Automatic classification of endogenous landslide seismicity using the Random Forest supervised classifier

Provost, Floriane; Hibert, Clément; Malet, Jean‐Philippe

doi:10.1002/2016gl070709

Cited by 121 publications

(134 citation statements)

References 25 publications

(41 reference statements)

Supporting

Mentioning

132

Contrasting

Order By: Relevance

“…Manconi et al, 2016) or multivariate classification approaches (e.g. Provost et al, 2017). However, the duration distributions of rockfalls versus earthquakes already allowed a sufficient discrimination.…”

Section: Rockfall Detection From Continuous Seismic Datamentioning

confidence: 99%

Seismic monitoring of small alpine rockfalls – validity, precision and limitations

Dietze

Mohadjer²,

Turowski

et al. 2017

Earth Surf. Dynam.

View full text Add to dashboard Cite

Abstract. Rockfall in deglaciated mountain valleys is perhaps the most important post-glacial geomorphic process for determining the rates and patterns of valley wall erosion. Furthermore, rockfall poses a significant hazard to inhabitants and motivates monitoring efforts in populated areas. Traditional rockfall detection methods, such as aerial photography and terrestrial laser scanning (TLS) data evaluation, provide constraints on the location and released volume of rock but have limitations due to significant time lags or integration times between surveys, and deliver limited information on rockfall triggering mechanisms and the dynamics of individual events. Environmental seismology, the study of seismic signals emitted by processes at the Earth's surface, provides a complementary solution to these shortcomings. However, this approach is predominantly limited by the strength of the signals emitted by a source and their transformation and attenuation towards receivers. To test the ability of seismic methods to identify and locate small rockfalls, and to characterise their dynamics, we surveyed a 2.16 km 2 large, near-vertical cliff section of the Lauterbrunnen Valley in the Swiss Alps with a TLS device and six broadband seismometers. During 37 days in autumn 2014, 10 TLS-detected rockfalls with volumes ranging from 0.053 ± 0.004 to 2.338 ± 0.085 m 3 were independently detected and located by the seismic approach, with a deviation of 81 +59 −29 m (about 7 % of the average inter-station distance of the seismometer network). Further potential rockfalls were detected outside the TLS-surveyed cliff area. The onset of individual events can be determined within a few milliseconds, and their dynamics can be resolved into distinct phases, such as detachment, free fall, intermittent impact, fragmentation, arrival at the talus slope and subsequent slope activity. The small rockfall volumes in this area require significant supervision during data processing: 2175 initially picked potential events reduced to 511 potential events after applying automatic rejection criteria. The 511 events needed to be inspected manually to reveal 19 short earthquakes and 37 potential rockfalls, including the 10 TLS-detected events. Rockfall volume does not show a relationship with released seismic energy or peak amplitude at this spatial scale due to the dominance of other, process-inherent factors, such as fall height, degree of fragmentation, and subsequent talus slope activity. The combination of TLS and environmental seismology provides, despite the significant amount of manual data processing, a detailed validation of seismic detection of small volume rockfalls, and revealed unprecedented temporal, spatial and geometric details about rockfalls in steep mountainous terrain.Published by Copernicus Publications on behalf of the European Geosciences Union.

show abstract

Section: Rockfall Detection From Continuous Seismic Datamentioning

confidence: 99%

Seismic monitoring of small alpine rockfalls – validity, precision and limitations

Dietze

Mohadjer²,

Turowski

et al. 2017

Earth Surf. Dynam.

View full text Add to dashboard Cite

show abstract

“…Rockfalls and avalanches signals were also characterized at steep debris slides and at rockslides (Helmstetter and Garambois, 2010;Walter et al, 2012;Tonnellier et al, 2013;Provost et al, 2017). In addition, a wide variety of tremor signals have been reported marginally (Gomberg et al, 1995;Brückl and Mertl, 2006;Mertl and Brückl, 2007;Spillmann et al, 2007;Gomberg et al, 2011;Walter et al, 2012;Tonnellier et al, 2013;Provost et al, 2017). No common typology has yet been 20…”

Section: Introductionmentioning

confidence: 94%

“…Gomberg et al (1995) and Gomberg et al (2011) However, a slide-generated source (slow rupture of faults or materials entrained within the faults like trees or boulders, or slow basal slip) is not excluded for tremor-like radiation devoid of gliding frequency and featuring the highest amplitudes at 30 the network most remote location from the road. These events last several minutes and show dominant energies distributed broadly above 30-50 Hz and diminishing toward the Nyquist at 125 Hz (Gomberg et al, 2011 observed with duration of a few seconds to tens of seconds (Walter et al, 2012;Tonnellier et al, 2013;Provost et al, 2017).…”

Section: Tremor Signals 431 Previous Observationsmentioning

confidence: 99%

“…Walter et al (2012) describe earthquake-like events with duration of up to 5 seconds and associated frequency content of 10-80 Hz, which they refer as slidequakes after Gomberg et al (1995). Tonnellier et al (2013) and Provost et al (2017) report quake-like signals with duration of about one second, dominant frequencies around 10 Hz, emergent first arrivals and undistinguishable P-and S-waves.…”

Section: Catalog Designmentioning

confidence: 99%

“…The study confirmed the existence of detectable brittle deformation processes associated to the 5 slide deformation. In Europe, investigated clayey landslides include the Heumoes slope in the Austrian Vorarlberg Alps (Walter and Joswig, 2008;Walter et al, 2011), the Super-Sauze landslide in the French Southwestern Alps (Walter and Joswig, 2009;Walter et al, 2012;Tonnellier et al, 2013;Provost et al, 2017) and the Valoria landslide in the Northern Apennines in Italy (Tonnellier et al, 2013). Case studies carried out at rockslides include for example the Randa rockslide in the Swiss Alps (Eberhardt et al, 2004;Spillmann et al, 2007); the Åknes rockslide in Norway (Roth et al, 2005;Fischer et 10 al., 2014); the Séchilienne rockslide in the Southeastern French Alps (Helmstetter and Garambois, 2010;Lacroix and Helmstetter, 2011); and the Gradenbach, Hochmais-Atemskopf and Niedergallmigg-Matekopf deep-seated rock slope deformations in the Eastern Austrian Alps (Brückl and Mertl, 2006;Mertl and Brückl, 2007;Brückl et al, 2013).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Characterizing the complexity of seismic signals at slow-moving clay-rich debris slides: The Super-Sauze (Southeastern France) and Pechgraben (Upper Austria) case studies

Vouillamoz¹,

Rothmund²,

Joswig³

2017

Preprint

View full text Add to dashboard Cite

Abstract. Soil and debris slides are prone to rapid and dramatic reactivation. Deformation within the instability is accommodated by sliding, whereby weak seismic energies are released through material deformation. Thus, passive microseismic monitoring provides information that relate to the slope dynamics. In this study, passive seismic data acquired at Super-Sauze (Southeastern France) and Pechgraben (Upper Austria) slow-moving clay-rich debris slides ("clayey 10 landslides") are investigated. Observations are benchmarked to previous similar case studies to provide a comprehensive and homogenized typology of seismic signals at clayey landslides. A well knowledge of the various seismic signals potentially triggered by the slope deformation is crucial for the future development of automatic detection systems to be implemented in early-warning systems. Detected seismic events range from short duration (< 2 s) quake-like signals to a wide variety of longer duration tremor-like radiations. Complex seismic velocity structures, low-quality signal onsets and non-optimal 15 seismic network geometry severely impedes the source location procedure, thus rendering source processes characterization challenging. Therefore, we constrain sources location using the prominent waveform attenuation pattern characteristic of near (< ~50 m) seismic events. A local magnitude scale (ML) for clayey landslides is empirically calibrated using calibration shots and hammer blows data. The derived ML scale returns landslide seismicity rates that correlate in general with higher displacement rates. However, high temporal and spatial resolution analyses of the landslide dynamics and hydrology are 20 required to better decipher the potential relations linking landslide-induced seismic signals to landslide deformation.

show abstract

Machine Learning improves warning systems of debris flows

Chmiel¹,

Walter²,

Wenner³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

modern seismic instrumentation has suggested new warning perspectives. Even at large distances (tens to hundreds of kilometers) seismometers can detect high-frequency (>1 Hz) ground unrest induced by debris flows (for a review see Allstadt et al., 2018). This may extend warning times compared to conventional instrumentation within or near torrents, which can only be installed in accessible terrain (e.g., Arattano

show abstract

Automatic classification of endogenous landslide seismicity using the Random Forest supervised classifier

Cited by 121 publications

References 25 publications

Seismic monitoring of small alpine rockfalls – validity, precision and limitations

Seismic monitoring of small alpine rockfalls – validity, precision and limitations

Characterizing the complexity of seismic signals at slow-moving clay-rich debris slides: The Super-Sauze (Southeastern France) and Pechgraben (Upper Austria) case studies

Machine Learning improves warning systems of debris flows

Contact Info

Product

Resources

About