How Much Shallow Coral Habitat Is There on the Great Barrier Reef?

Round 1

Reviewer 1 Report

Overall, this is a well-written manuscript describing an advance in reef mapping techniques that will benefit both the Great Barrier Reef and reefs globally. There are a number of areas where the manuscript can be improved, mainly for the purposes of providing clarity and sufficient detail to the reader.

Line 52-53: suggest including a reference for "(~5%) of the more than 3800 individual reefs... are targeted for monitoring..."

Line 75-77: "comprehensive" is a vague way to describe the bioregionalised characterisation of the GBR. Could this sentence be rewritten to more specifically describe what is comprehensive, and so that this sentence more precisely sets the scene for the next sentence that describes the limitations of current characterisation of the GBR.

Lines 98-99: "We define coral habitat as...". This is great and sets up an important concept for the paper. However, throughout the paper, "coral habitat" is used at some places and "suitable coral habitat" is used at others. Suggest use of "coral habitat", in place of "suitable coral habitat", for consistency and also because the importance of "suitability" for your definition of coral habitat is made clear on L98-99.

Line 126-127: It would be useful to make a comment, here or elsewhere, on the past success of this type of technology in separating rubble from rock, etc. Which benthic types are more likely to be confused with each other, and which ones are very distinct?

And, importantly, I suggest to state in the results section the discriminant ability of the machine learning pipeline for separating the four benthic types in this particular study. 

Line 131: "[52]" is missing after "Kennedy et al. (2020)"

Lines 138-140: Is the figure caption stating that (i) the flow chart is based on reference [18]; or (ii) the four modules are based on reference [18]; or (iii) the geomorphic and benthic cover classifications are based on reference [18]? Clarify.

Lines 141-142: Define "corrected" in the context of "Sentinel-2 (10 m x 10 m) corrected satellite surface reflectance image mosaic".

Lines 151-152: "Geomorphic reference samples were manually created by experts". It would be useful to provide, though briefly, a definition of geomorphic reference samples and explanation of the method for their creation - otherwise it may be hard for many readers to grasp this aspect of the study.

Line 164-167: Please expand on the use of depth estimates to create the slope component and the trigonometric multiplication of the planar mapped area by the slope component. Might be useful to include a diagram on these two aspects of the methods. Include enough detail for the reader to conceptualise how the 3D surface area was determined, for instance, was there a slope value for every pixel, or one slope value for the string of pixels that connected any shortest path between a point on the reef crest, and the lagoon floor?  

Figure 3: Minor point, is a palm tree the best symbol to use to represent vegetation on Great Barrier Reef islands? Up to you to decide whether to update this or not. Also, if relevant, include (in the figure caption or elsewhere) the attributions for the vector art used in this image. 

Line 172: define "extent"

Line 187-190: It is not clear how the broader generalisation of assuming that hard substrate is the dominant substrate type in the listed locations can account for the contribution of deeper reefs - clarify.

Line 196-197: "were created using a consistent mapping methodology". It seems like you could provide a stronger statement for the novelty and strength of your habitat maps here, either by providing additional information for the reader to understand why a consistent methodology is an advance, or by highlighting another important methodological advance that your study has made. 

Lines 206-207: This sentence implies that the 218 additional reefs contributed all of the additional surface area seen when mapping to 20m depth, which I presume is probably not the case. Maybe restructure sentence to " Mapping reefs to a depth of 20 m based on geomorphic zonation yielded a total 3D surface area of 28,261 km2 and provided information for an additional 218 reefs". 

Line 218: Should "Lewis et al. 2018" be "Lewis et al. 2003 [26]"? Otherwise there might be a reference missing from the reference list. 

Figure 4: The stacked bar chart gives a slightly confusing impression, as it is not clear that, to read the surface area from the axis, the planar area part of the bar and the surface area part of the bar should be added together. I suggest having a stand-alone planar area bar and stand-alone surface area bar, side by side, instead of a stacked bar. 

State in the figure caption which part of the GBR the example map is from, or illustrate in a map inset. 

Perhaps this is not relevant, but the figure illustrates clearly something that was not stated in the main text: that the 2D planar area of the new maps is reduced compared to the Lewis et al. (2003) maps. It might be useful to state this somewhere, along with a statement of whether or not this is significant and why, in order to help the reader interpret the Figure.

Line 269: "thematic detail": state, earlier in the manuscript, what was the best level of thematic detail among the previous attempts to map the GBR? 

Line 289: Remove "Hence," as its presence seems to make this line a non-sequitur.

Line 295-296: "Vertical areas cannot be mapped using earth observation technologies..." But, you can identify vertical walls, by sudden and sharp changes in depth, no? And once identified, could it be assumed that vertical areas are solid substrate and not rubble or sand? Or is it more complex than this?

Line 297-298: "variable (unknown) depth limits": what does this mean?

Lines 397-385 (Data availability): Perhaps include in this statement the availability of the old reef outlines (Lewis et al. 2003) which I believe are also available on GBRMPA's spatial data website. 

Lines 389-391: Some fragmentary sentences, consider revising. 

Reference 20: Correct the LTMP Procedures title. 

Table S2: I found it hard to understand this table. What does "user" and "producer" mean? What does "map" mean in the Benthic cover column, and what is the number reported beside "map"? The table caption could definitely be expanded to provide more explanation of the table.

General: the manuscript would benefit from a grammar check.

Reference details (page numbers, formatting etc.) in the list of references would benefit from review by a sharp eye. 

Author Response

We joined all the reviewers comments in one document as some of the comments is relevant for various reviewers.

Author Response File: Author Response.docx

Reviewer 2 Report

Your paper is of vital importance for multiple usages (support to other scientific studies, also foundation of environmental management/marine planning strategy/political decision... The reviewer believe that because of this, the author of the paper should provide more details on their methodology.  The reviewer believe that, the paper is missing some important points:

•  methodology should be described in more details, especially with respect to the extraction of information from the satellite images (estiamtion of the bathymetry, estimation of the nature of seafloor)
• Some boxes of Fig 2 do not seem to be used/described in the paper ( machine learning use; the wave data)
• estimation on the uncertainty of 1) the zonation and 2) surface coverage
• The reviewer also believe that Figure 5 is difficult to understand and should maybe either better described or even modified to be more explicit.

While the paper is well documented with previous studies, the improvement  provided by this paper to the overall problematic should be better described.

Author Response

We joined all the reviewers comments in one document as some of the comments is relevant for various reviewers.

Reviewer 3 Report

The paper presents a remote sensing study (Sentinel 2 satellite data) aimed at the inventory of shallow coral reefs of the Great Barrier Reef (GBR) in Australia, a major feature of the East Coast of the country. The paper has merits, and despite it ‘only’ provides inventories, coral reef basic inventories are still needed.

The paper is interesting but also relying, methodologically, on various papers, and is not self-standing. It is normal to use previously published papers, but here lots of critical information are therefore somewhat hidden, or unclear. Therefore, I believe the paper could be much more useful if several elements are added. First, many key method information come from these other papers. Although they are published, some background information would be usefully included here, for instance, what is the accuracy of the remote sensing-inferred depth product that is used here? It is used as it was perfectly accurate, but it is certainly not the case. Second, I find confusing to follow the results; it lacks a clear comparison with previous estimates and more importantly does not provide uncertainties in the areal estimates (especially for the benthos level); finally it lacks a critical assessment of what could be improved or how generalizable this work is for other large areas.

As it stands, it is not granted that the results are better, or at least more useful for managers and scientists than some previous information without these missing elements. Suggestions to improve all these elements should be taken into account for a revised version. I detail some aspects hereafter.

The paper objective is about providing new, hopefully better, reef inventories, but the description of what was previously available need to be improved. The paper should put in a Table (Introduction , or early in the Discussion) all the previous estimates of coral reef extent that may exist for the Great Barrier Reef, or part of it if large sub-areas, like here for the inshore reefs, were omitted or focus on (e.g., Torres Strait). The paper reports several times a number of studied reefs, which could be a good metrics to reflect the efforts of previous studies too, but it is eventually unclear what has been processed here for the different products, compared to other publications. I suggest to review briefly but systematically the methods, image data, extent (inshore, mid-shelf, shelf reef, or regions such as Torres Straits), surface area results and uncertainties (if any reported), number of processed reefs, and the range of depth and also its uncertainty. Depth limits is indeed a key information for sound comparison between all the previous estimates. The authors are quite clear of the limits they aim to address (0-10 and 0-20m depending on the classification levels), but here the depth is itself derived from remote sensing and it is very unclear how accurate this is. This does not help to evaluate if the computations of ‘3D area’ is itself accurate or not, and in which range. I am not sure as I do not have access at the moment but more information could be available than what is given here for historical work in the GBRMPA, with reports from Kuchler et al. (back from the 80s with Landsat), from the Hopley et al. books, and possibly from conference proceedings. Also, I could not find a pdf of Lewis et al. 2003, which I regret because  the reef outline of what is referenced in the figure as Lewis et al. 2018 (or is it 2003?) provided in Figure 4 is probably more accurate in term of coral reef ecosystem extent than the other, shallow (0-20m) ones. Bottom-line: the comparisons between the apples and the oranges that were used to estimate the extent of the GBR need to be much clearer. As it stands, I don’t necessarily believe the estimates from this paper are better or more accurate than previous one, globally for the GBR or for sections of it. Logically, the title also could be more appropriate and informative with “ A new estimate of GBR ~0-20 meters coral reef extent from Sentinel 2 imagery” for instance.

The authors report accuracy error (Table S2). The achieved accuracy per category are actually low, or even very low, for geomorphology, which should be very accurately mapped. As a reef scientist, there is little reason I would use such low accuracy map products. I guess same for managers or map users. Also some results are a bit strange such as the low accuracy of sand compared to a coral-algal complex, which goes against all published results. Any reasons? And I don’t believe it is because of mixing since large sand area exists on GBR reefs everywhere. I believe it is also necessary to point to these poor results, for objective presentation of their implications. As such, I think it is necessary to include explicitly the uncertainty in surface areas that these classification errors implies, for each category of benthos/reef types. So please use systematically a ‘mean +- error’ presentation when giving results, including in the figures (Figs. 4, 6).

Then, I find the discussion a bit thin, and lacking in-depth discussion of the limits of this work. The limits will be more obvious when uncertainties are reported in the surface areas extent, and what these uncertainties mean, especially when compared to the previous estimates (after all relevant classes are merged for a meaningful comparison). The statement in the abstract that a 31% increase is evidenced is not obvious. Also, it will be useful to have an idea of the cost of this initiative that seems to be labor and technology intensive. Sentinel iImages are free, but the reflectance mosaic that was created, the field work, the classification, etc. have a cost, and it will be useful to have an estimate, not just monetary, as it can vary by country, but in hours/days/weeks/months.

Other aspects can include:

-avoid statement such as ‘world most extensive coral reef ecosystem’ which is itself largely dependent of a clear definition of what is a ‘coral reef ecosystem’. The GBR itself is a mosaic of different sub-systems, or bioregions as it is said, that could be matched by other areas of continuous reef growth (Red Sea, Indonesia, etc..) if they are considered as one system.

-Some classes are unclear: what is a plateau? What defines a sheltered reef slope? Based on wave modelling? If so what are the thresholds in exposure?

- some statements may be argued (such as hyperspectral sensors can differentiate coral and algae), and there are several. So may be just use more precautions in these statements. Such as ‘should’ or ‘generally’ to balance the statement.

- I think a example map of a large section of the GBR reefs would be interesting, or even several in supplemental material. Three reefs are shown, including Heron Island which is shown in a lot of publications, but more would be interesting, from north to south. These three maps also show problems in the product (noise) and some obvious misclassification in both geomorphic and benthic categories (and indeed the sand is obviously misclassified and over represented). Hence, a A4 format map showing tens of reefs would be useful to see more of the good and the bad in these products.

Author Response

We joined all the reviewers comments in one document as some of the comments is relevant for various reviewers.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The reviewer thanks the author for the improvement brought to the paper.

Few questions are remaining though:

• How did you manage to vertically reference the estimated waterdepths to the MSL (tide measured/modelled surface) ?
• It is not clear which kind of machine learning classification algorithm you have use ? Could you detail or provide a reference ?
• Line 176-177. Likewise, please provide more details on the deep learning neural network modeled you used for the benthic composition derivation ?
• line 159: providing more information on the field bathymetric data used for the purpose of the comparison would help the reader to be more confident in the bathymetric information you generated. Please provide at least the type of sensor, the data at which those data were acquired, the number of samples, some element relative to the in-situ coverage
• Figure 4: Error bars are barely visible (which is nice in some ways, but at the present stage are meaningless on the figure)
• line 369-372: It is not clear if the bathymetric model you mention is part of your study. You have not introduced this information before
• line 424 and following: this section looks like a conclusio section rather than a new discussion item. You may want to title it accordingly.

Refenece 4 it should be R. (Rodolphe) Devillers

Author Response

Please see the attachment

Author Response File: Author Response.pdf