A Novel Hybrid Method for Landslide Susceptibility Mapping-Based GeoDetector and Machine Learning Cluster: A Case of Xiaojin County, China

Round 1

Reviewer 1 Report

The paper presents an automated procedure to analyze and process territorial data by selecting conditional factors, measuring and removing redundant ones, to perform reliable landslide susceptibility maps.

The algorythm is based on a cluster of four previsional models from statistics or AI, which act in parallel and led to a choise; the result comes from the method wich offers the best performance measured by classical methods.

The method allows a number of interesting features: evaluation of effectiveness of each selected conditional factor; evaluation of interactive factors; choice of a larger platform to best minimize errors between calculated and nodes properties.

The declared mission is also to abilitate not specialists to perform LSM, first step to knowledge and mitigation of risk of landslides.

The model has been applied to a very large area, about 5.500 sq.Km, with a grid of 60 m span, and 616 detected point of failure, by a remote sensing analysis.

In the presented form, the model works with conditional factors detected from automated procedure, rarely describing the physical conditions of the involved soils. Parameters like soil strenght, plasticity, saturation does not explicity appear in the conditional factors.

In the analysis of the weight of conditioning factors, the scarce relevance of the slope factor is highlighted. The circumstance is paradoxical, except in a context largely populated by rock lithologies.

Nevertheless results are encouraging and the best clusterized method reveals the well known logistic regression together with VSM.

Reading of the paperi is not easy, due the large and repeated number of acronyms. Could be better to repeat somewhere the acronym in the explicit form.

Preceding consideration allow us the evaluate the proposed method effective with poor expert operators and for large areas, with poor detailed local knowledges about physical land characterization.

The list of references contains a number of errors: lowercase instead of uppercase, omitted authors, etc. Please review

Line 17

Landslide susceptibility mapping (LSM) is an effective way to reduce landslide disasters and losses.

Better something like:

Lsm may be an effective way to reduce vulnerability from landslide disasters.

Line 35

The statement is valid also without the sentence: “With global warming and climate anomalies…”

Line 38

“A” must be in lower case

Line 46

physically-based and statistically-based methods

Statistics can use physically based parameters. Physically based methods should be modified in “ deterministic methods”.

Line 48

With the development of Geographic Information Science (GIS)

GIS is the acronym of “Geographic Information System”

Line 59

by many factors, and 2) new landslides are more likely to occur where landslides have occurred

not fully true. Also first failure landslides do occur. So i suggest to modify something like: …2) new landslides are more likely to occur where landslides have occurred or in similar conditions….”

Line 79 -81

Improve grammatical form (the verb is missing)

Line 136

Including..  the verb is missing. Probably: includes

Line 148

…Water infiltration is a necessary condition for landslides, and…

The sentence must be improved: if the infiltration turns into capillarity, for example, overall stability conditions get better.

Line 153

….the TWI was calculated by SAGA software.

Some reference should be added

Line 173

…with a resolution of 60 m (twice the resolution of DEM). Consequently, 1,709,680 mapping units are obtained. This makes the grid and the factors have a good correspondence with the unit of conditional factors.

The statement needs a better explanation. The width of the cell can allow significative variations inside of the conditional parameters

Line 341

The high areas were concentrated in urban

Better: the high susceptibility areas ….

Line 362

…….the location of roads and residential 362 areas, human

activities in this area have a strong effect on landslides.

Answer to this question: the strong effect on landslides comes from the proximity to the roads, or from the proximity to poorly constructed roads?

The paper presents an automated procedure to analyze and process territorial data by selecting conditional factors, measuring and removing redundant ones, to perform reliable landslide susceptibility maps.

The algorythm is based on a cluster of four previsional models from statistics or AI, which act in parallel and led to a choise; the result comes from the method wich offers the best performance measured by classical methods.

The method allows a number of interesting features: evaluation of effectiveness of each selected conditional factor; evaluation of interactive factors; choice of a larger platform to best minimize errors between calculated and nodes properties.

The declared mission is also to abilitate not specialists to perform LSM, first step to knowledge and mitigation of risk of landslides.

The model has been applied to a very large area, about 5.500 sq.Km, with a grid of 60 m span, and 616 detected point of failure, by a remote sensing analysis.

In the presented form, the model works with conditional factors detected from automated procedure, rarely describing the physical conditions of the involved soils. Parameters like soil strenght, plasticity, saturation does not explicity appear in the conditional factors.

In the analysis of the weight of conditioning factors, the scarce relevance of the slope factor is highlighted. The circumstance is paradoxical, except in a context largely populated by rock lithologies.

Nevertheless results are encouraging and the best clusterized method reveals the well known logistic regression together with VSM.

Reading of the paperi is not easy, due the large and repeated number of acronyms. Could be better to repeat somewhere the acronym in the explicit form.

Preceding consideration allow us the evaluate the proposed method effective with poor expert operators and for large areas, with poor detailed local knowledges about physical land characterization.

The list of references contains a number of errors: lowercase instead of uppercase, omitted authors, etc. Please review

Line 17

Landslide susceptibility mapping (LSM) is an effective way to reduce landslide disasters and losses.

Better something like:

Lsm may be an effective way to reduce vulnerability from landslide disasters.

Line 35

The statement is valid also without the sentence: “With global warming and climate anomalies…”

Line 38

“A” must be in lower case

Line 46

physically-based and statistically-based methods

Statistics can use physically based parameters. Physically based methods should be modified in “ deterministic methods”.

Line 48

With the development of Geographic Information Science (GIS)

GIS is the acronym of “Geographic Information System”

Line 59

by many factors, and 2) new landslides are more likely to occur where landslides have occurred

not fully true. Also first failure landslides do occur. So i suggest to modify something like: …2) new landslides are more likely to occur where landslides have occurred or in similar conditions….”

Line 79 -81

Improve grammatical form (the verb is missing)

Line 136

Including..  the verb is missing. Probably: includes

Line 148

…Water infiltration is a necessary condition for landslides, and…

The sentence must be improved: if the infiltration turns into capillarity, for example, overall stability conditions get better.

Line 153

….the TWI was calculated by SAGA software.

Some reference should be added

Line 173

…with a resolution of 60 m (twice the resolution of DEM). Consequently, 1,709,680 mapping units are obtained. This makes the grid and the factors have a good correspondence with the unit of conditional factors.

The statement needs a better explanation. The width of the cell can allow significative variations inside of the conditional parameters

Line 341

The high areas were concentrated in urban

Better: the high susceptibility areas ….

Line 362

…….the location of roads and residential 362 areas, human

activities in this area have a strong effect on landslides.

Answer to this question: the strong effect on landslides comes from the proximity to the roads, or from the proximity to poorly constructed roads?

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Authors present a design of a machine learning cluster including ANN, BN, LR, and SVM to obtain the optimal landslide susceptibility map automatically. Furthermore, they show a physically meaningful factor selection method by defining effective redundant factors to make the critical process of landslide conditional factors selection more reasonable.  

Overall presentation of the paper is good and I find it interesting for scientific community and stakeholders involved in decision-making related to land planning and landslides risk management.

I consider the manuscript with title "A novel hybrid method for landslide susceptibility mapping based GeoDetector and machine learning cluster: a case of Xiaojin County, China" should be accepted to be publish in the ISPRS International Journal of Geo-Information  

Here I include some comments and suggestions to improve this manuscript:·        

• In the introduction section, the inclusion of other references about Geodetector-approach for selecting factors and interesting reviews (such as Guzzetti et al. 1999) and Brenning, 2005) or comparison among the most common methodologies (similar to Nefeslioglu et al. (2008), Ermini et al  2005, and Baeza et al. 2010), would enrich the article significantly.
• Line 155 add reference/s about HAILS.
• Line 189, the description and/or information about Geodetector  “The GeoDetector software is freely available from  http://www.geodetector.cn/” should be appeared before in the manuscript (for example to line 168).
• Line 171 “the study area was divided into regular grids with a resolution of 60 m (twice the resolution of DEM). “ each regular grid is line a tile and is considered a unit map? please, clarify this sentence and terms (regular grid and mapping units), since it can be one of the important limitations of the method, as indicated in section 4.3· And, in order to improve and clarify the text, I suggest you to add “spatial” to the resolution term or change it by “cellsize”, because there are different kind of resolutions: spectral, temporal and spatial resolution.
• Line 230. Please, provide additional information in order to correctly justify this assertion about MLP model against the RBF?.
• Line 285 to 288. Check this text, I think it must be removed, because it is the caption of fig.6. Similar case between lines 337-338.·        
• Figure 1 b) change caption in order to be more clear (townships and roads on to a remote sensing image.
• Figure 2 l) change caption “settlement” by a better/correct description of the figure, for example ”area  of influence of settlements”

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

The article A novel hybrid method for landslide susceptibility mapping based GeoDetector and machine learning cluster: a case of Xiaojin County, China proposes a new method in assessing LSM in an area in China.

The overall assessment of the paper is good, not earth-shattering. The English of the manuscript needs to be checked by a native English speaker. At this stage, I would recommend Major Revisions; because there are some issues that need to be corrected and are highlighted below:

Introduction section – this section needs to be documented in a much detailed manner. You should give the paper an international context, and highlight the main discoveries in LSM until the present, in a very succinct manner. I could recommend some, but not limited to:

https://doi.org/10.1016/j.enggeo.2020.105776

https://doi.org/10.1016/j.scitotenv.2020.137231

https://doi.org/10.3390/su11205659

https://doi.org/10.1007/s11629-019-5702-6

https://doi.org/10.3390/sym12121954

L37: here you should continue the phrase, “every year….” Since what? It seems that the phrase needs an end

L73: Figure 1: as far as I can see, this figure has 3 parts, and none of them is identified in the main text. Please, correct

L83: maybe you could offer some more data about these two rivers. It will help the reader in better understanding the influence on the LSM. Some discharge data, length, the size of the catchment, etc.

L93-94: include some suggestive photos of the mentioned landslide types from the study area

L100: you can name the fifth cluster “Anthropogenic”

L154: you should better explain how you calculated the HAILS factor

L167: use “validation” instead of “verification”

L208: use the same Line Spacing for the entire manuscript

L328: replace “worst performance” with some milder words

Section 3.2. Please, include the SCAI method as an additional validation method. This will offer your study and your proposed method more confidence in the results presented

L336: why not 4 susceptibility classes?

Compare and Discuss your results with the following papers:

https://doi.org/10.1007/s10346-019-01286-5

https://doi.org/10.3390/su11226323

The References are not formatted according to the instructions for authors of ISPRS Int. J. Geo-Inf.; the use of DOI is highly indicated. Therefore, adhere to the Instructions for Authors for References as well

Good luck with the review!

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Dear Authors,

You did not address all my comments accordingly. SCAI method is very well known in the literature. You can find it within this paper: "GIS-based evaluation of diagnostic areas in landslide susceptibility analysis of Bahluieț River Basin (Moldavian Plateau, NE Romania). Are Neolithic sites in danger? Geomorphology 2018.

You almost did nothing for the Introduction Section (not to say that you did not add all the references). And the added references need to be discussed in the context of your present paper.

Also, in the Discussion section, you did not address my previous comment either. You did not discuss and/or compare your results with other studies/references that I have already recommended. You should also improve your Conclusions section.

Kind regards.

Author Response

please see the attachment

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.