Geophysical Journal International

Geophys. J. Int. (2013) 192, 346–358 doi: 10.1093/gji/ggs023

On measuring surface wave phase velocity from station–station

cross-correlation of ambient signal

Lapo Boschi,1,2 Cornelis Weemstra,3,4 Julie Verbeke,5 Göran Ekström,5 Andrea Zunino6
and Domenico Giardini3
1 UMR 7193 UPMC-CNRS, Laboratoire ISTeP, Université Pierre et Marie Curie, Paris, France. E-mail: lapo.boschi@upmc.fr
2 ETH Zurich, Main building, Rämistrasse 101, 8092 Zurich, Switzerland
3 Institute of Geophysics, ETH Zurich, Zurich, Switzerland
4 Spectraseis, Inc., Denver, Colorado, USA
5 Lamont-Doherty Earth Observatory, Columbia University, New York, NY, USA
6 Department of Informatics and Mathematical Modeling, Technical University of Denmark, Lyngby, Denmark

Accepted 2012 October 12. Received 2012 October 12; in original form 2012 July 30

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

SUMMARY
We apply two different algorithms to measure surface wave phase velocity, as a function of
frequency, from seismic ambient noise recorded at pairs of stations from a large European
network. The two methods are based on consistent theoretical formulations, but differ in the
implementation: one method involves the time-domain cross-correlation of signal recorded
at different stations; the other is based on frequency-domain cross-correlation, and requires
finding the zero-crossings of the real part of the cross-correlation spectrum. Furthermore,
the time-domain method, as implemented here and in the literature, practically involves the
important approximation that interstation distance be large compared to seismic wavelength.
In both cases, cross-correlations are ensemble-averaged over a relatively long period of time
(1 yr). We verify that the two algorithms give consistent results, and infer that phase velocity
can be successfully measured through ensemble-averaging of seismic ambient noise, further
validating earlier studies that had followed either approach. The description of our experiment
GJI Seismology

and its results is accompanied by a detailed though simplifed derivation of ambient-noise

theory, writing out explicitly the relationships between the surface wave Green’s function,
ambient-noise cross-correlation and phase and group velocities.
Key words: Time-series analysis; Interferometry; Seismic tomography; Theoretical
seismology; Crustal structure.

tions approximates the Green’s function associated with a point

1 I N T RO D U C T I O N
source acting at one of the stations’ location, and a receiver de-
The ability to observe coherent surface wave signal from the stacked ployed at the other’s. Such empirical Green’s function can then
cross-correlation of background noise recorded at different stations be analysed in different ways, with the ultimate goal of obtaining
is essential to improve our resolution of Earth structure via seismic information about Earth’s structure at various depths between the
imaging. Surface waves generated by earthquakes are best observed two stations. Most authors either extract group velocity v g from its
at teleseismic distances, where the body- and surface wave packets envelope (e.g. Shapiro et al. 2005; Stehly et al. 2006, 2009), or iso-
are well separated, and owing to different geometrical spreading, late the phase velocity v (e.g. Lin et al. 2008; Nishida et al. 2008;
surface waves are much more energetic than body waves; teleseis- Ekström et al. 2009; Yao & van der Hilst 2009). Fewer authors (e.g.
mic surface waves, however, are dominated by intermediate to long Tromp et al. 2010; Basini et al. 2012) attempt to explain (invert)
periods (30 s), and their speed of propagation is therefore related the entire ambient-noise waveform.
to mantle, rather than crustal structure (e.g. Boschi & Ekström Both v and v g are useful expressions of shallow Earth properties
2002). The averaged cross-correlated ambient-noise signal is in- between seismic source and receiver, or, in this case, between two
stead observed at periods roughly between 5 and 30 s (e.g. Stehly receivers. To measure v g one must be able to identify the peak of
et al. 2006, 2009), complementary to the period range of teleseismic the surface wave envelope. This, as a general rule, is easier than
surface waves, and allowing to extend imaging resolution upwards isolating the carrying sinusoidal wave (i.e. measuring v) at a given
into the lithosphere–asthenosphere boundary region and the crust. frequency. There are, however, several properties of v g that make
As first noted by Shapiro & Campillo (2004), the cross- phase-velocity observations useful and possibly preferable: (i) the
correlation of seismic ambient signal recorded at two different sta- envelope peak is less precisely defined than the phase of the carrying

346 
C The Authors 2012. Published by Oxford University Press on behalf of The Royal Astronomical Society.
 Measuring surface wave phase velocity 347

sinusoidal wave; (ii) at least so far as the surface wave fundamental

mode is concerned, v g depends on, and is in turn used to image,
structure over a narrower and shallower depth range than v (e.g.
Ritzwoller et al. 2001), so that v is particularly helpful to resolve
larger depths and (iii) a v g measurement needs to be made over a
wider time window than a v measurement, and contamination by
interfering phases is accordingly more likely.
While the validity of group-velocity estimates based on seismic
ambient noise is widely recognized, phase velocity is more elusive.
For instance, Yao et al. (2006) have noted an important discrepancy
between two-station observations of phase velocity obtained from
teleseismic versus ambient signal. The systematic application of a
far-field approximation, in the theoretical expression used to extract
the phase from cross-correlation observations (see eqs 41 and 35),
results in a π /4 shift with respect to the cross-correlation of ballistic Figure 1. Modified from Tsai (2009). Stations at x and y are separated by
signal (e.g. Harmon et al. 2008), which has caused some confusion a distance x > 0. Noise sources are far enough that the azimuth θ of any
as noted, for example, by Tsai (2009). We apply here two different given source is about the same with respect to either station.
approaches to measure interstation surface wave phase velocity
from 1 yr of continuous recording at a dense, large array of European Another important assumption of our and most other formula-
stations, first compiled by Verbeke et al. (2012). Both methods tions of ambient-noise theory is that the ambient signal be approxi-

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

can be derived from the same basic theoretical formulation (Tsai mately ‘diffuse’. In practice, this is not true at any moment in time,
& Moschetti 2010). One of them is based on time-domain cross- but can be at least partially achieved if the ambient signal recorded
correlation, and is implemented, here and elsewhere, using a far- over a very long time (e.g. 1 yr) is subdivided into shorter (e.g.
field approximation of the wavefield equation. The other is based 1-d long) intervals, which are then whitened and (after station–
on frequency domain cross-correlation, and on finding the roots station cross-correlation) stacked (Yang & Ritzwoller 2008; Mula-
of the real part of the cross-correlation spectrum; it involves no rgia 2012). This procedure is described in detail by Bensen et al.
far-field approximation. The consistency between the two methods’ (2007); we refer to it as ‘ensemble-average’, rather than time aver-
results further validates earlier phase-velocity tomography studies age, since shorter time intervals can be chosen to overlap (e.g. Seats
conducted with either approach (e.g. Lin et al. 2008; Ekström et al. et al. 2012; Weemstra et al. 2012). Over time, an array of seismic
2009; Yao & van der Hilst 2009; Fry et al. 2010; Verbeke et al. stations will record ambient signal generated over a wide range of
2012). azimuths and distances, and the process of stacking simulates the
superposition of simultaneously acting sources. Stehly et al. (2006)
show that, at least in the period range ∼5–15 s, most ambient-noise
2 T H E O RY signal is likely to be generated by the interaction between oceans
and the solid Earth (i.e. ocean storms), and the source distribu-
We study the properties of the cross-correlation Cxy (t, ω), function of tion of even the stacked ambient signal is accordingly non-uniform.
time t and frequency ω, of ambient surface wave signal u recorded at Yet, there are both empirical (Derode et al. 2003) and theoreti-
two seismic instruments, located at positions x and y. By definition cal (Snieder 2004) indications that as long as a significant fraction
 T of ambient signal hits a receiver pair along the receiver–receiver
1
C x y (t, ω) = u(x, τ, ω)u(y, t + τ, ω) dτ, (1) azimuth, ensemble-averaging will result in successful applications
2T −T
of ambient-noise methods. In our formulation, we treat sources as
with the parameter T defining the size of the window over which the uniformly distributed in azimuth with respect to the receiver pair.
cross-correlation is computed in practice. We limit our analysis to
sources sufficiently far from both receivers for the source–receiver
azimuth θ to be approximately the same. If we denote x the dis- 2.1 Monochromatic signal from a single source
tance separating the two receivers, it then follows, as illustrated in In the absence of strong lateral heterogeneity in elastic structure,
Fig. 1, that the surface wave of frequency ω and phase velocity v(ω) the momentum equation for a Love or Rayleigh wave can be de-
generated by a plane-wave source at azimuth θ hits the receiver at y coupled into a differential equation in the vertical, and another in
with an approximate delay the horizontal Cartesian coordinates. The latter coincides with the
td = x cos(θ )/v(ω) (2) Helmholtz equation and is solved by sinusoidal functions (e.g. Peter
et al. 2007).
with respect to the one at x. Seismic ambient noise can be thought of as the effect of a com-
Our treatment follows that of Tsai (2009) and Tsai (2011); we bination of sources more-or-less randomly distributed in space and
review the formulation carried out in those works, confirming the time. It is however convenient to start our treatment, following
theoretical consistency, and pointing out the practical differences Tsai (2009), from the simple case of a single source generating
between the data-analysis methods that we compare. The mathemat- a monochromatic signal of frequency ω. The first receiver then
ical treatment leads to complete expressions for cross-correlation records a signal
(Section 2.4), and group, as well as phase-velocity of the ambient
u(x, t) = S(x, ω) cos(ωt + φ), (3)
signal (Section 2.5). Like Tsai (2009), we assume, as mentioned,
that sources of ambient noise are far enough from our station pair where the constant phase delay φ is proportional to source–receiver
for the source–receiver azimuth to be approximately the same at the distance, and the amplitude term S(x, ω) is inversely proportional,
two stations. in the first approximation, to the square-root of source–receiver
348 L. Boschi et al.

distance (geometrical spreading). The signal (3) is observed at y generated by a single source localized in space, and not scattered,
with a delay td , that is, is only useful if the location of the source is known. It coincides
(once amplitude is normalized) with the response, at one station,
u(y, t) = S(y, ω) cos [ω(t + td ) + φ] . (4)
to a sinusoidal source located at the other, if and only if the two
(By virtue of eq. (2), td is negative when energy propagates from stations are aligned with the source, that is, azimuth θ = 0 or θ =
y to x (0 < θ < π /2) and positive when energy propagates from π , so that td = ±x/v.
x to y.)
Let us substitute (3) and (4) into (1), so that
 2.2 Monochromatic signal from a discrete set of sources
S(x)S(y) T
Cx y = cos(ωτ + φ) cos [ω(τ + t + td ) + φ] dτ. (5) Recorded seismic ambient noise is believed to be the cumulative
2T −T
effect of numerous localized sources, distributed almost randomly
It is convenient to substitute z = ωτ , to find all around our pair of recording instruments. The signal generated
 by a discrete set of monochromatic sources can be written as a
S(x)S(y) ωT
Cx y = cos(z + φ) cos [z + φ + ω(t + td )] dz. (6) superposition of single-source signals, eqs (3) and (4), resulting in
2ωT −ωT
We next make use of the general trigonometric identity cos (A + u(x, t) = Si (x, ω) cos(ωt + φi ) (14)
B) = cos Acos B − sin Asin B, valid for any A, B and i

 and
S(x)S(y) ωT  2

Cx y = cos (z) cos [φ + ω(t + td )] cos(φ)
2ωT −ωT u(y, t) = Si (y, ω) cos ω(t + tdi ) + φi , (15)

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

i
+ sin2 (z) sin [φ + ω(t + td )] sin(φ) where the summation is over the sources, φ i is the phase delay
associated with source i, and the time delay tdi between stations x
− sin(z) cos(z) cos(φ) sin [φ + ω(t + td )]
and y also changes with source azimuth, hence the superscript i. In
 analogy with Section 2.1, we next substitute (14) and (15) into (1),
− sin(z) cos(z) sin(φ) cos [φ + ω(t + td )] dz, (7)
and
which can be simplified if one notices that 1
 ωT  ωT Cx y =
1 + cos(2z) 1 2T
cos2 (z)dz = dz = ωT + sin(2ωT ), (8)
−ωT −ωT 2 2   T



Si (x)Sk (y) cos(ωτ + φi ) cos ω(τ + t + tdk ) + φk dτ .
 ωT  ωT −T
1 − cos(2z) 1 i,k
sin2 (z)dz = dz = ωT − sin(2ωT ), (9) (16)
−ωT −ωT 2 2
and finally Let us consider the
‘cross-terms’ (cross-correlations
 between
 ωT  2 ωT cos (ωτ + φ i ) and cos ω(τ + t + tdk ) + φk with i = k) in eq. (16):
sin(z) cos(z)dz =
sin (z)
= 0, (10) they are sinusoidal with the same frequency ω but randomly out of
−ωT 2 −ωT phase, and therefore do not interfere constructively. The remaining
(i = k) terms, on the other hand, interfere constructively, as we il-
where the notation [ f (z)] BA = f (B) − f (A).
lustrate below, so that, after the contribution of a sufficient number
After substituting the expressions (8), (9) and (10) into eq. (7),
of sources has been taken into account, the cross-term contribution
 
S(x)S(y) sin(2ωT ) becomes negligible relative to them. Following other derivations of
Cx y = 1+ cos(φ) cos [φ + ω(t + td )]
2 2ωT noise-correlation properties, we thus neglect cross-terms from this
   point on (e.g. Snieder 2004; Tsai 2009). We are left with a sum of
sin(2ωT ) integrals of the form (5), which we have proved in Section 2.1 to be
+ 1 sin(φ) sin [φ + ω(t + td )] . (11)
2ωT approximated by (13), so that
It then follows from simple trigonometric identities (cosine of the Si (x)Si (y)

Cx y ≈ cos ω(t + tdi ) . (17)
sum, sine of the sum) that 2
i
S(x)S(y)
Cx y =
2
  2.3 Continuous distribution of sources
sin(2ωT )
× cos [ω(t + td )] + cos [2φ + ω(t + td )] . (12) Eq. (17) can be further generalized to the case of a continuous
2ωT
distribution of sources,
This expression can be simplified if one considers that the size 2T
 xv
of the cross-correlation window should be large compared to the
period of the surface waves in question, that is, T  2π /ω, so that Cx y ≈ ρ(td , ω) cos [ω(t + td )] dtd , (18)
− x
v
2ωT  1. Eq. (12) then reduces to
where we have introduced the function ρ(td , ω), describing the
S(x)S(y) density of sources as a function of interstation delay td , or, which is
Cx y ≈ cos [ω(t + td )] (13)
2 the same (recall eq. 2), azimuth θ. Integration is accordingly over
(compare with eq. (1) of Tsai (2009)). From eq. (13) we infer that td , and the integration limits correspond, through eq. (2), to the
the station-station cross-correlation of a ‘ballistic’ signal, that is, interval of possible azimuths, from 0 to π . ρ is also a function of
 Measuring surface wave phase velocity 349

ω, as signal generated by differently located sources generally has a We next rewrite the integral in terms of Bessel and Struve func-
different frequency content. To keep the notation compact, we have tions. Let us first consider the 0-order Bessel function of the first
incorporated the continuous version of the source term Si (x, ω)Si (y, kind in its integral form
ω)/2 from eq. (17) in the source density function ρ(td , ω). 
1 π
In analogy with earlier formulations of ambient-noise theory, we J0 (z) = cos[z sin(θ)]dθ (25)
π 0
require the source distribution to be uniform with respect to azimuth
θ . To find the corresponding (not constant) expression of ρ as a [eq. (9.1.18) of Abramowitz & Stegun (1964)]. The integral from 0
function of td , we note that, for azimuthally constant source density, to π in (25) can be transformed into an integral from 0 to π /2
ρ(td ) multiplied by a positive increment |dtd | must coincide with the 
1 π
corresponding increment |dθ| times a constant factor. Formally, J0 (z) = cos[z sin(θ)]dθ
π 0
1  π  π
g(ω)|dθ| = ρ(td , ω)|dtd |, (19) 1 2
2π = cos[z sin(θ)]dθ + cos[z sin(θ)]dθ
π 0 π
where g(ω)/2π is the normalized value of uniform azimuthal source 2

density, selected so that its integral between 0 and 2π is exactly g(ω).  π 

0
The factor g(ω) serves to remind us that source amplitude generally 1 2
= cos[z sin(θ)]dθ − cos[z sin(π − θ
)]dθ

changes with frequency. After replacing |dtd | = xsin (θ)|dθ|/v, π 0 π

2

g(ω) x sin[θ(td )]  π  π

|dθ | = ρ(td , ω) |dθ|, (20) 1 2 2

2π v = cos[z sin(θ)]dθ + cos[z sin(θ )]dθ

π

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

0 0
or
v(ω)g(ω)  π
ρ(td , ω) = , (21) 2 2

2π x sin[θ(td )] = cos[z sin(θ)]dθ. (26)

π 0

which is the expression of ρ = ρ(td , ω) corresponding to azimuthally We then replace sin (θ) = cos (θ − π /2) and change the integration
uniform source density. variable θ = θ
+ π /2,
 π
2 2 π
J0 (z) = cos[z cos(θ − )]dθ
π 0 2
2.4 Cross-correlation and Green’s function
 0
2
It is convenient to separate the integral in eq. (18) into two integrals, = cos[z cos(θ
)]dθ

one over positive, and the other over negative td , π − π2

 0  π
2 2
Cx y ≈ ρ(td , ω) cos [ω(t + td )] dtd = cos[z cos(θ
)]dθ
, (27)
− x
v
π 0

 x and after substituting z with ωx/v,

v
+ ρ(td , ω) cos [ω(t + td )] dtd . (22)    π  
ωx 2 2 ωx
0 J0 = cos cos (θ) dθ
v π 0 v
The negative- and positive-time contributions to Cxy are usually
referred to as anticausal and causal, respectively.  π
2 2
=  eiωx cos(θ)/v
dθ . (28)
π 0

2.4.1 Positive-time (causal) contribution to the cross-correlation The 0-order Struve function also has an integral form
 π
Let us first consider the second term (td ≥ 0) at the right-hand side H0 (z) =
2 2
sin[z cos(θ)]dθ, (29)
of (22), which, since ρ(td , ω) is real (see eq. 21), can be rewritten π 0
  x which coincides with eq. (12.1.7) of Abramowitz & Stegun (1964) at
v √
C xtdy>0 ≈  eiωt ρ(td , ω)eiωtd dtd , (23) order 0 and substituting (1/2) = π , with denoting the Gamma
0 function. We replace, again, z with ωx/v, and
   π
where (. . .) equals the real part of its argument. It is convenient to ωx 2 2
replace ρ(td , ω) with its expression (21), and the integration variable H0 =  eiωx cos(θ)/v dθ , (30)
v π
td with θ . By differentiating eq. (2), dtd = −xsin (θ)dθ /v, while 0

the limits of integration 0, x/v correspond to azimuth θ = π /2, with the operator  mapping complex numbers to their imaginary
0, respectively, hence, using the symmetry of the cosine, part. It follows from (28) and (30) that
  π  π     
g(ω)eiωt 2 2 π ωx ωx
C xtdy>0 ≈  eiωx cos(θ)/v dθ . (24) eiωx cos(θ)/v dθ = J0 + iH0 , (31)
2π 0 0 2 v v
and substituting into (24)
(Recall that positive td corresponds to azimuth 0 < θ < π /2, while
     
the opposite holds for the td ≤ 0 term corresponding to π /2 < θ < g(ω)eiωt ωx ωx
C xtdy>0 ≈  J0 + iH0 . (32)
π .) 4 v v
350 L. Boschi et al.
   π 
π
Following Tsai (2009), or all other authors conducting ambient- 2 2

noise analysis in the time domain, we next assume that interstation = sin z sin θ + dθ
π 0 2
distance be much larger than the wavelength of the signal under
 π
consideration, that is, ωx/v  1. It then follows from eq. (9.2.1) 2
= sin[z sin(θ
)]dθ

of Abramowitz & Stegun (1964) that π π

 
2
  
ωx 2v ωx π  π
J0 ≈ cos − , (33) 2 2
v ωπ x v 4 = sin[z sin(−θ
)]dθ

π 0
and from eqs (12.1.34) and (9.2.2) of Abramowitz & Stegun (1964),
 π
       2 2
ωx ωx 2v ωx π =− sin[z sin(θ
)]dθ
. (39)
H0 ≈ Y0 ≈ sin − , (34) π 0
v v ωπ x v 4
Making use of eq. (39), and of expression (26) for the Bessel
with Y0 denoting the 0-order Bessel function of the second kind. function J0 , with z = ωx/v, in (38),
Substituting eqs (33) and (34) into (32),      
   g(ω)e−iωt ωx ωx
v
i(ωx/v−π/4)  C xtdy<0 ≈  J0 − iH0 , (40)
C xtdy>0 ≈  g(ω)eiωt e 4 v v
8π ωx
where only the sign of H0 at the right-hand side has changed with

v respect to eq. (32). We conclude that
= g(ω) cos [ω (x/v + t) − π/4] . (35) 
8π ωx v
C xtdy<0 ≈ g(ω) cos [ω (−x/v + t) + π/4] ,

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

(41)
Comparing eq. (35) to (13), we note a phase-shift π /4 between the 8π ωx
cross-correlated signal generated by a teleseismic event aligned with
that is, the negative-time phase-shift is symmetric to the positive-
the two stations, and that obtained from the ensemble-averaging of
time one, in agreement with Tsai (2009).
seismic ambient noise. π /4 is nWothing but the phase-shift between
Summing C xtdy<0 (eq. 41) and C xtdy>0 (eq. 35) one finds, according
a cosine and a Bessel function, for large values of the argument (i.e.
to eq. (22), an expression for Cxy valid at all, positive and negative
in the far field). In our experimental set-up, a cosine describes the
times. To verify its validity, we implement it numerically and com-
two-station cross-correlation of a plane wave hitting the receivers
pare it in Fig. 2 to the result of eq. (17) applied to a very large set of
from a single azimuth; the Bessel function (and hence the π /4 shift)
sources, for the same frequency and interstation distance. Confirm-
emerges from the combined effect of plane waves coming from all
ing earlier findings, the two differently computed cross-correlations
azimuths (i.e. focusing over the receiver array).
are practically coincident.

2.4.2 Negative-time (anticausal) contribution to the

2.5 Group and phase velocity
cross-correlation
We next consider the more general case of a seismogram formed
An analogous treatment applies to the negative-time cross-
by the superposition of surface waves with different frequencies.
correlation C xtdy<0 , that is, the first term at the right-hand side of
Let us start with our expression (35) for the cross-correlated signal,
eq. (22), which after the variable change from td to θ becomes
 grouping the amplitude terms in a generic positive factor S(ω). We

td <0 g(ω)eiωt π iωx cos(θ)/v then find the mathematical expression of a surface wave packet by
Cx y ≈  e dθ . (36) (i) discretizing the frequency band of interest into a set of closely-
2π π
2
spaced frequencies ωi identified by the subscript i and (ii) combin-
To express also this integral in terms of Bessel and Struve functions, ing different-frequency contributions by integration around each
we first notice that frequency ωi and summation over i, so that
 π  π  
2 π 
∞  ωi +ε  
x

π

f [cos(θ)]dθ = f cos θ
+ dθ u(x, t) = S(x, ω) cos ω +t − dω, (42)
π 0 2 v(ω) 4
i=1 ωi −ε
2

 π  π   π 
2 where ε  ωi . It is convenient to introduce the notation ψ =
= f cos(θ
) cos − sin(θ
) sin dθ

0 2 2 ω(x/v + t) − π /4, and, since ε is small, replace it with its Taylor

 π
expansion around ωi , that is,
2  
= f [− sin(θ
)]dθ
, (37) dψ
0 ψ(ω) ≈ ψ(ωi ) + (ω − ωi ) , (43)
dω ωi
for an arbitrary function f. From eq. (36) it then follows that
 where [ f (ω)]ωi denotes the value of any function f evaluated at ω =
 π
td <0 g(ω)eiωt 2
−iωx sin(θ)/v ωi . We rewrite eq. (42) accordingly, and find after some algebra that
Cx y ≈  e dθ . (38)
2π 0 the integral at its right-hand side
 ωi +ε    
Similar to eq. (27) in Section 2.4.1, we next replace cos (θ) = x π
S(ω) cos ω +t − dω
sin (θ + π /2) in expression (29) for the Struve function, and change ωi −ε v(ω) 4
the integration variable θ
= θ + π2 , 
 
 π 2 sin ε dψ dω ωi
H0 (z) =
2 2
sin[z cos(θ )]dθ ≈ S(ωi ) cos [ψ(ωi )]
dψ  (44)
π 0 dω ωi
 Measuring surface wave phase velocity 351

Figure 2. Numerical test of expression (41) + (35), with interstation distance of 500 km and wave speed of 3 km s−1 . Cxy resulting from the direct
implementation of (41) + (35) is denoted by a solid line. We compare it with the result of applying eq. (17) to model Cxy from the combined effect of 1000,
far, sinusoidal (with 4-s period) out-of-phase sources located at 200 different, uniformly distributed azimuths from the station couple. Finally, we also compute
Cxy from eq. (16) (crosses), neglecting the cross-terms i = k; a slight decay, with increasing lag, in the latter estimate of Cxy is caused by the finite length of
the time-integral in the implementation of (16). Amplitudes have been normalized. All modeled cross-correlations are perfectly in phase.

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

(valid in the assumption that S be a smooth function of ω). If one 3 H OW TO M E A S U R E P H A S E V E LO C I T Y
introduces a function
To evaluate whether phase velocity can be accurately observed in
v(ω) the ensemble-averaged cross-correlation of ambient noise, we use
vg (ω) = ω dv
, (45)
1 − v(ω) dω
two independent approaches to measure it from the same data.
Consistency of the results is then an indication of their validity.
it follows that dψ
takes the compact form The first approach (Section 3.1) consists of cross-correlating and
dω
stacking the surface wave signal (t-long records of ambient signal
 
dψ x in our case) to find the empirical Green’s function (Section 2.4),
= + t; (46) from which phase velocity can be measured (e.g. section 12.6.2 of
dω ωi vg (ωi )
Udı́as 1999). If, as is most often the case, one works in the far-field
we finally substitute it into (44) and substitute the resulting expres- approximation, this requires that a π /4 correction be applied to
sion into (42), to find the data as explained in Section 2.5, eq. (47). The other approach
we consider is based on the result of Aki (1957), confirmed by

∞
Ekström et al. (2009) for the frequency range of interest, that the
u(x, t) = S(ωi )
spectrum of the two-station cross-correlation of seismic ambient
i=1
noise should approximately coincide with a 0-order Bessel function
  
    x of the first kind (Section 3.2); in this case, no π /4 correction needs
x π 2 sin ε vg (ωi ) + t
× cos ωi +t −   . to be applied.
v(ωi ) 4 x
+tvg (ωi )

(47) 3.1 Time-domain cross-correlation

Each term at the right-hand side of eq. (47) is the product of a The procedure of ensemble-averaging ambient signal is described
wave of frequency ω and speed v(ω) with one of frequency ε  in detail, for example, by Bensen et al. (2007); a long (e.g. 1 yr) con-
ω and speed v g (ωi ). The latter factor, with much lower frequency, tinuous seismic record is subdivided into shorter t intervals. The
modulates the signal, and we call ‘group velocity’ its speed v g , records are whitened, and they are normalized in the time-domain
which coincides with the speed of the envelope of the signal. Eq. so that the effects of possible ballistic signal (i.e. large earthquakes)
(45) shows that, in the absence of dispersion (i.e. dω dv
= 0) phase present in the data are minimized. The cross-correlation between
and group velocities coincide. In practice, the values of v and v g are simultaneous t-long records from different stations is then com-
always comparable, and the large difference in frequency results in puted for all available t intervals, and the results for each station
a large difference in the wavelength of the phase and group terms. pair are stacked over the entire year.
Comparing eq. (47) to (42), it is important to notice that Bensen et al. (2007) measure group velocity from noise cross-
when phase velocity is measured from the station–station cross- correlations, and suggest that phase dispersion can be obtained
correlation of ambient signal, a phase correction of π /4 must first by integration of group dispersion curves. This approach however
be applied; the same is not true for group-velocity measurements. is not sufficient to identify phase velocity uniquely. Meier et al.
We have shown in Sections 2.4.1 and 2.4.2 that ambient-noise cross- (2004) provide an algorithm to derive phase velocity from the cross-
correlation coincides with a combination of Bessel functions, and correlation of teleseismic signals recorded by stations aligned with
that, for large values of their argument (corresponding to relatively the earthquake azimuth. Fry et al. (2010) and Verbeke et al. (in
large interstation distance), Bessel functions can be replaced by si- preparation) show that the algorithm of Meier et al. (2004) can be
nusoidal functions, whose argument coincides with the argument successfully applied to the ambient signal recorded at a regional-
of the Bessel functions minus π /4. The π /4-shift in (42) and (47) scale array of broadband stations. In reference to the study of Fry
arises precisely from this far-field approximation. et al. (2010) where it was first introduced, we shall dub this ap-
352 L. Boschi et al.

proach FRY. In the following we shall analyse a subset of the phase-

dispersion database compiled by Verbeke et al. (in preparation) via
their own automated implementation of FRY.
The phase-velocity measurements of Verbeke et al. (in prepara-
tion) are limited to the 0.02–0.1 Hz frequency range, where seismic
ambient noise is known to be strong (Stehly et al. 2009), most likely
as an effect of ocean storms and the coupling between oceans and the
solid Earth (Stehly et al. 2006). Frequency is discretized with incre-
ments whose length increases with increasing frequency (from 0.02
to 0.05 Hz). For each discrete frequency value, ensemble-averaged
cross-correlations are (i) bandpass filtered around the frequency in
question and (ii) windowed in the time-domain via a Gaussian win-
dow centred around the time of maximum amplitude of (filtered)
cross-correlation. Causal and anticausal parts are folded together
(i.e. stacked after reversing the time-dependence of the anticausal
one). The resulting time series is Fourier-transformed, and its phase
is identified as the arctangent of the ratio of the imaginary to real
part of the Fourier spectrum, as explained by (Udı́as 1999, section
12.6.1). Based on eq. (35), one must sum π /4 to the resulting folded
ensemble-averaged cross-correlation phase before applying eq. (3)

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

of Meier et al. (2004) (equivalent to eq. 12.56 of Udı́as 1999). Impor-
tantly, this π /4 shift is specific to ambient-noise cross-correlation,
and must not be applied in two-station analysis of ballistic surface
wave signal, as shown by eq. (13). Phase velocity is only known
up to a 2π n ‘multiple cycle ambiguity’, with n = 0, ±1, ±2, . . . .
After iterating over the entire frequency band, an array of disper-
sion curves is found, each corresponding to a value of n. Verbeke
et al. (in preparation) compare each curve (for all integer values of
n between −5 and 5) with phase velocity as predicted by PREM
(Dziewonski & Anderson 1981), and pick the one closest to PREM,
considering only the frequency range where the measurement is
reliable (no large jumps for small variations in frequency). [More
sophisticated procedures exist to resolve the ambiguity (e.g. Lin &
Ritzwoller 2011; Gouedard et al. 2012), but here we stick to the
simpler algorithm of Verbeke et al. (in preparation).]
Ensemble-averaged cross-correlations for two Swiss stations
(Fig. 3a) are shown in Fig. 3(b). At long period (compared to inter-
station distance divided by wave speed) the causal and anticausal
parts of the cross-correlation overlap, complicating the time-domain
analysis of cross-correlation, whose results are shown in Fig. 3(c).

3.2 Frequency-domain cross-correlation and

Bessel-function fitting
A different method, hereafter referred to as ‘AKI’, to extrapolate
phase velocity from the ambient signal recorded at two stations is
proposed by Ekström et al. (2009), based on much earlier work by
Aki (1957). The theoretical basis of this method has been recently Figure 3. Illustration of the FRY method. (a) Locations (triangles) of sta-
rederived by Nakahara (2006), Yokoi & Margaryan (2008) and Tsai tions TORNY and VDL, from the Swiss broadband network. (b) Ensemble-
& Moschetti (2010). As pointed out by Ekström et al. (2009), this averaged cross-correlation of continuous signal recorded at TORNY and
approach does not require that ωx/v  1, that is, it will work for VDL, filtered over different frequency bands as indicated; the bottom trace
wavelengths comparable to interstation distance. is the ‘full’ waveform. (c) Array of possible phase-velocity dispersion curves
According to AKI, ambient signal recorded over a long time (e.g. from cross-correlation of the continuous recordings made at TORNY and
1 yr) is, again, subdivided into shorter t intervals. Let us call VDL; each curve corresponds to a different value of n, identified by the
pi (ω) the frequency spectrum associated with a t-long record at curve colour as indicated. The black curve, closest to our selected reference
model (PREM), is our preferred one, but observations are only considered
station i (Fig. 4a, with t = 2 hr). After whitening, this is multiplied
valid in the frequency range marked by black squares.
with the simultaneous t-long recording made at another station j
(Fig. 4b), resulting in the cross-spectrum, or spectrum of the cross-
correlation between the two t-long records (Fig. 4c). This pro- Aki (1957),
cedure is repeated for all available t-intervals in the year, which     
pi p∗j ωx
are then stacked together, that is, ensemble-averaged (Fig. 4d). The  ∝ J0 , (48)
| pi | | p j | v(ω)
resulting quantity is usually referred to as ‘coherency’. Based on
 Measuring surface wave phase velocity 353

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

Figure 4. Illustration of the AKI approach. (a) Real part of the spectrum (ms) obtained Fourier-transforming two hours of ambient recording at station TORNY.
(b) same as (a), from the very same two hours of signal recorded at station VDL. (c) Product of (a) and (b) (coinciding with the real part of the spectrum of
the cross-correlation of the two time-domain signals) obtained after whitening both. (d) Results of ensemble-averaging an entire year of spectra like the one
at (c), for the same two stations: the blue and red lines identify values of real and imaginary parts found at various frequencies; the black solid line is the
linear combination of cubic splines that best-fits the observed real part of the spectrum. The locations of stations TORNY and VDL are shown in Fig. 3(a).
Cross-spectra in both (c) and (d) are implicitly normalized and hence unitless.

where ... denotes ensemble averaging, the left-hand side is pre- zero are identified. If ω = ωi for some i, the argument of (48) must
cisely what we call coherency, and the superscript ∗ marks the coincide with one of the known zeros zn (n = 1, 2, ...) of the Bessel
complex conjugate of a complex number. The quantities at the function J0 ,
right-hand side of (48) are defined as in Section 2.4 above, with x
ωi x
distance between stations i and j. [The alert reader might notice at = zn . (50)
this point that the right-hand side of eq. (48) is proportional to Cxy : v(ωi )
simply sum, according to eq. (22), its positive- and negative-time Eq. (50) can be solved for v,
contributions (32) and (40), respectively (Tsai & Moschetti 2010).]
Again based on Aki (1957), the ensemble-averaged imaginary ωi x
v(ωi ) = , (51)
part zn
  
pi p∗j and we now have an array of possible measurements of phase ve-
 = 0. (49) locity at the frequency ωi , each corresponding to a different value
| pi | | p j |
of n. Implementing (51) at all observed values of ωi , an array of
Importantly, both eqs (48) and (49) are shown by Aki (1957) to be dispersion curves is found. Much like in the case of FRY (Sec-
valid provided that the energy of ambient signal is approximately tion 3.1), a criterion must then be established to select a unique
uniform with respect to azimuth. As anticipated at the beginning of curve.
Section 2, this is typically not true at any moment in time, but can Importantly, the observation of ωi on ensemble-averaged cross-
be achieved, at least to some extent, by ensemble-averaging (Yang spectra like the one of Fig. 4(d) is complicated by small oscillations
& Ritzwoller 2008). that can be attributed to instrumental noise or inaccuracies related to
Eq. (48) can be used to determine phase dispersion. In practice, non-uniformity in the source distribution. Before identifying ωi , we
observed coherency is first of all plotted as a function of frequency determine the linear combination of cubic splines that best fits (in
(i.e. the ensemble-averaged, whitened cross-spectrum is plotted). least-squares sense, via the LSQR algorithm of Paige & Saunders
Values ωi (i = 1, 2, 3, . . . ) of frequency for which coherency is (1982)) observed coherency. Splines are equally spaced, and spacing
354 L. Boschi et al.

must be selected so that ‘splined’ coherency is sufficiently smooth

4 A P P L I C AT I O N T O C E N T R A L
(Fig. 4d).
E U R O P E A N D ATA A N D
Eqs (48) and (49) are rarely satisfied by seismic ambient noise
C R O S S - VA L I D AT I O N O F T H E T W O
as observed in the real world. At a given time, the wavefield associ-
METHODS
ated with ambient noise is not diffuse. The procedure of ensemble-
averaging over a long time serves precisely to mimic a diffuse Fig. 5 shows the set of ∼1000 randomly selected station pairs from
wavefield by combining non-diffuse ones. Yet, there are important Verbeke et al. (in preparation) that we shall analyse here. The corre-
systematic effects that ensemble-averaging does not remove: in Eu- sponding phase-velocity dispersion curves were measured by Ver-
rope, for example, most of the recorded seismic noise is generated in beke et al. (in preparation) following the procedure of Section 3.1,
the Atlantic Ocean (Stehly et al. 2006, 2009; Verbeke et al. 2012b), after subdividing the entire year 2006 into day-long intervals and
and the requirement of an azimuthally uniform source distribution ensemble-averaging the resulting day-long cross-correlations.
is accordingly not met. Presumably, scattering partly compensates We apply the AKI method of Section 3.2 to continuous records
for that, but the non-zero observed imaginary part of the coherency associated with the station pairs of Fig. 5. Our implementation was
shown, for example, in Fig. 4(d) indicates that the problem remains originally designed for reservoir-scale application (Weemstra et al.
(e.g. Cox 1973). The imaginary part should converge to zero if one 2012), but could be applied to our continent-scale array of data after
ensemble-averages not only over time, but also over station-pair az- only minor modifications. For each station, continuous recording
imuth (e.g. Weemstra et al. 2012), but then information on lateral for the entire year 2006 is subdivided into intervals of t = 2 hr,
Earth structure would be lost. with a very conservative 75 per cent overlap between neighbouring
It is practical to focus the analysis on zero crossings, rather than intervals to make sure that no coherent signal traveling from station
measuring the overall fit between J0 and measured coherency. The to station is neglected (Seats et al. 2012; Weemstra et al. 2012).

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

latter depends on the power spectrum of the noise sources, of which This results in as many as 45 spectra per d.
we know very little, and can be affected importantly by data pro- In Fig. 6, we compare our new phase-velocity measurements with
cessing (Ekström et al. 2009). those of Verbeke et al. (in preparation) for three example station

Figure 5. (a) Subset of European stations (circles) from Verbeke et al. (in preparation) that are also included in our analysis. We only compare phase-velocity
measurements associated with ∼1000 station pairs connected by solid lines. (b) Distribution of epicentral-distance values sampled by the data set at (a).
 Measuring surface wave phase velocity 355

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

Figure 6. Selected FRY phase-velocity dispersion measurements (black circles, connected by a black line) compared with analogous frequency-domain (AKI)
measurements (grey triangles), for three station pairs: (a) AQU and GIUL, in central Italy, only ∼90 km away from each other, with a north–south azimuth;
(b) TORNY and VDL (see Fig. 3a), with interstation distance of ∼190 km; (c) WTTA in western Austria and ZCCA in northern Italy, ∼330 km to the south.
Triangles in panel (d) mark the locations of all six stations considered here. We have not yet implemented an algorithm for automatic selection of a preferred
AKI dispersion curve, but the FRY curves clearly fit a single branch of AKI data points. At low frequencies, and particularly at shorter epicentral distances, the
match is less accurate. At longer epicentral distances and high frequencies, occasional one-cycle jumps as in (c) occur.

Figure 7. Frequency of observed phase-velocity misfit (AKI values subtracted from FRY ones) for the total set of ∼1000 analysed station pairs. The mean is
13 m s−1 and the standard deviation is 151 m s−1 .
356 L. Boschi et al.

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

Figure 8. FRY-AKI phase-velocity misfit, for the total set of ∼1000 analysed station pairs, averaged within (a) ∼0.3◦ interstation-distance bins, and (b) 2-mHz
frequency bins.

pairs. A visual analysis (which we repeated on many more pairs) Fig. 8(b) the misfit is likewise averaged within 2-mHz increments
suggests that the two methods provide very similar results. spanning the whole frequency range of interest. Fig. 8(a) shows that
To evaluate quantitatively their level of consistency, we first ex- FRY has a tendency to give slightly higher velocity estimates with
pand FRY dispersion curves over a set of cubic splines, and apply respect to AKI; this effect is reversed at very small and very large
spline interpolation to estimate FRY-based phase-velocity values at interstation distances. The misfit remains low (∼30 m s−1 or less)
the exact frequencies (associated with zero-crossings of the Bessel at most interstation distances.
function) where AKI measurements are available. We subtract the Fig. 8(b) shows clearly that misfit is systematically smaller
AKI phase velocities from the FRY ones interpolated at the same (20 m s−1 ) at relatively high frequencies (0.04 Hz) than it is
frequency, selecting at each frequency the AKI data point closest at low frequencies of ∼0.02–0.03 Hz. This is expected, as low
to the FRY one (we thus avoid the well known issue of multiple- frequency might result in relatively small ωx/v, which would
cycle ambiguity, that equally affects both approaches). We count deteriorate the performance of FRY (but not of AKI) for short inter-
the number of discrepancy observations, independent of frequency, station distance x: in practice, the causal and anticausal parts tend
falling in each of a set of 50 m s−1 intervals, and plot the associ- to overlap in the short-x time-domain cross-correlations, making
ated histogram in Fig. 7. Both mean and standard deviation of the it difficult to measure phase via the FRY method (e.g. Ekström et al.
FRY-AKI discrepancy are small (13 and 151 m s−1 , respectively), 2009).
and we conclude that, in our implementation, the two approaches The combined effect of short x and low frequency is per-
provide consistent results when applied to the data. Outliers exist haps better illustrated in Fig. 9(a), where both frequency- and x-
with misfit larger than ±1000 m s−1 , but they would not be visible dependence of misfit are shown in a single, 2-D plot. It emerges
in Fig. 7 even after extending the horizontal-axis range. that, even at low frequency, AKI and FRY are in good agreement
We next analyse the dependence of FRY-AKI discrepancy on for sufficiently large interstation distance. Fig. 9(b) shows that, not
interstation distance, through a second histogram (Fig. 8a) where surprisingly, sampling is not uniform with respect to frequency
the misfit is averaged within ∼0.3◦ interstation-distance bins. In and x; most seismic-ambient-noise energy in our station array is
 Measuring surface wave phase velocity 357

Figure 9. (a) FRY-AKI phase-velocity misfit, for the total set of ∼1000 analysed station pairs, averaged within (a) ∼0.2◦ × 0.04-Hz distance/frequency bins
and (b) number of pairs per distance/frequency bin.

found at frequencies around ∼0.05 Hz, and some of the discrepancy lies does not hold. We infer that Rayleigh-wave phase velocity
found at both higher and lower frequency (see in particular the top can be successfully observed, via ensemble averaging, from contin-

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

right-hand side of Fig. 8 a) presumably reflects the difficulty of find- uous recordings of seismic ambient noise, at least within the fre-
ing coherent signal in the absence of a sufficiently strong ambient quency (∼0.03–0.1 Hz) and interstation distance (∼0.5◦ –5◦ ) ranges
wavefield. analysed here. We further confirm the validity of published phase-
Overall, averaged discrepancies in Figs 8 and 9 remain velocity observations (e.g. Verbeke et al. 2012b) obtained through
50 m s−1 , with the exception of the lowest frequencies/shortest the time-domain approach.
epicentral distances considered, where averaged values can exceed
∼100 m s−1 . A velocity difference of 50 m s−1 can be considered
small if compared with the range of velocity heterogeneity in the
frequency range and geographic area of interest, that is, ∼1 km s−1
or more according to Verbeke et al. (2012b). We take this as an indi- AC K N OW L E D G M E N T S
cation that the AKI and FRY methods provide essentially consistent This study benefitted from our interactions with Michel Campillo,
results, and we infer that such results can be considered reliable. Bill Fry, Edi Kissling, Laurent Stehly, Victor Tsai and Yang Zha.
We are thankful to Eiichi Fukuyama, Mike Ritzwoller and one
anonymous reviewer for their insightful comments. A. Z. wishes to
5 C O N C LU S I O N S thank Daniele Spallarossa for his advice and constant support.
With this study we have conducted a detailed review of the theory
of ensemble-averaged cross-correlation of surface waves generated
by seismic ambient noise, as more tersely described by Tsai (2009,
2011) and Tsai & Moschetti (2010). With our rederivation we at-
tempt to focus the reader’s attention on the potential discrepancy be- REFERENCES
tween the time-domain and frequency-domain approaches in phase- Abramowitz, M. & Stegun, I.A., 1964. Handbook of Mathematical Func-
velocity measurements conducted on ambient-noise surface waves. tions with Formulas, Graphs and Mathematical Tables, 55, National
The possibly most important difference between the two methods Bureau of Standards Applied Mathematics Series, Washington, D.C.
resides in the far-field approximation that is generally applied by Aki, K., 1957. Space and time spectra of stationary waves with special
time-domain practitioners (e.g. Lin et al. 2008; Yao & van der Hilst reference to microtremors, Bull. Earthq. Res. Inst. Univ. Tokyo, 35, 415–
456.
2009; Fry et al. 2010; Verbeke et al. 2012b), and we have empha-
Basini, P., Nissen-Meyer, T., Boschi, L., Casarotti, E., Verbeke, J., Schenk, O.
sized how this approximation is inadequate for interstation distances & Giardini, D., 2012. Adjoint ambient-noise tomography of the European
comparable to the seismic wavelengths. The frequency-domain ap- lithosphere, Geochem., Geophys., Geosyst., submitted.
proach of Aki (1957) and Ekström et al. (2009) does not suffer Bensen, G.D., Ritzwoller, M.H., Barmin, M.P., Levshin, A.L., Lin, F.,
from this limitation: it is thus particularly useful for closely-spaced Moschetti, M.P., Shapiro, N.M. & Yang, Y., 2007. Processing seismic
stations, provided that precursory noise caused by inhomogeneities ambient noise data to obtain reliable broad-band surface wave dispersion
in the source distribution is negligible (e.g. Shapiro et al. 2006; Lin measurements, Geophys. J. Int., 169, 1239–1260.
et al. 2007; Villasenor et al. 2007; Zheng et al. 2011). Boschi, L. & Ekström, G., 2002. New images of the Earth’s upper mantle
We have employed our own implementations of the non- from measurements of surface wave phase velocity anomalies, J. geophys.
Res., 107, doi:10.1029/2000JB000,059.
asymptotic frequency-domain (AKI) and far-field asymptotic time-
Cox, H., 1973. Line array performance when signal coherence is spatially
domain (FRY) approaches, to measure Rayleigh-wave phase dis-
dependent, J. acoust. Soc. Am., 54, 1743–1746.
persion from a year of seismic noise recorded at a dense array of Derode, A., Larose, E., Campillo, M. & Fink, M., 2003. How to estimate
European stations (Verbeke et al. 2012b). The two approaches pro- the Green’s function of a heterogeneous medium between two passive
vide overall consistent results. As shown in Fig. 9, discrepancies are sensors? Application to acoustic waves, Appl. Phys. Lett., 83, 3054–3056.
limited to the lowest frequencies and shortest epicentral distances, Dziewonski, A.M. & Anderson, D.L., 1981. Preliminary reference earth
where the far-field approximation on which the FRY method re- model, Phys. Earth planet. Int., 25, 297–356.
358 L. Boschi et al.

Ekström, G., Abers, G.A. & Webb, S.C., 2009. Determination of surface- Shapiro, N.M., Ritzwoller, M.H. & Bensen, G.D., 2006. Source loca-
wave phase velocities across USArray from noise and Aki’s spectral tion of the 26 sec microseism from cross-correlations of ambient seis-
formulation, Geophys. Res. Lett., 36, doi:10.1029/2009GL039,131. mic noise, Geophys. Res. Lett., 33(L18310), doi:10.1029/2006GL027,
Fry, B., Deschamps, F., Kissling, E., Stehly, L. & Giardini, D., 2010. Layered 010.
azimuthal anisotropy of Rayleigh wave phase velocities in the European Snieder, R., 2004. Extracting the Green’s function from the correlation of
Alpine lithosphere inferred from ambient noise, Earth planet. Sci. Lett., coda waves: a derivation based on stationary phase, Phys. Rev. E, 69,
297, 95–102. doi:10.1103/PhysRevE.69.046,610.
Gouedard, P., Yao, H., Ernst, F. & van der Hilst, R.D., 2012. Surface-wave Stehly, L., Campillo, M. & Shapiro, N.M., 2006. A study of the seismic noise
eikonal tomography for dense geophysical arrays, Geophys. J. Int., 191(2), from its long-range correlation properties, J. geophys. Res., 111(B10306),
781–788. doi:10.1029/2005JB004237.
Harmon, N., Gerstoft, P., Rychert, C.A., Abers, G.A., de la Cruz, M.S. & Stehly, L., Fry, B., Campillo, M., Shapiro, N.M., Guilbert, J., Boschi, L. &
Fischer, K.M., 2008. Phase velocities from seismic noise using beam- Giardini, D., 2009. Tomography of the Alpine region from observations
forming and cross correlation in Costa Rica and Nicaragua, Geophys. of seismic ambient noise, Geophys. J. Int., 178, 338–350.
Res. Lett., 35(L19303), doi:10.1029/2008GL035,387. Tromp, J., Luo, Y., Hanasoge, S. & Peter, D., 2010. Noise cross-correlation
Lin, F.-C. & Ritzwoller, M.H., 2011. Helmholtz surface wave tomography sensitivity kernels, Geophys. J. Int., 183, 791–819.
for isotropic and azimuthally anisotropic structure, Geophys. J. Int., 186, Tsai, V.C., 2009. On establishing the accuracy of noise tomography travel-
1104–1120. time measurements in a realistic medium, Geophys. J. Int., 178, 1555–
Lin, F.-C., Ritzwoller, M.H., Townend, J., Bannister, S. & Savage, M.K., 1564.
2007. Ambient noise Rayleigh wave tomography of new Zealand, Geo- Tsai, V.C., 2011. Understanding the amplitudes of noise correla-
phys. J. Int., 170, 649–666. tion measurements, J. geophys. Res., 116(B09311), doi:10.1029/
Lin, F.-C., Moschetti, M.P. & Ritzwoller, M.H., 2008. Surface wave tomog- 2011JB008,483.
raphy of the western United States from ambient seismic noise: Rayleigh Tsai, V.C. & Moschetti, M.P., 2010. An explicit relationship between time-

Downloaded from http://gji.oxfordjournals.org/ at DTU Library on July 3, 2013

and Love wave phase velocity maps, Geophys. J. Int., 173, 281–298. domain noise correlation and spatial autocorrelation (SPAC) results, Geo-
Meier, T., Dietrich, K., Stockhert, B. & Harjes, H., 2004. One-dimensional phys. J. Int., 182, 454–460.
models of shear wave velocity for the eastern Mediterranean obtained Udı́as, A., 1999. Principles of Seismology, Cambridge University Press,
from the inversion of Rayleigh-wave phase velocities and tectonic impli- Cambridge, UK.
cations, Geophys. J. Int., 156, 45–58. Verbeke, J., Boschi, L., Stehly, L., Kissling, E. & Michelini, A., 2012. High-
Mulargia, F., 2012. The seismic noise wavefield is not diffuse, J. acoust. Soc. resolution Rayleigh-wave velocity maps of central Europe from a dense
Am., 131, 2853–2858. ambient-noise data set, Geophys. J. Int., 188, 1173D–1187.
Nakahara, H., 2006. A systematic study of theoretical relations between spa- Villasenor, A., Yang, Y., Ritzwoller, M.H. & Gallart, J., 2007. Ambi-
tial correlation and Green’s function in one-, two- and three-dimensional ent noise surface wave tomography of the Iberian Peninsula: implica-
random scalar wavefields, Geophys. J. Int., 167, 1097–1105. tions for shallow seismic structure, Geophys. Res. Lett., 34(L11304),
Nishida, K., Kawakatsu, H. & Obara, S., 2008. Three-dimensional doi:10.1029/2007GL030,164.
crustal S wave velocity structure in Japan using microseismic Weemstra, C., Boschi, L., Goertz, A. & Artman, B., 2012. Constraining
data recorded by Hi-net tiltmeters, J. geophys. Res., 113(B10302), seismic attenuation from recordings of ambient noise, Geophysics, in
doi:10.1029/2007JB005,395. press.
Paige, C. C. & Saunders, M.A., 1982. LSQR—an algorithm for sparse Yang, Y. & Ritzwoller, M.H., 2008. Characteristics of ambient seismic noise
linear-equations and sparse least-squares, ACM Trans. Math. Software, 8, as a source for surface wave tomography, Geochem., Geophys., Geosyst.,
43–71. 9, Q02008, doi:10.1029/2007GC001,814.
Peter, D., Tape, C., Boschi, L. & Woodhouse, J.H., 2007. Surface wave Yao, H. & van der Hilst, R.D., 2009. Analysis of ambient noise en-
tomography: global membrane waves and adjoint methods, Geophys. J. ergy distribution and phase velocity bias in ambient noise tomog-
Int., 171, 1098–1117. raphy, with application to SE Tibet, Geophys. J. Int., 179, 1113–
Ritzwoller, N.M., Shapiro, M.H., Levshin, A.L. & Leahy, G.M., 2001. 1132.
Crustal and upper mantle structure beneath Antarctica and surrounding Yao, H., van der Hilst, R.D. & de Hoop, M.V., 2006. Surface-wave ar-
oceans, J. geophys. Res., 106, 30 645–30 670. ray tomography in SE Tibet from ambient seismic noise and two-
Seats, K.J., Lawrence, J.F. & Prieto, G.A., 2012. Improved ambient noise station analysis—I. Phase velocity maps, Geophys. J. Int., 166, 732–
correlation functions using Welch’s method, Geophys. J. Int., 188, 513– 744.
523. Yokoi, T. & Margaryan, S., 2008. Consistency of the spatial autocorrela-
Shapiro, N.M. & Campillo, M., 2004. Emergence of broadband Rayleigh tion method with seismic interferometry and its consequence, Geophys.
waves from correlations of the ambient seismic noise, Geophys. Res. Lett., Prospect., 56, 435–451.
31(L07614), doi:10.1029/2004GL019491. Zheng, Y., Shen, W., Zhou, L., Yang, Y., Xie, Z. & Ritzwoller, M.H., 2011.
Shapiro, N.M., Campillo, M., Stehly, L. & Ritzwoller, M.H., 2005. High- Crust and uppermost mantle beneath the North China Craton, northeastern
resolution surface-wave tomography from ambient seismic noise, Science, China, and the Sea of Japan from ambient noise tomography, J. geophys.
307, 1615–1618. Res., 116(B12312), doi:10.1029/2011JB008,637.