Advancing Floating Macroplastic Detection from Space Using Experimental Hyperspectral Imagery

Round 1

Reviewer 1 Report

The present manuscript provides a new hyperspectral image database of virgin macroplastics and vegetation. Based on a Linear Discriminant Analysis (LDA), the authors indicate which wavelengths can contribute to discrimination between water and plastics, and plastics and vegetation. They also compare LDA results with products provided by existing satellite sensors such as Sentinel-2 and Worldview-3. Additionally, they evaluate NDVI and FDI indices to determine their drawbacks and how they could be improved. The topic is very interesting, and the manuscript matches the scope of special issue “Remote Sensing of Plastic Pollution”. The manuscript is well-organized and clear. However, it is recommended that authors provide access to the supplementary material. I have added the following comments as well:

line 121: Which were the seven polymer types? Which polymer type was the most frequent? In which river did the authors harvest the litter?

line 175: Were there any limitations (e.g., light conditions) or camera errors?

line 329: Although for NIR-SWIR there is an extended analysis, authors do not mention about the spectral behavior in VIS-NIR. Are there any particular spectral patterns due to different polymer types or different colored plastics in range VIS-NIR?

line 418: In which figure are these values depicted?

line 420: Authors perhaps mean floating object and water?

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear Editor and Authors, the paper is of interest of RS readers. It is well written and structured. Introduction presents the state-of-art optimally, Methods are clearly described, Results presented in a proper way.

I do not have specific request, only some (minor) suggestions.

1) Discussion (section 4) underlines the importance of the study, just repeating the state-of-art already presented in Introduction. Yet, comparisons with previous works are commented just from the methodological point of view, whereas it would be also of interest to read some comparison of results (spectral and/or indices).

2) I feel it is missing a paragraph regarding the limitations of the study. It is (rightly and fairly) stated that virgin materials were used for the experiment. It would be of interest to read some comments. For instance, in their experience/opinion, if and how the spectral responce may vary on degraded plastic.

 

Congrats on the work!

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors collected and analyzed a hyperspectral database for plastic, vegetation, and water, identifying which wavelengths are most efficient in discriminate those targets.  Also, a comparison based on Sentinel-2 and WordView-3 bands was carried out. In general, the article is well-written, emphasizing the advances, challenges, and efforts in monitoring macroplastic pollution using RS techniques. The research design is appropriate, and the methods are adequately described, allowing reproduction of the work. Additionally, the authors present a hyperspectral library for plastic samples acquired in a controlled environment.

Following, I have some punctual comments, suggestions, and questions:

Title: I suggest rethinking the title. The authors provided a comparison between hyperspectral data and orbital sensor bands, considered as an experimental design. Also, I do not agree that the authors provided a validation once the work doesn’t carry out an analysis using orbital images (Objective iii), either considered differences in observation-illumination angles (or BRDF effects, etc.) in satellite products.

I missed some results in the abstract. The authors focused mainly in describe what they did.

Line 18. I suggest writing “visible” instead of “visual”; and in the complete document;

Line 24. I suggest writing “Normalized” instead of “Normalised”; and in the complete document. NDVI is well-known as Normalized.

Line 100. I missed information on why to choose WorldView-3 and Sentinel-2;

Line 104. It would be interesting if the authors provided a figure workflow.

Figure 1. I suggest writing a box for “non-plastic”, as well as the authors did in the polymer categories.

Lines 131-133. How many samples for vegetation? Did was collected only one sample for water?

Line 198. And the water?

Line 199. It's confusing. The authors selected representative pixels manually, and the total of pixels was 20-95 million per category (line 206)?

Line 203. The authors considered the class “all plastic”, but there are also “floating items” in the water analysis, and “plastics” in the vegetation analysis. It's confusing because the differences are not completely clear.

Equation 6 and 7. The authors applied the spectral indices in hyperspectral images. I suggest specifying the bands in general instead of Sentinel-2 name bands. E.g., red and NIR, instead of B4 and B8. Moreover, the authors did not apply NDVI and FDI in Sentinel-2 and WorldView images.

Line 255. I suggest describing Worldview-3 MS-VNIR & MS-SWIR abbreviations.

Figure 3. Why vegetation has negative values and values above 1? If authors considered a reflectance factor, the values must be in a range of 0-1. The doubt is applied in the other spectral signatures. Also, the reflectance of water in the visible is extremely high (almost 1) considering the usual spectral signature.

Figure 4. I suggest overlaying the band's color in the gray color (no-data) to better visualize each bandwidth. Also, indicate what means the blue color.

Figure 6. Does not NDVI range shows negative values, as Figure 7?

Line 472 – 473. Compared with what?

Lines 542-550. Perhaps authors could cite information about sub-pixel analysis by using spectral mixing models.

Also, did the authors considered the possibility to calculate the mean of hyperspectral data in the same bandwidth range of each orbital sensor band (Sentinel-2 and WorldView)?

Line 502-584. The authors only compared NDVI and FDI. So, the work does not support this conclusion.

Line 602. The link is not working.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear Editor and Authors,

The manuscript has been sufficiently improved, and the authors have addressed all comments. I am recommending this manuscript be accepted and published.

Author Response

Thank you for your review and recommendation.

Reviewer 3 Report

The authors clarified all my remarks, and improved the readability and scientific quality of the manuscript. 
However, I still have some remarks that must to be emphasized in the manuscript:

Objective III. I suggest to shift "validation" to "comparison" considering all noted I mentioned in the first revision. 

Line 213-216. The authors explained to me, but not included information in the MS about "pixels were sampled by hand, using a paintbrush in the MATLAB software, ranging from 3x3 pixels all the way to 41x41 pixels". I suggest to write a brief information in the MS. Also, maintain 'manually" instead of "by hand". 

Figure 4. I still missing a specific explanation in the MS about the negative and above 1 values observed in the spectral signatures standard deviation. I believe that this statement must to be clarified in the MS to avoid misunderstandings.

Author Response

Please see the attachment

Author Response File: Author Response.pdf