DETER-R: An Operational Near-Real Time Tropical Forest Disturbance Warning System Based on Sentinel-1 Time Series Analysis

Round 1

Reviewer 1 Report

Overall, I believe that this is a good manuscript documenting a potentially impactful tool for tackling an important problem. The concept is clear and easy-to-follow, though on its own, I couldn't help feeling that the study didn't push scientific and technical boundaries as much as I might have hoped for in a journal article. That said, I did see that the technical details of the system were better described in previous work (reference 7 in particular) and I believe that this manuscript is overall worthy of publication.

I have a few major comments I'd request a response for, as well as some minor ones that wouldn't need a response if addressed as suggested.

It was nice to read an article that gives some hope to addressing the important problem of deforestation and uses SAR to do it, so I hope this work continues!

***

Major comments (response requested):

89: Specify how the conversion to decibel scale occurs: what's the baseline for 0 dB?

84-110: The significant trade-off for reducing false alarms in the manner described here isn't discussed at all: namely, that false negatives will also increase as the threshold is dropped. Considering that the goal is to prevent deforestation, this is potentially undesirable and perhaps even worse than false alarms. This is touched on much later on lines 415-417 and 425-437, but even there the authors don't really mention some of the other ways this problem can be tackled: for example, establishing a CFAR (constant false alarm rate) standard (often used for SAR-based maritime vessel detection), or using AI/ML to perform some pre-filtering that reduces false alarms without necessarily increase false negatives (alluded to in lines 448-450). The focus on keeping false positives down is understandable with respect to making sure that the detector remains useful for the field teams it is designed to support, but there should be a better discussion of what is potentially being sacrificed to achieve that (i.e., consequences of false negatives) and ways to mitigate it.

128 (Figure 1): Please describe how the test and validation AOIs were chosen? Are there any possible biases to the regions chosen: that is, are they typical of the rest of the Amazon, so that the algorithm will perform well elsewhere too? If this is described elsewhere in the text, I may have missed it: if so, please at least reference the text here.

218: Briefly describe the technique so that readers don't have to access the reference to understand how it's used here (they may not be able to)

268: The reasoning for limiting to 1 ha is only satisfactorily described much later on lines 417-420. It should be mentioned here to justify the limit of 1 ha. Since the reasons are stated to be cost and operations, I'd also request the authors discuss - either here or in the discussion - what the smallest MMU could technically be, and if this might enable earlier detection of deforestation if other resources were sufficient to enable quicker action.

384 (Table 2): The results seem good, but as a reader I felt I didn't have a good baseline to compare to. I'd request that the authors include in the discussion some mention of how their performance compares to similar systems, such as those in other countries as alluded to on lines 32-33 for Peru and Indonesia.

443: What about other systems such as Canada's RADARSAT Constellation Mission (available now) or TerraSAR-X / Tandem-X?

484-485: I'm not sure I agree that these algorithms need to "prove their suitability" any more than they already have. I think the fact that they aren't used more operationally is likely more a question of time needed to implement and deploy in an operational setting, and the accompanying investments, infrastructure, training, policies, etc. needed to accomplish that. I would leave this somewhat subjective - and, in my view, unsupported - assessment out and phrase this part and rephrase as "no NRT system has been able to apply these kind of techniques on an operational basis, although approaches based on deep learning algorithms have shown promising results [37]."

 

***

 

Minor edits (no response needed if corrected as suggested):

19: keywords missing

69: capitalize Bayesian

82: south Asia (no hyphen)

87: "an Amazon-wide" (an instead of a, capitalize Amazon)

111: Delete "In this context"

120: "4.21 million km^2"

123: The remaining area

124: "2.81 million km^2"

146-149: "During the first half of September 2021, issues with the image processing pipeline substantially reduced the availability of S1 images on the GEE platform. The images acquired during this period became available later in the same month."

150: "On December 23rd 2021, a power [...]"

153: Should say "May 2022" here instead of "May'21"

201: "The code used by DETER-R is open-source and can be downloaded at:"

206: Clarify here if the more complex procedure referenced here is actually being used, or if not that this is a limitation of the present work that might be addressed in the future.

208 (above, towards top of page): Add a reference that supports calling the despeckling procedure used here "standard"

381: Say "low" instead of "very low" ("very" is subjective)

384 (Table 2, caption): "parentheses" instead of "parenthesis"

413: "aim" instead of "aims"

435: "at the cost of a slight increase in the rate"

436: "such a process"

443: "scheduled for 2023"

471: "an additionally" doesn't work, would suggest "represents a 5.0% increase compared to optical alone."

472: "this figure increased to 8.1%"

474: "can be of the utmost importance, particularly in areas"

476: "which normally expand"

483: "low" instead of "very low"

Author Response

Thanks for your detailed review, it will certainly contribute to improve our paper. Please find attached our reply to your suggestions.

Author Response File: Author Response.pdf

Reviewer 2 Report

Protecting Brazil's forests and protecting the Amazon is equivalent to protecting the lungs of our planet. The manuscript is informative in that it uses Sentinel 1 time series data for monitoring forest degradation in the Amazon. However, this manuscript needs major revision in terms of writing as well as technical details.

1) The author need to think about whether the title of the paper needs to be changed? The paper uses not only Sentinel 1 SAR but also Sentinel 2 optical satellite data.

2）Lines 5-8，Please rewrite these sentences. Do not write them as products‘ manual, and write the full name the first time they appear.

3）Although it is stated in the abstract that the proposed method can monitor more forest area during the rainy season, how reliable and accurate are these data results? This needs to be made clear in the conclusion section.

4）Introduction: As an open access journal, it is recommended that the references are cited as up-to-date as possible. Could the authors briefly address the targets detecion by SAR, such as
Dynamic detection of offshore wind turbines by spatial machine learning from spaceborne synthetic aperture radar imagery, Journal of King Saud University - Computer and Information Sciences, 34(5): 1674–1686, 2022.

Forest and its detection: Jacob N M, Jincy L H, Abhiram B S, et al. Deforestation Detection using Geographical Information System//2022 Second International Conference on Artificial Intelligence and Smart Energy (ICAIS). IEEE, 2022: 892-896.

Nunes M H, Camargo J L C, Vincent G, et al. Forest fragmentation impacts the seasonality of Amazonian evergreen canopies. Nature communications, 2022, 13(1): 1-10.

Thank you.

5）Lines 103-111, has the DETER-R method already been published? If it has been published please provide full references. If the DETER-R method has not yet been published as a paper, please rework these two paragraphs.

6）FIgure 1: This image is not in clear font, clear provides a high resolution version.

7）FIgure 2: Figure 2 No avalible data of S1B starting in 2022, please explain.

8）Section 2.4: These validation methods are indirect and there is no validation information recorded from field site research. It would have been more convincing to have added validation information from field site research records.

9）Section 3. It is recommended that the method flow be presented in small paragraphs according to the steps drawn in Figure 3, so that the diagram and the content can be consistent and increase readability.

 

10）Page 7， Lines 207-208, speckle suppression has a significant impact on quantitative forest monitoring results. The averaged temporal filter lised in Equation 2 is unreasonable and it is recommended that the Lee filter be used instead.  In addition, It is suggested that the authors cite the latest reference  for Lee speckle: Jong-Sen Lee, Jen-Hung Wen, T. L. Ainsworth, Kun-Shan Chen and A. J. Chen, "Improved Sigma Filter for Speckle Filtering of SAR Imagery," IEEE Transactions on Geoscience and Remote Sensing, vol. 47, no. 1, pp. 202-213, Jan. 2009, doi: 10.1109/TGRS.2008.2002881. Thank you. 

11）Figure 4, the text in the middle plot is blurred and needs to be improved.

12）Section 3 does not provide an equation description of the proposed key monitoring algorithm, so it is suggested that several additional equations be added for DETER-R.

13）Figure 7,the text in the plots is blurred and also needs to be improved.

14) Figure 8 that makes it difficult to read and understand, so please improve.

15）Figure 9: The result shown looks like a young man's pimple, not beautiful, could it be improved?

16) Figures 10～Fig. 11: Could you clarify that the optical image is from that sensor, the result of the detection is from the SAR, and finally the two are superimposed and displayed together?

17) Section 5.1: Table 3, To clearly explain the different methods of comparison, add the appropriate references and explain why the different methods have different performance.

18）The conclusion section gives the impression that this manuscript is a product manual, so please revise and improve it carefully, thank you.

Author Response

Thanks for your detailed review, it will certainly contribute to improve our paper. Please find attached our reply to your suggestions.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Congratulations to the authors for improving the quality of their papers through revision.