A Knowledge-Guided Approach for Landslide Susceptibility Mapping Using Convolutional Neural Network and Graph Contrastive Learning
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper proposes a knowledge-guided framework for Landslide Susceptibility Mapping (LSM). A precise and reliable landslide susceptibility evaluation model is constructed based on convolutional neural networks, and maps of landslide susceptibility spatial distribution levels are generated via the Landslide Knowledge Fusion Cell (LKF-Cell) method. Therefore, landslide event characteristics with informative and semantically rich features were obtained. The results showed that the Convolutional Neural Networks (CNN) method outperforms traditional machine learning algorithms in predicting landslide probability. After sample optimization, the model's AUC improves by 3 to 6 percent. The AUC of the knowledge-guided model is 6 to 11 percent higher than that of the knowledge-free model. The paper is well written. The following comments can be addressed to improve the manuscript.
1. There is a big gap between the Section 2 and Section 3.
2. Most of the figures in the manuscript are not clear enough. The authors should improve the definition of the figures, especially the annotations in the figures. By the way, the authors should not use Chinese annotations in these figures, such as Figure 1.
3. There are too many abbreviations in the manuscript, which make the readers difficult to understand, such as FR and KDE. If the authors did not use the abbreviations more than ten times, I believe it is better to use the full name.
4. There are too many references, and some of them are too old.
Comments on the Quality of English LanguageThis manuscript should be carefully checked and read according to the typesetting requirements.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis manuscript presents a new approach (a knowledge-guided approach) for Landslide Susceptibility Mapping (LSM) using Convolutional Neural Networks. Application of the proposed approach study took place in Yunnan Province, China. The outcomes demonstrate the potential of this technique as a promising approach for LSM with high improvement in identification of landslides up to 15 %. Although this manuscript shows an interesting work, several aspects require further explanation and discussion. Additionally, the current presetation of the manuscript is considered poor. Therefore, revisions should be undertaken before publication. Below are some constructive suggestions and comments.
1. There is a lack of definition of GCL. Please provide the full name in the Paper title.
2. Please provide the full name of AUC at the beginning (Abstract or Introduction), not after the first 12 pages.
3. There is a lack of a major goal of this research in the Abstract and Introduction.
4. "Related work" can be briefly described in the Introduction.
5. What is the rationale behind the selection of 16 factors for assessment (Chapter 3.2)?
6. The “Methodology” chapter is too lengthy. The proposed approach should be presented before its application in the "Case Study".
7. Legend and text in Figures 1, 2, 5, 7, 8, and 12 is hardly visible.
8. Scientific content should be written in the passive voice. This is highlighted in the attached PDF version.
9. The manuscript is excessively long. Please consider cutting it down.
10. More comments are found in PDF.
11. The studied site covers an area of over 390,000 km2 (approximately the entire area of Japan or Germany), so this manuscript dealt with such large input data and uncertainty. It raises doubts about the reliability of the outcomes. Anyway, an improvement of 12-15% is not really high. The authors are aware of this issue and pointed it out in the conclusions. I do agree with you that it is necessary to "apply the proposed framework to smaller regions, employing more fine-grained knowledge guidance for landslide susceptibility prediction." Perhaps the outcomes would be more reliable with a higher improvement. Honestly, you should have carried out such research in much smaller area first, rather than in larger areas.
Please note that landslides often occur due to a combination of many factors (at the same time or one is triggered by another), so for each individual case, it requires individual analysis. No matter how high the improvement achieved, there is always uncertainty, so monitoring is required to prevent and mitigate landslides.
Comments for author File: Comments.pdf
The English needs improvement and should be reviewed by native speakers.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThank you for your responses
Just 2 comments regarding the quality of presentation are given. Please find them in the PDF version
The original version was 25 pages and now the revised version is 26 pages. Please cut down a couple of pages. Some suggestions in order to shorten the number of pages: Figure 10 and Figure 11 can be placed in 1 line, Table 3 and 4 can be placed in 1 table, Figure 2 is given then do you really need table 1?
The manuscript can be published after minor revision.
Comments for author File: Comments.pdf
I've noticed a sentence that needs to be rewritten in the passive voice (lines 432-434). Please find it attached.
English should be double-checked by native speakers
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
