Saito T, Shiomi K, Haryu Y (2015) Monitoring the instrument response derived from multi-mode ambient noise tomography with dense OBS
of the high-sensitivity seismograph network in Japan (Hi-net): effects network at the Japan trench. J Geophys Res. https://doi.org/10.1029/
of response changes on seismic interferometry analysis. Earth Planets 2021jb021789
Space 67(1):135. https://doi.org/10.1186/s40623-015-0305-0 Yang X, Gao H (2018) Full-wave seismic tomography in the northeastern
van der Walt S, Colbert SC, Varoquaux G (2011) The NumPy array: a structure United States: new insights into the uplift mechanism of the Adiron-
for efficient numerical computation. Comput Sci Eng 13(2):22–30. dack mountains. Geophys Res Lett 45(12):5992–6000. https://doi.org/
https://doi.org/10.1109/mcse.2011.37 10.1029/2018gl078438
Viens L, Perton M, Spica ZJ, Nishida K, Yamada T, Shinohara M (2022) Under- Yang Y, Ritzwoller MH, Levshin AL, Shapiro NM (2007) Ambient noise Rayleigh
standing surface wave modal content for high-resolution imaging of wave tomography across Europe. Geophys J Int 168(1):259–274. https://
submarine sediments with distributed acoustic sensing. Geophys J Int doi.org/10.1111/j.1365-246X.2006.03203.x
232(3):1668–1683. https://doi.org/10.1093/gji/ggac420 Yang Y, Ritzwoller MH, Lin F-C, Moschetti MP, Shapiro NM (2008) Structure of
Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T, Cournapeau D, the crust and uppermost mantle beneath the western United States
Burovski E, Peterson P, Weckesser W, Bright J, van der Walt SJ, Brett M, revealed by ambient noise and earthquake tomography. J Geophys
Wilson J, Millman KJ, Mayorov N, Nelson ARJ, Jones E, Kern R, Larson E, Res. https://doi.org/10.1029/2008jb005833
Carey CJ, Polat I, Feng Y, Moore EW, VanderPlas J, Laxalde D, Perktold J, Yang Y, Langston CA, Powell CA, Thomas WA (2022) Full waveform ambient
Cimrman R, Henriksen I, Quintero EA, Harris CR, Archibald AM, Ribeiro noise tomography for the northern Mississippi embayment. J Geophys
AH, Pedregosa F, van Mulbregt P, SciPy 1.0 Contributors (2020) SciPy 1.0: Res. https://doi.org/10.1029/2021jb022267
fundamental algorithms for scientific computing in Python. Nat Meth- Yao H, van der Hilst RD, Montagner J-P (2010) Heterogeneity and anisotropy
ods 17(3):261–272. https://doi.org/10.1038/s41592-019-0686-2 of the lithosphere of SE Tibet from surface wave array tomography. J
Wang J, Wu G, Chen X (2019) Frequency-Bessel transform method for effective Geophys Res. https://doi.org/10.1029/2009JB007142
imaging of higher-mode Rayleigh dispersion curves from ambient Yao H, Gouédard P, Collins JA, McGuire JJ, van der Hilst RD (2011) Structure
seismic noise data. J Geophys Res 124(4):3708–3723. https://doi.org/10. of young east pacific rise lithosphere from ambient noise correlation
1029/2018jb016595 analysis of fundamental- and higher-mode Scholte-Rayleigh waves. C R
Wang K, Liu Q, Yang Y (2019) Three-dimensional sensitivity kernels for Geosci 343(8–9):571–583. https://doi.org/10.1016/j.crte.2011.04.004
multicomponent empirical Green’s functions from ambient noise: Yoshizawa K, Kennett BLN (2002) Non-linear waveform inversion for surface
Methodology and application to adjoint tomography. J Geophys Res waves with a neighbourhood algorithm-application to multimode
124(6):5794–5810. https://doi.org/10.1029/2018jb017020 dispersion measurements. Geophys J Int 149(1):118–133. https://doi.
Wang Y, Lin F-C, Schmandt B, Farrell J (2017) Ambient noise tomography org/10.1046/j.1365-246X.2002.01634.x
across mount St. Helens using a dense seismic array. J Geophys Res Yoshizawa K, Kennett BLN (2004) Multimode surface wave tomography for
122(6):4492–4508. https://doi.org/10.1002/2016JB013769 the Australian region using a three-stage approach incorporating finite
Wang K, Jiang C, Yang Y, Schulte-Pelkum V, Liu Q (2020) Crustal deformation frequency effects. J Geophys Res 109(B2):1–19. https://doi.org/10.1029/
in southern California constrained by radial anisotropy from ambient 2002JB002254
noise adjoint tomography. Geophys Res Lett. https://doi.org/10.1029/ Yoshizawa K, Kennett BLN (2005) Sensitivity kernels for finite-frequency surface
2020gl088580 waves. Geophys J Int 162(3):910–926. https://doi.org/10.1111/j.1365-
Wapenaar K, Draganov D, Snieder R, Campman X, Verdel A (2010) Tutorial 246X.2005.02707.x
on seismic interferometry: part 1-basic principles and applications. Zagoskin A (2014) Quantum theory of many-body systems. Springer, New
Geophysics 75(5):75–195 York. https://doi.org/10.1007/978-3-319-07049-0
Zhao K, Yang Y, Luo Y (2020) Broadband finite frequency ambient noise

tomography: a case study in the western United States using USArray
stations. J Geophys Res. https://doi.org/10.1029/2019jb019314
Zheng S, Sun X, Song X, Yang Y, Ritzwoller MH (2008) Surface wave tomog-
raphy of China from ambient seismic noise correlation. Geochem
Geophys Geosyst 9(5):1–8. https://doi.org/10.1029/2008GC001981
Zhou L, Song X, Weaver RL (2020) Retrieval of amplitude and attenuation from
ambient seismic noise: synthetic data and practical considerations.
Geophys J Int 222(1):544–559. https://doi.org/10.1093/gji/ggaa194

Ambient Noise Multimode Surface Wave Tomography: Review Open Access

Uploaded by

Copyright:

Available Formats

Ambient Noise Multimode Surface Wave Tomography: Review Open Access

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ambient Noise Multimode Surface Wave Tomography: Review Open Access

Uploaded by

Copyright:

Available Formats

Nishida et al.

Progress in Earth and

REVIEW Open Access

Ambient noise multimode surface wave

1 Introduction frequently. The source and station distributions limit the

(a) 1st step: secion 4 (b) 2nd step: section 5

ZZ, ZR, ZZ, ZR,

Reference dispersion curves Dispersion curves for each path

Frequency Frequency Frequency Frequency

(c) 3nd step: section 6 (d) 4th step: section 7

0th Rayleigh 1st Rayleigh

 L  2 Then, the CCF can be represented by

feasible for better fitting a synthetic cross-spectrum to

where am and bm are Fourier coefficients of the source ∞

y the higher frequency, the travel-time anomaly decreases

3 Data processing 3.1.2 Instrumental corrections

cross-spectrum ij is normalized by the amplitude (e.g.,

where α0 is the reference value at the separation distance

(a) Gaussian filter (b) Gaussian filter

3.2 Anomaly [%] 2

Distance [km] Anomaly [%]

windowing the CCF around the arrival time predicted

where N is an integer number. The π/4 originates from

(a) Station distribution (b) Histogram of separation distance

0.1 0.2 0.3 0.4 0.5

SPAC: VR The maximum is inferred using the grid search method.

and the slowness of 0.3 s/km. We used an array with a

πω  than the theoretical value. The bias is caused by the fall-

components even if Love and Rayleigh waves appear in ωn r

noise continuously (Takagi and Nishida 2022; Takeo et al.

−ik  +i(χ23 +χ13 ) 

√−ik −ik(χ13 +χ23 ) B(θ ) + eik(χ13 −χ23 ) (B(θ

√−ik ik(χ13 +χ23 ) B(θ )

− 100 − 50 0 50 100 − 100 − 50 0 50 100 − 100 − 50 0 50 100

Then, the phase kernel Kp can be written by

(a) Earthquake surface wave tomography (b) ANT

6.3.1 Regionalization using subarrays

6.3.2 Ray‑theoretical inversion of ANT

6.3.3 Finite‑frequency inversion of ANT 7 3‑D seismic velocity inversion using

(a) (b) 0th Rayleigh 1st Rayleigh

(c) 0th Rayleigh 1st Rayleigh 0th Love 1st Love

Phase velocity (km/s)

(a) Fundamental Rayleigh wave (d) Multi-mode (e) Fundamental mode

(c)Fundamental Rayleigh wave

difficult to measure, the spectral ratio of the horizontal 8 Conclusions

Received: 29 September 2023 Accepted: 23 December 2023

Zhao K, Yang Y, Luo Y (2020) Broadband finite frequency ambient noise

You might also like

L 2 Then, the CCF can be represented by

3 Data processing 3.1.2 Instrumental corrections

cross-spectrum ij is normalized by the amplitude (e.g.,

πω than the theoretical value. The bias is caused by the fall-

−ik +i(χ23 +χ13 )

6.3.1 Regionalization using subarrays

6.3.2 Ray‑theoretical inversion of ANT

6.3.3 Finite‑frequency inversion of ANT 7 3‑D seismic velocity inversion using

difficult to measure, the spectral ratio of the horizontal 8 Conclusions