Maximizing Impacts of Remote Sensing Surveys in Slope Stability—A Novel Method to Incorporate Discontinuities into Machine Learning Landslide Prediction
Round 1
Reviewer 1 Report
General: This paper named as “Maximising impacts of remote sensing surveys in slope stability- a novel method to incorporate discontinuities into machine learning landslide prediction” incorporated the information of discontinuities into the evaluation of rock slope stability by machine learning and revealed that the results of the evaluation with the information of discontinuities were better than that without the information. It has been recognized that the condition of discontinuities gives great influence on the stability of a rock slope in traditional rock mechanics and this report proved the importance of the discontinuities. Moreover, this paper proposed novel method of rock slope stability evaluation using machine learning.
But Some problems as follows should be solved to be published.
- “1. Introduction”; The negative impact and evaluation of slope stability of “landslide” are explained in first and second paragraphs, respectively. The target of this paper is the landslide in rock slope but slopes consist of not only rock but also soil. You should distinguish landslides in rock slope from those in soil slope. Especially the evaluation of slope stability for rock slope should be strictly differentiated from that for soil slope in second and third paragraph.
- Page 2, lines 52-55; Please provide references for extracting the factors affecting on slope stability from past literatures. Not only by ML methods but also traditional analysis. Please mention that these factors are only for rock slope or not.
- Figure 1; Right side of the figure was out of page. Please make small figure within the width of the page.
- Page 3, lines 98-99; Please show the reason for “5-meter decrease in elevation” for extraction of landslides. Is this definition reasonable especially for the slope with high inclination near vertical?
- Page 3, lines 107-108; Please explain the criteria for extracting landslide absence data. Especially explain the existence or absence of discontinuities on the slopes.
- Page 4, line 116-127; Please explain the characteristics of aerial lidar.
- Page 4, line 118-139; Please show the example of images derived from UAV photogrammetrie and aerial lidar and explain the method for surveying the dip direction or orientation of discontinuity, the minimum width by the method.
- Table 2; Please explain how to evaluate the material of bedrock in detail. Especially how to determine the shear strength.
- Page 6, line 183; Is assumed friction angle adequate? Provide the reason for the value.
- Related to friction angle along the discontinuity; Please refer past literatures and explain the friction angle along the discontinuity in rock slope. It might depends on not only the material of bedrock but also the degree of weathered or filling clay.
- 4, 5, 7, 8 and 9; Right side of the figure was out of page. Please make small figure within the width of the page. Explanation related to these figures cannot be understood due to these incomplete figures.
- “4. Results” and “5. Validation and discussion”; Explanation focused only on the importance of incorporating the information of discontinuities into the evaluation of rock slope. Please explain the effect of other factors shown in Table 2. Importance of the factors related to discontinuities rather than other factors cannot be understand because of incompleteness of Fig. 7.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Dear colleagues, thank you for your contribution on "maximising impacts of remote sensing survey s in slope stability". This is an actual and valueable topic.
Your contribution is clearly structures. Methods are well described and the outcome of your observations has been made clear in comparison figures 5, 6, 7.
The training and validation relationsship is mentioned in line 109. Wouldn´t it be helpful to highlight methods of training in the following chapter? For my point of view the important part of geological structure extraction is quite dominant in comparison to the ML training. It would be helpful to emphasize ML training and its data for your use case.
In section 3.3 the variables and conditions are described. Somehow geometric conditions (topographic characteristics) are related to geological conditions.
Have you observed any difficulties when combining those different data sources? Weren`t additional processing steps needed to harmonize the different sources? Could you add some sentences on occurring difficulties for the integration of different data sources?
In the end you propose your "novel method" to incorporate discontinuities into ML. Your specific "novel method" is not that clear. Could you please highlight or subsume the main characteristics and application field of your novel method in its own section?
Gratulation to your contribution.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Dear authors,
The manuscript presented is quite interesting and will be a valuable contribution to the landslide research community. However, I see a major issue that you should clarify in order to have your manuscript accepted: How is possible to compare two different datasets with different time windows and sum up them to carry out ML analysis. While LiDAR data spans from 2008-2014, UAV-derived data was carried out in 2019. Coastline erosion, especially in UK is a rapid mechanism for landslide development. If you compare different datasets and you rely on structures that perhaps do not exist in previous or present data is a bit difficult to compare them. Moreover, in general, public LiDAR data is little reliable for appropiate accuracy in landslide analysis (20-50 cm positioning errors).
Please, clarify this point providing the necessary support for the methodological approach you have developed in this manuscript.
Minor comments are attached in the pdf copy.
Best regards
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Revisions were made against most of the comments in previous review. But explanation on the methodology seems to be insufficient for the readers who do not know these method very well. Please explain the resolution of the extraction method in Fig. 3 especially. Minimum size of cracks or discontinuity by this method should be mentioned here especially for readers who don’t know this method very well.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Dear authors,
While 1 m resolution depends on the density of your point cloud (this is your gridding resolution, but not your models' accuracy), a different issue is related to the accuracy of the free LiDAR dataset. It is commonly observed that national LiDAR datasets are not always accurate enough. For instance, compare photogrammetric flights with accuracy from LiDAR point clouds, and you will find errors over 50 cm in many flights. Please, clarify in the text what are the accuracy errors of your LiDAR datasets.
Anyway, you have made a good effort, and the paper is well presented.
Best regards
Author Response
Please see the attachment.
Author Response File: Author Response.docx
