Landslide Susceptibility Research Combining Qualitative Analysis and Quantitative Evaluation: A Case Study of Yunyang County in Chongqing, China

Round 1

Reviewer 1 Report

The papers uses random forest methodology to derive landslide susceptibility maps in a region of China. The premise, and conclusion, is that stratifying the region in zones, guided by qualitative knowledge of the region's landslide behavior improves the results over a bulk analysis of the region. The result is not surprising but useful. In my opinion the actual performance improvement is marginal but the different selection of important explanatory variables is useful. Nevertheless, the work needs improvement before publishing. I have several comments:

1. This paper does not belong in Forests - it truly has little to do with the focus of the journal. It wold be addressing the wrong audience. A natural hazards journal would be more appropriate.

2. I started trying to edit the use of English - vocabulary, grammar, spelling - and could not get beyond the abstract. It would need a LOT of work. I am not a stickler for language perfection and certainly this is not about style - which I respect differences. It is about proper writing. Typos are common:  EVALUATION in figure 4, fore for for, tone vs tuned, etc.

3. The presentation is too long and could be shortened significantly. Foer example there is a lot of overlap in figures 9,10,11. There are other figures that are of marginal value.

4. The technical discussions, for example the description of the random forest approach, is full of jargon which always worries me. Do the authors fully understand what their packaged algorithm is doing?

5. 20 % verification set is less than ideal. The discussion of verification can be strengthened. I cannot tell what is concluded from the verification set and the training set. Are susceptibility maps truly representative of the history?

6. The format/font of all numbers is completely wrong and confusing - they appear almost like subscripts/superscripts.

7. The use of terms like fishnet vs grid is unusual to me.

Author Response

1. This paper does not belong in Forests - it truly has little to do with the focus of the journal. It would be addressing the wrong audience. A natural hazards journal would be more appropriate.

Reply: Thank you for your kindly reminder. The topic of this special issue is “Landslides in Forests around the World: Causes and Solution”. And this manuscript emphasized the importance of qualitative knowledge for improving the accuracy of landslide susceptibility mapping; thus, a more accurate map could be used as a better solution to reduce the loss of life and property by avoiding the development of landslide-prone areas. Furthermore, as a mountain city, the forest area of Chongqing is more than 54.5%, and that of Yunyang County even exceeds 58.5%. To emphasize the topic more clearly, we have added the related contents to this paper as follows (Lines 120~123): “According to the announcement of the Chongqing Forest Bureau, while the forest area of Chongqing city is more than 54.5%, that of Yunyang County exceeds 58.5%, making it one of the greenest counties in China.”

2. I started trying to edit the use of English - vocabulary, grammar, spelling - and could not get beyond the abstract. It would need a LOT of work. I am not a stickler for language perfection and certainly this is not about style - which I respect differences. It is about proper writing. Typos are common:  EVALUATION in figure 4, fore for for, tone vs tuned, etc.

Reply: Thank you a lot for your detailed review! We carefully reviewed the vocabulary, grammar, and spelling of the entire manuscript. And the related changes are marked in red in the revised article.

3. The presentation is too long and could be shortened significantly. Foer example there is a lot of overlap in figures 9,10,11. There are other figures that are of marginal value.

Reply: Thanks for your suggestion, we have optimized Figs. 1, 3, 4, 6, and 9.

4. The technical discussions, for example the description of the random forest approach, is full of jargon which always worries me. Do the authors fully understand what their packaged algorithm is doing?

Reply: We are sorry for such worries. To improve the understandability, we added the mathematical expression of the voting process (Lines 211~212).

We also adjusted the expression in “Bootstrapping” to make it fit to the equation (Lines 204~205). In addition, a flowchart of random forest is added (Fig. 5).

“Bootstrapping: In order to build M decision trees (estimators), M subsets will be generated as training sets from the original dataset by sampling with replacement.”

In fact, our team had done tons of work related to machine learning applications, such as “

Zhang, W., Wu, C., Zhong, H., Li, Y., & Wang, L. (2021). Prediction of undrained shear strength using extreme gradient boosting and random forest based on Bayesian optimization. Geoscience Frontiers, 12(1), 469–477. https://doi.org/10.1016/j.gsf.2020.03.007

Zhang, W., Zhang, R., Wu, C., Goh, A. T. C., Lacasse, S., Liu, Z., & Liu, H. (2020). State-of-the-art review of soft computing applications in underground excavations. Geoscience Frontiers, 11(4), 1095–1106. https://doi.org/10.1016/j.gsf.2019.12.003

Zhang, W., Wu, C., Li, Y., Wang, L., & Samui, P. (2019). Assessment of pile drivability using random forest regression and multivariate adaptive regression splines. Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, 15(1), 27–40. https://doi.org/10.1080/17499518.2019.1674340”.

Therefore, we believe that we have enough experience with how the algorithm is doing.

5. 20 % verification set is less than ideal. The discussion of verification can be strengthened. I cannot tell what is concluded from the verification set and the training set. Are susceptibility maps truly representative of the history?

Reply: Thanks for your suggestions. First, the 80/20 ratio was chosen based on the cases in previous studies (Nuh et al. 2020, Wang et al. 2021). More importantly, because the number of data in the sub-zones is kind of limited in this case, the ratio of 80/20 can guarantee that we have enough samples for model training.

In the previous discussion of verification, we mainly proved that the proposed models successfully generate reasonable landslide susceptibility maps for the study area. The maps were generated by entering all 5,831,382 cells in the study area of the proposed models. In each of the generated maps, the historical landslides/area ratio of the areas with very high landslide susceptibility is the highest, while that of the areas with very low landslide susceptibility is the lowest. To make it more clear, we have separated the results and discussion into two sections, and made some changes to section 4.2 by adding the following sentence (Lines 316~317):

“After obtaining well-trained models, we used all 5,831,382 cells prepared in section 4.1 as input to generate landslide susceptibility maps for the study area.”

6. The format/font of all numbers is completely wrong and confusing - they appear almost like subscripts/superscripts.

Reply: Thanks for your reminder, all numbers in the word file we originally uploaded were in the right format, but they turned to superscripts because of the auto-format tool of the submission system. We now fix this problem in the revision.

7. The use of terms like fishnet vs grid is unusual to me.

Reply: Both fishnet and grid are the terms that are commonly used in GIS. Grid is a raster data storage format native to Esri, it is like a continuous data layer, different colors represent different values. However, it cannot be used directly as input for random forest models. Using the Create Fishnet tool of GIS, a feature class containing a net of small rectangular cells can be created, and their central points can be used as containers to save the extracted data from all of the grid data layers. After that, they can be used for model training. In this case, the study area was separated into 5,831,382 cells, and the 20 grid data layers were extracted into each of them. So they can be used as the input for random forest models to generate landslide susceptibility maps. The related contents were added to this paper as follows (Lines 267~270):

“As the grid data layers cannot be directly used for model training, the Yunyang County fishnet with a cell size of 25 m25 m was created for the purpose of data extraction for model training, and measuring the distances from specific structures/natural sources, the total number of cells is 5,831,382.”

Author Response File: Author Response.docx

Reviewer 2 Report

This paper presents the results of research „ Landslide susceptibility research combining qualitative analysis and quantitative evaluation: A case study of Yunyang  County in Chongqing, China”

 The research was aimed at combining qualitative and quantitative analysis and examining its influence on the accuracy of the mapping, based on the importance of the features and related literature. The research area was Yunyang County, Chongqing City, China. These are promising results. I hope the authors will continue their research on this issue. I also hope that the results of their work will find application. Unfortunately, the authors made mistakes that need to be corrected.

More specifically, I have the following main comments:

ü  About study area

There are stupid naming errors here – look 105.

Mountainous areas are usually susceptible to mass movements due to the frequent

occurrence of conditional and triggering events- look 129

A banal and somewhat incomprehensible sentence. I would like a more detailed explanation.

ü  About Methodology

You consider the factors:

               -Distance to water- is not precise. Whether it is a distance from streams, rivers, or maybe other streams?

              -Average annual temperature

I wonder about the sensibility of using the temperature factor. Is it necessary? Evidence would be requested that this is an important factor in modeling landslide susceptibility.

ü  Results and Discussion

It is not a good idea to combine the results and discussions. It's getting a mess. Please separate the results from the discussion. Both chapters should be prepared separately.

ü  Conclusion

it is worth emphasizing more clearly what was new about the results of the research .

ü  References

The bibliography contains 56 items. This is a sufficient amount. The most recent items are cited.

Correct errors in figures: Fig.1,2,3,4,6,8,10 and 11.

For further detailed comments, see the accompanying manuscript.

Comments for author File: Comments.pdf

Author Response

1. About study area

There are stupid naming errors here – look 105.

Mountainous areas are usually susceptible to mass movements due to the frequent

occurrence of conditional and triggering events- look 129

A banal and somewhat incomprehensible sentence. I would like a more detailed explanation.

Reply: Thanks for your detailed review. We have fixed the related errors (Line 107), and added a more detailed explanation as follows (Lines 138~141):

“Mountainous areas are generally susceptible to mass movements due to preparatory and triggering causal factors [26], not only weathering effects, but anthropogenic activities in the region also commonly accelerate the formation of unstable areas on both the earth material and on hill slopes [27].”

[26] Nakileza, B. R., & Nedala, S. (2020). Topographic influence on landslides characteristics and implication for risk management in upper Manafwa catchment, Mt Elgon Uganda. Geoenvironmental Disasters, 7(1). https://doi.org/10.1186/s40677-020-00160-0

[27] Nefeslioglu HA, Gokceoglu C, Sonmez H, Gorum T (2011) Medium-scale hazard mapping for shallow landslide initiation: the Buyukkoy catchment area (Cayeli, Rize, Turkey). Landslides. 8, 459–483. https://doi.org/10.1007/s10346-011-0267-7

2. About Methodology

You consider the factors:

               -Distance to water- is not precise. Whether it is a distance from streams, rivers, or maybe other streams?

              -Average annual temperature

I wonder about the sensibility of using the temperature factor. Is it necessary? Evidence would be requested that this is an important factor in modeling landslide susceptibility.

Reply: Thank you for your kind reminder; we have changed the “distance to water” to “distance from rivers” to clarify the expression. To emphasize the necessity of using the temperature factor, we have strengthened the description as follows (Lines 379~381).

“Temperature has a remarkable effect on landslide formation, experimental results indicated that the shear strength of slip surface soils reduces with decreasing temperature, which will negatively affect slope instability [43].”

[43] Shibasaki, T., Matsuura, S., & Okamoto, T. (2016). Experimental evidence for shallow, slow-moving landslides activated by a decrease in ground temperature. Geophysical Research Letters, 43(13), 6975–6984. https://doi.org/10.1002/2016gl069604

3. Results and Discussion

It is not a good idea to combine the results and discussions. It's getting a mess. Please separate the results from the discussion. Both chapters should be prepared separately.

Reply: Thanks for your suggestion; we have properly separated those two sections. Among them, the results part majorly the outputs of different models, the AUC improvement of sub-models, and the feature importance (Lines 320~370). The discussion part majorly includes feature importance analysis and model comparison (Lines 371~438).

4. Conclusion

it is worth emphasizing more clearly what was new about the results of the research .

Reply: We have added a paragraph to emphasize the innovation and potential application of the research as below (Lines 454~469).

“However, more general information extracted from “mainstream” landslides would usually cover that of the “minority” landslides when treating a large region equally, resulting in low information utility and the inability to identify potential landslides under special geological conditions. With enough data points, experience-based zoning before modeling is proved to be an effective solution to the issue, the qualitative analysis serves the purpose of pre-classification based on the information from geological hazards exploration, which groups the landslides that occurred under similar geological conditions, and thus enables the models to obtain the specific knowledge under each condition. Therefore, in our case, while the traditional RF obtained the general prediction skill for the entire region of Yunyang County, all the sub-models have become “experts” in their respective sub-areas. The test AUC values of sub-model one to four are 8.8%, 2.3%, 1.9%, and 9.1% higher than those of the parent model. Furthermore, the proposed method also contributes to further revealing the key factors that include local landslide instability under specific geological conditions, which can be used by planners and policymakers for a more specific and accurate landslide control in certain areas, thus further improving the safety of life and public property.”

5. References

The bibliography contains 56 items. This is a sufficient amount. The most recent items are cited.

Correct errors in figures: Fig.1,2,3,4,6,8,10 and 11.

Reply: Thanks a lot for your support! We have corrected all the errors in the figures as you suggested.

6. For further detailed comments, see the accompanying manuscript.

Reply: Thanks for your detailed comments. We have added the tectonic map (Fig. 2) and revised some wrong expressions such as “distance from water” to “distance from rivers”, “distance from anticline” to “distance from syncline to distance from syncline axis”, etc.

For the problem in the conclusion section “How to explain that synclines are more susceptible to anticlines”. My response is that as feature importance generally represents the importance of a feature in the decision-making process of the model, the highest importance does not necessarily mean the feature is the key triggering factor of the landslide; it can also be the key factor for identifying non-landslide points. To make it clear, the following sentences were added (Lines 353~355).

“The importance of feature generally represents how much a specific feature contributes to the decision-making process of a model. In this case, the most important features can be the key factors in identifying landslide/non-landslide points.”

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Most of my comments were taken on board and the manuscript was revised. There is also a problem with fig. 1. It is still not very attractive. Surface waters and landslides are only shown schematically. This is why a slight change intervenes. Please move Fig.2 to Fig.1.Details in the manuscript.

Comments for author File: Comments.pdf

Author Response

Response to the reviewer’s comments:

The authors sincerely thank the editor and reviewers for their valuable feedback. We have used their comments to improve our manuscript. Based on the recommendations made by the editor, we have transferred Fig. 2 to Fig. 1, modified the error in Table 2b, and upgraded Fig. 11. We have also made further improvements to the article. These changes do not affect its content or framework. Changes are marked in red in the revised article. The editors’ comments are laid out, and our responses are given below. We hope that this revised article meets the journal’s publication requirements.

Reviewer #2:
Most of my comments were taken on board and the manuscript was revised. There is also a problem with fig. 1. It is still not very attractive. Surface waters and landslides are only shown schematically. This is why a slight change intervenes. Please move Fig.2 to Fig.1.

Reply: Thank you for your detailed review. We have moved previous Fig. 2 to Fig. 1. As the surface water and road distribution were displayed in Fig. 2 and Fig. 3, we deleted that from Fig. 1 to avoid the repetitive expression. Besides, we also have fixed the error in Table 2b. Moreover, to provide a better visualization, Fig. 11 has been upgraded.

Author Response File: Author Response.docx