Sunteți pe pagina 1din 19

Engineering Geology 182 (2014) 182200

Contents lists available at ScienceDirect

Engineering Geology
journal homepage: www.elsevier.com/locate/enggeo

What we can learn about slope response to earthquakes from ambient


noise analysis: An overview
Vincenzo Del Gaudio a,, Sandro Muscillo a, Janusz Wasowski b
a
b

Dipartimento di Scienze della Terra e Geoambientali, Universit degli Studi di Bari Aldo Moro, Italy
Istituto di Ricerca per la Protezione Idrogeologica, Consiglio Nazionale delle Ricerche, Bari, Italy

a r t i c l e

i n f o

Article history:
Accepted 17 May 2014
Available online 29 May 2014
Keywords:
Landslides
Earthquakes
Site amplication
Ambient noise
Nakamura's method
Cross-correlation analysis

a b s t r a c t
Earthquake induced slope failures are responsible for a signicant amount of life loss and damage, and their
effective mitigation requires further advancements in our comprehension of slope behaviour under seismic
shaking. One source of uncertainty in seismic landslide susceptibility assessment is the phenomenon of enhanced
amplication of ground motion along down slope directions. This implies a strength demand beyond that estimated by standard slope stability analysis. An extensive accelerometer monitoring of slope dynamic response
in areas exposed to seismic landslide hazard is unfeasible. An alternative approach can take advantage of recent
development of reconnaissance techniques based on the analysis of ambient noise recorded by portable instruments. The most popular technique, known as Nakamura or HVNR method, consists in analysing H/V spectral ratios between Horizontal and Vertical components of Noise Recording, and allows the recognition of site
resonance frequencies. The application of HVNR to complex site conditions typical of marginally stable slopes
is often difcult and requires the development of ad hoc procedures both for acquisition and analysis of
noise recording. Tests in different geologic and geomorphic settings show that an analysis of azimuthal variation
of spectral ratios can reveal the presence and orientation of directional resonance, whereas the recognition of
main resonance frequencies requires a proper selection of signals to be analysed. Efforts to evaluate amplication
factors currently rely on numerical simulations, which in turn require S-wave velocity of slope materials. Ambient noise analysis in terms of velocity models can contribute through the inversion of H/V spectral ratios and surface wave velocity dispersion curves derived from the processing of multiple simultaneous noise recordings.
However these applications require a correct identication of the nature of surface waves present in the noise
recording.
2014 Elsevier B.V. All rights reserved.

1. Introduction
In the strong earthquake scenario, widespread slope failures represent not only a potential source of life loss and costly damages, but
also a threat to road and lifeline networks essential for an effective
emergency management. Seismically triggered landslides can cause additional collateral hazards, e.g. disastrous ooding resulting from river
damming. For example, the moment magnitude (Mw) 7.9 Wenchuan
earthquake of 12 May 2008 induced over 60,000 landslides (Gorum
et al., 2011); these were directly responsible for about 20,000 victims,
caused extensive damages to irrigation channels, and interrupted highways and bridges, thus isolating several towns (Tang et al., 2011). The
event generated over 500 barrier lakes which threatened people living
downstream (Fan et al., 2012). Therefore, civil protection actions
aimed at mitigating earthquake damage and at increasing preparedness
need to focus on wide-area evaluations of slope response to strong
Corresponding author.
E-mail address: vincenzo.delgaudio@uniba.it (V. Del Gaudio).

http://dx.doi.org/10.1016/j.enggeo.2014.05.010
0013-7952/ 2014 Elsevier B.V. All rights reserved.

shaking that can be expected under foreseeable seismic hazard


scenarios.
The prediction of slope behaviour during future strong earthquakes
is made difcult by the complexity of the amplication phenomena related to a combination of topographic and soil/bedrock stratigraphic effects. Topographic effects were invoked by some authors to explain
anomalous concentration of landslides triggered by earthquakes near
ridge crests (e.g. Harp et al., 1981; Harp and Jibson, 2002; Seplveda
et al., 2005). Furthermore, numerical modelling indicated the potential
destabilising role of topographic amplication (Meunier et al., 2008;
Lenti and Martino, 2012). Numerical simulations also provided indications that impedance contrast between surface material and more
rigid substratum may cause seismic amplication effects and favour
slope failures or landslide reactivation (e.g. Bourdeau and Havenith,
2008; Bozzano et al., 2008).
Instrumental evidence of amplication affecting landslide prone
slopes have recently been reported in several studies (Del Gaudio and
Wasowski, 2007, 2011; Gallipoli and Mucciarelli, 2007; Garambois
et al., 2010; Moore et al., 2011), which showed that ground motion

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

amplication at site-specic frequencies can present directional maxima close to potential sliding directions (e.g. maximum slope). This
implies greater slope susceptibility to seismic failure with respect to
that estimated by standard slope stability analysis.
An example of ground motion amplication is illustrated in Fig. 1,
which shows the comparison of horizontal accelerations recorded during the 6 April 2009 L'Aquila earthquake mainshock (Mw 6.3) in the
area of Caramanico Terme (Abruzzo, central Italy), located about
60 km SE of the epicentre. The recordings were acquired at three sites,
one located on the head of a pre-existing landslide in colluvial materials
(CAR2), the other on the mudstone substratum outcrop (CAR1), distant
some 600 m from CAR2, and the third at a reference station on the
nearly at limestone outcrop (CAR4). The analysis showed that, while
on the mudstone slope acceleration was amplied by a factor of two
in comparison to the reference site, a signicantly larger amplication
affected the landslide slope. Furthermore, the landslide site was
characterised by a pronounced directional peak along a direction close
to that of maximum slope (ENEWSW). A similar observation was reported by Burjnek et al. (2010) who analysed recordings of small
earthquakes at the Randa rock slope (Switzerland). They found a maximum amplication of ground motion on an instable part of the slope,
oriented approximately along the steepest slope (Figure 2).
A comprehensive investigation of seismic landslide hazard would require a long term and diffuse accelerometer monitoring of earthquakeprone regions. However, such widespread monitoring of slopes cannot
be afforded. Although much desirable, the recordings of actual strong
motions affecting slopes are rather few and generally limited to the aftershock phases (Wasowski et al., 2011). The development and application
of quick and cost-effective reconnaissance techniques based on the analysis of ambient noise can represent a possible solution. One useful reconnaissance technique is the HVNR (horizontal-to-vertical noise spectral
ratio) or Nakamura method (Nogoshi and Igarashi, 1971; Nakamura,
1989). It consists of analysing H/V spectral ratios between horizontal
and vertical components of noise recording acquired for a few tens of
minutes. For each component, Fourier spectra are calculated on several
time windows of few tens of seconds extracted from the recording.
Then, after a smoothing, an average of spectral ratios between horizontal
and vertical components is derived for all the time windows (for more
details see the guidelines reported by Bard, 2004).
The HVNR technique is based on the assumption that a strong impedance contrast between a surface soft layer and a more rigid substratum causes an amplication of the horizontal components of noise

183

ground motion at the same frequencies at which the shear wave amplication reaches the maximum. Thus, site resonance conditions can be
revealed by nding a pronounced peak in the H/V spectral ratios around
a site specic frequency.
Although the method was originally devised to investigate at and
horizontally layered sites, it proved capable to reveal resonance properties also in the more complex settings typical of unstable or marginally
stable slopes (e.g. Gallipoli et al., 2000; Havenith et al., 2002; Mric et al.,
2007; Danneels et al., 2008; Jongmans et al., 2009). In particular, it was
found that an analysis of azimuthal variation of H/V ratios can reveal the
presence of directional resonance phenomena (Del Gaudio et al., 2008;
Burjnek et al., 2010).
Moreover, ambient noise data offer also the possibility to obtain relevant information on surface material properties to support evaluation
of slope behaviour under seismic shaking. In particular, S-wave velocities can be obtained by deriving surface wave velocity dispersion curves
from the processing of multiple simultaneous noise recordings acquired
by a geophone array (e.g. Louie, 2001; Ohori et al., 2002) or by a few
portable seismometers; in the latter case one can follow a local scale application of a correlation analysis technique developed for a broad range
of distances (cf. Nunziata et al., 2009).
In this paper we provide an overview focused on the information
that can be obtained from ambient noise analysis for the characterisation of slope stability and slope dynamic response during earthquakes.
This is done by considering our most recent experiences, as well as studies published by other workers. We rst examine the properties of seismic noise to establish what part of noise signal can be usefully exploited
when investigating landslide prone slopes, and also provide an overview of the instruments that can be used in eld measurements. Then
we report some recent results obtained from the application of different
techniques of noise analysis, discuss their potential and limits, and offer
some practical implementation guidelines.
2. Ambient noise properties and implications for studying landslide
prone slopes
2.1. Frequency range
Ambient noise consists of ground vibrations observed in a wide frequency range, non induced by seismic events. Signals at frequencies
above 1 Hz are commonly indicated as microtremors and mainly consist of a cultural noise generated by human activities (e.g. car trafc,

Fig. 1. Comparison of horizontal accelerations recorded at three accelerometer stations in Caramanico Terme during the 2009 L'Aquila Mw 6.3 earthquake mainshock (left) and map showing their location and local geology (right). Explanation: LmMiocene and older age stratied limestone; Meevaporitic succession including clayeysiltysandy sediments with evaporitic limestone breccias and gypsum deposits (Messinian); MpPliocene age mudstone; Bqcemented carbonate megabreccia (Late Villafranchian); Sqhcolluvial deposit (Quaternary);
numbers from 1 to 5 indicate locations of accelerometer stations CAR1, CAR2, etc.; dashed lines show limits of major landslides, including slope failures occurred in 1989 and 1627 marked
by letters a and b, respectively. CAR2 is sited on the head of the 1989 landslide that mobilised thick Quaternary colluvium overlying Pliocene mudstone (Mp) on which CAR1 is located;
CAR4 is a reference station located on Miocene limestone bedrock.

184

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Fig. 2. Map of spectral amplication (estimated as spectral ratio in comparison to a reference site) at the Randa rock slide slope at three frequencies for ground motion components oriented according to the black arrows. Black dots represent the location of seismic sensors and black and yellow lines indicate, respectively, the instability boundary and main discontinuities
(modied from Burjnek et al., 2010).

trains, machinery at work, industrial plants). Lower frequency components are produced by natural sources, with oceanic waves and largescale meteorological conditions acting at frequencies below 0.5 Hz and
winds and local meteorological conditions being responsible for signals
from 1 to few Hz (Bonnefoy-Claudet et al., 2006).
The rst question to face when planning ambient noise analysis is
the frequency range that can provide information useful for investigating landslide prone slopes. Special attention should be paid to the lower
limit of the frequency band, as this can guide the choice of the recording
instruments. Noise recordings should enable extraction of information
about site resonance frequencies and velocity dispersion curves down
to frequencies reecting S-wave velocities at the maximum depth of
interest. Thus, the frequency to be analysed depends mainly on the
ratio between the S-wave velocity Vs of the surface soft layer potentially
susceptible to mobilisation and its thickness H. If the layer's lateral extension is of the order of ten times greater or more than its thickness,
the resonance frequencies fn can be approximated to that of a laterally
innite layer, i.e.
fn

2n 1  V S
4H

1

where n is an integer representing the vibration mode. For n = 0, one


obtains the main resonance frequency corresponding to the fundamental mode
fo

VS
4H

2

According to the guidelines provided for the HVNR method by the


SESAME project (Bard, 2004), a good denition of H/V ratio peak requires an inspection of H/V curve down to a frequency equal to 1/4 of
fo; thus it is desirable to analyse frequencies down to 1/16 of the Vs/H
ratio.
With regard to the survey methods based on the analysis of the dispersion curve of surface wave velocities, the investigation depth can be

indicated from the 1/2 value of the longest wavelength analysed. Therefore, to obtain information on Vs values down to the base of the surface
soft layer, one should analyse frequencies at least down to Vs/2H,
although an extension of measurements down to 1/2 of such frequency
would be desirable to better constrain the substratum velocity. The identication of the site resonance frequency requires the analysis down to
frequencies around 1/16 of Vs/H, and this frequency requirements of
the HVNR measurements represents the main constraint on the lower
limit of the frequency range that can be investigated in ambient noise
analysis.
Considering the likely values of the Vs/H ratio in landslide prone
slopes, in most cases the analysis of microtremor frequencies can be sufcient. However there could be the need to extend observations below
1 Hz, especially in the case of very large slope failures, e.g. megalandslides like the Tsaoling landslide mobilised by the Mw = 7.6 ChiChi earthquake of 20 September 1999, characterised by a thickness up
to 180 m (Chigira et al., 2003).
Ambient noise frequencies below 1 Hz are dominated by a strong
signal, named microseismic, with a major peak around 0.2 Hz
(Peterson, 1993). This is commonly dened double frequency (DF)
peak, in that its frequency is twice that prevailing in ocean waves and
consist of Rayleigh waves excited by the perturbations of sea water
pressure on ocean bottom; this represents an effect of the collision between oppositely propagating waves directed to and reected by continental coasts (Longuet-Higgins, 1950).
2.2. Polarisation
Polarisation of microseismic signals needs to be considered in ambient noise analysis for its possible inuence on the determination of directional properties of site resonance. Analysing seven years of
recordings acquired all over the world, Schimmel et al. (2011) found
that the polarisation of microseisms is consistent with a location of
their sources in ocean areas. They also observed seasonal variations in
the sources positions related to the location of major storms and to

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

the consequent changes in ocean swell spatial distribution. This kind of


signals can propagate over thousands of kilometres. Therefore, an analysis of directional variation of ambient noise H/V ratios at microseismic
frequencies can be biased by the source-controlled Rayleigh-type
polarisation of microseisms, which can show a persistent orientation
for some time and changes from one season to the other.
The above observations indicate a limitation of an analysis of directional resonances at microseismic frequencies. However, according to
Bromirski et al. (2005), at frequency larger than 0.3 Hz microseismic
energy does not seem to propagate through ocean oor beyond a few
hundreds of km and signals observed at these frequencies are generally
excited by sea waves generated by winds in the nearest coastal areas.
Thus, when analysing site response directivity, frequencies down to
0.3 Hz can be taken into consideration, but with some caution especially
where sea coast is close to the study site. In particular, we recommend
i) a simultaneous acquisition of noise at different sites in the same
study area, in order to distinguish whether observed polarisation is
peculiar of certain sites or reects signal's regional properties attributable to the characteristics of a distant noise source; and ii) the repetition of measurements in different seasons, to reveal whether the
polarisation shows constant properties due to site response and not seasonal variations reecting noise source properties.
2.3. Noise composition
The nature of the waveelds present in ambient noise represents
another aspect that deserves to be thoroughly analysed. The composition of the noise signal is still a matter of a controversy about the actual
proportion of different kinds of body and surface waves (cf. BonnefoyClaudet et al., 2006; Albarello and Lunedei, 2009). In common practice
of ambient noise analysis it is often assumed that recordings mainly
consist of fundamental mode Rayleigh waves. However, BonnefoyClaudet et al. (2006) showed that, at frequencies N 1 Hz, noise vertical
components generally include a mix of P waves and Rayleigh waves of
different modes, while horizontal components include a mix of Love
and Rayleigh waves. The proportion between the different types of
waves varies with frequency and depends on site properties, source
characteristics and distance.
Nevertheless, the uncertainty regarding noise waveeld nature does
not compromise the detection of the resonance frequency. The detection is feasible at least in case of resonance caused by impedance contrast in a 1D layering, because, at a similar frequency, a maximum of
H/V ratio is observed both in body waves, as effect of S-wave amplication, and in surface waves, as effect of the vanishing of Rayleigh wave
vertical component (Bard, 1999). This is why consistent results are
obtained in the resonance frequency identication.
However, the recognition of the noise signal nature is essential to
extract additional information for the determination of amplication
factors, considering the inuence of different waveelds on the amplitude of the H/V ratio values (see Albarello and Lunedei, 2009). Thus,
with regard to the interpretation of H/V curves, it is desirable to develop
a reliable technique to extract, within the noise recording, the wavetrains of different types, identifying, in particular, the Rayleigh waves
whose H/V curves can be conveniently modelled in terms of ellipticity
of the Rayleigh wave particle motion.
With regard to the interpretation of dispersion curves derived from
noise data, the use of vertical component recordings should lead to reliable identication of Rayleigh waves, because in this case Love waves
are excluded and body waves have different velocities. However, some
uncertainties can remain about the vibration mode of the recorded
Rayleigh waves and, in general, the inversion procedure requires considerable caution. For instance in some applications it may be necessary
to test different hypotheses and conduct a multimodal inversion in
which different parts of the dispersion curve corresponding to anomalous velocity changes are tentatively attributed to different modes
(e.g. Coccia et al., 2010).

185

3. Instruments for noise measurements


Different instruments are suitable for ambient noise measurements,
depending on the frequency band of interest and the data processing
procedure. For HVNR measurements portable velocimeters with
eigenfrequency of 1 to a few Hz can be used for microtremor analysis,
whereas accelerometers should be excluded because they are not sensitive enough to record small amplitude ground vibration. The market
now offers specically devised compact meters of noise, named
tromographs, e.g. Tromino (see www.tromino.eu for details), which
can record signals down to frequencies of a few tens of Hz.
If the focus is on frequencies below 1 Hz, a better option could be to
use a portable broad-band sensor like Trillium Compact (produced by
Nanometrics) combined with a portable acquisition system: despite
its small size (of the order of 10 cm), inner electronic feedback circuits
allow extending sensor response to very low frequencies, keeping the
nominal amplitude response practically homogeneous in a wide
frequency range (differences not larger than 1% in the interval 0.02
50 Hz). The electronic feedback makes this instrument very sensitive
to environmental conditions (temperature, air pressure, supporting
surface deformations), whose variation during a measurement can
introduce additional noise at low frequencies. Such effects can be
mitigated by insulating the sensor under a foam-lined cover during
measurements.
The time needed to stabilise broad band sensors response from hysteretic drift induced by shocks during instrument transfer between
measurement sites can represent a potential practical limitation. However, we carried out several tests comparing two identical Trillium instruments placed side by side, one kept xed at a permanent seismic
station and the other moved around before starting data acquisition.
The test results demonstrated that after 1015 min the mobile Trillium response stabilises, becoming equal to that of the xed instrument
(Figure 3); furthermore, applying a linear detrend to each of the time
windows for which spectral ratios are calculated, the drift effects can
be removed from the data acquired within a few minutes after instrument installation.
Noise measurements aimed at deriving Rayleigh wave velocity dispersion curves require simultaneous recordings by multiple properly
synchronised sensors (Shapiro and Campillo, 2004). For investigations
involving relatively long distances (in the order of several hundreds of
metres), this can be obtained by deploying two or more compact seismometers or tromographs, whose recordings are synchronised by
GPS. When dealing with shorter distances, one can use geophone arrays,
selecting vertical sensors with eigen-frequencies as low as possible
(e.g. 4.5 Hz), according to the recommendations provided by Louie
(2001).

4. Determination of site resonance properties


The HVNR technique is commonly used to characterise site resonance properties in simple geological conditions that can be assimilated
to 1D layering. Although unstable slopes frequently show lateral variations in material properties, often the depth of a surface layer susceptible to sliding is small in comparison to its lateral extension. In these
cases the analysis of H/V spectral ratios should reveal resonance frequencies related to the combination of surface layer thickness and
mean Vs velocity (see Eq. (2)). Literature on landslides provides several
examples of the HVNR technique applications, in which the Eq. (2) and
the observed variations of resonance frequencies were used to estimate
lateral variations of landslide thickness. For instance, Danneels et al.
(2008), studying a thin, about 1 km long loess ow in Kyrgyzstan,
exploited changes in H/V peak frequencies to reveal that the thickness
of the mobilised layer varied from 3 to 12 m. Other examples of similar
applications were reported by Gallipoli et al. (2000), Mric et al. (2007),
Jongmans et al. (2009) and Torgoev et al. (2013).

186

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

a) REFERENCE

b) ROVER
15

Frequency (HZ)

Frequency (HZ)

15

10

10

0
0

10

20

30

40

50

10

20

time (minutes)

30

40

50

time (minutes)

Ratio
0

10

Fig. 3. Comparison of spectrograms of H/V ratios obtained for the North component of ground motion from the recordings of two Trillium instruments placed side by side. Spectral ratios
as function of frequency were calculated on successive 30 s windows and are represented as vertical bars through a colour scale as function of time, starting from the initial recording:
a) spectrogram of the permanent station; b) spectrogram of the mobile station.

However, the resonance properties of slopes affected by or susceptible to landsliding often appear more complex than those observed in
sites with simple horizontal layering. This is not surprising considering
that resonance phenomena are basically caused by constructive interference of waves reectedrefracteddiffracted at the free surface and
at interfaces separating materials characterised by strong impedance
contrast. The geometry of these surfaces can determine amplication
at multiple frequencies controlled by the topography and by the lateral
extent (in addition to thickness) of near-surface geological bodies
which trap seismic waves. Indeed 2D or 3D effects were found to lead
to complex pattern of resonance peaks in case of larger scale geological
features like sedimentary basin (cf. Haghshenas et al., 2008) and one
can expect that similar situation can also occur, at higher frequencies,
on slopes affected by landslides.

4.1. Identication of site response directivity


One consequence of the complex site response in landslide-prone
areas is the occurrence of resonance anisotropy reected by azimuthal
variation of the H/V spectral ratios. The occurrence of directional variability of site response has been reported in literature (e.g. Bonamassa
and Vidale, 1991; Spudich et al., 1996) in different geological and topographic conditions and, interestingly, in some cases the relation to the
presence of pre-existing landslides was postulated (cf. Rial, 1996; Xu
et al., 1996). Phenomena of resonance directional variability have
been taken into consideration by some workers who applied the
HVNR techniques to study landslide areas (e.g. Havenith et al., 2002;
Mric et al., 2007). In particular, the differences in H/V ratios calculated
along orthogonal directions have been interpreted as possibly related
to landslide or slope characteristics (e.g. slope direction, maximum/
minimum thickness variability).
The results of a rst comprehensive directional analysis of ambient
noise recordings to detect H/V anisotropies in landslide areas were reported by Del Gaudio et al. (2008). Their study focused on the test
area of Caramanico Terme (central Italy), where an ongoing long term
accelerometer monitoring of landslide prone slopes offered the opportunity of comparing the results of noise analysis with site amplications
characteristics inferred from seismic event recordings. In particular,
following the methodology proposed by Borcherdt (1970), the analysis
of the Standard Spectral Ratios (SSR), i.e. the average spectral ratios
between homologous component recordings of the same events on

different slopes and at a reference site on rock, revealed that some


sites were constantly characterised by pronounced directional maxima
of amplications along a site specic preferential direction, regardless
of the event source location and mechanism (Del Gaudio and
Wasowski, 2007, 2011). The HVNR analysis at the same sites showed
a good correlation with systematic directional maxima of H/V spectral
ratio approximately oriented as the site response directivity revealed
by SSR analysis (Del Gaudio et al., 2008, 2013). A similar consistency
of the results was reported in the case of the Randa rock slide
(Switzerland), where, in the unstable part of the slope, an analysis
aimed at identifying strike and ellipticity of noise polarisation, revealed
a persistent maximum of polarisation oriented as the SSR maxima
(Burjnek et al., 2010; Moore et al., 2011).
On the basis of our experiences, we propose the following diagnostic
criteria for the reliable recognition of the occurrence and the orientation
of site response directivity from HVNR measurements (Del Gaudio et al.,
2008, 2013):
1) Presence, in the average H/V spectral ratios, of relative maxima with
amplitude larger than 2;
2) Azimuthal variation of H/V ratios of such maxima, with their amplitude at the peak frequency showing a decrease down to a directional
minimum (typically orthogonal to maximum), which should not be
larger than 2/3 of the maximum;
3) Consistent orientation (within 2030) of major directional peaks
in the average H/V ratios;
4) Dominant presence (in quantitative terms), in the recording session,
of time windows showing directional peaks satisfying the criteria
1) and 2) along a common orientation.
The criterion 1) is diagnostic with regard to the presence of amplication conditions: the H/V ratio at bedrock is expected to be equal to 1
within an approximation factor of 2, thus H/V N 2 values are requested
to evidence that surface lithology and/or morphology can cause ground
motion amplication. The criterion 2) is requested as evidence of a
signicant anisotropy of site response and the criterion 3) allows
recognising the presence of a direction along which maximum ground
motion tend to be concentrated.
Finally, the criterion 4), introduced under the acronym DHVPOR
(Directional H/V Peak Occurrence Rate: see Del Gaudio et al., 2013), assures that H/V directional maxima reect the persistence of a coherent
polarisation during the recording session, excluding any bias by the

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

187

because the variability of directional H/V ratios is larger than that of


their average between horizontal components; this greater variability
is due to noise polarisation controlled by properties of sources sparsely
distributed around the measurement site. Thus, the criterion 4) is a
softer way of correcting the bias that can result from isolated polarised
wave trains.

temporary recording of a strongly polarised wave train. In standard


HVNR method applications this bias is corrected by eliminating from
the calculations of average H/V ratios those time windows that can be
responsible for a strong increase of standard deviation of spectral ratios
(cf. Bard, 2004; Castellaro and Mulargia, 2009). However such a procedure could be too restrictive in the case of HVNR directional analysis

Percentage

10 20 30 40 50 60 70 80 90100

Percentage

b) Jun 2010

10 20 30 40 50 60 70 80 90100

a) Jul 2007

16

16

14

14

12

12

en

cy

z)

(H

z)

(H

60

180

qu

cy

40

160

10

Fre

10

en

qu

Fre

Azim

20

140
120
00
1
80
t

60

180

Azim

Percentage

10 20 30 40 50 60 70 80 90 100

Percentage

40

160

d) Jun 2012

10 20 30 40 50 60 70 80 90 100

c) May 2011

20

140
120
00
1
80
t

16

16

14

14

12

12

cy

140
120

u
Azim

z)

(H

100

en

z)

40

80

180

qu

(H

20

60

160

10

Fre

cy

en

qu

10

Fre

140
120

20

40

60

80

100

160

180

u
Azim

mean H/V
0

10

Fig. 4. Histograms of the DHVPOR values calculated from noise recordings carried out at site CAR2 at different times. Bar heights are proportional to the percentage of time windows that
show directional peaks satisfying signicance criteria (see text), for different combinations of azimuths (spaced by 10) and frequency (binned by 0.5 Hz intervals). Colour scale represents
the average of the amplitudes of the H/V peaks belonging to each azimuth-frequency bin; note that dark bars represent azimuth-frequency bins for which the average H/V peak ratio exceed the colour scale maximum.

188

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

pattern of H/V peak polarisation from occasional conditions of peak


iso-orientation controlled by the noise sources.
Figs. 4 and 5 show an example of this approach through 3D histograms representing the distribution of DHVPOR values as function of
azimuth and frequency: in these diagrams bar height is proportional
to the percentage of time windows including signicant directional
peaks (in the sense of criteria 1 and 2) and colours represent the averages of H/V peak values falling within each azimuth-frequency
bin.

Finding all the 4 mentioned criteria satised in a single noise recording, however, does not assure that the studied site is characterised by
site directional resonance. Indeed it is possible that environmental conditions determine the presence of a polarised noise spanning through
several frequencies: for instance, the presence of a continuous dominant
wind can induce vibration of trees, poles, buildings of different sizes,
causing noise with coherent polarisation at different frequencies. Thus
noise recordings should be repeated at different times and under different environmental conditions to distinguish a constant site specic

Percentage

10 20 30 40 50 60 70 80 90 100

Percentage

b) Jun 2010

10 20 30 40 50 60 70 80 90 100

a) Jul 2007

16

16

14

14

12

12

6
4

t
imu

Az

20

40

60

80

100

160

180

t
imu

Az

Percentage

10 20 30 40 50 60 70 80 90 100

d) Jun 2012

10 20 30 40 50 60 70 80 90 100

c) May 2011

Percentage

40
20 1

40

100

20

80

180

(Hz

(Hz

60

160

ncy

ncy

que

10

Fre

10

que

Fre

40
20 1

16

16

14

14

12

12

6
4

ut

m
Azi

20

60
40

80

100

14
120

ut

m
Azi

mean H/V
0

10

Fig. 5. Histograms of the DHVPOR values represented as in Fig. 3, derived from noise recordings carried out at site CAR5 at different times.

80

60 1

01

(Hz

)
100

ncy

(Hz

80

que

10

Fre

ncy

que

10

Fre

20

60
40

80

60 1

01

14
120

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

and those of spectral amplication, and this makes the identication


of the main frequencies affected by amplication more uncertain.
One can expect that similar situations can occur also at the scale of
smaller geological features like landslides. For example, on the basis of
an extensive microseismic survey of the Kalai Nav landslide in Tien
Shan, Torgoev et al. (2013) indicated a considerable reduction of H/V
spectral ratio amplitude in the head and crown zone in comparison to
its central part. Also our observations of H/V curves obtained from the
Caramanico accelerometer site CAR2 at the landslide head conrmed
both the presence of i) a more complex SSR pattern (in comparison to
what could be expected on the basis of landslide thickness and impedance contrast only), and ii) discrepancies between amplitude of H/V
and SSR peaks, the former being much lower than the latter (see
Figure 6).
The recognition of main resonance frequencies from H/V spectral ratios of noise recordings is further complicated by the possible occurrence of vertical component amplication, which can reduce the H/V
ratios. Such amplication was observed at CAR2 site just at the frequency of about 2.5 Hz characterised by the highest level of directional amplication of the horizontal components (see Figure 6b and Del
Gaudio et al., 2013). Furthermore, measurement repetitions at different
times revealed that the H/V peak at 2.5 Hz, evident in the JuneJuly data,
was hardly distinguishable in other periods; this in turn implies a possible inuence of seasonal variations (Figure 7).
This phenomenon could perhaps be linked to the variation in water
content of colluvial deposits constituting the landslide body at CAR2.
The increase in water content following rainfall and snow melting in
winterspring months could be responsible of the rise of P-wave velocity: this, in turn, could result in an increase of H/V ratio, both for the reduction of the amplication of body wave vertical component and for
the attening of the ellipticity of Rayleigh wave ground motion
resulting from the Poisson ratio increase (cf. Tuan et al., 2011).
Other difculties in recognition of directivity can arise from the
weak excitation of Rayleigh waves by the ambient noise sources. Rayleigh waves appear the most effective source of information on site directional resonance, but the results of DHVPOR analysis have provided
evidence that a large portion of a noise recording can contain limited
amount of well polarised wave-trains of Rayleigh type (Del Gaudio
et al., 2013). This means that the average of H/V spectral ratios is lessened by the contribution of non-polarised signals, which do not reect
site directional resonance properties. The implication is that a careful
selection of Rayleigh wave trains in the noise recording should lead to

DHVPOR values shown in Figs. 4 and 5 were obtained from different


measurement campaigns carried out at two nearby sites of the
Caramanico test area, i.e. the already cited accelerometer station CAR2,
located on a landslide, and another accelerometer station, CAR5, placed
on the same material of the landslide but about 150 m upslope of the
landslide crown (see Figure 1 for location).
Accelerometer data revealed no systematic preferential orientation
of maximum acceleration is present at CAR5, which is contrary to
what was observed at CAR2. The results of noise analysis also showed
a preferential orientation of H/V peaks at CAR2 in all the noise measurement campaigns, consistent with that of seismic ground motion amplication (Figure 4), and a more disperse and time varying distribution of
H/V peaks at CAR5 (Figure 5). However the results of a campaign in
2012 indicated a preferential orientation of peaks at CAR5 as well, so
that on the basis of this single measurement one could erroneously
infer the presence of a site response directivity. This example underlines
the importance of measurement repetitions.

4.2. Identication of directional resonance frequency


Although the presence and orientation of a site response directivity
can be quite easily recognised with a few repetitions of noise recordings,
the identication of main directional resonance frequencies is not
always straightforward. We can expect difculties in at least four situations: a) the presence of 2D or 3D effects; b) the occurrence of amplication of the vertical component of ground motion; c) the presence of a
weak excitation of Rayleigh waves by the ambient noise source; d) the
presence of a strong signal polarisation controlled by noise source
properties.
Site resonance properties can be complex in the presence of nearsurface geological bodies whose geometry signicantly deviates from
1D layering. With reference to larger scale structures like sedimentary
basin, Haghshenas et al. (2008) showed that the spectral amplication
of seismic waves tends to be complicated by the superimposition of interference between waves travelling not only vertically between free
surface and substratum top surface, but also horizontally between the
lateral boundaries of the basin. On the other hand, ambient noise H/V
spectral ratios, when measured near the edge of 2D or 3D geological
bodies, show lower and broader maxima in comparison to those observed on a 1D layering of the same thickness. These phenomena can
cause considerable dissimilarities between the curves of H/V ratios

a) HVNR

b) SSR (23 events)


15

Frequency (HZ)

15

Frequency (HZ)

189

10

10

0
0

30

60

90

120

150

180

30

60

Azimuth

90

120

150

180

Azimuth

Spectral ratio
0

10

Fig. 6. Comparison between azimuthal variations of HVNR (left) and SSR (right) values at CAR2 site. HVNR derived from noise measurements carried out in July 2007; SSR values derived
averaging spectral ratios between recordings obtained at CAR2 and at the reference site CAR4 for 23 seismic events recorded from 2007 to 2009. Vertical bar to the right shows the SSR
values relative to the vertical component.

190

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Jul 2007

Jun 2010

May 2011

15

10

15

Frequency (HZ)

Frequency (HZ)

Frequency (HZ)

15

10

0
0

30 60 90 120 150 180

0
0

30 60 90 120 150 180

Azimuth

30 60 90 120 150 180

Azimuth

Azimuth

Dec 2011

Jul 2013

15

15

Frequency (HZ)

Fr equency (HZ)

10

10

10

0
0

30 60 90 120 150 180

Azimuth

30 60 90 120 150 180

Azimuth

mean H/V
0

Fig. 7. Azimuthal distribution of HVNR values derived from noise measurements at site CAR2 repeated in different periods of year.

(Mw = 6.6), affected a slope with alternating beds of sandstone and


siltstones, and caused the damming of the Imogawa river (Chigira and
Yagi, 2006; Sassa et al., 2006). To recognise site specic resonance frequencies we analysed the DHVPOR values from recordings carried out
simultaneously with two instruments. One of these was kept xed at
site (TJ0), located at the toe of the landslide and used as reference.
Fig. 9 shows an example relative to the comparison with the recordings
acquired at a site (TJ1) on the head of the landslide.

SSR 90

10
HVNR 2007 Tromino
HVNR 2010 Trillium

8
HVNR 2011 Tromino
HVNR 2011 Trillium

Ratio

the improvement of analysis and better results. In this context, some


workers proposed methods based on time-frequency analysis to identify signal portion having a signicant energy in the vertical component
(Fh et al., 2001; Poggi et al., 2012), and thereby exclude the presence
of signicant contribution from SH-type polarised signals (Love or
body waves) to noise waveeld.
A simplied method to enhance the contribution of Rayleigh wavetrains to the H/V ratio calculation was proposed by Del Gaudio et al.
(2013) in the framework of the DHVPOR analysis. According to this
method, the calculation of the average H/V ratios is restricted to the
peak values (found in each recording session time window) that satisfy
the signicance criteria 1) and 2) described in Section 4.1. This excludes
H/V ratios related to non-polarised signals. Applying this method to the
noise data from CAR2 site, the presence of a signicant resonance frequency at 2.5 Hz can be highlighted despite the considerable uctuations of H/V ratio amplitude related to seasonal variation of site
conditions (Figure 8).
Uncertainty in directional resonance frequency analysis can derive
also from the possible presence of a source of strongly polarised noise
acting during data acquisition. This will be especially relevant at relatively low frequencies, which are characterised by a lower attenuation
and can be recorded even at large distance from the noise source. This
can be the case of dominant winds blowing in an almost constant direction or sea waves impact on the nearest coast. A procedure that can help
distinguish such cases from the occurrence of site directional resonance
consists in acquiring data simultaneously at more sites.
In particular, with at least two noise meters available, one of them
can be conveniently maintained in continuous recording at a x site
used as reference, while the other one is employed as a rover and
moved to different sites for shorter recordings. A comparison between
simultaneous recordings at the reference and the rover sites can reveal
whether coherently polarised signals reect site specic resonance
properties or are simply due to a regional waveeld coming from external sources and acting throughout the study area.
To illustrate the above case we present an example from our study
conducted in the area of the Terano landslide, Japan. The failure, triggered on 23 October 2004 by the mid Niigata prefecture earthquake

6
mean HVNR

0
0

10

12

14

16

Frequency (Hz)
Fig. 8. Diagram of mean values of peak spectral ratios at site CAR2 along a direction (EW)
characterised by the highest DHVPOR values. The red solid line represents the SSR values
obtained along the same azimuth; the other lines represent the average of the H/V peaks
values in time windows of different recordings (specied in the legend), grouped into frequencies bins of 0.5 Hz. The thick solid line represents the average of the curves relative to
different measurements.

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

191

Percentage

102030405060708090100

Percentage

b) TJ1

102030405060708090100

a) TJ0

16

16

14

14

12
10

q
Fre

10

180

6
4

)
Hz

)
Hz

60

y(
nc

y(
nc

Azim

40

160

ue

ue

q
Fre

12

20

140
120
100
0
8
t

20

40

60

140
120
100
0
8
t

160

180

Azim

mean H/V
0

10

c) DHVPOR TJ0 - TJ1

10

TJ0

100

TJ1

90

80

60
50

40

DHVPOR

H/V Ratio

70

30
2

20
10

0
0

10

12

14

16

Frequency (Hz)
Fig. 9. Histograms of DHVPOR values obtained for noise measurements carried out in 2012 on the slope affected by the Terano landslide (Japan). Part a) and b) represent the results obtained from simultaneous noise recordings carried out, respectively, at the sites TJ0, on the landslide toe, and TJ1, on the landslide head (see Figure 10 for location). For the direction
characterised by the highest DHVPOR values, part c) shows a comparison of data of two stations distinguished by colours (red for TJ0, blue for TJ1), i.e.: the DHVPOR values of TJO and
TJ1 indicated as bars put side by side within each frequency bin of 0.5 Hz (scale to the right); the average (solid lines) and the average one standard deviation (dashed lines) of H/V
peak values calculated as in Fig. 8 (scale to the left).

Following the above described procedure, curves of the average H/V


peak values for frequency bins of 0.5 Hz were examined along an approximately EW direction which is characterised by a preferential orientation of H/V directional peaks. This was done assuming as signicant
those H/V peaks corresponding to high values of DHVPOR (to exclude
peaks derived from measurements including a small number of time
windows). Apart from a strong polarised signal at frequencies below 1
Hz, which appears ubiquitous at all the measurement sites and is likely
related to a source-controlled polarisation, the results showed two

distinct signicant peaks (with H/V ratio 5) at 2.53 Hz and 3.5


4 Hz, respectively for TJ0 and TJ1. Other peaks at TJ1 at frequencies of
12 and 13 Hz are not considered signicant, because the corresponding
high H/V ratios (with amplitude of about 4) are derived from only about
5% of the recording data (Figure 9). Such low percentage suggests that
these peaks reect the strong polarisation of a short duration wavetrain rather than a site effect.
Fig. 10 synthesises the results of the ambient noise measurement
campaign in terms of orientation and frequencies of signicant H/V

192

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Fig. 10. Map of the Terano landslide and results of ambient noise analysis in terms of direction (indicated by arrows) and frequencies (indicated by colours) of main H/V directional peaks
derived from the DHVPOR analysis. Black arrows indicate that H/V peaks were found only at frequencies less than 1 Hz. Solid and dashed lines outline the boundary of 2004 landslide and of
a previous slope failure.

directional peaks. Although local site-specic variation of peak frequencies is apparent throughout the slope affected by the landslide, all peaks
share a common preferential EW orientation. However, to verify
whether they reect sites' directional resonance frequency, a repetition
of measurements under different environmental conditions is needed
to exclude the inuence of localised source of polarised noise (e.g. different height trees shaken by constantly directed wind).
5. Determination of S-wave velocity models
Although in some cases the amplitude of H/V spectral ratio was
similar to SSR values (cf. Lermo and Chvez-Garca, 1994), the former
cannot be condently used as a reliable estimate of site amplication
factor at the resonance frequency. Typically a correlation is found between H/V ratio amplitude and spectral amplication, considering that
both increase with the impedance contrast between surface layer and
substratum (Fh et al., 2001). However, in general, the values of the
H/V spectral ratios can considerably differ from the site amplication
factor, being inuenced by the ellipticity of the Rayleigh waves and by
the amount of SH-type wave (Love and body waves) contributing to
the ambient noise. The relative weights of different body and surface
waves composing the noise waveeld were found to depend on certain
characteristics of the measurement sites (e.g. presence of a more or less
large impedance contrast, interface geometry deviation from 1D
layering) and of the noise sources (e.g. distance from the measurement
site) (Haghshenas et al., 2008).
Nevertheless, amplication factor can be estimated through a numerical modelling of slope behaviour under seismic shaking, which
can be incorporated in dynamic slope stability assessment, e.g. through
permanent displacement analysis techniques that represent a good
compromise between prediction reliability and computational simplicity (Jibson, 2011). To include the effect of slope dynamic response, such
methods use decoupled and coupled approaches, according to
whether site amplication is assessed before or during permanent displacement computation. In both cases the modelling of slope dynamic
response depends primarily on elastic characteristics of slope materials
that can be derived from or represented by shear wave velocities of lithological layers present above and below the slip surface (Jibson, 2011).
Several computer programmes are available to calculate seismic
ground motion amplication. For example, one can use 1D modelling
(e.g. STRATA by Kottke and Rathje, 2008), which is adequate when amplication is due only to impedance contrast between surface layer and

bedrock, or 2D nite element analysis (e.g. QUAD4M, Hudson et al.,


2003), which in addition can account for topographic amplications
and boundary effects. The main required input is shear modulus or
S-wave velocity of the geological bodies being modelled as constituting
the site subsoil.
S-wave velocities are commonly obtained using several methods of
active seismic survey, however, passive methods based on ambient
noise processing can also be used. The latter offer some practical advantages, because they do not require an articial source of seismic waves,
which can present logistic and safety problems in the context of unstable slopes and rough topography typical of landslide areas.
Ambient noise processing aimed at Vs determination can be conducted in different ways using different instruments and acquisition
procedures. In particular, one can derive velocity model i) by interpreting the same H/V curve as function of frequency, ii) from analysis
of correlation between simultaneous recordings carried out according
to different congurations of deployed sensors, and iii) from the
space-time sampling of the noise waveeld along a geophone array
aimed at pointing out signal corresponding to passing through Rayleigh
wave trains.

5.1. H/V curve modelling


The inversion of H/V ratio spectrum in terms of S-wave velocity can
be carried out searching velocity models capable to reproduce the observed H/V curve in numerical simulations (e.g. Arai and Tokimatsu,
2004). In the trial and error procedure the search of solutions compatible with the observation data (within the measurement uncertainties)
relies on assumptions regarding the presence and proportion of different wave types (S-waves, different modes of Rayleigh and Love
waves) in the ambient noise recording. Furthermore, some independent
constraints are desirable (e.g. the depth of a major vertical discontinuity) to limit the number of possible solutions (Castellaro and Mulargia,
2009).
One advantage of the above method is that it is less inuenced by
lateral variation of soil properties with respect to measurements carried
out with a sensor array, because H/V values are based on measurements
carried out at a single point. However, one should take into account that,
as previously noted, H/V curve can be altered by 2D/3D effects in case of
measurement sites located close to lateral boundaries of geological
bodies.

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

A major drawback of the method is the dependence of H/V curve


interpretation on a considerable number of variables (including the
P-wave velocities), which multiplies the number of possible models
compatible with the experimental data. Thus, the method should not
be used without the support of other data and/or techniques, but can
be best exploited to provide additional constraints needed in other
geophysical investigations.

5.2. Correlation methods


A second method to determine Vs values from noise data processing
consists in calculating the cross-correlation between simultaneous
noise recordings at couple of sensors deployed in the study area. The
calculation is carried out by averaging the cross-correlation obtained
for a large number of time windows extracted from a long recording:
these time windows need to be several times longer than the maximum
period to be analysed. The cross-correlation obtained for different time
shifts between the recordings points out coherent wave-trains travelling from one sensor to the other and this allows calculating the velocity
of such wave-trains. By analysing the velocity for different frequencies,
it is possible to recognise dispersive surface waves and determine their
velocity variations with frequency, which mainly depend on vertical
changes in Vs.
To avoid ambiguous identication of surface waves, vertical component sensors should preferably be used, thus excluding Love waves from
data acquisition. The obtained velocity dispersion curves can be then
interpreted as pertinent to Rayleigh waves only and inverted to determine the Vs vertical distribution according to a 1D model.
This kind of an approach was rst proposed by Aki (1957) and is
now known under the acronym SPAC (Spatial Auto-Correlation). The
application requires the deployment of circular arrays of sensors around
a central sensor, which allows obtaining an azimuthal average of the
normalised correlations between the recordings at the central sensor
and at each of the other equidistant sensors. Aki (1957) showed that
the correlation between a couple of recordings ltered around dened
frequencies has the form of a zero order Bessel function of the rst
kind, whose argument is the product of the distance between sensors
by the wave number associated to that frequency: thus the wave number for different frequencies can be found and phase velocity
determined.
One challenging aspect of the application of this method is the need
to deploy the array with a circular conguration; this may not be easy in
landslide areas. Furthermore, since the results reect an azimuthal average, the method cannot reveal anisotropies of mechanical properties in
slope materials.
One recent application of this technique to unstable slopes was reported by Mric et al. (2007), who investigated two landslides in the
French Alps. The obtained results were in agreement with those of the
active seismic survey, and the authors also used the outcomes of H/V
measurements to provide additional constraints for the dispersion
curve inversion.
The difculty of deploying the sensors array according to a circular
conguration can be overcome following a methodological development known under the acronym ESAC (Extended Spatial AutoCorrelation), which was proposed by Ling and Okada (1993) and
popularised by Ohori et al. (2002). Through this technique simultaneous
noise recordings acquired by different sensors are rst ltered around
different central frequencies. Then, for each analysed frequency, cross
correlation is calculated between recordings obtained at different couples of sensors and compared to a Bessel function expressed as function
of sensor distance and wave velocity. Finally, for the latter parameter
the value is searched that provides the best tting of the Bessel function
with the cross correlation values obtained for different inter-sensor distances. This approach requires the deployment of an array of sensors
distributed at different distances from each other in order to have

193

enough cross-correlation values for a reliable estimate of velocity providing the best tting.
However, recently Ekstrm et al. (2009) proposed an alternative
way to obtain Rayleigh wave velocities as function of frequency. In
this alternative distinct dispersion curves can be calculated for each
couple of sensors and mean Rayleigh wave velocity along the path
connecting the two sensors is obtained. Most recently Pilz et al.
(2013) used this approach to obtain, from a set of multi-sensor ambient
noise recordings, 3D tomographies of S-wave velocities within a slope
affected by a landslide in the Fergana valley (Kyrgyzstan).
Another technique exploiting a cross-correlation analysis is based on
the properties of random diffuse waveelds demonstrated by Lobkis
and Weaver (2001) in acoustics. The technique was rst applied in ambient noise analysis by Shapiro and Campillo (2004), who found that
from cross correlation between noise recordings carried out simultaneously at two stations one can derive a signal proportional to the
Green's function relative to the station pair, i.e. the signal recorded at
one of the station as effect of an impulsive instantaneous force applied
at the other station site. The noise signal can be processed with a
frequency-time analysis (FTAN technique: Dziewonski et al., 1969;
Levshin et al., 1972) to obtain a dispersion curve of surface wave
group velocities, which, in turn, can be inverted in terms of S-wave vertical distribution. This approach has recently received an increased attention for its application potential to study different scale geophysical
phenomena using a variable frequency range (from crustal-mantle
structure to local site investigations: see Nunziata et al., 2009). A comprehensive description of the data processing methodology can be
found in Bensen et al. (2007).
In principle, velocity measurements can be obtained from a single
pair of sensors. However, a potential problem derives from the assumption of isotropic distribution of noise sources, which is plausible considering that recorded waveeld is scattered by sparsely distributed
subsoil heterogeneities. The isotropy assumption is imposed by the
method theory to obtain from data processing real velocities rather
than apparent ones biased by the absence of waves propagating parallel
to the line joining the sensor pair. Recent experimental tests demonstrated that the isotropy of ambient noise source distribution is hardly
satised (Mulargia, 2012). However, with multiple sensors deployed
to secure a coverage of different azimuths in noise data sampling, the
bias of an anisotropic distribution of noise sources can be contained
through a proper data processing (Mulargia and Castellaro, 2010).
The cross-correlation analysis of ambient noise can use data acquired for H/V spectral ratio calculations provided that simultaneous
reference-rover recordings rely on an accurate synchronisation system.
Thus the acquired data can be exploited for two different types of analysis. In the context of landslide area investigation, however, one can
have to face the presence of an anisotropy both in noise source azimuthal distribution and in slope material properties, so that it is not simple to
distinguish between these two effects.
Fig. 11 shows the rst results obtained in the Caramanico study area
by applying the cross-correlation method to noise recordings carried
out simultaneously at three sites: CAR2 on the slide head and at two
other sites located outside the landslide, one (CAR5) located 150 m
away upslope of CAR2 and the other (G12) 180 m away along the
slope direction. The data were processed using codes included in the
software package CPS (Computer Programs in Seismology: Herrmann,
2010). Importantly, the application of FTAN method allowed the recognition of Rayleigh waves of two different modes (fundamental and 1st
higher mode), which helped to better constrain Vs velocity inversion.
The results of the inversion showed a signicant difference in velocity
(up to 40% in shallow layers) between the alignments parallel and perpendicular to the maximum slope direction, the former being
characterised by lower velocity. We cannot exclude that this difference
is an artefact due to anisotropy of noise sources, which would tend to
cause an overestimate along alignment lacking noise sources. However
the difference in velocity appears consistent with the local structural

194

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Fig. 11. Results of the inversion of the dispersion curves of Rayleigh wave group velocities derived from cross-correlation analysis carried out between the recording site pairs CAR2CAR5
and CAR2G12: a) Vs velocity vertical prole providing dispersion curves best tting the experimental data for the pair CAR2CAR5 (blue line) and CAR2G12 (red line); b) location of the
two investigated alignments on a DEM of the study area showing lithological boundaries (white lines); c) and d) theoretical curve (continuous line) and experimental values (dots) of
Rayleigh wave group velocity for CAR2CAR5 and CAR2G12, respectively (note that the two distinct curves in each diagrams are referred, from the right to the left, to the fundamental
and the rst higher mode).

setting characterised by the presence of discontinuities roughly perpendicular to the maximum slope direction (Wasowski and Del Gaudio,
2000).
An interesting application of correlation-based method of velocity
determination has recently been reported by Mainsant et al. (2012),
who conducted continuous monitoring of ambient noise on two sites
near the opposite lateral boundaries of the Pont Bourquin landslide in
Switzerland (Figure 12). It is a case of about 10 m thick earthow
mobilised after intense rainfall on August 2010. The continuous acquisition of ambient noise offered the possibility of measuring the time variation of Rayleigh waves in slope material on a daily basis. The analysis
of velocity variation at frequencies of 1012 Hz showed a continuous
gradual decrease during the month preceding the failure and an accelerated drop in velocity (about 7%) in the last 4 days before the landslide
activation (Figure 12). Numerical modelling showed that the change
in Rayleigh wave velocity is attributable to a decrease of Vs in an
about 2 m thick layer at the base of the earthow; this can be related

to weakening of slope material shear strength. The case study reported


by Mainsant et al. (2012) suggests an interesting potential of ambient
noise measurements for implementation in slope failure warning
systems.
5.3. Geophone array methods
Another approach that can exploit ambient noise to determine slope
material Vs values consists in the application of a survey method developed by Louie (2001), commonly known under the acronym ReMi
(Refraction Microtremor). This method makes use of an array of vertical
geophones of relatively low frequency (typically about 4 Hz), which acquire ambient noise for some tens of minutes. Then recordings are
subdivided into time windows of few tens of seconds. Each set of recordings at different geophones represents a space-time sampling of
noise waveeld: data are processed to transform them into a matrix of
mean normalised spectral powers which is named pf transform. It is

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Fig. 12. Location of two stations (S1 and S2) of ambient noise continuous monitoring at the
opposite sides of the Pont Bourquin landslide (top) and diagram showing Rayleigh wave
velocity variations (as percentage) derived from noise analysis (grey dots) compared to
water table level variations (dark grey line), as function of time before and during the period of landslide reactivation (marked by vertical grey bar). Arrows (1) and (2) mark the
times of the beginning of velocity reduction and of the major velocity drop, respectively
(modied from Mainsant et al., 2012).

mapped as function of frequency f and apparent slowness p (the inverse


of apparent velocity) of the signals passed through the array alignment
(see Figure 13a). On the pf map coherent signals corresponding to
Rayleigh waves appear as trends of high power spectrum values. Picking
some points along this trends, the corresponding combination of slowness p and frequency f provide a sampling of the Rayleigh wave phase
velocity dispersion curves: these can then be interpreted in terms of
Vs velocity.
High spectral power values can be found in relation to energetic
wave trains crossing the geophone array along different directions:
the signal apparent velocity is given by the ratio between intergeophone distance and time difference of signal passage. Apparent
velocity coincides with the real velocity only if waves propagate parallel
to the geophone array: for waves having any other propagation direction apparent velocity will be larger than the real one. Thus picking
pf values along the lower boundary of the trend of high power spectrum values, corresponding to the minimum velocity shown by coherent signals, provides the best approximation of the real velocity
dispersion curve (see example in Figure 13a). In theory, this should
assure that the picked dispersion curve corresponds to the Rayleigh
fundamental mode (which is the slowest). However, some authors
(Xia et al., 2003; Luo et al., 2007) observed that at frequencies larger
than 1718 Hz higher modes tend to be dominant: this is likely the
reason why it is generally recommended to limit data interpretation
to frequencies not larger than 2.5 times the sensor eigenfrequency.
We tested the application of ReMi technique to landslides in the
Caramanico study area (Coccia et al., 2010), using a free software package (DINVER: Wathelet, 2005) for the inversion of the dispersion
curves. Considering the difculties in the array deployment posed by

195

irregular topography and lateral heterogeneities of soil material, we


used arrays shorter than 100 m, which is the minimum length recommended in standard applications (Louie, 2001). Nonetheless, thanks to
some additional constraints from independent subsurface data, the
method proved capable of providing information about slope material
Vs velocity down to a depth even a little larger than the commonly
assumed limit of half of the array length.
Measurements were also carried out deploying an L-shape array
with two orthogonal branches, one of which approximately oriented
along the maximum slope direction. The dispersion curves obtained at
sites outside the landslide body did not show directional differences,
whereas, at a site on the landslide head, 1020% lower velocities were
observed along the maximum slope direction for all the investigated
frequency range. This difference appears small, at the limit of uncertainty of the method, and inuence of anisotropy of noise source distribution cannot be excluded. Nonetheless, it is noteworthy that the
directional difference is qualitatively consistent with that derived by
cross-correlation analysis carried out along the alignment CAR5
CAR2G12 (Section 5.2).
A signicant outcome of this application was the recognition of portions of dispersion curves clearly belonging to different modes (see
Figure 13). However the repetition of measurements under different
environmental conditions shows that mode excitation can undergo
changes, so that surveys conducted in different periods can sample dispersion curves of different modes (Coccia et al., 2010). This poses some
problems in mode identication that can, however, be resolved through
a careful comparative analysis of dispersion curves obtained from data
acquired at different times in nearby sites. Finally, the presence of a
multi-modal signal in ambient noise has also a positive implication in
that, once modes are correctly identied, multi-modal curve inversion
provides a better resolution of model parameters (cf. Xia et al., 2000).
6. Discussion
The review of the recent literature on the dynamic response of
landslide-prone slopes and on the use of ambient noise analysis in
investigations of seismic slope behaviour has raised important questions that are discussed in the following. The issues of interest include
i) factors controlling site response directivity, ii) effects of directional
resonance on slope stability and iii) contribution of ambient noise analysis to improve slope hazard assessment.
6.1. Factors controlling site response directivity
The origins of the site directivity effects observed in landslide areas
are still unclear in part because it is difcult (or impossible) to recognise
a common predominant causative factor among all the case studies. This
possibly reects the existence of different mechanisms that can generate directional resonance, so that somewhat different explanations
could be proposed case by case.
One factor that seems to play a clear role in determining azimuthal
variation of site response is topography. This is evident especially in
case of elongated ridges, where a topographic effect reects the anisotropy of the relief's shape. Below we describe a case from the Apennine
Mountains, where we could demonstrate the presence of topographic
effect with the aid of both accelerometer and ambient noise data.
The site of interest includes the accelerometer station of Castiglione
Messer Marino (CMM), belonging to the Italian National Accelerometer
Network. The station is located about 10 m below the top of a sandstone
hill elongated for about 800 m in approximately NS direction, being
about 500 m wide and having a local relief of 80 m. The analysis of accelerometer recordings of 9 seismic events pointed out a systematic pronounced directional maximum of shaking energy in the direction
transversal to the hill elongation, even though this shaking was not
particularly strong. Ambient noise measurements, analysed through
the DHVPOR approach provided evidence of an approximately EW

196

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Fig. 13. Results of ReMi surveys conducted at site CAR2 repeating measurements at different times and with different congurations of the geophone array. a) map of the slowness
frequency (pf) matrix obtained from the rst measurement campaign and the results of the picking of points on pf map, chosen to sample the Rayleigh wave velocity dispersion
curve (open squares); continuous curves represent isolines joining points whose combinations of pf values correspond to xed wavelengths (expressed in metres) annotated on
each isoline. One can notice that the series of picked points crosses in more points the same isoline: this implies that different velocity values are associated to the same wavelength,
which is an evidence of the multi-modal nature of the resulting dispersion curve. b) Results of picking on pf matrixes derived from different noise data acquisitions (distinguished
by different symbols) at the same site; the picking results are represented in terms of velocities as function of wavelength and these data clearly outline at least two distinct modes
of Rayleigh waves interpreted as the fundamental and the rst higher mode (modied from Coccia et al., 2010).

directivity of H/V peak. Furthermore, although a comparison of H/V


peak mean values of the noise recordings acquired at the hill's base
and top showed a similar pattern, the presence of larger H/V ratio on
the hilltop (Figure 14) suggested a role of topographic amplication in
generating the directivity.
A well known literature case of directional resonance is that of the
Tarzana hill (California), where the observed directions of major and
secondary amplications were found, respectively, transversal and
parallel to the hill elongation axis (Spudich et al., 1996). This led the
authors to infer a topographic control of the phenomenon.
However a purely topographic effect can hardly account for amplication factors larger than two, which, however, seem common in case of
directional resonance (Del Gaudio and Wasowski, 2011). This points to
the geology, local structural characteristics, heterogeneities and anisotropies of rock mechanical properties, which could all have inuence on
directional variations of resonance.
An anisotropy of rock mechanical properties was invoked by Moore
et al. (2011) to explain the earlier cited case of the Randa rock slide: the
stronger amplitude of ground motion affecting the unstable part of the
slope in the maximum slope direction was attributed to the presence
of tension fractures oriented approximately perpendicular to the

displacement direction. These local structural features were the cause


of an anisotropy of slope material deformability under the effect of seismic shaking.
A similar effect could be present also in landslides affecting softer
materials, considering an anisotropy of shear strength induced by gravitational mass movement. For instance, at the interface between substratum and surface layer, one can envision a directional variation of
wave velocity resulting in an increase of transmission coefcient for
waves polarised transversally to the prevailing ssuring direction.
However, the behaviour of slope materials under dynamic conditions is not controlled only by the mechanical effects of gravitational
movements: it can depend also on structural features inherited from
tectonics and on 3D geometry of the near-surface geological bodies.
The interaction of these different factors can explain the presence of
directivity diverging from maximum slope direction.
Recently Pilz et al. (2013) found evidence of a complex pattern of
resonance phenomena on the aforementioned landslide in Kyrgyzstan.
Field measurements of H/V revealed the presence of a peak at frequency
of 22.5 Hz with an EW directivity (approximately along the maximum slope direction) throughout a 360 m long and 300 m wide landslide body consisting of limestones and claystones. The common peak

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Percentage

102030405060708090100

Percentage

b) CMM2

102030405060708090100

a) CMM1

197

16

16

14

14

12
10

q
Fre

10

180

6
4

)
Hz

)
Hz

y(
nc

y(
nc

60

160

ue

ue

q
Fre

12

Azim

40
20

140
120
100
0
8
t

40
20

60

80

140
120
100

160

180

ut

Azim

mean H/V
0

10

c) DHVPOR CMM1 - CMM2


10

CMM1

100

CMM2

90

80

60
50

40

DHVPOR

H/V Ratio

70

30
2

20
10

0
0

10

12

14

16

Frequency (Hz)
Fig. 14. Histograms of the DHVPOR values calculated from noise recordings carried out at the site of CMM accelerometer station, located 10 m below the top of a hill (CMM1a), and on the
top of the same hill (CMM2b), and comparison of the results obtained along the direction of maximum DHVPOR (c) (CMM1 in red, CMM2 in blue); data representation as in Fig. 9.

frequency appeared independent from the landslide thickness and was


interpreted as resonance effect of the whole landslide body according to
a vibration mode polarised along a direction characterised by weaker
mechanical constraints (corresponding to the sliding direction). In addition, the local occurrence of some minor peaks with higher frequencies
and variable directions was related to sub-units of the mass movement
characterised by different mobility.
On the whole, the presence of a directional variability of ground motion amplication reveals an anisotropy of geological body deformability,
which can depend on morphological, structural and lithological factors
(or their combination), acting at different frequencies. This implies that
the direction of maximum amplication does not necessarily coincide

with that of maximum slope along which potential sliding can occur.
Clearly, directional resonance will have an increased destabilising effect
on slopes where these two directions are similar and the size of potential
failures could be particularly sensitive to the site resonance frequencies.
6.2. Impact of slope resonance on slope stability
Another important question is whether amplication effect can have
a signicant impact on slope stability and whether directional variations
of the amplication can signicantly modify such an impact. Quantication of amplication factors on marginally stable slopes, based on a
comparison of accelerometer with a nearby reference site, was obtained

198

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

only in a few case studies. These include the Utiku landslide (New
ZealandGarambois et al., 2010), Randa rockslope 2010 (Switerland
Moore et al., 2011) and the Caramanico hillslopes (ItalyDel Gaudio
and Wasowski, 2007, 2011). The studies reported quite consistent
values of spectral amplications in a range of factors between 6 and
10, with variations by a factor 23 between directional maximum and
minimum. For the Italian site, Del Gaudio and Wasowski (2011) also
analysed amplication factors in terms of total shaking energy
expressed in terms of Arias intensity. They noted a considerable variability of amplication factors at low magnitude events, which, at
higher magnitudes tended to stabilise around a factor of 20 in direction
of maximum amplication (or around factor of 10 along a perpendicular
direction).
An estimate of the impact of amplication on slope destabilisation
can be obtained from empirical relations that link coseismic landslide
displacement to shaking energy (e.g. Jibson, 2007). Such relations indicate that the logarithm of permanent displacement induced by shaking
along a sliding surface is proportional to the logarithm of shaking energy according to a factor of 1.5 or more (range in estimates reects different datasets used). This implies that for an increase of Arias intensity by
a factor of 10 or 20 (depending whether the direction of maximum or
minimum amplication is considered), site amplication can cause a
median increase of displacement by a factor between 30 and 90. Thus
the impact in terms of increasing slope failure likelihood can be very
signicant.
6.3. Contribution of noise analysis to slope hazard estimates
At present the occurrence and the orientation of directional resonance cannot be determined from the examination of general slope
characteristics. Therefore, noise analysis can represent a quick and
cost-effective tool to provide insight on the strength demand for slopes
to resist failure during seismic events. The data on slopes directional resonance obtained from ambient noise measurements could help in a
better denition of a seismic scenario at regional scale, following approaches like that proposed by Jibson et al. (2000). If specic accurate
estimates of seismic shaking energy amplication factors are not available, an indicative range of 1020 can be suggested, the extreme values
being adopted when potential sliding direction is perpendicular or parallel, respectively, to maximum amplication direction derived from
noise measurements.
For stability assessment at a single slope scale, noise analysis can
provide additional useful information concerning in particular major
resonance frequencies and amplication factors. With regard to resonance frequencies, in addition to a simple examination of azimuthal
variation of HVNR values, a more advanced analysis of azimuthal variation of HVNR values can be needed when noise spectra show a complex
pattern with multiple peaks and seasonal variability of their amplitude.
In this context, methods aimed at extracting from the noise recording
the contribution of Rayleigh waves can help to recognise noise spectrum
properties that more closely reect site resonance frequencies.
Regarding amplication factors, the possibility of deriving them directly from H/V ratio amplitude should be dealt with some caution.
With reference to the complex site conditions often affecting marginally
stable slope, there are very few literature examples of a comparison between HVNR data and spectral amplications resulting from seismic
event recordings at the same site. The cases described by Burjnek
et al. (2010) and Del Gaudio et al. (2013) suggest that an acceptable
agreement between H/V ratios from noise recording and amplication
factors relative to a reference site can be found in case of rock slopes.
On the contrary, one case of a soil slope affected by landslide (Del
Gaudio et al., 2013) indicated that HVNR can lead to a considerable
underestimate of amplication (approximately by a factor of 2) at resonance frequencies, together with a possible seasonal variability attributable to vertical component amplication. Thus in those cases the H/V
ratio can be considered a lower boundary of the possible amplication

factor. A more reliable estimate can be obtained through an opportune


choice of the measurement season: it seems better to conduct measurements in periods when the water table is at the highest level and, therefore, P-wave velocity of the surface layer at its maximum.
However, for a slope-specic hazard assessment, a reliable evaluation of amplication may require numerical modelling of the slope response, using 1D or nite element approaches, according to whether
potential slide surface depth is small or not in comparison to landslide
size. In addition to information on material density, another basic ingredient required for numerical analysis is a model of shear-wave velocities
of slope material. This information could be retrieved directly from the
ambient noise data. An effective approach could consist of acquiring
noise records by using at least a couple of simultaneous synchronised
recorders. In this way, together with the acquisition of data for HVNR
analysis, one can also obtain the data needed to derive Vs values of
the slope material through correlation analysis.
7. Conclusions
Investigations of slope dynamic response to seismic shaking can
benet from different techniques that exploit short term ambient
noise recordings and derive information on site resonance properties.
The use of natural noise signals is attractive in studies on marginally
stable slopes, since other approaches require more time and money
(e.g. monitoring based on permanent in situ accelerometer stations)
or pose problems in the deployment of more cumbersome instrumentation (e.g. active geophysical survey methods).
The lack of control over signal source characteristics makes noise
data processing complex and interpretation difcult, especially as
regards the uncertainties in noise source location and noise waveeld
nature (possibly including body waves and multimodal surface
waves). Therefore, ambient noise analysis generally requires independent data to provide supplementary constraints for modelling and
thereby reduce the range of solutions compatible with the experimental
data. Ambient noise data can be effectively exploited in wide area
assessments of slope dynamic response, when some ground-truth is
available (auxiliary data from active geophysics or geotechnical investigations) thus making cross-comparisons possible. Wide area acquisition
of noise data is feasible thanks to portable instruments like compact
tromographs or geophones, which can also be employed at sites with
difcult access.
However, when investigating landslide prone slopes through ambient noise analysis, ad hoc adaptations of data acquisition and processing
procedures may be needed to select the data containing the most relevant information. In particular, an improved capacity to identify
Rayleigh wave trains in noise recording is desirable.
The information that can be relatively easily extracted from noise
analysis is the presence and orientation of site directional resonance.
This can be inferred from the search for persistent coherent sitespecic directional maxima in H/V spectral ratios derived using the
Nakamura's technique and a procedure designed to reveal azimuthal
variations. More complex appears the identication of main resonance
frequencies, which can require i) the repetition of measurements
under different environmental conditions, paying attention to groundwater conditions, and ii) the simultaneous acquisitions of noise data at
nearby sites to distinguish site specic resonance properties from
source controlled characteristics of the recorded waveeld.
Unfortunately, site amplication factors may not be condently
derived from ambient noise. However, noise data can be processed
using different techniques (H/V curve inversion, cross-correlation analysis, ReMi) to model Vs velocity of slope material, which represents a
fundamental input in numerical modelling of site dynamic response.
The application of these techniques on landslide prone slopes requires
some particular caution, taking into account possible effects of anisotropy both in source noise spatial distribution and in slope material
properties.

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Acknowledgement
We thank the editors of this special issue and two anonymous reviewers for their comments, which help to greatly improve this paper.

References
Aki, K., 1957. Space and time spectra of stationary stochastic waves, with special reference
to microtremors. Bull. Earthq. Res. Inst. 35, 415457.
Albarello, D., Lunedei, E., 2009. Alternative interpretations of horizontal to vertical spectral ratios of ambient vibrations: new insights from theoretical modeling. Bull. Earthquake Eng. 8, 519534. http://dx.doi.org/10.1007/s10518-009-9110-0.
Arai, H., Tokimatsu, K., 2004. S-wave velocity proling by inversion of microtremor H/V
spectrum. Bull. Seismol. Soc. Am. 94, 5363.
Bard, P.-Y., 1999. Microtremor measurements: a tool for site effect estimation? In: Irikura,
K., Kudo, K., Okada, H., Sasatani, T. (Eds.), The Effects of Surface Geology on Seismic
Motion. Balkema, Rotterdam, pp. 12511279.
Bard, P. Y., (coordinator) and The SESAME Team, 2004. Guidelines for the implementation
of the H/V spectral ratio technique on ambient vibrations: measurements, processing
and interpretation. SESAME European research project, WP12Deliverable D23.12,
http://sesame-fp5.obs.ujf-grenoble.fr/Papers/HV_User_Guidelines.pdf.
Bensen, G.D., Ritzwoller, M.H., Barmin, M.P., Levshin, A.L., Lin, F., Moschetti, M.P., Shapiro,
N.M., Yang, Y., 2007. Processing seismic ambient noise data to obtain reliable broadband surface wave dispersion measurements. Geophys. J. Int. 169, 12391260.
Bonamassa, O., Vidale, J.E., 1991. Directional site resonances observed from aftershocks of
the 18 October 1989 Loma Prieta earthquake. Bull. Seismol. Soc. Am. 81, 19451957.
Bonnefoy-Claudet, S., Cotton, F., Bard, P.-Y., 2006. The nature of seismic noise waveeld
and its implications for site effects studiesa literature review. Earth-Sci. Rev. 79,
205227.
Borcherdt, R.D., 1970. Effects of local geology on ground motion near San Francisco Bay.
Bull. Seismol. Soc. Am. 60, 2961.
Bourdeau, C., Havenith, H.-B., 2008. Site effects modeling applied to the slope affected by
the Suusamyr earthquake (Kyrgyzstan, 1992). Eng. Geol. 97, 126145.
Bozzano, F., Lenti, L., Martino, M., Paciello, A., Scarascia Mugnozza, G., 2008. Self-excitation
process due to local seismic amplication responsible for the reactivation of the
Salcito landslide (Italy) on 31 October 2002. J. Geophys. Res. 113, B10312. http://dx.
doi.org/10.1029/2007JB005309.
Bromirski, P.D., Duennebier, F.K., Stephen, R.A., 2005. Midocean microseisms. Geochem.
Geophys. Geosy. 6 (4), Q04009. http://dx.doi.org/10.1029/2004GC000768.
Burjnek, J., Gassner-Stamm, G., Poggi, V., Moore, J.R., Fh, D., 2010. Ambient vibration
analysis of an unstable mountain slope. Geophys. J. Int. 180, 820828.
Castellaro, S., Mulargia, F., 2009. VS30 estimates using constrained H/V measurements.
Bull. Seismol. Soc. Am. 99 (2A), 761773.
Chigira, M., Yagi, H., 2006. Geological and geomorphological characteristics of landslides
triggered by the 2004 Mid Niigta prefecture earthquake in Japan. Eng. Geol. 82,
202221.
Chigira, M., Wang, W.-N., Furuya, T., Kamai, T., 2003. Geological causes and geomorphological precursors of the Tsaoling landslide triggered by the 1999 Chi-Chi earthquake,
Taiwan. Eng. Geol. 68, 259273.
Coccia, S., Del Gaudio, V., Venisti, N., Wasowski, J., 2010. Application of Refraction
Microtremor (ReMi) technique for determination of 1-D shear wave velocity in a
landslide area. J. Appl. Geophys. 71, 7189. http://dx.doi.org/10.1016/j.jappgeo.
2010.05.001.
Danneels, G., Bourdeau, C., Torgoev, I., Havenith, H.B., 2008. Geophysical investigation and
dynamic modelling of unstable slopes: case-study of Kainama (Kyrgyzstan).
Geophys. J. Int. 175, 1734.
Del Gaudio, V., Wasowski, J., 2007. Directivity of slope dynamic response to seismic shaking. Geophys. Res. Lett. 34, L12301. http://dx.doi.org/10.1029/GL029842.
Del Gaudio, V., Wasowski, J., 2011. Advances and problems in understanding the seismic
response of potentially unstable slopes. Eng. Geol. 122, 7383.
Del Gaudio, V., Coccia, S., Wasowski, J., Gallipoli, M.R., Mucciarelli, M., 2008. Detection of
directivity in seismic site response from microtremor spectral analysis. Nat. Hazards
Earth Syst. Sci. 8, 751762.
Del Gaudio, V., Wasowski, J., Muscillo, S., 2013. New developments in ambient noise analysis to characterise the seismic response of landslide prone slopes. Nat. Hazards Earth
Syst. Sci. 13 (20752087), 2013. http://dx.doi.org/10.5194/nhess-13-2075-2013.
Dziewonski, A.M., Bloch, S., Landisman, M., 1969. A technique for the analysis of transient
seismic signals. Bull. Seismol. Soc. Am. 59, 427444.
Ekstrm, G., Abers, G.A., Webb, S.C., 2009. Determination of surface-wave phase velocities
across USArray from noise and Aki's spectral formulation. Geophys. Res. Lett. 36,
L18301. http://dx.doi.org/10.1029/2009GL039131.
Fh, D., Kind, F., Giardini, D., 2001. A theoretical investigation of average H/V ratios.
Geophys. J. Int. 145, 535549.
Fan, X., van Westen, C.J., Korup, O., Gorum, T., Xu, X., Fuchu Dai, F., Huang, R., Wang, G.,
2012. Transient water and sediment storage of the decaying landslide dams induced
by the 2008 Wenchuan earthquake, China. Geomorphology 171172, 5868.
Gallipoli, M.R., Mucciarelli, M., 2007. Effetti direzionali in registrazioni sismometriche in
aree in frana e bordi di bacino. XXVI GNGTS Conference, Rome, 1315 November
2007 (http://www2.ogs.trieste.it/gngts/gngts/convegniprecedenti/2007/presentazioni/
2_22/2_22_14_Gallipoli_frane.pdf).
Gallipoli, M.R., Lapenna, V., Lorenzo, P., Mucciarelli, M., Perrone, A., Piscitelli, S., Sdao, F.,
2000. Comparison of geological and geophysical prospecting techniques in the
study of a landslide in southern Italy. Eur. J. Environ. Eng. Geophys. 4, 117128.

199

Garambois, S., Quintero, A., Massey, C., Voisin, C., 2010. Azimuthal and thickness variabilities of seismic site effect response of the Utiku landslide (North Island, NewZealand). EGU General Assembly 2010, Vienna, 27 May, 2010. Geophys. Res. Abstr.
12, EGU2010EGU2430.
Gorum, T., Fan, X., van Westen, C.J., Huang, R., Xu, Q., Tang, C., Wang, G., 2011. Distribution
pattern of earthquake-induced landslides triggered by the 12 May 2008 Wenchuan
earthquake. Geomorphology 133, 152167.
Haghshenas, E., Bard, P.-Y., Theodulis, N., SESAME WP04 Team, 2008. Empirical evaluation
of microtremor H/V spectral ratio. Bull. Earthq. Eng. 6, 75108.
Harp, E.L., Jibson, R.W., 2002. Anomalous concentrations of seismically triggered rock falls
in Paicoma Canyon: are they caused by highly susceptible slopes or local amplication of seismic shaking? Bull. Seismol. Soc. Am. 92, 31803189.
Harp, E.L., Wilson, R.C., Wieczorek, G.F., 1981. Landslides from the February 4, 1976,
Guatemala earthquake. USGS Professional Paper 1204-A (40 pp.).
Havenith, H.B., Jongmans, D., Faccioli, E., Abdrakhmatov, K., Bard, P.-Y., 2002. Site effect
analysis around the seismically induced Ananevo rockslide, Kyrgyzstan. Bull. Seismol.
Soc. Am. 92, 31903209.
Herrmann, R.B. (Ed.), 2010. An overview of synthetic seismogram computation. Computer Programs in Seismology. Saint Louis University (Version 3.30 [electronic], http://
www.eas.slu.edu/eqc/eqccps.html).
Hudson, M.B., Idriss, I.M., Beikae, M., 2003. QUAD4M, a computer program to evaluate the seismic response of soil structures using nite element procedures and incorporating a compliant base. Pacic Earthquake Engineering Research Center, University of California,
Berkeley, California, (http://nisee.berkeley.edu/documents/SW/QUAD4M.pdf).
Jibson, R.W., 2007. Regression models for estimating coseismic landslide displacement.
Eng. Geol. 91, 209218.
Jibson, R.W., 2011. Methods for assessing the stability of slopes during earthquakesa
retrospective. Eng. Geol. 122, 4350.
Jibson, R.W., Harp, E.L., Michael, J.A., 2000. A method for producing digital probabilistic
seismic landslide hazard maps. Eng. Geol. 58, 271289.
Jongmans, D., Bievre, G., Renalier, F., Schwartz, S., Beaurez, N., Orengo, Y., 2009. Geophysical investigation of a large landslide in glaciolacustrine clays in the Trives area
(French Alps). Eng. Geol. 109, 4556.
Kottke, A.R., Rathje, E.M., 2008. Technical manual for STRATA. Pacic Earthquake Engineering Research Center Report 2008/10. University of California, Berkeley, California.
Lenti, L., Martino, S., 2012. The interaction of seismic waves with step-like slopes and its
inuence on landslide movements. Eng. Geol. 126, 1936. http://dx.doi.org/10.1016/
j.enggeo. 2011.12.002.
Lermo, J., Chvez-Garca, F.J., 1994. Are microtremors useful in site response evaluation?
Bull. Seismol. Soc. Am. 84, 13501364.
Levshin, A.L., Pisarenko, V.F., Pogrebinsky, G.A., 1972. On a frequencytime analysis of
oscillations. Ann. Geophys. 28, 211218.
Ling, S., Okada, H., 1993. An extended use of the spatial autocorrelation method for
the estimation of structure using microtremors. Proc. of the 89th SEGJ Conference,
Nagoya, Japan, 1214 October 1993. Society of Exploration Geophysicists of Japan,
pp. 4448 (in Japanese).
Lobkis, O.I., Weaver, R.L., 2001. On the emergence of the Greens function in the correlations of a diffuse eld. J. Acoust. Soc. Am. 110, 30113017.
Longuet-Higgins, M.S., 1950. A theory of the origin of microseisms. Philos. Trans. R. Soc. A
Math. Phys. Sci. 243 (857), 135.
Louie, J.N., 2001. Faster, better: shear-wave velocity to 100 meters depth from refraction
microtremor arrays. Bull. Seismol. Soc. Am. 91, 347364.
Luo, Y., Xia, J., Liu, J., Liu, Q., Xu, S., 2007. Joint inversion of high-frequency surface waves
with fundamental and higher modes. J. Appl. Geophys. 62, 375384.
Mainsant, G., Larose, E., Brnnimann, C., Jongmans, D., Michoud, C., Jaboyedoff, M., 2012.
Ambient seismic noise monitoring of a clay landslide: toward failure prediction.
J. Geophys. Res. 117, F01030. http://dx.doi.org/10.1029/2011JF002159.
Mric, O., Garambois, S., Malet, J.P., Cadet, H., Gueguen, P., Jongmans, D., 2007. Seismic
noise-based methods for soft-rock landslide characterization. Bull. Soc. Geol. Fr. 178,
137148.
Meunier, P., Hovius, N., Haines, J.A., 2008. Topographic site effects and the location of
earthquake induced landslides. Earth Planet. Sci. Lett. 275, 221232.
Moore, J.R., Gischig, V., Burjanek, J., Loew, S., Fh, D., 2011. Site effects in unstable rock
slopes: dynamic behavior of the Randa instability (Switzerland). Bull. Seismol. Soc.
Am. 101 (6), 31103116.
Mulargia, F., 2012. The seismic noise waveeld is not diffuse. J. Acoust. Soc. Am. 131,
28532858.
Mulargia, F., Castellaro, S., 2010. Nondiffuse elastic and anelastic passive imaging. J. Acoust.
Soc. Am. 127, 13911396.
Nakamura, Y., 1989. A method for dynamic characteristics estimation of subsurface using
microtremor on the ground surface. Q. Rep. Railw. Tech. Res. Inst. 30, 2533.
Nogoshi, M., Igarashi, T., 1971. On the amplitude characteristics of microtremor (part 2)
(in Japanese with English abstract). J. Seismol. Soc. Jpn. 24, 2640.
Nunziata, C., De Nisco, G., Panza, G.F., 2009. S-waves proles from noise cross correlation
at small scale. Eng. Geol. 105, 161170.
Ohori, M., Nobata, A., Wakamatsu, K., 2002. A comparison of ESAC and FK methods of
estimating phase velocity using arbitrarily shaped microtremor arrays. Bull. Seismol.
Soc. Am. 92, 23232332.
Peterson, J., 1993. Observation and modeling of background seismic noise. US Geol. Surv.
Open-File Rept. 93-322, Albuquerque.
Pilz, M., Parolai, S., Bindi, D., Saponaro, A., Abdybachaev, U., 2013. Combining seismic
noise techniques for landslide characterization. Pure Appl. Geophys. http://dx.doi.
org/10.1007/s00024-013-0733-3.
Poggi, V., Fh, D., Burjnek, J., Giardini, D., 2012. The use of Rayleigh-wave ellipticity for
site-specic hazard assessment and microzonation: application to the city of Lucerne,
Switzerland. Geophys. J. Int. 188, 11541172.

200

V. Del Gaudio et al. / Engineering Geology 182 (2014) 182200

Rial, J.A., 1996. The anomalous seismic response of the ground at the Tarzana Hill site during the Northridge 1994 Southern California earthquake: a resonant, sliding block?
Bull. Seismol. Soc. Am. 86, 17141723.
Sassa, K., Fukuoka, H., Wang, F., Wang, G., Okada, K., Marui, H., 2006. Landslide disasters
triggered by the 2004 Mid-Niigata Prefecture earthquake in Japan. Ann. Disaster
Prev. Res. Inst., Kyoto Univ. 49, 119136.
Schimmel, M., Stutzmann, E., Ardhuin, F., Gallart, J., 2011. Polarized Earth's ambient
microseismic noise. Geochem. Geophys. Geosyst. 12 (7), Q07014. http://dx.doi.org/
10.1029/2011GC003661.
Seplveda, S.A., Murphy, W., Jibson, R.W., Petley, D.N., 2005. Seismically induced rock
slope failures resulting from topographic amplication of strong ground motions:
the case of Pacoima Canyon California. Eng. Geol. 80, 336348.
Shapiro, N.M., Campillo, M., 2004. Emergence of broadband Rayleigh waves from correlations of the ambient seismic noise. Geophys. Res. Lett. 31, L07614. http://dx.doi.org/
10.1029/2004GL019491.
Spudich, P., Hellweg, M., Lee, W.H.K., 1996. Directional topographic site response at
Tarzana observed in aftershocks of the 1994 Northridge, California, earthquake:
implications for mainshock motions. Bull. Seismol. Soc. Am. 86 (1B), S193S208.
Tang, C., Zhu, J., Xin, Q., Ding, J., 2011. Landslides induced by the Wenchuan earthquake
and the subsequent strong rainfall event: a case study in the Beichuan area of
China. Eng. Geol. 122, 2233.
Torgoev, A., Lamair, L., Torgoev, I., Havenith, H.-B., 2013. A review of recent case studies of
landslides investigated in the Tien Shan Using microseismic and other geophysical
methods. In: Ugai, K., Yagi, H., Wakai, A. (Eds.), Earthquake-Induced Landslides,

Proceedings of the International Symposium on Earthquake-Induced Landslides,


Kiryu (Japan), 79 November, 2012. Springer. ISBN: 978-3-642-32237-2, pp.
285294. http://dx.doi.org/10.1007/978-3-642-32238-9_29.
Tuan, T.T., Scherbaum, F., Malischewsky, P.G., 2011. On the relationship of peaks and
troughs of the ellipticity (H/V) of Rayleigh waves and the transmission response of
single layer over half-space models. Geophys. J. Int. 184, 793800.
Wasowski, J., Del Gaudio, V., 2000. Evaluating seismically-induced mass movement hazard in Caramanico Terme (Italy). Eng. Geol. 58 (3/4), 291311.
Wasowski, J., Lee, C.T., Keefer, D., 2011. Toward the next generation of research on earthquake induced landslides: current issues and future challenges. Eng. Geol. 122 (18),
2011. http://dx.doi.org/10.1016/j.enggeo.2011.06.001.
Wathelet, M., 2005. Array recordings of ambient vibrations: surface-wave inversion. (PhD
thesis) Universit de Lige, Belgium.
Xia, J., Miller, R.D., Park, C.B., 2000. Advantage of calculating shear-wave velocity from surface waves with higher mode. Technical program with biographies, SEG, 70th Annual
Meeting, Calgary, Canada, pp. 12951298.
Xia, J., Miller, R.D., Park, C.B., Tian, G., 2003. Inversion of high frequency surface waves
with fundamental and higher modes. J. Appl. Geophys. 52, 4557.
Xu, Z., Schwartz, S.Y., Lay, T., 1996. Seismic wave-eld observations at a dense, small aperture array located on a landslide in the Santa Cruz Mountains, California. Bull.
Seismol. Soc. Am. 86, 655669.

S-ar putea să vă placă și