Comparison of RGB and Multispectral Unmanned Aerial Vehicle for Monitoring Vegetation Coverage Changes on a Landslide Area
Round 1
Reviewer 1 Report
The document is an interesting application of drone-based surveys, comparing the results obtained using an RGB UAV with those obtained from a Multispectral. The work is clear and interesting, the paper is well organized and complete with all the necessary paragraphs.
A moderate English revision is suggested, in particular by checking the consecutio temporum of verbs and avoiding the use of long sentences.
Here attached I report my observation, expressed directly in the pdf file. My main observation concerns the title and topic of the work. The paper aim at dealing with the use of drones for the characterization of landslides, but I could find only few explanations of how the results of the 4 surveys carried out, both with RGB and MS UAV, are interpreted for the characterization of instabilities. There is no mention in any way of what type of landslide it refers to, nor in what terms it is aimed at characterizing it. The distinction is made only on the 3 classes of material (vegetation, dead matter and bare soil), without however reasoning focused on evaluating the effect of these on the possibility of the presence of a landslide, nor on the characterization of the landslide in question. This is not even mentioned in the conclusions. The only references to landslides are found on lines 69-72 and 266-268, but even in this case, no characterization is present in the few comments.
The title is therefore misleading with respect to the content of the paper.
For the rest, the paper is an intersting work, including a detailed description of the strategy adopted for soil pixel-based classification. Anyway, in my opinion there are still some improvements or modifications that have to be included in the revision for the paper acceptance.
Comments for author File: 
Comments.pdf
Author Response
Point 1: A moderate English revision is suggested, in particular by checking the consecutio temporum of verbs and avoiding the use of long sentences.
Response 1: Thank you for your comment. In accordance with your advice, a review was made focusing on the consecutio temporum of verbs and long sentences.
Point 2: The paper aim at dealing with the use of drones for the characterization of landslides, but I could find only few explanations of how the results of the 4 surveys carried out, both with RGB and MS UAV, are interpreted for the characterization of instabilities. There is no mention in any way of what type of landslide it refers to, nor in what terms it is aimed at characterizing it. The distinction is made only on the 3 classes of material (vegetation, dead matter and bare soil), without however reasoning focused on evaluating the effect of these on the possibility of the presence of a landslide, nor on the characterization of the landslide in question. This is not even mentioned in the conclusions. The only references to landslides are found on lines 69-72 and 266-268, but even in this case, no characterization is present in the few comments.
The title is therefore misleading with respect to the content of the paper.
Response 2: We totally agree with your comment and the title was changed to “Comparison of RGB and Multispectral Unmanned Aerial Vehicles for Monitoring Vegetation Coverage Changes on a Landslide Area”. This study is focused on the comparison between the sensor types to map vegetation on a landslide area, instead of focusing on the stability of the landslide characterization. [L81 – L84]
The type of landslide, the aim, and the process for the classification and validation were also explained in more detail. [L99 – L104, L175 – L194]
We appreciate your report feedback and considered your observations.
Reviewer 2 Report
The authors have used two different types of drone images primarily to detect spatio-temporal changes in vegetation. Though the vegetation contrast is considered as a proxy of hillslope erosion (or landslide) in this study, discussion regarding the landslides seem secondary.
Further, the usage of such studies for landslide studies is limited to the hillslopes having moderate-high vegetation, which is not the case always. Therefore, authors are suggested to discuss the limitations of such an approach in view of different types of landslides.
Otherwise, approach, research design, and content presentation are well crafted. Minor English grammar improvements are required.
Author Response
Point 1: Though the vegetation contrast is considered as a proxy of hillslope erosion (or landslide) in this study, discussion regarding the landslides seem secondary.
Response 1: Thank you for your advice. We reconsidered the focus of this study and set our focus as a comparison between sensor types to map vegetation on a landslide area. We changed the title of our manuscript to “Comparison of RGB and Multispectral Unmanned Aerial Vehicles for Monitoring Vegetation Coverage Changes on a Landslide Area”. We improved our discussion about the impact of the classes on the landslide area. [L320 – L340]
Point 2: Further, the usage of such studies for landslide studies is limited to the hillslopes having moderate-high vegetation, which is not the case always. Therefore, authors are suggested to discuss the limitations of such an approach in view of different types of landslides
Response 2: Thank you for your observation, we explained some of the limitations of the methodology according to the dominant forest type [L323 – L326]
Point 3: Otherwise, approach, research design, and content presentation are well crafted. Minor English grammar improvements are required.
Response 3: Thank you very much, we reviewed our grammar according to your suggestion.
Reviewer 3 Report
The manuscript aims to compare vegetation classifications derived from RGB and multispectral drone imagery (several dates) on an area affected by a landslide. It uses an SVM pixel approach and assesses accuracy using a 5 k-folds cross-validation method. The title of the manuscript is deceiving as no landslide characterization is attempted. Similarly, section 2.4's title is misleading as no actual feature classification on landslides is presented. Judging the manuscript by its research aim, I suggest against publication on its present form due to a lack of focus, originality and critical discussion.
In order to improve the manuscript I suggest the following:
- Modify title to reflect contents (is it about landslides or vegetation mapping?)
- If mapping, focus introduction on land cover mapping using drone imagery, including current approaches beyond pixel based classification (e.g. object-based image analysis and/or including other machine learning and neural networks algorithms.
- If landslides, include landslide potential mapping!
- Add the assessment of parameters such as cloud cover (and pixel resolution?) to research objectives
- Justify rationale behind chosen land classes
- Suggest validation using an independent dataset as opposed to cross-validation
- Separate Results and Discussion sections
Author Response
Point 1: The title of the manuscript is deceiving as no landslide characterization is attempted. Similarly, section 2.4's title is misleading as no actual feature classification on landslides is presented. Judging the manuscript by its research aim, I suggest against publication on its present form due to a lack of focus, originality and critical discussion.
Response 1: We totally agree with your comment. The previous title did not adequately describe the purpose or content. The title was changed to: “Comparison of RGB and Multispectral Unmanned Aerial Vehicles for Monitoring Vegetation Coverage Changes on a Landslide Area”. And the title of section 2.4 was changed to “Classification and Accuracy Assessment” where we explained in detail why the cross-validation method was selected for this study. [L175 – L194]
Point 2: In order to improve the manuscript I suggest the following:
Modify title to reflect contents (is it about landslides or vegetation mapping?)
If mapping, focus introduction on land cover mapping using drone imagery, including current approaches beyond pixel based classification (e.g. object-based image analysis and/or including other machine learning and neural networks algorithms.
If landslides, include landslide potential mapping!
Response 2: We appreciate your suggestions and agreed that the manuscript was lacking focus. We tried to focus our study on vegetation monitoring, considering three different classes along the months, focusing on the comparison of the two UAVs sensor type [L79 – L92].
Point 3: Add the assessment of parameters such as cloud cover (and pixel resolution?) to research objectives
Justify rationale behind chosen land classes
Response 3: We appreciate your consideration. Additional information and explanation regarding your suggestions were added to our manuscript. [L71 – L81]
Point 4: Suggest validation using an independent dataset as opposed to cross-validation
Response 4: Thank you for your suggestion. We considered using other methods for validation such as the thematic map accuracy assessment. But as far as our understanding based on many previous studies ([45], [47]), the cross-validation presented a fairer method to compare the sensor outputs. The usage of the same reference dataset for the RGB and Multispectral UAV tried to reduce the variability of the training dataset in producing the final classification; focusing on each sensor type performance. We have added references to show that our validation method is preferred. [L188 – L194]
Point 5: Separate Results and Discussion sections
Response 5: We considered your suggestion, and the Results and Discussion sections were separated. Thank you for your suggestions.
Round 2
Reviewer 3 Report
The manuscript has significantly improved in focus and quality based on the author's revisions. I recommend the article be accepted in its present form.
