An Ensemble Broad Learning System (BLS) for Evaluating Landslide Susceptibility in Taiyuan City, Northern China
Round 1
Reviewer 1 Report
The authors provided a new method of landslide potential evaluation. The evaluation model proposed in the manuscript is worth further study. The comments about the manuscript is addressed below.
1. Total of 114 landslide locations were used in the model. The areas of these landslides should be provided to better understand the scale of each landslide.
2. There are five categories of each factor used in the model. Authors should briefly explain the consideration of using five categories.
3. The top three important factors, elevation, NDVI, and TWI, contribute more than 50% importance in the factor selection. The top 5 important factors contribute about 70% importance in the factor selection. It seems that not all factors are necessary to be included in the model, since the variable importance was evaluated. Authors should address a brief discussion about the difference between using all factors and using only top three (or top 5) factors in the model.
4. Figure 1. The word is not clear in the model charts. Enlarge the word size.
5. Figure 3. The word of the legends is too small to read. Enlarge the word size.
6. Line 124: "Data used", no "."
7. Line 133: "S"election of landslide conditioning factors...
The English language in the manuscript is fine for reading. Only few sentences and wording need to be revised.
Author Response
Reply to Editor and Reviewers’ Comments
Manuscript title：An ensemble broad learning system (BLS) for landslide susceptibility evaluation in Taiyuan City, Northern China
Dear editor and reviewers,
We feel great thanks for your professional review work on our article. These comments are very helpful to improve the quality of the manuscript. As you are concerned, there are several problems that need to be addressed. According to your professional comments, based on your professional opinion, we have made corrections and additions. In this revised version, changes to our manuscript were all highlighted within the document by using red-colored text. Point-by-point responses to the nice editor and two nice reviewers are listed below this letter to our previous draft, the detailed corrections are listed below.
- Total of 114 landslide locations were used in the model. The areas of these landslides should be provided to better understand the scale of each landslide.
Reply: We really appreciate your professional comments. We are very sorry that we do not have a detailed description of the area affected by each landslide.
The height of the mountains in the study area varies widely, ranging from 760 m to 2,708 m. In particular, the mountainous area in western Taiyuan is mainly composed of sandstones and shales from the Carboniferous and Permian periods and loess from the Quaternary period. The terrain is intricate and complex, and the natural geological conditions are very unfavorable. Coupled with relatively frequent human activities, such as mining, landslides are frequent in the study area. The larger volume resulted in not having more detailed geological survey data. The data that was taken was more to emphasize the damage caused to people.
For this issue, we will certainly have a more detailed study of the extent of all landslides in a subsequent study.
- There are five categories of each factor used in the model. Authors should briefly explain the consideration of using five categories.
Reply: We feel great thanks for your professional comments. We are really sorry that our statement is not accurate enough. After our review of the literature as well as specific research, now there is no consensus on the issue of factor grading. In the article (Convolutional Neural Network (CNN) with Metaheuristic Optimization Algorithms for Landslide Susceptibility Mapping in Icheon, South Korea. J. Environ. Manage. 2022, 305, 114367.) elevation, slope, and TWI are classified into five categories. In the article (Chen, X.; Chen, W. GIS-Based Landslide Susceptibility Assessment Using Optimized Hybrid Machine Learning Methods. CATENA 2021, 196, 104833.) NDVI, distance from road, distance from water, and plan curvature are classified into five categories.
In this study, the grading of all factors except slope direction was done by the natural interruption point grading method. This method is based on the statistical method of grading and classification according to the law of numerical statistical distribution, which can maximize the difference between classes. And the slope aspect is divided into nine classes at 45-degree intervals: North (N), Northeast (NE), East (E), Southeast (SE), South (S), Southwest (SW), West (W), Northwest (NW), and Plains.
3.The top three important factors, elevation, NDVI, and TWI, contribute more than 50% importance in the factor selection. The top 5 important factors contribute about 70% importance in the factor selection. It seems that not all factors are necessary to be included in the model, since the variable importance was evaluated. Authors should address a brief discussion about the difference between using all factors and using only top three (or top 5) factors in the model.
Reply: We are very grateful for your professional opinion. According to the literature (Panchal, S.; Shrivastava, A.Kr. Landslide Hazard Assessment Using Analytic Hierarchy Process (AHP): A Case Study of National Highway 5 in India. Ain Shams Eng. J. 2022, 13, 101626. Chen, W.; Zhang, S.; Li, R.; Shahabi, H. Performance Evaluation of the GIS-Based Data Mining Techniques of Best-First Decision Tree, Random Forest, and Naïve Bayes Tree for Landslide Susceptibility Modeling. Sci. Total Environ. 2018, 644, 1006–1018.), 12 landslide evaluation factors were selected. The accuracy is ensured while the computational power of the model is guaranteed. Also as shown in Table 1, all evaluation factors have VI values greater than 0, which means that all evaluation factors contribute to the prediction. As you said, the contribution of the first five factors is 70%, but removing the other factors still has an impact on the effect of the model.
- Figure 1. The word is not clear in the model charts. Enlarge the word size.
Reply: We feel very grateful for your professional comments. As you are concerned, Figure 1 is more of a narrative about the flow of the entire study. All the influencing factors, model structure, landslide susceptibility map, and ROC curves are shown in Figure 1. It has been enlarged as much as possible. The details of these maps are shown in later sections. They can be seen in detail in Figure 3 - Figure 10.
- Figure 3. The word of the legends is too small to read. Enlarge the word size.
Reply: Based on your professional advice we have modified Figure 3 accordingly. We hope it can bring better sensation to the readers.
- Line 124: "Data used", no "."
Reply: Thank you for your comment and we apologize for our lack of care. We have made the appropriate changes based on your suggestions. And the changes are highlighted in red.
- Line 133: "S"election of landslide conditioning factors...
Reply: Thank you for your comment and we apologize for our lack of care. We have made the appropriate changes based on your suggestions. And the changes are highlighted in red on line 313.
Once again, we thank the experts for their suggestions, and we will strengthen these areas of research in subsequent papers.
Sincerely yours
Corresponding author: Guorui Feng
Date: July. 5, 2023
Author Response File: 
Author Response.docx
Reviewer 2 Report
The effort applying ensemble broad learning system for landslide susceptibility evaluation is quite interesting and will bring a significant contribution in this field.
Besides this, the manuscript needs some improvements.
It is a good idea to rearrange slope values to any slope classification system. It is strange to include slope with high steepness into one category (>13) but divide undulating and flat areas. Color scheme for slope map (figure 3 should be changed too). Visually, it is hard to distinguish one class from another.
Also, it is a good idea to add data sources for DEM, RS data for NDVI etc. with date of acquisition.
I wish that my comment would be helpful in improving the quality of this research.
Thank you.
Author Response
Reply to Editor and Reviewers’ Comments
Manuscript title：An ensemble broad learning system (BLS) for landslide susceptibility evaluation in Taiyuan City, Northern China
Dear editor and reviewers,
We feel great thanks for your professional review work on our article. These comments are very helpful to improve the quality of the manuscript. As you are concerned, there are several problems that need to be addressed. According to your professional comments, based on your professional opinion, we have made corrections and additions. In this revised version, changes to our manuscript were all highlighted within the document by using red-colored text. Point-by-point responses to the nice editor and two nice reviewers are listed below this letter to our previous draft, the detailed corrections are listed below.
Thank you for your professional advice, and we apologize for our lack of rigor. We have adjusted the slope chart according to your suggestion. For the grading of the slope map, we have used the natural interruption point grading method to automatically complete the grading. This method maximizes the difference between classes. The FR values of each gradation of slope in Table 1 also show that landslides are more likely to occur in the area with slope of 3-13. rather than areas with slope less than 3.
Dem data were downloaded from the website（https://www.gscloud.cn/sources/index? pid=1&rootid=1&title=dem&sort=priority&page=1）. Dem data and remote sensing data were obtained using ASTER GDEM 30M, Landsat 8 OLI_TIRS respectively.
Once again, we thank the experts for their suggestions, and we will strengthen these areas of research in subsequent papers.
Sincerely yours
Corresponding author: Guorui Feng
Date: July. 5, 2023
Author Response File: Author Response.docx
