Sentinel-1 Shadows Used to Quantify Canopy Loss from Selective Logging in Gabon

Round 1

Reviewer 1 Report

1）Please explicitly clarify the novelty of the paper, especially with respect to the methodology.

 

2) To highlight the novelty, in the experiments, the proposed approach should be compared with other related methods.

Author Response

Dear Reviewer 1,

Thank you for your feedback. We appreciate your patience in waiting for our response: our lead author was conducting fieldwork in Gabon last month.

You asked us to explain what makes our paper unique, with reference to previous work:

• Please explicitly clarify the novelty of the paper, especially with respect to the methodology… the proposed approach should be compared with other related methods.

Thank you for raising this most important concern. The paper is novel in a number of ways: (1) in the type of data collected; (2) in the methods used; and (3) in the insights it provides.

Firstly, we would like to draw your attention to the following part of the abstract:

“previous work has relied on optical satellite data for calibration and validation, which has inherent uncertainties, leaving unanswered questions about the minimum magnitude and area of canopy loss this method can detect. Here, we use a novel bi-temporal LiDAR dataset in a forest degradation experiment in Gabon to show that canopy gaps as small as 0.02 ha (two 10 m x 10 m pixels) can be detected by Sentinel-1.”

Our bi-temporal LiDAR dataset is key to the novelty of this work: this type of ground validation is much richer, more reliable, and higher resolution than the validation data used in previous studies. Thanks to this advance, we were able to test the sensitivity of Sentinel-1 SAR right down the limit of its resolution. We feel this aspect of novelty is made explicit already in the above extract, and in the final three paragraphs of the introduction where the same idea is explained in more detail. In these paragraphs we refer to five related papers.

Secondly, we have edited the following part of our methods section in order to make it more explicit. It now reads:

“Building on the methods of \cite{bouvet2018use, mermoz2021continuous, ballere2021sar} we quantified shadow emergence through a radar change ratio (RCR).

Our approach differed in that we made use of multiple polarisations where previous studies tended to use VV only; analysed longer time series of images after disturbance events compared to previous work that prioritised near real-time capability; and adopted a purely pixel-based algorithm as opposed to reconstructing forest loss patches as objects.”

Here, we aim to be totally transparent that our method is based on previous work, while also stressing three important ways in which our approach is different. We then provide equations so that there can be no doubt to the exact algorithm we apply. We assume that you are not asking us to include the full algorithms of previous papers, as this information can easily be found by following the references.

Thirdly, we would like to highlight the following sentence from the abstract:

“By applying our method to 3 years’ worth of imagery over Gabon, we produce the first national scale map of small-magnitude canopy cover loss.”

To the best of our knowledge, this is the first time that gross canopy cover loss (as opposed to area of deforestation) has been quantified over an entire country. It was only possible to attempt such a map because of the uniqueness of our field data and the improvements we made to Sentinel-1 processing algorithms.

Again, we feel that this aspect of novelty is already made explicit in the abstract, and that sufficient comparison is made to previous, related products: indeed there is an entire subsection of the results devoted to such comparisons (3.6).

Reviewer 2 Report

The article deals with a very important topic related to deforestation of forest areas. This process occurs in different parts of the world with varying intensity, contributing negatively to the intensifiacation of climate changes.

For this reason, an important issue is the monitoring of changes in the forest environment that allows the determination of the size of the harvesting work carried out. This task can be performed with the use of modern technologies based on the analysis of satellite data. The technology used allowed for the capture of canopy forest gaps as small as 0.02 ha.

The article was written correctly. The Introduction presents the background of the issue using 46 references. This chapter ends with assumed hypotheses and a clearly defined goal of the work. The other parts of the article also do not raise any major objections. The methodological part, the results and the discussion are properly presented. Personally I prefer not to use subchapters  in the Discussion and this part of the article should  be carried out on the relatively general level. Conclusions respond to the assumed hypotheses while referring to the achieved results.

Detailed comments.

1.       The article should be written impersonally avoiding the words "we", "our" etc.

2.       Please check all units as there are ones that there is a space missing between number and unit. E.g. not 10m, but should be 10 m.

3.       Line 45. Please avoid statements like "we belive".

4.       In References, according to the editorial requirements, please use the journal abbreviations.

Author Response

Dear reviewer 2,

Thank you for your comments and also your patience: our response was delayed due to the lead author being in Gabon collecting more LiDAR data.

Please see italic text below for our specific responses to your feedback.

1. The article should be written impersonally avoiding the words "we", "our" etc.

While we fully respect your preference for impersonal language, we prefer to write in this style as we find it easier to read. Use of the word “we” appears to be common in MDPI Remote Sensing (e.g. we find tens of instances in all of the latest four articles) and other journals actively encourage this kind of scientific writing (e.g. Nature, see paragraph 4 on https://www.nature.com/nature-portfolio/for-authors/write).

2. Please check all units as there are ones that there is a space missing between number and unit. E.g. not 10m, but should be 10 m.

Thank you for spotting this detail. We have checked throughout the paper and believe this has now been fixed.

3. Line 45. Please avoid statements like "we believe".

We have changed this sentence to read “The Sentinel-1 (S-1) mission funded by the European Union and operated by the European Space Agency (ESA) is a promising tool for quantifying pantropical degradation as it provides regular, reliable and high resolution imagery over the entire tropical landmass.”

In line with our response to point #1, however, we have not changed the overall style of the article.

4. In References, according to the editorial requirements, please use the journal abbreviations

Thank you again for your attention to detail. We have used the MDPI Latex Template which includes the correct bibliography style, but unfortunately we do not understand why the journal abbreviations are not appearing. We will raise this with the editor to ensure it is fixed before publication.

Reviewer 3 Report

The paper from Carstairs et al. aims at mapping forest degradation using Sentinel-1 data, with a minimum mapping unit of 0.02ha. There is an important need of automatic and accurate forest degradation detection, and very little associated methods. This paper is thus potentially very important and requires a careful review.

 

I found the paper remarkable, i.e., clearly written and presenting a very simple and efficient detection method. Using lidar for validation is great and separating anthropic from natural disturbances is crucial. The previous works in this field are also well described, and commented in a diplomatic manner. Using the Fodex method calibrated in a very small area to map the whole Gabon, without any further validation, is a strong limitation, but I would have done the same.

 

This paper should be published but requires important clarifications before. Please find below major and minor comments.

 

Major

 

L259-L267: I attached a figure to this review, please look at it. In the situation shown in the figure attached, do you consider that Fodex accurately detected the canopy loss area detected using lidar? If yes, this is a serious concern that needs to be addressed. You should define a clear rule, such as: we consider our area (thus, object) as accurately detected when at least 50% of this area is detected using lidar.  And in this case, you have to recompute all your validation results.

 

I am also wondering why there is no pixel-based validation (e.g., following the Olofsson’s methods), especially given the fact that you corrected geometric shifts. You must fix that as it would be of great value. You use pixels for the 1ha and 5ha validations, but it is not the same at all (you may have the same amount of detected pixels at different places).

 

 

Minor

 

- L3: “Not yet available” please add “at this scale”

- L16: Please correct: “We produced the first national scale map”

- L104: “The principAl limitation

- L105-115: I disagree, the 5m resolution Planet data may potentially allow finer detection that Sentinel-1 data

- Figure 2 at the top: When reading the paper and trying to understand the whole story, I realized that some zooms of the figure would help the ready to figure out how the detected areas look like. Please add.

- The radar change ratio (RCR) has been defined in previous paper, for example in equation 1 in Tanase et al. (2018). Detection of windthrows and insect outbreaks by L-band SAR. RSE. You can’t say you are using RCR when using – (minus) RCR. Please correct

- Equation 1: You should add “max” just before RVRvh for clarity. Please explain why you are using sigma instead of gamma?

- Equation 2: I am surprised that you are using the same image for computing the mean backscatter value before and after this image (see the same t in the two sums).

- L206: please precise that alpha<=1

- If the aim of Equation 3 is only to show that you are merging at least two pixels to define a disturbed area, then it is really unnecessary and confusing. In addition you probably lean <=2 instead of <2? Please remove Equation 3 anyway.

- I really don’t know why you selected the ESA’s worldcover product? AGEOS has a great forest non forest mask for example. And there is no definition of forest at all in the paper, despite the mapping of the whole Gabon. Please comment.

- In the paper, please indicate the theoretical number of looks for each for the configurations. This is very easy to add and useful to the reader.

- L282-L287: I regret the lack of physical explanations there, and the purely statistical and rough bias correction. The point is that Fodex overestimates lidar based areas, and it would good to understand why and fix this in further studies.

- Figure 3: I am wondering why the false alarm rate paradoxically increases when n>25. Any ideas? Sorry if I missed something.

- Figure 5 caption: please rephrase, because the equation allowed for the calculus of the conversion factor, but is not the conversion factor.

Table2: How did you get these numbers ? Pixel-counting, or rather using the 1ha/5ha windows? Sorry if I missed something.

L436: Six months later??? It is very surprising.

L506: Not only at all. It is because the methods provide better results with higher MMU. And 5m resolution Planet data were used in previous studies.  

L517: But this is expensive and complicated.

L522: I am not sure to understand the word “filtering” here, please rephrase clearly.

L536-L546: You provide here very important information that is not reflected in the abstract. Please add 1 ou 2 sentences in the abstract.

L559: more false alarms but maybe more detections as well ? Trying to use both ascending and descending pass would be great!

L580-581: How do you know that?

L533: So there is 23% of areas detected by RADD that you do not detect?

 

@font-face {font-family:"Cambria Math"; panose-1:2 4 5 3 5 4 6 3 2 4; mso-font-charset:0; mso-generic-font-family:roman; mso-font-pitch:variable; mso-font-signature:3 0 0 0 1 0;}@font-face {font-family:Calibri; panose-1:2 15 5 2 2 2 4 3 2 4; mso-font-charset:0; mso-generic-font-family:swiss; mso-font-pitch:variable; mso-font-signature:-469750017 -1073732485 9 0 511 0;}p.MsoNormal, li.MsoNormal, div.MsoNormal {mso-style-unhide:no; mso-style-qformat:yes; mso-style-parent:""; margin:0cm; mso-pagination:widow-orphan; font-size:12.0pt; font-family:"Calibri",sans-serif; mso-ascii-font-family:Calibri; mso-ascii-theme-font:minor-latin; mso-fareast-font-family:Calibri; mso-fareast-theme-font:minor-latin; mso-hansi-font-family:Calibri; mso-hansi-theme-font:minor-latin; mso-bidi-font-family:"Times New Roman"; mso-bidi-theme-font:minor-bidi; mso-fareast-language:EN-US;}.MsoChpDefault {mso-style-type:export-only; mso-default-props:yes; font-family:"Calibri",sans-serif; mso-ascii-font-family:Calibri; mso-ascii-theme-font:minor-latin; mso-fareast-font-family:Calibri; mso-fareast-theme-font:minor-latin; mso-hansi-font-family:Calibri; mso-hansi-theme-font:minor-latin; mso-bidi-font-family:"Times New Roman"; mso-bidi-theme-font:minor-bidi; mso-fareast-language:EN-US;}div.WordSection1 {page:WordSection1;}

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

The article proposes a new algorithm for estimation of selective logging. The authors have done a lot of painstaking work and described the results of this work well. These results can be used to identify illegal logging or to refine forest biomass in carbon flux estimates.

Author Response

Dear reviewer 2,

Thank you for your comments and also your patience: our response was delayed due to the lead author being in Gabon collecting more LiDAR data.

We are very encouraged to receive your positive feedback, and believe there were no specific concerns to address in your review.

Reviewer 5 Report

Dear Authors,

  the paper is good and it is well-written and well-organised. The methods used as well as the data processing are clear and robust. My only comment is related to the Introduction. It could be improved with a little bit more background literature related to the topic. I would suggest to include the following two papers as an example: i) https://agritrop.cirad.fr/592787/1/Mercieretal_2019.pdf; ii) https://eo4society.esa.int/wp-content/uploads/2021/06/SOFT_FR_v2.0.pdf

Based on this I would suggest the publication after a minor revision.

Author Response

Dear reviewer 5,

Thank you for your comments and also your patience: our response was delayed due to the lead author being in Gabon collecting more LiDAR data.

We are very encouraged to receive your positive feedback, and very much appreciate your suggestions regarding the introduction.

The report on the SOFT project (https://eo4society.esa.int/wp-content/uploads/2021/06/SOFT_FR_v2.0.pdf) is certainly relevant and we were very interested to read it. We have added a reference to this document on line 94.

The second paper you mentioned was Evaluation of Sentinel-1 and 2 Time Series for Land Cover Classification of Forest–Agriculture Mosaics in Temperate and Tropical Landscapes by Mercier et al. We took from this that you wanted to see a mention of other applications of Sentinel-1 data in tropical forests besides degradation monitoring. In response, we have added a sentence to the introduction (line 65) that references the Mercier paper alongside a few others to provide this context.

Round 2

Reviewer 1 Report

The paper could be accepted. 

Author Response

Thank you again for taking the time to review our paper.