Automatic Detection of Marine Litter: A General Framework to Leverage Synthetic Data
Round 1
Reviewer 1 Report
In this manuscript, the authors Nagy et al. propose a more advanced simulation data training algorithm for marine litter monitoring based on satellite remote sensing data, which solves the current problem caused by the lack of field measurement data for automatic plastic litter detection. From a technical point of view, the authors are relatively fair in their analysis of the strengths and weaknesses of the algorithms and frameworks they have developed. However, there is one point of view that is worth discussing with the authors: effective monitoring of marine litter is ultimately for marine litter management, so in situ measurement and management is also encouraged in the future - it may be expensive or time-consuming - but it is necessary for the actual solution of the plastic litter problem. Therefore, from the introduction of the idea and the subsequent discussion of the methodology, I suggest that the authors can make some discussion in this regard, and should encourage the expansion of the field measurement dataset, as well as the algorithm enhancement, and the combination of the two, which can eventually lead to the subsequent management of plastic litter.
Overall, this is interesting, informative and generally well-written and structured.
Before the manuscript is accepted, there are several questions that need to be responded to by the author.
1. Please add the phrase represented by ML in line 15, machine learning?
2. Several recently published important references in the field are missing, such as:
1) Z. Yang et al., UAV remote sensing applications in marine monitoring: Knowledge visualization and review. Science of The Total Environment 838, 155939 (2022).
2) Taggio, et al., A Combination of Machine Learning Algorithms for Marine Plastic Litter Detection Exploiting Hyperspectral PRISMA Data. Remote Sens. 2022, 14(15) 3) Goddijn-Murphy, et al., Using a UAV Thermal Infrared Camera for Monitoring Floating Marine Plastic Litter. Remote Sens. 2022, 14(13) 4) Cao and Huang, A coarse-to-fine weakly supervised learning method for green plastic cover segmentation using high-resolution remote sensing images, ISPRS Journal of Photogrammetry and Remote Sensing. 2022, 188, 157-176.
-- These two papers provide a good introduction to the application of remote sensing (satellite or UAV remote sensing) in plastic pollution monitoring. They will be very helpful for the introduction of topics or the discussion of arguments in this manuscript.
3. In lines 63-65, "To date, only---", please do not use such an absolute tone when a comprehensive literature search has not been performed, and please revise this sentence.
4. Lines 84-86, "---is not a sustainable solution,---", it is still very necessary to add more measurement activities in the future for plastic waste estimation and management, so it is enough to talk about this method being expensive and time-consuming here, not to say it is “not a sustainable solution”.
5. In the discussion section, please add a section for future monitoring and management of marine litter based on the advantages and applications of the algorithm or framework.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper proposes a methodology to leverage synthetic data to automatically detect marine litter from remote sensing images. This addresses the critical issue of lack of training data for automated marine litter detection from remote sensing images. The authors then apply the method on Sentinel 2 images, showing the results of the application of the approach in “real” images. This is a very relevant topic and the document is quite detailed regarding the method and the code is published online aiding researchers who want to make use of the method. Some overall remarks and details to be addressed below:
Overall remarks:
It is easy to get lost in the paper (e.g. there are 2 sub sections named with data and classifier). The sections and sub-sections titles should be more related to the specifics of their contents in a bid to contextualizing them in the document and aid a potential reader navigate the paper. I would also propose at the end of the introduction to detail how the paper is organized, how the sections connect with each other and their contents.
Miss a more nuanced view regarding the resolution of Sentinel imagery. A lot of studies have been using UAV and aerial imagery to map litter. While these platforms do not have the areal coverage or economic benefit of Sentinel based approaches, these also have higher resolution. This should at least be addressed in the introduction, even if briefly.
Section 1.3 should be extended, given this being a Remote Sensing journal, the reader may not be familiarized with the generation of synthetic data. Hence, a more comprehensive review of the matter should be performed. Did any studies have already use synthetic data for remote sensing purposes? How does that relate with the authors work? An overview of the literature regarding synthetic data generation should be added. Moreover, without such review is difficult to position the paper within the literature. Then in the discussion the authors can come back to this section and contextualize the findings within the current literature (which is also missing).
Other remarks:
Line 109 and 110 : “close enough”, “reasonable confidence”, avoid wording like this; e.g. close enough may vary regarding the reader. There a few throughout the paper.
line 145: “some inacuracies” see above
Figure 1 should appear after the text which addresses it, the same for Figure 2.
Line 269: I would put the link to the code as a citation
Line 332: link to figure broken.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Dear Editor and Authors, the revision was made with scrupulous attention to my previous review. I believe the paper can now be accepted for publication. Congrats on the work.
