Quantifying the Potential Contribution of Submerged Aquatic Vegetation to Coastal Carbon Capture in a Delta System from Field and Landsat 8/9-Operational Land Imager (OLI) Data with Deep Convolutional Neural Network
Round 1
Reviewer 1 Report (Previous Reviewer 2)
Thank you for revising the manuscript and addressing most of my comments. There is still minor revision needed before it is ready for publication. Please address following comments:
1) Figure 2 says it is Sentinel 2 MSI. However, this is Landsat 8. Please revise.
2) Throughout the revised manuscript, I see Figures and Tables are written twice. Please correct it.
3) Table1: Why there are two types of products mentioned here? One is L1TP and other is L2SP. Which one was actually used? For October 15, 2022 the satellite sensor mentioned is Landsat 9. However, the product name suggest it is Landsat 8. Did you use Landsat 9 or Landsat 8? Also, for each year you only used one satellite scene. Why two satellite scene for 2022? If there were both Landsat 8 and Landsat 9 were available for 2022 then why did you use Landsat 9? It would be better to use Landsat 8 to be consistent with other years.
Finally, I have a major concern regarding authors justification for significant reduction is SAV during 2019 because of heavy flooding. It might be possible that the satellite scene used for classification was highly turbid and hence the model was not able to capture the SAV full extent. There was lack of cloud-free image for next year and just using one scene per year and drawing a conclusion that Hurricane and Flooding reduced SAV coverage is not convincing. I would suggest to mention some of these limitations and be careful to draw such conclusion based on limited satellite images.
Author Response
Please see the attachment, thanks!
Author Response File: 
Author Response.pdf
Reviewer 2 Report (New Reviewer)
The authors used deep convolutional neural network (DCNN) techniques to assess the carbon sequestration of Submerged aquatic vegetation (SAV) habitats in the Atchafalaya River Delta. Additionally, the authors have estimated carbon fluxes, net greenhouse gas (GHG) sink potential of SAV habitats and showed the impact of Hurricanes on the SAV. The methodology was established in a good way, but there might be questions about not including an air-sea gas exchange in the Carbon balance model. The application of convolutional neural networks in SAV identification is interesting, although the field observations and literature-based data are not sufficient for training the WSAV-Net model. The tsunami impacts could be easily observed from the habitat maps provided by the authors. The effect of SAV was established with the SAV scenario and without the SAV scenario. The authors could compare the results obtained from the study with similar studies elsewhere. The manuscript is acceptable, but the authors need to clarify the comments below:
Comments:
1. The authors have mentioned using 0.25 m2 quadrats for SAV mapping in this study. The 0.25 m2 quadrats are too small, considering Landsat data with 30 m accuracy has been used to classify SAV.
2. In supplement Table S2, there is a habitat named Brackish Marsh which is not present in Figure 3b. Why brackish marsh has not been considered in training samples?
3. In lines 182-186, the authors mentioned using six Landsat bands with NDVI and MNDWI, which makes the total band number 8, not 7. But the authors described it as a 7-band composite image. Please clarify.
4. The authors are requested to elaborate on the remote sensing data preprocessing steps and mention the classification scheme used for generating Figure 3b. A flowchart is often a good idea for representing the methods.
5. The authors have mentioned, “Subsequently, CH4 emissions were converted to their carbon dioxide equivalent (CO2e) by multiplying by a Global Warming Potential (GWP) index value of 25”. The authors are requested to mention the data source of CH4 emission as one can argue that some amount of CO2 and CH4 is mixed into the air via interaction with the seawater (air-sea exchange), which is not reflected in equation 2.
6. In Table 3, there are three data mentioned about some lakes and wetlands in China. These three data are the only source of GHG fluxes. How do Chinese GHG fluxes apply to Atchafalaya River Delta? The authors are requested to clarify the issue.
7. The data sources are very limited, as seen in Table 3, which makes the argument whether only two or three data sources are enough to represent the whole study area.
8. The figures and table numbers are mentioned twice in the manuscript.
Author Response
Please see the attachment, thanks!
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
The study is highly fascinating and seems almost ready for publication pending a few minor adjustments. However, the authors will need to invest some effort in completing the necessary tasks. Furthermore, certain aspects require additional attention and revision.
1. Line 41 and 280, please add the citation
2. Enhance the resolution of Figure 3 to improve its visual clarity and detail.
3. Line 223, Habitat % cover I, please rewrite this term.
4. Line 292, please summarize the equation variables
5. Could you kindly include Figure 4, which shows the learning curve for the training and validation accuracy?
6. Please remove the years "2015, 2016, 2017, 2019, 2021, and 2022" from line 429 of the text.
7. In line 597, the scale of 000’s of hectares, what do you mean?
8. Could you please provide a brief summary of the conclusion section? It appears to be quite lengthy.
9. Please revise the list of references by removing outdated citations and prioritizing the latest and most relevant sources.
10. Kindly edit the manuscript's English language to enhance its coherence and readability.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Overall Comment:
It was great to read this paper focused on quantifying the contribution of submerged aquatic vegetation (SAV) to carbon sequestration. Authors have found significant impact associated with passage of Hurricanes Barry (2019) and Ida (2021) with SAV and net GHG flux which was very interesting to me. However, I don’t think using just one scene per year would be sufficient for comparison where seasonality can play a major role in variability of SAV. Authors also did not include the satellite image classified maps for 2018 whereas they have mentioned that the 2018 image was downloaded for this study. Moreover, I was surprised that the authors did not mention Hurricane Harvey that made landfall in August 2017 and a major flooding occurred post-hurricane landfall in southwest Louisiana. In addition, another Hurricane, Nate, made landfall in early October 2017. I was wondering if there was such a prolonged effect because of Hurricane Barrey that made landfall in 2019, then why not such an effect was visible in the 2017 classified map? Also, I was wondering that the authors have used Landsat 8 for 2015-2021 and Landsat 9 for 2022, was there any effect of sensor differences in classified maps, especially because all Landsat 9 scenes are being re-processed (USGS Earth Explorer). My concern is differences due to atmospheric correction. Lastly, I would suggest for future study to use harmonized Landsat-Sentinel 2 images to fill the data gap issue.
Specific Comments:
Title: Title is too long and can be shortened.
Abstract:
Lines 9-10: It is not clear which one is recognized as a valuable component in climate change..submerged aquatic vegetation or carbon sequestration? Need to revise the sentence.
Line 20: Using the acronym GHG without expanding, please expand.
Line 26: What is 000’s of hectares? Is this a typo?
Keywords: No need to include words already present in the title.
Introduction:
Line 101: “2022” is not the current year anymore.
Materials and Methods:
Line 106: Why not use square km instead of ha as standard unit?
Table 1: Why were those particular Landsat dates selected? I understand for 19th September 2015 and October 15, 2022 were closest to field sampling but what about other dates? What criteria was followed to select other dates?
Lines 408-409: Why not use Sentinel 2 imagery to fill the data gap? There are already harmonized Landsat-Sentinel 2 products available. Therefore, models developed for Landsat can be tested using Sentinel 2 data as well.
Figure 5: Please include the dates of these individual images. Also, it is not clear to me how one scene can represent one year? If there is seasonality involved in the variability then we cannot compare the end of October (30 October 2016) almost November with early September (5 September 2019) for example. They are almost 2-months apart if we consider monthly variability. Also, 2021 and 2022 maps were produced using Landsat 9 scenes. Do you think Landsat 8 and Landsat 9 might have some differences and effect on result because one is a relatively new sensor and currently all Landsat 9 data are being reprocessed (as message is appearing in EarthExplorer website)?
Figure 5: Please correct the typo in the legends “Intermediate mash”, “Brackish mash”. Also, is there any “Brackish marsh” available in the classified maps? I don’t see that color on any of the classified maps. If not, no need to include that legend.
Figure 5: Why is the 2018 map missing from this figure? I saw Landsat 8 scenes from 2018 included in Table 1.
Figure 6. According to Figure 5, there were more classes for SAV such as high, medium, and low. Did you combine all three together here? If so, why not produce a combined classified map as well where you combine all SAV (three classes) into one class and Marsh (three classes) into one class. This will help to relate Figure 6 with such a classified map. Again, I don’t see brackish marsh class here which was present in Figure 5 legend.
Figure 7. I am just wondering how accurate these classified maps are? Mainly because when I see the high resolution Google Earth image for this region and compare it with SAV and FAV maps, I suspect regions corresponding to SAV medium and FAV medium might be misclassified. Again, it’s difficult to compare when you have three subclasses within a broader class.
Figure 10: Again, I am wondering why 2018 is missing from this chart if you have processed the 2018 scene?
There was major flooding in Southwest Louisiana after Hurricane Harvey in August 2017. If there was significant loss of SAV after Hurricane Barrey then why not such an effect after Hurricane Harvey? Also, there was another Hurricane (Hurricane Nate) that made Landfall near the Louisiana coast in early October 2017. So, basically the Landsat image used for classification in 2017 was acquired after the passage of two hurricanes and we don’t see a significant effect on SAV as we see in 2019, why?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
This Manuscript presents interesting and well-elaborated research about the spatiotemporal variations in Submerged Aquatic Vegetation distribution and the potential role of Submerged Aquatic Vegetation in carbon sequestration in a subtropical shallow and fluvially dominated delta system in Louisiana, USA, using a novel combination of Deep Convolutional Neural Network techniques and field and remote sensing observations. Some technical issues were founded that needed to be corrected.
There are 186 listed references in the Reference section. However, only 122 appear in the text of the manuscript. Also, many error messages (Error! Reference source not found) appear in the text. Please make corrections.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The paper has improved significantly; however, it still needs some revisions in terms of the English language.
Thank you
Reviewer 2 Report
Thank you for responding to my comments and revising the manuscript. Though, I am not satisfied with the response regarding the satellite image dates used in this study. First of all, they excluded 2018 image from table (after my comment) citing the issue related to cloud in the image, fine!
Now, I checked other satellite image dates that they have used in this study. When I checked the first image date, i.e., September 19, 2015, what I found is there was no Landsat scene on that date which covered the study region completely. Only the partial study site was covered by Landsat scene on that date (see the Landsat true color images attached for that date). Then how come authors were able to produce the classified map for the entire study region?
Next, I was wondering significant reduction in SAV area on September 05, 2019, image which authors claimed is because of hurricane impact and they also responded to my comment that other hurricanes (Harvey and Nate) might not have such impact as Hurricane Barrey in 2019. However, when I checked the quality of satellite image for that particular date (September 05, 2019), that authors have used in classification, it was contaminated by cloud (marked in circle in attached document). Why did the author use such cloud-contaminated image? This can significantly affect the SAV area classification and hence the discussion and conclusion of study. I tried to see if there was any other cloud-free image available in 2019 or not, then I found there was a cloud free image available on October 23, 2019 (see in attached images). Then, why authors did not use this cloud-free image instead of cloudy September 05, 2019 image?
Again, when I checked image from September 10, 2021, which authors have used in classification had a thin cloud (marked in red circle below) over study region. From the same month, there was a complete cloud-free image was available on September 26, 2021 (see below). Why not to use that complete cloud-free image?
Authors have also mentioned in their response that when images were available in both September and October months, then they chose September month. Then why authors did not use September 28, 2016 image (see attached image), instead they used October 30, 2016 image in classification?
Finally, authors have mentioned the image date 23 September 2022 but there was no Landsat scene for that particular date for this study region. Please double check.
I would like to see how the results change when using a complete cloud-free images for 2019, 2021 and also need justification regarding images (image dates) used in classification for year 2015 and 2022.
Comments for author File: Comments.pdf
