Near-real-time automated classification of seismic signals  of slope failures with continuous random forests

Wenner, Michaela; Hibert, Clément; van Herwijnen, Alec; Meier, Lorenz; Walter, Fabian

doi:https://doi.org/10.5194/nhess-21-339-2021

Articles | Volume 21, issue 1

https://doi.org/10.5194/nhess-21-339-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/nhess-21-339-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 21, issue 1

Research article

|

27 Jan 2021

Research article |

| 27 Jan 2021

Near-real-time automated classification of seismic signals of slope failures with continuous random forests

Michaela Wenner, Clément Hibert, Alec van Herwijnen, Lorenz Meier, and Fabian Walter

Download

Final revised paper (published on 27 Jan 2021)
Preprint (discussion started on 08 Jul 2020)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'Referee comments', Anonymous Referee #1, 27 Jul 2020
- AC1: 'Answer to reviewer comments NHESS-2020-200', Michaela Wenner, 11 Sep 2020
RC2: 'Referee comments', Anonymous Referee #2, 10 Aug 2020
- AC2: 'Answer to reviewer comments NHESS-2020-200', Michaela Wenner, 11 Sep 2020

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

ED: Reconsider after major revisions (further review by editor and referees) (11 Sep 2020) by Yves Bühler

AR by Michaela Wenner on behalf of the Authors (21 Nov 2020) Manuscript

ED: Referee Nomination & Report Request started (24 Nov 2020) by Yves Bühler

RR by Anonymous Referee #1 (07 Dec 2020)

Suggestions for revision or reasons for rejection

The revised version of the text improved in many occasions as the authors did a thorough job. I consider the addition of a further study site a valuable asset that gives the study a bolder shape. Below I give a few comments and suggestions that involve in my view minor changes to the text. Thus, I would not insist on seeing the manuscript another time unless the editor sees an urgent need to decide otherwise.

In general, the addition of the Illgraben site should prompt a (slight) change in the motivation and pitch of the manuscript. My impression is that you could motivate the study by saying that the approach has been shown to work under ideal conditions, which are in many use cases rarely met, and that you want to explore the effects of i) less than ideal network geometries but high SNR events (Pizzo Cengalo), ii) well suited network geometry but low SNR events (Illgraben), and iii) imbalanced training data sets (both? sites). This would give a proper justification of your study design. The new pitch should be better reflected in the abstract (apart from just mentioning you work on two different data sets) and also the end of the introduction.

L 7, change “earthquakes” to “earthquake”

L 8, mention the names of the two locations (Illgraben, Pizzo Cengalo) when listing the data sets

L 9–12, level of detail is imbalanced between the two study site. Consider more equal representation of the results between the two sites.

L 24, LaTEX bibliography issue

L 37, define what you mean with “large events”. Is that for example millions of m³? Or “events that affect entire slopes”? Depending on who reads the article, there may be different personal definitions about “large events”.

L 41, I would add the Ekstrom & Stark reference at the end of the sentence, because this is a nice example of continuous monitoring of large (global) areas.

L 76–80, this is where the adjustments of scope/pitch/motivation could be implemented (see my general comment above).

L 85, does this last sentence not make more sense if it were to occur further up, just after the sentence ending in the middle on line 83?

L 185, change “train” to “training”

L192, “does not allow” … “nor does it allow”, “nor” requires use of “neither” before. Consider rewriting.

L 357–359, can this guessed implication be tested? Otherwise, consider to explicitly mention that you cannot test this. Currently, the “might” is a bit odd and leaves us unsatisfied about what the sentence should imply.

L 392–396, move this part to the results and focus here on the actual discussion of these results, on explaining why the two-step approach performed less well.

L 400–408, this part appears out of context with what is discussed before and after, and especially with what the section title implies.

L 430, “break-off”, do you mean “detachment” (as process) or “detachment zone/site/area” (as spatial description)?

L 434, “bad weather conditions during snowfall periods”, please clarify what “bad” means. Can this be an artifact?

L 437, add “snow” before “avalanches”

L 476–480, I am not quite sure if I can follow this argumentation. According to l 331, a classifier trained on both sites increased the minority class TP (?) value by 10 %. Thus, you might want to relax the statement of “greatly improve” (l 479) a bit, especially when considering the small amount of underlying slope movement events.

Hide

ED: Publish subject to minor revisions (review by editor) (07 Dec 2020) by Yves Bühler

AR by Michaela Wenner on behalf of the Authors (08 Dec 2020) Author's response Manuscript

ED: Publish as is (08 Dec 2020) by Yves Bühler

AR by Michaela Wenner on behalf of the Authors (08 Dec 2020)

Short summary

Mass movements constitute a risk to property and human life. In this study we use machine learning to automatically detect and classify slope failure events using ground vibrations. We explore the influence of non-ideal though commonly encountered conditions: poor network coverage, small number of events, and low signal-to-noise ratios. Our approach enables us to detect the occurrence of rare events of high interest in a large data set of more than a million windowed seismic signals.