Assessing the importance of feature selection in Landslide Susceptibility for Belluno province (Veneto Region, NE Italy)

Meena, Sansar Raj; Puliero, Silvia; Bhuyan, Kushanav; Floris, Mario; Catani, Filippo

doi:https://doi.org/10.5194/nhess-2021-299

Preprints

https://doi.org/10.5194/nhess-2021-299

Preprints

18 Nov 2021

Assessing the importance of feature selection in Landslide Susceptibility for Belluno province (Veneto Region, NE Italy)

Sansar Raj Meena^1,2, Silvia Puliero¹, Kushanav Bhuyan^1,2, Mario Floris¹, and Filippo Catani¹ Sansar Raj Meena et al.

Show author details

¹Department of Geosciences, University of Padova, Padova, Italy
²Faculty of Geo-Information Science and Earth Observation (ITC), University of Twente, Enschede, Netherlands

Received: 11 Oct 2021 – Discussion started: 18 Nov 2021

Abstract. In the domain of landslide risk science, landslide susceptibility mapping (LSM) is very important as it helps spatially identify potential landslide-prone regions. This study used a statistical ensemble model (Frequency Ratio and Evidence Belief Function) and two machine learning (ML) models (Random Forest and XG-Boost) for LSM in the Belluno province (Veneto Region, NE Italy). The study investigated the importance of the conditioning factors in predicting landslide occurrences using the mentioned models. In this paper, we evaluated the importance of the conditioning factors (features) in the overall prediction capabilities of the statistical and ML algorithms. By the trial-and-error method, we eliminated the least "important" features by using a common threshold. Conclusively, we found that removing the least "important" features does not impact the overall accuracy of the LSM for all three models. Based on the results of our study, the most commonly available features, for example, the topographic features, contributes to comparable results after removing the least "important" ones. This confirms that the requirement for the important factor maps can be assessed based on the physiography of the region. Based on the analysis of the three models, it was observed that most commonly available feature data can be useful for carrying out LSM at regional scale, eliminating the least available ones in most of the use cases due to data scarcity. Identifying LSMs at regional scale has implications for understanding landslide phenomena in the region and post-event relief measures, planning disaster risk reduction, mitigation, and evaluating potentially affected areas.

Download & links

Preprint (PDF, 4002 KB)

Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Preprint (4002 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (4002 KB)
BibTeX
EndNote

Journal article(s) based on this preprint

Sansar Raj Meena et al.

Interactive discussion

Status: closed

RC1:
'Comment on nhess-2021-299', Anonymous Referee #1, 01 Dec 2021

I have revised the manuscript "Assessing the importance of feature selection in Landslide Susceptibility for Belluno province (Veneto Region, NE Italy)", submitted by Sansar Raj Meena, Silvia Puliero, Kushanav Bhuyan, Mario Floris, Filippo Catani, focused on the the importance of feature selection for landslide susceptibility zonation. The manuscript could be interesting for the journal but requires a strong revision. Data are not clearly presented, some definition are confusing, the structure needs to be reorganized. Moreover, the manuscript is not written in a good English and some sentences are difficult to understand. I recommend the authors to submit a revised version of the manuscript after a revision by an English-speaking person. Detailed comments are throughout the text.

Citation: https://doi.org/10.5194/nhess-2021-299-RC1
- AC2: 'Reply on RC1', Sansar Raj Meena, 19 Jan 2022
  
  Thank you for reviewing the manuscript, our replies to reviewers comments are in the supplement file.
  
  Citation: https://doi.org/10.5194/nhess-2021-299-AC2
RC2:
'Comment on nhess-2021-299', Anonymous Referee #2, 07 Dec 2021

I have gone through the manuscript and found that the quality of work is very good and applied. I have some observation needs to be correct before its goes to final publication.

1. The words features and factors are used interchangeably in the paper. Better to stick to one word.

2. Definition between pre and post predictions are not clear. Making it difficult to read and understand the context sometimes.

3. An image of the affected area for example could be very insightful to comment on the extent of the damage over the area.

4. Scale of the maps that are taken into the experimentation are missing.

5. Explanation of the methodology can be better, especially the starting paragraph.

6. The mapping units are not defined as of yet, which must be mentioned.

7. Better explanation of the models is required, mostly in the case of LSM and how these models learn and predict using the models for LSM.

8. The reasoning for 0.3 as the threshold must be reasoned better.

9. Graph axes have no labels.

10. No definition of training and testing datasets for model prediction. Need a section for that.

Citation: https://doi.org/10.5194/nhess-2021-299-RC2
- AC1:
  'Reply on RC2', Sansar Raj Meena, 19 Jan 2022
  Comment on nhess-2021-299
  Anonymous Referee #2
  I have gone through the manuscript and found that the quality of work is very good and applied. I have some observation needs to be correct before its goes to final publication.
  The words features and factors are used interchangeably in the paper. Better to stick to one word.
  
  Ans: Thank you for your suggestions. We use the words “conditioning factors” throughout the manuscript to avoid further confusion.
  
  2. Definition between pre and post predictions are not clear. Making it difficult to read and understand the context sometimes.
  
  Ans: The “pre-predictions” refer to the moment before actually training the model and thus, we refer to some literature that performs factor importance prior to model training. Similarly, “post-predictions” refer to factor importance after model training, which we perform in our study here. Nonetheless, we have edited these two words in the document to avoid confusion.
  
  3. An image of the affected area for example could be very insightful to comment on the extent of the damage over the area.
  
  Ans: Thank you for your comment. We have added some images in figure 1.
  4. Scale of the maps that are taken into the experimentation are missing.
  Ans: We have added the scale information in the manuscript.
  
  5. Explanation of the methodology can be better, especially the starting paragraph.
  
  Ans: We have re-arranged the paragraphs along with the conceptual framework diagram to make a more comprehensive and suitable readability experience.
  
  6. The mapping units are not defined as of yet, which must be mentioned.
  
  Ans: We have added this information in the manuscript. We have done the analysis at pixel level for our study area.
  
  7. Better explanation of the models is required, mostly in the case of LSM and how these models learn and predict using the models for LSM.
  
  Ans: We have explained the models better keeping in mind the usage of them in the context of LSM. Please refer to sections 3.2.1 and 3.2.2.
  
  8. The reasoning for 0.3 as the threshold must be reasoned better.
  
  Ans: We have given the reasoning of this in line 398-400. But to recall, we tried countless values as a cut-off or threshold value to see which of the conditioning factors gave the best accuracy for the susceptibility after removal of the factors based on the cut-off value.
  
  9. Graph axes have no labels.
  
  Ans: We have added the y-axes labels in the graphs. We refer you to figures 6 and 10.
  
  10. No definition of training and testing datasets for model prediction. Need a section for that.
  
  Ans: We have added the definition of training and testing datasets for model prediction in section 2.2.
  
  Citation: https://doi.org/10.5194/nhess-2021-299-AC1

Peer review completion

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

ED: Reconsider after major revisions (further review by editor and referees) (24 Jan 2022) by Daniele Giordan

AR by Sansar Raj Meena on behalf of the Authors (26 Jan 2022) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (09 Feb 2022) by Daniele Giordan

RR by Anonymous Referee #3 (21 Feb 2022)

Suggestions for revision or reasons for rejection

Dear authors, I have read and carefully aevaluated your manuscript "Assessing the importance of feature selection in Landslide Susceptibility for Belluno province (Veneto Region, NE Italy)". I found that the idea of the work is original and the outcomes are interesting, thus I recommend publication after a round of minor revisions. Please consider my comments hereafter.

GENERAL COMMENTS
- English is understandable but a further revision from the most experienced coauthors would be useful to improve it.
- The idea behind this work is interesting and original. The analysis performed is incomplete. The comparison between different models and different parameters’ configurations is performed only in terms of performance metrics and visual inspections. In my opinion this is not enough to characterize the differences between the maps, a more thorough comparison is needed. I strongly recommend following the approach of Xiao et al. 2020, which is quite simple and could be done in a few minutes, adding a nice part in the discussion: the authors could make a comparison between the “full” and “reduced” version of each model by making the arithmetical difference of the susceptibility values calculated by the 14-factors-version of the model and the 9-factors-version of the same model. The same comparison could be repeated for the three models. The calculated difference would be useful to check if the differences are randomly distributed or if they consist in systematic errors driven by the spatial distribution of some variable (maybe one of the discarded ones? That would be even more interesting). That would add a nice value to the discussion and interpretation of the differences encountered.
- It is not clear how you actually implement the models. Do you use Matlab, R, or other software?

SPECIFIC COMMENTS
L21= which threshold? This is not clear
L22-24= I suggest being more precise and accurate. I would name clearly which are the variables discarded and the most important ones.
L26-30= I suggest deleting. The first sentence is a repetition. The second one is not fundamental.
L34= A reference is needed here about impacts of landslides, e.g. Froude et al., 2018
L47-48= what do you mean with “the significance of landslide studies”? I suggest rephrasing.
L51= reference needed (e.g. Reichenbach et al., 2018)
L62= data driven is a broader “family” of approaches, it does not apply to FR alone.
Section 2.1= I would move the figure of the study area here.
L165 (and elsewhere)= please use a more specific term instead of “precipitation”: which variable are you using? Mean annual precipitation? Mean monthly rainfall?
Table 1: I have some concerns about some factors. Elevation: I think at this latitude it is important because it largely influences climate, including the amount, intensity and distribution of rainfall. Slope: I would use the range 0°-86°. Aspect: motivation is quite confused. TWI: please change “measures” with “estimates”. TRI: please, rephrase the last part of the description (fluctuation, moves): it seems that the topographic surface moves. 7 and 8: please add the values of the ranges. 10: which feature of the rainfall regime are you considering? Mean monthly rainfall? 11: I don’t know the study cited here and I couldn’t find it in the reference list; I suggest to cite a paper specifically focused on the importance of lithology in LSM (e.g. Segoni et al., 2020). 11 and 13: I would include all the classes in the third column and not “etc.”.
L306 and 310: here I understand that you use natural breaks (l306) […] which is better than natural breaks (l310). Please, revise.
Fig 6 and 10: it would be nice to mark somehow the discarded factors. E.g. by coloring them in a different way or by adding a horizontal red line at 0.3 to show the cut-off value.
L402: 0.910
L408: after this discussion, it would be interesting to show the comparison I suggested in my general comment. You can show the three maps FREBF14 - FREBF9, WGB14-XG9 and RF14-RF9. And briefly describe the spatial patterns of the differences shown: if you remove 5 factors, are the differences random or do they follow a systematic pattern?
L460: instead of making reference to two figures that are not here, you could provide another figure, or a table, listing the factors ranked by importance. It would be meaningful to observe if the ranks are similar and if the discarded variables are similar.

REFERENCES
Froude, M. J., & Petley, D. N. (2018). Global fatal landslide occurrence from 2004 to 2016. Natural Hazards and Earth System Sciences, 18(8), 2161-2181.
Reichenbach, P., Rossi, M., Malamud, B. D., Mihir, M., & Guzzetti, F. (2018). A review of statistically-based landslide susceptibility models. Earth-science reviews, 180, 60-91.
Segoni, S., Pappafico, G., Luti, T., & Catani, F. (2020). Landslide susceptibility assessment in complex geological settings: Sensitivity to geological information and insights on its parameterization. Landslides, 17(10), 2443-2453.
Xiao, T., Segoni, S., Chen, L., Yin, K., & Casagli, N. (2020). A step beyond landslide susceptibility maps: a simple method to investigate and explain the different outcomes obtained by different approaches. Landslides, 17(3), 627-640.

Hide

RR by Anonymous Referee #2 (22 Feb 2022)

ED: Publish subject to minor revisions (review by editor) (03 Mar 2022) by Daniele Giordan

AR by Sansar Raj Meena on behalf of the Authors (11 Mar 2022) Author's response Author's tracked changes

ED: Publish as is (21 Mar 2022) by Daniele Giordan

Interactive discussion

Status: closed

RC1:
'Comment on nhess-2021-299', Anonymous Referee #1, 01 Dec 2021

I have revised the manuscript "Assessing the importance of feature selection in Landslide Susceptibility for Belluno province (Veneto Region, NE Italy)", submitted by Sansar Raj Meena, Silvia Puliero, Kushanav Bhuyan, Mario Floris, Filippo Catani, focused on the the importance of feature selection for landslide susceptibility zonation. The manuscript could be interesting for the journal but requires a strong revision. Data are not clearly presented, some definition are confusing, the structure needs to be reorganized. Moreover, the manuscript is not written in a good English and some sentences are difficult to understand. I recommend the authors to submit a revised version of the manuscript after a revision by an English-speaking person. Detailed comments are throughout the text.

Citation: https://doi.org/10.5194/nhess-2021-299-RC1
- AC2: 'Reply on RC1', Sansar Raj Meena, 19 Jan 2022
  
  Thank you for reviewing the manuscript, our replies to reviewers comments are in the supplement file.
  
  Citation: https://doi.org/10.5194/nhess-2021-299-AC2
RC2:
'Comment on nhess-2021-299', Anonymous Referee #2, 07 Dec 2021

I have gone through the manuscript and found that the quality of work is very good and applied. I have some observation needs to be correct before its goes to final publication.

1. The words features and factors are used interchangeably in the paper. Better to stick to one word.

2. Definition between pre and post predictions are not clear. Making it difficult to read and understand the context sometimes.

3. An image of the affected area for example could be very insightful to comment on the extent of the damage over the area.

4. Scale of the maps that are taken into the experimentation are missing.

5. Explanation of the methodology can be better, especially the starting paragraph.

6. The mapping units are not defined as of yet, which must be mentioned.

7. Better explanation of the models is required, mostly in the case of LSM and how these models learn and predict using the models for LSM.

8. The reasoning for 0.3 as the threshold must be reasoned better.

9. Graph axes have no labels.

10. No definition of training and testing datasets for model prediction. Need a section for that.

Citation: https://doi.org/10.5194/nhess-2021-299-RC2
- AC1:
  'Reply on RC2', Sansar Raj Meena, 19 Jan 2022
  Comment on nhess-2021-299
  Anonymous Referee #2
  I have gone through the manuscript and found that the quality of work is very good and applied. I have some observation needs to be correct before its goes to final publication.
  The words features and factors are used interchangeably in the paper. Better to stick to one word.
  
  Ans: Thank you for your suggestions. We use the words “conditioning factors” throughout the manuscript to avoid further confusion.
  
  2. Definition between pre and post predictions are not clear. Making it difficult to read and understand the context sometimes.
  
  Ans: The “pre-predictions” refer to the moment before actually training the model and thus, we refer to some literature that performs factor importance prior to model training. Similarly, “post-predictions” refer to factor importance after model training, which we perform in our study here. Nonetheless, we have edited these two words in the document to avoid confusion.
  
  3. An image of the affected area for example could be very insightful to comment on the extent of the damage over the area.
  
  Ans: Thank you for your comment. We have added some images in figure 1.
  4. Scale of the maps that are taken into the experimentation are missing.
  Ans: We have added the scale information in the manuscript.
  
  5. Explanation of the methodology can be better, especially the starting paragraph.
  
  Ans: We have re-arranged the paragraphs along with the conceptual framework diagram to make a more comprehensive and suitable readability experience.
  
  6. The mapping units are not defined as of yet, which must be mentioned.
  
  Ans: We have added this information in the manuscript. We have done the analysis at pixel level for our study area.
  
  7. Better explanation of the models is required, mostly in the case of LSM and how these models learn and predict using the models for LSM.
  
  Ans: We have explained the models better keeping in mind the usage of them in the context of LSM. Please refer to sections 3.2.1 and 3.2.2.
  
  8. The reasoning for 0.3 as the threshold must be reasoned better.
  
  Ans: We have given the reasoning of this in line 398-400. But to recall, we tried countless values as a cut-off or threshold value to see which of the conditioning factors gave the best accuracy for the susceptibility after removal of the factors based on the cut-off value.
  
  9. Graph axes have no labels.
  
  Ans: We have added the y-axes labels in the graphs. We refer you to figures 6 and 10.
  
  10. No definition of training and testing datasets for model prediction. Need a section for that.
  
  Ans: We have added the definition of training and testing datasets for model prediction in section 2.2.
  
  Citation: https://doi.org/10.5194/nhess-2021-299-AC1

Peer review completion

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

ED: Reconsider after major revisions (further review by editor and referees) (24 Jan 2022) by Daniele Giordan

AR by Sansar Raj Meena on behalf of the Authors (26 Jan 2022) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (09 Feb 2022) by Daniele Giordan

RR by Anonymous Referee #3 (21 Feb 2022)

Suggestions for revision or reasons for rejection

Hide

RR by Anonymous Referee #2 (22 Feb 2022)

ED: Publish subject to minor revisions (review by editor) (03 Mar 2022) by Daniele Giordan

AR by Sansar Raj Meena on behalf of the Authors (11 Mar 2022) Author's response Author's tracked changes

ED: Publish as is (21 Mar 2022) by Daniele Giordan

Journal article(s) based on this preprint

Sansar Raj Meena et al.

Viewed

Total article views: 613 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
412	184	17	613	12	9

HTML: 412
PDF: 184
XML: 17
Total: 613
BibTeX: 12
EndNote: 9

Views and downloads (calculated since 18 Nov 2021)

Month	HTML	PDF	XML	Total
Nov 2021	162	45	2	209
Dec 2021	101	33	2	136
Jan 2022	64	24	5	93
Feb 2022	23	8	0	31
Mar 2022	27	12	2	41
Apr 2022	17	16	4	37
May 2022	2	8	1	11
Jun 2022	3	5	1	9
Jul 2022	1	4	0	5
Aug 2022	2	3	0	5
Sep 2022	1	5	0	6
Oct 2022	2	4	0	6
Nov 2022	2	5	0	7
Dec 2022	3	5	0	8
Jan 2023	2	7	0	9

Cumulative views and downloads (calculated since 18 Nov 2021)

Month	HTML	PDF	XML	Total
Nov 2021	162	45	2	209
Dec 2021	101	33	2	136
Jan 2022	64	24	5	93
Feb 2022	23	8	0	31
Mar 2022	27	12	2	41
Apr 2022	17	16	4	37
May 2022	2	8	1	11
Jun 2022	3	5	1	9
Jul 2022	1	4	0	5
Aug 2022	2	3	0	5
Sep 2022	1	5	0	6
Oct 2022	2	4	0	6
Nov 2022	2	5	0	7
Dec 2022	3	5	0	8
Jan 2023	2	7	0	9

Viewed (geographical distribution)

Total article views: 584 (including HTML, PDF, and XML) Thereof 584 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 10 Jan 2023

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (4002 KB)
Metadata XML


Total:	0
HTML:	0
PDF:	0
XML:	0