You are on page 1of 12

Geophys. J. Int. (2022) 229, 299–310 https://doi.org/10.

1093/gji/ggab473
Advance Access publication 2021 November 23
GJI Seismology

Monitoring the changing seismic site response of a fast-moving


rockslide (Brienz/Brinzauls, Switzerland)

Mauro Häusler ,1 Valentin Gischig,2 Reto Thöny,3 Franziska Glueer1 and Fäh Donat1
1 SwissSeismological Service, ETH Zurich, Zurich 8092, Switzerland. E-mail: mauro.haeusler@sed.ethz.ch
2 CSD Ingenieure AG, Berne 3097, Switzerland
3 BTG Büro für Technische Geologie AG, Sargans 7320, Switzerland

Accepted 2021 November 17. Received 2021 November 17; in original form 2021 June 21

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


SUMMARY
Seismic measurements on unstable rock slopes are a complementary tool to surface displace-
ment surveys to characterize and monitor landslides. A key parameter is seismic amplification,
which tends to scale with the degree of rock mass degradation. Amplification also provides a
direct measure of how the wavefield is intensified during seismic loading, eventually leading to
coseismic failure. Here we present the dynamic response of the fast-moving Brienz/Brinzauls
rock slope instability in Switzerland (10 × 106 to 25 × 106 m3 ), which threatens settlements
and infrastructure in the area. The rockslide shows strong seismic amplification at two resonant
frequencies with factors of up to 11 and wavefield polarization influenced by the local fracture
network orientation. We monitored the dynamic response over a period of 30 months using
ambient vibrations and regional earthquake recordings. We observed a change in wavefield
polarization of up to 50◦ , coinciding with a rotation of the relative surface displacement vector
field measured by geodetic systems, highlighting the linkage between wavefield polarization
and stress field (i.e. rock mass kinematics). For the analysis of secondary, relative surface
displacements, we propose a singular value filtering of the displacement field to remove the
principal component of landslide motion. In addition, we found increased seismic amplifica-
tion values after periods of strong precipitation, providing empirical field evidence that the
local precipitation history is a key parameter for assessing the hazard of earthquake-induced
slope failure.
Key words: Geodetic instrumentation; Earthquake hazards; Seismic noise; Site effects;
Surface waves and free oscillations.

are not representative of the kinematics at larger depth. Since the


1 I N T RO D U C T I O N
acceleration of rockslides is often sensitive to hydrological forcing,
Deep-seated gravitational slope deformations can produce signifi- alternative approaches involve modeling of the acceleration based
cant socio-economical costs due to progressive damage and result- on environmental parameters, such as precipitation and pore water
ing maintenance of engineered structures such as lifelines and build- pressure (Fernández-Merodo et al. 2014; Vallet et al. 2016).
ings. Moreover, such instabilities are prone to secondary mass move- In the last decade, ambient vibrations have been increasingly
ments and can eventually accelerate towards catastrophic failure. used to characterize and monitor natural landforms (e.g. Larose et
With the advent of airborne and satellite based remote sensing, al. 2015; Bottelin et al. 2017; Burjánek et al. 2018; Kleinbrod et
slope instabilities can be rapidly detected and mapped with high al. 2019; Geimer et al. 2020; Häusler et al. 2021). Decreasing rock
spatial and temporal resolution over large regional extents (e.g. stiffness prior to a collapse (e.g. due to failing rock bridges) was ob-
Jaboyedoff et al. 2012; Casagli et al. 2016; Bickel et al. 2018; Dini served to result in a decrease in resonant frequency and shear wave
et al. 2019). Nonetheless, to monitor a specific high-risk slope and to velocity and can serve as a precursor for slope instabilities (Lévy et
investigate catastrophic failure scenarios, extensive on-site survey- al. 2010; Fiolleau et al. 2020; Colombero et al. 2021; Le Breton et
ing is indispensable. Such monitoring is usually based on measuring al. 2021). Bontemps et al. (2020) observed a drop in seismic shear
surface displacements, either by optical systems or radar-based tech- wave velocity in a soil landslide during phases of acceleration and
nology (Loew et al. 2017; Manconi et al. 2018; Glueer et al. 2019). as a result of earthquake loading. Seismic loading itself can act as
Although surface displacements might allow geologists to estimate trigger for slope failure, resulting in the most fatal coseismic hazard
failure, the prediction can be biased if the surface displacements besides tsunamis (e.g. Budimir et al. 2014). Numerical models and


C The Author(s) 2021. Published by Oxford University Press on behalf of The Royal Astronomical Society. This is an Open Access

article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which
permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
299
300 M. Häusler et al.

observations of coseismic landslides suggest a strong impact of en- ∼2.5 × 106 m3 , called ‘Igl Rutsch’, was released in 1878, reaching
vironmental influences and local geomechanical conditions on the average velocities of about 1 m d–1 with increased activity during
seismic stability (Gischig et al. 2015; Massey et al. 2018; Nowicki spring and fall of up to 5 m d–1 (Heim 1932). Fig. 1(d) summarizes
Jessee et al. 2018; Bradley et al. 2019). However, in situ monitoring the geological model of the slope instability. Displacement rates
of seismic site responses of rock instabilities remains rare. of both subsystems show simultaneous phases of acceleration, and
In this study, we explore the potential of ambient vibration and observations of short-term accelerations and local rock fall activity
earthquake monitoring on the increasingly active slope instabil- indicate a strong correlation with precipitation rates.
ity at Brinzauls (Switzerland). We apply array-based seismolog-
ical mapping to analyse amplification and normal modes, which
helps characterize the landslide, for example, by compartmentaliz-
3 METHODS
ing the unstable rock mass into structural units based on vibrational
properties. We track the dynamic response over 30 months with a
3.1 Ambient vibration array data acquisition and
seismometer installed directly on the slope instability and report
processing
on significant changes in wavefield polarization, coinciding with
changes in landslide kinematics and demonstrate that precipitation To characterize the dynamic response on the Plateau, we deployed a

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


history directly influences the seismic site response of the landslide. temporary array of 16 seismometers on 18 June 2018 and recorded
105 min of ambient vibrations (Fig. 2). We used LE-3D/5 s seis-
mometers from Lennartz and Nanometrics Taurus and Centaur digi-
tizers with 200 Hz sampling frequency. All temporary stations were
2 LANDSLIDE GENESIS AND ACTIVITY
mounted on metal tripods in shallow holes in the soil (∼30 cm), on
The active slope instability of Brinzauls (German: Brienz) covers an large boulders or, where applicable, on solid bedrock. Seismometers
area of about 3 km2 on the southern slope of Piz Linard in Canton of were oriented to magnetic north by compass with accuracy better
Grisons, Switzerland, and is part of a deep-seated gravitational slope than 5◦ (magnetic declination during survey: +1◦ ). Every digitizer
deformation, which extends over an area of about 10 km2 (Fig. 1a; was equipped with a GPS antenna for accurate timing informa-
Krähenbühl & Nänni 2017). The active part of the landslide can be tion. During the array, the permanent seismic station BRIZ2 was
divided into two interlinked domains. recording simultaneously (Section 3.2).
The lower domain consists of a cultivated terrace around the Empirical and numerical studies on unstable rock slopes demon-
village Brinzauls, which is limited by the Albula river to the south strated that such sites exhibit large seismic wavefield amplifications
and sharp lateral boundaries towards the east and west. A drilling (>4; Bozzano et al. 2008; Del Gaudio & Wasowski 2011; Gischig
campaign in 2018/2019, covering the entire lower domain, revealed et al. 2015). Local relative wavefield amplification can be estimated
the presence of a distinct, relatively flat (10–15◦ ) basal sliding plane, using the site-to-reference spectral ratio technique using ambient
reaching maximum depth of 153 m at the village of Brinzauls vibration data (SRSR; Borcherdt 1970; Perron et al. 2018). Hor-
(borehole KB1) and daylighting at the recent riverbed of the Albula izontal amplitude spectra at the test stations are divided by the
(BTG 2021). The sliding mass has a volume of 100 × 106 to 200 spectra of a reference station placed on presumably stable ground
× 106 m3 and consists mainly of Flysch (turbidity flow deposit, not exhibiting amplification. In our study, we used the geometrical
slate with sandy interbeds), which run over the fluvial deposits of mean of both horizontal components and station 317 as a reference
the Albula over a distance of roughly 300 m. Sliding has begun station, which is located on the alpine meadow Propissi Sot outside
in prehistoric times presumably due to glacial retreat after the last the rock instability. Spectra with high resolution and low spectral
glacial maximum. Sliding velocity around the village increased leakage are computed using the multitaper method with four ta-
from 0.2 m yr–1 prior to November 2011 to 1.2 m yr–1 in November pers, a time-bandwidth product of 2.5 and 100 s time windows
2020 (Fig. 1b), resulting in severe damages on farmland, buildings (Thomson 1982; Prieto et al. 2009). The high frequency resolution
and infrastructure. of the spectra provided by multitaper analysis allows for detecting
The second, upper domain comprises the rock formations above near-monochromatic noise sources, which are most likely caused
the village, a ridge named Caltgeras and a less steep area called by anthropogenic sources, such as machinery. If the distance be-
Plateau, consisting of the Allgäu formation, overlaid by Raibler tween reference and test station is small, such noise signals are
layers (Rauwacke and dolomites) and Arlberg dolomite. Litholog- successfully cancelled by normalizing to the reference spectrum.
ical boundaries are of tectonic origin and dip moderately into the Relative amplification curves were automatically clustered using
slope. A drilling campaign in 2020 consisted of cored and instru- the heuristic k-means++ clustering algorithm, which is a more effi-
mented boreholes, and revealed different kinematic mechanisms cient version of the classic k-means clustering algorithm and allows
along Caltgeras: sliding at the toe of the ridge (at 125 m in KB8) us to group stations with similar seismological fingerprints and pre-
and in the Arlberg dolomite on top (at 57 m in KB10) and top- sumably similar site conditions (Lloyd 1982; Arthur & Vassilvitskii
pling in the Allgäu formation in the midslope regions (at 183 m in 2007). The k-means++ algorithm is applied with the numbers of
KB11). The fracture network in the Arlberg dolomite consists of clusters (k) ranging from one to five (maximum number of clusters
steeply inclined SW–NE and NW–SE striking fractures and sub- expected). We used the squared Euclidean distance and 500 repli-
dominant N–S trending fracture zones. Displacement velocities at cates, that is the clustering is repeated 500 times to avoid converging
the Plateau were 0.6 m yr–1 in October 2011 and reached 3.8 m yr–1 to a local minimum. For every k, the silhouette clustering criterion
in November 2020 (Fig. 1b) with a sliding dip angle >35◦ (see also (Rousseeuw 1987) is computed and the run with the highest silhou-
Gojcic et al. 2021). The volume of this second domain is estimated ette value determines the optimal amount of clusters. In our study,
to be on the order of 10 × 106 to 25 × 106 m3 . The area is heavily we applied the clustering algorithm in two frequency bands: 1.1–
fractured, resulting in frequent rock fall events towards the village 2.3 Hz and 2.3–4.0 Hz. These frequency bands were determined
of Brienz/Brinzauls. Fig. 1(c) provides an overview of the principal based on a first inspection of the power spectra of all components
morphological structures. A previous collateral landslide of about registered, which revealed a peak of elevated power in each of these
Site response of the Brinzauls slope instability 301

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023

Figure 1. Overview of the deep-seated landslide at Brinzauls, with focus on the Plateau subdomain. (a) Photograph of the Brinzauls rock instability (view
towards northwest) with drilling locations, GNSS (Global Navigation Satellite System) monitoring positions and seismometer location. (b) Displacement
time-series at selected locations marked on panel (a). (c) Elevation model and simplified geomorphological map of the Plateau (see Fig. S1 for details). (d)
Geological profile along the ridge Caltgeras. Photograph panel (a) by courtesy of SRF Einstein. Elevation model in panel (c) provided by courtesy of AWN.
Coordinate frame LV03 (WGS84: 46.679, 9.593). More photographs of the site and the monitoring installation are provided in Fig. S2.
302 M. Häusler et al.

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


Figure 2. Overview map of the experimental setup at Caltgeras and the Plateau, including temporary seismometers (red squares), one permanent seismic
station (red triangle) and GNSS markers (yellow and green circles). Note that marker 1010 is excluded from the singular value decomposition (SVD), as it is
located on the currently inactive part of the landslide. Geodata source AWN and Federal Office of Topography.

bands. We chose 1.1 Hz instead of 1.0 Hz for the lowest frequency (Brincker et al. 2001; Michel et al. 2010; Bottelin et al. 2013; Poggi
to avoid a strong monochromatic signal at 1 Hz recorded at one sen- et al. 2015; Häusler et al. 2021; Mercerat et al. 2021). The cross-
sor, which is attributed to noise. The same bands can be obtained power spectral density matrix of all input traces of ambient vibration
by analysing the amplification function based on empirical spectral data is computed and decomposed in its singular values and singular
modelling (Section 3.3). The k-means++ algorithm can be found vectors. The singular values represent the auto spectral densities of
as built-in function of MATLAB. each modal coordinate, which peak at the resonant frequencies of
High amplification in distinct frequency bands might indicate the system. The singular vector at the resonant frequency directly
normal-mode behaviour, that is a standing wave phenomenon represents the 3-D mode shape vector at the recording station. The
caused by waves trapped between fractures. To investigate this phe- advantage of FDD lies in the combination of all input traces in a
nomenon, we applied the frequency domain decomposition (FDD) single plot and in the direct interpretation of the mode shapes by con-
technique for normal-mode analysis, which is a standard technique serving the phase information between different recording stations.
in structural engineering but has found broad application on geolog- This property can, for example, be used to map the fracture net-
ical features such as sedimentary valleys and rock slope instabilities work of rock instabilities by delineating the zero-crossings between
Site response of the Brinzauls slope instability 303

different vibration phases of a mode shape (Häusler et al. 2019). In Service. The empirical spectral modeling amplification curve is rep-
our analysis, we included the reference station 317 and all stations resented by the geometric mean of the horizontal components of all
on the instability except station 301, because of its corrupted vertical observations. An observation only contributes to the computation
component. To better distinguish true resonant frequencies of the if the signal-to-noise ratio of a recording exceeds three. Therefore,
system from harmonic oscillations arising from noise sources such for some events only one horizontal component (i.e. observation)
as machineries, we additionally evaluated the mean of all power is available. Note that this method does not consider non-linear site
spectra computed with the multitaper method. Near-harmonic sig- effects and is, thus likely to overestimate amplification during strong
nals appear as sharp spikes, supporting the detection of these noise ground shaking (Beresnev & Wen 1996).
sources on the singular value plot of the FDD. In our study, 72 observations from 40 regional seismic events are
contributing to the site amplification function at station BRIZ2. We
provide an overview of the contributing events in Table S1 and Fig.
3.2 Ambient vibration monitoring and processing S6. The regional earthquakes reached local magnitudes between
1.6 and 4.3 and occurred at distances between 16 and 187 km. The
We measured ambient vibrations at the seismic station BRIZ2 using
catalogue is dominated by the Elm sequences located at 43 km
a Lennartz LE-3Dlite/1s seismometer, recording at 200 Hz using
distance, making up 13 out of 40 seismic events or 23 out of 72
a solar-powered Nanometrics Taurus digitizer and mobile phone

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


observations.
network for data transmission. The sensor was installed on 30 May
2018 and was inspected on 13 September 2019 and 19 November
2020 while no deviation from magnetic north was observed in the
precision of a hand-held magnetic compass. The sensor tilted by
less than 2◦ towards south across the monitoring period. 3.4 GNSS-based displacement measurements
Data are processed every full hour using data of the past 60 The landslide area is monitored semi-annually by measuring fixed
min. Data handling and preparatory signal processing is performed field markers with a hand-held differential GNSS system (yellow
using the seismological processing package ObsPy (Beyreuther et and green circles in Fig. 2). In addition, a permanent GNSS sensor
al. 2010). For every data package of 60 min, power spectra were (station 101) is installed in the centre of the village (see Figs 1a and
computed using the multitaper technique (described in the previous S1). Both systems have an accuracy of about 2 cm.
section) and wavefield polarization analysis was performed using To analyse relative secondary displacement vectors, we retrieved
three frequency bands (A: 1–2.4 Hz, B: 2.4–4 Hz, and C: 4–10 Hz.). and removed the principal component of the displacement field
Bands A and B correspond to the bandwidths used for clustering by applying a singular value filter, a common technique in image
relative amplification curves (see Section 3.1), that is each band processing (Andrews & Patterson 1976). For every interval be-
includes one amplification peak. We added the third band (C) up tween two subsequent surveys, a matrix M is set up, containing
to 10 Hz to preserve some higher frequencies and complete the the displacement of the three Cartesian components in its columns
1–10 Hz bandwidth. For polarization analysis, we used the time- (x, y, z) and the number of GNSS survey locations n in the
frequency polarization technique by Burjánek et al. (2010), which rows:
is a generalized particle motion polarization analysis adapted from ⎡ ⎤
Vidale (1986). The method computes the coherency matrix between x1 y1 z 1
⎢ ⎥
all three components using the continuous wavelet transform. The M = ⎣ ... . . . ... ⎦ . (1)
eigenvalue decomposition provides the directionality of the seismic
xn yn z n
signal by evaluating the eigenvector of the largest eigenvalue. The
resulting parameters are the orientation of the wavefield in azimuthal The singular value decomposition of Mi takes the form
direction from north, the inclination of the wavefield with respect to
3
horizontal and the ellipticity, describing the degree of polarization of M = USVT = u j s j vTj , (2)
the wavefield. For display purpose, resulting azimuth values were j=1

smoothed using a 72-hr median filter and hourly measurements where U and V are orthogonal matrices containing the left and right
values of all polarization parameters are illustrated in Figs S3–S5 singular vectors, that is U = [u1 , . . . , un ] and V = [v1 , v2 , v3 ]. Note
in the supplementary information. that U is a n × n matrix and V is always a 3 × 3 matrix. S is a n × 3
matrix containing the positive singular values in descending order
in the diagonal elements with off-diagonal elements being zero. T
3.3 Absolute amplification by empirical spectral modelling denotes the transpose of a matrix.
We retrieved information on the absolute linear seismic site am- The principal component is removed by withdrawing the first
plification by applying the empirical spectral modelling method singular value, that is by neglecting j = 1:
introduced by Edwards et al. (2013). This approach compares the 3
A = U S V =
T
expected spectra from a Brune ω2 source model (Brune 1970) based u j s j vTj . (3)
j=2
on a localized hypocentre of an earthquake with those recorded in
the field by the Swiss national seismic network (SED 1983). After The resulting matrix A describes the displacement vector field
each earthquake registered by the network, the model is inverted for without the principal component and, thus represents secondary
damping and the local site amplification relative to a Swiss generic displacements only. Stations 1010 and the permanent GNSS station
rock reference model. This reference model is based on 27 seismic 101 were excluded from the analysis, as station 1010 is situated on
stations where shear wave velocity profiles are available (Poggi et the inactive part north of the rear fracture and station 101 is located
al. 2011). Due to the event-by-event approach, the resulting am- in the village of Brinzauls and, thus in the lower domain of the
plification becomes increasingly robust over time. The technique is landslide, which experiences different kinematics than the upper
implemented in the routine monitoring of the Swiss Seismological Caltgeras-Plateau domain.
304 M. Häusler et al.

3.5 Meteorological and environmental data deviations from the horizontal plane between 3◦ and 8◦ (see Table
S2 in the electronic supplement for azimuth and dip information).
Precipitation data were taken from the weather station operated
This polarization pattern coincides well with the direct surface dis-
by the Swiss Federal Office of Meteorology and Climatology
placement field measured with GNSS between May and November
(www.meteoswiss.admin.ch) in Valbella (station code VAB, 8.9 km
2018, that is in the period of the seismic array measurement. Such
distance to BRIZ2, 1568 m a.s.l., WGS84: 46.75503, 9.55443). We
alignment of the wavefield polarization perpendicular to dominant
used a normalized precipitation index to model the decaying con-
fracture systems could be observed in field and synthetic data ac-
tribution of a given precipitation event to the water content in the
quired on other rock instabilities and around tectonic faults before
subsurface. The index describes the daily precipitation sums and
(Burjánek et al. 2010, 2019; Pischiutta et al. 2012, 2013; Häusler et
includes the daily precipitation sums of all previous days with a
al. 2019). Since all stations are resonating in-phase, these findings
weight exponentially decaying with time. We followed the formu-
support our interpretation that this frequency band is sensitive to
lation given by Helmstetter & Garambois (2010) and normalized
one large unstable volume.
the index by the time window length considered. The normalized
A second, more pronounced resonant frequency f2 can be ob-
precipitation index PC at time t j is given by:
served at 2.8 Hz. The corresponding normal-mode vectors reveal
i ti −t j a heterogeneous pattern: stations west and southwest of BRIZ2
P t j exp −

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


j=0 tc
Pc (ti ) = , (4) show a NE–SW in-phase polarization pattern, while BRIZ2 and
i ti −t j
j=0 exp − tc
stations towards east and southeast are characterized by a strong
NW–SE polarization (Fig. 4c). We interpret this higher resonant
where P(t j ) is the daily precipitation sum at time t j and tc cor- frequency to represent secondary volumes (Häusler et al. 2019).
responds to the characteristic time, which was set to 14 d in our Therefore, we remove the principal displacement component (i.e.
study. N–S) from the displacement field measured by GNSS to obtain sec-
Snow height data were taken from an automated station op- ondary displacements between individual markers. This is achieved
erated by the WSL Institute for Snow and Avalanche Research by a singular value filter and removing the first singular vector. The
(www.slf.ch) located at Piz Martegnas at an elevation of 2429 m resulting vector field describes the surface displacements in absence
a.s.l. (station code PMA2, WGS84: 46.58008, 9.53739) and with a of the dominating N–S component. The horizontal component of
distance of 11.8 km to BRIZ2. Discharge data of the Albula river the secondary displacement vectors is perpendicular to the cliff at
were recorded at Tiefencastel, which is located adjacent to the deep- the western edge of the instability and points towards east at BRIZ2
seated gravitational slope deformation of Brinzauls. The recording and markers east of the seismic station. The secondary displace-
station (code 2141) is operated by the Swiss Federal Office of the ment field and normal-mode vectors of the second mode are not
Environment (www.bafu.admin.ch). in perfect alignment at all locations. However, the secondary dis-
placement vector field shows an E–W diverging component, which
is supported by the E–W polarization of the wavefield. The dividing
4 R E S U LT S
line appears to strike N–S in close proximity of BRIZ2. Table S2
provides the azimuth and dip of all stations. Normal-mode vectors
4.1 Seismological mapping
are close to horizontal with deviations between 2◦ and 22◦ to the
In the frequency band 1.1–2.3 Hz, the amplification pattern of the horizontal plane.
seismic stations clusters into two groups (Figs 3a and c). The sta- The resonant frequencies determined by FDD (1.7 and 2.8 Hz,
tions of the first group show a mean amplification factor of about Fig. 4) were observed at lower frequencies than the peaks in SRSR
five peaking at 2 Hz and are all located south of the active rear (2.0 and 3.0 Hz, Fig. 3). By normalizing the power spectra to a
fracture. The second cluster is characterized by stations with weak reference site, source and path effects are removed from the SRSR,
amplifications not exceeding three. Note that the amplification pat- given that vibration sources are at significantly larger distance than
tern clearly distinguishes the active from the inactive part of the the interstation distance. In contrast, no such normalization is per-
landslide. formed for FDD. Therefore, source and path effects might bias the
In the frequency band 2.3–4.0 Hz, again two clusters are distin- dynamic response, as the underlying assumption of equally dis-
guished (Figs 3b and d). However, the clustering now segregates tributed sources with white noise characteristics is not fully met.
a small group of eight stations with a mean amplification factor On the other hand, FDD only considers energy correlated between
of eight with a peak at 3 Hz. However, amplifications reach up to different recording channels and stations by computing the cross-
eleven in the surroundings of BRIZ2. This cluster separates from power spectral densities, while SRSR only considers the power and
a cluster including all stations north of the rear fracture and three not the phase information. Therefore, the SRSR might contain en-
stations on the Plateau with lower amplifications. Based on this am- ergy of local amplification phenomena that are not related to an
plification pattern, we interpret that the resonant frequency at 3 Hz underlying normal-mode behaviour. Despite we are unable to ex-
corresponds to one or several secondary volumes of the landslide. tract the specific reason for these frequency shifts, we still interpret
The singular value plot derived from FDD and the mean of all that both methods are sensitive to the same resonant phenomena due
power spectra are shown in Fig. 4(a) and indicate the presence of sev- to the large width of both resonant peaks (1–2.3 Hz and 2.3–4.0 Hz).
eral near-monochromatic noise sources, leading to narrow-banded
spikes in the spectra and joint elevation of higher singular values
(Hjortenberg & Risbo 1975; Brincker et al. 2000). Nevertheless, the
4.2 Ambient vibration monitoring
singular values suggest a weak resonant frequency f1 at 1.7 Hz. The
normal-mode shape at this frequency shows a predominant in-phase The spectrogram of the permanent seismic station BRIZ2 reveals
north–south oriented polarization. The red-black arrows in Fig. 4(b) that f2 at about 3 Hz can be observed continuously without sig-
show the normal-mode vectors with the colour corresponding to nificant variations of the resonant frequency (Fig. 5a). Fig. 5(b)
the phase information. The polarization is nearly horizontal with shows the power spectra of all three seismometer components on
Site response of the Brinzauls slope instability 305

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


Figure 3. Seismological mapping based on site-to-reference spectral ratios (SRSR). (a) and (b) Relative seismic amplifications (marker colour) at (2 and 3)
Hz, respectively, clustered into two groups (marker shape) by a k-means++ clustering algorithm. The reference station (317) is assumed to be located outside
the instability on stable ground. (c) and (d) Mean amplification curves and envelope of all observations of each cluster between (1.1 and 2.3) Hz and (2.3 and
4.0) Hz, respectively. Geodata source: Federal Office of Topography.

21 January 2019, which is representative for most times of the year, amplification effect due to changing site response caused by an
showing a broad spectral peak at 3 Hz. Considerably higher am- elevated ground water level during snow melt cannot be excluded.
plitudes are visible between June and December 2020, related to Wavefield polarization (Burjánek et al. 2012) azimuths of con-
drilling activities in proximity of station BRIZ2. f1 between (1.5 tinuous seismic data for three frequency bands (A: 1–2.3 Hz, B:
and 2.0) Hz is only visible during distinct periods, for example, 2.3–4 Hz, C: 4–10 Hz) are illustrated in Fig. 5(f). The azimuth
from 2 June to 8 July 2019 (Fig. 5c). During these windows, the in frequency band A fluctuates around 150◦ N, increasing towards
amplitude of f2 is also elevated. These events correlate with the 190◦ N in fall 2020. Contrary to the polarization derived by FDD,
increased Albula river discharge (>25 m3 s–1 ) and is likely a source the polarization vector is not strictly parallel to the principal dis-
effect (e.g. Dı́az et al. 2014; Fig. 5d). River discharge is primarily placement vector but perpendicular to the southeastern cliff of the
governed by snow melt and only secondarily by strong precipitation rock instability. After the clockwise rotation in fall 2020, polar-
events (compare the precipitation index in Fig. 5(d) to snow height ization and principal displacement vectors are in good agreement
in Fig. 5e). The peak in river discharge in spring 2019 was caused (∼190◦ ). A particular feature can be observed during periods when
by large snow amounts in combination with high snow melting rate the resonant frequency f1 at about 1.7 Hz is excited (e.g. 2 June
(snow height and snow height gradient in Fig. 5e). We hypothesize to 8 July 2019). Polarization changes during these periods to an
that this increased signal power emerging from river discharge is azimuth of about 110◦ N, and thus in a similar range as bands B
necessary to excite resonant frequency f1 . However, an additional and C. We might argue that the elevated signal power enables the
306 M. Häusler et al.

band B to reach 64◦ N in December 2020. This counter-clockwise


rotation is observable in the GNSS-based displacement field af-
ter removing the first singular vector. GNSS markers surrounding
BRIZ2 show a rotation from ∼100◦ N towards 40◦ N between May
2019 and November 2020 (marker 1009 in Fig. 5g). No such rota-
tion can be observed at further distance to BRIZ2. The initialization
of the rapid rotation in June 2019 coincides with a general accel-
eration of the landslide and an increase in sliding dip angle at the
Plateau during spring 2019 (Figs 5h and i).
Based on the good agreement with rotating secondary displace-
ment vectors we interpret that changing polarization is linked to
landslide kinematics. We interpret that in the beginning of the seis-
mological monitoring, extension along a SW–NE oriented frac-
ture network dominated, corresponding to the 120◦ N polarization
at ∼3 Hz and causing the superficial dolomite block to deform to-

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


wards SE. We hypothesize that in spring 2019, related to the overall
acceleration of the slide, a N–S to NW–SE oriented fracture system
accommodated a significant part of the deformation, forcing the
blocks to diverge E–W to NE–SW and inducing subsidence at the
Plateau, manifested by the increasing sliding dip angle. The wave-
field polarizes perpendicular to the fracture network and changes
its orientation towards northeast as well. Pischiutta et al. (2013)
showed in numerical models that the stress field, orientation of ex-
tension fractures and polarization of ambient vibrations are closely
related. Polarization monitoring of frequency band B at Brinzauls
showed that the counter-clockwise rotation was already in process
in July 2018 (by 5◦ until March 2019), thus, before a rotation was
visible in surface displacement vectors. Following the interpreta-
tions by Pischiutta et al. (2013), we suggest that this early rotation
was an effect of changing stress conditions, which was followed
by changing landslide kinematics in spring 2019. However, while
wavefield polarization could be used as an immediate proxy for
the orientation of the dominating fracture network and thus for the
anticipated surface deformation vector field in extensional regimes,
there is no direct causality between displacement and polarization.
Even though significant polarization can be measured, the rock for-
mation can show no deformation (Burjánek et al. 2012; Kleinbrod
et al. 2017) or deformation parallel to the fracture plane.

4.3 Earthquake monitoring


Figure 4. Normal-mode analysis on the Plateau. (a) First nine singular
values (black) of the frequency domain decomposition modal analysis and Fig. 6(a) shows the empirical spectral modelling amplification of
mean multitaper power spectrum (red), indicating two resonant modes f1 40 regional seismic events recorded at station BRIZ2. We observe
and f2 (red markers). Sharp peaks in the power spectrum designate near- a first broad amplification peak between (1 and 2.3) Hz reaching
monochromatic noise signals (grey-shaded areas), which are broadened in a factor of three. A second amplification peak appears at 3.2 Hz,
the singular value plot due to lower frequency resolution and should not be reaching a factor of four, confirming the dynamic response deter-
interpreted as normal modes. (b) Normal-mode vectors (red-black double mined by ambient vibration surveys. However, amplification values
arrows, indicating the phase of the vibration) of the weak fundamental mode might differ between SRSR and empirical spectral modelling due
f1 at 1.7 Hz and displacement vectors of GNSS markers between May and
to different definitions for the reference.
November 2018 (blue). (c) Normal-mode vectors of the more dominant
mode f2 at 2.8 Hz and relative GNSS displacement field after removal of the
Fig. 6(b) displays each amplification curve with increasing ob-
first singular vector. Geodata source: Federal Office of Topography. servation number, whereas the temporal distribution for all observa-
tions is shown in Fig. 6(d). We computed the median amplification
excitation of a second fundamental mode, dominating the otherwise factor between (1 and 4) Hz of each observation and compare it
persistent N–S directed polarization. However, no array data exist to the precipitation index (Fig. 6c). The precipitation index (Helm-
from a period with an elevated excitation level to describe the modal stetter & Garambois 2010) is based on daily precipitation sums and
behaviour. accounts for the decreasing effect of water on the rock mass with
Frequency band B shows a counter-clockwise rotation of the az- time due to drainage. The comparison of precipitation index and
imuth starting at 120◦ N in June 2018, accelerating in March 2019 median amplification indicates a direct correlation. We highlight
at 115◦ N and flattening towards the end of 2020 to reach 70◦ N. The observation 10 (label A), when a high precipitation index coincides
azimuth of band C slightly decreased from 107◦ N in June 2018 to with a high median amplification factor of about five. Note that
103◦ N in July 2019 and followed the counter-clockwise rotation of observation 10 also coincides with the high Albula river discharge
Site response of the Brinzauls slope instability 307

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


Figure 5. Ambient vibration monitoring of the Brinzauls rockslide. (a) Spectrogram of station BRIZ2, geometric mean of both horizontal components. (b)
and (c) Power spectra of all components of BRIZ2 during a period without (21 January 2019) and with (28 June 2019) strong resonance at f1 . Z: vertical
component, N: north component, E: east component. (d) Precipitation index with tc of 14 d at Valbella and Albula river discharge at Tiefencastel. (e) Snow
height at Piz Martegnas on 2430 m elevation and temporal gradient of snow height. (f) Azimuth of wavefield polarization for frequency bands of 1.0–2.3 Hz
(A), 2.3–4.0 Hz (B) and 4.0–10 Hz (C). A 3-d running median filter was applied, hourly measurement results and standard deviation are given in Fig. S3. (g)
First singular vector of the singular value decomposition of the displacement field at the Plateau (black) and secondary displacement vectors at GNSS markers
1002 (orange) and 1009 (red). (h) Displacement rates of GNSS markers 1002 and 1009 (overlapping). (i) Sliding dip angle measured at GNSS markers 1002
and 1009. Note that panels (g) to (i) have a discontinuous time axis to focus on the seismological monitoring period.

during a period of rapid snow melting. Amplification factors were directly results in higher seismic amplification. However, outliers
between three and four during summer 2019 and lower during a rel- of high amplification also occurred during dry periods. Therefore,
atively dry period starting with observation 23 (label B). This period source and path effects might play a role. A potential reason for
ends with a heavy precipitation event on 27 May 2020 (label C), increased seismic amplification after precipitation periods might be
coinciding with a significant increase in seismic amplification, fol- the reduction in rock strength and stiffness (i.e. shear modulus) due
lowed by a slow decay over time. This indicates that high water input to water infiltration (Wong et al. 2016; Yurikov et al. 2018).
308 M. Häusler et al.

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


Figure 6. Monitoring of seismic amplification of the Brinzauls landslide based on earthquake recordings. (a) Empirical spectral amplification based on 72
observations from 40 regional seismic events. The solid black line represents the geometric mean of all observations, first standard deviation is given by the
shaded area. (b) Amplification for every observation between (0.8 and 10) Hz. For most seismic events, two observations (north and east component) are
available. However, a low signal-to-noise ration eventually leads to the rejection of one of both. Red lines (labels A to C) indicate the beginning of periods with
significantly lower and higher amplification, respectively. No amplification value is available for white areas due to a signal-to-noise ratio smaller than three at
the corresponding frequency. (c) Precipitation index with respect to the past 14 d at the day of the seismic event (blue) and median amplification value between
(1 and 4) Hz for every seismic event (red). (d) All seismic events used for computation of amplifications with increasing observation number (as shown in
panels b and c). Circle size represents the event magnitude.

5 C O N C L U S I O N S A N D I M P L I C AT I O N S proximity of BRIZ2, which we interpret as an effect of changes in


FOR LANDSLIDE MONITORING the local stress field. This information could be retrieved in real-time
using a single seismic station, highlighting the potential of ambient
The dominating south-directed surface displacement vectors of the
vibration observations as a complementary technique to displace-
landslide coincide with the wavefield polarization at 1.7 Hz. The in-
ment surveys for landslide monitoring on this and potentially other
phase vibration pattern across the Plateau indicates a large but weak
geological mass movements. Nevertheless, the empirical observa-
fundamental resonator limited by the rear fracture and hypotheti-
tions presented here should be verified by laboratory tests or 3-D
cally constrained by a lithological or structural boundary at depth
numerical simulations by monitoring wavefield polarization under
(e.g. by the Raibler layers, see Fig. 1d). The heterogeneous polar-
changing stress conditions over time.
ization at 2.8 Hz indicates fragmentation into separate blocks in the
Seismic amplifications during small seismic events revealed a
dolomite with a predominating east-west component in proximity
large temporal variance of the seismic response that correlates with
of station BRIZ2. These findings are supported by GNSS-based
precipitation history. This has direct implication for the hazard as-
surface displacement measurements. Polarization and amplification
sessment of earthquake-induced slope failure, as the slope is likely
patterns can be used as proxies for displacement vectors and land-
to be more susceptible to failure during periods of high ampli-
slide disaggregation and can be rapidly retrieved by temporal seis-
fication following heavy precipitation events and possibly during
mic arrays. This is especially beneficial before or in the early stage of
periods of increased snow melt. Tracking the seismic response on
continuous landslide monitoring to plan the monitoring setup and to
slope instabilities could be used to calibrate geomechanical and cou-
bypass the early monitoring period, when measured displacements
pled seismo-hydromechanical models in the linear domain to model
are too small to reliably determine displacement vectors. However,
failure during (possibly non-linear) seismic loading (Allstadt et al.
studies on other rock instabilities showed that polarization is not
2018; Chen et al. 2020).
solely governed by fracture geometry, but possibly also by other
parameters such as material stiffness distribution (Häusler et al.
2021).
Monitoring the polarization at BRIZ2 showed an irreversible AC K N OW L E D G E M E N T S
counter-clockwise rotation for frequencies above 2 Hz, accelerat- This work was financed by ETH fund 0-20361-17. We thank A.
ing in July 2019. This rotation is also evident in the secondary Huwiler and A. Largiadèr (AWN, office for forest and natural haz-
surface displacement vectors, thus supporting our conclusion that ards, Canton of Grisons) and the commune Albula/Alvra for their
wavefield polarization and landslide kinematics are closely linked. collaboration, shared data and unrestricted access to the site. We
The transition from a polarization azimuth of 120◦ N towards 70◦ N thank the companies CSD Thusis and BTG Sargans for sharing re-
coincides with a change of the direction of secondary extension in ports and insights into the geological model and kinematics of the
Site response of the Brinzauls slope instability 309

site. We thank our technician (R. Moser, ETH) and field helpers (J. Bottelin, P., et al., 2017. Monitoring rock reinforcement works with ambient
Aaron, M. Faber, D. Farsky, J. Holland, U. Kleinbrod, J. Laukenmann vibrations: la Bourne case study (Vercors, France), Eng. Geol., 226, 136–
and M. Studer) for their efforts in the field. We are thankful to R. 145.
Racine (Swiss Seismological Service) for maintaining the empirical Bottelin, P., Lévy, C., Baillet, L., Jongmans, D. & Guéguen, P., 2013. Modal
and thermal analysis of Les Arches unstable rock column (Vercors massif,
spectral modelling amplification software. Any use of trade, firm or
French Alps), Geophys. J. Int., 194(2), 849–858.
product names is for descriptive purposes only and does not imply
Bozzano, F., Lenti, L., Martino, S., Paciello, A. & Scarascia Mugnozza, G.,
endorsement by ETH Zurich or the Swiss government. 2008. Self-excitation process due to local seismic amplification respon-
sible for the reactivation of the Salcito landslide (Italy) on 31 October
2002, J. geophys. Res., 113(B10), doi:10.1029/2007JB005309.
D ATA AVA I L A B I L I T Y Bradley, K. et al., 2019. Earthquake-triggered 2018 Palu Valley landslides
enabled by wet rice cultivation, Nat. Geosci., 12(11), 935–939.
Data generated in this study and their corresponding metadata are
Brincker, R., Andersen, P. & Møller, N., 2000. Indicator for separation of
available on the ETH Research Collection via https://doi.org/10.3 structural and harmonic modes in output-only modal testing, in Proceed-
929/ethz-b-000474116. Data from third party data providers are ings of SPIE: The International Society for Optical Engineering, 4052.
accessible via the website of the different data providers. Instruc- Brincker, R., Zhang, L. & Andersen, P., 2001. Modal identification of output-
tions for data access are given in the metadata on the ETH Research only systems using frequency domain decomposition, Smart Mater.

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


Collection. Struct., 10(3), 441–445.
Brune, J. N., 1970. Tectonic stress and the spectra of seismic shear waves
from earthquakes, J. geophys. Res. (1896-1977), 75(26), 4997–5009.
AU T H O R C O N T R I B U T I O N S TAT E M E N T BTG, 2021. Report 5897-13, Rutschung Brienz/Brinzauls GR, Geologis-
che Detailuntersuchungen 2018 bis 2021, Sondierbohrungen. Geological
MH and DF designed the study, MH acquired, processed, and anal- report. BTG Büro für Technische Geologie AG.
ysed the seismological data. MH and VG analysed displacement Budimir, M. E. A., Atkinson, P. M. & Lewis, H. G., 2014. Earthquake-
data. RT and VG compiled the geological model and provided the and-landslide events are associated with more fatalities than earthquakes
geological description of the site. FG provided the geomorpholog- alone, Nat. Hazards, 72(2), 895–914.
ical map and supported geological interpretation. All Authors dis- Burjánek, J., Gassner-Stamm, G., Poggi, V., Moore, J. R. & Fäh, D., 2010.
cussed and interpreted the results. MH wrote the manuscript with Ambient vibration analysis of an unstable mountain slope, Geophys. J.
Int., 180(2), 820–828.
contributions from all co-authors.
Burjánek, J., Gischig, V., Moore, J. R. & Fä,h, D., 2018. Ambient vibra-
tion characterization and monitoring of a rock slope close to collapse,
Geophys. J. Int., 212(1), 297–310.
C O M P E T I N G I N T E R E S T S S TAT E M E N T Burjánek, J., Kleinbrod, U. & Fäh, D., 2019. Modeling the seismic response
Authors VG and RT are employed by companies, which are both of unstable rock mass with deep compliant fractures, J. geophys. Res.,
active in business on the landslide of Brienz/Brinzauls for early 124(12), 13 039–13 059.
Burjánek, J., Moore, J. R., Molina, Yugsi, F., X. & Fäh, D., 2012. Instrumental
warning service and geological analyses. However, we declare to
evidence of normal mode rock slope vibration, Geophys. J. Int., 188(2),
have no direct competing interests, as there is no financial support 559–569.
of the research project by the companies. Similarly, the research Casagli, N. et al., 2016. Landslide mapping and monitoring by using radar
project is not supporting any of the two companies. Both co-authors and optical remote sensing: examples from the EC-FP7 project SAFER,
are collaborating scientifically on an informal basis in the project. Rem. Sens. Applicat.: Soc. Environ., 4, 92–108.
Chen, L., Mei, L., Zeng, B., Yin, K., Shrestha, D. P. & Du, J., 2020. Failure
probability assessment of landslides triggered by earthquakes and rainfall:
REFERENCES a case study in Yadong County, Tibet, China, Sci. Rep., 10(1), 16531.
Allstadt, K. E., Jibson, R. W., Thompson, E. M., Massey, C. I., Wald, D. J., Colombero, C., Jongmans, D., Fiolleau, S., Valentin, J., Baillet, L. & Bièvre,
Godt, J. W. & Rengers, F. K., 2018. Improving near-real-time coseismic G., 2021. Seismic noise parameters as indicators of reversible modifica-
landslide models: lessons learned from the 2016 Kaikōura, New Zealand, tions in slope stability: a review, Surv. Geophys., 42(2), 339–375.
earthquake, Bull. seism. Soc. Am., 108(3B), 1649–1664. Del Gaudio, V. & Wasowski, J., 2011. Advances and problems in under-
Andrews, H. & Patterson, C., 1976. Singular value decompositions and standing the seismic response of potentially unstable slopes, Eng. Geol.,
digital image processing, IEEE Trans. Acoust. Speech Signal Process., 122(1), 73–83.
24(1), 26–53. Dı́az, J., Ruı́z, M., Crescentini, L., Amoruso, A. & Gallart, J., 2014. Seismic
Arthur, D. & Vassilvitskii, S., 2007. k-means++: the advantages of careful monitoring of an Alpine mountain river, J. geophys. Res., 119(4), 3276–
seeding, in Proceedings of the 18th Annual ACM-SIAM Symposium on 3289.
Discrete Algorithms, New Orleans, Louisiana, January 7–9, Society for Dini, B., Manconi, A. & Loew, S., 2019. Investigation of slope instabilities
Industrial and Applied Mathematics, pp. 1027–1035. in NW Bhutan as derived from systematic DInSAR analyses, Eng. Geol.,
Beresnev, I. A. & Wen, K.-L., 1996. Nonlinear soil response—a reality?, 259, 105111.
Bull. seism. Soc. Am., 86(6), 1964–1978. Edwards, B., Michel, C., Poggi, V. & Fäh, D., 2013. Determination of site
Beyreuther, M., Barsch, R., Krischer, L., Megies, T., Behr, Y. & Wassermann, amplification from regional seismicity: application to the Swiss National
J., 2010.ObsPy: a Python toolbox for seismology, Seismol. Res. Lett., Seismic Networks, Seismol. Res. Lett., 84(4), 611–621.
81(3), 530–533. Fernández-Merodo, J. A., Garcı́a-Davalillo, J. C., Herrera, G., Mira, P. &
Bickel, V. T., Manconi, A. & Amann, F., 2018. Quantitative assessment of Pastor, M., 2014. 2D viscoplastic finite element modelling of slow land-
digital image correlation methods to detect and monitor surface displace- slides: the Portalet case study (Spain), Landslides, 11(1), 29–42.
ments of large slope instabilities, Rem. Sens., 10(6), 865. Fiolleau, S., Jongmans, D., Bièvre, G., Chambon, G., Baillet, L. & Vial,
Bontemps, N., Lacroix, P., Larose, E., Jara, J. & Taipe, E., 2020. Rain and B., 2020. Seismic characterization of a clay-block rupture in Harmalière
small earthquakes maintain a slow-moving landslide in a persistent critical landslide, French Western Alps, Geophys. J. Int., 221(3), 1777–1788.
state, Nat. Commun., 11(1), 780. Geimer, P. R., Finnegan, R. & Moore, J. R., 2020. Sparse ambient resonance
Borcherdt, R. D., 1970. Effects of local geology on ground motion near San measurements reveal dynamic properties of freestanding rock arches.
Francisco Bay∗, Bull. seism. Soc. Am., 60(1), 29–61. Geophys. Res. Lett., 47(9), e2020GL087239.
310 M. Häusler et al.

Gischig, V. S., Eberhardt, E., Moore, J. R. & Hungr, O., 2015. On the Michel, C., Guéguen, P., Lestuzzi, P. & Bard, P. Y., 2010. Comparison be-
seismic response of deep-seated rock slope instabilities — Insights from tween seismic vulnerability models and experimental dynamic properties
numerical modeling, Eng. Geol., 193, 1–18. of existing buildings in France, Bull. Earthq. Eng., 8(6), 1295–1307.
Glueer, F., Loew, S., Manconi, A. & Aaron, J., 2019. From toppling to Nowicki Jessee, M. A. et al., 2018. A global empirical model for near-
sliding: progressive evolution of the Moosfluh Landslide, Switzerland, J. real-time assessment of seismically induced landslides, J. geophys. Res.,
geophys. Res., 124(12), 2899–2919. 123(8), 1835–1859.
Gojcic, Z., Schmid, L. & Wieser, A., 2021. Dense 3D displacement vector Perron, V., Gélis, C., Froment, B., Hollender, F., Bard, P.-Y., Cultrera, G.
fields for point cloud-based landslide monitoring, Landslides, 18, 3821– & Cushing, E. M., 2018. Can broad-band earthquake site responses be
3832. predicted by the ambient noise spectral ratio? Insight from observations
Häusler, M., Michel, C., Burjánek, J. & Fäh, D., 2019. Fracture net- at two sedimentary basins, Geophys. J. Int., 215(2), 1442–1454.
work imaging on rock slope instabilities using resonance mode analysis, Pischiutta, M., Rovelli, A., Salvini, F., Di Giulio, G. & Ben-Zion, Y., 2013.
Geophys. Res. Lett., 46(12), 6497–6506. Directional resonance variations across the Pernicana Fault, Mt Etna, in
Häusler, M., Michel, C., Burjánek, J. & Fäh, D., 2021. Monitoring the relation to brittle deformation fields, Geophys. J. Int., 193(2), 986–996.
Preonzo rock slope instability using resonance mode analysis, J. geophys. Pischiutta, M., Salvini, F., Fletcher, J., Rovelli, A. & Ben-Zion, Y., 2012.
Res., 126, e2020JF005709, doi: 10.1029/2020JF005709. Horizontal polarization of ground motion in the Hayward fault zone at
Heim, A., 1932. Bergsturz und Menschenleben, Beiblatt zur Vierteljahrss- Fremont, California: dominant fault-high-angle polarization and fault-
chrift der Naturforschenden Gesellschaft in Zürich, 77, https://www.gra- induced cracks, Geophys. J. Int., 188(3), 1255–1272.

Downloaded from https://academic.oup.com/gji/article/229/1/299/6433627 by guest on 03 February 2023


nat.ch/system/media/2410/original/77BB20 0.pdf ?1422022711. Poggi, V., Edwards, B. & Fäh, D., 2011. Derivation of a reference shear-wave
Helmstetter, A. & Garambois, S., 2010. Seismic monitoring of Séchilienne velocity model from empirical site amplification, Bull. seism. Soc. Am.,
rockslide (French Alps): analysis of seismic signals and their correlation 101(1), 258–274.
with rainfalls, J. geophys. Res., 115(F3), doi:10.1029/2009JF001532. Poggi, V., Ermert, L., Burjanek, J., Michel, C. & Fäh, D., 2015. Modal
Hjortenberg, E. & Risbo, T., 1975. Monochromatic components of analysis of 2-D sedimentary basin from frequency domain decomposition
the seismic noise in the NORSAR area, Geophys. J. Int., 42(2), of ambient vibration array recordings, Geophys. J. Int., 200(1), 615–626.
547–554. Prieto, G. A., Parker, R. L. & Vernon Iii, F. L., 2009. A Fortran 90 library
Jaboyedoff, M., Oppikofer, T., Abellán, A., Derron, M.-H., Loye, A., Met- for multitaper spectrum analysis, Comput. Geosci., 35(8), 1701–1710.
zger, R. & Pedrazzini, A., 2012. Use of LIDAR in landslide investigations: Rousseeuw, P. J., 1987. Silhouettes: a graphical aid to the interpretation and
a review, Nat. Hazards, 61(1), 5–28. validation of cluster analysis, J. Comput. Appl. Math., 20, 53–65.
Kleinbrod, U., Burjánek, J. & Fäh, D., 2019. Ambient vibration classifi- Swiss Seismological Service (SED) at ETH Zurich, 1983. National Seismic
cation of unstable rock slopes: a systematic approach, Eng. Geol., 249, Networks of Switzerland. https://doi.org/10.12686/sed/networks/ch.
198–217. Thomson, D. J., 1982. Spectrum estimation and harmonic analysis, Proc.
Kleinbrod, U., Burjánek, J., Hugentobler, M., Amann, F. & Fäh, D., 2017. A IEEE, 70(9), 1055–1096.
comparative study on seismic response of two unstable rock slopes within Vallet, A., Charlier, J. B., Fabbri, O., Bertrand, C., Carry, N. & Mudry, J.,
same tectonic setting but different activity level, Geophys. J. Int., 211(3), 2016. Functioning and precipitation-displacement modelling of rainfall-
1428–1448. induced deep-seated landslides subject to creep deformation, Landslides,
Krähenbühl, R. & Nänni, C., 2017. Ist das Dorf Brienz-Brinzauls Bergsturz 13(4), 653–670.
gefährdet?, Swiss Bull. angew. Geol., 22(2), 33–47. Vidale, J. E., 1986. Complex polarization analysis of particle motion, Bull.
Larose, E. et al., 2015. Environmental seismology: what can we learn on seism. Soc. Am., 76(5), 1393–1405.
earth surface processes with ambient noise?, J. appl. Geophys., 116, 62– Wong, L. N. Y., Maruvanchery, V. & Liu, G., 2016. Water effects on rock
74. strength and stiffness degradation, Acta Geotech., 11(4), 713–737.
Le Breton, M., Bontemps, N., Guillemot, A., Baillet, L. & Larose, É., 2021. Yurikov, A., Lebedev, M., Gor, G. Y. & Gurevich, B., 2018. Sorption-induced
Landslide monitoring using seismic ambient noise correlation: challenges deformation and elastic weakening of Bentheim sandstone, J. geophys.
and applications, Earth Sci. Rev., 216, 103518. Res., 123(10), 8589–8601.
Lévy, C., Baillet, L., Jongmans, D., Mourot, P. & Hantz, D., 2010. Dy-
namic response of the Chamousset rock column (Western Alps, France),
J. geophys. Res., 115(F4), doi:10.1029/2009JF001606. S U P P O RT I N G I N F O R M AT I O N
Lloyd, S.(1982). Least squares quantization in PCM, IEEE Trans. Inf. The-
ory, 28(2), 129–137. Supplementary data are available at GJ I online.
Loew, S., Gschwind, S., Gischig, V., Keller-Signer, A. & Valenti, G., 2017. Figure S1. Geomorphological map of the Brinzauls landslide.
Monitoring and early warning of the 2012 Preonzo catastrophic rockslope Figure S2. Photographs of the Brinzauls landslide and geophysical
failure, Landslides, 14(1), 141–154. survey setup.
Manconi, A., Kourkouli, P., Caduff, R., Strozzi, T. & Loew, S., 2018. Mon- Figure S3. Polarization azimuth at BRIZ2 in three frequency bands.
itoring surface deformation over a failing rock slope with the ESA Sen- Figure S4. Polarization dip at BRIZ2 in three frequency bands.
tinels: insights from Moosfluh Instability, Swiss Alps, Rem. Sens., 10(5),
Figure S5. Polarization ellipticity at BRIZ2 in three frequency
672.
bands.
Massey, C. et al., 2018. Landslides triggered by the 14 November 2016 Mw
7.8 Kaikōura earthquake, New Zealand, Bull. seism. Soc. Am., 108(3B), Figure S6. Overview map of regional earthquakes contributing to
1630–1648. the empirical spectral modelling amplification analysis.
Mercerat, E. D., Payeur, J. B., Bertrand, E., Malascrabes, M., Pernoud, M. Tables S1. Regional earthquakes contributing to the empirical spec-
& Chamberland, Y., 2021. Deciphering the dynamics of a heterogeneous tral modelling amplification analysis at station BRIZ2.
sea cliff using ambient vibrations: case study of the Sutta-Rocca overhang Tables S2. Normal-mode shape vectors determined by FDD at
(southern Corsica, France), Geophys. J. Int., 224(2), 813–824. 1.7 and 2.8 Hz.

You might also like