Mapping Post-Earthquake Landslide Susceptibility: A U-Net Like Approach
Artemi CerdÃ
Round 1
Reviewer 1 Report
Dear Authors,
I reade your paper entitled Use of a U-net like model for mapping post-2 earthquake landslide susceptibility.
I noticed it was already revised by several reviewers and that you modified your paper according to their comments.
I believe that you improved your paper and I have only few comments, that you can find in the attached document.
I believe it can be considered for publication after few minor revisions.
Best Regards
Comments for author File: 
Comments.pdf
Author Response
Thank you very much for your review. the reply is attached.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The pape is very interesting
See minor comments to improve the paper
Comments for author File: Comments.pdf
Author Response
Thank you very much for your review. the reply is attached.
Author Response File: Author Response.docx
Reviewer 3 Report
Dear Editor
I just finished the manuscript entitled " Use of a U-net like model for mapping post-earthquake landslide susceptibility.", tried to apply a U-net model for mapping post-earthquake landslide susceptibility. Although this model appears applied for the first time, (as the authors mentioned) there are several significant problems need to be resolved before publication
General comments
Title
The title needs modification and more attractions to the readers
Abstract
1-L 23 Why TM not ETM+ which date ??
2- L31-33 It is better to highlight on the main benefits to the environment instead of the influence of a pixel
Introduction
L-41 satellite, airborne and UAV are carrier not sensors. So, delete sensors please
L-42-4 Not clear ...try to reword it please
L-87-89 Not correct --reword
Study area
Record of the historical earthquake locations need to be added to figures 1 and 3 .. and brief description is needed to be explained.
Data and methods
The source of the data and the construction of conditioning factors are not clear and need more details
L159 This section needs more explanation on how these factors were built and constructed. Roads should be distance from roads, stream network should
Results
Spatial analysis between conditioning parameters and earthquakes and related landslides needs to be explained and discussed
Discussion
This section is not related to results section
Subtitles in the methods, results and discussion should be similar
Conclusion
Conclusion needs reword in clear way
Finally, the manuscript ca be accept for publication after major corrections
Author Response
Thank you very much for your review. the reply is attached.
Author Response File: Author Response.docx
Reviewer 4 Report
This paper presents a study on a U-net like model for mapping post-earthquake landslide susceptibility. The performance of the model is compared with that of other existing models. Overall this paper deals with an interesting topic. The reviewer has the following comments.
First, the presentation of the paper is poor – it seems that authors forgot to turn off the “tracking” mode. Authors should prepare manuscripts more carefully.
Line 16, the period should be deleted.
Line 134 and the subsequent texts, authors should better explain the sources of these data.
Table 1, what do the figure names refer to?
Sect 4, what about the dependence of the model accuracy on the sample/training size? What about the computational cost of the U-net like model compared with that of other models?
Author Response
Thank you very much for your review. the reply is attached.
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
Dear Editor
I just finished the manuscript entitled " Use of a U-net like model for mapping post-earthquake landslide susceptibility."
The manuscript appears in good shape. However, some minor corrections still need to resolve before publication. I highlighted some comments on the pdf version
I recommend the manuscript to be published and resolve the minor corrections
Comments for author File: Comments.pdf
Reviewer 4 Report
The paper now can be accepted for publication.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Introduction
- Please add damage cost of landslide in this part
- It should add hybrid models such as ANFIS-metaheuristic and SVM-metaheuristic in lecture
Study Area
- Depth of earthquake, peak ground acceleration (PGA) are important of each earthquake. Please add those information.
Landslide inventory
- This part isn’t clear. How do you detect landslide. Please describe.
Topography
- There are several small landslides. But, you utilized ALOS or other image. The resolution of those images are 30 meter. How do you detect small landslide?!
Lithology and Human activity
- Please add scale of maps.
Seismic parameters
- Why do use macro seismic intensity (MI)?! PGA is more important than MI.
Model architecture
- here are many well established traditional AI based statistical models which successfully applied in water resources application such as precipitation, soil moisture etc. Why did you use Convolutional neural Network over other AI based blended method?
- Authors need to provide significant information such as epoch, learning rate, learning momentum, weight decay.
- What Optimizer, loss function did you use? In the activation layer you should also generate feature maps. Authors should provide few feature maps which will provide clear idea about the DL network.
- Why authors did not apply ResNet-101 backbone network which is immensely popular r in pixel-wise classification? I would expect Resnet-101 for Spatial Prediction of flood hazard.
- Deep learning neural network characterized by more than single hidden layer. How do you chose layer number? Is Layer densely or parsley connected which is also missing in your manuscript?
- How did you create robust the model without showing any fundamental results? Several plots are mandatory to justify the model set up
Epoch vs accuracy
Results
There are several methods for calculating accuracy such as Confusion matrix. Please add those methods.
Author Response
Thank you very much for your detailed review.
My response could be found in the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report
Interesting paper with another landslide susceptibility mapping for earthquake triggered landslides at the scale of a large region, with the more positive aspects being the use of a methods which considers the influence of adjacent pixels and not only the information of each pixel, and the use of pre-earthquake terrain conditions .
General comments: I think that the methods and procedures need a more detailed description to enable a broader use of the proposed methods.
There is the common problem of referring recent works for topics which were presented, debated or created many years before, neglecting the original contributions.
The text needs improvements.
The figures should have a higher resolution to enable readability. For example the road network is not perceivable.
The references need revision namely is not acceptable that a paper is referred as: 1st author et al., - all author names should be presented.
Detail comments:
Line 50 – Reference on LR could also be:
Guzzetti, F., Reichenbach, P., Cardinali, M., Galli, M., Ardizzone, F. (2005) Probabilistic landslide hazard assessment at the basin scale, Geomorphology, 72, 272-299.
Line 51 – It would be advisable to quote the first well known application:
Yin, KL, Yan, TZ (1988) Statistical prediction model for slope instability of metamorphosed rocks. Bonnard, C. (Ed.), Proc. 5th Int Symp Landslides, Lausanne. A.A. Balkema, Rotterdam, 2:1269-1272.
Line 80 - This earthquake is was the most destructive
Line 147 – This is basic soil mechanics (although many of the landslides occurred in rock masses). Early reference could be the more than classical Terzaghi (1950).
Line 149 – Another very recent reference on topics which were already discussed in much more old references.
Lines 201-211 – Improve phrasing and clarify the procedures.
Line 253 – More appropriate reference would be, for example: Bi, J, Bennet, KP (2003) Regression Error Characteristic Curves. Proc 20th Int Conf Machine Learning (ICML-2003), Washington DC, 43-50
Line 282 – values in Fig. 10 are probabilities? Clarify.
Lines 308-320 – The results obtained with the different methods require further discussion: in test area 1 LR had a clear worst performance, while in test area 2 the performance of the 3 methods is the same, within the error margins. The apparently anomalous results in the plain with LR can be explained? (local slope? Geology? Roads?), and are in contrast with almost the same AUC for the 3 methods.
Lines 344-348 – Clarify the need for sampling and why all the study area pixels could not be used for the study, dividing the landslide inventory in different groups of two sub sets, with one sub-set for training and the other for validation.
Line 420 – Figures of % of landslides are switched.
Improve phrasing, lines: 92, 95-96, 118, 130-131, 202, 203
Author Response
Thank you very much for your detailed review.
My response could be found in the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Revision of “Use of a U-net model for mapping post-earthquake 2 landslide susceptibility” by Chen et al. (846957)
The paper present a new technique to identify the landslide area. I’m not an expert in this field, but I found very interesting the paper and very useful as a possible tool to coordinate the post-earthquake activities mainly of building reconstructions (as the author said to avoid problem of landslides in the selected place to reconstruct the town). I have small questions and suggestions that you can found below in the specific points.
Specific points:
Page 1, line 23. Please define the acronym TM (Thematic Mapper ?)
Page 2, line 79-80. Please specify some other details, like earthquake depth, , fault type, tectonic settings, etc..
Page 3, line 109. 30 m/pixel
Page 5, line 150.Please check singular/plural of this sentence (at least movements).
Page 6, line 157. Please, could you specify better how the moving windos is applied? Is it a circle? Is it moved in lat and lon? Which step?
Figure4a I suggest to start the colorbar not from 0m a.s.l. but from the minimum altitude that I mean it is about 450 m.a.s.l. of Chengdu town (if I read properly the map it is down on the right?).
Page 6, line 169. Perhaps it is better: “Lithology and faults can affect…”
Figure 6a The street are too concentrate, so I suggest to show also a zoom of a small area inside the same picture.
Page 8, line 190. I suggest to end the sentence with “maps” because all these product provided by USGS just about within half hour from seismic event are maps.
Page 8, line 191. The estimation of Macroseismic Intensity are made by USGS using the seismograph recorded by various instruments around the interested region. Why don’t use the macroseismic intensity calculate by China Earthquake Administration based on real survey? I suggest to substitute by this one that is more accurate (the disadvantage is that it requires some days or in this case probably months to survey all the hit towns…)
Paragraph 3.4.1 and 3.4.2 I suggest to list clearly the 29 layers of input (perhaps in a dedicated table or in the way the authors think is better).
Equation 3. I done some calculus and I see that F1 is equal to 2TP / (2TP + FN + FP) maybe you can insert?
Page 11 line 273. “NVIDIA” instead of “NVIDA”
Page 12 line 296. I think it is better to include also high susceptibility in this sentence of 0.5-0.7, as it has been included in the binary selection of landslide-area.
Page 16 line 347-348. This is just an example, but I have doubt how it fit as it is not an obtained result (all pixel with output = 0). Or does any combinations produce this result?
Page 16 line 353. Is it better “penalty weight” or “”the price to be paid” ?
Figure 17 (text line 399). Could you please explain better the graph? What is it the Epoch of the horizontal axis, for example?
Page 18 line 420. I think you can underline better that taking into account “extremely high” and “very high” level areas about the 74% of the total landslide areas are identified.
Author Response
Thank you very much for your detailed review and helpful suggestions.
My response could be found in the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
The literature basis of the text should include more state of the art new positions. Authors should add the literature review section to the paper with a more in-deep analysis of previous researches.
The author(s) should provide a step-by-step account of their research methods. Because at the moment, the basic methodological elements of the article are not clearly stated, explained and justified. There seem to be a mixed up of all the element which impede the transparency and accountability of the research process itself.
Author Response
Thank you very much for your review.
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I have read the revised version of the manuscript carefully. The authors have dealt with the comments and suggestions of reviewers in a highly satisfactory and constructive manner. The revised manuscript clearly meets the standards that have been required by reviewers. I propose the acceptance of the manuscript
