Rainfall-induced shallow landslides and soil wetness: comparison of physically based and probabilistic predictions

Leonarduzzi, Elena; McArdell, Brian W.; Molnar, Peter

doi:https://doi.org/10.5194/hess-25-5937-2021

Articles | Volume 25, issue 11

https://doi.org/10.5194/hess-25-5937-2021

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

https://doi.org/10.5194/hess-25-5937-2021

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

Articles | Volume 25, issue 11

Research article

|

15 Nov 2021

Research article |

| 15 Nov 2021

Rainfall-induced shallow landslides and soil wetness: comparison of physically based and probabilistic predictions

Elena Leonarduzzi, Brian W. McArdell, and Peter Molnar

Download

Final revised paper (published on 15 Nov 2021)
Preprint (discussion started on 08 Dec 2020)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'Comments and Suggestions', Anonymous Referee #1, 07 Feb 2021
- AC1: 'Reply on RC1', Elena Leonarduzzi, 12 Mar 2021
RC2: 'Reviewer report on manuscript “Rainfall-induced shallow landslides and soil wetness: comparison of physically-based and probabilistic predictions”, by Leonarduzzi et al.', Anonymous Referee #2, 08 Feb 2021
- AC2: 'Reply on RC2', Elena Leonarduzzi, 12 Mar 2021

Peer-review completion

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

ED: Reconsider after major revisions (further review by editor and referees) (22 Mar 2021) by Carlo De Michele

AR by Elena Leonarduzzi on behalf of the Authors (16 Jun 2021) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (19 Jun 2021) by Carlo De Michele

RR by Anonymous Referee #2 (12 Jul 2021)

Suggestions for revision or reasons for rejection

The authors have carefully considered my comments and the issues I raised during the first round of review. I think that this revised version of the manuscript is acceptable for publication in HESS.
I invite the authors to consider only the following considerations, that could lead to some minor revision of the Discussion and, maybe, to modify something also in the Conclusions.

The comparison between the landslide predictions carried out with the (coarse-gridded) physically based model, with those obtained with a conceptual hydrological model (with finer resolution), shows that the latter still outperforms the first one, although in this new version the Authors exploit in a better way the outcome of the physically based model and obtain some different indications, compared to the former analyses.
In the Discussion section, they ascribe the limits of the physically based analysis mainly to the coarse resolution (of both the model itself and of input data about geomorphological and geotechnical parameters).
I agree with this intepretation, but I would like the authors to consider also another point (and add some comment in this respect, if they see my point).
They state that the coarse-gridded physically based model cannot reproduce the (lateral) fluxes between cells, as they are too distant from each other (the model grid is 12.5x12.5 km), as the high number of events during which nothing happens to soil moisture seems to demonstrate. Differently, the (conceptual) hydrological model, providing soil moisture at 0.5x0.5 km resolution, gives more valuable information to predict landslides.
What I want to stress here is that in initially unsaturated (shallow) soil covers, the prevailing direction of water fluxes is close to the orthogonal to the ground surface (e.g. Lu et al., 2011), owing to the top (atmosphere) and bottom (whatever it is) boundary conditions, that make the orthogonal hydraulic gradient much much larger than the one parallel to the slope (because all the verticals share more or less the same hydraulic conditions, and so there is very little gradient along the slope (which is at least one order of magnitude longer than the thickness of the soil cover). Only when saturation is reached somewhere within the slope cover, then, in that saturated part, the orthogonal hydraulic gradient becomes of the same order of the one parallel to the slope (depending on slope and bedrock inclination), and so lateral fluxes become significant (leading to subsurface runoff generation). However, I guess that most of the conditionally unstable slopes would have already failed before saturation was reached. I don't expect this picture to change significantly if the model was run over a 0.5x0.5 km gird instead of the 12.5x12.5 grid.
Far from saturation the only way infiltrating water can be drained out of the slope cover is either evapotranspiration (a too slow process over the time scale of 1 to 6 days of rainfall events) or drainage through the soil-bedrock interface, which becomes the most delicate point of the physically based model, about which the authors do not give any information to the reader.
Although I did not look in the cited paper where the PREVAH model is described, I expect that water exchange mechanisms between the three storage modules are somehow introduced in such model, and if the model is somehow calibrated in order to provide reliable results, these mechanisms consider the water exchange between the two upper storage modules and the lower one. This could explain why the dynamics of soil saturation estimated by the PREVAH model better matches with the effects of single rainfall events.
So, at least this is my opinion, the physically based predictions are limited not only by resolution and/or accuracy of input data issues, but also by the unsuitability of thechosen model to correctly assess the effects of one of the (major) processes controlling the dynamics of soil moisture in the unsaturated zone, i.e. the leakage towards the underlying saturated zone.

N. Lu, B.S. Kaya, J.W. Godt (2011). Direction of unsaturated flow in a homogeneous and isotropic hillslope. Water Resources Research 47(2), https://doi.org/10.1029/2010WR010003

Hide

RR by Anonymous Referee #3 (16 Jul 2021)

Suggestions for revision or reasons for rejection

Review Hess-2020-624 : Rainfall-induced shallow landslides and soil wetness: comparison of physically based and probabilistic predictions
By: Elena Leonarduzzi, Brian W. McArdell, and Peter Molnar

This paper studies the prediction of landslide initiation by adding hydrological information to the well-known meteorological thresholds most frequently used and compares probabilistic and physically based modelling approaches for inclusion of antecedent soil wetness state (or water table) in the prediction. This review is on the revised paper. I have not been involved in the first round of reviews.
This paper is timely and contributes to an important research question in landslide research: can landslide forecasting on regional scale be improved by including hydrological information compared to traditional meteorological threshold based forecasting. And if so, how? The paper contains a huge amount of work and data. The methods are sound and clearly described. It studies the obvious question to what extend landslide forecasting can make use of existing regional hydrological model outputs. It tests this using two different regional models with different concepts and resolutions. The underlying work is in my opinion a creative and innovative contribution to this research field. The paper is well structured and written.
Overall, the study concludes that the contribution of the output of an existing large-scale hydrological model with coarse spatial resolution combined with infinite slope FoS calculation in the current state does not lead to an improved landslide forecasting. However, adding the hydrological information from a water balance model (with higher spatial resolution) modestly improve landslide forecasting in Switzerland. It shows that antecedent soil moisture information improves the landslide prediction using the probabilistic framework, but in a quite rudimentary way: by splitting the data sets in dry and wet antecedent conditions (just like proposed frequently before in literature, such as Figure 5.2 published in the Sidle-Ochiai, 2006).
Looking at the first round of reviews, I am of the opinion the authors did a profound job addressing the valuable comments and improved the paper significantly. The fact that this study uses existing models and combines those in landslide forecasting does not mean it is not novel, on the contrary.
I think the paper can be accepted for publication in Hess with some minor revisions.

Comments
- The authors sometimes use words or abbreviations of words in formula’s and in figure caption and axis titles. I personally find those difficult to read and suggest to use 1 symbol for a variable or parameter and subscripts (e.g. x-axis of Fig 4, 5).
- L256: does a FoS needs to be ‘exceptionally’ low? I guess just a certain degree lower could be sufficient? And Figure 4 shows that a bit lower starting point is sufficient.
- L278: Would it not be easier for the reader if you use one dominant terminology: False alarm, misses, TP, TN (and later on T, above/below, NT_above/below). So use TP, TN, FP, FN (and add false alarm or missed alarms in parenthesis to it a few times)? Same for Figure 6.
- L298: layout exponent, not in superscript
- L300: Figure 8: Honestly, is the TSS really that much different for the different graphs? I would argue you can use all these graphs to split the data set. The entire range of the TSS as function of antecedent wetness is 0.1…. that is not a lot, is it?
- L323: vegetation cohesion is not a correct terminology. It is soil cohesion and apparent root cohesion: combined used in the infinite slope model.
- L324: for shallow slip surfaces (<2m) cohesion is a very sensitive parameter, by definition. Especially if apparent root cohesion is added.
- L346: I challenge the authors with this statement. Why would a physically based model per definition (or theoretically) be superior? The so-called physical laws do not apply to the scale of the model but have another Representative Elemental Volume. Equally, could we not argue that on the regional scale we need another set of physical laws. So why can we not say that a physically conceptual model will be theoretically superior to describe the hydrological system on regional scale better than a bottom-up physically based model applied to the ‘wrong’ scale? At least you should address the scale issue related to the used ‘physics’. You could elaborate a bit more on this.
- L360: Maybe the authors can reflect a little why using regional hydrological models have only modest improvement to landslide forecasts and how that compares to hydrometeorological thresholds for landslides using in-situ hydrological (soil moisture) measurements (in Switzerland).
- L367: I am a bit surprised to read the first conclusion. The first conclusion is about the effect of cohesion (in your case soil cohesion and apparent root cohesion) or not. I do not consider this the first and most important conclusion. As mentioned above as well, for shallow landslides cohesion can be the dominant resisting force, so putting that to zero is not surprisingly changing the areas where landslides can take place. I suggest to make this not the first conclusion but rather the 3rd or so.
- L376: This conclusion is not fully justified, or at least not telling the complete story. Another way of framing this conclusion could be stressing that in your analysis 65% of the country is US (35% when C=0 kpa) That leaves 35% (or 65%) to be potentially ‘triggerable’. This is how I read this result.

Hide

RR by Anonymous Referee #1 (25 Jul 2021)

ED: Publish subject to minor revisions (review by editor) (02 Sep 2021) by Carlo De Michele

AR by Elena Leonarduzzi on behalf of the Authors (24 Sep 2021) Author's response Author's tracked changes Manuscript

ED: Publish as is (04 Oct 2021) by Carlo De Michele

Short summary

Landslides are a dangerous natural hazard affecting alpine regions, calling for effective warning systems. Here we consider different approaches for the prediction of rainfall-induced shallow landslides at the regional scale, based on open-access datasets and operational hydrological forecasting systems. We find antecedent wetness useful to improve upon the classical rainfall thresholds and the resolution of the hydrological model used for its estimate to be a critical aspect.