See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/235623159

Single station determination of Rayleigh wave ellipticity by using the random

decrement technique (RayDec)

Article  in  Geophysical Research Letters · July 2009

DOI: 10.1029/2009GL038863

CITATIONS READS

70 737

4 authors:

Manuel Hobiger Pierre-Yves Bard

ETH Zurich Université Gustave Eiffel
39 PUBLICATIONS   468 CITATIONS    447 PUBLICATIONS   11,906 CITATIONS   

SEE PROFILE SEE PROFILE

Cécile Cornou Nicolas Le Bihan

University Joseph Fourier - Grenoble 1 French National Centre for Scientific Research
157 PUBLICATIONS   3,160 CITATIONS    115 PUBLICATIONS   2,075 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

SINAPS@ View project

ANR EXAMIN (https://anr-examin.brgm.fr) View project

All content following this page was uploaded by Manuel Hobiger on 17 February 2015.

The user has requested enhancement of the downloaded file.

 GEOPHYSICAL RESEARCH LETTERS, VOL. 36, L14303, doi:10.1029/2009GL038863, 2009

Single station determination of Rayleigh wave ellipticity by using

the random decrement technique (RayDec)
M. Hobiger,1 P.-Y. Bard,1 C. Cornou,1 and N. Le Bihan2
Received 24 April 2009; revised 13 June 2009; accepted 22 June 2009; published 28 July 2009.
[1] Rayleigh wave ellipticity as a function of frequency is types has to be accounted for in inversion algorithms.
closely linked to underground structure, i.e., shear wave Therefore, the proportion between Rayleigh and Love
velocity profile and sediment thickness. The possibility to waves in the noise wave field has to be inferred or
calculate these underground properties by inverting assumed prior to the inversion [Scherbaum et al., 2003;
ellipticity curves has recently been shown. We propose a Arai and Tokimatsu, 2004; Köhler et al., 2006]. Extrac-
new technique enabling the Rayleigh wave ellipticity to be tion of Rayleigh wave ellipticity and its subsequent
recovered over a wide frequency range by using ambient inversion [Boore and Toksöz, 1969; Fäh et al., 2001;
noise recordings. Based on the random decrement technique Malischewsky and Scherbaum, 2004] would avoid making
commonly used to characterize dynamic parameters of such assumptions.
buildings, this method eliminates all wave types except [3] We propose a new method (RayDec) to estimate the
Rayleigh waves. We apply the method to noise synthetics Rayleigh wave ellipticity spectrum by using the random
simulated for different underground structures and show its decrement technique [Asmussen, 1997]. A usual application
applicability to real seismic noise data. Citation: Hobiger, of the random decrement technique is the measurement
M., P.-Y. Bard, C. Cornou, and N. Le Bihan (2009), Single station of resonance frequencies and damping parameters of
determination of Rayleigh wave ellipticity by using the random buildings [Dunand, 2005; Michel et al., 2008]. We propose
decrement technique (RayDec), Geophys. Res. Lett., 36, L14303, to apply the random decrement technique to three-
doi:10.1029/2009GL038863. component seismometer records of seismic noise. By using
the vertical component as a master trigger and stacking a
1. Introduction large number of horizontal and vertical signals, Rayleigh
waves will be emphasized with respect to Love and body
[2] In order to properly evaluate local site amplification it waves. Therefore, the obtained ellipticity curve will be closer
is important to have a good knowledge of the local soil to the true curve than the H/V curve. In this paper, we will
properties, i.e., shear-wave velocities and sediment thick- present the method, apply it to synthetic seismic noise and
ness. These soil properties can be obtained by inverting finally show its applicability to real seismic noise records
dispersion curves. One way is to use the dispersion curves with one example.
of surface waves, as derived from active or passive seismic
array measurements. However, it usually provides reliable
information at frequencies higher than the resonance fre- 2. Methodology
quency and the inverted velocity profiles do not carry [4] Assume that a three-component sensor record con-
information on the deeper structures. Additional informa- sists of three time-series of seismic noise (e(t) in east-west
tion around the resonance frequency can be added by using direction, n(t) in north-south direction and v(t) in vertical
H/V spectra. Recent studies have shown the possibility to direction). Each of these time-series has a time length T
evaluate the S-wave velocity profile by inversion of H/V and consists of N data points. In order to extract the
spectra [Fäh et al., 2003; Arai and Tokimatsu, 2004] or joint ellipticity of Rayleigh waves we suggest the following
inversion of H/V spectra and dispersion curves [Arai and method, based on the random decrement technique
Tokimatsu, 2005; Picozzi et al., 2005]. The classical H/V [Asmussen, 1997]. The principles of the method are to
technique [Nakamura, 1989] calculates the ratio between emphasize Rayleigh waves with respect to other wave
the horizontal and the vertical spectrum of ambient vibra- types by summing a large number of specially tuned signal
tions recordedpffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
on a single three-component seismic sensor windows, calculate the energy on the vertical and horizon-
jEð f Þj2 þjN ð f Þj2 tal summed signals and henceforward estimate Rayleigh
by H/V( f ) = pffiffi where E( f ), N( f ) and Z( f ) are
2jZð f Þj wave ellipticity.
the spectra of the east – west, north – south and vertical [5] The first step consists of filtering the signals by using
displacements of the sensor. Both Rayleigh and Love a narrow-band Chebyshev filter of order 4 and bandwidth
waves contribute to the H/V spectrum [Bonnefoy-Claudet df centered on frequency f. We use the Chebyshev filter
et al., 2008] and the presence of both surface wave because of its sharpness. Filtering was performed on a
linear frequency scale. The resulting signals are called
ef (t), nf (t) and vf (t).
1
[6] Then we search for all times t i where the signal on
Laboratoire de Géophysique Interne et Tectonophysique, IRD, LCPC,
CNRS, Université Joseph Fourier, Grenoble, France.
the vertical component changes its sign from negative to
2
GIPSA-Lab, CNRS, Grenoble, France. positive (vf (t i)  0, vf (t i + dt) > 0). Every time this
condition is satisfied, we store signals of length D on all
Copyright 2009 by the American Geophysical Union. three components, leading to buffered signals vf,b,i (t),
0094-8276/09/2009GL038863

L14303 1 of 5
L14303 HOBIGER ET AL.: RAYDEC L14303

ef,b,i (t) and nf,b,i (t) defined in terms of the original signals [8] The signals vf,b,i (t) and hf,b,i (t) are weighted by
as (for 0  t  D): c2f,b,i and summed for all times t i to obtain two buffered
signal windows of length D:
vf ; b; i ðt Þ ¼ vf ðt i þ tÞ; X
  vf ; S ðt Þ ¼ c2f ; b; i vf ; b; i ðt Þ
1 i
ef ; b; i ðt Þ ¼ ef t i  þt ; ð1Þ X
4f hf ; S ðt Þ ¼ c2f ; b; i hf ; b; i ðt Þ: ð6Þ
 
1 i
nf ; b; i ðt Þ ¼ nf t i  þt :
4f
[9] Finally, the ellipticity  is calculated as the square root
The shift of 1
on the eastern and northern components of the ratio of the energies in the buffered signal windows
4f hf,S (t) and vf,S (t):
accounts for the 90° phase shift between the vertical and
horizontal components of Rayleigh waves. Then we build a vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uR D
2
horizontal signal hf,b,i (t) by projecting the east-west and u hf ; S ðt Þ dt
north-south components onto an axis forming an azimuth  ¼ tR0D
2 : ð7Þ
angle Ji with the northern direction by 0 vf ; S ðt Þ dt

By repeated filtering of the signal over the entire frequency

hf ; b; i ðt Þ ¼ sinðJi Þ ef ; b; i ðt Þ þ cosðJi Þ nf ; b; i ðt Þ: ð2Þ range and applying the presented method the ellipticity of
Rayleigh waves as a function of frequency is determined.
The angle Ji is chosen in such a way that it maximizes the The free parameters of the method are the width of the
correlation between the vertical signal vf,b,i (t) and the frequency filter df and the length of the buffered signal D.
horizontal signal hf,b,i (t) given by: [10] It could be supposed that Love and body waves
influence the results. In an assumed isotropic medium,
Love waves are not present on the vertical component and
Z D their horizontal movements are thus uncorrelated with the
Cf ; b; i ðJi Þ ¼ vf ; b; i ðt Þ hf ; b; i ðt Þdt vertical signal. Although incidental correlations of Love
0
Z D and Rayleigh waves arriving at the same time cannot be
Cf ; b; i ðJi Þ ¼ sinðJi Þ vf ; b; i ðtÞ ef ; b; i ðt Þdt excluded, such effects will rapidly lose importance because
0
Z of stacking and can thus be neglected. Arriving P- and
D
S-waves can be present on the vertical and horizontal
þ cosðJi Þ vf ; b; i ðt Þ nf ; b; i ðt Þdt: ð3Þ
0 components, but as their movements are not phase delayed,
they will also disappear due to averaging over a sufficiently
large number of windows.
The correlation is maximized by

RD ! 3. Application to Synthetic Noise

vf ; b; i ðt Þ ef ; b; i ðtÞdt [11] In order to test the method, we used a set of synthetic
Ji ¼ tan1 0
RD : ð4Þ
0 vf ; b; i ðt Þ nf ; b; i ðtÞdt noise simulated for the models N101, N102, N103 and
N104 of the ESG 2006 benchmark [Cornou et al., 2009].
The compressional (P) and shear wave (S) velocity profiles
There are two possible Ji 2 [0°, 360°[ satisfying equation (4), of these models are displayed in Figure 1.
with Ji,2 = Ji,1 + 180°, one of them leading to a negative, the [12] The noise synthetics were obtained in the following
other to a positive correlation. The Ji leading to a positive way: Noise sources were approximated by forces of random
correlation is chosen. The method does not distinguish orientation and amplitude located 0.5 m below the surface
between prograde and retrograde particle motion as prograde [Moczo and Kristek, 2002]. The forces are randomly dis-
Rayleigh waves are treated as retrograde waves arriving from tributed in time and space. The time function of each source
the opposite azimuth. is a delta-like signal with a flat Fourier frequency-amplitude
[7] Although the chosen angle Ji maximizes the cor- spectrum between 0.1 and 20 Hz. In this frequency range,
relation between vf,b,i (t) and hf,b,i (t), the correlation can be the associated wave field has been computed by using the
low. Thus, we make use of the correlation factor defined wavenumber-based technique for horizontally layered struc-
as tures proposed by Hisada [1994, 1995]. All types of surface
and body waves are included in the simulations.
RD
0 vf ; b; i ðt Þ hf ; b; i ðt Þdt 3.1. Influence of the Parameters of the Method
cf ; b; i ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
RD 2 RD 2 : ð5Þ
v ðt Þdt h ðt Þdt [13] As mentioned above, there are two free parameters
0 f ; b; i 0 f ; b; i
that can be adjusted to improve the performance of the
method: df, the width of the frequency filter, and D, the
In the case of perfect correlation between vf,b,i (t) and length of the buffered signal. To investigate the optimal
hf,b,i (t) this factor equals 1. On the other side, it will parameters, we used 30 minutes of noise simulated for the
vanish when the vertical and horizontal signals are not model N102. The noise sources are located in the far field as
correlated at all. well as in the local field.

2 of 5
L14303 HOBIGER ET AL.: RAYDEC L14303

Figure 1. S- and P-wave velocity profiles for the models N101, N102, N103 and N104 [Cornou et al., 2009] and the
Colfiorito borehole [Di Giulio et al., 2006].

[14] Figure 2a shows the calculated ellipticity for differ- curves are very close. This illustrates the temporal stability
ent bandwidths df of the frequency filter compared to the of the results.
theoretical ellipticity of the model. The used bandwidths are
proportional to the central frequencies. Too weak filtering 3.3. Results for the Models N101, N103 and N104
(df = 0.5f ) as well as too strong filtering (df = 0.1f ) do not [18] RayDec has been applied to synthetic noise simu-
give satisfying results. An intermediate setting of 0.2f lated for different models. The results of these simulations
reproduces the real curve the best. Therefore, we fixed the are shown in Figure 3. For model N101 (Figure 3a), the
bandwidth to 0.2f in the following. curve is close to the real curve over the whole frequency
[15] Figure 2b shows the resulting ellipticity for different range, although the peak ellipticity is overestimated. For
lengths D of the buffered signal windows in comparison to this model layout, reflected body waves can deteriorate the
the theoretical curve. We fix D as a function of frequency, results. Therefore, more signal would be needed to obtain a
so that the signals consist of the same number of periods at more robust ellipticity curve. For model N103 (Figure 3b),
all frequencies. As can be seen in Figure 2b, at least ten
cycles are necessary for a good coverage of the signal
whereas further increasing of the window length does not
improve the results. Therefore, D = 10/f is used in the
following.
3.2. Minimum Required Signal Length and Temporal
Stability of the Results
[16] The required signal length is illustrated in Figure 2c.
The same signals as in subsection 3.1 were used and the first
1800, 600, 300 and 120 seconds of signal were analyzed.
The results of the analysis for signals of 1800, 600 and
300 seconds only slightly differ. However, the curve obtained
for 120 seconds of signal is less smooth than the other curves.
At low frequencies, the signal statistics deteriorate due to the
smaller number of buffered windows. It can be deduced that -
for sites with resonance frequencies around 1 Hz - a signal
length of five minutes can be sufficient to measure an
accurate ellipticity curve, as long as enough Rayleigh waves Figure 2. Sensitivity of the results with respect to the
are present in the signal. For smaller underground resonance parameters of the method demonstrated on simulated seismic
frequencies, the measurement has to be longer. Indeed, longer noise for the N102 model. The real ellipticity curve of the
measurements improve the statistical features and the reli- model is included for comparison. Curves of Figures 2a
ability of the results at all frequencies. and 2b were calculated for 30 minutes of simulated seismic
[17] The temporal stability of the obtained ellipticity noise. (a) Effect of the filter bandwidth df on the resulting
curves is shown in Figure 2d. Five different synthetic ellipticity curve (D = 10/f) for df = 0.5f, df = 0.3f, df = 0.2f
signals of ten minutes each were simulated and the ellip- and df = 0.1f. (b) Effect of the length D of the buffered
ticity curves were calculated for each window. The overall signal (df = 0.2f) for D = 1/f, D = 5/f, D = 10/f and D = 20/f.
behavior of the curves is identical. Only the fourth ten- (c) Effect of the length of the signals (time length T,
minute-window shows some instabilities between 0.1 and df = 0.2f, D = 10/f ) for T = 1800s, T = 600s, T = 300s
0.2 Hz emphasizing the need for longer signal lengths at low and T = 120s. (d) Analysis of five different signals of
frequencies. For the other frequencies, the five ellipticity ten minutes each (df = 0.2f, D = 10/f ).

3 of 5
L14303 HOBIGER ET AL.: RAYDEC L14303

Colfiorito basin in central Italy. Colfiorito basin is a well-

known site which has been extensively investigated by
different geophysical methods [Di Giulio et al., 2006].
The theoretical ellipticity curve is calculated on the basis
of S- and P-wave velocity profiles obtained by a borehole
measurement (see Figure 1). The parameters used were
df = 0.2f and D = 10/f. The hour of signal was cut into six
ten-minute signals which were analyzed separately. The
resulting curves were then averaged. When comparing with
the theoretical ellipticity curve, one should keep in mind
that a borehole measurement is made at a selected point,
whereas a seismic noise record is influenced by the whole
surrounding underground structure. The RayDec curve is in
better agreement with the theoretical curve than the H/V
curve, especially on the slopes of the fundamental resonance
frequency peak (Figure 3d). However, both the H/V and
Figure 3. Ellipticity curves obtained for ten minutes of RayDec curve’s peak frequencies are shifted with respect to
synthetic noise simulated for three different underground the borehole model’s peak frequency.
models ((a) N101, (b) N103 and (c) N104) and for one hour
of real noise from Colfiorito basin (d). The used parameters 5. Conclusion
are df = 0.2f and D = 10/f. In the four graphs, the results are [21] RayDec is a fast and simple method to determine the
compared to the real ellipticity curves of the fundamental ellipticity of Rayleigh waves. As it suppresses Love and
mode and the first harmonic mode (only in Figure 3c) and to body waves efficiently, its results are closer to theory than
the results obtained by the H/V technique. (d) Curve also H/V results. The necessary signal length depends on the
contains the error ranges of the H/V and RayDec curves. resonance frequency of the underground structure, but some
minutes of signal can be sufficient to get satisfying results,
as long as enough Rayleigh waves are present in the wave
there is a good agreement between the simulated and the field, although longer signal lengths are preferable. How-
theoretical curves around the resonance frequency and at ever, as it is a one-sensor technique, RayDec gives two
higher frequencies, with some minor deviations between 0.3 possible azimuths and cannot distinguish between prograde
and 2 Hz. Below 0.15 Hz, the curve differs considerably and retrograde particle motion. By coupling RayDec with
from the theoretical curve. Model N104 (Figure 3c) exhibits array measurements the correct azimuth and the sense of
some unfavorable behavior, as the fundamental mode and particle motion could be retrieved. Besides that, it could be
the first harmonic mode of Rayleigh wave motion overlap envisaged to estimate the proportion of Rayleigh waves in
above 1.4 Hz. RayDec cannot distinguish between two the wave field by comparing RayDec results including only
modes. Therefore, its results are a mixture of the two modes Rayleigh waves with H/V measurements including all wave
in this frequency range. For lower frequencies, RayDec types. The future work consists of trying to improve the
reproduces well the theoretical curve around the first peak inversion of S-wave velocity profiles by using RayDec’s
of the fundamental mode. The H/V curve follows the same ellipticity estimations. This is a non-trivial endeavor which
trend, but the absolute values of the RayDec curve better will be addressed in a future paper.
reproduce the right flank of the fundamental peak.
[19] It can be stated that, in general, the results of RayDec [22] Acknowledgments. Noise synthetics were performed at the
are much closer to the theoretical curves than the results Service Commun de Calcul Intensif (SCCI) of Grenoble observatory,
obtained with the H/V method. However, RayDec regularly France (OSUG). Ambient noise data from Colfiorito site were collected
during the SESAME project (EVG1-CT-2000-00026). This work is sup-
fails in reproducing very small ellipticity values. At these ported by the NERIES European project and the ASIRAN ANR project. We
points, the horizontal components of the Rayleigh wave would like to thank two anonymous reviewers whose comments helped to
vanish. Therefore, it is problematic to find horizontal signals improve the manuscript.
correlated to the vertical signal. The same reasoning is
valid for the peak frequency ellipticity in the case of a References
theoretically vanishing vertical component. This explains Arai, H., and K. Tokimatsu (2004), S-wave velocity profiling by inversion
why the peak ellipticity cannot be satisfactorily estimated. of microtremor H/V spectrum, Bull. Seismol. Soc. Am., 94, 53 – 63.
On the other hand, small ellipticities can also be hidden by Arai, H., and K. Tokimatsu (2005), S-wave velocity profiling by joint
the presence of higher Rayleigh wave modes. In cases inversion of microtremor dispersion curve and horizontal-to-vertical
(H/V) spectrum, Bull. Seismol. Soc. Am., 95, 1766 – 1778, doi:10.1785/
where higher Rayleigh wave modes are close to the funda- 0120040243.
mental modes, RayDec only indicates an apparent ellipticity Asmussen, J. C. (1997), Modal analysis based on the random decrement
which cannot be simply related with the true ellipticity of technique, Ph.D. thesis, Univ. of Aalborg, Aalborg, Denmark.
Bonnefoy-Claudet, S., A. Köhler, C. Cornou, M. Wathelet, and P.-Y. Bard
each mode. (2008), Effects of Love waves on microtremor H/V ratio, Bull. Seismol.
Soc. Am., 98, 288 – 300, doi:10.1785/0120070063.
Boore, D. M., and M. N. Toksöz (1969), Rayleigh wave particle motion and
4. Application to Real Noise Data crustal structure, Bull. Seismol. Soc. Am., 59, 331 – 346.
[20] To test the RayDec method on real data, we applied Cornou, C., M. Ohrnberger, D. M. Boore, K. Kudo, and P.-Y. Bard
(2009), Derivation of structural models from ambient vibration array
it to one hour of seismic noise recorded in the center of recordings: Results from an international blind test, in ESG 2006:

4 of 5
 L14303 HOBIGER ET AL.: RAYDEC L14303

Third International Symposium on the Effects of Surface Geology on of Surface Geology on Seismic Motion, vol. 1, edited by P.-Y. Bard et al.,
Seismic Motion, vol. 2, edited by P.-Y. Bard et al., Lab. Cent. des pp. 351 – 360, Lab. Cent. des Ponts et Chaussées, Paris.
Ponts et Chaussées, Paris, in press. Malischewsky, P. G., and F. Scherbaum (2004), Love’s formula and H/V-
Di Giulio, G., C. Cornou, M. Ohrnberger, M. Wathelet, and A. Rovelli ratio (ellipticity) of Rayleigh waves, Wave Motion, 40, 57 – 67.
(2006), Deriving wavefield characteristics and shear-velocity profiles Michel, C., P. Guéguen, and P.-Y. Bard (2008), Dynamic parameters of
from two-dimensional small-aperture arrays analysis of ambient vibrations structures extracted from ambient vibration measurements: An aid for
in a small-size alluvial basin, Colfiorito, Italy, Bull. Seismol. Soc. Am., 96, the seismic vulnerability assessment of existing buildings in moderate
1915 – 1933, doi:10.1785/0120060119. seismic hazard regions, Soil Dyn. Earthquake Eng., 28, 593 – 604.
Dunand, F. (2005), Pertinence du bruit de fond sismique pour la caractér- Moczo, P., and J. Kristek (2002), FD code to generate noise synthetics,
isation dynamique et l’aide au diagnostic sismique des structures de génie Sesame Rep. D02.09, IGSAS, Bratislava. (Available at http://SESAME-
civil, Ph.D. thesis, Univ. Joseph Fourier, Grenoble, France. FP5.obs.ujf-grenoble.fr)
Fäh, D., F. Kind, and D. Giardini (2001), A theoretical investigation of Nakamura, Y. (1989), A method for dynamic characteristics estimation of
average H/V ratios, Geophys. J. Int., 145, 535 – 549. subsurface using microtremor on the ground surface, Q. Rep. RTRI, 30,
Fäh, D., F. Kind, and D. Giardini (2003), Inversion of local S-wave velocity 25 – 33.
structures from average H/V ratios, and their use for the estimation of Picozzi, M., S. Parolai, and S. M. Richwalski (2005), Joint inversion of H/V
site-effects, J. Seismol., 7, 449 – 467. ratios and dispersion curves from seismic noise: Estimating the S-wave
Hisada, Y. (1994), An efficient method for computing Green’s functions for velocity of bedrock, Geophys. Res. Lett., 32, L11308, doi:10.1029/
a layered halfspace with sources and receivers at close depths, Bull. 2005GL022878.
Seismol. Soc. Am., 84, 1456 – 1472. Scherbaum, F., F. K. Hinzen, and M. Ohrnberger (2003), Determination of
Hisada, Y. (1995), An efficient method for computing Green’s functions for shallow shear wave velocity profiles in the Cologne, Germany area using
a layered halfspace with sources and receivers at close depths (part 2), ambient vibrations, Geophys. J. Int., 152, 597 – 612.
Bull. Seismol. Soc. Am., 85, 1080 – 1093.
Köhler, A., M. Ohrnberger, and F. Scherbaum (2006), The relative fraction 
of Rayleigh and Love waves in ambient vibration wavefields at different P.-Y. Bard, C. Cornou, and M. Hobiger, LGIT, BP 53, F-38041 Grenoble
European sites, ESG 2006: Third International Symposium on the Effects CEDEX 9, France. (manuel.hobiger@obs.ujf-grenoble.fr)
N. Le Bihan, GIPSA-Lab, BP 46, F-38402 Grenoble CEDEX, France.

5 of 5

View publication stats