ESurfEarth Surface DynamicsESurfEarth Surf. Dynam.2196-632XCopernicus PublicationsGöttingen, Germany10.5194/esurf-6-955-2018Seismic detection of rockslides at regional scale: examples from the Eastern Alps and feasibility of kurtosis-based event locationSeismic detection of rockfallsFuchsFlorianflorian.fuchs@univie.ac.athttps://orcid.org/0000-0002-2023-5611LenhardtWolfganghttps://orcid.org/0000-0001-9031-3753BokelmannGötzthe AlpArray Working GroupDepartment of Meteorology and Geophysics, University of Vienna, Althanstraße 14, UZA 2, 1090 Vienna, AustriaCentral Institute for Meteorology and Geodynamics, ZAMG, Vienna, AustriaFor further information regarding the team, please visit the link the end of the paper.Florian Fuchs (florian.fuchs@univie.ac.at)29October20186495597028June201810July201819September201816October2018This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/4.0/This article is available from https://esurf.copernicus.org/articles/6/955/2018/esurf-6-955-2018.htmlThe full text article is available as a PDF file from https://esurf.copernicus.org/articles/6/955/2018/esurf-6-955-2018.pdf
Seismic records can provide detailed insight into the mechanisms
of gravitational mass movements. Catastrophic events that generate
long-period seismic radiation have been studied in detail, and monitoring
systems have been developed for applications on a very local scale. Here we
demonstrate that similar techniques can also be applied to regional seismic
networks, which show great potential for real-time and large-scale monitoring
and analysis of rockslide activity. This paper studies 19 moderate-sized
to large rockslides in the Eastern Alps that were recorded by regional
seismic networks within distances of a few tens of kilometers to more than 200 km.
We develop a simple and fully automatic processing chain that detects,
locates, and classifies rockslides based on vertical-component seismic
records. We show that a kurtosis-based onset picker is suitable to detect the
very emergent onsets of rockslide signals and to locate the rockslides
within a few kilometers from the true origin using a grid search and a 1-D
seismic velocity model. Automatic discrimination between rockslides and local
earthquakes is possible by a combination of characteristic parameters
extracted from the seismic records, such as kurtosis or maximum-to-mean
amplitude ratios. We attempt to relate the amplitude of the seismic records
to the documented rockslide volume and reveal a potential power law in
agreement with earlier studies. Since our approach is based on simplified
methods we suggest and discuss how each step of the automatic processing
could be expanded and improved to achieve more detailed results in the
future.
Introduction
Gravitational mass movements shape the surface of
our planet and pose sincere hazards to the human population, in particular in
densely populated mountain regions such as the European Alps. Understanding
the triggers of slope failures allows us to better evaluate their impact on the
evolution of geomorphology and to design mitigation measures or early warning
systems. However, such events may occur spontaneously and in remote areas and
thus remain undetected in many cases. This can introduce significant
uncertainty to, e.g., event inventories and triggering studies. Yet,
comprehensive knowledge and reliable event data are of particular importance
for the assessment of hazards imposed by rapid gravitational mass movements
. This renders remote and preferably real-time
detection methods for rapid gravitational mass movements highly desirable.
Classical approaches such as remote sensing via satellite imagery or
stationary slope monitoring systems are usually limited in either temporal or
spatial resolution and cannot cover vast areas in real time.
In recent years seismology has gained attention for being able to provide
both temporal and spatial resolution for the detection and characterization
or even forecasting of various kinds of mass movements. This includes
landslides , rockfalls
, avalanches
, debris flows
, and bed load transport . Most of the studies that demonstrate the large potential of
seismology for the event characterization of mass movements utilize long-period
seismic radiation created by catastrophic landslides . Seismic broadband observations of such events
allow us to invert for the 3-D landslide force history and time-dependent center
of mass position and – in combination with topography data – enable
seismologists to fully describe a mass wasting event from remote (hundreds to
thousands of kilometers of distance) observations. Such observations have
revealed scaling laws that link seismic observables to the mass and momentum
of massive landslides , help to constrain numerical models
of landslides , and support observations of
frictional weakening during sliding events .
Short-period seismic radiation generated by mass movements is more complex
and challenging to interpret due to complex source mechanisms, the influence of
topography, directional effects, and strong near-surface scattering and
attenuation. report relations between the bulk momentum
of catastrophic landslides and the 3–10 Hz bandpass-filtered envelopes of
the respective seismic signals. At smaller scale, controlled experiments
study the generation of high-frequency seismic waves by mass impact under
field or laboratory conditions . Only
a few studies try to utilize high-frequency seismic waves to detect and
characterize mass movements at local or regional scales. The majority of such
studies rely on seismic data acquired in close proximity to the events,
e.g., for monitoring of unstable slopes or avalanches
. Thus, although such approaches are powerful
at small scale they are limited in spatial coverage .
demonstrate a robust automatic detection and location
scheme for rockfalls inside a volcanic crater on La Réunion island.
first documented a set of rockfalls recorded by a
regional seismic network in the Western Alps, and
document statistical relations between rockfall characteristics and seismic
recordings obtained from the Swiss permanent seismic network. Recently, there
have been efforts to utilize existing regional seismic networks for the
detection and characterization of mass movements . Such networks, which were designed for earthquake monitoring
purposes, usually consist of well-installed and sensitive seismic stations
providing high-quality seismic data in real time and thus offer promising
datasets, especially for the study of rockfalls and rockslides.
Map of the study area in eastern Austria and neighboring countries.
Rockslides are marked by red circles. Bright and dark triangles denote
permanent and temporary seismic stations, respectively. The yellow lines mark
country borders. The inset marks the location of the study area in Europe.
Here we present a study of 19 rockfalls and rockslides that occurred in or
near Austria in the years 2007 to 2017 and were well recorded by permanent
national seismic networks in the Alps during routine earthquake monitoring.
We use this dataset of confirmed events to develop and test automatic
detection and locating algorithms that could be used to systematically search
for additional events in existing and future data. Exploring the feasibility
of a country-wide real-time detection scheme for rockfalls, we focus on
developing simple automatic location routines to automatically
distinguish such events from regional earthquakes.
Dataset
This work is based on seismic recordings of 19 rockfall and
rockslide events that occurred in Austria and the neighboring countries
Switzerland and Italy during the years 2007–2017 (see Fig.
and Table ). The event database was compiled by the
Austrian earthquake service and focuses on rockslides and rockfalls from
Austria and South Tyrol (Italy). These events were manually detected and
classified during routine earthquake monitoring by the Austrian earthquake
service (Central Institute for Meteorology and Geodynamics, ZAMG) and
verified in cooperation with the Austrian Geological Service (GBA). We
additionally include two large-scale rockslides that occurred in Switzerland,
but were also detected by the Austrian colleagues and assigned a magnitude.
Out of these 19 events, 16 rockslides have been independently studied by
field observations. All verified events were either first recognized by an
analyst during routine national earthquake monitoring and later confirmed
by field observations or were first recognized in the field and later clearly
associated with seismic waveforms by analysts at ZAMG. For photographs of the
individual events please follow the references listed at the end of the
paper.
During routine processing of the seismic events, a local magnitude Ml was
assigned by ZAMG to all rockfalls and rockslides based on distance and
maximum amplitude as read from the seismic records, just as if the events
were earthquakes. The measured local magnitude ranges between 0.0 and 2.7.
For all events ground-truth reference coordinates are available from field
observations. However, other than the date and coordinates, few reliable event
parameters are available since most of the events were not studied or mapped
in detail on-site or because proper documentation could not be found.
We performed internet searches for all events listed in Table to obtain on-site photographs and to find information on
the volume of rock that was displaced. For almost all events we were able to
retrieve the volume that was usually reported in local newspapers based on
on-site estimates by local geological surveys. Note that these values might
be subject to large uncertainties and should rather be considered as an
order-of-magnitude estimation.
We obtained continuous waveform data for all 19 events from the European
Integrated Data Archive (EIDA), which hosts data from the permanent broadband
seismic stations in the Alps. For each rockfall we identified stations within
a 300 km radius around the event and downloaded all available data for all
three components (Z, N, E) and at the highest sampling rate available (see Fig.
for network geometry). All data since 2016 are provided at 100 sps sampling rate, while earlier data are partially only available at 25 sps.
For events after 1 January 2016 we also used data from the temporary
AlpArray broadband stations (100 sps), which covered the entire alpine
region,
and densify the seismic network, in particular in Austria (; ).
We use this dataset of confirmed rockslides to develop and
test automatic detection and locating algorithms, which we describe in the
following.
List of rockslides studied in this paper. Origin times are
calculated from the seismic records. The coordinates denote the true location
of the events obtained from field observations. The stations column denotes
the number of stations that show a positive STA / LTA trigger. The distance
column indicates the minimum and maximum distance from the events for these
stations. Slide volumes were obtained from a web search and are usually based
on local newspaper reports; please refer to the Acknowledgements section
for source references. Events that are rockfalls rather than rockslides are
marked with an asterisk (*). Local magnitude Ml as estimated
by the Austrian seismological service (ZAMG). The magnitude refers to the first event in the sequence. The volume
estimates the total mass loss over all stages.
aFor an STA / LTA threshold of
4.0 (see Sect. ); bnot independently verified, no
reference coordinates available; cthe Mellental event occurred in three
stages.
Automatic processing
The first step within the automatic processing chain is the identification of
a rockfall event within the continuous background signal. We cut the seismic
traces to 8 min segments around the known origin time (180 s prior to
and 300 s after origin time) to simplify the processing and to avoid
potential false alarms at this stage of development. We also restrict our
processing to the vertical component only. Prior to any further processing,
we remove the instrument response, apply a 1–5 Hz bandpass filter, and taper
and detrend the sliced data. Note that bandpass filtering is required to
enhance the signal-to-noise ratio, especially to suppress microseism and
long-period noise. Indeed, several earlier studies report this frequency band
as dominant for regional seismic records of gravitational mass movements
. Since many of the older
waveform data are only available at 25 sps sampling rate, we cannot
reasonably extend the bandpass window to higher frequencies. For consistency
we use the same settings even for 100 sps data.
Event detection
For simplicity we first implemented a recursive STA / LTA coincidence trigger
to detect the rockfall signals . We used the following
parameters for event detection: STA window, 5 s; LTA window, 120 s;
trigger-on threshold ratio, 4.0; trigger-off ratio, 1.5; minimum number of
stations, four. All events in our dataset created seismic waves strong enough
to in principle be detected with the values stated above. Table
lists the number of stations with a positive STA / LTA
trigger for each rockfall. The number of stations used for single event
analysis in this study ranges from the minimum of four stations to more than
70 stations. The activation time of the STA / LTA trigger also serves as
the initial signal onset time for further processing.
Kurtosis onset picker
Once our algorithm identified stations with a detectable seismic rockfall
signal via the STA / LTA coincidence trigger it automatically determines the
signal onset at each station. Unlike earthquakes, rockfalls and rockslides
commonly show rather emergent signal onsets and hence we cannot use the
STA / LTA trigger times as event starting times because the trigger-on
threshold is always reached after the signal onset. Since
successfully demonstrated the applicability to rockfall signals, we
implemented a kurtosis-based phase picker to determine the onset of the
emergent rockfall signals. The kurtosis is a statistical value, in this case
characterizing the shape of a given amplitude distribution. It is a positive
scalar defined as the standardized fourth moment about the mean. In discrete
form it is written as
β=1n∑i=1n+1(xi-x‾)41n∑i=1n+1(xi-x‾)22,
where n is the total number of samples, xi the value of the ith
sample, and x‾ the mean over n samples. The kurtosis of a normal
distribution is β=3 and any deviations from this value (i.e., excess
kurtosis) can be used for the detection of potential seismic signals on top
of regular background noise.
Similar to the processing described in and
, we calculate a characteristic function CF(t) of the
seismic signal s(t) within a sliding window of size ΔT.
CF(t)=βs(t-ΔT),…,s(t)
The time window is set to ΔT=5 s and t slides between 10 s
before and 1 s after the preliminary onset time determined by the STA / LTA
trigger. CF(t) has a maximum near the true signal onset, when the kurtosis
β of the seismic amplitude distribution within the sliding window
ΔT is maximized; that is, when the entire time window is dominated by
seismic signals from the event (see Fig. ). However, for
location purposes we are interested in the very first onset of the seismic
signal, which is the first time t at which the characteristic function
CF(t) starts to deviate from the background level. Thus, we adopt the
procedure of and modify CF(t) as follows.
cCF(k)=∑i=1kαiwithαi=CFi+1-CFiif(CFi+1-CFi)≥0αi=0otherwise
The function cCF can be read as the cumulative sum of the slope of CF,
and its value increases most drastically at the time of the signal onset.
Thus, we define the time t at which the time derivative d(cCF)/dt is
maximized as the final signal onset time. If several maxima of
d(cCF)/dt lie close to each other we define the first one as the signal onset time (see
Fig. ).
Examples for the performance of the kurtosis picker. All waveforms are
from 1–5 Hz bandpass-filtered vertical components. Panels (a) and
(b) show an example of the 19 August 2016, Kleine Gaisl, Italy,
rockslide from station OE.SQTA at 95 km of distance. Panels (c) and
(d) show an example of the 1 May 2012, Gamsgrube, Austria, rockslide
from station OE.FETA at 82 km of distance. Panels (b) and (d)
show close-ups of the grey-shaded parts of the waveforms in (a) and
(c), respectively. The vertical axes in (b) and
(d) indicate the values of CF. For perfectly Gaussian noise we
expect a value CF =3.0, which is marked by the dashed horizontal lines.
Vertical lines denote picks for the event onset and end. Solid red line:
onset pick based on maximum d(cCF)/dt. Dashed red line: onset time
of STA / LTA trigger. Solid blue line: event end time as given by the
1.1 times the pre-event noise condition (see Sect. 3). Dashed blue line: end
time of STA / LTA trigger (for comparison; not used for any processing).
Location quality based on kurtosis picks. The deviation indicates
the discrepancy between the final location result and the true location of
the event. Four events could not be located due to an insufficient number of
picks.
aNumber of
stations (number of picks) used for location routine;
this number may deviate from the number of stations that passed the STA / LTA
trigger (see Table ) because the kurtosis algorithm may
not have found viable onset picks. bOnly the strongest event from the
sequence is listed.
Record sections (signal vs. distance) of the vertical component for
two large rockslides. All data are bandpass-filtered between 1 and 5 Hz.
(a) Kleine Gaisl, Italy, 19 August 2016, is an event example that does not
show a clear second arrival. (b) Mellental, Austria, 25 March 2016,
does show a distinct second arrival for stations farther than 50 km from the
origin. Black lines mark expected arrival times for a constant travel time of
5.0 and 3.0 km s-1, respectively.
Origin time and event location
Figure shows seismic record sections for two large-scale
rockslides in different areas of the Eastern Alps with patterns of
distinct seismic phase arrivals, which are common for most of the rockslides
in this study. Despite the emergent character of the rockslide signals we can
identify a first arrival that travels with an apparent velocity of
approximately 5.0 km s-1. We thus assume that this arrival is a P wave. For
eight events (Einserkofel, Hochwand, Gamsgrube, Trins, Stubaital,
Dobratsch, Mellental, Zwölferkofel) a distinct second arrival is visible,
which is usually stronger than the first arrival and sometimes (in the case
of a
low signal-to-noise ratio) is the only visible signal. This arrival travels
with an apparent velocity of approximately 3.0 km s-1 and we suggest that it is
due to S waves or surface waves (see Discussion section). We exclude the possibility that the
two distinct arrivals reflect two separate events, since with increasing
distance we observe increasing separation time. In addition, no such
separation is visible on the records of the stations closest to the
rockslide.
We run a grid search to estimate the origin time and location of the
rockslides based on the onset times determined by the kurtosis picker. The
search area is a rectangle with 5 km grid spacing spanned by all seismic
stations with positive STA / LTA detection. Time is scanned in steps of 2 s
between the earliest measured onset time (latest possible origin time) and
an estimated earliest possible origin time that is set as the first onset
pick minus the maximum travel time along the grid diagonal. For each grid
point and each time step we calculate the theoretical arrival time (fixed
velocity of 5.0 km s-1, no topography) for all stations and its
difference (residual) to the measured onset time. The grid point and time at which the
root mean square (RMS) value of the set of station residuals is minimized is
set as the preliminary origin time and event location (see Fig. ). For one-third of the rockslides analyzed within this
study the simple grid search location is already quite satisfactory, with
results that are significantly less than 10 km from the true rockslide
location.
Example for a grid search result (rockfall in Tscheppaschlucht,
Austria, 23 October 2011). Black triangles mark the stations used for the
grid search. Colors indicate the root mean square travel time residuals among
all stations (for the best-fitting origin time and for a fixed velocity of
5.0 km s-1). Note that colors are smoothed between grid points (small
black dots). The green dot represents the grid point that minimizes the set
of travel time residuals and thus marks the preliminary location of the
rockslide.
To overcome the simplifications of the grid search we subsequently perform an
iterative location routine as is done for earthquakes using the HYPOCENTER
code and a simple 1-D velocity model suitable for the
Eastern Alps . The kurtosis-based onset picks are treated
as crustal P waves. The results are summarized in Table and
demonstrate that good location accuracy can be
achieved with regional seismic networks even for emergent rockslide signals.
Eight of 18 tested events were located less than 6 km from the true location. Four
events could not be located due to a very low signal-to-noise ratio or
insufficient number of stations. We discuss possible limitations and reasons
for outliers as well as the robustness of the results in the Discussion
section.
Discrimination from regional earthquakes
A key aspect for automatic processing of seismic rockslide data is to
distinguish such events from earthquakes and other potential sources of
seismicity. suggest a set of parameters that are
extracted from the seismic signal and are systematically different for
earthquakes and rockslides. Here we explore if this simple concept that was
successfully applied on a local scale can be extended to the regional scale.
For each rockslide signal on each available station we extract the following
three parameters (see Fig. ): (1) the kurtosis of the
envelope of the entire signal (EnvKurto); (2) the ratio between
maximum amplitude and mean amplitude (Max / Mean); and (3) the ratio of the
duration (Inc / Dec) of the increasing signal flank (signal start to
maximum amplitude) compared to the duration of the decreasing signal flank
(maximum amplitude to signal end). The end time of the event is defined as
the time at which the 2 s moving average of the signal envelope decayed to 1.1
times the pre-event levels. The pre-event amplitude is estimated as the
mean amplitude within a 60 s window 5 s prior to the first signal onset.
We extract the same three parameters from a set of regional earthquake
records in order to identify potential differences between rockslides and
earthquakes. We downloaded data for 31 earthquakes (Ml<3.5) within
August 2015 and January 2016 that occurred in or near western Austria. Thus, the
earthquakes occurred in the same area as the rockslides and induced similar
levels of shaking (see Fig. S1 and Table S1 and the Supplement for details). The processing of the earthquake data was the same as
for the rockslide data and we read the parameters described above for each
earthquake at each available station.
Figure shows the distribution of potential
discrimination parameters extracted from rockslides and earthquakes. For all
parameters both distributions overlap but they peak at different values.
Notably, for rockslides all values measured for the kurtosis of the envelope
(EnvKurto) and the ratio of maximum to mean amplitude
(Max / Mean) stay below a certain threshold compared to
earthquakes. We make use of this observation and define a simple decision
criterion for whether an event should be declared as a rockslide or earthquake. An
event is considered a rockslide if the mean value measured over all
stations satisfies the following condition.
This way all 19 rockslides and all 31 regional earthquakes are correctly
identified and we demonstrate that even on a regional scale it might be
possible to distinguish rockslides from earthquakes based on a few simple
criteria. We introduce potential extensions of this scheme in the Discussion
section.
Volume–magnitude relation
Besides the event location the event volume is a crucial parameter for an
assessment of a rockslide. Thus we attempt to relate the slide volume to the
local magnitude Ml, a parameter that is routinely determined for seismic
events by any seismological service. Several studies attempt to relate the volume of mass
movements to the measured seismic energy or amplitude. However, derived
scaling relations are often only loosely constrained due to, e.g., a limited
number of events, generally large scatter, or insufficient information about
the event. From the 19 events studied here, 15 rockslides have a magnitude
assigned by ZAMG and a volume estimate available (see Table ).
Figure shows the local magnitude
as a function of the event volume. Note that we exclude the data pair
(Ml=0.0, V=150000; Schwaz event) since the volume estimate is likely
wrong. Although the proposed fit is not well constrained (R2=0.60) due
to large scatter and limited data points, the distribution suggests a linear
relation between the local magnitude Ml and the logarithmic volume
V.
Ml=-0.60+0.44logV
Since the local magnitude Ml=log(A/A0) is a logarithmic measure of the
seismic amplitude A this translates into a power-law relation between the
seismic amplitude A and the rockslide volume V, including a regional
correction term A0 that depends on the epicentral distance corrections
applied during the calculation of Ml.
A=A00.25+V0.44
Distributions of the three different discrimination parameters for
rockslides and earthquakes. Panels (a), (c), and
(e) show the definition of the respective parameters.
Panels (b), (d), and (f) show the frequentness of
the respective parameters in logarithmic scale. Note that the total number of
parameter reads is slightly higher for earthquakes than for rockslides and
the distributions are not normalized. Green colors mark the values read from
rockslide records, and blue colors mark the values read from earthquake records.
The red lines in (b, d, f) mark the respective thresholds for the
decision criterion (see Eq. ).
Discussion
Here we demonstrated that regional seismic networks can be used to reliably
detect moderate- to large-sized rockslides to distances up to more than 200 km. Such seismic networks cover vast areas and record data
continuously, and many networks provide data in real time. Thus, they allow
for the regional monitoring of potentially catastrophic mass movements, and they
additionally provide a temporal resolution that is unmatched by classical
methods such as remote sensing. Here we suggest several processing steps to
analyze the seismic signal generated by rockslides and show that simple
concepts and easy-to-integrate tools already provide reasonable insight into
the events. This demonstrates that even large datasets may be screened for
rockslide data automatically. While this shows the potential of regional
seismic records to study gravitational mass movements, there is much room for
improvement that may strongly increase the quality of the extractable
information. All processing steps including the event location and
characterization were performed completely automatically without the intervention
of a human analyst. In particular, no attempt was made to remove outliers or,
e.g., wrong onset picks, which in some cases greatly reduces the quality of
the location result. Still, our simplistic approach may be complemented in
most of the processing steps to increase the robustness of the results.
Event detection
We have shown that all moderate- to large-sized rockslides in this study could
in principle be detected with an STA / LTA coincidence detector that is widely
used by, e.g., seismological observatories and generally serves as a fast
algorithm to screen datasets for events. However, STA / LTA detectors need to
be balanced between sensitivity and the rate of false alarms. While the STA / LTA
settings reported above safely detect all of our events we did not check
how many false alarms would be introduced if a continuous data stream was
analyzed (we cut the data to 8 min around the events). However, the
STA / LTA triggering threshold level of 4.0 used in this study is commonly used
for sites that are quiet on average . Increasing the number of stations
needed for a positive result can in this case be used to lower the false alarm
rate. Generally, there are more sensitive yet sometimes more computationally
intensive algorithms to detect events in continuous seismic data.
demonstrate how alpine rockslides can be automatically
detected on regional networks using hidden Markov models, which allows us to
simultaneously detect and classify mass movements within seismic records.
report that the predictive multi-band detector
FilterPicker is suitable to detect and phase-pick
emergent seismic signals of rockslides. Lassie is a stack-and-delay-based coherence detector to find and locate events at the same time
and may also be applicable to rockslide signals.
demonstrate how coherent volcanic tremor signals can be
detected and classified on a regional seismic network based on network
covariance matrices. Since rockslide signals in several aspects resemble
tremor signals (emergent onset, long duration, frequency content) this
concept might also be applicable to rockslide detection. Template matching
and subspace detectors are commonly used for earthquake
and tremor detection, but we speculate that such methods may not be suitable
for rockslide detection, as for every event waveforms are highly individual
because of the complexity and variability in source mechanisms.
Local magnitude of all rockslides versus their volume (black dots).
The distribution indicates a linear relation (blue line) between magnitude
and logarithmic volume. The equation with the best-fitting parameters and the
coefficient of determination R2 are indicated above the graph. The data
pair (Ml=0.0, V=150000 m3; marked red) is likely an
outlier due to wrong volume estimate. We thus excluded this point from the
linear fit.
Kurtosis picker performance and location accuracy
designed a robust onset picker for rockslide signals
based on a transition in the kurtosis. However, the method was only applied
at a very local scale (network extension of a few kilometers) around a volcano.
also document the performance of a kurtosis picker for
earthquake localization on regional seismic networks. Here we show that this
concept could also be applied to the rather emergent signals induced by
gravitational mass movements at regional distances. Eight of 14 locatable
events in this study could be located within a few kilometers of deviation from
the true location (see Table ), which shows that based on
onset picks a similar precision as for earthquakes is possible. However, some
of the locations should be considered lucky hits, as the number
of stations is low and the azimuthal gap is large, strikingly for some of the
most well-located events. We do in fact observe that the location results
currently lack robustness and may change by a few kilometers when certain
parameters of the kurtosis picker (e.g., the length of the moving window,
bandpass filter corner frequencies) are adjusted. This is most likely due to
both unfavorable noise conditions and to the simplistic processing we
used for demonstration purposes. Additionally, we did not implement automatic
outlier handling at this stage. Several of the bad locations listed in Table
can be explained by strong outliers in the kurtosis
picks due to noise. We expect that picking accuracy can be greatly improved
if measures are taken to make the kurtosis picker more robust and to exclude
outliers. Future work should include all three components of the seismic
record and use different narrow frequency bands for comparison, as suggested
by . We expect that evaluating the kurtosis pick among
different frequency bands would suppress outliers (due to noise) and thus
make the onset determination more robust and precise. Yet, in this study –
due to low sampling rate for older records – we could not extend the
processing to higher frequencies. Lower frequencies are very weak in
amplitude or absent for almost all rockslides in this study. This is in line
with observations from several other studies that report the 1–5 Hz
frequency range as the dominant one for regional seismic records of
rockslides .
Besides kurtosis methods, pickers based on, e.g., autoregressive prediction
might be very suitable for emergent onset picks, as
they include frequency and phase information in addition to the amplitude
(kurtosis pickers are only based on amplitudes). Since determining the onset
of an emergent signal is anyways challenging, pickless location routines such
as waveform correlation should also be explored for
mass movements. suggest combining the location probabilities
obtained from seismic waves with location probabilities based on terrain
slopes to narrow down the potential source areas.
For location purposes we assumed the first onset of the rockslide signals to
be a direct, i.e., crustal, P wave. The observed average phase velocity of the
first arrival is approximately 5.0 km s-1 (see Fig. ), which
is similar to the observations by and represents a
typical value for P-wave velocities in the upper crust of the Eastern Alps
. For some events
(Einserkofel, Hochwand, Gamsgrube, Trins, Stubaital, Dobratsch,
Mellental, Zwölferkofel) a very distinct second arrival is visible (see
Fig. b) that travels at lower velocities of approximately
3.0 km s-1. In this velocity range we potentially expect both crustal S waves
and surface waves. If the type of wave was clearly identifiable a second phase
pick would be available that could drastically increase the location
accuracy. The majority of events (Fig. a) show no clear
second onset and amplitudes gradually increase towards the maximum after the
first onset. This cigar-type shape is more commonly found in other
seismic studies of landslides and rockslides . For such events we observe that the signal group
around the maximum amplitude travels slower than the first onset, which
suggests that P waves and other types of waves mix within the signal and
complicate any in-detail analysis of the seismic phases or polarization. The
mechanism of each individual rockslide event likely influences the relative
strength at which certain wave types are generated. We also suggest that
depending on the slide mechanism, e.g., P waves and S waves must not
necessarily be excited at the same time during the event. Additionally, a
rockslide is potentially a very directional source of seismic energy that
may introduce anisotropic radiation patterns for the seismic energy.
point out the influence of scattering at surface topography
for location purposes and we should note that gravitational mass movements
might be particularly affected by such effects since they occur in areas of
pronounced topography and at the earth's surface.
Event discrimination
We show that rockslides and earthquakes from the same source region can be
discriminated by a few simple parameters such as the ratio between the maximum and
mean amplitude of the seismic signal or the amplitude distribution.
present a robust decision criterion only based on the
ratio Ml/Md of the local magnitude Ml and the duration magnitude
Md. proposed combining several criteria within a
simple fuzzy logic decision algorithm and we suggest that similar approaches
can also safely distinguish rockslides from earthquakes on a regional scale.
Note, however, that each region where such methods is applied might require
individual modification of the decision thresholds for each parameter.
Recently, more sophisticated decision algorithms based on machine learning
have been developed that allow us to classify any kind of seismic event within a
huge event database with great precision, after being trained by selected
known events. demonstrate how a single training event
can be used to scan continuous data for rockslides based on hidden Markov
models. Classifiers based on random forest algorithms were successfully
applied to classify gravitational mass movements and other events in several
different settings, such as volcanoes and slow-moving
landslides , and show great potential for application
to regional seismic networks . Random forest classifiers
work more reliably the more training events are available. Recent studies
demonstrate that sensitivities higher than 85 % can be achieved if just 10 %
of the events inside a dataset are used to train the algorithm
. In the work of this
corresponds to 20–40 training events per event category, which is the
same order of magnitude as the number of events in this study, suggesting
that these could be sufficient to screen larger datasets.
Volume estimation
Extracting reliable volume or mass information from seismic records of
mass movement remains challenging and requires more research on the factors
influencing the efficiency of seismic wave generation. Among these factors
are the bulk mass, the drop mechanisms (free fall and impact versus
sliding), the slope, and the runout distance. For the 19 events in this study
we can only estimate the drop mechanism from photographs, which is not always
conclusive. While the majority of events would classify as rockslides, some
may include a free-fall phase and could rather be regarded as rockfalls (see
Table ). For catastrophic events that generate strong
long-period signals, such properties can be inverted from the seismic
data . Short-period radiation
is more complex to interpret, however. report simple
scaling relations between the bulk mass momentum and short-period seismic
amplitudes for catastrophic landslides from within the same source area if
source mechanisms are comparable among different events. They also report similar
observations for controlled single-block fall experiments
. At local scale, knowledge of the topography and a large
number of events helps to constrain parameter estimates based on the seismic
signals . At regional scale, however, unknown scattering,
attenuation, and propagation of short-period seismic waves may obscure
any potential scaling relations.
point out that regional attenuation relations extracted
from earthquakes may not be applicable to rockfall records and thus local
magnitudes may not properly reflect the amount of seismic energy released by
the source. They suggest that peak ground velocity is not a good measure to
characterize rockfall signals. In contrast, deduct
reasonably well-constrained relationships between rockslide parameters and
the seismic peak ground velocity. This is in agreement with our findings that
show an acceptable power-law relation between the averaged maximum seismic
amplitude and the slide volume. Note, however, that apart from the volume
estimate the local magnitude may not be very well defined, especially
for low-magnitude (Ml<2) events with only a few amplitude readings
available. suggest that the regional propagation and
attenuation of rockslide signals is strongly influenced by topography. In
addition, several studies observe that seismic efficiency – the ratio of
available potential energy over the released seismic energy – is usually low
for gravitational mass movements . This may in part explain the poor correlations between seismic
amplitudes and the rockslide volumes for several studies (including this
one), since it suggests that a large part of the potential energy is released
through other processes (e.g., friction, cracking, plastic deformation) and
not transmitted seismically . attempt
to derive a scaling law for rockslide volume not based on seismic
amplitudes but on the duration magnitude Md, and they show a reasonable
empirical correlation even for events of very different mechanisms and origin
areas.
A general drawback of many studies (including this one) that aim to identify
scaling relations for seismic energy created by gravitational mass movements
at regional scale is the limited number of events . This is partly due to the limited availability of
high-quality seismic data (network density, sampling rate), geographical
restrictions (e.g., country borders), or lack of reliable event information
(e.g., volume). Advancing our knowledge about short-period seismic radiation
created by gravitational mass movements now calls for several actions:
merging or cross-checking national event databases, which unfortunately
often end at country borders, should greatly improve the number of events
available for analysis and the robustness of the event parameters.
Multidisciplinary approaches should also be explored to constrain event parameters
routinely via, e.g., remote sensing. Finally, efficient data screening
algorithms will allow us to detect and classify gravitational mass movements
inside huge datasets, such as the AlpArray seismic network
. This will drastically increase the number of events to
study and thus opens new possibilities to investigate the triggers and
mechanisms of gravitational mass movements.
Conclusions
We have outlined simple methods on how to search for seismic
signatures of rockslides in data from regional seismic networks up to more
than 200 km from the origin. Kurtosis-based phase pickers allow us to reliably
detect the onset of rockslide signals despite their emergent character.
Resulting location accuracies are in the range of a few kilometers and can
potentially be further reduced by incorporating proper handling of outliers
and if secondary phases can be clearly associated. Automatic discrimination
from earthquakes and other local or regional sources is possible by a simple
combination of three decision parameters, such as maximum-to-mean amplitude
ratio. Based on a larger set of similar parameters, the future application of
machine-learning techniques to data from regional seismic networks promises
automatic event classification with great accuracy. This will likely increase
the number of seismically detected rockslide events at regional scale. Larger
and better parameterized datasets of rockslides will clarify scaling
relations between event parameters and seismic observables and will help to
better understand the seismic waves created by gravitational mass movements.
Regional seismic networks can cover vast areas and at the same time provide
continuous data for very long time series. This combination of spatial
coverage and temporal resolution is currently unmatched by other geophysical
methods. Thus, seismic networks are ideally suited to remotely study
time-dependent rockslide activity. This may include long-term variations
in rockslide activity potentially linked to climate change, fore- and
after-slides of a main event, and more detailed insight into rockslide
triggering factors.
The majority of the seismic waveform data used in this study are openly
available for download at the European Integrated Data Archive (EIDA;
http://www.orfeus-eu.org/data/eida/index.html, last access: October 2018).
Waveform data with network code Z3 were acquired from the temporary AlpArray Seismic Network , which at the time of publication was not openly available
according to the decision of the AlpArray Working Group. Please visit
http://www.alparray.ethz.ch/en/seismic_network/backbone/data-access/ (last access: October 2018) for a
complete description of data access.
All processing required for this paper was done using the ObsPy toolbox
. For location purposes we made use of certain
modules of the SEISAN analysis software package .
Rockslide photographs and references for volume estimations in
Table are as follows (last access date for all links below: October 2018).
The Supplement related to this article is available online at: https://doi.org/10.5194/esurf-6-955-2018-supplement
The AlpArray Working Group: http://www.alparray.ethz.ch/en/seismic_network/backbone/data-policy-and-citation/ (last access: 25 October 2018).
FF led the study, developed the codes, and wrote the paper.
WL compiled the event list, provided event details, and analyzed individual events.
GB supervised the study and helped compile the paper. The AlpArray Working Group jointly
installed the seismic network Z3 and developed guidelines to ensure data quality.
The authors declare that they have no conflict of
interest.
This article is part of the special issue “From process to
signal – advancing environmental seismology”. It is a result of the EGU
Galileo conference, Ohlstadt, Germany, 6–9 June 2017.
Acknowledgements
This work was funded by the Austrian Science Fund FWF project number P26391.
This work benefited from fruitful discussions at the EGU Galileo conference
on Environmental Seismology 2017, Ohlstadt, Germany.
We thank Helmut Hausmann (ZAMG) for his help in compiling the event parameters
and independent information. Nils Tilch and Alexandra Haberler of the
Geological Survey of Austria (GBA) are thanked for the cooperation and help
in compiling the event database, verification of seismic data, and alerting us
to new rockslides.
We acknowledge the use of data from the AlpArray network (code Z3;
); please visit the project home page at http://www.alparray.ethz.ch (last access: 25 October 2018)
for a full list of people contributing to the AlpArray
seismic network.
For this study we used seismic data from several permanent seismic networks
and we appreciate the continuous operation of these seismic networks by the
responsible institutions: BW net , CH net , CR net,
FR net , GN net , GU net , GR net, IV
net , MN net , NI net , OE net, OX net
, SI net, SL net , and ST net . We
acknowledge ORFEUS and EIDA for providing the tools to access the seismic
data. Edited by: Michael Dietze
Reviewed by: Naomi Vouillamoz and one anonymous referee
ReferencesAllstadt, K.: Extracting source characteristics and dynamics of the August
2010
Mount Meager landslide from broadband seismograms, J. Geophys. Res.-Earth, 118, 1472–1490, 10.1002/jgrf.20110, 2013.AlpArray Seismic Network: AlpArray Seismic Network (AASN) temporary
component, AlpArray Working Group, Datacite link: http://data.datacite.org/10.12686/alparray/z3_2015 (last access: 25 October 2018),
Project webpage: http://www.alparray.ethz.ch (last access: 25 October 2018), 10.12686/alparray/z3_2015,
2015.Arrowsmith, S., Young, C., Ballard, S., Slinkard, M., and Pankow, K.:
Pickless Event Detection and Location The Waveform Correlation Event Detection System
(WCEDS) Revisited, B. Seismol. Soc. Am., 106,
2037–2044, 10.1785/0120160018, 2016.Baillard, C., Crawford, W. C., Ballu, V., Hibert, C., and Mangeney, A.: An
Automatic Kurtosis-Based P- and S-Phase Picker Designed for Local Seismic
Networks, B. Seismol. Soc. Am., 104, 394–409,
10.1785/0120120347, 2014.Burtin, A., Hovius, N., Milodowski, D. T., Chen, Y.-G., Wu, Y.-M., Lin,
C.-W.,
Chen, H., Emberson, R., and Leu, P.-L.: Continuous catchment-scale monitoring
of geomorphic processes with a 2-D seismological array, J. Geophys. Res.-Earth, 118, 1956–1974,
10.1002/jgrf.20137, 2013.Burtin, A., Hovius, N., and Turowski, J. M.: Seismic monitoring of torrential
and fluvial processes, Earth Surf. Dynam., 4, 285–307,
10.5194/esurf-4-285-2016, 2016.Dammeier, F., Moore, J. R., Haslinger, F., and Loew, S.: Characterization of
alpine rockslides using statistical analysis of seismic signals, J. Geophys. Res., 116, F04024, 10.1029/2011JF002037, 2011.Dammeier, F., Moore, J. R., Hammer, C., Haslinger, F., and Loew, S.:
Automatic
detection of alpine rockslides in continuous seismic data using hidden Markov
models, J. Geophys. Res.-Earth, 121, 351–371,
10.1002/2015JF003647, 2016.Delannay, R., Valance, A., Mangeney, A., Roche, O., and Richard, P.: Granular
and particle-laden flows: from laboratory experiments to field observations,
J. Phys. D. Appl. Phys., 50, 053001, 10.1088/1361-6463/50/5/053001, 2017.Deparis, J., Jongmans, D., Cotton, F., Baillet, L., Thouvenot, F., and Hantz,
D.: Analysis of Rock-Fall and Rock-Fall Avalanche Seismograms in the French
Alps, B. Seismol. Soc. Am., 98, 1781–1796,
10.1785/0120070082, 2008.Department of Earth and Environmental Sciences, Geophysical Observatory,
University of Munchen: BayernNetz, International Federation of Digital
Seismograph Networks, Other/Seismic Network, 10.7914/SN/BW, 2001.Ekström, G. and Stark, C. P.: Simple scaling of catastrophic landslide
dynamics, Science, 339, 1416–1419, 10.1126/science.1232887, 2013.Farin, M., Mangeney, A., de Rosny, J., Toussaint, R., Sainte-Marie, J., and
Shapiro, N. M.: Experimental validation of theoretical methods to estimate
the energy radiated by elastic waves during an impact, J. Sound Vib., 362, 176–202, 10.1016/j.jsv.2015.10.003, 2016.
Feng, Z.: The seismic signatures of the 2009 Shiaolin landslide in Taiwan,
Nat. Hazards Earth Syst. Sci., 11, 1559–1569,
10.5194/nhess-11-1559-2011, 2011.Fuchs, F., Kolínský, P., Gröschl, G., Apoloner, M.-T., Qorbani, E.,
Schneider, F., and Bokelmann, G.: Site selection for a countrywide temporary
network in Austria: noise analysis and preliminary performance, Adv. Geosci.,
41, 25–33, 10.5194/adgeo-41-25-2015, 2015.Fuchs, F., Kolínský, P., Gröschl, G., Bokelmann, G., and the AlpArray
Working Group: AlpArray in Austria and Slovakia: technical realization, site
description and noise characterization, Adv. Geosci., 43, 1–13,
10.5194/adgeo-43-1-2016, 2016.Geological Survey-Provincia Autonoma di Trento: Trentino Seismic Network,
International Federation of Digital Seismograph Networks,
10.7914/SN/ST, 1981.Gualtieri, L. and Ekström, G.: Seismic Reconstruction of the 2012 Palisades
Rockfall Using the Analytical Solution to Lamb's Problem, B. Seismol. Soc. Am., 107, 63–71, 10.1785/0120160238,
2017.Hammer, C., Fäh, D., and Ohrnberger, M.: Automatic detection of wet-snow
avalanche seismic signals, Nat. Hazards, 86, 601–618,
10.1007/s11069-016-2707-0, 2017.
Hausmann, H., Hoyer, S., Schurr, B., Bruckl, E., Houseman, G., and Stuart,
G.:
New seismic data improve earthquake location in the Vienna Basin area,
Austria, Austrian J. Earth Sci., 103, 2–14, 2010.Havskov, J. and Ottemoller, L.: SeisAn Earthquake analysis software, Seismol. Res. Lett., 70, 532–534,
10.1785/gssrl.70.5.532, 1999.
Heimann, S., Matos, C., Cesca, S., Rio, I., and Custodia, S.: Lassie: A
versatile tool to detect and locate seismic activity, in preparation, Note:
interested users to preview Lassie can write to:
sebastian.heimann@gfz-potsdam.de, 2018.Helmstetter, A. and Garambois, S.: Seismic monitoring of Sechilienne
rockslide
(French Alps): Analysis of seismic signals and their correlation with
rainfalls, J. Geophys. Res., 115, F03016,
10.1029/2009JF001532, 2010.Hetenyi, G., Molinari, I., Clinton, J., Bokelmann, G., Bondar, I., Crawford,
W. C., Dessa, J.-X., Doubre, C., Friederich, W., Fuchs, F., Giardini, D.,
Graczer, Z., Handy, M. R., Herak, M., Jia, Y., Kissling, E., Kopp, H., Korn,
M., Margheriti, L., Meier, T., Mucciarelli, M., Paul, A., Pesaresi, D.,
Piromallo, C., Plenefisch, T., Plomerova, J., Ritter, J., Rumpker, G., Sipka,
V., Spallarossa, D., Thomas, C., Tilmann, F., Wassermann, J., Weber, M.,
Weber, Z., Wesztergom, V., Zivcic, M., the AlpArray Seismic Network Team,
the AlpArray OBS Cruise Crew, and the AlpArray Working Group: The
AlpArray Seismic Network: A Large-Scale European Experiment to Image the
Alpine Orogen, Surv. Geophys., 39, 1009–1033,
10.1007/s10712-018-9472-4, 2018.Hibert, C., Mangeney, A., Grandjean, G., and Shapiro, N. M.: Slope
instabilities in Dolomieu crater, Reunion Island: From seismic signals to
rockfall characteristics, J. Geophys. Res., 116, F04032,
10.1029/2011JF002038, 2011.Hibert, C., Mangeney, A., Grandjean, G., Baillard, C., Rivet, D., Shapiro,
N. M., Satriano, C., Maggi, A., Boissier, P., Ferrazzini, V., and Crawford,
W.: Automated identification, location, and volume estimation of rockfalls at
Piton de la Fournaise volcano, J. Geophys. Res.-Earth, 119, 1082–1105, 10.1002/2013JF002970, 2014a.Hibert, C., Ekström, G., and Stark, C. P.: Dynamics of the Bingham Canyon
Mine
landslides from seismic signal analysis, Geophys. Res. Lett., 41,
4535–4541, 10.1002/2014GL060592, 2014b.Hibert, C., Malet, J.-P., Bourrier, F., Provost, F., Berger, F., Bornemann,
P., Tardif, P., and Mermin, E.: Single-block rockfall dynamics inferred from
seismic signal analysis, Earth Surf. Dynam., 5, 283–292,
10.5194/esurf-5-283-2017, 2017a.Hibert, C., Ekström, G., and Stark, C. P.: The relationship between
bulk-mass
momentum and short-period seismic radiation in catastrophic landslides,
J. Geophys. Res.-Earth, 122, 1201–1215,
10.1002/2016JF004027, 2017b.
Hibert, C., Michea, D., Provost, F., Malet, J.-P., and Geertsema, M.: 20 years
of landslide activity in Alaska from automated machine-learning based seismic
detection, Geophysical Research Abstracts, EGU General Assembly 2018, 20,
EGU2018–8595–1, 2018.Husen, S., Kissling, E., Deichmann, N., Wiemer, S., Giardini, D., and Baer,
M.:
Probabilistic earthquake location in complex three-dimensional velocity
models: Application to Switzerland, J. Geophys. Res., 108,
2077, 10.1029/2002JB001778, 2003.INGV Seismological Data Centre: Rete Sismica Nazionale (RSN), Istituto
Nazionale di Geofisica e Vulcanologia (INGV), Italy,
10.13127/SD/X0FXnH7QfY, 1997.Institut de Physique du Globe de Paris (IPGP) & Ecole et Observatoire des
Sciences de la Terre de Strasbourg (EOST): GEOSCOPE, French Global Network
of broad band seismic stations, Institut de Physique du Globe de Paris
(IPGP), 10.18715/GEOSCOPE.G, 1982.Krischer, L., Megies, T., Barsch, R., Beyreuther, M., Lecocq, T., Caudron,
C.,
and Wassermann, J.: ObsPy: a bridge for seismology into the scientific Python
ecosystem, Computational Science & Discovery, 8, 014003,
10.1088/1749-4699/8/1/014003, 2015.Küperkoch, L., Meier, T., Brüstle, A., Lee, J., Friederich, W., and working
group, E.: Automated determination of S phase arrival times using
autoregressive prediction: application to local and regional distances,
Geophys. J. Int., 188, 687–702,
10.1111/j.1365-246X.2011.05292.x, 2012.Lacroix, P., Grasso, J.-R., Roulle, J., Giraud, G., Goetz, D., Morin, S., and
Helmstetter, A.: Monitoring of snow avalanches using a seismic array:
Location, speed estimation, and relationships to meteorological variables,
J. Geophys. Res., 117, F01034, 10.1029/2011JF002106,
2012.Levy, C., Mangeney, A., Bonilla, F., Hibert, C., Calder, E. S., and Smith,
P. J.: Friction weakening in granular flows deduced from seismic records at
the Soufrière Hills Volcano, Montserrat, J. Geophys. Res.-Sol. Ea., 120, 7536–7557, 10.1002/2015JB012151, 2015.Lima, P., Steger, S., Glade, T., Tilch, N., Schwarz, L., and Kociu, A.:
Landslide Susceptibility Mapping at National Scale: A First Attempt for
Austria, in: Advancing Culture of Living with Landslides, edited by: Mikos,
M., Tiwari, B., Yin, Y., and Sassa, K., WLF 2017, Springer, Cham,
10.1007/978-3-319-53498-5_107, 2017.Loew, S., Gschwind, S., Gischig, V., Keller-Signer, A., and Valenti, G.:
Monitoring and early warning of the 2012 Preonzo catastrophic rockslope
failure, Landslides, 14, 141–154, 10.1007/s10346-016-0701-y, 2017.Lomax, A., Satriano, C., and Vassallo, M.: Automatic Picker Developments and
Optimization: FilterPicker – a Robust, Broadband Picker for Real-Time Seismic
Monitoring and Earthquake Early Warning, Seismol. Res. Lett., 83,
531–540, 10.1785/gssrl.83.3.531, 2012.Lopez Comino, J. A., Heimann, S., Cesca, S., Milkereit, C., Dahm, T., and
Zang,
A.: Automated Full Waveform Detection and Location Algorithm of Acoustic
Emissions from Hydraulic Fracturing Experiment, Procedia Engineer., 191,
697–702, 10.1016/j.proeng.2017.05.234, 2017.Lucas, A., Mangeney, A., and Ampuero, J. P.: Frictional velocity-weakening in
landslides on Earth and on other planetary bodies, Nat. Commun., 5,
3417, 10.1038/ncomms4417, 2014.Maceira, M., Rowe, C. A., Beroza, G., and Anderson, D.: Identification of
low-frequency earthquakes in non-volcanic tremor using the subspace
detector method, Geophys. Res. Lett., 37, L06303,
10.1029/2009GL041876, 2010.Maggi, A., Ferrazzini, V., Hibert, C., Beauducel, F., Boissier, P., and
Amemoutou, A.: Implementation of a Multistation Approach for Automated Event
Classification at Piton de la Fournaise Volcano, Seismol. Res. Lett., 88, 878–891, 10.1785/0220160189, 2017.Manconi, A., Picozzi, M., Coviello, V., de Santis, F., and Elia, L.:
Real-time
detection, location, and characterization of rockslides using broadband
regional seismic networks, Geophys. Res. Lett., 43, 6960–6967,
10.1002/2016GL069572, 2016.MedNet project partner institutions: Mediterranean Very Broadband
Seismographic Network (MedNet), Istituto Nazionale di Geofisica e
Vulcanologia (INGV), Italy, 10.13127/SD/fBBBtDtd6q, 1988.Moore, J. R., Pankow, K. L., Ford, S. R., Koper, K. D., Hale, J. M., Aaron,
J.,
and Larsen, C. F.: Dynamics of the Bingham Canyon rock avalanches (Utah, USA)
resolved from topographic, seismic, and infrasound data, J. Geophys. Res.-Earth, 122, 615–640,
10.1002/2016JF004036, 2017.Moretti, L., Mangeney, A., Capdeville, Y., Stutzmann, E., Huggel, C.,
Schneider, D., and Bouchut, F.: Numerical modeling of the Mount Steller
landslide flow history and of the generated long period seismic waves,
Geophys. Res. Lett., 39, L16402, 10.1029/2012GL052511, 2012.Moretti, L., Allstadt, K., Mangeney, A., Capdeville, Y., Stutzmann, E., and
Bouchut, F.: Numerical modeling of the Mount Meager landslide constrained by
its force history derived from seismic data, J. Geophys. Res.-Sol. Ea., 120, 2579–2599, 10.1002/2014JB011426, 2015.OGS (Istituto Nazionale di Oceanografia e di Geofisica Sperimentale) and
University of Trieste: North-East Italy Broadband Network, International
Federation of Digital Seismograph Networks, 10.7914/SN/NI, 2002.OGS (Istituto Nazionale di Oceanografia e di Geofisica Sperimentale):
North-East Italy Seismic Network, International Federation of Digital
Seismograph Networks, 10.7914/SN/OX, 2016.Petschko, H., Brenning, A., Bell, R., Goetz, J., and Glade, T.: Assessing the
quality of landslide susceptibility maps – case study Lower Austria, Nat.
Hazards Earth Syst. Sci., 14, 95–118,
10.5194/nhess-14-95-2014, 2014.Provost, F., Hibert, C., and Malet, J.-P.: Automatic classification of
endogenous landslide seismicity using the Random Forest supervised
classifier, Geophys. Res. Lett., 44, 113–120,
10.1002/2016gl070709, 2017.RESIF: RESIF-RLBP French Broad-band network, RESIF-RAP strong motion
network
and other seismic stations in metropolitan France, RESIF – Reseau
sismologique & geodesique francais, 10.15778/RESIF.FR, 1995.Roth, D. L., Finnegan, N. J., Brodsky, E. E., Rickenmann, D., Turowski,
J. M.,
Badoux, A., and Gimbert, F.: Bed load transport and boundary roughness
changes as competing causes of hysteresis in the relationship between river
discharge and seismic amplitude recorded near a steep mountain stream,
J. Geophys. Res.-Earth, 122, 1182–1200,
10.1002/2016JF004062, 2017.Schmandt, B., Aster, R. C., Scherler, D., Tsai, V. C., and Karlstrom, K.:
Multiple fluvial processes detected by riverside seismic and infrasound
monitoring of a controlled flood in the Grand Canyon, Geophys. Res. Lett., 40, 4858–4863, 10.1002/grl.50953, 2013.Slovenian Environment Agency: Trentino Seismic Network, International
Federation of Digital Seismograph Networks, 10.7914/SN/SL, 2001.Soubestre, J., Shapiro, N. M., Seydoux, L., de Rosny, J., Droznin, D. V.,
Droznina, S. Y., Senyukov, S. L., and Gordeev, E. I.: Network-Based Detection
and Classification of Seismovolcanic Tremors: Example From the Klyuchevskoy
Volcanic Group in Kamchatka, J. Geophys. Res.-Sol. Ea.,
123, 564–582, 10.1002/2017JB014726, 2018.Swiss Seismological Service (SED) at ETH Zurich: National Seismic Networks
of
Switzerland, ETH Zurich, 10.12686/sed/networks/ch, 1983.The ObsPy Development Team: (27 February 2017) ObsPy 1.0.3, Zenodo,
10.5281/zenodo.165134, 2017.Trnkoczy, A.: Understanding and parameter setting of STA/LTA trigger
algorithm,
in: New Manual of Seismological Observatory Practice 2 (NMSOP2), edited by:
Bormann, P., 1–20, Deutsches GeoForschungsZentrum GFZ, Potsdam,
10.2312/GFZ.NMSOP-2_ch4, 2012.University of Genova: Regional Seismic Network of North Western Italy,
International Federation of Digital Seismograph Networks,
10.7914/SN/GU, 1967.van Herwijnen, A. and Schweizer, J.: Monitoring avalanche activity using a
seismic sensor, Cold Reg. Sci. Technol., 69, 165–176,
10.1016/j.coldregions.2011.06.008, 2011.van Herwijnen, A., Heck, M., and Schweizer, J.: Forecasting snow avalanches
using avalanche activity data obtained through seismic monitoring, Cold Reg. Sci. Technol., 132, 68–80,
10.1016/j.coldregions.2016.09.014, 2016.Walter, F., Burtin, A., McArdell, B. W., Hovius, N., Weder, B., and Turowski,
J. M.: Testing seismic amplitude source location for fast debris-flow
detection at Illgraben, Switzerland, Nat. Hazards Earth Syst. Sci., 17,
939–955, 10.5194/nhess-17-939-2017, 2017.Walter, M., Schwaderer, U., and Joswig, M.: Seismic monitoring of precursory
fracture signals from a destructive rockfall in the Vorarlberg Alps, Austria,
Nat. Hazards Earth Syst. Sci., 12, 3545–3555,
10.5194/nhess-12-3545-2012, 2012.Wang, N., Shen, Y., Flinders, A., and Zhang, W.: Accurate source location
from
waves scattered by surface topography, J. Geophys. Res.-Sol. Ea., 121, 4538–4552, 10.1002/2016JB012814, 2016.
Ye, S., Ansorge, J., Kissling, E., and Mueller, S.: Crustal structure beneath
the eastern Swiss Alps derived from seismic refraction data, Tectonophysics,
242, 199–221, 10.1016/0040-1951(94)00209-R, 1995.