Optimising Sampling Strategies in Coral Reefs Using Large-Area Mosaics

Round 1

Reviewer 1 Report

It is highly appreciated that the authors conducted a large-scale coral reef sampling using a new underwater camera system, and analysed coral classification and coverage by careful data analysis.
In addition, it is also a very unique research content that the confirmation of the effectiveness of the conventional photoquadrat-based method by an advanced mosaic data, and applying to make a guideline for the conventional coral sampling.

Comments to the authors;
1. There are 8 survey points, and data in 8 locations are written in Table 2, but there are only 7 locations written in the description on lines 93 to 94 in the text.
2. I couldn't understand the part in 5.conclusions, "sampling designs covering two percent (with 0.5 x 0.5 m quadrats) or three percent (with 2 x 2 m quadrats) of the total area are needed". What the underlying data are. Is it Fig.6?
3. When considering sampling strategies as a long-term coral reef monitoring, it must be assumed that coral bleaching and a significant decrease in coverage will occur at the sampling point in the future.
Is it meaningful to optimize the long-term sampling plan and create guidelines, based on the current coverage and taxonomic abundant?

Author Response

Point-by-point response uploaded as a PDF file

Author Response File: Author Response.pdf

Reviewer 2 Report

Line 37 suggest ‘multiple stressors’, rather than simply ‘stressors’

Line 38 Please reword this sentence as it is really the crux of why the work presented is important. Rigorous quantification of marine habitats does not satisfy ecological or management criteria, rather, provides the data necessary to understanding ecological processes and so informing management actions. I don’t believe ref 13-14 really get at the tenet of this statement.  

Line 45-49 These three sentences are quite opaque. It would benefit the reader to be more specific. I suspect the point being raised is that information is required at a range of scales dependant on the  processes of interest?  

Line 47. Suggest remove the word “Fortunately” and replace “allow considerations at large spatial scales” with something along the lines of “have vastly increased the resolution of information that can be efficiently gathered over large areas of habitat”.

Line 63. Refs 24 and 25 are not focused on sampling protocols per se.

Line 103  is “do capturing” a typo?

Table one Comments row 1, should the word “keys”, be simply “key”

Line 150 is the word “the” omitted from “covered by the mesh”

Table 2. Including the abbreviated names of the sampled sites under the head transects is confusing. Perhaps include the word location rather than transect to identify the different sample sites. Also the label quadrat size appears to apply to the transects and should be moved to span the columns to which the label actually applies.

Line 180 the word is “altogether” not “all together” as written.

Figure 2. It would be valuable to define the range of cover types, indicated on the plots.

Figure 3. I suspect the box plots are not correctly described. Typically the box would include the second and third quartiles about the median, not the first and third as described. The span of the whiskers should also be described.

Figure 4, again boxplot is not correctly described.

Figure 6. It would seem appropriate to provide figures for comparable areas in each plot. Sampling area of 1 2x2m quad would equate to 4 1x1m quadrats and 16 0.5X0.5m quadrats. So in the plot the x-axis  for Figure A at 160 should align with Figure b at 40 and  Figure c at 10.

Line 309 stating the 2 x 2 m quadrat size was the most efficient is misleading and needs qualification as sampling the larger quadrats is not comparable to sampling the smaller ones.  Figure 5a suggests slight larger quadrats require higher proportion of the area be surveyed for any given abundance.

Line 320. Reference to point intercept here is unwarranted. The study relates to photo quadrats the estimate cover from the entire quadrat. Point intercept methods sub sample the photo (quadrat). The results presented do not provide insight into sampling strategy for point intercept photos.

Line 325. The point relating to higher resolution is sound, however this is not something this work provides evidence of. Simply the results suggest that what is important in the proportion of area covered, with little or no difference in precision due to quadrat size at a given proportion of total area sampled.

Line 330 the suggestion is that sub-sampling of large photos will be an option to reduce analysis effort however this completely ignores that the precision of information returned will be substantially reduced form a few points compared to the annotation of the entire area.

#345 the word “are” should be “area”

#343 it is not clear to me how the two components of this sentence relate. The sentence starts by suggesting there is a limitation to estimating the true value of the parameter being quantified, which as a sampling issue. And then goes on to suggest the high-res large area imaging provides a point of truth. This needs a bit of clarification.

#354-355. It needs to be clearly stated that differences in sampling of different benthic classes cannot be differentiated from differences due to the abundance of those classes in this work.

#356 It is not clear where the figures of 2 and 3% are derived. Considering Figure 6 the proportion of categories returning representative cover estimates at the desired precision increases from 2.4 % (a single, presumably the most abundant, category) with the use of 27 0.5x0.5m quadrats = (3% of the 250m2 transects) or three 2x2m quadrats (5% of the 250 m2 transects). Allowing for the fact that not the full 250m of transects could accommodate quadrats (Table 2) the actual proportion of cover was greater. As the main point of this work it is critical to capture this correctly.

it is critical that a better job is done of summarising the results in a way that provides clear guidance.

Ultimately, I feel the manuscript is overly complicated. As precision of estimates is largely a mathematical function of sample size and variability in the distribution of targets, I can’t help concluding that the primary result of this work is presented in Figure 2 (A). It is this figure that provides the loss of precision given cover of a target and level of subsampling.

Much of the following analysis and presentation of results does not provide substantially clearer guidance (stated objective (3) lines86-88), for benthic sampling designs.  All the ANOVA’s of individual taxa and Figure 3 boxplots appear to be confounded by differences in cover among transects that is clearly evident in Figure 2A. The results of these analyses are barely discussed, and this in itself suggests the authors where at a loss as to their value.

In several places point intercept methods are raised but not in a way that is supported by the work undertaken. The way methods not investigated by the work undertaken are considered needs careful evaluation. For example, at line 332 it is suggested that subsampling quadrats using point intercept could be considered however this would completely alter the precision of the estimates compared to those of a quadrat based sampling design.

 

 

Author Response

Point-by-point response uploaded as a PDF file

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript by Lechene et al. presents an analysis of the effects of sampling (number of quadrats of a given size) on precision of cover estimates of biota on a coral reef. I find the subject matter to be important given the need to track changes in benthic ecosystems through time and the development of new technologies which can help researchers achieve this goal. The manuscript is generally well written; however, I currently have a few major concerns which I feel would need addressing before this work could be published.

Subsampling of large datasets of marine imagery for detailed scoring is common due to the large volume of imagery that is often collected. It is not currently clear why you might need to do this for the technology you have deployed in this study. Given that total area enumeration of a transect took 4-24 hours, why wouldn’t you just score the whole mosaic? While I think you have generated a useful data set to answer research questions for those faced with the need to subsample, you should discuss these points in more detail. Furthermore, it is not currently clear in your analysis if the aim is to get precise estimates of cover for each transect or over the whole study area of Heron Island. It appears that percent cover has been pooled for all eight transects. No transect level data is presented, or variability between transects. Was the analysis of random subsampling of quadrats done within a transect and then the results averaged over the eight transects? This is not clear and yet crucial to interpreting the results. Also, is the aim to have precise transect level estimates, or whole reef level estimates? You really should include a map of your study area and the transect locations. I am left wondering how representative the transects are of the reef, and how a long-term monitoring program using this approach would be designed.

I have major reservations about the methods used for the subsampling of quadrats across each mosaic. You state that subsamples are taken “with replacement” in order to achieve a specified total percentage of coverage. This means that the same quadrat can be selected two (or more) times in order to get your estimate. The selection of the same quadrat multiple times is more likely for the larger quadrats, particularly where you are aiming to achieve higher percentage of total coverage. This will lead to a bias in your results. Furthermore, this is not what a researcher would do if they were choosing to subsample images from a large area of benthos. It would be pointless to score the same image or quadrat twice. Researchers would be far more interested in how the subsampled images were spread across the area: which is better? Random? Systematic? Spatially balanced? How does the distribution of the target biota interact with these choices?

I feel that further thought needs to be put into how the results may be applied to studies elsewhere, or even to future work using this methodology. There is currently little discussion around how similar studies are conducted and how these results might be applicable. For example, point counts within images are often used, which will add more uncertainty to cover estimates. The digitised cover you have used is likely to be much more accurate than this. While comparing a point count approach to a digitising approach may be out of the scope of your manuscript you should at least discuss this point. You should also discuss how the methods you have developed might be used in a monitoring context. For example, a more detailed discussion around the trade-off in terms of time spent in the field vs scoring is warranted. Would you be better off conducting more transects and then subsampling these?

I provide some more specific feedback below:

 

 

Line11: Surveys often cover less than 5-10% of the area. I think you should support this statement with some references in the main text.

Line 14: Here you mention ~300 m², but in the main text 250 m² is used.

Lines 18-19: Here and elsewhere: It’s not surprising that larger quadrats are able to provide higher precision when number of quadrats is used as the metric. With 0.5 m quadrats you have 0.25 m^2 covered per quadrat, whereas with 2 m quadrats you have 4 m^2 which is 16 times the area. I’m not sure how useful it is to analyse things in this way.

Lines 29-32: I don’t think this is the best way to lead into the introduction. While this is true, this study does not look into the relationship between reef structure and biodiversity. The main focus should be on the importance of developing monitoring programs capable of sufficient precision to track change.

Line 45: “…with respect to the spatial extent” of what? Please qualify this statement.

Line 58: By “computer analyses” I assume you mean image scoring/annotation. Consider changing the wording as this brings to mind statistical analysis of the resulting data.

Lines76-78: In previous lines you have been discussing precision in percent cover, but here you have switched to talking about taxonomic richness. Be careful as these are two different things. Precision required to “adequately” describe taxonomic richness will depend on the taxonomic level chosen.

Line 81: I think you are testing sampling “effort” rather than “design” as you have only tested random sampling.

Lines 91-94: Considering you state that this is a “well studied coral cay” I find it unusual that you don’t cite any other studies in your discussion and compare your findings. Please include a map of the study area and the location where the data was collected.

Lines 98-99: It’s not clear why these visual references were necessary. Either include this information or delete this line and refer readers to the Figueira (2015) paper for further details.

Table 1: Consider moving this to an Appendix. This means little to someone who is not familiar with the software.

Line 120: Here you mention 250m², whereas in the introduction and abstract you use 300m². I would recommend stating 250m² throughout as this is what you used in the analysis.

Line 123: Why were these subclasses chosen? Is this what is commonly used? What level could be identified from imagery?

Figure 1: In the “virtual sampling” panel the 2 m quadrats clearly cover a smaller portion of the whole transect. I think you should at least discuss that this may have lead to some bias as you are not comparing the same areas. I think it would have been preferable to have 2 m cells with complete coverage (as in the figure) and then break these into the smaller cells.

Line 143: If you can, please cite some studies that use these scales.

Line 151: did you really use “sampling with replacement”? Please see my earlier comments.

Lines 151-154: Coefficient of variation is more commonly used, and would be the same thing as RSE.

Lines 165-166: You state that this power relationship was always followed. How closely? In my experience these relationships are followed less clearly with rarer biota. Did you use this fitted relationship for the precision estimate? If so, I’m left wondering how this might have influenced the results, as you are then calculating where this curve begins to level off (i.e. a “smooth of a smooth”).

Lines 179-182: I agree that the important thing is that a common criterion is used. However, see my above comment. Perhaps you could show how well the power curve fit for some example biota in an appendix figure.

Line185: Did you pool across all transects for the final results? It seems so, but you need to be clear here.

Line 196: Here and in your title: be careful of using the word “optimal”. Optimal for what? In this case 100% is always optimal as it gives the most precise result!

Line 208: What do you mean by “specific identity”? It seems that percent cover was the thing you analysed and thus was the only criteria used.

Line 232: I think this should be < 1%.

Table 3: It appears from this table that there was considerable variability between transects. For example, one of the highest percent cover groups (bleached staghorn coral) had 17% mean cover, but a se of 17 and only occurred in 2 transects. I think this part of your study is crucially important to discuss. Where are these two transects? Are they closely located? By pooling your analysis of quadrats this gets lost. Is it more important to spread you transects out over the reef than to sample intensively within transects? Considering it seems to be reef level cover you’re interested in, what proportion of the entire reef is sampled by the transects?

Line 239: Variability in what?

Figure 3 and 4: It appears you have the labels for (B) and (C) around the wrong way in the figure legend.

Lines 256-262: I’m not convinced of the usefulness of analysing the number of quadrats used given the huge difference in total area sampled. Also, here you refer to “low sampling effort” as >30% of the total area. Given that you state that most benthic surveys sample 5-10 % of the area, this would not be considered low! You need to discuss this fact.

Lines 271-279: See my comments above regarding the level of sampling and what is typically achieved.

Figure 5: Relative abundance is usually used when referring to counts of organisms. While percent cover is a measure of relative abundance, I think it would be preferable here and in the text to just use percent cover.

Figure 6: This would be more useful if the x axes were equivalent in terms of the total area sampled. For example, the lowest levels of sampling aren’t equivalent: 27*0.5 m quadrats is 6.75 m²,  10*1m quadrats is 10 m² and 3*2m quadrats is 12m². Plotting 3 lines on the same plot but using percent area would be more informative to see if the ere was a difference.

Line 308: What do you mean by “no significant difference in sample size”?

Line 309-310: Is this based on using number of quadrats? If so, this is totally expected given the large increase in area surveyed.

Line 312: Do you mean to say “precision” here rather than “sampling effort”?

Lines 318-319: What does this mean? Doesn’t sampling effort always scale with the number of quadrats?

Lines 320-322: Here you need to elaborate on how other approaches may compare to what you’ve done. You have precise quadrat level estimates due to digitising, whereas point intercept methods will have lower precision. What are the implications?

Lines 325-326: “Eventually” seems a strange word to use. Also, the spatial distribution of a target is important here.

Line 345: Change “are” to “area”

Lines 336-350: I think this paragraph needs to be broken up and expanded on. This is the important stuff! Yes, this approach seems like a powerful tool. However, thought needs to be given to how it is used. How many transects are necessary? Do you aim to repeat transects over time or take a random sample of the reef? If the latter, then there is also an interaction between the number of quadrats required and the number of transects required as you will be sampling different patches of reef each time.

Lines 355-357: What?!? Where did the two or three percent come from? This is the first mention of this! Is this of the whole reef system? It appears that the aims of the study were not clearly defined and part of the analysis not presented. If the aim was to determine the proportion of the whole reef that should be sampled then this should be clearly stated up front. Also, if this is the case then how representative of the reef are the transects?

 

Author Response

Point-by-point response uploaded as a PDF file

Author Response File: Author Response.pdf