the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Predicting the thickness of shallow landslides in Switzerland using machine learning
Abstract. Landslide thickness is a key parameter in various types of models used to simulate landslide susceptibility. In this study, we developed a model providing improved predictions of potential shallow landslide thickness in Switzerland. We tested three machine learning models based on random forests, generalized additive model, and linear regression and compared the results to three existing models that link soil thickness to slope and elevation. The models were calibrated using data from two field inventories in Switzerland ("HMDB" with 709 records and "KtBE" with 517 records). We explored 37 different covariates including metrics on terrain, geomorphology, vegetation height, and lithology at three different cell sizes. To train the machine learning models, 21 variables were chosen based on the variable importance derived from random forest models and expert judgement. Our results show that the machine learning models consistently outperformed the existing models by reducing the mean absolute error by at least 17 %. The random forests models produced a mean absolute error of 0.25 m for the HMDB and 0.19 m for the KtBE data. Models based on machine learning substantially improve the prediction of landslide thickness, offering refined input for enhancing the performance of slope stability simulations.
- Preprint
(9104 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 03 Jul 2024)
-
RC1: 'Comment on nhess-2024-76', Anonymous Referee #1, 23 Jun 2024
reply
Dear Authors,, I apologize for the late upload of my review report. The good news is that my review will be very positive: I liked the manuscript and I think it addresses an important topic, well inside the aims and scopes of NHESS. It has an ambitious goal (soil thickness prediction at national scale is very challenging and, to the best of my knowledge, never attempted before in these terms), which is pursued with a robust and original approach. The manuscript is well structured, well written and very clear. Overall, the manuscript surely deserves publication and wit will be a valuable contribution to the journal. I just have a few comments. None of them is a big issue, and the manuscript can be accepted after minor revisions.
GENERAL COMMENTS
1 - The thing I miss the most in your study is a better linkage with the geology.
1a- I would add a test site description section, to make clear that you work at national scale (an amazing feature of your work) and briefly describing the main features of Switzerland (geology, climate).
1b- Geology is important in influencing both landsliding and soil thickness distribution. You use geology/lithology in your study by means of bedrock density, which you consider a proxy for bedrock lithology. In my opinion, this strategy needs a better justification. I guess you did this to have a numerical variable instead of a categorical variable (e.g. lithology classes). However, rock/soil cohesion, internal friction angle or hydraulic conductivity (just to name a few) may sometimes have a better relation with soil properties or slope stability. Some more reasoning on your strategy would be welcome.
1c- Building on previous comment, I recommend adding a table in which you list the main lithologies and the density value you assigned to them (or viceversa, as it suits you better). This may be also linked to a figure with main lithologies (see comment about test site description). Does Swisstopo contain thematic layers about other geotechnical properties? If yes, why didn't you consider them as well? When accounting for morphology, you took into account many parameters, so geology could be also accounted for by different parameters.
1d- In section 6.1 you analysed landslide distribution by slope class. However, the first thing that came up in my mind was to evaluate their distribution across lithologies. Is it possible to quickly perform a similar analysis?
SPECIFIC COMMENTS
L2: I would mention the areal extension of Switzerland. I think the width of your test site is a big constrain to the work; as such, it should be emphasized as an additional point of strength (you may consider to stress it also in the discussion and conclusion).
L13: Would you consider adding also https://doi.org/10.1016/j.jeem.2024.102942 ? I think it is pertinent, as it reports on indirect impacts of hydrogeological processes, which are rarely accounted for (most studies focus only on casualties or direct economic damages).
L7 - I suggest to also convert CHF amount to USD. A few European readers may be aware of the value of CHF, but maybe not all the international readers are familiar with this currency.
L21 - My English is not better than yours, but isn't "carried out within" more appropriate than "carried out at within"?
L25 - I found odd to read the landslide definition in the middle of the introduction... Isn't it better to move it earlier in the text?
L51 and L55 are mentioning three and two landslide inventories, respectively. At this stage, this is confusing.
Section 2.1 - Just a comment: another point clearly highlighting the importance of soil thickness is the math formulae used in slope stability models. As scientific knowledge advances, the complexity of models has always been increasing, integrating new parameters and new processes in the stability equation. However, soil thickness has always been there: since the first pioneering equations (e.g. Skempton, A. W., & DeLory, F. A. (1957). Stability of natural slopes in London clay. Thomas Telford Publishing, London, UK, 15, 378-381.), soil thickness has always been there, among the uncontested key parameters!
Tab1 - GIST-RF was recently applied to another case of study. If you want, you can add https://doi.org/10.1016/j.catena.2024.108024 along with Xiao's work.
Section 3.1 - I suggest to be clearer on one point: the geometry used to map landslides in the inventory. Are they mapped as polygons or points? In case they are points, it would make a big difference if the mapped point is the triggering point or the impact point, especially in case of shallow landslides with large runout distances.
L133 - Geological bedrock instead of geological underground?
L148-153 - see comment 1c.
L212 (and elsewhere) - Shouldn't it be CARET, with capital letters? If yes, please adjust all occurrences in the text.
L311 I was surprised in seeing errors higher than the maximum expected value (2.5m) Didn't you applied a upper constrain to the modeled thickness values?
L330 I would make reference to Fig B1
L376-379- Could it be that by adding many 0 values in calibration the overall average modeled thickness is lowered?
Citation: https://doi.org/10.5194/nhess-2024-76-RC1 -
RC2: 'Comment on nhess-2024-76', Anonymous Referee #2, 24 Jun 2024
reply
Model code and software
HAFL-WWI/Landslide_Thickness_Prediction: Release for Predicting shallow landslide thickness using ML v0.1.1 Christoph Schaller https://doi.org/10.5281/zenodo.11032083
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
281 | 84 | 14 | 379 | 10 | 11 |
- HTML: 281
- PDF: 84
- XML: 14
- Total: 379
- BibTeX: 10
- EndNote: 11
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1