Research on the Uncertainty of Landslide Susceptibility Prediction Using Various Data-Driven Models and Attribute Interval Division
Round 1
Reviewer 1 Report
In this paper, five continuous landslide impact factor interval attribute classifications and three data-driven models (DBN, RF, andBP) were used for a total of 15 different scenarios of landslide susceptibility prediction studies, and the landslide susceptibility index uncertainty analysis was performed. But the quality of preseatation is very poor and there are little novelty points in this paper. Besides, the selection basis for the interval attribute classifications should be well clarified.
Other edits comments:
1. The title of section 5.1 is same as 5.2.
2. Line 86-90, the research production expression should be further refined.
3. There are lots of grammar mistakes in the text and should be corrected.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Landslide susceptibility prediction has proven to be a fundamental and effective tool for predicting the spatial occurrence of landslide hazards in an endangered region. The process of landslide susceptibility prediction is highly affected by the input reliability and implementation models. Various approaches with different sets of assumptions and procedures were developed to improve accuracy of this process, including heuristic, physically based, statistical and advanced machine learning methods. The authors have analyzed the uncertainty of landslide susceptibility prediction using various data-driven models and attribute interval. They used three data-driven models (Deep 19 Belief Networks (DBN), Random Forest (RF), and Neural Network (Back Propagation (BP). The obtained results can be very useful for efficient and accurate construction of the landslide susceptibility prediction models.
Detailed comments:
1. Landslide susceptibility prediction is a prerequisite for preventing and reducing landslide hazards, as one of the significant natural hazards that cause great damage to human life and the regional economy. Therefore, the authors in the introduction should expand the references and emphasize the relationship between landslide susceptibility, hazard and risk assessment (recommended paper: Skrzypczak et al. Landslide Hazard Assessment Map as an Element Supporting Spatial Planning: The Flysch Carpathians Region Study. Remote Sens. 2021, 13, 317, https://doi.org/10.3390/rs13020317)
2. Some figures are unclear or too small and require correction (e.g. Fig. 6)
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
i. This topic is better, but the content expounded by the author is mostly a comparison of several prediction methods, and I think the main purpose of his research is not well explained.
ii. I think it is very important that the uncertainty of landslide prediction is related to model establishment, the choice of prediction method, and the impact factor. The sensitivity of impact factors and predictions is not sufficiently elaborated, which will lead to uncertainty in the results of this paper.
iii.The introduction of several methods accounts for a certain amount of space in this paper, and these methods are also commonly used at present. It is suggested to simplify this part and increase the elaboration of the impact factors of landslide occurrence.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
In this paper, the uncertainty of landslide susceptibility prediction is studied based on different data-driven models and attribute interval segmentation. The paper is logical in structure and innovative in content.
It is recommended to be accepted after minor revision.
I list below some suggested revisions that I hope will help improve the quality of the manuscript:
Line 117:(ROC) should be in front of curve
Line 146:Out of Bag (OOB),Some abbreviations of this article are lowercase and some are uppercase. Please make the whole article the same.
Line 203: 2441.4km2 ，The number should be expressed in English. If the number exceeds 1000, separate it with commas. km2 should be km2. The following 13000m2 and 6000m2 are the same problem, please check the full text of the same problem and modify.
Line 195:The abbreviation MSE should be written with its full name： mean-square error
Line 264: 4, 8, 12, 16 and 20 should be 4, 8, 12, 16, and 20 . The same problem will not be repeated in the future. Please modify it uniformly
Table 1-3:These three tables show a lot of results, but this presentation is clunky and unreadable, so try to use graphs to show results.
The text in figure 6 is not clear. Please improve the quality of the figure
Line 403: (Table 6) cannot be a single sentence, please put it in the text to describe.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
l The research in this paper is interesting, however, the result graphs are not as ideal as the authors suggest and there are some possible academic errors.
The authors point out lines 23-26 “The results demonstrate that for the same model, as the interval attribute value rises from 4 to 8 and 24 finally to 20, the forecast accuracy of landslide susceptibility initially increases gradually, then progressively grows until stable.”
However, this conclusion could not be obtained from the figure 6, and the calculations in figure 6 and table 5 do not match. Let’s take a look one by one.
l The authors used the following method to test the model. “Three perspectives were used to analyze the uncertainty of the prediction results: 128 the receiver operation characteristic (ROC) curve accuracy evaluation, the susceptibility 129 index difference, and its distribution law.” The actual situation of the landslide samples of the testing set falling into each partition is an important factual basis for judging the prediction results, and it is highly recommended to add this content.
l As the authors state, there is no uniform quantity standard (AIN) for the segmentation of factors based on frequency ratios. In order to get a relatively suitable segmentation situation, scholars assign the segmentation criteria suitable for each of them based on the span of different factor value range intervals after comprehensive consideration. The author presupposes the same AIN for continuous variables in this paper, can such segmentation ensure the maximum effectiveness of factors?
l I did not see the screening process for the factors in the manuscript, which was actually one of the elements that influenced the results. How did the authors consider this?
l The landslide inventories and all data in the manuscript need to be clearly labeled with the access to the source and cited.
l Line 50-51:“Landslide susceptibility evaluation is so crucial that it has recently emerged as one of the main areas of interest in landslide risk studies”
This may not be the appropriate expression, as the topic of landslide susceptibility dates back to the last century.
l Line 223, 227, 234, 262 ...
The expression of numbers and units is relatively uncommon, perhaps it would be better to modify them.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The Paper can be accepted
Author Response
Thank you again for giving us so many valuable comments, which are indeed helpful for improving our submitted manuscript. Your comments are also of great significance to deepen our future researches.
