Using Persistent Scatterer Interferometry for Post-Earthquake Landslide Susceptibility Mapping in Jiuzhaigou
, Yun Shao 1,2,3, Chou Xie 1,3,*, Bangsen Tian 1,*, Yu Zhu 1, Yihong Guo 1, Qing Yang 4 and Ying Yang 1
Round 1
Reviewer 1 Report
The manuscript deals with a topic of great scientific interest and addresses well-known methodological procedures. However, the data related to landslide distribution are quite concentrated in very different situations, which raise the question whether they should not be considered separately. I think this is a point that should be discussed at the beginning of the methodological proposition or in the discussions item.
The results presented on the LSM maps, especially the high susceptibility classes show high control by attributes and their classes. But, due to the dimensions of the figures of each group of attributes in Figure 3, it becomes impossible to evaluate the relationships. Thus, it is based on the discussion item that the authors discuss the relationships between the susceptibility classes of the maps of Figures 9 and 11 with the attributes.
The figures need to be improved and placed in sizes that readers can relate the basic data and the results obtained.
Further remarks can be found in the manuscript that is attached.
Comments for author File: 
Comments.pdf
Author Response
Dear Editor and reviewers:
Thank you for your letter and the reviewers’ comments on our manuscript entitled "Using Persistent Scatterer Interferometry for Post-earthquake Landslide Susceptibility Mapping in Jiuzhaigou" (ID：applsci-1859129). Those comments are very helpful for revising and improving our paper, as well as the important guiding significance to other research. We have studied the comments carefully and made corrections which we hope meet with approval. The main corrections are in the manuscript and the responds to the reviewers’ comments are as follows (the replies are highlighted in blue ).
Replies to the reviewers’ comments:
Reviewer #1:
1 The manuscript deals with a topic of great scientific interest and addresses well-known methodological procedures. However, the data related to landslide distribution are quite concentrated in very different situations, which raise the question whether they should not be considered separately. I think this is a point that should be discussed at the beginning of the methodological proposition or in the discussions item.
Answer：We have added a comparison of pre and post-earthquake landslide susceptibility maps (Fig. 11), which are discussed in the Results and Discussion section. Specifically in lines 409 and 540 of the article.
2 The results presented on the LSM maps, especially the high susceptibility classes show high control by attributes and their classes. But, due to the dimensions of the figures of each group of attributes in Figure 3, it becomes impossible to evaluate the relationships. Thus, it is based on the discussion item that the authors discuss the relationships between the susceptibility classes of the maps of Figures 9 and 11 with the attributes.
Answer：All figures have been modified in the paper and could be easily identified.
3 Both landslide distributions are associated with fault-affected zones? If so, an integration of Figure 2 with Figure 3h would be interesting!
Answer：According to the modified Figure 2b, pre-earthquake landslides in Jiuzhaigou are mainly distributed on both sides of the Tazang Fault, with the earthquake-induced landslides located between the Minjiang Fault and the Tazang Fault.
4 What are shallow rock slides and rockfalls?
Answer：This phenomenon(shallow rock slides) is mostly found in areas of karst action, where water flows down through the solution holes in the overlying rock, and the karst water and mudstone layers produce mudification by water-rock interaction. This mudification causes a reduction in internal friction, resulting in a significant reduction in rock stability.
Rockfalls means a sudden downward fall of soil or rock on a slope under the action of gravity. Precipitation, weathering and earthquakes are often the triggers for the rockfalls of unstable slopes.
5 Figure 3:Figures are very small! it is impossible to see the data!!
Answer：All figures have been modified in the paper and could be easily identified.
6 212:How were the probabilities obtained? Or it was obtained - Frequency????
Answer：This probability represents the probability that a landslide will occur in the area, and this probability value is a predicted value calculated by statistical or machine learning models from occurred landslides and landslide factors. Landslide susceptibility is a binary classification problem, i.e. whether it is a landslide or not, and the output is two nodes and that satisfy +=1. The x is the input factor for each indicator category and A is the landslide occurrence event. Then the classification model function :
The is a softmax or sigmod function that turns the output values of the output nodes into a probability distribution and constrains the sum of the output values of each output node to 1.
Therefore, the probability of landslide mentioned in the paper is the .
7 The figures need to be improved and placed in sizes that readers can relate the basic data and the results obtained.
Answer：We have modified the figures in these graphs and they can now be easily identified.
8 Tabel 2:
Answer：The LULC is land use and land cover. We have replaced it.
9 Figure 6：This figure needs more explanation!!
Answer：We have added an explanation for Figure 6 to the legend and the article, modified the size of the text in the figure and they are now clearly identifiable.
10 Figure 7：Put the specified area on a more detailed scale!!!
Answer：We have modified this figure on a more detailed scale and added google optics images.
11 Figure 10 The differences in the values are not significant!
Answer：In fact, the variation in susceptibility is mainly due to InSAR corrections, and the cells that need to be corrected are areas with large changes in surface deformation (>20 mm/year), which represent a very small part of the whole area (Figure 6). At the same time, landslide areas are also a phenomenon in parts of the study area. Therefore, according to the correction matrix and InSAR results, 4.87% of the cells were corrected.
12 In the manuscript, on the inventory map, in figure 2 there is no separation between before and after ecological restoration! There are not 2 NVDI maps considering the differences! It is difficult to verify the conclusion!
Answer：Ecological restoration is now measured by indicators, such as NDVI. We added NDVI for the earthquake period and after ecological restoration (2017.8-2017.12 and 2021.8-2021.12). in Figure 15 in the article, we provide changes in NDVI for these two periods, illustrating the effect of ecological restoration. Moreover, Jiuzhaigou news and UNESCO reports all illustrated the ecological restoration.(https://news.cgtn.com/news/2021-10-03/China-s-ecological-conservation-and-restoration-efforts-are-paying-off-143vWDsACje/index.html) In fact, the measures consisted of two main components. ecological management and disaster mitigation management. These measures have reduced the susceptibility to landslides.
Author Response File: Author Response.pdf
Reviewer 2 Report
Generally the subject of the paper is of interest for the scientific community, as well as for the local administration/population. Therefore I feel that major revisions are necessary to prepare the manuscript ready for publication.
Major comments:
(1) There is a feeling that the main aim of the study is the recognition of landslide events form InSAR images. For careful prediction of earthquake-induced landslide there should be used strong motion parameters. This information is not provided in the article. For the post-earthquake landslide prediction the post-seismic landslides should be used in the learning algorithms. I did not find any reference to the post-earthquake landslides considered as a training subset. How can you provide the probability (or susceptibility) changes if you did not register any post-earthquake landslides? It is not clear which type of landslide were considered as a training subset (seismically induced landslides or pre-earthquake landslides or post-earthquake landslides). I feel that Abstract, Introduction and Methods should be rewritten so, that reader could understand the main principles of data processing. The Methods should be explained carefully stet by step.
(2) The equations and coefficient obtained by the logistic regression tool should be added to the text and briefly discussed.
(3) For the imbalanced dataset the threshold probability of the logit-model may be differ from 0.5. In this case the tuning of the threshold should be performed. After that the performance metrics should be recalculated if threshold probability significantly differ from 0.5. It is strongly recommended to analyze the threshold and provide a brief description.
(4) The counts of landslide and non-landslide events considered in the analysis should be mentioned in the article.
(5) It is strongly recommended to insert the table with the full description of causative factors considered in the current study. The appropriate references to the data used in the analysis should be provided in the table.
Some specific comments:
(1) Line 62: There is excess dot.
(2) Line 144: In addition, for training and validation, we randomly created the same number of non-landslide points, ensuring that the distance to the landslide points was greater than 500m and non-plain areas.
It looks like incomplete sentence: …and non-plain areas...
(3) Abstract: There are grammatical and orthographic errors in the first sentence. It should be rewritten.
(4) Generally, the language needs to be improved; it is recommended to check the manuscript by a native speaker.
(5) Fig. 2: It is not clear from the figure and figure caption which landslide are induced by the Ms 7.0 earthquake and which landslides are referred to the pre-earthquake ones.
(6) Fig. 2: The epicenter marker should be replaced by a fault line since the magnitude of the earthquake is significant and it cannot be associated with the point source.
(7) Fig. 3f: It is not clear which rupture plane is associated with the Ms 7.0 earthquake considered in the study. It should be clarified.
Author Response
Dear Editor and reviewers:
Thank you for your letter and the reviewers’ comments on our manuscript entitled "Using Persistent Scatterer Interferometry for Post-earthquake Landslide Susceptibility Mapping in Jiuzhaigou" (ID：applsci-1859129). Those comments are very helpful for revising and improving our paper, as well as the important guiding significance to other research. We have studied the comments carefully and made corrections which we hope meet with approval. The main corrections are in the manuscript and the responds to the reviewers’ comments are as follows (the replies are highlighted in blue ).
Replies to the reviewers’ comments:
Reviewer #2:
1 Generally the subject of the paper is of interest for the scientific community, as well as for the local administration/population. Therefore I feel that major revisions are necessary to prepare the manuscript ready for publication.
Answer: Jiuzhaigou is a world-renowned natural heritage and tourist area with great human and ecological values. However, the earthquake in 2017 caused great damage to the ecological and geological environment of the region, with landslides and other phenomena occurring from time to time, posing a threat to the scenic spots and people's lives. After four years of ecological restoration and landslide management, the ecological environment has been effectively improved. We applied InSAR techniques and random forest models to dynamically assess the landslide susceptibility of the region 4 years after the earthquake, proving the effectiveness of the dynamic update LSM method and assessing the current level of landslide susceptibility. All these works provide a reference for the subsequent management and construction of the region.
2 There is a feeling that the main aim of the study is the recognition of landslide events form InSAR images. For careful prediction of earthquake-induced landslide there should be used strong motion parameters. This information is not provided in the article. For the post-earthquake landslide prediction the post-seismic landslides should be used in the learning algorithms. I did not find any reference to the post-earthquake landslides considered as a training subset. How can you provide the probability (or susceptibility) changes if you did not register any post-earthquake landslides? It is not clear which type of landslide were considered as a training subset (seismically induced landslides or pre-earthquake landslides or post-earthquake landslides). I feel that Abstract, Introduction and Methods should be rewritten so, that reader could understand the main principles of data processing. The Methods should be explained carefully stet by step.
Answer: Movement parameters are an important factor when predicting earthquake-induced landslides. This is why we have used InSAR data to correct for initial landslide susceptibility. In the paper, we have used InSAR to monitor deformation in the Jiuzhaigou area and converted the rate on the radar line of sight into a slope-oriented velocity according to Formula 1(in papaer). Many studies have demonstrated that converting slope-oriented deformation rates is a good motion parameter.(https://doi.org/10.1007/s10346-017-0940-6 et.) The training subset is composed of pre-earthquake landslides, which were provided by the Sichuan Ecological and Environmental Restoration Institute, and post-earthquake landslides, which were provided and requested for access by Professor Xu Chong et al. in the web(DOI:10.12072/ncdc.LANDSLIDE.db2067.2022 ). These landslides were obtained using satellite imagery for interpretation. We have included a description of this data in the revised manuscript.
We have rewritten the Abstract, Introduction and Methods sections to clarify the main processes and principles of data processing.
3 The equations and coefficient obtained by the logistic regression tool should be added to the text and briefly discussed.
Answer: we include an analysis and discussion of logistic regression tools in row 308 and coefficient obtained by the logistic regression tool is added in row 393.
4 For the imbalanced dataset the threshold probability of the logit-model may be differ from 0.5. In this case the tuning of the threshold should be performed. After that the performance metrics should be recalculated if threshold probability significantly differ from 0.5. It is strongly recommended to analyze the threshold and provide a brief description.
Answer: The datasets on landslides and non-landslides are balanced. In total there are 410 pre-earthquake landslide points 97 post-earthquake landslide points and 507 non-landslide points.
5 The counts of landslide and non-landslide events considered in the analysis should be mentioned in the article.
Answer: an equal number of landslide and non-landslide points were used as shown in the figure below. In total there are 410 pre-earthquake landslide points 97 post-earthquake landslide points and 507 non-landslide points.
6 It is strongly recommended to insert the table with the full description of causative factors considered in the current study. The appropriate references to the data used in the analysis should be provided in the table.
Answer:We have included a full paragraph of discussion of landslide factors, the exact location of which is in row 164.
Some specific comments:
1 Line 62: There is excess dot.
Answer:We have deleted it.
2 Line 144: In addition, for training and validation, we randomly created the same number of non-landslide points, ensuring that the distance to the landslide points was greater than 500m and non-plain areas.
It looks like incomplete sentence: …and non-plain areas...
Answer:We have rewritten it.
3 Abstract: There are grammatical and orthographic errors in the first sentence. It should be rewritten.
Answer:We have rewritten it.
4 Generally, the language needs to be improved; it is recommended to check the manuscript by a native speaker.
Answer:We have checked manuscript by a native speaker.
5 Fig. 2: It is not clear from the figure and figure caption which landslide are induced by the Ms 7.0 earthquake and which landslides are referred to the pre-earthquake ones.
Answer:We have redrawn Figure 2 and distinguished between earthquake-induced landslides and pre-earthquake landslides.
6 Fig. 2: The epicenter marker should be replaced by a fault line since the magnitude of the earthquake is significant and it cannot be associated with the point source.
Answer:We have added a schematic, including historical earthquakes and fault lines, and illustrated it in Figure 2.
7 Fig. 3f: It is not clear which rupture plane is associated with the Ms 7.0 earthquake considered in the study. It should be clarified.
Answer:According to the modified Figure 2b, pre-earthquake landslides in Jiuzhaigou are mainly distributed on both sides of the Tazang Fault, with the earthquake-induced landslides located between the Minjiang Fault and the Tazang Fault. We have added relevant content to the paper.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have made changes to the manuscript that make the methodology and results more interesting to a reader. Only a few analyses, as pointed out in the manuscript, have been included.
Comments for author File: Comments.pdf
Author Response
Dear Editor and reviewers,
Thank you for your letter and the reviewers’ comments on our manuscript entitled "Using Persistent Scatterer Interferometry for Post-earthquake Landslide Susceptibility Mapping in Jiuzhaigou" (ID：applsci-1859129). Those comments are very helpful for revising and improving our paper, as well as the important guiding significance to other research. We have studied the comments carefully and made corrections which we hope meet with approval. The main corrections are in the manuscript and the responds to the reviewers’ comments are as follows (the replies are highlighted in blue ).
Replies to the reviewers’ comments:
Reviewer #1:
1 Row 521: CONCLUSION????
Answer: we have modified this section and moved this sentence to the conclusion in row 573(the revisions are highlighted red in the manuscript).
2 Row 548: CONCLUSION!!
Answer: We have moved this sentence to the conclusion section in row 578(the revisions are highlighted red in the manuscript).
3 English language and style are fine/minor spell check required
Answer: we have made some minor revisions to the English language style and spelling in the manuscript (the revisions are highlighted red in the manuscript).
Kind regards,
Authors
Author Response File: Author Response.pdf
Reviewer 2 Report
The manuscript is accepted in present form
Author Response
Dear Editor and reviewers,
Thank you for your letter and the reviewers’ comments on our manuscript entitled "Using Persistent Scatterer Interferometry for Post-earthquake Landslide Susceptibility Mapping in Jiuzhaigou" (ID：applsci-1859129). Those comments are very helpful for revising and improving our paper, as well as the important guiding significance to other research. We have studied the comments carefully and made corrections which we hope meet with approval. The main corrections are in the manuscript and the responds to the reviewers’ comments are as follows (the replies are highlighted in blue ).
Replies to the reviewers’ comments:
Round 2 Reviewer #2:
1 English language and style are fine/minor spell check required
Answer: we have made some minor revisions to the English language style and spelling in the manuscript (the revisions are highlighted red in the manuscript).
Kind regards,
Authors
Author Response File: Author Response.pdf
