Detecting the Spatial Variability of Seagrass Meadows and Their Consequences on Associated Macrofauna Benthic Activity Using Novel Drone Technology

Round 1

Reviewer 1 Report

The ms describes novel applications of RS observations of seagrass meadows, and macrofaunal burrowing activity. Overall, results are interesting and relevant to the journal, but some issues in my view require addressing before publication.

Main comments:

• You state that VIS+NIR results are more accurate than the VIS only, and yet you provide no proof in what way they are ‘better’. In fact, your performance analysis summarized in table 1 clearly indicates that both have the same classification accuracy, and paragraphs 4.1.2 (VIS) and 4.1.3 VIS+NIR are identical texts (remove one of them….) indicating that there is ‘zero’ difference between them (your purely statistical values of 90, 91 and 98 & accuracy (VIS) versus 92, 94 and 98 % (VIS+NIR) are (statistically) identical between the two methods). Yet, Figs 7 and Fig 8 indicate that there is a substantial difference in change detection across the sensors in all seasons.

 

• These differences must be addressed. Also, provide a critical appraisal of sensor performance, rather than just claiming that spectral resolution is more important than spatial resolution. This may well be so but you need to contrast your RS data against ground truthed data, and the results of relevant statistical analyses of the data needs to reflect this. As is, the analytical and statistical treatment does not demonstrate advantage of VIS+NIR over VIS, and I am eft with the question which one is actually correct (in Fig 8 for example)?

 

• You seem to have only one GCP as indicated in the figures? You spend a lot of time on statistics but do not discuss positional accuracy which is very much related to gcp density and distribution (you say you are highly accurate but provide no proof). In fact, panel d in Fig 7 shows a systematic offset between seasonal data sets which fairly clearly indicates a problem with positional accuracy in the order of 2m or so. This evidently is a problem in fine scale analyses and needs to be adressed, and included in your performance analysis.

 

 

Additional comments :

• In my view lots of somewhat over-complicated statistics used in order to provide a classification scheme that is not that difficult to obtain by simpler methods, but the analyses are ok. Given the complexity of the analyses, the interpretations are light weight.
• State which drone was used not only the camera. Positional accuracy depends amongst other things on the model.
• you talk about detecting macrofaunal communities – it is more correct to refer to benthic activity rather than communities (including in the title).
• You state 1.35 cm and 3 cm pixel resolution respectively but you detect feeding burrows. A priori it is a little surprising you can detect burrows at this spatial scale. You make no mention which size the burrows in your specific environment have but it seems to me you are operating close to detection limit. Comment on this, include flight altitude and other parameters playing a role here (on spatial resolution) and justify why you believe you capture the major bioturbation activity. This has important implications: maybe burrow size is just simply smaller in winter and you don’t detect it because of limitations in resolution?
• Include a scale in fig 4. Note that an L1 GNSS solution obtained with a mobile phone is extremely inaccurate (many meters of positional uncertainty) and thus not useful in the context of a study that has a vocation of high accuracy.
• Line 343 : you state you achieved a repeatable technique with high positional accuracy but you do not demonstrate this at all. Again, your data in Fig 7 panel d casts doubts on this.
• Line 300-301 : you state there is a direct relationship between seagrass density and macrofaunal activity but I can’t see this. more seagrass = more activity ok (and not surprising / well known), but you should discuss more the quantitative relationships (how much more), spatial distribution of activity (eg on the edge or in the middle of a new seagrass patch etc etc ). Why restricting yourself to average values when you have spatial information which is the large advantage of RS techniques?
• Section 4.3 is almost entirely a repeat of previous text and should really be removed, which leaves the ms with a short discussion. It is unfortunate that the authors don’t make more use of the spatial information their data contains.
• Line 449-450 : it is claimed that both sensors provide consistent results. I strongly disagree, fig 8 clearly shows important differences.
• As a general matter, review the text to reduce the many currently appesring repetitions.

Author Response

Dear Reviewer, 1,

Thank you for the comments and suggestions for the manuscript. Attached are the responses and changes based on your reviews. They were insightful in identifying the shortfalls which added value to this manuscript, and we would like to thank you for your time in reviewing this manuscript. 

Author Response File: Author Response.pdf

Reviewer 2 Report

The article takes an interesting approach to using drone imaging for seagrass mapping.
The methodology is clear and encourages replication in other coastal environments.

I emphasize the need to include some information:

Include the information: Which RPA was used?
Better describe the number of classes and training points (218 – 219 and 232-233).
Improve the description of sample size accuracy (222 – 223).

Author Response

Dear Reviewer, 2,
Thank you for the comments and suggestions for the manuscript. Attached are the responses and changes based on your reviews. They were insightful in identifying the shortfalls which added value to this manuscript, and we would like to thank you for your time in reviewing this manuscript. 

Author Response File: Author Response.pdf

Reviewer 3 Report

In this study, the authors investigated  and tested the application possibilities of RPAS VIS and VIS+NIR sensors to detect fine-scale seasonal seagrass changes in a dynamic nearshore marine environment using several spectral indices and SVM supervised classification technique. In addition, the effects of seagrass gain and loss on the abundance and distribution of macrofauna communities was investigated.  Both the methodology used and the research results are noteworthy and worthy of publication. The comments and questions can be found in the attached file.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer, 3,
Thank you for the comments and suggestions for the manuscript. Attached are the responses and changes based on your reviews. They were insightful in identifying the shortfalls which added value to this manuscript, and we would like to thank you for your time in reviewing this manuscript.