Drone-Based Characterization of Seagrass Habitats in the Tropical Waters of Zanzibar

Round 1

Reviewer 1 Report

Dear authors,

I read carefully the submitted manuscript on the use of UAS for seagrass mapping. I find it well structured overall but some parts are missing or can be improved. Below in the bullets you can find my comments and suggestions.

• How Sentinel 2 has been used to map SAV and identify sites? No information is available and is needed to have a separate chapter on that. Others have work in nearby region as well (https://zslpublications.onlinelibrary.wiley.com/doi/full/10.1002/rse2.187).
• the SDB product is open access? Any information on methodology and accuracy? the cited papers are generic papers on RTM and not on  DHI GRAS product methodology.
• line 179 - Camera model model PDC-1 250MP - maybe 25mp?
• A suggestion from my side is to make all field data and final products as open access in Zenodo or other repositories for the sake of further developments on the domain and to support other multiscale approaches in the data poor region.
• ML classification is quite basic - what about move the segmenteted layer from ArcGIS to EnMAP plugin in the open access GIS app QGIS to access advanced machine learning approaches like Random Forests or Support Vector Machines (see https://www.sciencedirect.com/science/article/pii/S0303243418311656). The OA values are not bad but a comparison will provide insights on the usefullness of advanced algorithms in UAS data analysis (if any).
• Figure 7 and 8. What about using colors instead of patterns to seperate %SAVcover and species?
• Figure 15. The legent talks about "Relationship between patchiness/seagrass cover with sea urchin abundance" but no relation (sensu regression plot) is shown.
• You can also read that work https://www.unep.org/resources/report/out-blue-value-seagrasses-environment-and-people which is a diluton of the current knowledge in a simple way

Best regards

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper detailed describe seagrass distribution around the zanzibar islands. The contents of paper can interest scientists who do researches on seagrass. The paper can be accepted with minor revisions, some details were shown as followings:

• Seagrass mapping with drone-based RGB map should consider geometry correction, how did the author do the work? The content of geometry correction cannot be found in the paper.
• What type of drone did the author use in the paper? Rotor drone or fixed wing drone?
• What type of weather when drone work? which will affect seagrass detection accuracy.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Review for Manuscript: remotesensing-1545586

 

The authors present a novel approach for small scale mapping of seagrass habitats by using drone and in-situ imagery. This study suggests that high resolution proximal sensing data are valuable in evaluation of ecological aspects at landscape scale. The presented approach is scientifically sound however there are a few things missing and thus a minor review is suggested. Particularly there are some additions that would increase the overall quality of this manuscript. Please consider the comments below for further details.    

 

Introduction

Please mention the major limitations occurring in current seagrass mapping studies. This will further assist with making the scope of this study more clear.

Please include a small paragraph with information about earlier studies using OBIA and/or drone/satellite imagery for mapping benthic habitats. Example literature:

Fallati, L., Saponari, L., Savini, A., Marchese, F., Corselli, C., Galli, P., 2020. Multi-Temporal UAV Data and Object-Based Image Analysis (OBIA) for Estimation of Substrate Changes in a Post-Bleaching Scenario on a Maldivian Reef. Remote Sens. 12, 2093. https://doi.org/10.3390/rs12132093

Lyons, M., Phinn, S., Roelfsema, C., 2011. Integrating Quickbird Multi-Spectral Satellite and Field Data: Mapping Bathymetry, Seagrass Cover, Seagrass Species and Change in Moreton Bay, Australia in 2004 and 2007. Remote Sens. 3, 42–64. https://doi.org/10.3390/rs3010042

Román, A., Tovar-Sánchez, A., Olivé, I., Navarro, G., 2021. Using a UAV-Mounted Multispectral Camera for the Monitoring of Marine Macrophytes. Front. Mar. Sci. 8, 722698. https://doi.org/10.3389/fmars.2021.722698

Zhang, C., 2015. Applying data fusion techniques for benthic habitat mapping and monitoring in a coral reef ecosystem. ISPRS J. Photogramm. Remote Sens. 104, 213–223. https://doi.org/10.1016/j.isprsjprs.2014.06.005

 

 

 

 

 

Methods

Line 95: What do you mean with “1m and 10m”? It is unclear.

Paragraph 2.2.4: Please provide some quality metrics about the SDB, i.e.: mean absolute error and bias.

Please mention which image/landscape features were used for classification and provide a justification about selecting those. Ideally provide a list with these parameters.

Please consider to make drone and in-situ data openly available. This would result in a greater impact of your study and increase its visibility.

 

 

Results

Figure 4 is quite large and supports only visualization of species, therefore it should go in the Appendix.

Figure 5: Please provide an explanation of the enlarged areas within the red triangles on a,b,c. It should be clear to the audience why these enlargements are important on the figure. The sub-plots of the enlargements should also have an identification letter/number.

Figure 6 is quite unclear and confusing. Is there a linkage between a and b? If yes then it is not clear. An alternative figure could be presented based on the suggestion below:

One important aspect of this study is to examine the between- and within-class variability.

Therefore it is expected that the authors should provide some plots showing the performance of classification accuracy between different benthic classes and seagrass species. In addition, a confusion matrix should be presented for each area showing the classification accuracy for each species and each benthic class separately (in the Appendix).

Figure 11: Please provide more informative titles and units for the Y-axes in a and b.

Figure 12: The entire figure is confusing and the caption is unclear. Please consider another type of figure (e.g.: a diagram) for illustrating benthic patchiness.

Line 573: please introduce and explain MDS.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

This manuscript uses drone-based imagery and other methods to document the spatial distribution of seagrasses and other response variables in the tropical water of Zanzibar. The manuscript is quite well written, but it is way too long, contains repetition and unnecessary detail and has too many tables and figures. The whole thing would be better received by readers if the text was reduced by 30% and limited to six figures and six tables. My most important comment is - make it significantly shorter, please.

 

Line 10: Delete “around the globe” as it is not necessary.

 

Line 21: You only need to mention that seagrass cover was negatively correlated to urchin density.

 

Line 23: ANOSIM is a hypothesis testing tool. What were your apriori hypotheses? If you were just looking for clusters, other tools are more appropriate.

 

Line 47: Delete “mechanical”

 

Line 50: Overexploitation of what resources?

 

Line 79: Avoid the confusion and just call them drones. https://doi.org/10.1139/juvs-2014-0009

 

Line 141: As there were no accurate (XYZ) ground control points or an RTK or a PPK drone used, all the maps were not orthorectified. It is probably OK for a one-off sampling, but this limitation needs discussion. You will not be able to sensibly compare between times.

 

Section 2.1: How did you deal with glare issues, as you were mapping in NADIR with drone deploy.

Giles A, Davies JE, Ren K, Kelaher BP (2021) A deep learning algorithm to detect and classify sun glint from high-resolution aerial imagery over water. Journal of Photogrammetry and Remote Sensing 181: 20–26

 

Table 2: Resolution need units (e.g. cm per pixel)

 

Line 192: When was the bathymetry data collected? If it was not recent, it is not clear how relevant it is to the seagrass distributions. Depth in these habitats can vary significantly over time.

 

Multivariate analysis: If you were looking for clusters, you should do the MDS first. Perhaps then you could do pairwise testing using ANOSIM to look for significant differences.

 

Delete Figure 4 or move to supplementary materials. This information is freely available on google.

 

Move Table 5 to supplementary materials.

 

Line 411: remove “significantly”

 

The number of tables and figures should be reduced to 6 of each. In many cases, they are simply not needed in the main body of the text, as they are barely referred to. Move them to supplementary materials, if you like.

 

Line 543. It is not clear how this correlation was done? Were the replicates for each site combined. If so, why? Was the correlation a linear analysis, in which case the line on the graph should be straight? Here, also grazing impacts are assumed, but correlation does not equal causation. There could, for example, be fewer sea urchin predators in sparely populated seagrass. You should only discuss the relationship in the results section and leave speculation about the mechanism in the discussion.

 

Line 561-564. This is repetitive of the methods. Please remove.

 

Line 566. This result means there is significant variation among sites and not overall differences. Furthermore, the pairwise tests among 9 groups with few replicates are meaningless, as they need to be corrected for type 1 error. How did you get the replication? There should have just been one value per site. The depth analysis needed to be two-way analysis (e.g. PERMANOVA with an interaction term). You should really have someone with a good knowledge of PRIMER take a look at this and fix it for you. Figure 16 is not very professional. MDS plots can be made so much better than the basic graph turned out by PRIMER.

 

Section 4.12 You need to discuss the spatial limitations of your non-orthorectified maps. Your methods would be improved by using RTK or PPK drones over water and ground control points over land where possible. Similarly, the resolution of your cameras was pretty average, with cameras up to 100 MP available for drones. Similarly, the coverage could be improved by using fixed-wing rather than small quadcopter drones.

 

The discussion is really long and should be decreased by 30% with many of the sub-headings removed.

 

Line 841: What about aerial photos from crewed aircraft? These have been used for years to map seagrasses. Companies like NEARMAP make it easy (but expensive) to access large scale orthomosiac imagery with 6-7 cm captured every few months. While it is not available in Zanzibar, it is being used in other places like Australia and US.

 

Conclusion: If the aim is to do ongoing monitoring of seagrass, you need to orthorectify your maps with ground control points, RTK, PPK or other enhanced GNSS systems.

 

Author Response

Please see the attachement.

Author Response File: Author Response.docx