Journal of Applied Geophysics 71 (2010) 7189

Contents lists available at ScienceDirect

Journal of Applied Geophysics

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j a p p g e o

Application of Refraction Microtremor (ReMi) technique for determination of 1-D

shear wave velocity in a landslide area
S. Coccia a, V. Del Gaudio a,, N. Venisti b, J. Wasowski c
a
Dipartimento di Geologia e Geosica, Universit degli Studi di Bari, via E. Orabona, 4, 70125 Bari, Italy
b
Osservatorio Sismologico, Universit degli Studi di Bari, Italy
c
Istituto di Ricerca per la Protezione Idrogeologica, Consiglio Nazionale delle Ricerche, Bari, Italy

a r t i c l e i n f o a b s t r a c t

Article history: The application of the Refraction Microtremor (ReMi) method on slopes affected by or prone to landsliding is
Received 10 March 2009 complicated by the presence of lateral lithological heterogeneities and irregular topography, which may
Accepted 5 May 2010 hinder the extension of the geophone array to the minimum lengths (100200 m) usually adopted in
standard applications of this technique. We focus on deriving one-dimensional shear-wave velocity (Vs)
Keywords:
vertical proles from the analysis of microtremor recordings carried out in the municipality of Caramanico
Landslide-prone slope
Refraction Microtremor analysis
Terme (central Italy) where the seismic response has been monitored with a local accelerometer network
Rayleigh waves since 2002. The stability of the ReMi data acquisitions and the reliability of the results in irregular landslide
S-wave velocity terrain were tested by using ReMi campaigns in three different periods and different acquisition parameters
Vibration modes (seismograph channel number, geophone frequency and spacing). We also investigated the possible
Neighbourhood algorithm presence of directional variations in soil properties by carrying out noise recordings along L-shaped arrays.
The inuence of changing environmental conditions and of different acquisition parameters was tested by
comparing the data obtained from different campaigns, using the same acquisition parameters, with the data
from simultaneous acquisitions using different parameters. The tests showed that stable results can be
obtained under different acquisition conditions provided that i) the ratio between the coherent and
incoherent part of ambient noise is sufciently high and ii) spatial aliasing does not contaminate the signal in
the p (slowness)f (frequency) matrix near the picking area: the latter condition can be satised by selecting
geophone frequency and spacing appropriate for the site characteristics and for the investigation purpose.
The differences in Vs measured in two orthogonal directions did not exceed 1020 % and their analysis
suggests that these directional variations are most likely due to anisotropy in noise source distribution rather
than in material properties.
The Rayleigh wave velocity dispersion curves obtained from microseimic noise recordings were then
inverted with the DINVER software package to derive vertical distribution of the Vs. We reconstructed
vertical proles of Vs through the joint inversion of fundamental and higher modes, constrained by borehole
information. The results from a site of a recent deep-seated slope failure showed that both colluvium few
tens of meters thick, and the underlying mudstone have lower velocities than those of the same formations
present in the surrounding area not involved in mass movements. This suggests that the mudstone has been
affected by slope deformations at this site.
2010 Elsevier B.V. All rights reserved.

1. Introduction of relevant geotechnical parameters of slope materials is difcult in sites

Considerable amount of earthquake damage in hilly and mountain-
ous areas is caused by slope failures triggered by seismic shaking
(Keefer, 1984). However our understanding of seismic slope response is
still limited because of the scarcity of ground motion recordings on
landslide-prone slopes. Furthermore, numerical modelling of slope
behaviour under earthquake shaking is not easy because the acquisition
Corresponding author. Tel.: +39 080 5442279; fax: +39 080 5442625.
E-mail address: delga@geo.uniba.it (V. Del Gaudio).
characterised by rough topography and sharp lateral lithological and/or
physical heterogeneities. The assessment of subsurface geology through
borehole or active geophysical surveying is expensive and is typically
limited to post-factum (post-failure) local scale investigations. In this
context it is of interest to explore the capability of recently developed
cheaper and quicker passive geophysical techniques that exploit
natural signal sources, like the Refraction Microtremor analysis
technique (ReMi) (Louie, 2001), also known as Noise Analysis of
Surface Waves (NASW). This methods is based on ambient noise
measurements that are carried out with seismic arrays to obtain
information on surface wave velocity dispersion.

0926-9851/$ see front matter 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.jappgeo.2010.05.001
72 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189
The inversion of dispersion curves can provide a one-dimensional
shear wave velocity (Vs) model down to a depth related to the length
of the array, and thereby furnish useful data for the numerical model-
ling of seismic site response (Havenith et al., 2007). On the other
hand, uncertainties affecting inversion results pose some problems to
a complete stand-alone application (Dal Moro et al., 2006). Therefore,
within a research aimed at improving the understanding of factors
controlling seismically induced landsliding, we tested the effective-
ness of ReMi in providing parameters relevant for the assessment of
seismic response of landslide-prone slopes.
We applied this method in the municipality of Caramanico Terme
(Central Italy), an area often subjected to mass movements caused by
earthquakes and meteorological events (Wasowski and Del Gaudio,
2000). In this area an accelerometric monitoring has been conducted
since 2002 to study seismic response of unstable slopes. Strong motion
data are being acquired through a specically designed local accel-
erometric network composed of 5 stations sited in different lithos-
tratigraphic and topographic settings. The results of the accelerometric
monitoring provided evidence of amplications on a landslide that in
1989 mobilized 3040 m thick Quaternary colluvial deposits overlying
Pliocene mudstones (Del Gaudio and Wasowski, 2007).
ReMi data for an analysis of seismic slope response, and specically
to determine Vs vertical proles, were acquired at sites near the
accelerometric stations both within and around the landslide. On
hillslopes affected by landslides considerable difculties can arise
from local eld conditions that often limit the extension of geophone
linear spread below the lengths (100200 m) commonly adopted in
ReMi standard practise. Moreover, the distance from anthropic
sources of microtremors can cause unfavourable noise conditions in
comparison to other well established cases of application.
Here we rst describe several tests conducted to assess the
repeatability of data acquisition and the uncertainty of the estimated
Rayleigh wave dispersion curves in case of a landslide-prone slope.
The assessment relies on comparisons of the data acquired by
conducting ReMi campaigns in three different periods. Then we
illustrate how vertical distributions of Vs were determined from
Rayleigh wave dispersion curves inverted with the software package
DINVER (Wathelet, 2005), which provided a set of models compatible
with the observed dispersion curves within the data uncertainty
limits. Finally, we discuss the implications of the data inversion results
for a hillslope affected by a recent deep-seated landslide.
2. Description of the study area
The town of Caramanico Terme is located in a seismically active
part of the Apennine chain, in a fault-bounded valley (Fig. 1). The
complex geological setting and the unfavourable hydrogeological
conditions that characterise the area around the town determine a
diffuse presence of marginally stable slopes with a long record of
historical landsliding, typical of mountainous settings subjected to
relatively high average precipitation and seismic activity. High local
relief, steep slopes, gully erosion, and strong river down cutting are
the main geomorphologic factors predisposing the recurrence of mass
movements (Wasowski and Del Gaudio, 2000).
The main lithologic units of the Caramanico area are (Fig. 1):
1) Miocene age limestones; 2) evaporitic succession (Messinian);
2) marly mudstones of Early Pliocene age, which form the substratum
that crops out along the slopes of the Orta River Valley; 3) carbonate
megabreccias (Quaternary?), which form the caprock of the Carama-
Fig. 1. (a) Orthophoto of the study area showing lithological units and position of the ve accelerometric stations: inset shows the geographical location of Caramanico Terme. Major
landslides are indicated in green, geological contacts in light brown (after Wasowski and Del Gaudio, 2000). Explanation: Lm = Miocene limestones; Me = Messinian sandy-silty
deposits with carbonate breccia; Mp = Pliocene mudstones; Bq = Quaternary limestone megabreccias; Sqh = Quaternary and Holocene soils (colluvium and articial ground); red
marks show the location of the accelerometric stations CAR1CAR5; the labels 1989 and 1627 indicate the locations and years of occurrence of the landslides discussed in the text.
(b) Geologic proles of CAR1-5 sites (modied after Del Gaudio and Wasowski, 2007), drawn along the maximum slope directions; m a.s.l. stands for meters above sea level.
nico hillslopes; 4) soils (colluvial materials, landslide deposits, water-
laid and eluvial sediments, articial ground - Holocene). The presence
of these different lithologies, with their lateral and vertical changes,
results in a hydrogeologic setting with high permeability contrasts
and complex groundwater ow patterns.
A majority of recent damaging slope failures have been linked to
long periods of precipitation and typically involved remobilisations of
older landslides. The 1989 landslide is a typical example of reactivation
of pre-existing mass movements and was triggered by two major
storms that occurred during a 15 day span (Wasowski, 1998). This
movement, mobilized about 40 m thick colluvium overlying Pliocene
mudstones and locally affected the thick accumulations of carbonate
debris.
In addition, several historical records describe landslides and
ground deformations induced by earthquakes in the Caramanico area
(Wasowski and Del Gaudio, 2000). For example a large mass
movement was caused by an earthquake of 1627 (estimated
magnitude between 6.7 and 7.0), with epicentre located in northern
Apulia, about 100 km from Caramanico.
Seismic response of marginally stable peri-urban slopes at
Caramanico Terme is currently investigated with the aid of a per-
manent network of 5 accelerometric stations. The monitored sites are
characterised by different topographic and lithostratigraphic condi-
tions (Fig. 1): (1) CAR1, sited on Pliocene mudstones, within a slope
dipping 18 to WSW ; (2) CAR2, 600 m to SSE of CAR1, within the
same hillslope dipping 11 to WSW, but on the head of 1989 landslide;
(3) CAR3, on the rim of an over 50 m deep gorge locally oriented
WNWENE, on 10 m thick Messinian carbonate breccia overlying
Miocene limestones; (4) CAR4 (reference station), 2.5 km SE of
Caramanico, on the same Miocene limestones as in CAR3, forming a
gentle slope (8) dipping to NW; (5) CAR5, 200 m upslope CAR2, on
the same thick colluvium as in CAR2, but not involved in landsliding,
in a gently inclined (b 7) area.
For two out of ve monitored sites, the numerous recordings
acquired since 2002 showed the presence of signicant amplications
with pronounced maxima systematically oriented along an almost
constant site-specic direction, regardless of the event source location.
This direction appears related to local topography features and in one
case (CAR2) coincides with the main direction of movement of a deep-
seated landslide. These observations lead to suggest the presence of a
directional resonance in the dynamic response of the two sites to
seismic shaking, which has subsequently been conrmed by analyzing
the directional variations of average horizontal to vertical spectral
ratios (HVSR) calculated both for the recorded seismic events (Del
Gaudio and Wasowski, 2007) and for microtremor recordings carried
out using portable seismograph with velocimetric characteristics (Del
Gaudio et al., 2008). The HVSR data also supported the hypothesis that
the presence of a landslide body contributes to determine conditions
of site response directivity, possibly caused by a concomitance of more
factors (e.g. local topography, lateral changes of thickness, lithology
and physical characteristics, anisotropy of landslide body material
properties controlled by gravity driven movements). The inuence of
these factors could be investigated by modelling slope behaviour
under dynamic conditions, but this requires additional data and one of
the critical parameters for the numerical modelling is shear-wave
velocity Vs.
In the following we report on the application of ReMi technique to
obtain Vs down to typical depths of mass movements at Caramanico.
This technique is based on the identication of Rayleigh wave trains in
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189 73

recordings carried out by a geophone array that provides a space-time 1984, and MASWPark et al., 1999), it does not make use of con-
sampling of seismic wave elds. Unlike other techniques based on trolled articial source of seismic waves, but exploits ambient noise of
surface wave spectral analysis (e.g. SASW - Nazarian and Stokoe, natural or anthropic origin.
74 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189

3. Data acquisition
In the Caramanico study area three ReMi data acquisition
campaigns were conducted from 2005 to 2007. In all the campaigns,
vertical geophone arrays were used to acquire 1012 records of 30 s at
each site, with a 2 ms sampling interval.
First investigation (September 2005) was carried out at CAR2 and
CAR3, i.e. the sites affected by the largest directional amplications
revealed by accelerometric recordings. Moreover, during the same
campaign, a preliminary test was made at the site planned for the
installation of the station CAR4, in order to check if the site conditions
were adequate to use this station as a reference for the comparative
analysis of the seismic site response at Caramanico. A 12 channel
seismograph was used for the recordings with an array of 10 Hz
geophones spaced each other by 5 m; only at CAR4 10 m spacing was
adopted considering that site conditions allow to extend the
geophone array for more than 100 m.
A second campaign was conducted in June 2006 at the same
previously investigated sites and also at three other sites, i.e. CAR1,
CAR5, installed in the meantime, and the inferred location of the
historical landslide triggered by the 1627 earthquake. In this
campaign a 24 channel seismograph was employed with 4.5 Hz
geophones spaced by 2.5 m, except for site CAR4, where, as in the
2005 campaign, the spacing was doubled.
A comparison of the data obtained in the two campaigns showed
some discrepancies at the sites where measurements had been
repeated. Therefore a third campaign was carried out on May 2007 to
investigate the stability of data acquisitions. Considering that the rst

Fig. 2. Pf maps obtained at the 1627 landslide site using an array of 4.5 Hz geophones with: a) 24 geophones (spacing 2.5 m); b) 12 geophones (spacing 5 m). Contour lines
represent iso curves connecting points for which the combination pf corresponds to equal wavelength () values. Open squares represent the picked pf pairs.
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189 75

two campaigns were characterised by differences both in environ-
mental conditions and in the geophone arrays employed, the third
ReMi surveying was arranged to investigate the possible inuence of
these factors on the results obtained; this was done by i) repeating
measurements with the same data acquisition parameters of the
previous campaigns and ii) conducting simultaneous data acquisition
with arrays having different characteristics. In particular, the data
acquisitions were carried out employing simultaneously two arrays of
12 and 24 geophones, both using 10 Hz or 4.5 Hz sensors, and two
arrays with 12 geophones of different frequency (4.5 and 10 Hz).
Furthermore, during 2007 campaign noise recording acquisitions
were carried out also with L-shaped array, in order to evaluate

Fig. 3. Maps of spectral power differences between simultaneous acquisitions carried out with 12 channels (geophone spacing 5 m) and 24 channels (geophone spacing 2.5 m) for
the 1627 landslide site: results obtained using 4.5 Hz (a) and 10 Hz geophones (b). Note the divergent results of picking below geophone resonant frequency.
76 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189
directional variations in the results of velocity measurements and
their possible relation with the observed directivity of seismic site
response. These tests were conducted at the reference station CAR4
located on rock and at sites close to the stations CAR2 and CAR5,
lithologically similar but differing for the presence at CAR2 of the
landslide and associated site response directivity.
Seismic data processing was performed using the module
ReMiVspect of the commercial software package SeisOpt ReMi v.
4.0, to transform seismic data recordings into a slowness (p)
frequency (f) matrix of normalised spectral power ratios (pf
transformation). The pf matrices of different records are stacked to
improve the signal/noise ratio, which in this context is the ratio
between the spectral power of the coherent part of the ambient noise
recording caused by Rayleigh waves and that of incoherent noise.
Then, dispersion curve trend of Rayleigh waves is delineated in
stacked pf spectrum by picking points along the lower boundary of
the maximum amplitude region. According to Louie (2001), this
procedure should provide phase velocity generally corresponding to
the Rayleigh fundamental mode.
In each campaign the same arrays that recorded microtremors
were also used for active seismic refraction measurements and the
data were interpreted following a tomographic approach using the
code SeismOpt@2D v. 3.5. In this way P-wave velocity comparative
data were also obtained for the shallowest layers and used to
constrain some parameters of the dispersion curve inversion.
4. Test results
The acquisition of data in three different campaigns with different
settings of the geophone arrays offered the possibility to check the
stability of the results provided by the ReMi method and to estimate
measurement uncertainty under the harsh surveying conditions of
the study area. With the only exception of the site CAR4, the local
conditions did not allow the deployment of a geophone array longer
than 60 m, which is below the recommended minimum of about
100 m (cf. Louie, 2001). To optimise the data acquisition procedure
and to verify the results reliability in such conditions, we carried out
three tests described in the following.
4.1. Acquisition setting test
One test was focused on the inuence of different acquisition
system parameters (seismograph channel number, sensor resonant
frequency, inter-geophone spacing) on the nal results. For this
purpose during the 2007 campaign we arranged simultaneous data
acquisitions at sites CAR2, CAR4, CAR5 and at the 1627 landslide with
two geophone arrays of equal length, but differing for geophone
frequency or spacing. In particular, comparisons were made between
(1) data obtained with geophones of equal frequency but with
different inter-geophone spacing and (2) data recorded with the same
geophone spacing but using different geophone frequencies.
For simultaneous acquisitions with arrays having different inter-
geophone spacing, the recorded noise sources are the same, and the
only difference is the resolution of the wave eld sampling. Most of
the recorded signals proved to be clear far beyond the frequency
2.5 times higher than that of geophones, recommended as upper limit
by Louie (2001). Indeed, with 4.5 Hz geophones, picking was possible
even up to 35 Hz (Fig. 2), which are frequencies particularly useful for
investigating shear wave velocities of the subsurface layers.
Picking obtained with different spacing generally provided
consistent results for most part of the frequency interval, but
occasionally some discrepancies were found near the extremities of
the dispersion curves. With reference to the site on the 1627
landslide, Fig. 3 shows the map of differences between spectral
powers obtained from simultaneous acquisition with geophone
spacing of 2.5 and 5 m, together with the relevant picked points,
both for recordings carried out with 4.5 and 10 Hz geophones. At
frequencies below geophone resonant frequency, discrepancies
appear clearly related to the worsening of the signal/noise ratio.
At high frequencies the anelastic attenuation contributes to signal
weakening, allowing the recording only of noise from very close
sources, but the mismatch is also inuenced by spatial aliasing. This
is responsible for the artefacts appearing in the pf map as signals
beneath the Nyquist wavelength N (equal to two inter-geophone
spacings), which actually represent an approximate mirror-like
replication of longer wavelength signals (see Fig. 2a and b, where N
is 5 and 10 m, respectively). The displacement of N towards higher
wavelengths at larger inter-geophone spacings can produce a signal
contamination by spatial aliasing at the highest frequencies of the
picking region (Fig. 2b).
The spatial aliasing, combined with the signal/noise ratio worsen-
ing as the number of geophones decreases, can make picking more
uncertain. However, this does not necessarily jeopardise the picking
results: in case of sites characterised by fast subsurface layers, like the
site CAR4 on rock, the signal wavelengths are longer and the picking
region can be well separated from the inuence of the spatial aliasing.
In this case consistent results are obtained also using a low number of
channels (see Fig. 4).
Regarding the comparison of simultaneous data recorded with the
same geophone spacing, the results obtained with 4.5 and 10 Hz
geophones are in a good agreement. There are some discrepancies at
the lowest frequencies where the picking of data acquired with 10 Hz
geophones is not possible or uncertain (Fig. 5). This indicates that
with 10 Hz geophones a reliable investigation may be limited to
shallower depths (in the order of few tens of meters).
4.2. Directivity test
To investigate the possible presence of directional variations of soil
properties, during the 2007 campaign we also conducted tests to
compare simultaneous noise recordings acquired along orthogonal
directions by deploying L-shaped arrays. The acquisition tests were
carried out in three areas: i) near the station CAR2 located on the 1989
landslide, a site that provided evidence of directivity in seismic site
response; ii) at the station CAR5 in lithologically similar setting, but
outside the landslide and not affected by directivity phenomena;
iii) at the reference station CAR4, on rock. In some cases bad quality
data were obtained and therefore, for the subsequent comparative
study, we used only simultaneous 24 channel acquisitions with 4.5
and 10 Hz geophones, which gave good results.
At the site CAR4 no differences were found along the two
orthogonal arrays (Fig. 6). Near CAR2, directional variations of limited
amount (up to 1020%, at most) were observed in the phase velocity
values (Fig. 7): the highest velocity values were found along NS
array, which is the direction of minimum amplication of the soil
motion. If directivity in site response had been caused by anisotropy of
attenuation properties (e.g. as an effect of preferentially oriented
ssuring systems, in analogy with the observations by Carter and
Kendall (2006) on oil elds), one would have expected a coincidence
of higher velocities with maximum amplication direction. Since the
opposite is observed, it seems that the velocity difference, rather than
reecting material property anisotropy, is due to anisotropy of
distribution of noise sources, which causes overestimate of velocities
if noise sources aligned with array are lacking. At the site CAR5 (Fig. 8)
there is a good agreement between the two dispersion curves until
22 Hz: at this frequency the two curves start to diverge with the
highest velocity values obtained along NS array. Again, as in the case
of CAR2, the disagreement between the two curves is probably caused
by an anisotropy of sources of noise with frequencies N 22 Hz. Indeed,
the CAR5 site is not characterized by phenomena of seismic site
response directivity.
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189
Fig. 4. (a) Maps of spectral power and picking results obtained using 24 (top) and 12 (bottom) channels for site CAR4; (b) map of spectral power differences.

77
 78
S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189
Fig. 5. (a) Maps of spectral power and picking results obtained with array of 12 channels using 4.5 Hz (top) and 10 Hz geophones (bottom) for site CAR4; (b) map of spectral power differences.
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189 79

Fig. 6. Maps of spectral power differences between orthogonal directions (NorthSouth and EastWest) and picking results obtained for site CAR4 from L-shaped 24 channel arrays
with 4.5 (a) and 10 Hz (b) geophones.

4.3. Environmental condition test
To identify differences due to variable environmental conditions,
we compared noise recordings made at different times, but with the
same acquisition parameters. In particular, we made comparisons of
the data from the rst and the third campaign at the site CAR2, and of
the data from second and the third campaign at the sites CAR2, CAR5
and the 1627 landslide.
In general, the measurements provided consistent results in terms
of repeatability (Fig. 9), as long as the signal/noise ratio was good and
80 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189

Fig. 7. Maps of spectral power differences between orthogonal directions and picking results obtained for site CAR2 from L-shaped 24 channel arrays with 4.5 (a) and 10 Hz
(b) geophones.

the acquisition setting allowed preventing spatial aliasing from
affecting the picking region on the (pf) map. However, few cases
with apparently discrepant data were also encountered.
For example, some difculties arouse when comparing the CAR2
data from the 2005 and the 2007 surveys with a 12 geophone array
(Fig. 10). The picked dispersion curves diverge considerably at
frequencies higher than 9 Hz. This can in part be attributed to the
unfavourable environmental conditions during the second acquisi-
tion; indeed the pf map indicates a rather low signal/noise ratio.
As a general principle, the lower velocity values should be
considered more reliable because the misalignment of the seismic
array with the noise sources tends to produce systematic velocity
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189 81

Fig. 8. Maps of spectral power differences between orthogonal directions and picking results obtained for site CAR5 from L-shaped 24 channel arrays with 4.5 (a) and 10 Hz
(b) geophones.

overestimates. However, an examination of the pf pairs picked from
the 2005 campaign data shows an inversion of the velocityfrequency
curve from a decreasing trend at low frequencies to an increasing one at
higher frequencies, so that the dispersion curve crosses the same iso
curve at different points. This in turn results in pf pairs for which
different velocities are associated with the same wavelength. The same
problems were also noticed in the data acquired during other campaigns
(e.g. Figs. 6 and 9). All this suggests that the obtained signals belong to
different vibration modes so that picking can jump between dispersion
curves of different modes (cf. Socco and Strobbia, 2004).
82 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189

Fig. 9. Maps of spectral power differences between the results of second (2006) and third (2007) campaign at site CAR5 (a) and at site CAR2 (b), obtained using 24 channel arrays of
4.5 Hz geophones.

This hypothesis was studied in detail by comparing all the pickings
from different acquisitions and drawing diagrams showing velocity as
function of wavelength. For instance, the pf pairs picked from the
results of different campaigns at the CAR2 site (Fig. 11) form at least
two distinct trends, which can be attributed to the fundamental and
the rst higher modes.
According to the results of our tests, the recording of one or the other
of vibration modes during different acquisitions and the frequency
intervals at which this occurs have no systematic relation with the
adopted acquisition parameters. This phenomenon seems related to
variable ambient noise conditions, possibly depending on the depth and
nature of the wave source (cf. Aki and Richards, 2002). Thus, it is possible
that apparently discrepant results between different surveys are due to
difference of vibration modes recorded at certain frequency intervals.
Some authors (Xia et al., 2003; Luo et al., 2007) found that the
energy of higher modes tends to become dominant at frequencies
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189
Fig. 10. Maps of spectral power derived from acquisitions carried out at site CAR2 using 12 channel arrays with 10 Hz geophones in the rst (2005) and third (2007) campaign (a) and map of relative spectral power differences (b).

83
84 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189
Fig. 11. Diagram showing the velocity of Rayleigh waves as function of wavelength at site CAR2 for all the pf pairs picked from the results of three campaigns (1,2,3 cam in legend)
with different acquisition settings. Note the presence of two distinct trends, highlighted by ellipses, corresponding to the fundamental and the rst higher mode, respectively.
higher than 1718 Hz. At these frequencies the fundamental mode
energy is relatively weak and this implies that dispersion curves
picked on the basis of signal energy may not be representative of the
surface wave fundamental mode as assumed by Louie (2001). On the
other hand, Xia et al. (2003) and Luo et al. (2007) also demonstrated
that at frequencies N 1718 Hz dispersion curves of higher modes
are more sensitive to the velocity of deeper layers than that of
fundamental mode: thus, the joint inversion of different modes can
increase the investigation depth and has greater potential to improve
the quality of the inversion of S-wave velocity model, provided that
different modes are correctly identied.
5. Dispersion curve inversion
Surface wave dispersion curve inversion in terms of Vs vertical
proles presents some uncertainties due to its multi-modal and highly
non linear nature. Moreover, the solution of the dispersion curve
inversion is generally non-unique (Dal Moro et al., 2006). This poses
some problems in data interpretation and requires that the uncer-
tainties affecting the results be taken into account. Thus we veried
whether and under what circumstances the data obtained from ReMi
measurements can provide useful information in the context of harsh
operating conditions.
The Rayleigh wave velocity dispersion curves, obtained from
microseismic vibration recordings during the three campaigns, were
inverted with the tool Surface Wave Inversion of the software
package DINVER (Wathelet, 2005) to reconstruct Vs vertical distri-
bution. This software uses the so-called neighbourhood algorithm,
which is a stochastic direct-search method for nding models of
acceptable data t inside a multidimensional parameter space
(Sambridge, 1999).
The comparison of the computation results with the measured
dispersion curve provides one mist value given by the expression:
s
nF
x x 2
misfit = di 2 ci
i=1 i nF
where
xdi is the velocity observed at the frequency fi,
xci is the velocity calculated at the frequency fi,
i is the uncertainty of the frequency samples considered,
nF is the number of frequency samples.
When various modes are observed and identied their inversions
require a multi-modal mist (Wathelet, 2005).
Through an iterative procedure the DINVER algorithm samples the
parameter space to nd models with minimum mist, using the
results of previous iterations to guide the search for improved models.
For this purpose, the parameter space is rst decomposed into Ns0
cells centred on randomly generated models; then, at each successive
iteration, the Nr models characterised by the lowest mist are
selected among those generated in all the previous iterations and Ns
new models are randomly generated within the cells of the best Nr
models, decomposing the parameter space around the new models
into smaller cells. Thus, the zones of the parameter space where no
plausible solution can be found are less sampled than promising areas
(Wathelet, 2008). The procedure is repeated for a prexed number of
iterations, but one can extend the number of iterations until the value
of the minimum mist become stable.
In our modelling the dispersion curve computation was designed
for layers with homogeneous properties. For surface wave inversion
the main empirical parameters, which control the inversion process
are: the S-wave velocity Vs, the P-wave velocity Vp, the density and
the thickness of each layer. However material density has limited
inuence on dispersion curves, so it can be xed at a constant value in
each layer. Vp and Vs are linked together by Poisson' ratio, which
constrains the possible range of variation, whereas thickness can be
constrained by xing the bottom depths of some layers. The Rayleigh
wave dispersion curve is most sensitive to Vs variation with depth,
thus the inuence of shear wave velocity on the Rayleigh wave data
inversion is critical, whereas that of the other structural parameters is
relatively small (Xia et al., 1999).
The procedure for deriving velocity proles from dispersion
curves was applied at the sites CAR1 (on an outcrop of the 1989
landslide substratum), CAR2 (on colluvium mobilised by the 1989
landslide) and CAR5 (sited 200 m upslope CAR2, on colluvium not
affected by the landslide). Thus, we were able to compare Vs values of
the same lithologies inside and outside the area involved in recent
1 landsliding.
A critical point in the interpretation process is the correct iden-
tication of the modes of different parts of Rayleigh wave dispersion
curves. The mode identication can be facilitated by the comparison
of measurements repeated under different ambient noise conditions,
which can point out signals of different modes, and by the comparison
of dispersion curves obtained for neighbouring sites with similar
geological characteristics, which can give indications on the consis-
tency of mode identications.
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189 85

As shown in Fig. 11, pf pairs picked in different acquisitions can Table 1

show separate trends, clearly attributable to different modes, but Frequencyslowness pairs used for the inversion of the Rayleigh wave dispersion
curves derived from ReMi surveys carried out at the sites of the accelerometric stations
there is also a clouds of pf pairs of uncertain nature where these CAR1, CAR2 and CAR5.
trends converge; this results from picking carried out in the
transition zone between the spectral intervals where the energy CAR1 CAR2 CAR5

of two different modes are prevailing. Plotting the distribution of Frequency Slowness Frequency Slowness Frequency Slowness
velocity values derived from picking against wavelength, as in Fig. 11, (Hz) (s/m) (Hz) (s/m) (Hz) (s/m)
helps to exclude uncertain data from the inversion and thus identify First higher mode Fundamental mode First higher mode
pf pairs that certainly belong to different modes. Fig. 12 shows 8 0.00076 3 0.00177 26 0.00174
together pf pairs from all the pickings of all the surveys carried out at 9 0.00087 4 0.00199 27 0.00180
10 0.00099 5 0.00219 28 0.00186
sites CAR1, CAR2 and CAR5, having selected only those belonging to 11 0.00110 6 0.00238 29 0.00193
clearly distinct modes. Considering the lithological similarity of these 12 0.00121 7 0.00255 30 0.00199
sites, one can expect that dispersion curves for the same mode should 13 0.00133 8 0.00271 31 0.00206
also be similar. Thus, the pf pairs were grouped in three clusters that 14 0.00144 9 0.00286 32 0.00212
15 0.00155 10 0.00299 33 0.00218
correspond to three different modes: the rst cluster to the left was
16 0.00166 11 0.00311 34 0.00224
identied as fundamental mode, the others as higher modes. 17 0.00178 12 0.00321 35 0.00230
The dispersion curve of each mode was approximated through a 18 0.00189 13 0.00330 36 0.00236
regression to a polynomial curve of a degree dened according to the 19 0.00200 14 0.00337 37 0.00241
trend of the picking data (mostly third degree, except for the rst 20 0.00211 15 0.00343 38 0.00246
21 0.00222 First higher mode 39 0.00250
higher mode of CAR1 and the second mode of CAR5, which were 22 0.00234 30 0.00266 Second higher mode
approximated to polynomials of second and fourth degree, respec- 23 0.00245 31 0.00271 29 0.00102
tively). Then regularly spaced pairs (p, f) were interpolated and the 32 0.00276 30 0.00113
regression standard error was used to estimate the uncertainty i 33 0.00280 31 0.00124
34 0.00283 32 0.00135
needed to calculate the mist according to the Eq. (1). Table 1 reports
35 0.00286 33 0.00143
the pf pairs used for the inversion. 36 0.00288 34 0.00149
The dispersion curve inversion through the package DINVER 37 0.00291 35 0.00153
requires the denition of the model parameter intervals that can be 38 0.00293 36 0.00155
explored during the search of solutions. The parameterisations 39 0.00295 37 0.00155
40 0.00298 38 0.00156
adopted at the examined sites are summarised in Table 2. To choose
41 0.00302
the number of layers, the depth of interfaces and the density values, 42 0.00307
we used information from boreholes (Buccolini et al., 1995) located 43 0.00313
nearby the accelerometric stations CAR2 and CAR5 (see Fig. 1b for 44 0.00321
45 0.00332
location) and considered also subsurface data uncertainties in
dening interface depth range. In some cases a supplementary
subdivision into sub-layers was introduced to account for possible
vertical variations of physical parameters within the same lithostrati- The inversion procedure can provide multiple solutions, thus we
graphic unit. Supplementary information on subsurface layering was considered only those compatible with experimental dispersion
derived from the results of seismic refractions measurements, which curves within the data uncertainty limits, i.e. with mist 1 for the
also provided data on the velocity of the P-waves. The ranges of S- curves of different modes. Some trials were also carried out
wave velocities were dened on the basis of plausible values for the attributing alternative identications to the mode of dispersion
lithologies identied by the boreholes. curves, but this did not lead to mode identications different from

Fig. 12. Diagram of the pf pairs selected from the surveys carried out at stations CAR1, CAR2 and CAR5 for the inversion of the Rayleigh wave dispersion curves. The points selected
are those for which the identication of the mode is considered reliable. Groups of pf pairs corresponding to the same mode are highlighted by ellipses.
86 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189

Table 2 1400 m/s at depths corresponding to the top of the substratum at

Range of parameter values examined in the search of solution for the inversion of the CAR2 and at CAR5; these values are higher than the velocity of CAR2
experimental Rayleigh wave velocity dispersion curves obtained at sites CAR1, CAR2
and CAR5: Depth in meters is that of the bottom of each layer, Vp and Vs are P- and S-
substratum (4001000 m/s) and are consistent with the range of
wave velocity, respectively. solutions found for CAR5 substratum (10001500 m/s). The presence
of lower Vs velocities at CAR2 suggests that the mudstone underlying
Layer Depth (m) Vp (m/s) Vs (m/s) Poisson ratio Density (Kg/m3)
the landslide was to some extent also involved in recent and/or past
CAR1 slope movements. It is also possible that the velocity data reect some
1 06 600 250300 0.20.5 2000
subtle differences in the physical properties of the substratum
2 67.5 600 300350 0.20.5 2000
3 36 15001800 350400 0.20.5 2000 material that were not captured during the borehole investigation.
4 70 20003000 3501000 0.20.5 2000
5 3000 5001300 0.20.5 2000 6. Discussion and conclusions

CAR2
Conventional methods for determining Vs (boreholes and active
1 16 500 200300 0.20.5 2000
2 612 800 300500 0.20.5 2000 seismics) are relatively expensive and often difcult to apply on
3 1240 1200 500600 0.20.5 2000 slopes affected by landslides. The ReMi technique, which is a faster
4 4060 12002600 6001200 0.20.5 2000 and cheaper method, represents a promising tool for obtaining shear
5 26003000 12001500 0.20.5 2000
wave prole in the surcial materials and underlying bedrock.
CAR5 However, the in situ conditions in marginally stable slope areas
1 112 500800 100400 0.20.5 18002000 typically limit the possibility of deploying long seismic arrays. In
2 1235 18002200 4001000 0.20.5 18002000 addition, the use of seismic wave natural source instead of controlled
3 3537 18002200 4001000 0.20.5 18002000 ones introduces a supplementary factor of uncertainty in data
4 5070 25003000 6001500 0.20.5 18002000
interpretation.
5 30003000 10002500 0.20.5 18002000
The tests of repeatability, carried out on landslide-prone slopes at
Caramanico, showed that changing environmental conditions do not
determine data inconsistency as long as a high signal/noise ratio (i.e. a
those shown in Fig. 12: for instance, no plausible solution with mist high energy of the coherent part of the ambient noise recording
1 was found interpreting the dispersion curve obtained at CAR1 as caused by Rayleigh waves in comparison to incoherent noise) is
fundamental mode. maintained. Test results indicate that the use of a larger number of
Applying the neighbourhood algorithm of DINVER package, the channels improves the signal/noise ratio, but the employment of a
initial number Ns0 of models and the number Ns of models generated minimum number of sensor (e.g. 12) does not necessarily jeopardise
at each iteration were both chosen equal to 50, whereas the number Nr the picking results, provided the geophone spacing is set up so to
of the best solutions around which the parameter space was sampled prevent spatial aliasing from contaminating the signal in the p
at each iteration was 10. For each inversion the number of iterations (slowness)f (frequency) matrix near the picking area. In this regard,
was initially xed to 100, but for the inversions of CAR2 and CAR5 data an a priori estimate of expected Rayleigh wavelengths can be made
an extension to 300 iterations was needed to obtain a stabilisation of (e.g. through simple preliminary numerical modelling taking into
the minimum mist. Thus the nal number of computed models account local lithology and possible range of velocity), to guide the
ranges from a minimum of about 5000 to a maximum of 15000. choice of geophone spacing and resonant frequencies in relation to
As dispersion data to be inverted, for the site CAR5 we identied the desired depth of investigation.
two groups of pf pairs as belonging to the rst (in the range 26 The tests conducted with simultaneous microtremor acquisition
39 Hz) and the second higher mode (in the range 2938 Hz: see along orthogonal directions did not reveal any major directional
Table 1). Inversion obtained using the rst higher mode alone and variations in velocity (up to 1020% variation, at most). These dif-
without additional constraints would have produced a large number ferences were more likely due to an anisotropic distribution of the
of models, but a signicant reduction was obtained through a joint noise sources rather than to anisotropy of material property.
inversion of the two higher modes and by introducing additional The tests showed also that consistent Rayleigh dispersion data can
independent information (Fig. 13). In particular, for CAR5 we used the be obtained even above frequencies considerably higher than the
subsurface data from the nearby borehole (see Fig. 1b for location), generally recommended limit of 2.5 times the geophone resonant
which allowed xing the depth of the interface between colluvial frequency. However, it should be taken into account that at higher
deposits and the underlying mudstone (Fig. 13c). frequencies the recorded signals show a prevalence of vibration
With regard to CAR2, pf pairs of the fundamental and of the rst modes higher than the fundamental one.
higher mode were identied in the frequency ranges 315 Hz and 30 Overall, the results of our tests demonstrate that in marginally
45 Hz, respectively (Fig. 14a). As for CAR5, the inversion was carried stable slope areas the ReMi method can provide the data needed to
out by using a joint inversion of two modes, as well as constraints calculate shear-wave velocity vertical proles down to depths of few
from the nearby borehole (see Fig. 1b for location). tens of meters. This can be achieved even using array shorter than the
Apart from the uppermost few meters, the velocity results commonly recommended minimum value of 100 m.
obtained for the colluvial deposits at CAR2 and at CAR5, are The Vs vertical distribution can be derived from inversion of
considerably different, Vs being 300600 m/s at the former site on Rayleigh wave dispersion curve. Generally, surface wave dispersion
the 1989 landslide and 6001000 m/s at the latter site, located 200 m data collected in geophysical eld trials are often noisy, limited in
upslope. This difference can be attributed to the effects of loosening terms of frequency range, can contain a mixture of data points from
and fracturing of the colluvium at CAR2 caused by the landslide. multiple mode dispersion, and therefore, are not easy to deal with. In
For CAR1, no borehole stratigraphy was available to constrain this study, the comparison of the results from several acquisitions
modelling (Fig. 14b). Furthermore, only one ReMi campaign was conducted under different conditions showed that in some cases
carried out there and picking provided data in the frequency range 8 higher modes can take more energy than the fundamental mode
23 Hz attributable to a single vibration mode identied as the rst throughout the available frequency spectrum of the recorded signal.
higher mode. Consequently the inversion results present larger In such cases higher modes remain the only ones usable for data
uncertainties. However the solutions compatible with the dispersion inversion. Xia et al. (2003) found that energy of higher modes tends to
curves indicate a progressive increase of Vs that reaches a range 700 become predominant as the source distance becomes larger and that
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189
Fig. 13. Comparison between experimental dispersion curves at site CAR5 (black solid lines with measurement error bars) and theoretical ones (top) obtained for ground prole models (bottom), with line colours representing the mist of the
theoretical curves according to the colour scale: (a) Dispersion curve of the rst higher mode and models obtained for CAR5 using this mode without independent constraints; (b) Dispersion curves of the two higher modes and models
obtained with their joint inversion; (c) Dispersion curves and models obtained with a joint inversion of the two higher modes, constrained by subsurface information from a nearby borehole whose stratigraphy is shown in the lower right
(borehole stratigraphy modied from Buccolini et al., 1995).

87
88 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189

Fig. 14. Comparison between experimental dispersion curves at sites CAR1 and CAR2 (black solid lines with measurement error bars) and theoretical ones (top) obtained for ground
prole models (bottom), with line colours representing the mist of the theoretical curves according to the colour scale: (a) Dispersion curve of the fundamental and the rst higher
mode and models obtained for CAR2 with a joint inversion of the two modes, constrained by subsurface information from a nearby borehole. (b) Dispersion curves of the rst higher
mode and models obtained for CAR1 using this mode alone and information from seismic refraction data.

shorter wavelength components of fundamental mode Rayleigh
waves are generally obscured by the higher frequency data: this
could be a common case in surveys carried out in landslide-prone
areas that are distant from sources of anthropic noise. However, the
positive side of this phenomenon is that higher modes have a high
potential in improving accuracy of S-wave velocity model inversion,
provided the parts of dispersion curves pertaining to different modes
are correctly identied. In this regard, the repetition of acquisitions
under different noise conditions and the comparison of picking results
obtained at lithologically similar nearby sites can help the identica-
tion of vibration modes by pointing out the different ranges of the
velocitywavelength relation corresponding to the different modes
present in the recorded signals.
Our tests showed that, in spite of the mode identication
difculties, the simultaneous inversion of dispersion curves of
different modes (fundamental and higher ones) can improve the
resolution of S-wave velocity model inversion (cf. Xia et al., 2000,
2003). Furthermore, with the support of independent constraints (e.g.
subsurface data from boreholes and seismic refraction), it is possible
to optimise the random inverse searching and to estimate Vs proles,
which are important for a reliable analysis of seismic site response.
Since there are no theoretical limitations to the length of ReMi
arrays, considerable depths of investigation could be obtained
(Havenith et al., 2007). In practice, however, in situ operating
conditions will often limit this possibility and if the length of the
array is not appropriate in comparison to the Rayleigh wavelengths
relevant for the investigation target, the constraints for inversion will
be weak. In such cases the number of models compatible with data
uncertainties can increase so much as to make the data practically
useless.
In general, arrays 23 times longer than the planned investigation
depth are recommended, but this work shows that satisfactory results
can also be obtained with shorter arrays through the exploitation of
supplementary constraints (e.g. borehole stratigraphy). In particular,
with 60 m long arrays we were able to reveal plausible velocity
differences, down to 3040 m depths, between the in situ slope
materials and those involved in recent mass movements. These results
indicate that ReMi can provide useful information in investigations of
landslide-prone areas as long as the multi-modal nature of the
acquired signal is correctly recognised and properly taken into
account in data inversion; the latter may require the use of
independent information (e.g. regarding the depth of major litholog-
ical interfaces). If applied as a complementary geophysical tool, ReMi
can also help to optimise the use of more expensive and logistically
demanding geotechnical investigations as well as other geophysical
methods.
 S. Coccia et al. / Journal of Applied Geophysics 71 (2010) 7189
Acknowledgements
We are grateful to Satish Pullammanappallil (Optim, Reno,
Nevada) for providing a routine that made possible the comparative
analysis of ReMi intermediate output data and Giancarlo Dal Moro
(Dipartimento di Scienze Geologiche Ambientali e Marine - Universit
di Trieste) for useful discussions on inversion process. One of two
anonymous reviewers provided useful contribution to improve this
paper. Pierpaolo Pierri (a colleague from the Dipartimento di
Geologia e Geosica of the University of Bari) and two undergraduate
students, Annamaria Saponaro and Maria Zupo, took part in some of
the eld measurements and in initial data processing. This work was
carried out with the nancial support of the Italian Ministry of
Education, University and Research through the Project of National
Interest (PRIN05) entitled Induced seismic hazard: analysis, model-
ling and predictive scenarios of earthquake triggered landslides.
References
Aki, K., Richards, P., 2002. Quantitative Seismology, Second ed. University Science
Books, Sausalito, California.
Buccolini, M., Fiorillo, F., Lollino, G., Rusi, S., Sciarra, N., Wasowski, J., 1995. La frana di
Caramanico Terme dell'ottobre 1989. Dipartimento di Scienze, Universit G.
d'Annunzio (Pescara) e gruppo nazionale C.N.R. Deformazioni gravitative profonde
di versante.
Carter, A.J., Kendall, J.M., 2006. Attenuation anisotropy and the relative frequency
content of split shear waves. Geophys. J. Int. 165, 865874.
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., 2007. Directivity of slope dynamic response to seismic
shaking. Geophys. Res. Lett. 34, L12301. doi:10.1029/GL029842.
89
Dal Moro, G., Pipan, M., Gabrielli, P., 2006. Rayleigh wave dispersion curve inversion via
genetic algorithms and marginal posterior probability density estimation.
Geophysics 61, 3955. doi:10.1016/j.appgeo.2006.04.002.
Havenith, H.B., Fh, D., Polom, U., Roull, A., 2007. S-wave velocity measurements
applied to the seismic microzonation of Basel, Upper Rhine Graben. Geophys. J. Int.
170, 346358.
Keefer, D., 1984. Landslides caused by earthquakes. GSA Bulletin 95, 406421.
Louie, J.N., 2001. Shear wave velocity to 100 meters depth from refraction microtremor
arrays. Bull. Seismol. Soc. Am. 91 (2), 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.
Nazarian, S., Stokoe II, K.H., 1984. In situ shear wave velocity from spectral analysis
of surface waves. Proc. of the 8th World Conf. on Earthquake Engineering 3,
pp. 3138.
Park, C.B., Miller, R.D., Xia, J., 1999. Multi-channel analysis of surface waves (MASW).
Geophysics 64, 800808.
Sambridge, M., 1999. Geophysical inversion with a neighbourhood algorithm: I.
Searching a parameter space. Geophys. J. Int. 138, 479494.
Socco, L.V., Strobbia, C., 2004. Surface-wave method for near-surface characterization: a
tutorial. Near Surf. Geophys. 2 (4), 165185.
Wasowski, J., Del Gaudio, V., 2000. Evaluating seismically induced mass movement
hazard in Caramanico Terme (Italy). Eng. Geol. 58 (34), 291311.
Wasowski, J., 1998. Understanding rainfalllandslide relationship in man modied
environment: a casa history from Caramanico Terme (Italy). Environ. Geol. 35,
197209.
Wathelet, M., 2005. Array recordings of ambient vibrations: surface-wave inversion.
PhD thesis, Universit de Lige, Belgium.
Wathelet, M., 2008. An improved neighborhood algorithm: parameter conditions and
dynamic scaling. Geophys. Res. Lett. 35, L09301. doi:10.1029/2008GL033256.
Xia, J., Miller, R.D., Park, C.B., 1999. Estimation of near surface shear-wave velocity by
inversion of Rayleigh wave. Geophysics 64, 691700.
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 (1), 4557.