Robust Landslide Recognition Using UAV Datasets: A Case Study in Baihetan Reservoir
, Huai-Xian Xiao 3, Zi-Hao Niu 1, Nan Jiang 3,*, Hai-Bo Li 4 and Yu-Xiang Hu 4Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsDear authors：
I think that the paper is well written and deals with an important issue which automatic recognition of weak-feature landslides. This paper proposes a new semantic segmentation model DeepLab4LS for landslide recognition. This model is based on the existing DeepLabV3+ model and incorporates a dual-encoder design to reconstruct optical and topographic images, aligning more closely with the recognition pathways of human experts. A novel loss function—Positive Enhanced loss (PE loss)—is introduced to enhance the model's understanding of positive samples. Through comparative experiments and ablation studies, the effectiveness of the model is validated.
Just a few comments and suggestions are provided below so that the authors take them under consideration.
1. The highlight of the manuscript is the introduction of terrain feature recognition for landslides. Please showcase this highlight in the abstract.
2. The introduction provides a very clear overview of the research background and the current state of research methods. However, personally, I feel that it is too lengthy, and I recommend condensing the literature review section. Some related are suggested (https://doi.org/10.1007/s10346-013-0436-y ; https://doi.org/10.1016/j.ijrmms.2024.105655).
3. --L202-L211-- For slopes with different aspects and gradients, the characteristics of hillshade vary significantly under different lighting conditions, which greatly affects the final results. How to determine the lighting parameters (such as height and angle) for creating the hillshade?
4. You introduced the topographic input in the proposed deep learning model, which is apparently not available in other situations. Specifically, what are the exact data requirements for the model?
5. The study area is quite large, but the data acquisition for this study used a quadcopter rather than a fixed-wing aircraft, so I presume that the data collection effort for this data collection was quite high, and I hope that the authors will provide more details of the flight parameters and plans.
6. Please adjust the formatting of the tables or images in the article. For example, can the first item in Table 1 be changed to fit into a single line to make it easier for readers to view?
Author Response
Dear reviewer 1,
Thanks for your important comments on this manuscript. The authors appreciate the constructive comments and technical questions provided by the reviewers. Those comments have been very helpful for the authors to improve the quality of this manuscript.
Please see the attachment for detailed response.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsAuthors performed manuscript on the landslide recognition approach applying UAV data. From one side manuscript proposes an interesting solution from the other there are many flaws. I’m doubting that the project represents a case study. This is because authors proposes a general solution using data only as an example. Authors do not focus on the particular case (btw, please correct the title, because “A Reservoir Area” is surely not a name of the site). Proposition of modified goggle model application is interesting however from the other side merging/fusion of orthophoto and DTM/DSM for landslide recognition, monitoring and risk assessment is practically a standard tool since a long time. To emphasize the proposed solution authors manipulate with citations and they compare their approach with “current methods” citing an article from 1978 (line 73). Please note that we are in the other century and there is 2024 now. So, please keep a basis of merit for arguments. Data from 1978 is called historical data.
Detail comments:
1. L.2–Authors use a term “multi-source UAV”–please clarify why multi not dual? I see two datasets here: as optical image and DEM.
2. L. 3–“A Reservoir Area” it is not a name and please do not use capital.
3. L. 61–63–This sentence is unclear, please clarify.
4. L.81–82–Authors go too much in details. This part can be easily cut.
5. L.88-93–The same as above.
6. Fig. 1 What data is presented, please provide a legend, please also check the caption.
Fig.1a–What means “suspected landslide area”? Please clarify and provide citation for the term.
Fig. 1c–What does the blue rectangle mean? Please provide a legend
7. L.95–Should be “methods”.
8. L. 96–What does “common” mean in the context? Please explain.
9. L.97–118–Please consider cutting material. Authors still go too much into details.
10. L.121–122–Please do not use capital letters if “reservoir area” is not a name.
11. L.123–129–Numeration 1–3 is unnecessary. Please reorganize it.
12. L.133–136–The sentence should be set after the next one. Please shift it and keep an order concerning the title.
13. Fig.2–Please provide a detail caption. Captions should describe elements of the figure, and it should start from left to right.
14. L.155–“This”? Please name the object.
15. L.160–Are you sure “global” is a correct “scaling”? I would use “regional”.
16. L.169–175–In my opinion this is very bizarre data division. These are 2 types of input data: UAV optical image and DEM. Then you extract slopes, aspects, hillshades from DEM. Labels are nothing else as attached text/numeration information. Authors mislead and mix the data levels. (Btw, authors also write about two data types in L.329?) So, how many data types authors present?
17. Fig.4. Please provide a detail caption.
18. L.181–What kind of energy authors write about? Please clarify.
19. L. 184–186–Material is out of the section’s scope.
20. L.193–212– Material is out of the section’s scope. Please reorganize the structure.
21. L.206–“from different times” is incorrect. It should be ”at different times”.
22. Fig.5–Please start describing the figure elements from 1 to 5.
23. Fig.6–Caption is unclear.
24. L.239–Should be “Methods”
25. L.268–272–Material is unclear. Please clarify.
26. Fig.8a–What is “landslide image”? Please clarify.
27. L.279–Please reconsider the new subsection from the line 279 (or 268).
28. L.294–307–Material should be set in Results or Discussion sections.
I noticed a free thoughts flow rather than an ordered manuscript structure here.
29. L.334–What means “final prediction map” please clarify.
30. From L.346–This should be a new subsection, otherwise the structure is unclear and unordered.
31. From L.360–This should be a new subsection including Fig.11, otherwise the structure is disordered.
32. L.398–Which structure, please clarify.
33. “L.469–Please do not start the section from “Table 2 shows...” Please change it.
34. L.472–If MobileNetV2 has the highest mIoU and authors start describing table 2 with this factor, why order of the table is absolutely different? Authors do not adopt any rules for presentation.
35. Fig.12–The same remark as above. Authors start describing the figure with c) and d) and then write about a) and b)? Please clarify numbers 1–8.
36. L.500–What means “prediction map”. Please clarify.
37. L.546–What a “topographic image” is? Please clarify. Maybe authors tried to describe “topographic features”.
38. Fig.13–What a “feature map” is? Please clarify?
39. L.618–Formula should be set out the text.
40. L.629–Please clarify the term “topographic image”.
41. L.638–Please defend the term “multi-source UAV dataset”. In my opinion authors used 2 datasets.
42. L.690–“SAUCHYN”–Only first letter should be capital, please correct the style of writing.
Author Response
Dear reviewer 2,
Thanks for your important comments on this manuscript. The authors appreciate the constructive comments and technical questions provided by the reviewers. Those comments have been very helpful for the authors to improve the quality of this manuscript.
Please see the attachment for detailed response.
Author Response File: Author Response.pdf
