A Critique of the Proposed National Education Policy Reform
Site effects in the city: Seismic Risk Evaluation
1. Site Effects in the City of Lourdes, France, from H/V Measurements:
Implications for Seismic-Risk Evaluation
by Annie Souriau, Agathe Roullé, and Christian Ponsolles
Abstract The pilgrimage city of Lourdes in the French Pyrenees has been heavily
damaged by earthquakes in the seventeenth and eighteenth centuries. Assessments of
local seismic hazard have been performed through a seismic microzonation of the city.
A campaign of urban noise recording has been conducted at about 180 points to
determine seismic wave amplification by means of spectral ratios of the horizon-
tal-to-vertical (H/V) components. A comparison of this ratio obtained from ambient
noise measurements with spectral ratios obtained from earthquake S-wave recording
(H/Href) reveals that the fundamental resonance frequency is well estimated but that
the site amplification is generally underestimated. Resonance frequencies and ampli-
tudes exhibit coherent patterns that relate to the geological structure. Bedrock sites are
generally characterized by flat spectra whatever the geological nature of bedrock, un-
less topography or karst are present. Along the river, in particular at the pilgrimage
site, thick sediments induce low resonance frequencies with large amplifications.
Along and across valleys, north and east of the city, H/V spectra show systematic
variations according to sediment filling properties. Finally, on the hill flanks east
of the city, H/V spectra rather relate to topography.
An additional issue with H/V spectral ratios based on ambient noise is whether this
indicator reflects meaningful geological variations. This question is tested on sites
where the S-wave ground velocity structure is known. A first test consists of estimat-
ing H/V spectral ratios by means of simulation of urban noise through the summation
of synthetic seismograms with random properties. A second test consists of estimating
H/Href spectral ratios by propagating an earthquake record through a 1D soil profile.
These two modelings successfully predict most of the observed spectrum character-
istics. In addition, they give some information on the nature of the urban noise, which
may include a significant amount of shear energy.
Introduction
In countries with relatively low seismicity, the instru-
mentally based seismic hazard assessment is difficult, be-
cause short instrumental databases lack large events. This
is the case for France, where the national seismic zonation
is currently based on the large scale effects of historical
earthquakes. Even with the new, probabilistic zonation (cur-
rently debated among experts), the spatial resolution is too
coarse for urban planning at the city scale, in particular when
local strong site effects occur. Localized strong amplifica-
tions of seismic motion, called site effects, generally occur
above unconsolidated sediments and/or rugged topography.
The city of Lourdes, in the central French Pyrenean foreland,
straddles both configurations. It is built partly at the conflu-
ence of several alluvial and glacial valleys and partly on
steep Pyrenean hill slopes. Over the past century, this city
has become an important pilgrimage site, with more than five
million visitors each year. Lourdes has been partly destroyed
twice, in the seventeenth and eighteenth centuries, by seismic
events of intensity VIII–IX located only some 10 km from
the city. A better assessment of the expected ground motions
inside the city is thus a key element for urban and civil pro-
tection planning.
A preliminary site effect experiment with 10 seismic
stations deployed on various geological units has been per-
formed by Dubos et al. (2003) at the city scale. They found
that the spectral ratio of the S-wave horizontal component
with respect to a reference station at rock site condition
(H/Href) revealed very rapid lateral variations of site effects.
In some cases, two sites separated by less than 500 m exhibit
variations of H/Href amplitudes of one order of magnitude at
some particular frequencies. It thus appears necessary to den-
sify spatial sampling to investigate site effects inside the city.
2118
Bulletin of the Seismological Society of America, Vol. 97, No. 6, pp. 2118–2136, December 2007, doi: 10.1785/0120060224
2. However, because the H/Href method depends on the record-
ing of natural earthquakes, which may take several months to
occur given the tectonic context, we preferred a method
based on estimating the ratio between horizontal and vertical
component spectra of ambient noise, called H/V in the fol-
lowing discussion (Nakamura, 1989; Bard, 1999). This tech-
nique requires recording for only a few tens of minutes at
each site. Although this method is less reliable, joint analysis
of densely sampled H/V values, H/Href ratios previously de-
termined, and geological and geotechnical information pro-
vide the major components for a seismic zonation of the city.
After a brief recall of the tectonic context and principle and
limitations of the H/V method, we present the results of the
H/V experiment, some data interpretation based on 1D mod-
eling, and a discussion of the global results in relation to the
seismic zonation of the city.
Geological and Tectonic Context
The Pyrenees result from the convergence of the Iberian
and Eurasian plates for the last 65 Ma, which took place after
an extension episode opening a shallow sea between the two
plates (Choukroune, 1992). The limit between the two plates,
the North Pyrenean fault, corresponds to a 15 km Moho jump
with a thicker crust to the south (Hirn et al., 1980), whereas
the paleo-rift remains as a weak zone (the North Pyrenean
Zone) with thick sediments north of the North Pyrenean fault
(Fig. 1). South of the North Pyrenean fault, the Paleozoic
Axial Zone contains the highest summits (3400 m in the cen-
tral part of the range). The convergence continues today at a
low rate, probably less than 1 mm=yr (Nocquet and Calais,
2003), and the seismic activity remains generally moderate.
The historical seismicity in the Pyrenees is well docu-
mented back to the fourteenth century (Lambert et al., 1996).
It reveals a maximum of activity in the central part of the
range on the French side (Fig. 1), with about 25 events of
Medvedev–Sponheuer–Karnik (MSK) intensity larger than
VII since the beginning of the seventeenth century (intensi-
ties are given in the MSK scale [Medvedev et al.,1967] com-
monly used in Europe). Part of the cities of Lourdes and
Bagnères-de-Bigorre were destroyed in 1660 by an event
of MSK intensity IX that occurred 10 km south of Lourdes
and killed 30 persons. Its magnitude is estimated as 6.0–
6.1 (Levret et al., 1994). Two other events caused severe
damage, one in 1750 with intensity VIII and another in
1854 with intensity VII, both located only 5–7 km south
of Lourdes.
Instrumental seismicity is recorded since the 1960s, but
dense networks for a global survey of the whole Pyrenean
range were deployed only in 1989 (Souriau and Pauchet,
1998), and for this specific region only in 1996. In addition,
some temporary experiments were performed to specify the
exact location of the active faults and the focal mechanism of
earthquakes in this region (Rigo et al., 2005). They reveal
a general east–west trend of seismicity north of the North
Pyrenean fault with several clusters, one of them being lo-
cated immediately south of Lourdes (Fig. 1).
Lourdes is located in the North Pyrenean Zone, in a very
complex geological setting (Fig. 2) inherited from the Pyr-
enean convergence and from the Quaternary glaciations,
Figure 1. Simplified seismotectonic map of the region of Lourdes (see location in the inset) with instrumental seismicity for the period
1989–2006 and historical seismicity for the last five centuries. North Pyrenean fault, NPF.
Site Effects in the City of Lourdes, France, from H/V Measurements 2119
3. Figure 2. (a) Topographic map with the Gave-de-Pau river and Lourdes Lake, and (b) geological map of the city of Lourdes. Also
reported are the stations used for the 2001 experiment of site effect measurements (Dubos et al., 2003) using spectral ratios with a reference
station at rock (ROC). Sanctuary, SAN; limits of the Lourdes district, dashed line.
2120 A. Souriau, A. Roullé, and C. Ponsolles
4. whose northern limit was at the latitude of Lourdes. Lourdes
is built at the junction of five valley branches (Fig. 2a). Dur-
ing successive glaciations, the Lourdes glacier flowed from
south to north across Lourdes and through the valleys north-
northeast and northwest of the town. Today, these two val-
leys are dry. The river, Gave-de-Pau, flows from south to
north into town and bends west along the axial drainage. This
Quaternary landscape evolution explains the present filling
with glacial moraines with large erratic blocks in the south,
north-northeast, and northwest valleys and with alluvial se-
diments along the axial drainage (Fig. 2b). The sanctuary
(SAN in Fig. 2b) is located along this westward running seg-
ment of the river. The castle (CHA in Fig. 2b), in the middle
of the city, is built on a 50-m high hill corresponding to a
recessional moraine. Two summits flank the valley to the
south of the city: the Béout (altitude 719 m) to the west
and the Pic du Jer (altitude 1948 m) to the east.
Site Effect Determinations: The H/Href and
H/V Methods
Three methods, based on spectral ratio computations,
are commonly used to determine the soil response to a seis-
mic excitation (e.g., Field and Jacob, 1995): The first one,
considered the most reliable one (Borcherdt, 1970), consists
of computing the spectral ratio between each component of a
site station and of a reference station without site effect, gen-
erally located at a bedrock site without topography. It is
called H/Href for the horizontal components spectral ratio
and V/Vref for the vertical components spectral ratio. The
second method (the horizontal-to-vertical spectral ratio
[HVSR] method) consists of estimating the amplitude spec-
tral ratio of the horizontal-to-vertical components of seismic
S waves from natural earthquakes. Finally, the third method
(H/V) consists of calculating the ratio between the horizon-
tal and vertical component spectra from ambient noise
records (Nakamura, 1989). The first two methods (HVSR
and H/Href) are based on records of natural earthquakes:
their main drawback is to rely on the occurrence of earth-
quakes at some 10–100 km from the station. If the seismicity
is moderate, a long-period recording (a few weeks to a few
months) will be necessary. The third method (H/V) is based
on ambient noise recordings and may thus be applied easily
in urban environments. Its main advantage is its promptitude:
only a few tens of minutes of record are needed, without use
of a reference station. It is thus very convenient for practical
engineering purposes. However, it leads to results that are
more difficult to interpret.
The H/Href method assumes that earthquake induced
ground motion at the bedrock–sediment interface is similar
to that of a nearby station located on rock. Under this as-
sumption, the H/Href ratio gives the transfer function of
the sedimentary layer for a station installed on a sedimentary
layer above bedrock. This assumption hypothesizes that the
incident wave is the same beneath both stations, which im-
plies that the distance between earthquake sources and sta-
tions must be large compared to the distance between the
station and the reference site. Source and propagation effects
could be safely neglected. A difficulty of this method is find-
ing a good reference site, which can effectively be charac-
terized as strong rock, without topographic effect. The
method is generally applied to S waves, which are respon-
sible for most of the damage.
The HVSR method is based on the assumption that, for
a station at a rock site, the earthquake induced H/V ratio
is close to unity (Lermo and Chavez-Garcia, 1993). This
is generally the case for a seismic ray path with incidence
angles close to 45° (thus for local and regional events, again
with distances from a few tens to a few hundreds of kilo-
meters); it is not true for teleseisms, for which rays arrive
nearly vertically beneath the station. These generally give
an S signal larger on horizontal components than on vertical
components (e.g., Kulhánek, 1990). It may be shown that, if
sedimentary layers are present, the amplitude of the vertical
component is only weakly modified by these layers, whereas
the horizontal components are strongly amplified at some
particular frequencies. The HVSR provides the resonance fre-
quency peaks of the transfer function, as well as an estimate
of local amplification for simple structures (Lermo and
Chavez-Garcia, 1993).
The H/V method (Bard, 1999) relies on the nature of the
ambient noise and on the relationships between soil structure
and amplification of the different seismic waves present in
the noise. Noise includes natural microseisms, mainly com-
posed of Rayleigh waves with characteristic periods in the
range of 4 to 10 s, and microtremors due to urban activity.
Urban noise appears predominantly at periods smaller than
1 s and includes body waves (P and S) and surface waves
(Rayleigh and Love waves) generated by traffic, industrial
and domestic activity, and wind. Because such activities
generate mostly superficial compressional sources, the shear
signal will be dominated by the fundamental mode of Ray-
leigh waves. Higher Rayleigh modes, Love modes, and shear
body waves are also present, due in particular to diffraction
on scatterers. The relative contribution of these different
waves (body waves and fundamental and higher modes of
surface waves) in the noise seems, however, to exhibit sig-
nificant variability (see the review by Bonnefoy-Claudet
et al. [2006]).
The Rayleigh-wave ellipticity depends on the structure,
on the period, and on the considered mode (Harkrider, 1970).
For high impedance contrasts at the sediment interfaces,
peaks in the H/V ratio are generated by resonance inside
the uppermost layers. For a single plane layer overlaying
the bedrock and without any topography, the resonance peak
frequency fR can be roughly estimated by fR vS=4h,
where vS is the shear velocity in the layer of thickness h.
It has been shown that the most prominent low-frequency
peak differs generally by less than 10% from the fundamen-
tal S-wave resonance frequency (Bard, 1999). This is no
longer valid if the impedance contrast between superficial
layers and substratum is weak (Malishewski and Scherbaum,
Site Effects in the City of Lourdes, France, from H/V Measurements 2121
5. 2004) or if several layers are present. The peak may be gen-
erated by the vanishing of the horizontal component, for
example, if the Rayleigh-wave ellipticity becomes pro-
grade, which occurs if a thick soft sediment layer is present
(Tanimoto and Rivera, 2005). It may also occur for some
higher modes of Rayleigh waves (e.g., Harkrider, 1970).
Considering now the amplitude of the H/V spectral ratio,
the peak amplitude is not clearly related to the S-wave am-
plification and is generally smaller. It is sensitive to the Pois-
son’s ratio. It also depends on the smoothing function
applied to the spectra (Lachet and Bard, 1994). A ratio close
to unity is expected for the fundamental mode of Rayleigh
waves at conditions close to a half-space (H=V ∼ 0:7, for a
half-space with a Poisson’s ratio of 0.25).
All three methods described in the preceding discussion
have been applied to the city of Lourdes.
The Site Effect Experiments at Lourdes
The 2001 H/Href Experiment (Dubos et al., 2003)
The H/Href method has been applied to data collected
during a seven month experiment performed by Dubos et al.
(2003) at 10 sites located on various geological units in the
city (Fig. 2b). The reference station (ROC) corresponds to a
rocky site with flat topography in the center of the city. Its
good quality as a reference site is confirmed by its flat H/V
ratio over the whole range of frequencies. The H/Href ratios,
computed for the horizontal components in the frequency
range 0.2–25 Hz, reveal very rapid spatial variations of
the site responses inside the city (Fig. 3). The V/Vref ratios
have also been computed for the vertical components: their
maxima are generally at higher frequency than that of the
H/Href spectra (see Dubos et al. [2003], their fig. 7). The
two horizontal components (north and east) exhibit nearly
similar spectra and could thus be combined in a single signal,
according to H N2
E2 1=2
. Some sites exhibit a flat
spectral ratio (HOP and EDF). Most of them have one peak
between 2 and 10 Hz on the horizontal component, which
sometimes broadens (SAN) or splits into two peaks (PMP)
at a high frequency (Fig. 3). For most of the 10 sites, the
H/V ratios could be computed for seismic signal and for noise
(Dubos, 2003). These spectral ratios generally retrieve the
fundamental frequency, but they are unable to restitute the
complexity of the spectra, and they underestimate the peak
amplitude (Fig. 3).
The 2005 Ambient Noise H/V Experiment
In 2005, an ambient noise experiment was conducted by
two teams, one from the Observatoire Midi-Pyrénées (OMP)
and one from the Bureau de Recherches Géologiques et
Minières (French Geological Survey, BRGM). It was con-
ducted during the winter for two reasons: winter conditions
provide a stronger natural microseismic noise, and touristic
activity in summer continuously generates noise very close
Figure 3. Results of the site effect determined from three different methods for the sites reported in Fig. 2b. S-wave spectral ratio from
earthquakes, with reference station (H/Href), solid line; H/V spectral ratio measured on earthquakes (H/V signal), dashed line; H/V spectral
ratio measured on noise (H/V noise), gray line. For clarity, the 1σ confidence level is reported for H/Href only; it is of the same order for the
two other methods (after Dubos, 2003).
2122 A. Souriau, A. Roullé, and C. Ponsolles
6. to recording stations. To further diminish human distur-
bance, most of the records were performed during the night.
Each OMP recording station includes a velocimeter
CDJ-S2A from Chongqing Geological Instruments whose
bandpass is broadened at low frequencies up to 0.1 Hz
(J. F. Fels, personal comm., 2004). For a few sites, a Guralp
CMG-40 velocimeter was also set up. Each seismometer is
connected to a Minititan3 Agecodagis recorder with a sam-
pling rate of 125 Hz. The experimental set-up adopted by
OMP during the survey campaign used two stations a few
meters apart and a real time visual display. Duplicate stations
with visual display prevent recording very local perturba-
tions (e.g., underground pipes). When possible, the seism-
ometers were partly buried in the ground, to remove the
resonance frequencies above 25 Hz (Fig. 4a) caused by the
coupling between the legs and the body of the instrument and
to decrease the perturbing signal due to the wind. Up to three
recordings of 300 sec were performed at each site.
The BRGM operated at each site a 5-sec Lennartz velo-
cimeter connected to a GSR-24 recorder from GeoSig. The
seismometer was up on its legs. The band pass is about 0.2–
40 Hz. The sampling rate is 100 Hz, and the record length is
660 sec.
At each site, the spectra are computed for each compo-
nent using a fast Fourier transform and are smoothed on win-
dows with the width set to 20% of the central frequency. This
removes of instabilities and preserves the resonance peaks.
The two horizontal components are merged according to
H E2
N2 1=2
, which implicitly assumes that they have
no systematic phase coherence. The mean and standard de-
viation are deduced from the processing of tens of indepen-
dent signal sampling windows. Each window includes
4096 points (32 sec) for OMP. The data processing is glob-
ally the same for BRGM, in particular the same smoothing of
the spectra has been applied, so that the results are compa-
tible with each other. As a test, a few sites have been instru-
mented simultaneously by both OMP and BRGM. The results
are similar (Fig. 4b). In some cases H/V ratios increased at
very low frequencies due to strong horizontal shaking caused
by windy conditions during the experiment, in particular for
BRGM measurements: this increase must not be interpreted
as being representative of ground conditions.
Figure 5 maps the 189 points sampled by OMP (dots)
and BRGM (squares) superimposed on the geographic–
topographic map. In the city center, the distance between
points is on the order of 200–300 m. Dense sampling has
also been performed in the valley to the north, because
of the building density. The two mountains to the south
and the forested region to the northwest were only sparsely
sampled, generally in the vicinity of special equipment
(e.g., a garbage collection site or a water pumping or sew-
age works).
In what follows, sites occupied by OMP will be noted by
an (for n from 1 to 111); those occupied by BRGM will be
noted by bm (for m from 1 to 68). The 10 sites of the H/Href
experiments previously carried out by OMP will be noted
with their three-character code.
Experimental Results
The following section presents the results illustrating
specific characteristics of the method, or specific features
of the geology in Lourdes, with their implications for
seismic-risk evaluation.
Bedrock Sites
Bedrock sites in topographically flat areas are character-
ized by flat H/V ratios with an ordinate value close to 1 over
the whole range of frequencies. This property is used to iden-
tify a “good” reference station in the H/Href method. In other
cases, bedrock sites may present spectra that deviate from a
flat line. This occurs when bedrock is altered (Steidl et al.,
1996)—an extreme example of which is karstic limestone
bedrock with underground cavities. Complex spectra also
occur when the site is in rugged topography.
Figure 6 gives H/V ratios for some rock sites in Lourdes,
corresponding to different geological units (see their location
Figure 4. Tests of the experimental method. (a) Influence of the
seismometer set up (on legs or half buried), showing the resonance
frequency due to the coupling between the legs and the body of the
instrument; (b) comparison of the results of OMP and BRGM at the
same site (with 1σ confidence levels).
Site Effects in the City of Lourdes, France, from H/V Measurements 2123
7. in Fig. 5). ROC and a17 are hard Upper Cretaceous coral reef
limestone, a97, a30, and b43 are Middle Cretaceous flysch,
and a57 is located at an ophitic massif. All these sites have
either no or very mild topography. The spectral ratios are
reasonably flat over the whole frequency domain, with am-
plitudes between 1 and 3.
There is no obvious dependence of the spectral ratios on
the type of rock. The station ROC, located on a rock outcrop
in the center of the city, exhibits the typical flat spectrum of a
high quality rock site, which justified its choice as a refer-
ence site for the H/Href method (Dubos et al., 2003). By con-
trast, site a17, which is located on the same type of rock west
Figure 5. Points sampled with the H/V method for site effect determination in the city of Lourdes, operated by OMP, dots, and BRGM,
squares. Ten more points are located in the western part of the Lourdes district (not shown). Numbered sites correspond to rock sites shown in
Fig. 6 (bold numbers) and to the Gave-de-Pau bank sites in Fig. 7 (underlined numbers). The three boxes refer to Figs. 8, 9, and 10. Geo-
logical cross section of Fig. 8 (after BRGM, 2006), AB. Topographic background is from the Institut Géographique National.
2124 A. Souriau, A. Roullé, and C. Ponsolles
8. of the city at the entrance of a cave, exhibits a strong ampli-
fication at high frequencies, likely due to the underground
cavities. The station CHA (Fig. 3) is also located on the same
type of rock but at the flank of a 50-m high hill, mid distance
from the top. The deamplification observed at high fre-
quencies (f > 3 Hz) has been ascribed to a topographic ef-
fect (Dubos et al., 2003). It is also interesting to compare a57
(Fig. 5) with EDF (Fig. 3), two points close together on an
ophitic massif at the foot of a cliff. The main difference
in the spectra concerns the mean value, close to 3 for
57, which is directly at the foot of the cliff, and close to
1, as for a flat rock site, for EDF, which is located only
100 m away. Similar shifts in the mean value have been ob-
served for other points close to cliff walls, either at the top or
at the foot.
Sites Along the Gave-de-Pau River
Quaternary soft sediments of glacial (Würm and Riss)
and fluvial origin bound the Gave-de-Pau river along most
of its course. Few people live there in permanent houses
as this zone is subject to floods. However, some buildings
in the sanctuary, an electric power plant, a sewage works sta-
tion, and several car parks, are built on soft soils along the
Gave-de-Pau. This area thus deserves some consideration for
risk assessment. Figure 7 shows the spectral ratios obtained
for these points, sorted as natural sites (river bank) and man-
modified sites (park places) (see their locations in Fig. 5).
For the natural sites (a81, a67, and a109), a low-fre-
quency peak is observed at 1–3 Hz, with a spectral ratio am-
plitude of 20. The same resonance frequency was obtained at
the sanctuary (Fig. 3), which is also located along the river on
soft sediments. However, the comparison of the H/Href and
the H/V results (Fig. 3) suggests that the resonance frequency
bands may include higher frequencies. A seismic refraction
experiment performed in the sanctuary garden along the
river reveals the presence of a 26-m thick layer with S velo-
city of 300 m sec 1
(Dubos et al., 2003). The relationship
f vS=4h predicts a resonance frequency of 2.9 Hz, in good
agreement with the observation (Fig. 3, site SAN). As will be
seen in the next section, a modeling with synthetic seismo-
grams based on spectral analysis of surface waves (SASW)
results also predicts this peak. SASW experiments performed
by BRGM along the river (on the north side and the west
part of the river) reveal a generally thinner, 10- to 15-m
thick, soft soil layer with mean vS values on the order of
250–300 m sec 1
(BRGM, 2006), which would predict reso-
nance frequencies higher than those observed near the sanc-
tuary and along the south part of the river. It is a clear
indicator of a great spatial variability of the structure. Thus,
comparison of observations and predictions is only possible
for sites with a nearby SASW profile available.
The points sampled on river sides transformed as car
parks show H/V spectra broadened at low frequencies, com-
pared to points on natural sites. Geological data indicate that
it may not be due to an anomalously thick sedimentary layer.
It is very likely due to horizontal modes of resonance of the
concrete work set up to reinforce the stability of the parks, as
the largest park place (site a59) also has the broader reso-
nance domain at low frequencies. A rough estimate of the
fundamental frequency f0 for this l 550 m long park place
may be made using l λ=2, where λ is the wavelength
(λ vS=f0, where vS is S-wave velocity). It leads to fre-
quencies on the order of 0.3–0.5 Hz, significantly lower than
the resonance frequency of the natural sites. Horizontal re-
sonances are also sometimes observed in narrow sedimen-
Figure 6. H/V spectral ratios determined from noise at some rock sites with no significant topography. ROC and a17 are on Cretaceous
limestone (with a cave at a17); a97, a30, and b43 are on Cretaceous flysch, and a57 is on ophite (see location of points in Fig. 5, except a97,
which is located at the very northwest edge of the Lourdes district).
Site Effects in the City of Lourdes, France, from H/V Measurements 2125
9. tary basins (e.g., Cornou et al., 2003); they may perhaps ex-
plain some low-frequency bumps in the spectra, as observed
at 0.3 Hz for a81 (Fig. 7).
The Saux Valley to the North-Northeast
This sediment-filled valley (Fig. 8a) north-northeast
of Lourdes is the former bed of the Gave-de-Pau river, before
moraine deposits deviated its flow to the west. A small creek
is now running along this valley from north to south, with
former swamps at some places, which are now drained
(Monge swamp, site a36). The valley is the main access
to the city, and many buildings of economical interest are
located there (e.g., commercial centers, factories, a swim-
ming pool, an aquarium). From a geological point of view,
the uppermost formation is rather well known thanks to sev-
eral boreholes drilled along the valley (Fig. 8a and 8c) and
SASW experiments performed by BRGM (see locations on
Fig. 8a). SASW results are given in Table 1. They reveal
a structure that has great similarities to the north (near site
a25) and to the very south (near site b13) but that is very
different in the Monge swamp. The sediment thickness is
greatest along the axis of the valley and decreases to zero
at the border. We thus expect responses dependent on the
position of the sites within the valley (see, e.g., King and
Tucker [1984]). Figure 8b shows a series of H/V spectra
along the valley axis. They exhibit a great variability from
north to south. Sites to the north (a24, a25, and b44) and to
the very south (b13) generate two peaks, one at 3–5 Hz, the
other at 10–20 Hz; these peak positions are well predicted by
modeling from the vS profile (see next section). For the sta-
tions in the central part of the profile (a22, a23, a36, and
a26), the spectra include lower frequencies in the range of
1–2 Hz. Basin-edge-induced surface waves (Kawase and
Aki, 1989; Narayan, 2005) are very likely at the origin of
this energy at low frequency, as has also been identified
in the Grenoble Basin in the Alps (Cornou et al., 2003). Note
also the sharp resonance peak at a low frequency inside the
Monge swamp (site a36, fR 0:9 Hz) and the relatively flat
response for b15, located on a small promontory.
Two parameters can contribute to the spatial variations
observed on the H/V spectral ratio within the Saux valley: the
effect of the topography (considering both the shape of the
sediment–bedrock interface and the shape of the hills) and
the spatial variations of sediment properties. Several theore-
tical studies have aimed to model the topographic effect of a
valley on S-wave or Rayleigh-wave propagation (e.g., King
and Tucker, 1984; Geli et al., 1988; Kawase and Aki, 1989;
Sánchez-Sesma and Campillo, 1993; Savage, 2004). In the
absence of filling by sedimentary layers, the horizontal
motion of the Rayleigh wave is strongly deamplified at
the bottom of valleys, compared to a flat half-space, for
wavelengths of the same order of magnitude as the valley
width (Savage, 2004). On the other hand, the horizontal com-
ponent of the S wave exhibits a complex amplification–
deamplification pattern, which is nonsymmetrical about
Figure 7. H/V spectral ratios determined from noise for sites along the Gave-de-Pau river. (a) Natural sites: the peaks are due to the
vertical resonance in the sedimentary layers; (b) Sites on park places: low-frequency ratios are possibly due to horizontal resonance modes
(see location of points in Fig. 5).
2126 A. Souriau, A. Roullé, and C. Ponsolles
10. the valley axis, depending on the direction of wave arrival,
on the nature of the incident wave (SV or SH), and on the
wave incidence (Sánchez-Sesma and Campillo, 1993). How-
ever, the filling of the valley seems to contribute more to
amplitude perturbation than to topography (Geli et al.,
1988). The spatial variability of the H/V spectral ratio in
Figure 8. (a) H/V experiment along the Saux valley to the north-northeast of Lourdes (see location in Fig. 5) with sampled sites, dots;
soundings, diamonds; SASW experiments, triangles; and the northern segment A′B of the geological cross section AB. (b) Some of the
representative H/V spectral ratios determined from noise. (c) Geological cross section with indication of the soundings and locations of
sites. Topographic background is from the Institut Géographique National.
Site Effects in the City of Lourdes, France, from H/V Measurements 2127
11. the Saux valley may then be primarily related to the spatial
variability of the geological layers.
The Sarsan Hill
The Sarsan hill is a topographic feature located to the
northeast of the city (Fig. 9a). It is made of Campanian
(Upper Cretaceous) flysch. It culminates at 200 m above
the valley floor and has a half-width of 900 m (Fig. 9b).
Houses are built on the southwest flank of the hill, which
is the only one that could be sampled with an easy access.
The H/V spectral ratio (Fig. 9c) exhibits a systematic
trend from the base (b11) to the top (a21). At the base, a
weak high-frequency peak (∼7 Hz) is observed. In the lower
half of the hill, this peak broadens and presents lower fre-
quencies with increasing height. In the upper half of the hill,
the peak weakens and the spectrum loses its low-frequency
content. At 50 m from the summit, the spectrum becomes
almost flat. Note, however, that for this last point (a21)
the ratio is 2 and not 1, as it would normally be for rocky
sites in flat areas.
The complex amplification and deamplification pattern
observed on the Sarsan hill has been compared to simulations
of topographic effects on hills. Most of the existing models
concern the amplitude of the horizontal S motion, not the H/V
ratio. They predict an amplification of the horizontal signal at
hill tops for both SH waves and Rayleigh waves (e.g., Geli
Table 1
Mean Uppermost Structure in the Saux Valley to the North-
Northeast of Lourdes (after BRGM [2006])
Rock Type Thickness (m) vS (m s 1)
North Sand and gravels 10 300
Gravels 26 630
Substratum — 860
Monge swamp Peat 8.4 220
Mud 8.6 170
Substratum — —
South Sandy loam 5 300
Clayey mud 8 650
Substratum — 840
Figure 9. Results for the Sarsan hill, to the northeast of Lourdes (see location in Fig. 5). (a) Sampled sites of the H/V experiment, dots;
soundings, diamonds; and SASW experiment, triangle. (b) Topographic profile: origin of distances corresponds to the hill foot. (c) H/V
spectral ratios. Topographic background is from the Institut Géographique National.
2128 A. Souriau, A. Roullé, and C. Ponsolles
12. et al., 1988; Sánchez-Sesma and Campillo, 1993; Bouchon
et al., 1996; Savage, 2004). Amplification at the top is usual-
ly observed for wavelengths comparable to the hill width,
which corresponds to the fundamental transverse oscillatory
resonance mode of the hill (Geli et al., 1988). The Rayleigh-
wave model of Savage for a ridgelike topography predicts an
increase of horizontal amplitude and a reduction of vertical
amplitude at the ridge top, thus a high H/V ratio. At the same
time, it predicts a deamplification at the base of the hill. This
is very different from our observations.
As noted by Bouchon and Barker (1996), the response
of the hill is very sensitive to topographic details, and it is
difficult to model the observed resonance pattern for real si-
tuations. For a structure somewhat similar to the Sarsan hill,
but with 10 times the reduction in size (height 20 m; half-
width 90 m) and also a small plateau at the top, Bouchon
and Barker (1996) performed a numerical simulation with
SH-waves input. At low frequencies, they predict amplifica-
tion at the hill top whereas, for high frequencies, they predict
a deamplification at the top and amplification in the upper
part of the flank. We observe a more complex pattern. This
discrepancy between this model and our observations could
be due to the fact that urban noise includes mostly Rayleigh
waves, so that these modeled results cannot be easily com-
pared to our field observations.
The Anclades-Pic-du-Jer Profile
This profile crosses the valley to the east and part of the
mountain to the southeast. For reasons of site access, the pro-
file bends and does not follow the steepest slope of the
mountain (Fig. 10a). The highest point (a84) is therefore lo-
cated at a pass rather than on the summit. The mountain is
made of folded Cretaceous marls and reef limestone, with
some glacial deposits to the east. The valley floor at the bot-
tom is flat and filled with sediments. Four SASW profiles
were performed in the middle and on the edges of the valley.
The mean structure includes a 10-m thick sand layer over-
laying a 35-m thick compact loam (Table 2). The substratum
is reached at a depth of 37 m to the south of the valley and at
45 m in its central part.
The H/V spectral ratios exhibit a clear single peak for the
sites located in the valley. The central frequency of this peak
decreases from 5–7 Hz for the points located at the border of
the valley (b22 to the north and a35 to the south) to about
1 Hz for the point located close to the valley axis (a33), in
agreement with the increasing sediment thickness. At the
same time, the peak width broadens, suggesting reverbera-
tions due to a more complex sediment stack, to 3D effects,
or to waves bouncing off the basin edges.
For points sited on the mountain flank, the spectra are
complex and vary from one site to the other, with two to three
peaks in the frequency range of 0.2–20 Hz. Such amplifica-
tion variations with oscillations are expected even for homo-
geneous mountains, at least for an incident SH wave (Geli
et al., 1988). Site a83 may be affected by the presence of
glacial deposits, whereas the other sites are on hard rock.
The peak at about 1.5 Hz (sites a84 and a85) corresponds
to the resonance of the mountain in its transverse direction.
As for the points located along the river, a rough estimate of
the resonance frequency f0 may be made using the relation-
ship l λ=2, where l is the mountain width and λ vS=f0.
With a mean P velocity of 2:8 km sec 1
in the limestone
(Press, 1966), a vP=vS ratio of 1.7, and a mountain width
of 1.5 km (estimated at altitude 550 m), we get
f0 1:1 Hz, which is close to the observed value. However,
we do not have enough knowledge of the 3D-geological
structure to model its response with any confidence.
1D Modeling of H/V with Synthetic Seismograms
Modeling may help to better estimate the sensitivity of
the H/V ratio to small structure perturbations. It allows us to
check the ability of the H/V ratio to predict the effective re-
sonance frequencies and amplitudes. It also gives insights on
the nature of the urban noise.
We have noted previously that the simple relation fR
vS=4h holds true only for a vertical resonance inside a sin-
gle layer with a strong impedance contrast with the substra-
tum. If this is not the case, or if several sediment layers are
present, more sophisticated modeling is necessary. We have
performed two kinds of modeling: a modeling of the H/V
ratio based on the computation of noise from synthetic seis-
mograms and a computation of the H/Href ratio of S waves
based on the vertical propagation of a real event though a
soil profile.
Methods
Generation of Noise from Synthetic Seismograms
Following Lachet and Bard (1994), urban noise is gen-
erated from a summation of synthetic seismograms. It allows
us to generate simultaneously surface modes and body
waves, which probably have a nonnegligible contribution
as noted previously for bedrock sites.
Noise has been generated by summing 1000 synthetic
seismograms with a 1024-points length and a sampling rate
of 125 Hz (8.2 sec long elementary seismograms). For a gi-
ven structure, each elementary seismogram is generated
using the reflectivity method (Müller, 1985), with parameters
chosen randomly inside specified domains. The variable
parameters are the distance, the type of source, the far-field
shape, the dominant frequency, the amplitude, and the time
delay. The sources are randomly distributed around the site at
distances between 100 and 500 m. As in the field experiment,
we have avoided sources too close to the recording station.
Sources are either explosions or single vertical forces. De-
spite the fact that shear sources are supposed to be marginally
present in the generation of urban noise, we also perform
simulations including a proportion of pure shear sources
in order to simulate conversions from compressional energy
Site Effects in the City of Lourdes, France, from H/V Measurements 2129
13. to shear energy at scatterers and at basin edges. Three shapes
of far-field sources are possible: a single pulse, a single os-
cillation, or a three-arch oscillation. Their dominant period
ranges from 0.03 to 0.5 sec with a logarithmic distribution,
in order to generate more high frequencies than low frequen-
cies, like in real noise. The summation of the 1000 seismo-
grams is performed ascribing an amplitude between 1 and 1
and a time delay between 0 and 4096 points (0 to 32.8 sec) to
each of them. The length of the synthetic records obtained by
summation is 4096 points, as for the experimental ones. Ten
different noise series of 4096 points have been generated in
this way for each structure and have been processed exactly
in the same way as the experimental data to obtain a mean
and standard deviation of each spectrum.
Propagation of an Earthquake through a Soil Profile
Another approach to modeling site effect is to propagate
vertically a real or synthetic earthquake signal (used as input
on bedrock at the bottom of the soil column) through the
1D known structure and to analyze the output signal at
the surface. A nonlinear 1D dynamic model was developed
Figure 10. Results for a profile perpendicular to the Anclades valley to the east (see location in Fig. 5) with an uphill segment along the
Pic-du-Jer. (a) Topographic map with H/V sites, dots; soundings, diamonds; and SASW experiments, triangles. (b) Topography along the
profile. (c) H/V ratio for some of the sites, showing the decrease of the peak frequency in the middle of the valley and the complex response on
the hill flank. Topographic background is from the Institut Géographique National.
2130 A. Souriau, A. Roullé, and C. Ponsolles
14. at BRGM to simulate the response of soil profiles under
various hydraulic conditions (e.g., Bernardie et al., 2006);
it is implemented in a commercial piece of software, Cyber-
Quake (Modaressi et al., 1995). The main advantage of this
approach is to take into account the mechanical soil proper-
ties at the origin of nonlinear responses of the soils, giving
a realistic simulation of soil response during both weak
and strong earthquakes. The results allow us to estimate
the H/Href ratio (i.e., the transfer function of the sediment
layer) using the S-wave input signal, at the bottom of the
column, as reference. In our case, output signals were used
to compare the experimental H/V response with the predicted
H/Href resonance frequencies for earthquakes, thus, to eval-
uate the credibility of our analyses for seismic-risk assess-
ment. The chosen input is an event of M 5.0 (17 Novem-
ber 2006) recorded at a distance of 20 km from Lourdes
at a rock site station (PYLS) without site effect (Drouet
et al., 2005).
Modeling Results
All the results we present in the following discussion
concern geological structures without topography. Forward
modeling of synthetic spectral ratios based on SASW results
will be compared to the experimental spectral ratios obtained
at nearby sites.
Bedrock Sites
Synthetic noise has been computed for the rock model
shown in Figure 11a (right). The H/V ratio computed for this
model is close to 1 at any frequency (Fig. 11a, left), in agree-
ment with the experimental values. The weak frequency de-
pendence of the modeled H/V ratio is due to arbitrary
multilayering in the substratum of our model, which is intro-
duced for computational reasons.
As previously mentioned, the H/V ratio expected for a
pure fundamental Rayleigh mode is close to 0.7 (Rayleigh-
wave ellipticity) at typical bedrock sites (considered as a
half-space) with Poisson’s ratios in the range of 0.23–
0.27. For the fundamental Rayleigh mode, an ellipticity close
to unity (H=V 1) may be obtained only in the unrealistic
case of a negative Poisson’s ratio (Malischewski and Scher-
baum, 2004). By contrast, S waves from local sources may
induce large H/V values as well higher modes of Rayleigh
waves (e.g., Harkrider, 1970; remember, however, that high-
er modes are not excited in a real half-space). Therefore, one
has to invoke the presence of body waves in the noise to
explain the H/V ratio of 1. The independence of Rayleigh-
wave ellipticity and body-wave incidence to frequency in
the case of a half-space explains why the H/V ratio is flat
over the whole frequency domain. The introduction of a
shear sources has no effect (dashed line in Fig. 11a), because
Love waves are not generated in a half-space.
The Gave Bank at the Sanctuary
A 1D model is given down to 30 m by a SASW profile
(Fig. 11b, right) performed close to site SAN (Fig. 2). Fig-
ure 11b (left, gray line) shows the H/V ratio predicted by the
SASW model when sources do not include shear (here, arbi-
trary, 50% explosion, and 50% single vertical force; this bal-
ance has no critical influence). The observed peak position
and the amplification close to 5 of the H/V ratio are rather
well reproduced by the model. By contrast, the slow decrease
at low frequency is poorly reproduced if no shear is present
in the noise, due to a deficit of shear energy in the signal at
low frequencies. If sources include pure shear energy (here,
one-third pure shear, one-third explosion, and one-third sin-
gle force), the low-frequency trend is much better reproduced
(dashed line), and at the same time the peak is slightly shifted
toward lower frequencies. This fits very well with the obser-
vations. The fit of the low-frequency trend is, however,
rather sensitive to the proportion of shear energy.
The simulation of the H/Href ratio with CyberQuake
(Fig. 12a, gray line) shows several peaks of resonance with,
for the fundamental peak, an amplification that exceeds 10.
The frequency of this peak is well expressed on the H/V ratio
of SAN, but its amplitude is strongly underestimated. This
was also the conclusion drawn from the comparison of H/V
and H/Href on natural earthquakes (Fig. 3) at the same site
and for eight other sites in Lourdes. Note that the H/Href
ratio shape predicted from CyberQuake is significantly dif-
ferent from that measured from earthquakes (dashed line),
where there is a plateau between 1.5 and 20 Hz, as observed
in the case of dipping interfaces (a case which cannot be si-
mulated by 1D modeling). However, the maximum ampli-
tudes of the two spectra are similar.
The Saux Valley to the North
The numerous SASW profiles performed in the valley
(Fig. 8) have revealed a great variability in the thickness
and nature of the top 50 m (Table 1); thus, a 1D modeling
may not be appropriate. We present here the results for only
the northernmost point (a24). The corresponding SASW pro-
file (Fig. 11c, right) is defined down to 37 m, but it presents a
velocity inversion near 10 m, so that its reliability at a greater
depth may be poorer than for the sanctuary. The agreement of
the model with the H/V observations (Fig. 11c, left)
is, however, globally satisfactory. The experimental low-
frequency peak position is well reproduced, but its ampli-
tude is slightly overestimated. Again, the inclusion of shear
Table 2
Mean Uppermost Structure in the Anclades Valley to the East (after
BRGM [2006])
Rock Type Thickness (m) vS (m sec 1)
Sand 10 240
Loam 35 445
Substratum (Upper Cretaceous flysch) — 800
Site Effects in the City of Lourdes, France, from H/V Measurements 2131
15. Figure 11. Left: Modeling of the H/V ratio from urban noise generated by a summation of a large number of synthetic seismograms with
randomly defined characteristics. Only compressional sources, gray line; compressional and shear sources (see text), dashed line; the 1σ
confidence level is reported. Observed H/V ratio, black line. Right: vS structure deduced from SASW profiles. (a) Rock site; (b) sanctuary;
(c) Saux valley; (d) Anclades basin.
2132 A. Souriau, A. Roullé, and C. Ponsolles
16. source signals as for the sanctuary (dashed line) leads to a
better fit of the observations at low frequencies; moreover,
it generates the peak at 15 Hz, but with an amplitude much
lower than the observed one.
The simulation of the H/Href ratio from CyberQuake
again reveals that the H/V ratio is able to predict the position
of only the lowest frequency peak and that it underestimates
the expected amplification.
The Anclades Basin to the East
Three SASW profiles are available (Fig. 10), but LOU02
does not exceed 15 meters, and LOU25 exhibits poorly con-
trolled velocity inversions. We thus considered LOU01,
although it is located at the basin’s edge. This profile is de-
fined down to 37 m (Fig. 11d, right). As for the sanctuary,
the agreement between the observed and the predicted H/V
ratio is quite good when shear sources are included in the
noise (Fig. 11d). The H/Href simulation (Fig. 12) again
shows very nicely that the H/V ratio detects correctly the re-
sonance peak at the lowest frequency but that it underesti-
mates the resonance amplitude.
Synthesis and Discussion of H/V Results
In the previous sections, we have presented the results
for a few particular geological and topographic structures of
interest in Lourdes. We have also shown that some of the
observed features of the H/V spectra may be correctly repli-
cated with simple 1D models but that in many cases model-
ing is limited by complex 3D structures that are poorly
known. Thus, site effects for seismic-risk evaluation will still
rely on experimental results.
A global analysis of the results is made on the basis of
the frequencies and amplitudes of the highest peaks observed
in the spectra. Figure 13 gives a map of the H/V sites with
these spectrum characteristics. Stations without site ef-
fects are those for which amplitude remains below 2 over
the whole spectrum (0.5–20 Hz). For the other sites, ampli-
tudes (amp) are represented with two classes: amp < 5
and amp ≥ 5. Five classes are considered for the dominant
frequency: the highest frequency domain being for f >
10 Hz, and the lowest one being for f < 2 Hz. Roughly,
the highest frequency range corresponds to a frequency do-
main that has almost no impact on buildings; the lowest fre-
quency range concerns five-storey and taller buildings; and
the other classes concern individual houses and small build-
ings. This statement is based on the relationship fR 10=N,
where fR is the resonance frequency (in Hz) of the struc-
ture and N is the number of storeys, a relationship which
holds for only one standard type of reinforced concrete build-
ing. For seismic-risk assessment, we have to keep in mind
that the H/V method gives the position of only the lowest
frequency peak and that the H/V peak amplitude is lower than
the real soil amplification (Figs. 3 and 12). Additional infor-
mation, concerning in particular geology and seismic ground
velocities, are thus necessary to establish maps of seismic
risk (BRGM, 2006). Here, we will limit our analysis to the
H/V spectra.
Figure 13 reveals large zones where coherent features
are observed. The points without site effect (squares) or with
a very small amplification (open circles) are mostly located
on rock areas to the north and southwest of the city. In the
city center, large amplifications in the frequency range of
Figure 12. H/Href ratios of S waves modeled from the propa-
gation of an earthquake through the soil profiles given in Figure 11,
gray line, compared to the experimental H/V ratios, black line.
For the sanctuary, the experimental H/Href ratio is also reported,
dashed line.
Site Effects in the City of Lourdes, France, from H/V Measurements 2133
17. 1–4 Hz are generally observed (see also Fig. 3), but the am-
plifications tend to decrease to the edges of the basin. Low
resonance frequencies with moderate amplifications (small
black dots) are also observed to the west-northwest of the
city along the Lourdes lake, which is contained in a basin
of thick recent deposits. Results in the valley to the north-
northeast are very scattered, with a predominance of reso-
nance frequencies in the range of 4–6 Hz, but again we
observe generally smaller amplifications at the borders of
the valley. It may be related to a thinning of the sedimentary
layers. Points immediately along the river correspond to
large amplifications with various resonance frequencies.
Sites south of the river along the east–west oriented segment
of the river bed exhibit rather homogeneous characteristics,
with moderate amplifications (amp < 5) in the frequency
range of 2–6 Hz. From a general point of view, the regions
with similar spectra are also those with similar geological
characteristics; the topography seems to be of secondary
importance.
Conclusion
In the city of Lourdes, H/V measurements performed on
urban noise have been used to get a dense spatial sampling of
site effects. They come in complement to H/Href measure-
ments, applied to S waves of natural earthquakes, which were
previously obtained at a few places in the city. If the H/V
method is easy and rapid to implement, it has, however,
some drawbacks. Its limitations for predicting the site re-
sponse during earthquakes is clearly shown by the compar-
ison of H/V ratios with experimental and synthetic H/Href
values: H/V spectral ratios predict correctly the position of
the lowest resonance frequency peak but not the peak width
nor the amplification value, which is generally underesti-
mated. It is, however, a valuable tool for seismic-risk assess-
ment so long as it is used in combination with other methods.
Because of the difficulty to perform reliable numerical mod-
eling, due to the natural complexity of surficial geology, the
H/V measurements on noise will long be the most efficient
method to quantify soil responses inside cities. In the case of
Figure 13. Summary of the results of the H/V ratios with indication of the dominant frequency, depicted by the level of gray, and the H/V
amplitude at this frequency, depicted by the size of the dot. Sites for which a resonance frequency could not be defined, squares. Topographic
background is from the Institut Géographique National.
2134 A. Souriau, A. Roullé, and C. Ponsolles
18. 1D simple geological settings, two numerical methods based
on (1) the modeling of urban noise from synthetic seismo-
grams, and (2) on the propagation of earthquake data through
a 1D soil column show good coherency between geological
data, SASW results, and H/V measurements.
In the case of Lourdes, we have been able to identify
regions with similar characteristics of the H/V ratios corre-
sponding generally to regions with similar geological char-
acteristics. The comparison between the simulation results
and the different kind of available observations or experi-
mental results (e.g., H/Href, SASW profiles, and soundings)
shows that, despite their limitations, the measurements of the
H/V-noise ratios are very informative for seismic-risk evalua-
tion because they provide a confident estimate of the reso-
nance frequency. They are thus a key input in urban seismic
microzonation, in the elaboration of a large earthquake sce-
nario, and in seismic hazard mitigation.
The H/V method remains mostly empirical: first because
the input signal is still poorly understood and second because
there have been few theoretical studies of standard situations
such as hills, valleys, cliffs, karsts, etc. We have shown from
1D modeling with synthetic seismograms at a few sites
that the urban noise cannot be reduced to the fundamental
Rayleigh mode. Noise includes body waves, higher modes
of surface waves, and an important contribution of shear en-
ergy. A complete modeling of the observations will require
not only an accurate knowledge of the 3D structures and to-
pography, but also a complete model of all the waves that
propagate through these structures.
Acknowledgments
Financial support has been provided by the French Ministry of Ecol-
ogy, by the Bureau de Recherches Géologiques et Minières, and by the city
of Lourdes. We thank Sébastien Benahmed and Stéphane Drouet for their
participation in the field experiments and the city of Lourdes for its logistical
help. A special thank you to Georges Delpont for his interpretation of the
geological data. Many thanks also to two anonymous reviewers and to the
Associate Editor Arben Pitarka for their helpful comments on the manuscript
and to Thomas Dewez for his help in correcting it. Topographic maps are
from the French Institut Géographique National (IGN), Saint-Mandé, France
(Convention 8869/IGN-BRGM); Geological maps are from the Bureau de
Recherches Géologiques et Minières (BRGM), Orléans, France.
References
Bard, P.-Y. (1999). Microtremor measurements: a tool for site effect estima-
tion? in The Effects of Surface Geology on Seismic Motion, K. Irikura,
K. Kudo, H. Okada and T. Sasatani (Editors), Balkema, Rotterdam.
Bernadie, S., E. Foerster, and H. Modaressi (2006). Non-linear site response
simulations in the Chang-Hwa region during the 1999 Chi-chi earth-
quake, Taiwan, Soil Dyn. Earthq. Eng. 26, 1038–1048.
Bonnefoy-Claudet, S., F. Cotton, and P.-Y. Bard (2006). The nature of noise
wavefield and its applications for site effects studies. A literature re-
view, Earth Sci. Rev., 79, 205–227, doi 10.1016/j.earscirev.2006
.07.004.
Borcherdt, R. D. (1970). Effects of local geology on ground motion near
San Francisco Bay, Bull. Seismol. Soc. Am. 60, 29–81.
Bouchon, M., and J. S. Barker (1996). Seismic response of a hill: the ex-
ample of Tarzana, California, Bull. Seismol. Soc. Am. 86, 66–72.
Bouchon, M., C. A. Schultz, and M. N. Toksöz (1996). Effect of three-
dimensional topography on seismic motion, J. Geophys. Res. 101,
5835–5846.
Bureau de Recherches Géologiques et Minières (BRGM) (2006). BRGM/
RP-53846-FR Microzonage sismique de Lourdes, Rapport Final.
222 pp.
Choukroune, P. (1992). Tectonic evolution of the Pyrenees, Annu. Rev. Earth
Planet. Lett. 20, 143–158.
Cornou, C., P. Y. Bard, and M. Dietrich (2003). Contribution of dense array
analysis to the identification and quantification of basin-edge-induced
waves, part II: Application to Grenoble basin (French Alps), Bull. Seis-
mol. Soc. Am. 93, 2624–2648.
Drouet, S., A. Souriau, and F. Cotton (2005). Attenuation, seismic moments,
and site effects for weak-motion events: application to the Pyrenees,
Bull. Seismol. Soc. Am. 95, 1731–1748.
Dubos, N. (2003). Contribution à l’évaluation du risque sismique dans les
Pyrénées centrales, Ph.D. Thesis, University Paul Sabatier, Toulouse
III, 211 pp.
Dubos, N., A. Souriau, C. Ponsolles, and J. F. Fels (2003). Etude des effets
de site dans la ville de Lourdes (Pyrénées, France) par la méthode des
rapports spectraux, Bull. Soc. Géol. Fr. 174, 33–44.
Field, E. H., and K. H. Jacob (1995). A comparison and test of various site-
response estimation techniques, including three that are not reference-
site dependent, Bull. Seismol. Soc. Am. 85, 1127–1143.
Geli, L., P.-Y. Bard, and B. Jullien (1988). The effect of topography on
earthquake ground motion: a review and new results, Bull. Seismol.
Soc. Am. 78, 42–63.
Harkrider, D. G. (1970). Surface waves in multilayered elastic media, part II:
Higher mode spectra and spectral ratios from point sources in plane
layered Earth models, Bull. Seismol. Soc. Am. 60, 1937–1987.
Hirn, A., M. Daignières, J. Gallart, and M. Vadell (1980). Explosion seismic
sounding of throws and dips in the continental Moho, Geophys. Res.
Lett. 7, 263–266.
Kawase, H., and K. Aki (1989). A study on the response of a soft basin for
incident P, S and Rayleigh waves with special reference to the long
duration observed in Mexico City, Bull. Seismol. Soc. Am. 79,
1361–1382.
King, J. L., and B. E. Tucker (1984). Observed variations of earthquake mo-
tion across a sediment-filled valley, Bull. Seismol. Soc. Am. 74,
136–151.
Kulhánek, O. (1990). Anatomy of seismograms, in Developments in Solid
Earth Geophysics, Vol. 18, Elsevier, Amsterdam, 178 pp.
Lachet, C., and P.-Y. Bard (1994). Numerical and theoretical investigations
on the possibilities and limitations of Nakamura’s technique, J. Phys.
Earth 42, 377–397.
Lambert, J., A. Levret-Albaret, M. Cushing, and C. Durouchoux (1996).
Mille Ans de Séismes en France, Ouest Editions, Nantes, France,
70 pp.
Lermo, J., and F. J. Chavez-Garcia (1993). Site effect evaluation using spec-
tral ratios with only one station, Bull. Seismol. Soc. Am. 83, 1574–
1594.
Levret, A., J. C. Backe, and M. Cushing (1994). Atlas of macroseismic maps
for French earthquakes with their principal characteristics, Nat. Ha-
zard 10, 19–46.
Malischewsky, P. G., and F. Scherbaum (2004). Love’s formula and H/V-
ratio (ellipticity) of Rayleigh waves, Wave Motion 40, 57–67.
Medvedev, S., W. Sponheuer, and V. Karnik (1967). Seismic intensity scale
version 1964, in Inst. Geody. Publ., Vol. 48, Inst. Geody., Jena,
Germany.
Modaressi, H., E. Foerster, D. Aubry, and A. Modaressi (1995). Research
and professional computer-aided dynamic analysis of soils, First
International Conference on Earthquake Geotechnical Engineering,
Tokyo, Japan, 1171–1176.
Müller, G. (1985). The reflectivity method: a tutorial, J. Geophys. 58, 153–
174.
Site Effects in the City of Lourdes, France, from H/V Measurements 2135
19. Nakamura, Y. (1989). A method for dynamic characteristics estimation of
subsurface using microtremor on the ground surface, Q. Rep. Railw.
Tech. Res. Inst. 30, no. 1, 25–33.
Narayan, J. P. (2005). Study of basin-edge effects on the ground motion
characteristics using 2.5-D modelling, Pure Appl. Geophys. 162,
273–289.
Nocquet, J. M., and E. Calais (2003). Crustal velocity field for Western Eur-
ope from permanent GPS array solutions, 1996–2001, Geophys. J. Int.
154, 72–88.
Press, F. (1966). Seismic velocities, in Handbook of Physical Constants,
S. P. Clark (Editor), Publications of the Geological Society of Amer-
ica, Vol. 97, Yale Univ. Press, New Haven, Connecticut, 195–218.
Rigo, A., A. Souriau, N. Dubos, M. Sylvander, and C. Ponsolles (2005).
Seismotectonic interpretation of a microseismic analysis in the central
part of the Pyrenees (France), J. Seism. 9, 211–222.
Sánchez-Sesma, F. J., and M. Campillo (1993). Topographic effects for in-
cident P, SV and Rayleigh vaves, Tectonophys. 218, 113–125.
Savage, W. Z. (2004). An exact solution of effects of topography on free
surface waves, Bull. Seismol. Soc. Am. 94, 1706–1727.
Souriau, A., and H. Pauchet (1998). A new synthesis of the Pyrenean seis-
micity and its tectonic implications, Tectonophys. 290, 221–244.
Steidl, J. H., A. G. Tumarkin, and R. J. Archuleta (1996). What is a reference
site, Bull. Seismol. Soc. Am. 86, 1733–1748.
Tanimoto, T., and L. Rivera (2005). Prograde Rayleigh wave particule mo-
tion. Geophys. J. Int. 162, 399–405.
CNRS, Laboratoire de Dynamique Terrestre et Planétaire
Observatoire Midi-Pyrénées
14, Avenue Edouard Belin
31400 Toulouse, France
(A.S., C.P.)
BRGM/ARN/RIS
3 Avenue Claude Guillemin
45060 Orléans, Cedex 2, France
(A.R.)
Manuscript received 25 October 2006
2136 A. Souriau, A. Roullé, and C. Ponsolles